xref: /openbmc/linux/net/dccp/ipv4.c (revision c4a7b9b5)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  net/dccp/ipv4.c
4  *
5  *  An implementation of the DCCP protocol
6  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7  */
8 
9 #include <linux/dccp.h>
10 #include <linux/icmp.h>
11 #include <linux/slab.h>
12 #include <linux/module.h>
13 #include <linux/skbuff.h>
14 #include <linux/random.h>
15 
16 #include <net/icmp.h>
17 #include <net/inet_common.h>
18 #include <net/inet_hashtables.h>
19 #include <net/inet_sock.h>
20 #include <net/protocol.h>
21 #include <net/sock.h>
22 #include <net/timewait_sock.h>
23 #include <net/tcp_states.h>
24 #include <net/xfrm.h>
25 #include <net/secure_seq.h>
26 #include <net/netns/generic.h>
27 
28 #include "ackvec.h"
29 #include "ccid.h"
30 #include "dccp.h"
31 #include "feat.h"
32 
33 struct dccp_v4_pernet {
34 	struct sock *v4_ctl_sk;
35 };
36 
37 static unsigned int dccp_v4_pernet_id __read_mostly;
38 
39 /*
40  * The per-net v4_ctl_sk socket is used for responding to
41  * the Out-of-the-blue (OOTB) packets. A control sock will be created
42  * for this socket at the initialization time.
43  */
44 
45 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
46 {
47 	const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
48 	struct inet_bind_hashbucket *prev_addr_hashbucket = NULL;
49 	__be32 daddr, nexthop, prev_sk_rcv_saddr;
50 	struct inet_sock *inet = inet_sk(sk);
51 	struct dccp_sock *dp = dccp_sk(sk);
52 	__be16 orig_sport, orig_dport;
53 	struct flowi4 *fl4;
54 	struct rtable *rt;
55 	int err;
56 	struct ip_options_rcu *inet_opt;
57 
58 	dp->dccps_role = DCCP_ROLE_CLIENT;
59 
60 	if (addr_len < sizeof(struct sockaddr_in))
61 		return -EINVAL;
62 
63 	if (usin->sin_family != AF_INET)
64 		return -EAFNOSUPPORT;
65 
66 	nexthop = daddr = usin->sin_addr.s_addr;
67 
68 	inet_opt = rcu_dereference_protected(inet->inet_opt,
69 					     lockdep_sock_is_held(sk));
70 	if (inet_opt != NULL && inet_opt->opt.srr) {
71 		if (daddr == 0)
72 			return -EINVAL;
73 		nexthop = inet_opt->opt.faddr;
74 	}
75 
76 	orig_sport = inet->inet_sport;
77 	orig_dport = usin->sin_port;
78 	fl4 = &inet->cork.fl.u.ip4;
79 	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
80 			      sk->sk_bound_dev_if, IPPROTO_DCCP, orig_sport,
81 			      orig_dport, sk);
82 	if (IS_ERR(rt))
83 		return PTR_ERR(rt);
84 
85 	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
86 		ip_rt_put(rt);
87 		return -ENETUNREACH;
88 	}
89 
90 	if (inet_opt == NULL || !inet_opt->opt.srr)
91 		daddr = fl4->daddr;
92 
93 	if (inet->inet_saddr == 0) {
94 		if (inet_csk(sk)->icsk_bind2_hash) {
95 			prev_addr_hashbucket =
96 				inet_bhashfn_portaddr(&dccp_hashinfo, sk,
97 						      sock_net(sk),
98 						      inet->inet_num);
99 			prev_sk_rcv_saddr = sk->sk_rcv_saddr;
100 		}
101 		inet->inet_saddr = fl4->saddr;
102 	}
103 
104 	sk_rcv_saddr_set(sk, inet->inet_saddr);
105 
106 	if (prev_addr_hashbucket) {
107 		err = inet_bhash2_update_saddr(prev_addr_hashbucket, sk);
108 		if (err) {
109 			inet->inet_saddr = 0;
110 			sk_rcv_saddr_set(sk, prev_sk_rcv_saddr);
111 			ip_rt_put(rt);
112 			return err;
113 		}
114 	}
115 
116 	inet->inet_dport = usin->sin_port;
117 	sk_daddr_set(sk, daddr);
118 
119 	inet_csk(sk)->icsk_ext_hdr_len = 0;
120 	if (inet_opt)
121 		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
122 	/*
123 	 * Socket identity is still unknown (sport may be zero).
124 	 * However we set state to DCCP_REQUESTING and not releasing socket
125 	 * lock select source port, enter ourselves into the hash tables and
126 	 * complete initialization after this.
