xref: /linux/net/dccp/ipv4.c (revision 3c4fc7bf4c9e66fe71abcbf93f62f4ddb89b7f15)
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_sock *inet = inet_sk(sk);
49 	struct dccp_sock *dp = dccp_sk(sk);
50 	__be16 orig_sport, orig_dport;
51 	__be32 daddr, nexthop;
52 	struct flowi4 *fl4;
53 	struct rtable *rt;
54 	int err;
55 	struct ip_options_rcu *inet_opt;
56 
57 	dp->dccps_role = DCCP_ROLE_CLIENT;
58 
59 	if (addr_len < sizeof(struct sockaddr_in))
60 		return -EINVAL;
61 
62 	if (usin->sin_family != AF_INET)
63 		return -EAFNOSUPPORT;
64 
65 	nexthop = daddr = usin->sin_addr.s_addr;
66 
67 	inet_opt = rcu_dereference_protected(inet->inet_opt,
68 					     lockdep_sock_is_held(sk));
69 	if (inet_opt != NULL && inet_opt->opt.srr) {
70 		if (daddr == 0)
71 			return -EINVAL;
72 		nexthop = inet_opt->opt.faddr;
73 	}
74 
75 	orig_sport = inet->inet_sport;
76 	orig_dport = usin->sin_port;
77 	fl4 = &inet->cork.fl.u.ip4;
78 	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
79 			      sk->sk_bound_dev_if, IPPROTO_DCCP, orig_sport,
80 			      orig_dport, sk);
81 	if (IS_ERR(rt))
82 		return PTR_ERR(rt);
83 
84 	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
85 		ip_rt_put(rt);
86 		return -ENETUNREACH;
87 	}
88 
89 	if (inet_opt == NULL || !inet_opt->opt.srr)
90 		daddr = fl4->daddr;
91 
92 	if (inet->inet_saddr == 0)
93 		inet->inet_saddr = fl4->saddr;
94 	sk_rcv_saddr_set(sk, inet->inet_saddr);
95 	inet->inet_dport = usin->sin_port;
96 	sk_daddr_set(sk, daddr);
97 
98 	inet_csk(sk)->icsk_ext_hdr_len = 0;
99 	if (inet_opt)
100 		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
101 	/*
102 	 * Socket identity is still unknown (sport may be zero).
103 	 * However we set state to DCCP_REQUESTING and not releasing socket
104 	 * lock select source port, enter ourselves into the hash tables and
105 	 * complete initialization after this.
106 	 */
107 	dccp_set_state(sk, DCCP_REQUESTING);
108 	err = inet_hash_connect(&dccp_death_row, sk);
109 	if (err != 0)
110 		goto failure;
111 
112 	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
113 			       inet->inet_sport, inet->inet_dport, sk);
114 	if (IS_ERR(rt)) {
115 		err = PTR_ERR(rt);
116 		rt = NULL;
117 		goto failure;
118 	}
119 	/* OK, now commit destination to socket.  */
120 	sk_setup_caps(sk, &rt->dst);
121 
122 	dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
123 						    inet->inet_daddr,
124 						    inet->inet_sport,
125 						    inet->inet_dport);
126 	inet->inet_id = prandom_u32();
127 
128 	err = dccp_connect(sk);
129 	rt = NULL;
130 	if (err != 0)
131 		goto failure;
132 out:
133 	return err;
134 failure:
135 	/*
136 	 * This unhashes the socket and releases the local port, if necessary.
137 	 */
138 	dccp_set_state(sk, DCCP_CLOSED);
139 	ip_rt_put(rt);
140 	sk->sk_route_caps = 0;
141 	inet->inet_dport = 0;
142 	goto out;
143 }
144 EXPORT_SYMBOL_GPL(dccp_v4_connect);
145 
146 /*
147  * This routine does path mtu discovery as defined in RFC1191.
148  */
149 static inline void dccp_do_pmtu_discovery(struct sock *sk,
150 					  const struct iphdr *iph,
151 					  u32 mtu)
152 {
153 	struct dst_entry *dst;
154 	const struct inet_sock *inet = inet_sk(sk);
155 	const struct dccp_sock *dp = dccp_sk(sk);
156 
157 	/* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
158 	 * send out by Linux are always < 576bytes so they should go through
159 	 * unfragmented).
