xref: /openbmc/linux/net/dccp/ipv4.c (revision 07c7c6bf)
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 int 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), 0);
260 	if (!sk) {
261 		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
262 		return -ENOENT;
263 	}
264 
265 	if (sk->sk_state == DCCP_TIME_WAIT) {
266 		inet_twsk_put(inet_twsk(sk));
267 		return 0;
268 	}
269 	seq = dccp_hdr_seq(dh);
270 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
271 		dccp_req_err(sk, seq);
272 		return 0;
273 	}
274 
275 	bh_lock_sock(sk);
276 	/* If too many ICMPs get dropped on busy
277 	 * servers this needs to be solved differently.
278 	 */
279 	if (sock_owned_by_user(sk))
280 		__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
281 
282 	if (sk->sk_state == DCCP_CLOSED)
283 		goto out;
284 
285 	dp = dccp_sk(sk);
286 	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
287 	    !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
288 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
289 		goto out;
290 	}
291 
292 	switch (type) {
293 	case ICMP_REDIRECT:
294 		if (!sock_owned_by_user(sk))
295 			dccp_do_redirect(skb, sk);
296 		goto out;
297 	case ICMP_SOURCE_QUENCH:
298 		/* Just silently ignore these. */
299 		goto out;
300 	case ICMP_PARAMETERPROB:
301 		err = EPROTO;
302 		break;
303 	case ICMP_DEST_UNREACH:
304 		if (code > NR_ICMP_UNREACH)
305 			goto out;
306 
307 		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
308 			if (!sock_owned_by_user(sk))
309 				dccp_do_pmtu_discovery(sk, iph, info);
310 			goto out;
311 		}
312 
313 		err = icmp_err_convert[code].errno;
314 		break;
315 	case ICMP_TIME_EXCEEDED:
316 		err = EHOSTUNREACH;
317 		break;
318 	default:
319 		goto out;
320 	}
321 
322 	switch (sk->sk_state) {
323 	case DCCP_REQUESTING:
324 	case DCCP_RESPOND:
325 		if (!sock_owned_by_user(sk)) {
326 			__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
327 			sk->sk_err = err;
328 
329 			sk->sk_error_report(sk);
330 
331 			dccp_done(sk);
332 		} else
333 			sk->sk_err_soft = err;
334 		goto out;
335 	}
336 
337 	/* If we've already connected we will keep trying
338 	 * until we time out, or the user gives up.
339 	 *
340 	 * rfc1122 4.2.3.9 allows to consider as hard errors
341 	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
342 	 * but it is obsoleted by pmtu discovery).
343 	 *
344 	 * Note, that in modern internet, where routing is unreliable
345 	 * and in each dark corner broken firewalls sit, sending random
346 	 * errors ordered by their masters even this two messages finally lose
347 	 * their original sense (even Linux sends invalid PORT_UNREACHs)
348 	 *
349 	 * Now we are in compliance with RFCs.
350 	 *							--ANK (980905)
351 	 */
352 
353 	inet = inet_sk(sk);
354 	if (!sock_owned_by_user(sk) && inet->recverr) {
355 		sk->sk_err = err;
356 		sk->sk_error_report(sk);
357 	} else /* Only an error on timeout */
358 		sk->sk_err_soft = err;
359 out:
360 	bh_unlock_sock(sk);
361 	sock_put(sk);
362 	return 0;
363 }
364 
365 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
366 				      __be32 src, __be32 dst)
367 {
368 	return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
369 }
370 
371 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
372 {
373 	const struct inet_sock *inet = inet_sk(sk);
374 	struct dccp_hdr *dh = dccp_hdr(skb);
375 
376 	dccp_csum_outgoing(skb);
377 	dh->dccph_checksum = dccp_v4_csum_finish(skb,
378 						 inet->inet_saddr,
379 						 inet->inet_daddr);
380 }
381 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
382 
383 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
384 {
385 	return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
386 					   ip_hdr(skb)->saddr,
387 					   dccp_hdr(skb)->dccph_dport,
388 					   dccp_hdr(skb)->dccph_sport);
389 }
390 
391 /*
392  * The three way handshake has completed - we got a valid ACK or DATAACK -
393  * now create the new socket.
