xref: /openbmc/linux/net/dccp/ipv4.c (revision 6db6b729)
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 		err = inet_bhash2_update_saddr(sk,  &fl4->saddr, AF_INET);
94 		if (err) {
95 			ip_rt_put(rt);
96 			return err;
97 		}
98 	} else {
99 		sk_rcv_saddr_set(sk, inet->inet_saddr);
100 	}
101 
102 	inet->inet_dport = usin->sin_port;
103 	sk_daddr_set(sk, daddr);
104 
105 	inet_csk(sk)->icsk_ext_hdr_len = 0;
106 	if (inet_opt)
107 		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
108 	/*
109 	 * Socket identity is still unknown (sport may be zero).
110 	 * However we set state to DCCP_REQUESTING and not releasing socket
111 	 * lock select source port, enter ourselves into the hash tables and
112 	 * complete initialization after this.
113 	 */
114 	dccp_set_state(sk, DCCP_REQUESTING);
115 	err = inet_hash_connect(&dccp_death_row, sk);
116 	if (err != 0)
117 		goto failure;
118 
119 	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
120 			       inet->inet_sport, inet->inet_dport, sk);
121 	if (IS_ERR(rt)) {
122 		err = PTR_ERR(rt);
123 		rt = NULL;
124 		goto failure;
125 	}
126 	/* OK, now commit destination to socket.  */
127 	sk_setup_caps(sk, &rt->dst);
128 
129 	dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
130 						    inet->inet_daddr,
131 						    inet->inet_sport,
132 						    inet->inet_dport);
133 	atomic_set(&inet->inet_id, get_random_u16());
134 
135 	err = dccp_connect(sk);
136 	rt = NULL;
137 	if (err != 0)
138 		goto failure;
139 out:
140 	return err;
141 failure:
142 	/*
143 	 * This unhashes the socket and releases the local port, if necessary.
144 	 */
145 	dccp_set_state(sk, DCCP_CLOSED);
146 	inet_bhash2_reset_saddr(sk);
147 	ip_rt_put(rt);
148 	sk->sk_route_caps = 0;
149 	inet->inet_dport = 0;
150 	goto out;
151 }
152 EXPORT_SYMBOL_GPL(dccp_v4_connect);
153 
154 /*
155  * This routine does path mtu discovery as defined in RFC1191.
156  */
157 static inline void dccp_do_pmtu_discovery(struct sock *sk,
158 					  const struct iphdr *iph,
159 					  u32 mtu)
160 {
161 	struct dst_entry *dst;
162 	const struct inet_sock *inet = inet_sk(sk);
163 	const struct dccp_sock *dp = dccp_sk(sk);
164 
165 	/* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
166 	 * send out by Linux are always < 576bytes so they should go through
167 	 * unfragmented).
168 	 */
169 	if (sk->sk_state == DCCP_LISTEN)
170 		return;
171 
172 	dst = inet_csk_update_pmtu(sk, mtu);
173 	if (!dst)
174 		return;
175 
176 	/* Something is about to be wrong... Remember soft error
177 	 * for the case, if this connection will not able to recover.
178 	 */
179 	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
180 		WRITE_ONCE(sk->sk_err_soft, EMSGSIZE);
181 
182 	mtu = dst_mtu(dst);
183 
184 	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
185 	    ip_sk_accept_pmtu(sk) &&
186 	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
187 		dccp_sync_mss(sk, mtu);
188 
189 		/*
190 		 * From RFC 4340, sec. 14.1:
191 		 *
192 		 *	DCCP-Sync packets are the best choice for upward
193 		 *	probing, since DCCP-Sync probes do not risk application
194 		 *	data loss.
195 		 */
196 		dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
197 	} /* else let the usual retransmit timer handle it */
198 }
199 
200 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
201 {
202 	struct dst_entry *dst = __sk_dst_check(sk, 0);
203 
204 	if (dst)
205 		dst->ops->redirect(dst, sk, skb);
206 }
207 
208 void dccp_req_err(struct sock *sk, u64 seq)
209 	{
210 	struct request_sock *req = inet_reqsk(sk);
211 	struct net *net = sock_net(sk);
212 
213 	/*
214 	 * ICMPs are not backlogged, hence we cannot get an established
215 	 * socket here.