160 	 */
161 	if (sk->sk_state == DCCP_LISTEN)
162 		return;
163 
164 	dst = inet_csk_update_pmtu(sk, mtu);
165 	if (!dst)
166 		return;
167 
168 	/* Something is about to be wrong... Remember soft error
169 	 * for the case, if this connection will not able to recover.
170 	 */
171 	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
172 		sk->sk_err_soft = EMSGSIZE;
173 
174 	mtu = dst_mtu(dst);
175 
176 	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
177 	    ip_sk_accept_pmtu(sk) &&
178 	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
179 		dccp_sync_mss(sk, mtu);
180 
181 		/*
182 		 * From RFC 4340, sec. 14.1:
183 		 *
184 		 *	DCCP-Sync packets are the best choice for upward
185 		 *	probing, since DCCP-Sync probes do not risk application
186 		 *	data loss.
187 		 */
188 		dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
189 	} /* else let the usual retransmit timer handle it */
190 }
191 
192 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
193 {
194 	struct dst_entry *dst = __sk_dst_check(sk, 0);
195 
196 	if (dst)
197 		dst->ops->redirect(dst, sk, skb);
198 }
199 
200 void dccp_req_err(struct sock *sk, u64 seq)
201 	{
202 	struct request_sock *req = inet_reqsk(sk);
203 	struct net *net = sock_net(sk);
204 
205 	/*
206 	 * ICMPs are not backlogged, hence we cannot get an established
207 	 * socket here.
208 	 */
209 	if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
210 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
211 	} else {
212 		/*
213 		 * Still in RESPOND, just remove it silently.
214 		 * There is no good way to pass the error to the newly
215 		 * created socket, and POSIX does not want network
216 		 * errors returned from accept().
217 		 */
218 		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
219 	}
220 	reqsk_put(req);
221 }
222 EXPORT_SYMBOL(dccp_req_err);
223 
224 /*
225  * This routine is called by the ICMP module when it gets some sort of error
226  * condition. If err < 0 then the socket should be closed and the error
227  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
228  * After adjustment header points to the first 8 bytes of the tcp header. We
229  * need to find the appropriate port.
230  *
231  * The locking strategy used here is very "optimistic". When someone else
232  * accesses the socket the ICMP is just dropped and for some paths there is no
233  * check at all. A more general error queue to queue errors for later handling
234  * is probably better.
235  */
236 static int dccp_v4_err(struct sk_buff *skb, u32 info)
237 {
238 	const struct iphdr *iph = (struct iphdr *)skb->data;
239 	const u8 offset = iph->ihl << 2;
240 	const struct dccp_hdr *dh;
241 	struct dccp_sock *dp;
242 	struct inet_sock *inet;
243 	const int type = icmp_hdr(skb)->type;
244 	const int code = icmp_hdr(skb)->code;
245 	struct sock *sk;
246 	__u64 seq;
247 	int err;
248 	struct net *net = dev_net(skb->dev);
249 
250 	/* Only need dccph_dport & dccph_sport which are the first
251 	 * 4 bytes in dccp header.
252 	 * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
253 	 */
254 	BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
255 	BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
256 	dh = (struct dccp_hdr *)(skb->data + offset);
257 
258 	sk = __inet_lookup_established(net, &dccp_hashinfo,
259 				       iph->daddr, dh->dccph_dport,
260 				       iph->saddr, ntohs(dh->dccph_sport),
261 				       inet_iif(skb), 0);
262 	if (!sk) {
263 		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
264 		return -ENOENT;
265 	}
266 
267 	if (sk->sk_state == DCCP_TIME_WAIT) {
268 		inet_twsk_put(inet_twsk(sk));
269 		return 0;
270 	}
271 	seq = dccp_hdr_seq(dh);
272 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
273 		dccp_req_err(sk, seq);
274 		return 0;
275 	}
276 
277 	bh_lock_sock(sk);
278 	/* If too many ICMPs get dropped on busy
279 	 * servers this needs to be solved differently.
