xref: /linux/net/dccp/ipv4.c (revision 9e9f60108423f18a99c9cc93ef7f23490ecc709b)
1 /*
2  *  net/dccp/ipv4.c
3  *
4  *  An implementation of the DCCP protocol
5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
6  *
7  *	This program is free software; you can redistribute it and/or
8  *	modify it under the terms of the GNU General Public License
9  *	as published by the Free Software Foundation; either version
10  *	2 of the License, or (at your option) any later version.
11  */
12 
13 #include <linux/dccp.h>
14 #include <linux/icmp.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/random.h>
19 
20 #include <net/icmp.h>
21 #include <net/inet_common.h>
22 #include <net/inet_hashtables.h>
23 #include <net/inet_sock.h>
24 #include <net/protocol.h>
25 #include <net/sock.h>
26 #include <net/timewait_sock.h>
27 #include <net/tcp_states.h>
28 #include <net/xfrm.h>
29 #include <net/secure_seq.h>
30 
31 #include "ackvec.h"
32 #include "ccid.h"
33 #include "dccp.h"
34 #include "feat.h"
35 
36 /*
37  * The per-net dccp.v4_ctl_sk socket is used for responding to
38  * the Out-of-the-blue (OOTB) packets. A control sock will be created
39  * for this socket at the initialization time.
40  */
41 
42 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
43 {
44 	const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
45 	struct inet_sock *inet = inet_sk(sk);
46 	struct dccp_sock *dp = dccp_sk(sk);
47 	__be16 orig_sport, orig_dport;
48 	__be32 daddr, nexthop;
49 	struct flowi4 *fl4;
50 	struct rtable *rt;
51 	int err;
52 	struct ip_options_rcu *inet_opt;
53 
54 	dp->dccps_role = DCCP_ROLE_CLIENT;
55 
56 	if (addr_len < sizeof(struct sockaddr_in))
57 		return -EINVAL;
58 
59 	if (usin->sin_family != AF_INET)
60 		return -EAFNOSUPPORT;
61 
62 	nexthop = daddr = usin->sin_addr.s_addr;
63 
64 	inet_opt = rcu_dereference_protected(inet->inet_opt,
65 					     sock_owned_by_user(sk));
66 	if (inet_opt != NULL && inet_opt->opt.srr) {
67 		if (daddr == 0)
68 			return -EINVAL;
69 		nexthop = inet_opt->opt.faddr;
70 	}
71 
72 	orig_sport = inet->inet_sport;
73 	orig_dport = usin->sin_port;
74 	fl4 = &inet->cork.fl.u.ip4;
75 	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
76 			      RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
77 			      IPPROTO_DCCP,
78 			      orig_sport, orig_dport, sk);
79 	if (IS_ERR(rt))
80 		return PTR_ERR(rt);
81 
82 	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
83 		ip_rt_put(rt);
84 		return -ENETUNREACH;
85 	}
86 
87 	if (inet_opt == NULL || !inet_opt->opt.srr)
88 		daddr = fl4->daddr;
89 
90 	if (inet->inet_saddr == 0)
91 		inet->inet_saddr = fl4->saddr;
92 	inet->inet_rcv_saddr = inet->inet_saddr;
93 
94 	inet->inet_dport = usin->sin_port;
95 	inet->inet_daddr = daddr;
96 
97 	inet_csk(sk)->icsk_ext_hdr_len = 0;
98 	if (inet_opt)
99 		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
100 	/*
101 	 * Socket identity is still unknown (sport may be zero).
102 	 * However we set state to DCCP_REQUESTING and not releasing socket
103 	 * lock select source port, enter ourselves into the hash tables and
104 	 * complete initialization after this.
105 	 */
106 	dccp_set_state(sk, DCCP_REQUESTING);
107 	err = inet_hash_connect(&dccp_death_row, sk);
108 	if (err != 0)
109 		goto failure;
110 
111 	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
112 			       inet->inet_sport, inet->inet_dport, sk);
113 	if (IS_ERR(rt)) {
114 		err = PTR_ERR(rt);
115 		rt = NULL;
116 		goto failure;
117 	}
118 	/* OK, now commit destination to socket.  */
119 	sk_setup_caps(sk, &rt->dst);
120 
121 	dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
122 						    inet->inet_daddr,
123 						    inet->inet_sport,
124 						    inet->inet_dport);
125 	inet->inet_id = dp->dccps_iss ^ jiffies;
126 
127 	err = dccp_connect(sk);
128 	rt = NULL;
129 	if (err != 0)
130 		goto failure;
131 out:
132 	return err;
133 failure:
134 	/*
135 	 * This unhashes the socket and releases the local port, if necessary.