216 	 */
217 	if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
218 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
219 	} else {
220 		/*
221 		 * Still in RESPOND, just remove it silently.
222 		 * There is no good way to pass the error to the newly
223 		 * created socket, and POSIX does not want network
224 		 * errors returned from accept().
225 		 */
226 		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
227 	}
228 	reqsk_put(req);
229 }
230 EXPORT_SYMBOL(dccp_req_err);
231 
232 /*
233  * This routine is called by the ICMP module when it gets some sort of error
234  * condition. If err < 0 then the socket should be closed and the error
235  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
236  * After adjustment header points to the first 8 bytes of the tcp header. We
237  * need to find the appropriate port.
238  *
239  * The locking strategy used here is very "optimistic". When someone else
240  * accesses the socket the ICMP is just dropped and for some paths there is no
241  * check at all. A more general error queue to queue errors for later handling
242  * is probably better.
243  */
244 static int dccp_v4_err(struct sk_buff *skb, u32 info)
245 {
246 	const struct iphdr *iph = (struct iphdr *)skb->data;
247 	const u8 offset = iph->ihl << 2;
248 	const struct dccp_hdr *dh;
249 	struct dccp_sock *dp;
250 	const int type = icmp_hdr(skb)->type;
251 	const int code = icmp_hdr(skb)->code;
252 	struct sock *sk;
253 	__u64 seq;
254 	int err;
255 	struct net *net = dev_net(skb->dev);
256 
257 	if (!pskb_may_pull(skb, offset + sizeof(*dh)))
258 		return -EINVAL;
259 	dh = (struct dccp_hdr *)(skb->data + offset);
260 	if (!pskb_may_pull(skb, offset + __dccp_basic_hdr_len(dh)))
261 		return -EINVAL;
262 	iph = (struct iphdr *)skb->data;
263 	dh = (struct dccp_hdr *)(skb->data + offset);
264 
265 	sk = __inet_lookup_established(net, &dccp_hashinfo,
266 				       iph->daddr, dh->dccph_dport,
267 				       iph->saddr, ntohs(dh->dccph_sport),
268 				       inet_iif(skb), 0);
269 	if (!sk) {
270 		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
271 		return -ENOENT;
272 	}
273 
274 	if (sk->sk_state == DCCP_TIME_WAIT) {
275 		inet_twsk_put(inet_twsk(sk));
276 		return 0;
277 	}
278 	seq = dccp_hdr_seq(dh);
279 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
280 		dccp_req_err(sk, seq);
281 		return 0;
282 	}
283 
284 	bh_lock_sock(sk);
285 	/* If too many ICMPs get dropped on busy
286 	 * servers this needs to be solved differently.
287 	 */
288 	if (sock_owned_by_user(sk))
289 		__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
290 
291 	if (sk->sk_state == DCCP_CLOSED)
292 		goto out;
293 
294 	dp = dccp_sk(sk);
295 	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
296 	    !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
297 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
298 		goto out;
299 	}
300 
301 	switch (type) {
302 	case ICMP_REDIRECT:
303 		if (!sock_owned_by_user(sk))
304 			dccp_do_redirect(skb, sk);
305 		goto out;
306 	case ICMP_SOURCE_QUENCH:
307 		/* Just silently ignore these. */
308 		goto out;
309 	case ICMP_PARAMETERPROB:
310 		err = EPROTO;
311 		break;
312 	case ICMP_DEST_UNREACH:
313 		if (code > NR_ICMP_UNREACH)
314 			goto out;
315 
316 		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
317 			if (!sock_owned_by_user(sk))
318 				dccp_do_pmtu_discovery(sk, iph, info);
319 			goto out;
320 		}
321 
322 		err = icmp_err_convert[code].errno;
323 		break;
324 	case ICMP_TIME_EXCEEDED:
325 		err = EHOSTUNREACH;
326 		break;
327 	default:
328 		goto out;
329 	}
330 
331 	switch (sk->sk_state) {
332 	case DCCP_REQUESTING:
333 	case DCCP_RESPOND:
334 		if (!sock_owned_by_user(sk)) {
335 			__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
336 			sk->sk_err = err;
337 
338 			sk_error_report(sk);
339 
340 			dccp_done(sk);
341 		} else {
342 			WRITE_ONCE(sk->sk_err_soft, err);
343 		}
344 		goto out;
345 	}
346 
347 	/* If we've already connected we will keep trying
348 	 * until we time out, or the user gives up.