127 	 */
128 	dccp_set_state(sk, DCCP_REQUESTING);
129 	err = inet_hash_connect(&dccp_death_row, sk);
130 	if (err != 0)
131 		goto failure;
132 
133 	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
134 			       inet->inet_sport, inet->inet_dport, sk);
135 	if (IS_ERR(rt)) {
136 		err = PTR_ERR(rt);
137 		rt = NULL;
138 		goto failure;
139 	}
140 	/* OK, now commit destination to socket.  */
141 	sk_setup_caps(sk, &rt->dst);
142 
143 	dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
144 						    inet->inet_daddr,
145 						    inet->inet_sport,
146 						    inet->inet_dport);
147 	inet->inet_id = get_random_u16();
148 
149 	err = dccp_connect(sk);
150 	rt = NULL;
151 	if (err != 0)
152 		goto failure;
153 out:
154 	return err;
155 failure:
156 	/*
157 	 * This unhashes the socket and releases the local port, if necessary.
158 	 */
159 	dccp_set_state(sk, DCCP_CLOSED);
160 	ip_rt_put(rt);
161 	sk->sk_route_caps = 0;
162 	inet->inet_dport = 0;
163 	goto out;
164 }
165 EXPORT_SYMBOL_GPL(dccp_v4_connect);
166 
167 /*
168  * This routine does path mtu discovery as defined in RFC1191.
169  */
170 static inline void dccp_do_pmtu_discovery(struct sock *sk,
171 					  const struct iphdr *iph,
172 					  u32 mtu)
173 {
174 	struct dst_entry *dst;
175 	const struct inet_sock *inet = inet_sk(sk);
176 	const struct dccp_sock *dp = dccp_sk(sk);
177 
178 	/* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
179 	 * send out by Linux are always < 576bytes so they should go through
180 	 * unfragmented).
181 	 */
182 	if (sk->sk_state == DCCP_LISTEN)
183 		return;
184 
185 	dst = inet_csk_update_pmtu(sk, mtu);
186 	if (!dst)
187 		return;
188 
189 	/* Something is about to be wrong... Remember soft error
190 	 * for the case, if this connection will not able to recover.
191 	 */
192 	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
193 		sk->sk_err_soft = EMSGSIZE;
194 
195 	mtu = dst_mtu(dst);
196 
197 	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
198 	    ip_sk_accept_pmtu(sk) &&
199 	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
200 		dccp_sync_mss(sk, mtu);
201 
202 		/*
203 		 * From RFC 4340, sec. 14.1:
204 		 *
205 		 *	DCCP-Sync packets are the best choice for upward
206 		 *	probing, since DCCP-Sync probes do not risk application
207 		 *	data loss.
208 		 */
209 		dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
210 	} /* else let the usual retransmit timer handle it */
211 }
212 
213 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
214 {
215 	struct dst_entry *dst = __sk_dst_check(sk, 0);
216 
217 	if (dst)
218 		dst->ops->redirect(dst, sk, skb);
219 }
220 
221 void dccp_req_err(struct sock *sk, u64 seq)
222 	{
223 	struct request_sock *req = inet_reqsk(sk);
224 	struct net *net = sock_net(sk);
225 
226 	/*
227 	 * ICMPs are not backlogged, hence we cannot get an established
228 	 * socket here.
229 	 */
230 	if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
231 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
232 	} else {
233 		/*
234 		 * Still in RESPOND, just remove it silently.
235 		 * There is no good way to pass the error to the newly
236 		 * created socket, and POSIX does not want network
237 		 * errors returned from accept().
238 		 */
239 		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
240 	}
241 	reqsk_put(req);
242 }
243 EXPORT_SYMBOL(dccp_req_err);
244 
245 /*
246  * This routine is called by the ICMP module when it gets some sort of error
247  * condition. If err < 0 then the socket should be closed and the error
248  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
249  * After adjustment header points to the first 8 bytes of the tcp header. We
250  * need to find the appropriate port.
251  *
252  * The locking strategy used here is very "optimistic". When someone else
253  * accesses the socket the ICMP is just dropped and for some paths there is no
254  * check at all. A more general error queue to queue errors for later handling
255  * is probably better.
256  */
257 static int dccp_v4_err(struct sk_buff *skb, u32 info)
258 {
259 	const struct iphdr *iph = (struct iphdr *)skb->data;
260 	const u8 offset = iph->ihl << 2;
261 	const struct dccp_hdr *dh;
262 	struct dccp_sock *dp;
263 	struct inet_sock *inet;
264 	const int type = icmp_hdr(skb)->type;
265 	const int code = icmp_hdr(skb)->code;
266 	struct sock *sk;
267 	__u64 seq;
268 	int err;
269 	struct net *net = dev_net(skb->dev);
270 
271 	/* Only need dccph_dport & dccph_sport which are the first
272 	 * 4 bytes in dccp header.