280 	 */
281 	if (sock_owned_by_user(sk))
282 		__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
283 
284 	if (sk->sk_state == DCCP_CLOSED)
285 		goto out;
286 
287 	dp = dccp_sk(sk);
288 	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
289 	    !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
290 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
291 		goto out;
292 	}
293 
294 	switch (type) {
295 	case ICMP_REDIRECT:
296 		if (!sock_owned_by_user(sk))
297 			dccp_do_redirect(skb, sk);
298 		goto out;
299 	case ICMP_SOURCE_QUENCH:
300 		/* Just silently ignore these. */
301 		goto out;
302 	case ICMP_PARAMETERPROB:
303 		err = EPROTO;
304 		break;
305 	case ICMP_DEST_UNREACH:
306 		if (code > NR_ICMP_UNREACH)
307 			goto out;
308 
309 		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
310 			if (!sock_owned_by_user(sk))
311 				dccp_do_pmtu_discovery(sk, iph, info);
312 			goto out;
313 		}
314 
315 		err = icmp_err_convert[code].errno;
316 		break;
317 	case ICMP_TIME_EXCEEDED:
318 		err = EHOSTUNREACH;
319 		break;
320 	default:
321 		goto out;
322 	}
323 
324 	switch (sk->sk_state) {
325 	case DCCP_REQUESTING:
326 	case DCCP_RESPOND:
327 		if (!sock_owned_by_user(sk)) {
328 			__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
329 			sk->sk_err = err;
330 
331 			sk_error_report(sk);
332 
333 			dccp_done(sk);
334 		} else
335 			sk->sk_err_soft = err;
336 		goto out;
337 	}
338 
339 	/* If we've already connected we will keep trying
340 	 * until we time out, or the user gives up.
341 	 *
342 	 * rfc1122 4.2.3.9 allows to consider as hard errors
343 	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
344 	 * but it is obsoleted by pmtu discovery).
345 	 *
346 	 * Note, that in modern internet, where routing is unreliable
347 	 * and in each dark corner broken firewalls sit, sending random
348 	 * errors ordered by their masters even this two messages finally lose
349 	 * their original sense (even Linux sends invalid PORT_UNREACHs)
350 	 *
351 	 * Now we are in compliance with RFCs.
352 	 *							--ANK (980905)
353 	 */
354 
355 	inet = inet_sk(sk);
356 	if (!sock_owned_by_user(sk) && inet->recverr) {
357 		sk->sk_err = err;
358 		sk_error_report(sk);
359 	} else /* Only an error on timeout */
360 		sk->sk_err_soft = err;
361 out:
362 	bh_unlock_sock(sk);
363 	sock_put(sk);
364 	return 0;
365 }
366 
367 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
368 				      __be32 src, __be32 dst)
369 {
370 	return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
371 }
372 
373 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
374 {
375 	const struct inet_sock *inet = inet_sk(sk);
376 	struct dccp_hdr *dh = dccp_hdr(skb);
377 
378 	dccp_csum_outgoing(skb);
379 	dh->dccph_checksum = dccp_v4_csum_finish(skb,
380 						 inet->inet_saddr,
381 						 inet->inet_daddr);
382 }
383 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
384 
385 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
386 {
387 	return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
388 					   ip_hdr(skb)->saddr,
389 					   dccp_hdr(skb)->dccph_dport,
390 					   dccp_hdr(skb)->dccph_sport);
391 }
392 
393 /*
394  * The three way handshake has completed - we got a valid ACK or DATAACK -
395  * now create the new socket.