349 	 *
350 	 * rfc1122 4.2.3.9 allows to consider as hard errors
351 	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
352 	 * but it is obsoleted by pmtu discovery).
353 	 *
354 	 * Note, that in modern internet, where routing is unreliable
355 	 * and in each dark corner broken firewalls sit, sending random
356 	 * errors ordered by their masters even this two messages finally lose
357 	 * their original sense (even Linux sends invalid PORT_UNREACHs)
358 	 *
359 	 * Now we are in compliance with RFCs.
360 	 *							--ANK (980905)
361 	 */
362 
363 	if (!sock_owned_by_user(sk) && inet_test_bit(RECVERR, sk)) {
364 		sk->sk_err = err;
365 		sk_error_report(sk);
366 	} else { /* Only an error on timeout */
367 		WRITE_ONCE(sk->sk_err_soft, err);
368 	}
369 out:
370 	bh_unlock_sock(sk);
371 	sock_put(sk);
372 	return 0;
373 }
374 
375 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
376 				      __be32 src, __be32 dst)
377 {
378 	return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
379 }
380 
381 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
382 {
383 	const struct inet_sock *inet = inet_sk(sk);
384 	struct dccp_hdr *dh = dccp_hdr(skb);
385 
386 	dccp_csum_outgoing(skb);
387 	dh->dccph_checksum = dccp_v4_csum_finish(skb,
388 						 inet->inet_saddr,
389 						 inet->inet_daddr);
390 }
391 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
392 
393 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
394 {
395 	return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
396 					   ip_hdr(skb)->saddr,
397 					   dccp_hdr(skb)->dccph_dport,
398 					   dccp_hdr(skb)->dccph_sport);
399 }
400 
401 /*
402  * The three way handshake has completed - we got a valid ACK or DATAACK -
403  * now create the new socket.
404  *
405  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
406  */
407 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
408 				       struct sk_buff *skb,
409 				       struct request_sock *req,
410 				       struct dst_entry *dst,
411 				       struct request_sock *req_unhash,
412 				       bool *own_req)
413 {
414 	struct inet_request_sock *ireq;
415 	struct inet_sock *newinet;
416 	struct sock *newsk;
417 
418 	if (sk_acceptq_is_full(sk))
419 		goto exit_overflow;
420 
421 	newsk = dccp_create_openreq_child(sk, req, skb);
422 	if (newsk == NULL)
423 		goto exit_nonewsk;
424 
425 	newinet		   = inet_sk(newsk);
426 	ireq		   = inet_rsk(req);
427 	sk_daddr_set(newsk, ireq->ir_rmt_addr);
428 	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
429 	newinet->inet_saddr	= ireq->ir_loc_addr;
430 	RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
431 	newinet->mc_index  = inet_iif(skb);
432 	newinet->mc_ttl	   = ip_hdr(skb)->ttl;
433 	atomic_set(&newinet->inet_id, get_random_u16());
434 
435 	if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
436 		goto put_and_exit;
437 
438 	sk_setup_caps(newsk, dst);
439 
440 	dccp_sync_mss(newsk, dst_mtu(dst));
441 
442 	if (__inet_inherit_port(sk, newsk) < 0)
443 		goto put_and_exit;
444 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
445 	if (*own_req)
446 		ireq->ireq_opt = NULL;
447 	else
448 		newinet->inet_opt = NULL;
449 	return newsk;
450 
451 exit_overflow:
452 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
453 exit_nonewsk:
454 	dst_release(dst);
455 exit:
456 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
457 	return NULL;
458 put_and_exit:
459 	newinet->inet_opt = NULL;
460 	inet_csk_prepare_forced_close(newsk);
461 	dccp_done(newsk);
462 	goto exit;
463 }
464 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
465 
466 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
467 					   struct sk_buff *skb)
468 {
469 	struct rtable *rt;
470 	const struct iphdr *iph = ip_hdr(skb);
471 	struct flowi4 fl4 = {
472 		.flowi4_oif = inet_iif(skb),
473 		.daddr = iph->saddr,
474 		.saddr = iph->daddr,
475 		.flowi4_tos = ip_sock_rt_tos(sk),
476 		.flowi4_scope = ip_sock_rt_scope(sk),
477 		.flowi4_proto = sk->sk_protocol,