136 	 */
137 	dccp_set_state(sk, DCCP_CLOSED);
138 	ip_rt_put(rt);
139 	sk->sk_route_caps = 0;
140 	inet->inet_dport = 0;
141 	goto out;
142 }
143 EXPORT_SYMBOL_GPL(dccp_v4_connect);
144 
145 /*
146  * This routine does path mtu discovery as defined in RFC1191.
147  */
148 static inline void dccp_do_pmtu_discovery(struct sock *sk,
149 					  const struct iphdr *iph,
150 					  u32 mtu)
151 {
152 	struct dst_entry *dst;
153 	const struct inet_sock *inet = inet_sk(sk);
154 	const struct dccp_sock *dp = dccp_sk(sk);
155 
156 	/* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
157 	 * send out by Linux are always < 576bytes so they should go through
158 	 * unfragmented).
159 	 */
160 	if (sk->sk_state == DCCP_LISTEN)
161 		return;
162 
163 	dst = inet_csk_update_pmtu(sk, mtu);
164 	if (!dst)
165 		return;
166 
167 	/* Something is about to be wrong... Remember soft error
168 	 * for the case, if this connection will not able to recover.
169 	 */
170 	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
171 		sk->sk_err_soft = EMSGSIZE;
172 
173 	mtu = dst_mtu(dst);
174 
175 	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
176 	    ip_sk_accept_pmtu(sk) &&
177 	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
178 		dccp_sync_mss(sk, mtu);
179 
180 		/*
181 		 * From RFC 4340, sec. 14.1:
182 		 *
183 		 *	DCCP-Sync packets are the best choice for upward
184 		 *	probing, since DCCP-Sync probes do not risk application
185 		 *	data loss.
186 		 */
187 		dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
188 	} /* else let the usual retransmit timer handle it */
189 }
190 
191 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
192 {
193 	struct dst_entry *dst = __sk_dst_check(sk, 0);
194 
195 	if (dst)
196 		dst->ops->redirect(dst, sk, skb);
197 }
198 
199 /*
200  * This routine is called by the ICMP module when it gets some sort of error
201  * condition. If err < 0 then the socket should be closed and the error
202  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
203  * After adjustment header points to the first 8 bytes of the tcp header. We
204  * need to find the appropriate port.
205  *
206  * The locking strategy used here is very "optimistic". When someone else
207  * accesses the socket the ICMP is just dropped and for some paths there is no
208  * check at all. A more general error queue to queue errors for later handling
209  * is probably better.
210  */
211 static void dccp_v4_err(struct sk_buff *skb, u32 info)
212 {
213 	const struct iphdr *iph = (struct iphdr *)skb->data;
214 	const u8 offset = iph->ihl << 2;
215 	const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
216 	struct dccp_sock *dp;
217 	struct inet_sock *inet;
218 	const int type = icmp_hdr(skb)->type;
219 	const int code = icmp_hdr(skb)->code;
220 	struct sock *sk;
221 	__u64 seq;
222 	int err;
223 	struct net *net = dev_net(skb->dev);
224 
225 	if (skb->len < offset + sizeof(*dh) ||
226 	    skb->len < offset + __dccp_basic_hdr_len(dh)) {
227 		ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
228 		return;
229 	}
230 
231 	sk = inet_lookup(net, &dccp_hashinfo,
232 			iph->daddr, dh->dccph_dport,
233 			iph->saddr, dh->dccph_sport, inet_iif(skb));
234 	if (sk == NULL) {
235 		ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
236 		return;
237 	}
238 
239 	if (sk->sk_state == DCCP_TIME_WAIT) {
240 		inet_twsk_put(inet_twsk(sk));
241 		return;
242 	}
243 
244 	bh_lock_sock(sk);
245 	/* If too many ICMPs get dropped on busy
246 	 * servers this needs to be solved differently.