396  *
397  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
398  */
399 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
400 				       struct sk_buff *skb,
401 				       struct request_sock *req,
402 				       struct dst_entry *dst,
403 				       struct request_sock *req_unhash,
404 				       bool *own_req)
405 {
406 	struct inet_request_sock *ireq;
407 	struct inet_sock *newinet;
408 	struct sock *newsk;
409 
410 	if (sk_acceptq_is_full(sk))
411 		goto exit_overflow;
412 
413 	newsk = dccp_create_openreq_child(sk, req, skb);
414 	if (newsk == NULL)
415 		goto exit_nonewsk;
416 
417 	newinet		   = inet_sk(newsk);
418 	ireq		   = inet_rsk(req);
419 	sk_daddr_set(newsk, ireq->ir_rmt_addr);
420 	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
421 	newinet->inet_saddr	= ireq->ir_loc_addr;
422 	RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
423 	newinet->mc_index  = inet_iif(skb);
424 	newinet->mc_ttl	   = ip_hdr(skb)->ttl;
425 	newinet->inet_id   = prandom_u32();
426 
427 	if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
428 		goto put_and_exit;
429 
430 	sk_setup_caps(newsk, dst);
431 
432 	dccp_sync_mss(newsk, dst_mtu(dst));
433 
434 	if (__inet_inherit_port(sk, newsk) < 0)
435 		goto put_and_exit;
436 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
437 	if (*own_req)
438 		ireq->ireq_opt = NULL;
439 	else
440 		newinet->inet_opt = NULL;
441 	return newsk;
442 
443 exit_overflow:
444 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
445 exit_nonewsk:
446 	dst_release(dst);
447 exit:
448 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
449 	return NULL;
450 put_and_exit:
451 	newinet->inet_opt = NULL;
452 	inet_csk_prepare_forced_close(newsk);
453 	dccp_done(newsk);
454 	goto exit;
455 }
456 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
457 
458 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
459 					   struct sk_buff *skb)
460 {
461 	struct rtable *rt;
462 	const struct iphdr *iph = ip_hdr(skb);
463 	struct flowi4 fl4 = {
464 		.flowi4_oif = inet_iif(skb),
465 		.daddr = iph->saddr,
466 		.saddr = iph->daddr,
467 		.flowi4_tos = RT_CONN_FLAGS(sk),
468 		.flowi4_proto = sk->sk_protocol,
469 		.fl4_sport = dccp_hdr(skb)->dccph_dport,
470 		.fl4_dport = dccp_hdr(skb)->dccph_sport,
471 	};
472 
473 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
474 	rt = ip_route_output_flow(net, &fl4, sk);
475 	if (IS_ERR(rt)) {
476 		IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
477 		return NULL;
478 	}
479 
480 	return &rt->dst;
481 }
482 
483 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
484 {
485 	int err = -1;
486 	struct sk_buff *skb;
487 	struct dst_entry *dst;
488 	struct flowi4 fl4;
489 
490 	dst = inet_csk_route_req(sk, &fl4, req);
491 	if (dst == NULL)
492 		goto out;
493 
494 	skb = dccp_make_response(sk, dst, req);
495 	if (skb != NULL) {
496 		const struct inet_request_sock *ireq = inet_rsk(req);
497 		struct dccp_hdr *dh = dccp_hdr(skb);
498 
499 		dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
500 							      ireq->ir_rmt_addr);
501 		rcu_read_lock();
502 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
503 					    ireq->ir_rmt_addr,
504 					    rcu_dereference(ireq->ireq_opt),
505 					    inet_sk(sk)->tos);
506 		rcu_read_unlock();
507 		err = net_xmit_eval(err);
508 	}
509 
510 out:
511 	dst_release(dst);
512 	return err;
513 }
514 
515 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
516 {
517 	int err;
518 	const struct iphdr *rxiph;
519 	struct sk_buff *skb;
520 	struct dst_entry *dst;
521 	struct net *net = dev_net(skb_dst(rxskb)->dev);
522 	struct dccp_v4_pernet *pn;
523 	struct sock *ctl_sk;
524 
525 	/* Never send a reset in response to a reset. */
526 	if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
527 		return;
528 
529 	if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
530 		return;
531 
532 	pn = net_generic(net, dccp_v4_pernet_id);
533 	ctl_sk = pn->v4_ctl_sk;
534 	dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
535 	if (dst == NULL)
536 		return;
537 
538 	skb = dccp_ctl_make_reset(ctl_sk, rxskb);
539 	if (skb == NULL)
540 		goto out;
541 
542 	rxiph = ip_hdr(rxskb);
543 	dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
544 								 rxiph->daddr);
545 	skb_dst_set(skb, dst_clone(dst));
546 
547 	local_bh_disable();
548 	bh_lock_sock(ctl_sk);
549 	err = ip_build_and_send_pkt(skb, ctl_sk,
550 				    rxiph->daddr, rxiph->saddr, NULL,
551 				    inet_sk(ctl_sk)->tos);
552 	bh_unlock_sock(ctl_sk);
553 
554 	if (net_xmit_eval(err) == 0) {
555 		__DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
556 		__DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
557 	}
558 	local_bh_enable();
559 out:
560 	dst_release(dst);
561 }
562 
563 static void dccp_v4_reqsk_destructor(struct request_sock *req)