478 		.fl4_sport = dccp_hdr(skb)->dccph_dport,
479 		.fl4_dport = dccp_hdr(skb)->dccph_sport,
480 	};
481 
482 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
483 	rt = ip_route_output_flow(net, &fl4, sk);
484 	if (IS_ERR(rt)) {
485 		IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
486 		return NULL;
487 	}
488 
489 	return &rt->dst;
490 }
491 
492 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
493 {
494 	int err = -1;
495 	struct sk_buff *skb;
496 	struct dst_entry *dst;
497 	struct flowi4 fl4;
498 
499 	dst = inet_csk_route_req(sk, &fl4, req);
500 	if (dst == NULL)
501 		goto out;
502 
503 	skb = dccp_make_response(sk, dst, req);
504 	if (skb != NULL) {
505 		const struct inet_request_sock *ireq = inet_rsk(req);
506 		struct dccp_hdr *dh = dccp_hdr(skb);
507 
508 		dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
509 							      ireq->ir_rmt_addr);
510 		rcu_read_lock();
511 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
512 					    ireq->ir_rmt_addr,
513 					    rcu_dereference(ireq->ireq_opt),
514 					    inet_sk(sk)->tos);
515 		rcu_read_unlock();
516 		err = net_xmit_eval(err);
517 	}
518 
519 out:
520 	dst_release(dst);
521 	return err;
522 }
523 
524 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
525 {
526 	int err;
527 	const struct iphdr *rxiph;
528 	struct sk_buff *skb;
529 	struct dst_entry *dst;
530 	struct net *net = dev_net(skb_dst(rxskb)->dev);
531 	struct dccp_v4_pernet *pn;
532 	struct sock *ctl_sk;
533 
534 	/* Never send a reset in response to a reset. */
535 	if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
536 		return;
537 
538 	if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
539 		return;
540 
541 	pn = net_generic(net, dccp_v4_pernet_id);
542 	ctl_sk = pn->v4_ctl_sk;
543 	dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
544 	if (dst == NULL)
545 		return;
546 
547 	skb = dccp_ctl_make_reset(ctl_sk, rxskb);
548 	if (skb == NULL)
549 		goto out;
550 
551 	rxiph = ip_hdr(rxskb);
552 	dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
553 								 rxiph->daddr);
554 	skb_dst_set(skb, dst_clone(dst));
555 
556 	local_bh_disable();
557 	bh_lock_sock(ctl_sk);
558 	err = ip_build_and_send_pkt(skb, ctl_sk,
559 				    rxiph->daddr, rxiph->saddr, NULL,
560 				    inet_sk(ctl_sk)->tos);
561 	bh_unlock_sock(ctl_sk);
562 
563 	if (net_xmit_eval(err) == 0) {
564 		__DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
565 		__DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
566 	}
567 	local_bh_enable();
568 out:
569 	dst_release(dst);
570 }
571 
572 static void dccp_v4_reqsk_destructor(struct request_sock *req)
573 {
574 	dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
575 	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
576 }
577 
578 void dccp_syn_ack_timeout(const struct request_sock *req)
579 {
580 }
581 EXPORT_SYMBOL(dccp_syn_ack_timeout);
582 
583 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
584 	.family		= PF_INET,
585 	.obj_size	= sizeof(struct dccp_request_sock),
586 	.rtx_syn_ack	= dccp_v4_send_response,
587 	.send_ack	= dccp_reqsk_send_ack,
588 	.destructor	= dccp_v4_reqsk_destructor,
589 	.send_reset	= dccp_v4_ctl_send_reset,
590 	.syn_ack_timeout = dccp_syn_ack_timeout,
591 };
592 
593 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
594 {
595 	struct inet_request_sock *ireq;
596 	struct request_sock *req;
597 	struct dccp_request_sock *dreq;
598 	const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
599 	struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
600 
601 	/* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
602 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
603 		return 0;	/* discard, don't send a reset here */
604 
605 	if (dccp_bad_service_code(sk, service)) {
606 		dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
607 		goto drop;
608 	}
609 	/*
610 	 * TW buckets are converted to open requests without
611 	 * limitations, they conserve resources and peer is
612 	 * evidently real one.