394  *
395  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
396  */
397 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
398 				       struct sk_buff *skb,
399 				       struct request_sock *req,
400 				       struct dst_entry *dst,
401 				       struct request_sock *req_unhash,
402 				       bool *own_req)
403 {
404 	struct inet_request_sock *ireq;
405 	struct inet_sock *newinet;
406 	struct sock *newsk;
407 
408 	if (sk_acceptq_is_full(sk))
409 		goto exit_overflow;
410 
411 	newsk = dccp_create_openreq_child(sk, req, skb);
412 	if (newsk == NULL)
413 		goto exit_nonewsk;
414 
415 	newinet		   = inet_sk(newsk);
416 	ireq		   = inet_rsk(req);
417 	sk_daddr_set(newsk, ireq->ir_rmt_addr);
418 	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
419 	newinet->inet_saddr	= ireq->ir_loc_addr;
420 	RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
421 	newinet->mc_index  = inet_iif(skb);
422 	newinet->mc_ttl	   = ip_hdr(skb)->ttl;
423 	newinet->inet_id   = jiffies;
424 
425 	if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
426 		goto put_and_exit;
427 
428 	sk_setup_caps(newsk, dst);
429 
430 	dccp_sync_mss(newsk, dst_mtu(dst));
431 
432 	if (__inet_inherit_port(sk, newsk) < 0)
433 		goto put_and_exit;
434 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
435 	if (*own_req)
436 		ireq->ireq_opt = NULL;
437 	else
438 		newinet->inet_opt = NULL;
439 	return newsk;
440 
441 exit_overflow:
442 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
443 exit_nonewsk:
444 	dst_release(dst);
445 exit:
446 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
447 	return NULL;
448 put_and_exit:
449 	newinet->inet_opt = NULL;
450 	inet_csk_prepare_forced_close(newsk);
451 	dccp_done(newsk);
452 	goto exit;
453 }
454 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
455 
456 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
457 					   struct sk_buff *skb)
458 {
459 	struct rtable *rt;
460 	const struct iphdr *iph = ip_hdr(skb);
461 	struct flowi4 fl4 = {
462 		.flowi4_oif = inet_iif(skb),
463 		.daddr = iph->saddr,
464 		.saddr = iph->daddr,
465 		.flowi4_tos = RT_CONN_FLAGS(sk),
466 		.flowi4_proto = sk->sk_protocol,
467 		.fl4_sport = dccp_hdr(skb)->dccph_dport,
468 		.fl4_dport = dccp_hdr(skb)->dccph_sport,
469 	};
470 
471 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
472 	rt = ip_route_output_flow(net, &fl4, sk);
473 	if (IS_ERR(rt)) {
474 		IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
475 		return NULL;
476 	}
477 
478 	return &rt->dst;
479 }
480 
481 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
482 {
483 	int err = -1;
484 	struct sk_buff *skb;
485 	struct dst_entry *dst;
486 	struct flowi4 fl4;
487 
488 	dst = inet_csk_route_req(sk, &fl4, req);
489 	if (dst == NULL)
490 		goto out;
491 
492 	skb = dccp_make_response(sk, dst, req);
493 	if (skb != NULL) {
494 		const struct inet_request_sock *ireq = inet_rsk(req);
495 		struct dccp_hdr *dh = dccp_hdr(skb);
496 
497 		dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
498 							      ireq->ir_rmt_addr);
499 		rcu_read_lock();
500 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
501 					    ireq->ir_rmt_addr,
502 					    rcu_dereference(ireq->ireq_opt));
503 		rcu_read_unlock();
504 		err = net_xmit_eval(err);
505 	}
506 
507 out:
508 	dst_release(dst);
509 	return err;
510 }
511 
512 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
513 {
514 	int err;
515 	const struct iphdr *rxiph;
516 	struct sk_buff *skb;
517 	struct dst_entry *dst;
518 	struct net *net = dev_net(skb_dst(rxskb)->dev);
519 	struct sock *ctl_sk = net->dccp.v4_ctl_sk;
520 
521 	/* Never send a reset in response to a reset. */
522 	if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
523 		return;
524 
525 	if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