273 	 * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
274 	 */
275 	BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
276 	BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
277 	dh = (struct dccp_hdr *)(skb->data + offset);
278 
279 	sk = __inet_lookup_established(net, &dccp_hashinfo,
280 				       iph->daddr, dh->dccph_dport,
281 				       iph->saddr, ntohs(dh->dccph_sport),
282 				       inet_iif(skb), 0);
283 	if (!sk) {
284 		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
285 		return -ENOENT;
286 	}
287 
288 	if (sk->sk_state == DCCP_TIME_WAIT) {
289 		inet_twsk_put(inet_twsk(sk));
290 		return 0;
291 	}
292 	seq = dccp_hdr_seq(dh);
293 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
294 		dccp_req_err(sk, seq);
295 		return 0;
296 	}
297 
298 	bh_lock_sock(sk);
299 	/* If too many ICMPs get dropped on busy
300 	 * servers this needs to be solved differently.
301 	 */
302 	if (sock_owned_by_user(sk))
303 		__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
304 
305 	if (sk->sk_state == DCCP_CLOSED)
306 		goto out;
307 
308 	dp = dccp_sk(sk);
309 	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
310 	    !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
311 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
312 		goto out;
313 	}
314 
315 	switch (type) {
316 	case ICMP_REDIRECT:
317 		if (!sock_owned_by_user(sk))
318 			dccp_do_redirect(skb, sk);
319 		goto out;
320 	case ICMP_SOURCE_QUENCH:
321 		/* Just silently ignore these. */
322 		goto out;
323 	case ICMP_PARAMETERPROB:
324 		err = EPROTO;
325 		break;
326 	case ICMP_DEST_UNREACH:
327 		if (code > NR_ICMP_UNREACH)
328 			goto out;
329 
330 		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
331 			if (!sock_owned_by_user(sk))
332 				dccp_do_pmtu_discovery(sk, iph, info);
333 			goto out;
334 		}
335 
336 		err = icmp_err_convert[code].errno;
337 		break;
338 	case ICMP_TIME_EXCEEDED:
339 		err = EHOSTUNREACH;
340 		break;
341 	default:
342 		goto out;
343 	}
344 
345 	switch (sk->sk_state) {
346 	case DCCP_REQUESTING:
347 	case DCCP_RESPOND:
348 		if (!sock_owned_by_user(sk)) {
349 			__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
350 			sk->sk_err = err;
351 
352 			sk_error_report(sk);
353 
354 			dccp_done(sk);
355 		} else
356 			sk->sk_err_soft = err;
357 		goto out;
358 	}
359 
360 	/* If we've already connected we will keep trying
361 	 * until we time out, or the user gives up.
362 	 *
363 	 * rfc1122 4.2.3.9 allows to consider as hard errors
364 	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
365 	 * but it is obsoleted by pmtu discovery).
366 	 *
367 	 * Note, that in modern internet, where routing is unreliable
368 	 * and in each dark corner broken firewalls sit, sending random
369 	 * errors ordered by their masters even this two messages finally lose
370 	 * their original sense (even Linux sends invalid PORT_UNREACHs)
371 	 *
372 	 * Now we are in compliance with RFCs.
373 	 *							--ANK (980905)
374 	 */
375 
376 	inet = inet_sk(sk);
377 	if (!sock_owned_by_user(sk) && inet->recverr) {
378 		sk->sk_err = err;
379 		sk_error_report(sk);
380 	} else /* Only an error on timeout */
381 		sk->sk_err_soft = err;
382 out:
383 	bh_unlock_sock(sk);
384 	sock_put(sk);
385 	return 0;
386 }
387 
388 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
389 				      __be32 src, __be32 dst)
390 {
391 	return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
392 }
393 
394 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
395 {
396 	const struct inet_sock *inet = inet_sk(sk);
397 	struct dccp_hdr *dh = dccp_hdr(skb);
398 
399 	dccp_csum_outgoing(skb);
400 	dh->dccph_checksum = dccp_v4_csum_finish(skb,
401 						 inet->inet_saddr,
402 						 inet->inet_daddr);
403 }
404 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
405 
406 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
407 {
408 	return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
409 					   ip_hdr(skb)->saddr,
410 					   dccp_hdr(skb)->dccph_dport,
411 					   dccp_hdr(skb)->dccph_sport);
412 }
413 
414 /*
415  * The three way handshake has completed - we got a valid ACK or DATAACK -
416  * now create the new socket.