613 	 */
614 	dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
615 	if (inet_csk_reqsk_queue_is_full(sk))
616 		goto drop;
617 
618 	if (sk_acceptq_is_full(sk))
619 		goto drop;
620 
621 	req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
622 	if (req == NULL)
623 		goto drop;
624 
625 	if (dccp_reqsk_init(req, dccp_sk(sk), skb))
626 		goto drop_and_free;
627 
628 	dreq = dccp_rsk(req);
629 	if (dccp_parse_options(sk, dreq, skb))
630 		goto drop_and_free;
631 
632 	if (security_inet_conn_request(sk, skb, req))
633 		goto drop_and_free;
634 
635 	ireq = inet_rsk(req);
636 	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
637 	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
638 	ireq->ir_mark = inet_request_mark(sk, skb);
639 	ireq->ireq_family = AF_INET;
640 	ireq->ir_iif = READ_ONCE(sk->sk_bound_dev_if);
641 
642 	/*
643 	 * Step 3: Process LISTEN state
644 	 *
645 	 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
646 	 *
647 	 * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
648 	 */
649 	dreq->dreq_isr	   = dcb->dccpd_seq;
650 	dreq->dreq_gsr	   = dreq->dreq_isr;
651 	dreq->dreq_iss	   = dccp_v4_init_sequence(skb);
652 	dreq->dreq_gss     = dreq->dreq_iss;
653 	dreq->dreq_service = service;
654 
655 	if (dccp_v4_send_response(sk, req))
656 		goto drop_and_free;
657 
658 	inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
659 	reqsk_put(req);
660 	return 0;
661 
662 drop_and_free:
663 	reqsk_free(req);
664 drop:
665 	__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
666 	return -1;
667 }
668 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
669 
670 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
671 {
672 	struct dccp_hdr *dh = dccp_hdr(skb);
673 
674 	if (sk->sk_state == DCCP_OPEN) { /* Fast path */
675 		if (dccp_rcv_established(sk, skb, dh, skb->len))
676 			goto reset;
677 		return 0;
678 	}
679 
680 	/*
681 	 *  Step 3: Process LISTEN state
682 	 *	 If P.type == Request or P contains a valid Init Cookie option,
683 	 *	      (* Must scan the packet's options to check for Init
684 	 *		 Cookies.  Only Init Cookies are processed here,
685 	 *		 however; other options are processed in Step 8.  This
686 	 *		 scan need only be performed if the endpoint uses Init
687 	 *		 Cookies *)
688 	 *	      (* Generate a new socket and switch to that socket *)
689 	 *	      Set S := new socket for this port pair
690 	 *	      S.state = RESPOND
691 	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
692 	 *	      Initialize S.GAR := S.ISS
693 	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
694 	 *	      Continue with S.state == RESPOND
695 	 *	      (* A Response packet will be generated in Step 11 *)
696 	 *	 Otherwise,
697 	 *	      Generate Reset(No Connection) unless P.type == Reset
698 	 *	      Drop packet and return
699 	 *
700 	 * NOTE: the check for the packet types is done in
701 	 *	 dccp_rcv_state_process
702 	 */
703 
704 	if (dccp_rcv_state_process(sk, skb, dh, skb->len))
705 		goto reset;
706 	return 0;
707 
708 reset:
709 	dccp_v4_ctl_send_reset(sk, skb);
710 	kfree_skb(skb);
711 	return 0;
712 }
713 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
714 
715 /**
716  *	dccp_invalid_packet  -  check for malformed packets
717  *	@skb: Packet to validate
718  *
719  *	Implements RFC 4340, 8.5:  Step 1: Check header basics
720  *	Packets that fail these checks are ignored and do not receive Resets.