564 {
565 	dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
566 	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
567 }
568 
569 void dccp_syn_ack_timeout(const struct request_sock *req)
570 {
571 }
572 EXPORT_SYMBOL(dccp_syn_ack_timeout);
573 
574 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
575 	.family		= PF_INET,
576 	.obj_size	= sizeof(struct dccp_request_sock),
577 	.rtx_syn_ack	= dccp_v4_send_response,
578 	.send_ack	= dccp_reqsk_send_ack,
579 	.destructor	= dccp_v4_reqsk_destructor,
580 	.send_reset	= dccp_v4_ctl_send_reset,
581 	.syn_ack_timeout = dccp_syn_ack_timeout,
582 };
583 
584 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
585 {
586 	struct inet_request_sock *ireq;
587 	struct request_sock *req;
588 	struct dccp_request_sock *dreq;
589 	const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
590 	struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
591 
592 	/* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
593 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
594 		return 0;	/* discard, don't send a reset here */
595 
596 	if (dccp_bad_service_code(sk, service)) {
597 		dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
598 		goto drop;
599 	}
600 	/*
601 	 * TW buckets are converted to open requests without
602 	 * limitations, they conserve resources and peer is
603 	 * evidently real one.
604 	 */
605 	dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
606 	if (inet_csk_reqsk_queue_is_full(sk))
607 		goto drop;
608 
609 	if (sk_acceptq_is_full(sk))
610 		goto drop;
611 
612 	req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
613 	if (req == NULL)
614 		goto drop;
615 
616 	if (dccp_reqsk_init(req, dccp_sk(sk), skb))
617 		goto drop_and_free;
618 
619 	dreq = dccp_rsk(req);
620 	if (dccp_parse_options(sk, dreq, skb))
621 		goto drop_and_free;
622 
623 	if (security_inet_conn_request(sk, skb, req))
624 		goto drop_and_free;
625 
626 	ireq = inet_rsk(req);
627 	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
628 	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
629 	ireq->ir_mark = inet_request_mark(sk, skb);
630 	ireq->ireq_family = AF_INET;
631 	ireq->ir_iif = READ_ONCE(sk->sk_bound_dev_if);
632 
633 	/*
634 	 * Step 3: Process LISTEN state
635 	 *
636 	 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
637 	 *
638 	 * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
639 	 */
640 	dreq->dreq_isr	   = dcb->dccpd_seq;
641 	dreq->dreq_gsr	   = dreq->dreq_isr;
642 	dreq->dreq_iss	   = dccp_v4_init_sequence(skb);
643 	dreq->dreq_gss     = dreq->dreq_iss;
644 	dreq->dreq_service = service;
645 
646 	if (dccp_v4_send_response(sk, req))
647 		goto drop_and_free;
648 
649 	inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
650 	reqsk_put(req);
651 	return 0;
652 
653 drop_and_free:
654 	reqsk_free(req);
655 drop:
656 	__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
657 	return -1;
658 }
659 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
660 
661 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
662 {
663 	struct dccp_hdr *dh = dccp_hdr(skb);
664 
665 	if (sk->sk_state == DCCP_OPEN) { /* Fast path */
666 		if (dccp_rcv_established(sk, skb, dh, skb->len))
667 			goto reset;
668 		return 0;
669 	}
670 
671 	/*
672 	 *  Step 3: Process LISTEN state
673 	 *	 If P.type == Request or P contains a valid Init Cookie option,
674 	 *	      (* Must scan the packet's options to check for Init
675 	 *		 Cookies.  Only Init Cookies are processed here,
676 	 *		 however; other options are processed in Step 8.  This
677 	 *		 scan need only be performed if the endpoint uses Init
678 	 *		 Cookies *)
679 	 *	      (* Generate a new socket and switch to that socket *)
680 	 *	      Set S := new socket for this port pair
681 	 *	      S.state = RESPOND
682 	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
683 	 *	      Initialize S.GAR := S.ISS
684 	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
685 	 *	      Continue with S.state == RESPOND
686 	 *	      (* A Response packet will be generated in Step 11 *)
687 	 *	 Otherwise,
688 	 *	      Generate Reset(No Connection) unless P.type == Reset
689 	 *	      Drop packet and return
690 	 *
691 	 * NOTE: the check for the packet types is done in
692 	 *	 dccp_rcv_state_process
693 	 */
694 
695 	if (dccp_rcv_state_process(sk, skb, dh, skb->len))
696 		goto reset;
697 	return 0;
698 
699 reset:
700 	dccp_v4_ctl_send_reset(sk, skb);
701 	kfree_skb(skb);
702 	return 0;
703 }
704 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
705 
706 /**
707  *	dccp_invalid_packet  -  check for malformed packets
708  *	@skb: Packet to validate
709  *
710  *	Implements RFC 4340, 8.5:  Step 1: Check header basics
711  *	Packets that fail these checks are ignored and do not receive Resets.