247 	 */
248 	if (sock_owned_by_user(sk))
249 		NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
250 
251 	if (sk->sk_state == DCCP_CLOSED)
252 		goto out;
253 
254 	dp = dccp_sk(sk);
255 	seq = dccp_hdr_seq(dh);
256 	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
257 	    !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
258 		NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
259 		goto out;
260 	}
261 
262 	switch (type) {
263 	case ICMP_REDIRECT:
264 		dccp_do_redirect(skb, sk);
265 		goto out;
266 	case ICMP_SOURCE_QUENCH:
267 		/* Just silently ignore these. */
268 		goto out;
269 	case ICMP_PARAMETERPROB:
270 		err = EPROTO;
271 		break;
272 	case ICMP_DEST_UNREACH:
273 		if (code > NR_ICMP_UNREACH)
274 			goto out;
275 
276 		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
277 			if (!sock_owned_by_user(sk))
278 				dccp_do_pmtu_discovery(sk, iph, info);
279 			goto out;
280 		}
281 
282 		err = icmp_err_convert[code].errno;
283 		break;
284 	case ICMP_TIME_EXCEEDED:
285 		err = EHOSTUNREACH;
286 		break;
287 	default:
288 		goto out;
289 	}
290 
291 	switch (sk->sk_state) {
292 		struct request_sock *req , **prev;
293 	case DCCP_LISTEN:
294 		if (sock_owned_by_user(sk))
295 			goto out;
296 		req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
297 					  iph->daddr, iph->saddr);
298 		if (!req)
299 			goto out;
300 
301 		/*
302 		 * ICMPs are not backlogged, hence we cannot get an established
303 		 * socket here.
304 		 */
305 		WARN_ON(req->sk);
306 
307 		if (!between48(seq, dccp_rsk(req)->dreq_iss,
308 				    dccp_rsk(req)->dreq_gss)) {
309 			NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
310 			goto out;
311 		}
312 		/*
313 		 * Still in RESPOND, just remove it silently.
314 		 * There is no good way to pass the error to the newly
315 		 * created socket, and POSIX does not want network
316 		 * errors returned from accept().
317 		 */
318 		inet_csk_reqsk_queue_drop(sk, req, prev);
319 		goto out;
320 
321 	case DCCP_REQUESTING:
322 	case DCCP_RESPOND:
323 		if (!sock_owned_by_user(sk)) {
324 			DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
325 			sk->sk_err = err;
326 
327 			sk->sk_error_report(sk);
328 
329 			dccp_done(sk);
330 		} else
331 			sk->sk_err_soft = err;
332 		goto out;
333 	}
334 
335 	/* If we've already connected we will keep trying
336 	 * until we time out, or the user gives up.
337 	 *
338 	 * rfc1122 4.2.3.9 allows to consider as hard errors
339 	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
340 	 * but it is obsoleted by pmtu discovery).
341 	 *
342 	 * Note, that in modern internet, where routing is unreliable
343 	 * and in each dark corner broken firewalls sit, sending random
344 	 * errors ordered by their masters even this two messages finally lose
345 	 * their original sense (even Linux sends invalid PORT_UNREACHs)
346 	 *
347 	 * Now we are in compliance with RFCs.
348 	 *							--ANK (980905)
349 	 */
350 
351 	inet = inet_sk(sk);
352 	if (!sock_owned_by_user(sk) && inet->recverr) {
353 		sk->sk_err = err;
354 		sk->sk_error_report(sk);
355 	} else /* Only an error on timeout */
356 		sk->sk_err_soft = err;
357 out:
358 	bh_unlock_sock(sk);
359 	sock_put(sk);
360 }
361 
362 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
363 				      __be32 src, __be32 dst)
364 {
365 	return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
366 }
367 
368 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
369 {
370 	const struct inet_sock *inet = inet_sk(sk);
371 	struct dccp_hdr *dh = dccp_hdr(skb);
372 
373 	dccp_csum_outgoing(skb);
374 	dh->dccph_checksum = dccp_v4_csum_finish(skb,
375 						 inet->inet_saddr,
376 						 inet->inet_daddr);
377 }
378 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
379 
380 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
381 {
382 	return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
383 					   ip_hdr(skb)->saddr,
384 					   dccp_hdr(skb)->dccph_dport,
385 					   dccp_hdr(skb)->dccph_sport);
386 }
387 
388 /*
389  * The three way handshake has completed - we got a valid ACK or DATAACK -
390  * now create the new socket.