721  */
722 int dccp_invalid_packet(struct sk_buff *skb)
723 {
724 	const struct dccp_hdr *dh;
725 	unsigned int cscov;
726 	u8 dccph_doff;
727 
728 	if (skb->pkt_type != PACKET_HOST)
729 		return 1;
730 
731 	/* If the packet is shorter than 12 bytes, drop packet and return */
732 	if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
733 		DCCP_WARN("pskb_may_pull failed\n");
734 		return 1;
735 	}
736 
737 	dh = dccp_hdr(skb);
738 
739 	/* If P.type is not understood, drop packet and return */
740 	if (dh->dccph_type >= DCCP_PKT_INVALID) {
741 		DCCP_WARN("invalid packet type\n");
742 		return 1;
743 	}
744 
745 	/*
746 	 * If P.Data Offset is too small for packet type, drop packet and return
747 	 */
748 	dccph_doff = dh->dccph_doff;
749 	if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
750 		DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
751 		return 1;
752 	}
753 	/*
754 	 * If P.Data Offset is too large for packet, drop packet and return
755 	 */
756 	if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
757 		DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
758 		return 1;
759 	}
760 	dh = dccp_hdr(skb);
761 	/*
762 	 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
763 	 * has short sequence numbers), drop packet and return
764 	 */
765 	if ((dh->dccph_type < DCCP_PKT_DATA    ||
766 	    dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
767 		DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
768 			  dccp_packet_name(dh->dccph_type));
769 		return 1;
770 	}
771 
772 	/*
773 	 * If P.CsCov is too large for the packet size, drop packet and return.
774 	 * This must come _before_ checksumming (not as RFC 4340 suggests).
775 	 */
776 	cscov = dccp_csum_coverage(skb);
777 	if (cscov > skb->len) {
778 		DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
779 			  dh->dccph_cscov, skb->len);
780 		return 1;
781 	}
782 
783 	/* If header checksum is incorrect, drop packet and return.
784 	 * (This step is completed in the AF-dependent functions.) */
785 	skb->csum = skb_checksum(skb, 0, cscov, 0);
786 
787 	return 0;
788 }
789 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
790 
791 /* this is called when real data arrives */
792 static int dccp_v4_rcv(struct sk_buff *skb)
793 {
794 	const struct dccp_hdr *dh;
795 	const struct iphdr *iph;
796 	bool refcounted;
797 	struct sock *sk;
798 	int min_cov;
799 
800 	/* Step 1: Check header basics */
801 
802 	if (dccp_invalid_packet(skb))
803 		goto discard_it;
804 
805 	iph = ip_hdr(skb);
806 	/* Step 1: If header checksum is incorrect, drop packet and return */
807 	if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
808 		DCCP_WARN("dropped packet with invalid checksum\n");
809 		goto discard_it;
810 	}
811 
812 	dh = dccp_hdr(skb);
813 
814 	DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
815 	DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
816 
817 	dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
818 		      dccp_packet_name(dh->dccph_type),
819 		      &iph->saddr, ntohs(dh->dccph_sport),
820 		      &iph->daddr, ntohs(dh->dccph_dport),
821 		      (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
822 
823 	if (dccp_packet_without_ack(skb)) {
824 		DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
825 		dccp_pr_debug_cat("\n");
826 	} else {
827 		DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
828 		dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
829 				  DCCP_SKB_CB(skb)->dccpd_ack_seq);
830 	}
831 
832 lookup:
833 	sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
834 			       dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
835 	if (!sk) {
836 		dccp_pr_debug("failed to look up flow ID in table and "
837 			      "get corresponding socket\n");
838 		goto no_dccp_socket;
839 	}
840 
841 	/*
842 	 * Step 2:
843 	 *	... or S.state == TIMEWAIT,
844 	 *		Generate Reset(No Connection) unless P.type == Reset
845 	 *		Drop packet and return
846 	 */
847 	if (sk->sk_state == DCCP_TIME_WAIT) {
848 		dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
849 		inet_twsk_put(inet_twsk(sk));
850 		goto no_dccp_socket;
851 	}
852 
853 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
854 		struct request_sock *req = inet_reqsk(sk);
855 		struct sock *nsk;
856 
857 		sk = req->rsk_listener;
858 		if (unlikely(sk->sk_state != DCCP_LISTEN)) {
859 			inet_csk_reqsk_queue_drop_and_put(sk, req);
860 			goto lookup;
861 		}
862 		sock_hold(sk);
863 		refcounted = true;
864 		nsk = dccp_check_req(sk, skb, req);
865 		if (!nsk) {
866 			reqsk_put(req);
867 			goto discard_and_relse;
868 		}
869 		if (nsk == sk) {
870 			reqsk_put(req);
871 		} else if (dccp_child_process(sk, nsk, skb)) {
872 			dccp_v4_ctl_send_reset(sk, skb);
873 			goto discard_and_relse;
874 		} else {
875 			sock_put(sk);
876 			return 0;
877 		}
878 	}
879 	/*
880 	 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
881 	 *	o if MinCsCov = 0, only packets with CsCov = 0 are accepted
882 	 *	o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
883 	 */
884 	min_cov = dccp_sk(sk)->dccps_pcrlen;
885 	if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
886 		dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
887 			      dh->dccph_cscov, min_cov);
888 		/* FIXME: "Such packets SHOULD be reported using Data Dropped
889 		 *         options (Section 11.7) with Drop Code 0, Protocol
890 		 *         Constraints."                                     */
891 		goto discard_and_relse;
892 	}
893 
894 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
895 		goto discard_and_relse;
896 	nf_reset_ct(skb);
897 
898 	return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
899 
900 no_dccp_socket:
901 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
902 		goto discard_it;
903 	/*
904 	 * Step 2:
905 	 *	If no socket ...