712  */
713 int dccp_invalid_packet(struct sk_buff *skb)
714 {
715 	const struct dccp_hdr *dh;
716 	unsigned int cscov;
717 	u8 dccph_doff;
718 
719 	if (skb->pkt_type != PACKET_HOST)
720 		return 1;
721 
722 	/* If the packet is shorter than 12 bytes, drop packet and return */
723 	if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
724 		DCCP_WARN("pskb_may_pull failed\n");
725 		return 1;
726 	}
727 
728 	dh = dccp_hdr(skb);
729 
730 	/* If P.type is not understood, drop packet and return */
731 	if (dh->dccph_type >= DCCP_PKT_INVALID) {
732 		DCCP_WARN("invalid packet type\n");
733 		return 1;
734 	}
735 
736 	/*
737 	 * If P.Data Offset is too small for packet type, drop packet and return
738 	 */
739 	dccph_doff = dh->dccph_doff;
740 	if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
741 		DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
742 		return 1;
743 	}
744 	/*
745 	 * If P.Data Offset is too large for packet, drop packet and return
746 	 */
747 	if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
748 		DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
749 		return 1;
750 	}
751 	dh = dccp_hdr(skb);
752 	/*
753 	 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
754 	 * has short sequence numbers), drop packet and return
755 	 */
756 	if ((dh->dccph_type < DCCP_PKT_DATA    ||
757 	    dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
758 		DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
759 			  dccp_packet_name(dh->dccph_type));
760 		return 1;
761 	}
762 
763 	/*
764 	 * If P.CsCov is too large for the packet size, drop packet and return.
765 	 * This must come _before_ checksumming (not as RFC 4340 suggests).
766 	 */
767 	cscov = dccp_csum_coverage(skb);
768 	if (cscov > skb->len) {
769 		DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
770 			  dh->dccph_cscov, skb->len);
771 		return 1;
772 	}
773 
774 	/* If header checksum is incorrect, drop packet and return.
775 	 * (This step is completed in the AF-dependent functions.) */
776 	skb->csum = skb_checksum(skb, 0, cscov, 0);
777 
778 	return 0;
779 }
780 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
781 
782 /* this is called when real data arrives */
783 static int dccp_v4_rcv(struct sk_buff *skb)
784 {
785 	const struct dccp_hdr *dh;
786 	const struct iphdr *iph;
787 	bool refcounted;
788 	struct sock *sk;
789 	int min_cov;
790 
791 	/* Step 1: Check header basics */
792 
793 	if (dccp_invalid_packet(skb))
794 		goto discard_it;
795 
796 	iph = ip_hdr(skb);
797 	/* Step 1: If header checksum is incorrect, drop packet and return */
798 	if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
799 		DCCP_WARN("dropped packet with invalid checksum\n");
800 		goto discard_it;
801 	}
802 
803 	dh = dccp_hdr(skb);
804 
805 	DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
806 	DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
807 
808 	dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
809 		      dccp_packet_name(dh->dccph_type),
810 		      &iph->saddr, ntohs(dh->dccph_sport),
811 		      &iph->daddr, ntohs(dh->dccph_dport),
812 		      (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
813 
814 	if (dccp_packet_without_ack(skb)) {
815 		DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
816 		dccp_pr_debug_cat("\n");
817 	} else {
818 		DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
819 		dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
820 				  DCCP_SKB_CB(skb)->dccpd_ack_seq);
821 	}
822 
823 lookup:
824 	sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