526 		return;
527 
528 	dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
529 	if (dst == NULL)
530 		return;
531 
532 	skb = dccp_ctl_make_reset(ctl_sk, rxskb);
533 	if (skb == NULL)
534 		goto out;
535 
536 	rxiph = ip_hdr(rxskb);
537 	dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
538 								 rxiph->daddr);
539 	skb_dst_set(skb, dst_clone(dst));
540 
541 	local_bh_disable();
542 	bh_lock_sock(ctl_sk);
543 	err = ip_build_and_send_pkt(skb, ctl_sk,
544 				    rxiph->daddr, rxiph->saddr, NULL);
545 	bh_unlock_sock(ctl_sk);
546 
547 	if (net_xmit_eval(err) == 0) {
548 		__DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
549 		__DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
550 	}
551 	local_bh_enable();
552 out:
553 	dst_release(dst);
554 }
555 
556 static void dccp_v4_reqsk_destructor(struct request_sock *req)
557 {
558 	dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
559 	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
560 }
561 
562 void dccp_syn_ack_timeout(const struct request_sock *req)
563 {
564 }
565 EXPORT_SYMBOL(dccp_syn_ack_timeout);
566 
567 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
568 	.family		= PF_INET,
569 	.obj_size	= sizeof(struct dccp_request_sock),
570 	.rtx_syn_ack	= dccp_v4_send_response,
571 	.send_ack	= dccp_reqsk_send_ack,
572 	.destructor	= dccp_v4_reqsk_destructor,
573 	.send_reset	= dccp_v4_ctl_send_reset,
574 	.syn_ack_timeout = dccp_syn_ack_timeout,
575 };
576 
577 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
578 {
579 	struct inet_request_sock *ireq;
580 	struct request_sock *req;
581 	struct dccp_request_sock *dreq;
582 	const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
583 	struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
584 
585 	/* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
586 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
587 		return 0;	/* discard, don't send a reset here */
588 
589 	if (dccp_bad_service_code(sk, service)) {
590 		dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
591 		goto drop;
592 	}
593 	/*
594 	 * TW buckets are converted to open requests without
595 	 * limitations, they conserve resources and peer is
596 	 * evidently real one.
597 	 */
598 	dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
599 	if (inet_csk_reqsk_queue_is_full(sk))
600 		goto drop;
601 
602 	if (sk_acceptq_is_full(sk))
603 		goto drop;
604 
605 	req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
606 	if (req == NULL)
607 		goto drop;
608 
609 	if (dccp_reqsk_init(req, dccp_sk(sk), skb))
610 		goto drop_and_free;
611 
612 	dreq = dccp_rsk(req);
613 	if (dccp_parse_options(sk, dreq, skb))
614 		goto drop_and_free;
615 
616 	if (security_inet_conn_request(sk, skb, req))
617 		goto drop_and_free;
618 
619 	ireq = inet_rsk(req);
620 	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
621 	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
622 	ireq->ir_mark = inet_request_mark(sk, skb);
623 	ireq->ireq_family = AF_INET;
624 	ireq->ir_iif = sk->sk_bound_dev_if;
625 
626 	/*
627 	 * Step 3: Process LISTEN state
628 	 *
629 	 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
630 	 *
631 	 * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
632 	 */
633 	dreq->dreq_isr	   = dcb->dccpd_seq;
634 	dreq->dreq_gsr	   = dreq->dreq_isr;
635 	dreq->dreq_iss	   = dccp_v4_init_sequence(skb);
636 	dreq->dreq_gss     = dreq->dreq_iss;
637 	dreq->dreq_service = service;
638 
639 	if (dccp_v4_send_response(sk, req))
640 		goto drop_and_free;
641 
642 	inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
643 	reqsk_put(req);
644 	return 0;
645 
646 drop_and_free:
647 	reqsk_free(req);
648 drop:
649 	__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
650 	return -1;