906 	 *		Generate Reset(No Connection) unless P.type == Reset
907 	 *		Drop packet and return
908 	 */
909 	if (dh->dccph_type != DCCP_PKT_RESET) {
910 		DCCP_SKB_CB(skb)->dccpd_reset_code =
911 					DCCP_RESET_CODE_NO_CONNECTION;
912 		dccp_v4_ctl_send_reset(sk, skb);
913 	}
914 
915 discard_it:
916 	kfree_skb(skb);
917 	return 0;
918 
919 discard_and_relse:
920 	if (refcounted)
921 		sock_put(sk);
922 	goto discard_it;
923 }
924 
925 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
926 	.queue_xmit	   = ip_queue_xmit,
927 	.send_check	   = dccp_v4_send_check,
928 	.rebuild_header	   = inet_sk_rebuild_header,
929 	.conn_request	   = dccp_v4_conn_request,
930 	.syn_recv_sock	   = dccp_v4_request_recv_sock,
931 	.net_header_len	   = sizeof(struct iphdr),
932 	.setsockopt	   = ip_setsockopt,
933 	.getsockopt	   = ip_getsockopt,
934 	.addr2sockaddr	   = inet_csk_addr2sockaddr,
935 	.sockaddr_len	   = sizeof(struct sockaddr_in),
936 };
937 
938 static int dccp_v4_init_sock(struct sock *sk)
939 {
940 	static __u8 dccp_v4_ctl_sock_initialized;
941 	int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
942 
943 	if (err == 0) {
944 		if (unlikely(!dccp_v4_ctl_sock_initialized))
945 			dccp_v4_ctl_sock_initialized = 1;
946 		inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
947 	}
948 
949 	return err;
950 }
951 
952 static struct timewait_sock_ops dccp_timewait_sock_ops = {
953 	.twsk_obj_size	= sizeof(struct inet_timewait_sock),
954 };
955 
956 static struct proto dccp_v4_prot = {
957 	.name			= "DCCP",
958 	.owner			= THIS_MODULE,
959 	.close			= dccp_close,
960 	.connect		= dccp_v4_connect,
961 	.disconnect		= dccp_disconnect,
962 	.ioctl			= dccp_ioctl,
963 	.init			= dccp_v4_init_sock,
964 	.setsockopt		= dccp_setsockopt,
965 	.getsockopt		= dccp_getsockopt,
966 	.sendmsg		= dccp_sendmsg,
967 	.recvmsg		= dccp_recvmsg,
968 	.backlog_rcv		= dccp_v4_do_rcv,
969 	.hash			= inet_hash,
970 	.unhash			= inet_unhash,
971 	.accept			= inet_csk_accept,
972 	.get_port		= inet_csk_get_port,
973 	.shutdown		= dccp_shutdown,
974 	.destroy		= dccp_destroy_sock,
975 	.orphan_count		= &dccp_orphan_count,
976 	.max_header		= MAX_DCCP_HEADER,
977 	.obj_size		= sizeof(struct dccp_sock),
978 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
979 	.rsk_prot		= &dccp_request_sock_ops,
980 	.twsk_prot		= &dccp_timewait_sock_ops,
981 	.h.hashinfo		= &dccp_hashinfo,
982 };
983 
984 static const struct net_protocol dccp_v4_protocol = {
985 	.handler	= dccp_v4_rcv,
986 	.err_handler	= dccp_v4_err,
987 	.no_policy	= 1,
988 	.icmp_strict_tag_validation = 1,
989 };
990 
991 static const struct proto_ops inet_dccp_ops = {
992 	.family		   = PF_INET,
993 	.owner		   = THIS_MODULE,
994 	.release	   = inet_release,
995 	.bind		   = inet_bind,
996 	.connect	   = inet_stream_connect,
997 	.socketpair	   = sock_no_socketpair,
998 	.accept		   = inet_accept,
999 	.getname	   = inet_getname,
1000 	/* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1001 	.poll		   = dccp_poll,