825 			       dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
826 	if (!sk) {
827 		dccp_pr_debug("failed to look up flow ID in table and "
828 			      "get corresponding socket\n");
829 		goto no_dccp_socket;
830 	}
831 
832 	/*
833 	 * Step 2:
834 	 *	... or S.state == TIMEWAIT,
835 	 *		Generate Reset(No Connection) unless P.type == Reset
836 	 *		Drop packet and return
837 	 */
838 	if (sk->sk_state == DCCP_TIME_WAIT) {
839 		dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
840 		inet_twsk_put(inet_twsk(sk));
841 		goto no_dccp_socket;
842 	}
843 
844 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
845 		struct request_sock *req = inet_reqsk(sk);
846 		struct sock *nsk;
847 
848 		sk = req->rsk_listener;
849 		if (unlikely(sk->sk_state != DCCP_LISTEN)) {
850 			inet_csk_reqsk_queue_drop_and_put(sk, req);
851 			goto lookup;
852 		}
853 		sock_hold(sk);
854 		refcounted = true;
855 		nsk = dccp_check_req(sk, skb, req);
856 		if (!nsk) {
857 			reqsk_put(req);
858 			goto discard_and_relse;
859 		}
860 		if (nsk == sk) {
861 			reqsk_put(req);
862 		} else if (dccp_child_process(sk, nsk, skb)) {
863 			dccp_v4_ctl_send_reset(sk, skb);
864 			goto discard_and_relse;
865 		} else {
866 			sock_put(sk);
867 			return 0;
868 		}
869 	}
870 	/*
871 	 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
872 	 *	o if MinCsCov = 0, only packets with CsCov = 0 are accepted
873 	 *	o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
874 	 */
875 	min_cov = dccp_sk(sk)->dccps_pcrlen;
876 	if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
877 		dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
878 			      dh->dccph_cscov, min_cov);
879 		/* FIXME: "Such packets SHOULD be reported using Data Dropped
880 		 *         options (Section 11.7) with Drop Code 0, Protocol
881 		 *         Constraints."                                     */
882 		goto discard_and_relse;
883 	}
884 
885 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
886 		goto discard_and_relse;
887 	nf_reset_ct(skb);
888 
889 	return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
890 
891 no_dccp_socket:
892 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
893 		goto discard_it;
894 	/*
895 	 * Step 2:
896 	 *	If no socket ...
897 	 *		Generate Reset(No Connection) unless P.type == Reset
898 	 *		Drop packet and return
899 	 */
900 	if (dh->dccph_type != DCCP_PKT_RESET) {
901 		DCCP_SKB_CB(skb)->dccpd_reset_code =
902 					DCCP_RESET_CODE_NO_CONNECTION;
903 		dccp_v4_ctl_send_reset(sk, skb);
904 	}
905 
906 discard_it:
907 	kfree_skb(skb);
908 	return 0;
909 
910 discard_and_relse:
911 	if (refcounted)
912 		sock_put(sk);
913 	goto discard_it;
914 }
915 
916 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
917 	.queue_xmit	   = ip_queue_xmit,
918 	.send_check	   = dccp_v4_send_check,
919 	.rebuild_header	   = inet_sk_rebuild_header,
920 	.conn_request	   = dccp_v4_conn_request,
921 	.syn_recv_sock	   = dccp_v4_request_recv_sock,
922 	.net_header_len	   = sizeof(struct iphdr),
923 	.setsockopt	   = ip_setsockopt,
924 	.getsockopt	   = ip_getsockopt,
925 	.addr2sockaddr	   = inet_csk_addr2sockaddr,
926 	.sockaddr_len	   = sizeof(struct sockaddr_in),
927 };
928 
929 static int dccp_v4_init_sock(struct sock *sk)
930 {
931 	static __u8 dccp_v4_ctl_sock_initialized;
932 	int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
933 
934 	if (err == 0) {
935 		if (unlikely(!dccp_v4_ctl_sock_initialized))
936 			dccp_v4_ctl_sock_initialized = 1;
937 		inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
938 	}
939 
940 	return err;
941 }
942 