417  *
418  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
419  */
420 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
421 				       struct sk_buff *skb,
422 				       struct request_sock *req,
423 				       struct dst_entry *dst,
424 				       struct request_sock *req_unhash,
425 				       bool *own_req)
426 {
427 	struct inet_request_sock *ireq;
428 	struct inet_sock *newinet;
429 	struct sock *newsk;
430 
431 	if (sk_acceptq_is_full(sk))
432 		goto exit_overflow;
433 
434 	newsk = dccp_create_openreq_child(sk, req, skb);
435 	if (newsk == NULL)
436 		goto exit_nonewsk;
437 
438 	newinet		   = inet_sk(newsk);
439 	ireq		   = inet_rsk(req);
440 	sk_daddr_set(newsk, ireq->ir_rmt_addr);
441 	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
442 	newinet->inet_saddr	= ireq->ir_loc_addr;
443 	RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
444 	newinet->mc_index  = inet_iif(skb);
445 	newinet->mc_ttl	   = ip_hdr(skb)->ttl;
446 	newinet->inet_id   = get_random_u16();
447 
448 	if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
449 		goto put_and_exit;
450 
451 	sk_setup_caps(newsk, dst);
452 
453 	dccp_sync_mss(newsk, dst_mtu(dst));
454 
455 	if (__inet_inherit_port(sk, newsk) < 0)
456 		goto put_and_exit;
457 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
458 	if (*own_req)
459 		ireq->ireq_opt = NULL;
460 	else
461 		newinet->inet_opt = NULL;
462 	return newsk;
463 
464 exit_overflow:
465 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
466 exit_nonewsk:
467 	dst_release(dst);
468 exit:
469 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
470 	return NULL;
471 put_and_exit:
472 	newinet->inet_opt = NULL;
473 	inet_csk_prepare_forced_close(newsk);
474 	dccp_done(newsk);
475 	goto exit;
476 }
477 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
478 
479 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
480 					   struct sk_buff *skb)
481 {
482 	struct rtable *rt;
483 	const struct iphdr *iph = ip_hdr(skb);
484 	struct flowi4 fl4 = {
485 		.flowi4_oif = inet_iif(skb),
486 		.daddr = iph->saddr,
487 		.saddr = iph->daddr,
488 		.flowi4_tos = RT_CONN_FLAGS(sk),
489 		.flowi4_proto = sk->sk_protocol,
490 		.fl4_sport = dccp_hdr(skb)->dccph_dport,
491 		.fl4_dport = dccp_hdr(skb)->dccph_sport,
492 	};
493 
494 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
495 	rt = ip_route_output_flow(net, &fl4, sk);
496 	if (IS_ERR(rt)) {
497 		IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
498 		return NULL;
499 	}
500 
501 	return &rt->dst;
502 }
503 
504 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
505 {
506 	int err = -1;
507 	struct sk_buff *skb;
508 	struct dst_entry *dst;
509 	struct flowi4 fl4;
510 
511 	dst = inet_csk_route_req(sk, &fl4, req);
512 	if (dst == NULL)
513 		goto out;
514 
515 	skb = dccp_make_response(sk, dst, req);
516 	if (skb != NULL) {
517 		const struct inet_request_sock *ireq = inet_rsk(req);
518 		struct dccp_hdr *dh = dccp_hdr(skb);
519 
520 		dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
521 							      ireq->ir_rmt_addr);
522 		rcu_read_lock();
523 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
524 					    ireq->ir_rmt_addr,
525 					    rcu_dereference(ireq->ireq_opt),
526 					    inet_sk(sk)->tos);
527 		rcu_read_unlock();
528 		err = net_xmit_eval(err);
529 	}
530 
531 out:
532 	dst_release(dst);
533 	return err;
534 }
535 
536 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
537 {
538 	int err;
539 	const struct iphdr *rxiph;
540 	struct sk_buff *skb;
541 	struct dst_entry *dst;
542 	struct net *net = dev_net(skb_dst(rxskb)->dev);
543 	struct dccp_v4_pernet *pn;
544 	struct sock *ctl_sk;
545 
546 	/* Never send a reset in response to a reset. */
547 	if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
548 		return;
549 
550 	if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
551 		return;
552 
553 	pn = net_generic(net, dccp_v4_pernet_id);
554 	ctl_sk = pn->v4_ctl_sk;
555 	dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
556 	if (dst == NULL)
557 		return;
558 
559 	skb = dccp_ctl_make_reset(ctl_sk, rxskb);
560 	if (skb == NULL)
561 		goto out;
562 
563 	rxiph = ip_hdr(rxskb);
564 	dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
565 								 rxiph->daddr);
566 	skb_dst_set(skb, dst_clone(dst));
567 
568 	local_bh_disable();
569 	bh_lock_sock(ctl_sk);
570 	err = ip_build_and_send_pkt(skb, ctl_sk,
571 				    rxiph->daddr, rxiph->saddr, NULL,
572 				    inet_sk(ctl_sk)->tos);
573 	bh_unlock_sock(ctl_sk);
574 
575 	if (net_xmit_eval(err) == 0) {
576 		__DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
577 		__DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
578 	}
579 	local_bh_enable();
580 out:
581 	dst_release(dst);
582 }
583 
584 static void dccp_v4_reqsk_destructor(struct request_sock *req)