651 }
652 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
653 
654 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
655 {
656 	struct dccp_hdr *dh = dccp_hdr(skb);
657 
658 	if (sk->sk_state == DCCP_OPEN) { /* Fast path */
659 		if (dccp_rcv_established(sk, skb, dh, skb->len))
660 			goto reset;
661 		return 0;
662 	}
663 
664 	/*
665 	 *  Step 3: Process LISTEN state
666 	 *	 If P.type == Request or P contains a valid Init Cookie option,
667 	 *	      (* Must scan the packet's options to check for Init
668 	 *		 Cookies.  Only Init Cookies are processed here,
669 	 *		 however; other options are processed in Step 8.  This
670 	 *		 scan need only be performed if the endpoint uses Init
671 	 *		 Cookies *)
672 	 *	      (* Generate a new socket and switch to that socket *)
673 	 *	      Set S := new socket for this port pair
674 	 *	      S.state = RESPOND
675 	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
676 	 *	      Initialize S.GAR := S.ISS
677 	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
678 	 *	      Continue with S.state == RESPOND
679 	 *	      (* A Response packet will be generated in Step 11 *)
680 	 *	 Otherwise,
681 	 *	      Generate Reset(No Connection) unless P.type == Reset
682 	 *	      Drop packet and return
683 	 *
684 	 * NOTE: the check for the packet types is done in
685 	 *	 dccp_rcv_state_process
686 	 */
687 
688 	if (dccp_rcv_state_process(sk, skb, dh, skb->len))
689 		goto reset;
690 	return 0;
691 
692 reset:
693 	dccp_v4_ctl_send_reset(sk, skb);
694 	kfree_skb(skb);
695 	return 0;
696 }
697 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
698 
699 /**
700  *	dccp_invalid_packet  -  check for malformed packets
701  *	Implements RFC 4340, 8.5:  Step 1: Check header basics
702  *	Packets that fail these checks are ignored and do not receive Resets.
703  */
704 int dccp_invalid_packet(struct sk_buff *skb)
705 {
706 	const struct dccp_hdr *dh;
707 	unsigned int cscov;
708 	u8 dccph_doff;
709 
710 	if (skb->pkt_type != PACKET_HOST)
711 		return 1;
712 
713 	/* If the packet is shorter than 12 bytes, drop packet and return */
714 	if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
715 		DCCP_WARN("pskb_may_pull failed\n");
716 		return 1;
717 	}
718 
719 	dh = dccp_hdr(skb);
720 
721 	/* If P.type is not understood, drop packet and return */
722 	if (dh->dccph_type >= DCCP_PKT_INVALID) {
723 		DCCP_WARN("invalid packet type\n");
724 		return 1;
725 	}
726 
727 	/*
728 	 * If P.Data Offset is too small for packet type, drop packet and return
729 	 */
730 	dccph_doff = dh->dccph_doff;
731 	if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
732 		DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
733 		return 1;
734 	}
735 	/*
736 	 * If P.Data Offset is too too large for packet, drop packet and return
737 	 */
738 	if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
739 		DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
740 		return 1;
741 	}
742 	dh = dccp_hdr(skb);
743 	/*
744 	 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
745 	 * has short sequence numbers), drop packet and return
746 	 */
747 	if ((dh->dccph_type < DCCP_PKT_DATA    ||
748 	    dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
749 		DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
750 			  dccp_packet_name(dh->dccph_type));
751 		return 1;
752 	}
753 
754 	/*
755 	 * If P.CsCov is too large for the packet size, drop packet and return.
756 	 * This must come _before_ checksumming (not as RFC 4340 suggests).
757 	 */
758 	cscov = dccp_csum_coverage(skb);
759 	if (cscov > skb->len) {
760 		DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
761 			  dh->dccph_cscov, skb->len);
762 		return 1;
763 	}
764 
765 	/* If header checksum is incorrect, drop packet and return.