1002 	.ioctl		   = inet_ioctl,
1003 	.gettstamp	   = sock_gettstamp,
1004 	/* FIXME: work on inet_listen to rename it to sock_common_listen */
1005 	.listen		   = inet_dccp_listen,
1006 	.shutdown	   = inet_shutdown,
1007 	.setsockopt	   = sock_common_setsockopt,
1008 	.getsockopt	   = sock_common_getsockopt,
1009 	.sendmsg	   = inet_sendmsg,
1010 	.recvmsg	   = sock_common_recvmsg,
1011 	.mmap		   = sock_no_mmap,
1012 };
1013 
1014 static struct inet_protosw dccp_v4_protosw = {
1015 	.type		= SOCK_DCCP,
1016 	.protocol	= IPPROTO_DCCP,
1017 	.prot		= &dccp_v4_prot,
1018 	.ops		= &inet_dccp_ops,
1019 	.flags		= INET_PROTOSW_ICSK,
1020 };
1021 
1022 static int __net_init dccp_v4_init_net(struct net *net)
1023 {
1024 	struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1025 
1026 	if (dccp_hashinfo.bhash == NULL)
1027 		return -ESOCKTNOSUPPORT;
1028 
1029 	return inet_ctl_sock_create(&pn->v4_ctl_sk, PF_INET,
1030 				    SOCK_DCCP, IPPROTO_DCCP, net);
1031 }
1032 
1033 static void __net_exit dccp_v4_exit_net(struct net *net)
1034 {
1035 	struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1036 
1037 	inet_ctl_sock_destroy(pn->v4_ctl_sk);
1038 }
1039 
1040 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1041 {
1042 	inet_twsk_purge(&dccp_hashinfo, AF_INET);
1043 }
1044 
1045 static struct pernet_operations dccp_v4_ops = {
1046 	.init	= dccp_v4_init_net,
1047 	.exit	= dccp_v4_exit_net,
1048 	.exit_batch = dccp_v4_exit_batch,
1049 	.id	= &dccp_v4_pernet_id,
1050 	.size   = sizeof(struct dccp_v4_pernet),
1051 };
1052 
1053 static int __init dccp_v4_init(void)
1054 {
1055 	int err = proto_register(&dccp_v4_prot, 1);
1056 
1057 	if (err)
1058 		goto out;
1059 
1060 	inet_register_protosw(&dccp_v4_protosw);
1061 
1062 	err = register_pernet_subsys(&dccp_v4_ops);
1063 	if (err)
1064 		goto out_destroy_ctl_sock;
1065 
1066 	err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1067 	if (err)
1068 		goto out_proto_unregister;
1069 
1070 out:
1071 	return err;
1072 out_proto_unregister:
1073 	unregister_pernet_subsys(&dccp_v4_ops);
1074 out_destroy_ctl_sock:
1075 	inet_unregister_protosw(&dccp_v4_protosw);
1076 	proto_unregister(&dccp_v4_prot);
1077 	goto out;
1078 }
1079 
1080 static void __exit dccp_v4_exit(void)
1081 {
1082 	inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1083 	unregister_pernet_subsys(&dccp_v4_ops);
1084 	inet_unregister_protosw(&dccp_v4_protosw);
1085 	proto_unregister(&dccp_v4_prot);
1086 }
1087 
1088 module_init(dccp_v4_init);
1089 module_exit(dccp_v4_exit);
1090 
1091 /*
1092  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1093  * values directly, Also cover the case where the protocol is not specified,
1094  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1095  */
1096 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1097 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1098 MODULE_LICENSE("GPL");
1099 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1100 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1101