585 {
586 	dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
587 	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
588 }
589 
590 void dccp_syn_ack_timeout(const struct request_sock *req)
591 {
592 }
593 EXPORT_SYMBOL(dccp_syn_ack_timeout);
594 
595 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
596 	.family		= PF_INET,
597 	.obj_size	= sizeof(struct dccp_request_sock),
598 	.rtx_syn_ack	= dccp_v4_send_response,
599 	.send_ack	= dccp_reqsk_send_ack,
600 	.destructor	= dccp_v4_reqsk_destructor,
601 	.send_reset	= dccp_v4_ctl_send_reset,
602 	.syn_ack_timeout = dccp_syn_ack_timeout,
603 };
604 
605 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
606 {
607 	struct inet_request_sock *ireq;
608 	struct request_sock *req;
609 	struct dccp_request_sock *dreq;
610 	const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
611 	struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
612 
613 	/* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
614 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
615 		return 0;	/* discard, don't send a reset here */
616 
617 	if (dccp_bad_service_code(sk, service)) {
618 		dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
619 		goto drop;
620 	}
621 	/*
622 	 * TW buckets are converted to open requests without
623 	 * limitations, they conserve resources and peer is
624 	 * evidently real one.
625 	 */
626 	dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
627 	if (inet_csk_reqsk_queue_is_full(sk))
628 		goto drop;
629 
630 	if (sk_acceptq_is_full(sk))
631 		goto drop;
632 
633 	req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
634 	if (req == NULL)
635 		goto drop;
636 
637 	if (dccp_reqsk_init(req, dccp_sk(sk), skb))
638 		goto drop_and_free;
639 
640 	dreq = dccp_rsk(req);
641 	if (dccp_parse_options(sk, dreq, skb))
642 		goto drop_and_free;
643 
644 	if (security_inet_conn_request(sk, skb, req))
645 		goto drop_and_free;
646 
647 	ireq = inet_rsk(req);
648 	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
649 	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
650 	ireq->ir_mark = inet_request_mark(sk, skb);
651 	ireq->ireq_family = AF_INET;
652 	ireq->ir_iif = READ_ONCE(sk->sk_bound_dev_if);
653 
654 	/*
655 	 * Step 3: Process LISTEN state
656 	 *
657 	 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
658 	 *
659 	 * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
660 	 */
661 	dreq->dreq_isr	   = dcb->dccpd_seq;
662 	dreq->dreq_gsr	   = dreq->dreq_isr;
663 	dreq->dreq_iss	   = dccp_v4_init_sequence(skb);
664 	dreq->dreq_gss     = dreq->dreq_iss;
665 	dreq->dreq_service = service;
666 
667 	if (dccp_v4_send_response(sk, req))
668 		goto drop_and_free;
669 
670 	inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
671 	reqsk_put(req);
672 	return 0;
673 
674 drop_and_free:
675 	reqsk_free(req);
676 drop:
677 	__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
678 	return -1;
679 }
680 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
681 
682 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
683 {
684 	struct dccp_hdr *dh = dccp_hdr(skb);
685 
686 	if (sk->sk_state == DCCP_OPEN) { /* Fast path */
687 		if (dccp_rcv_established(sk, skb, dh, skb->len))
688 			goto reset;
689 		return 0;
690 	}
691 
692 	/*
693 	 *  Step 3: Process LISTEN state
694 	 *	 If P.type == Request or P contains a valid Init Cookie option,
695 	 *	      (* Must scan the packet's options to check for Init
696 	 *		 Cookies.  Only Init Cookies are processed here,
697 	 *		 however; other options are processed in Step 8.  This
698 	 *		 scan need only be performed if the endpoint uses Init
699 	 *		 Cookies *)
700 	 *	      (* Generate a new socket and switch to that socket *)
701 	 *	      Set S := new socket for this port pair
702 	 *	      S.state = RESPOND
703 	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
704 	 *	      Initialize S.GAR := S.ISS
705 	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
706 	 *	      Continue with S.state == RESPOND
707 	 *	      (* A Response packet will be generated in Step 11 *)
708 	 *	 Otherwise,
709 	 *	      Generate Reset(No Connection) unless P.type == Reset
710 	 *	      Drop packet and return
711 	 *
712 	 * NOTE: the check for the packet types is done in
713 	 *	 dccp_rcv_state_process
714 	 */
715 
716 	if (dccp_rcv_state_process(sk, skb, dh, skb->len))
717 		goto reset;
718 	return 0;
719 
720 reset:
721 	dccp_v4_ctl_send_reset(sk, skb);
722 	kfree_skb(skb);
723 	return 0;
724 }
725 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
726 
727 /**
728  *	dccp_invalid_packet  -  check for malformed packets
729  *	@skb: Packet to validate
730  *
731  *	Implements RFC 4340, 8.5:  Step 1: Check header basics
732  *	Packets that fail these checks are ignored and do not receive Resets.