766 	 * (This step is completed in the AF-dependent functions.) */
767 	skb->csum = skb_checksum(skb, 0, cscov, 0);
768 
769 	return 0;
770 }
771 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
772 
773 /* this is called when real data arrives */
774 static int dccp_v4_rcv(struct sk_buff *skb)
775 {
776 	const struct dccp_hdr *dh;
777 	const struct iphdr *iph;
778 	bool refcounted;
779 	struct sock *sk;
780 	int min_cov;
781 
782 	/* Step 1: Check header basics */
783 
784 	if (dccp_invalid_packet(skb))
785 		goto discard_it;
786 
787 	iph = ip_hdr(skb);
788 	/* Step 1: If header checksum is incorrect, drop packet and return */
789 	if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
790 		DCCP_WARN("dropped packet with invalid checksum\n");
791 		goto discard_it;
792 	}
793 
794 	dh = dccp_hdr(skb);
795 
796 	DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
797 	DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
798 
799 	dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
800 		      dccp_packet_name(dh->dccph_type),
801 		      &iph->saddr, ntohs(dh->dccph_sport),
802 		      &iph->daddr, ntohs(dh->dccph_dport),
803 		      (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
804 
805 	if (dccp_packet_without_ack(skb)) {
806 		DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
807 		dccp_pr_debug_cat("\n");
808 	} else {
809 		DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
810 		dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
811 				  DCCP_SKB_CB(skb)->dccpd_ack_seq);
812 	}
813 
814 lookup:
815 	sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
816 			       dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
817 	if (!sk) {
818 		dccp_pr_debug("failed to look up flow ID in table and "
819 			      "get corresponding socket\n");
820 		goto no_dccp_socket;
821 	}
822 
823 	/*
824 	 * Step 2:
825 	 *	... or S.state == TIMEWAIT,
826 	 *		Generate Reset(No Connection) unless P.type == Reset
827 	 *		Drop packet and return
828 	 */
829 	if (sk->sk_state == DCCP_TIME_WAIT) {
830 		dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
831 		inet_twsk_put(inet_twsk(sk));
832 		goto no_dccp_socket;
833 	}
834 
835 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
836 		struct request_sock *req = inet_reqsk(sk);
837 		struct sock *nsk;
838 
839 		sk = req->rsk_listener;
840 		if (unlikely(sk->sk_state != DCCP_LISTEN)) {
841 			inet_csk_reqsk_queue_drop_and_put(sk, req);
842 			goto lookup;
843 		}
844 		sock_hold(sk);
845 		refcounted = true;
846 		nsk = dccp_check_req(sk, skb, req);
847 		if (!nsk) {
848 			reqsk_put(req);
849 			goto discard_and_relse;
850 		}
851 		if (nsk == sk) {
852 			reqsk_put(req);
853 		} else if (dccp_child_process(sk, nsk, skb)) {
854 			dccp_v4_ctl_send_reset(sk, skb);
855 			goto discard_and_relse;
856 		} else {
857 			sock_put(sk);
858 			return 0;
859 		}
860 	}
861 	/*
862 	 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
863 	 *	o if MinCsCov = 0, only packets with CsCov = 0 are accepted
864 	 *	o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
865 	 */
866 	min_cov = dccp_sk(sk)->dccps_pcrlen;
867 	if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
868 		dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
869 			      dh->dccph_cscov, min_cov);
870 		/* FIXME: "Such packets SHOULD be reported using Data Dropped
871 		 *         options (Section 11.7) with Drop Code 0, Protocol
872 		 *         Constraints."                                     */
873 		goto discard_and_relse;
874 	}
875 
876 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
877 		goto discard_and_relse;
878 	nf_reset(skb);
879 
880 	return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
881 
882 no_dccp_socket:
883 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
884 		goto discard_it;
885 	/*
886 	 * Step 2:
887 	 *	If no socket ...