943 static struct timewait_sock_ops dccp_timewait_sock_ops = {
944 	.twsk_obj_size	= sizeof(struct inet_timewait_sock),
945 };
946 
947 static struct proto dccp_v4_prot = {
948 	.name			= "DCCP",
949 	.owner			= THIS_MODULE,
950 	.close			= dccp_close,
951 	.connect		= dccp_v4_connect,
952 	.disconnect		= dccp_disconnect,
953 	.ioctl			= dccp_ioctl,
954 	.init			= dccp_v4_init_sock,
955 	.setsockopt		= dccp_setsockopt,
956 	.getsockopt		= dccp_getsockopt,
957 	.sendmsg		= dccp_sendmsg,
958 	.recvmsg		= dccp_recvmsg,
959 	.backlog_rcv		= dccp_v4_do_rcv,
960 	.hash			= inet_hash,
961 	.unhash			= inet_unhash,
962 	.accept			= inet_csk_accept,
963 	.get_port		= inet_csk_get_port,
964 	.shutdown		= dccp_shutdown,
965 	.destroy		= dccp_destroy_sock,
966 	.orphan_count		= &dccp_orphan_count,
967 	.max_header		= MAX_DCCP_HEADER,
968 	.obj_size		= sizeof(struct dccp_sock),
969 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
970 	.rsk_prot		= &dccp_request_sock_ops,
971 	.twsk_prot		= &dccp_timewait_sock_ops,
972 	.h.hashinfo		= &dccp_hashinfo,
973 };
974 
975 static const struct net_protocol dccp_v4_protocol = {
976 	.handler	= dccp_v4_rcv,
977 	.err_handler	= dccp_v4_err,
978 	.no_policy	= 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 };
1005 
1006 static struct inet_protosw dccp_v4_protosw = {
1007 	.type		= SOCK_DCCP,
1008 	.protocol	= IPPROTO_DCCP,
1009 	.prot		= &dccp_v4_prot,
1010 	.ops		= &inet_dccp_ops,
1011 	.flags		= INET_PROTOSW_ICSK,
1012 };
1013 
1014 static int __net_init dccp_v4_init_net(struct net *net)
1015 {
1016 	struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1017 
1018 	if (dccp_hashinfo.bhash == NULL)
1019 		return -ESOCKTNOSUPPORT;
1020 
1021 	return inet_ctl_sock_create(&pn->v4_ctl_sk, PF_INET,
1022 				    SOCK_DCCP, IPPROTO_DCCP, net);
1023 }
1024 
1025 static void __net_exit dccp_v4_exit_net(struct net *net)
1026 {
1027 	struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1028 
1029 	inet_ctl_sock_destroy(pn->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 	.id	= &dccp_v4_pernet_id,
1042 	.size   = sizeof(struct dccp_v4_pernet),
1043 };
1044 
1045 static int __init dccp_v4_init(void)
1046 {
1047 	int err = proto_register(&dccp_v4_prot, 1);
1048 
1049 	if (err)
1050 		goto out;
1051 
1052 	inet_register_protosw(&dccp_v4_protosw);
1053 
1054 	err = register_pernet_subsys(&dccp_v4_ops);
1055 	if (err)
1056 		goto out_destroy_ctl_sock;
1057 
1058 	err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1059 	if (err)
1060 		goto out_proto_unregister;
1061 
1062 out:
1063 	return err;
1064 out_proto_unregister:
1065 	unregister_pernet_subsys(&dccp_v4_ops);
1066 out_destroy_ctl_sock:
1067 	inet_unregister_protosw(&dccp_v4_protosw);
1068 	proto_unregister(&dccp_v4_prot);
1069 	goto out;
1070 }
1071 
1072 static void __exit dccp_v4_exit(void)
1073 {
1074 	inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1075 	unregister_pernet_subsys(&dccp_v4_ops);
1076 	inet_unregister_protosw(&dccp_v4_protosw);
1077 	proto_unregister(&dccp_v4_prot);
1078 }
1079 
1080 module_init(dccp_v4_init);
1081 module_exit(dccp_v4_exit);
1082 
1083 /*
1084  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1085  * values directly, Also cover the case where the protocol is not specified,
1086  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1087  */
1088 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1089 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1090 MODULE_LICENSE("GPL");
1091 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1092 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1093