733  */
734 int dccp_invalid_packet(struct sk_buff *skb)
735 {
736 	const struct dccp_hdr *dh;
737 	unsigned int cscov;
738 	u8 dccph_doff;
739 
740 	if (skb->pkt_type != PACKET_HOST)
741 		return 1;
742 
743 	/* If the packet is shorter than 12 bytes, drop packet and return */
744 	if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
745 		DCCP_WARN("pskb_may_pull failed\n");
746 		return 1;
747 	}
748 
749 	dh = dccp_hdr(skb);
750 
751 	/* If P.type is not understood, drop packet and return */
752 	if (dh->dccph_type >= DCCP_PKT_INVALID) {
753 		DCCP_WARN("invalid packet type\n");
754 		return 1;
755 	}
756 
757 	/*
758 	 * If P.Data Offset is too small for packet type, drop packet and return
759 	 */
760 	dccph_doff = dh->dccph_doff;
761 	if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
762 		DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
763 		return 1;
764 	}
765 	/*
766 	 * If P.Data Offset is too large for packet, drop packet and return
767 	 */
768 	if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
769 		DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
770 		return 1;
771 	}
772 	dh = dccp_hdr(skb);
773 	/*
774 	 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
775 	 * has short sequence numbers), drop packet and return
776 	 */
777 	if ((dh->dccph_type < DCCP_PKT_DATA    ||
778 	    dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
779 		DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
780 			  dccp_packet_name(dh->dccph_type));
781 		return 1;
782 	}
783 
784 	/*
785 	 * If P.CsCov is too large for the packet size, drop packet and return.
786 	 * This must come _before_ checksumming (not as RFC 4340 suggests).
787 	 */
788 	cscov = dccp_csum_coverage(skb);
789 	if (cscov > skb->len) {
790 		DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
791 			  dh->dccph_cscov, skb->len);
792 		return 1;
793 	}
794 
795 	/* If header checksum is incorrect, drop packet and return.
796 	 * (This step is completed in the AF-dependent functions.) */
797 	skb->csum = skb_checksum(skb, 0, cscov, 0);
798 
799 	return 0;
800 }
801 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
802 
803 /* this is called when real data arrives */
804 static int dccp_v4_rcv(struct sk_buff *skb)
805 {
806 	const struct dccp_hdr *dh;
807 	const struct iphdr *iph;
808 	bool refcounted;
809 	struct sock *sk;
810 	int min_cov;
811 
812 	/* Step 1: Check header basics */
813 
814 	if (dccp_invalid_packet(skb))
815 		goto discard_it;
816 
817 	iph = ip_hdr(skb);
818 	/* Step 1: If header checksum is incorrect, drop packet and return */
819 	if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
820 		DCCP_WARN("dropped packet with invalid checksum\n");
821 		goto discard_it;
822 	}
823 
824 	dh = dccp_hdr(skb);
825 
826 	DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
827 	DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
828 
829 	dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
830 		      dccp_packet_name(dh->dccph_type),
831 		      &iph->saddr, ntohs(dh->dccph_sport),
832 		      &iph->daddr, ntohs(dh->dccph_dport),
833 		      (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
834 
835 	if (dccp_packet_without_ack(skb)) {
836 		DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
837 		dccp_pr_debug_cat("\n");
838 	} else {
839 		DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
840 		dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
841 				  DCCP_SKB_CB(skb)->dccpd_ack_seq);
842 	}
843 
844 lookup:
845 	sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
846 			       dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
847 	if (!sk) {
848 		dccp_pr_debug("failed to look up flow ID in table and "
849 			      "get corresponding socket\n");
850 		goto no_dccp_socket;
851 	}
852 
853 	/*
854 	 * Step 2:
855 	 *	... or S.state == TIMEWAIT,
856 	 *		Generate Reset(No Connection) unless P.type == Reset
857 	 *		Drop packet and return
858 	 */