888 	 *		Generate Reset(No Connection) unless P.type == Reset
889 	 *		Drop packet and return
890 	 */
891 	if (dh->dccph_type != DCCP_PKT_RESET) {
892 		DCCP_SKB_CB(skb)->dccpd_reset_code =
893 					DCCP_RESET_CODE_NO_CONNECTION;
894 		dccp_v4_ctl_send_reset(sk, skb);
895 	}
896 
897 discard_it:
898 	kfree_skb(skb);
899 	return 0;
900 
901 discard_and_relse:
902 	if (refcounted)
903 		sock_put(sk);
904 	goto discard_it;
905 }
906 
907 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
908 	.queue_xmit	   = ip_queue_xmit,
909 	.send_check	   = dccp_v4_send_check,
910 	.rebuild_header	   = inet_sk_rebuild_header,
911 	.conn_request	   = dccp_v4_conn_request,
912 	.syn_recv_sock	   = dccp_v4_request_recv_sock,
913 	.net_header_len	   = sizeof(struct iphdr),
914 	.setsockopt	   = ip_setsockopt,
915 	.getsockopt	   = ip_getsockopt,
916 	.addr2sockaddr	   = inet_csk_addr2sockaddr,
917 	.sockaddr_len	   = sizeof(struct sockaddr_in),
918 #ifdef CONFIG_COMPAT
919 	.compat_setsockopt = compat_ip_setsockopt,
920 	.compat_getsockopt = compat_ip_getsockopt,
921 #endif
922 };
923 
924 static int dccp_v4_init_sock(struct sock *sk)
925 {
926 	static __u8 dccp_v4_ctl_sock_initialized;
927 	int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
928 
929 	if (err == 0) {
930 		if (unlikely(!dccp_v4_ctl_sock_initialized))
931 			dccp_v4_ctl_sock_initialized = 1;
932 		inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
933 	}
934 
935 	return err;
936 }
937 
938 static struct timewait_sock_ops dccp_timewait_sock_ops = {
939 	.twsk_obj_size	= sizeof(struct inet_timewait_sock),
940 };
941 
942 static struct proto dccp_v4_prot = {
943 	.name			= "DCCP",
944 	.owner			= THIS_MODULE,
945 	.close			= dccp_close,
946 	.connect		= dccp_v4_connect,
947 	.disconnect		= dccp_disconnect,
948 	.ioctl			= dccp_ioctl,
949 	.init			= dccp_v4_init_sock,
950 	.setsockopt		= dccp_setsockopt,
951 	.getsockopt		= dccp_getsockopt,
952 	.sendmsg		= dccp_sendmsg,
953 	.recvmsg		= dccp_recvmsg,
954 	.backlog_rcv		= dccp_v4_do_rcv,
955 	.hash			= inet_hash,
956 	.unhash			= inet_unhash,
957 	.accept			= inet_csk_accept,
958 	.get_port		= inet_csk_get_port,
959 	.shutdown		= dccp_shutdown,
960 	.destroy		= dccp_destroy_sock,
961 	.orphan_count		= &dccp_orphan_count,
962 	.max_header		= MAX_DCCP_HEADER,
963 	.obj_size		= sizeof(struct dccp_sock),
964 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
965 	.rsk_prot		= &dccp_request_sock_ops,
966 	.twsk_prot		= &dccp_timewait_sock_ops,
967 	.h.hashinfo		= &dccp_hashinfo,
968 #ifdef CONFIG_COMPAT
969 	.compat_setsockopt	= compat_dccp_setsockopt,
970 	.compat_getsockopt	= compat_dccp_getsockopt,
971 #endif
972 };
973 
974 static const struct net_protocol dccp_v4_protocol = {
975 	.handler	= dccp_v4_rcv,
976 	.err_handler	= dccp_v4_err,
977 	.no_policy	= 1,
978 	.netns_ok	= 1,
979 	.icmp_strict_tag_validation = 1,
980 };
981 
982 static const struct proto_ops inet_dccp_ops = {
983 	.family		   = PF_INET,
984 	.owner		   = THIS_MODULE,
985 	.release	   = inet_release,
986 	.bind		   = inet_bind,
987 	.connect	   = inet_stream_connect,
988 	.socketpair	   = sock_no_socketpair,