859 	if (sk->sk_state == DCCP_TIME_WAIT) {
860 		dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
861 		inet_twsk_put(inet_twsk(sk));
862 		goto no_dccp_socket;
863 	}
864 
865 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
866 		struct request_sock *req = inet_reqsk(sk);
867 		struct sock *nsk;
868 
869 		sk = req->rsk_listener;
870 		if (unlikely(sk->sk_state != DCCP_LISTEN)) {
871 			inet_csk_reqsk_queue_drop_and_put(sk, req);
872 			goto lookup;
873 		}
874 		sock_hold(sk);
875 		refcounted = true;
876 		nsk = dccp_check_req(sk, skb, req);
877 		if (!nsk) {
878 			reqsk_put(req);
879 			goto discard_and_relse;
880 		}
881 		if (nsk == sk) {
882 			reqsk_put(req);
883 		} else if (dccp_child_process(sk, nsk, skb)) {
884 			dccp_v4_ctl_send_reset(sk, skb);
885 			goto discard_and_relse;
886 		} else {
887 			sock_put(sk);
888 			return 0;
889 		}
890 	}
891 	/*
892 	 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
893 	 *	o if MinCsCov = 0, only packets with CsCov = 0 are accepted
894 	 *	o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
895 	 */
896 	min_cov = dccp_sk(sk)->dccps_pcrlen;
897 	if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
898 		dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
899 			      dh->dccph_cscov, min_cov);
900 		/* FIXME: "Such packets SHOULD be reported using Data Dropped
901 		 *         options (Section 11.7) with Drop Code 0, Protocol
902 		 *         Constraints."                                     */
903 		goto discard_and_relse;
904 	}
905 
906 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
907 		goto discard_and_relse;
908 	nf_reset_ct(skb);
909 
910 	return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
911 
912 no_dccp_socket:
913 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
914 		goto discard_it;
915 	/*
916 	 * Step 2:
917 	 *	If no socket ...
918 	 *		Generate Reset(No Connection) unless P.type == Reset
919 	 *		Drop packet and return
920 	 */
921 	if (dh->dccph_type != DCCP_PKT_RESET) {
922 		DCCP_SKB_CB(skb)->dccpd_reset_code =
923 					DCCP_RESET_CODE_NO_CONNECTION;
924 		dccp_v4_ctl_send_reset(sk, skb);
925 	}
926 
927 discard_it:
928 	kfree_skb(skb);
929 	return 0;
930 
931 discard_and_relse:
932 	if (refcounted)
933 		sock_put(sk);
934 	goto discard_it;
935 }
936 
937 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
938 	.queue_xmit	   = ip_queue_xmit,
939 	.send_check	   = dccp_v4_send_check,
940 	.rebuild_header	   = inet_sk_rebuild_header,
941 	.conn_request	   = dccp_v4_conn_request,
942 	.syn_recv_sock	   = dccp_v4_request_recv_sock,
943 	.net_header_len	   = sizeof(struct iphdr),
944 	.setsockopt	   = ip_setsockopt,
945 	.getsockopt	   = ip_getsockopt,
946 	.addr2sockaddr	   = inet_csk_addr2sockaddr,
947 	.sockaddr_len	   = sizeof(struct sockaddr_in),
948 };
949 
950 static int dccp_v4_init_sock(struct sock *sk)
951 {
952 	static __u8 dccp_v4_ctl_sock_initialized;
953 	int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
954 
955 	if (err == 0) {
956 		if (unlikely(!dccp_v4_ctl_sock_initialized))
957 			dccp_v4_ctl_sock_initialized = 1;
958 		inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
959 	}
960 
961 	return err;
962 }
963 
964 static struct timewait_sock_ops dccp_timewait_sock_ops = {
965 	.twsk_obj_size	= sizeof(struct inet_timewait_sock),
966 };
967 
968 static struct proto dccp_v4_prot = {
969 	.name			= "DCCP",
970 	.owner			= THIS_MODULE,
971 	.close			= dccp_close,
972 	.connect		= dccp_v4_connect,
973 	.disconnect		= dccp_disconnect,
974 	.ioctl			= dccp_ioctl,
975 	.init			= dccp_v4_init_sock,
976 	.setsockopt		= dccp_setsockopt,
977 	.getsockopt		= dccp_getsockopt,
978 	.sendmsg		= dccp_sendmsg,
979 	.recvmsg		= dccp_recvmsg,
980 	.backlog_rcv		= dccp_v4_do_rcv,
981 	.hash			= inet_hash,
982 	.unhash			= inet_unhash,
983 	.accept			= inet_csk_accept,
984 	.get_port		= inet_csk_get_port,
985 	.shutdown		= dccp_shutdown,
986 	.destroy		= dccp_destroy_sock,
987 	.orphan_count		= &dccp_orphan_count,
988 	.max_header		= MAX_DCCP_HEADER,
989 	.obj_size		= sizeof(struct dccp_sock),
990 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
991 	.rsk_prot		= &dccp_request_sock_ops,
992 	.twsk_prot		= &dccp_timewait_sock_ops,
993 	.h.hashinfo		= &dccp_hashinfo,
994 };
995 
996 static const struct net_protocol dccp_v4_protocol = {
997 	.handler	= dccp_v4_rcv,
998 	.err_handler	= dccp_v4_err,
999 	.no_policy	= 1,
1000 	.icmp_strict_tag_validation = 1,
1001 };
1002 
1003 static const struct proto_ops inet_dccp_ops = {
1004 	.family		   = PF_INET,
1005 	.owner		   = THIS_MODULE,
1006 	.release	   = inet_release,
1007 	.bind		   = inet_bind,
1008 	.connect	   = inet_stream_connect,
1009 	.socketpair	   = sock_no_socketpair,
1010 	.accept		   = inet_accept,
1011 	.getname	   = inet_getname,
1012 	/* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1013 	.poll		   = dccp_poll,
1014 	.ioctl		   = inet_ioctl,
1015 	.gettstamp	   = sock_gettstamp,
1016 	/* FIXME: work on inet_listen to rename it to sock_common_listen */
1017 	.listen		   = inet_dccp_listen,
1018 	.shutdown	   = inet_shutdown,
1019 	.setsockopt	   = sock_common_setsockopt,
1020 	.getsockopt	   = sock_common_getsockopt,
1021 	.sendmsg	   = inet_sendmsg,
1022 	.recvmsg	   = sock_common_recvmsg,
1023 	.mmap		   = sock_no_mmap,
1024 	.sendpage	   = sock_no_sendpage,
1025 };
1026 
1027 static struct inet_protosw dccp_v4_protosw = {
1028 	.type		= SOCK_DCCP,
1029 	.protocol	= IPPROTO_DCCP,
1030 	.prot		= &dccp_v4_prot,
1031 	.ops		= &inet_dccp_ops,
1032 	.flags		= INET_PROTOSW_ICSK,
1033 };
1034 
1035 static int __net_init dccp_v4_init_net(struct net *net)
1036 {
1037 	struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1038 
1039 	if (dccp_hashinfo.bhash == NULL)
1040 		return -ESOCKTNOSUPPORT;
1041 
1042 	return inet_ctl_sock_create(&pn->v4_ctl_sk, PF_INET,
1043 				    SOCK_DCCP, IPPROTO_DCCP, net);
1044 }
1045 
1046 static void __net_exit dccp_v4_exit_net(struct net *net)
1047 {
1048 	struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1049 
1050 	inet_ctl_sock_destroy(pn->v4_ctl_sk);
1051 }
1052 
1053 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1054 {
1055 	inet_twsk_purge(&dccp_hashinfo, AF_INET);
1056 }
1057 
1058 static struct pernet_operations dccp_v4_ops = {
1059 	.init	= dccp_v4_init_net,
1060 	.exit	= dccp_v4_exit_net,
1061 	.exit_batch = dccp_v4_exit_batch,
1062 	.id	= &dccp_v4_pernet_id,
1063 	.size   = sizeof(struct dccp_v4_pernet),
1064 };
1065 
1066 static int __init dccp_v4_init(void)
1067 {
1068 	int err = proto_register(&dccp_v4_prot, 1);
1069 
1070 	if (err)
1071 		goto out;
1072 
1073 	inet_register_protosw(&dccp_v4_protosw);
1074 
1075 	err = register_pernet_subsys(&dccp_v4_ops);
1076 	if (err)
1077 		goto out_destroy_ctl_sock;
1078 
1079 	err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1080 	if (err)
1081 		goto out_proto_unregister;
1082 
1083 out:
1084 	return err;
1085 out_proto_unregister:
1086 	unregister_pernet_subsys(&dccp_v4_ops);
1087 out_destroy_ctl_sock:
1088 	inet_unregister_protosw(&dccp_v4_protosw);
1089 	proto_unregister(&dccp_v4_prot);
1090 	goto out;
1091 }
1092 
1093 static void __exit dccp_v4_exit(void)
1094 {
1095 	inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1096 	unregister_pernet_subsys(&dccp_v4_ops);
1097 	inet_unregister_protosw(&dccp_v4_protosw);
1098 	proto_unregister(&dccp_v4_prot);
1099 }
1100 
1101 module_init(dccp_v4_init);
1102 module_exit(dccp_v4_exit);
1103 
1104 /*
1105  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1106  * values directly, Also cover the case where the protocol is not specified,
1107  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1108  */
1109 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1110 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1111 MODULE_LICENSE("GPL");
1112 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1113 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1114