989 	.accept		   = inet_accept,
990 	.getname	   = inet_getname,
991 	/* FIXME: work on tcp_poll to rename it to inet_csk_poll */
992 	.poll		   = dccp_poll,
993 	.ioctl		   = inet_ioctl,
994 	.gettstamp	   = sock_gettstamp,
995 	/* FIXME: work on inet_listen to rename it to sock_common_listen */
996 	.listen		   = inet_dccp_listen,
997 	.shutdown	   = inet_shutdown,
998 	.setsockopt	   = sock_common_setsockopt,
999 	.getsockopt	   = sock_common_getsockopt,
1000 	.sendmsg	   = inet_sendmsg,
1001 	.recvmsg	   = sock_common_recvmsg,
1002 	.mmap		   = sock_no_mmap,
1003 	.sendpage	   = sock_no_sendpage,
1004 #ifdef CONFIG_COMPAT
1005 	.compat_setsockopt = compat_sock_common_setsockopt,
1006 	.compat_getsockopt = compat_sock_common_getsockopt,
1007 #endif
1008 };
1009 
1010 static struct inet_protosw dccp_v4_protosw = {
1011 	.type		= SOCK_DCCP,
1012 	.protocol	= IPPROTO_DCCP,
1013 	.prot		= &dccp_v4_prot,
1014 	.ops		= &inet_dccp_ops,
1015 	.flags		= INET_PROTOSW_ICSK,
1016 };
1017 
1018 static int __net_init dccp_v4_init_net(struct net *net)
1019 {
1020 	if (dccp_hashinfo.bhash == NULL)
1021 		return -ESOCKTNOSUPPORT;
1022 
1023 	return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1024 				    SOCK_DCCP, IPPROTO_DCCP, net);
1025 }
1026 
1027 static void __net_exit dccp_v4_exit_net(struct net *net)
1028 {
1029 	inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1030 }
1031 
1032 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1033 {
1034 	inet_twsk_purge(&dccp_hashinfo, AF_INET);
1035 }
1036 
1037 static struct pernet_operations dccp_v4_ops = {
1038 	.init	= dccp_v4_init_net,
1039 	.exit	= dccp_v4_exit_net,
1040 	.exit_batch = dccp_v4_exit_batch,
1041 };
1042 
1043 static int __init dccp_v4_init(void)
1044 {
1045 	int err = proto_register(&dccp_v4_prot, 1);
1046 
1047 	if (err)
1048 		goto out;
1049 
1050 	inet_register_protosw(&dccp_v4_protosw);
1051 
1052 	err = register_pernet_subsys(&dccp_v4_ops);
1053 	if (err)
1054 		goto out_destroy_ctl_sock;
1055 
1056 	err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1057 	if (err)
1058 		goto out_proto_unregister;
1059 
1060 out:
1061 	return err;
1062 out_proto_unregister:
1063 	unregister_pernet_subsys(&dccp_v4_ops);
1064 out_destroy_ctl_sock:
1065 	inet_unregister_protosw(&dccp_v4_protosw);
1066 	proto_unregister(&dccp_v4_prot);
1067 	goto out;
1068 }
1069 
1070 static void __exit dccp_v4_exit(void)
1071 {
1072 	inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1073 	unregister_pernet_subsys(&dccp_v4_ops);
1074 	inet_unregister_protosw(&dccp_v4_protosw);
1075 	proto_unregister(&dccp_v4_prot);
1076 }
1077 
1078 module_init(dccp_v4_init);
1079 module_exit(dccp_v4_exit);
1080 
1081 /*
1082  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1083  * values directly, Also cover the case where the protocol is not specified,
1084  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1085  */
1086 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1087 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1088 MODULE_LICENSE("GPL");
1089 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1090 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1091