391  *
392  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
393  */
394 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
395 				       struct request_sock *req,
396 				       struct dst_entry *dst)
397 {
398 	struct inet_request_sock *ireq;
399 	struct inet_sock *newinet;
400 	struct sock *newsk;
401 
402 	if (sk_acceptq_is_full(sk))
403 		goto exit_overflow;
404 
405 	newsk = dccp_create_openreq_child(sk, req, skb);
406 	if (newsk == NULL)
407 		goto exit_nonewsk;
408 
409 	newinet		   = inet_sk(newsk);
410 	ireq		   = inet_rsk(req);
411 	newinet->inet_daddr	= ireq->ir_rmt_addr;
412 	newinet->inet_rcv_saddr = ireq->ir_loc_addr;
413 	newinet->inet_saddr	= ireq->ir_loc_addr;
414 	newinet->inet_opt	= ireq->opt;
415 	ireq->opt	   = NULL;
416 	newinet->mc_index  = inet_iif(skb);
417 	newinet->mc_ttl	   = ip_hdr(skb)->ttl;
418 	newinet->inet_id   = jiffies;
419 
420 	if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
421 		goto put_and_exit;
422 
423 	sk_setup_caps(newsk, dst);
424 
425 	dccp_sync_mss(newsk, dst_mtu(dst));
426 
427 	if (__inet_inherit_port(sk, newsk) < 0)
428 		goto put_and_exit;
429 	__inet_hash_nolisten(newsk, NULL);
430 
431 	return newsk;
432 
433 exit_overflow:
434 	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
435 exit_nonewsk:
436 	dst_release(dst);
437 exit:
438 	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
439 	return NULL;
440 put_and_exit:
441 	inet_csk_prepare_forced_close(newsk);
442 	dccp_done(newsk);
443 	goto exit;
444 }
445 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
446 
447 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
448 {
449 	const struct dccp_hdr *dh = dccp_hdr(skb);
450 	const struct iphdr *iph = ip_hdr(skb);
451 	struct sock *nsk;
452 	struct request_sock **prev;
453 	/* Find possible connection requests. */
454 	struct request_sock *req = inet_csk_search_req(sk, &prev,
455 						       dh->dccph_sport,
456 						       iph->saddr, iph->daddr);
457 	if (req != NULL)
458 		return dccp_check_req(sk, skb, req, prev);
459 
460 	nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
461 				      iph->saddr, dh->dccph_sport,
462 				      iph->daddr, dh->dccph_dport,
463 				      inet_iif(skb));
464 	if (nsk != NULL) {
465 		if (nsk->sk_state != DCCP_TIME_WAIT) {
466 			bh_lock_sock(nsk);
467 			return nsk;
468 		}
469 		inet_twsk_put(inet_twsk(nsk));
470 		return NULL;
471 	}
472 
473 	return sk;
474 }
475 
476 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
477 					   struct sk_buff *skb)
478 {
479 	struct rtable *rt;
480 	const struct iphdr *iph = ip_hdr(skb);
481 	struct flowi4 fl4 = {
482 		.flowi4_oif = inet_iif(skb),
483 		.daddr = iph->saddr,
484 		.saddr = iph->daddr,
485 		.flowi4_tos = RT_CONN_FLAGS(sk),
486 		.flowi4_proto = sk->sk_protocol,
487 		.fl4_sport = dccp_hdr(skb)->dccph_dport,
488 		.fl4_dport = dccp_hdr(skb)->dccph_sport,
489 	};
490 
491 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
492 	rt = ip_route_output_flow(net, &fl4, sk);
493 	if (IS_ERR(rt)) {
494 		IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
495 		return NULL;
496 	}
497 
498 	return &rt->dst;
499 }
500 
501 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req)
502 {
503 	int err = -1;
504 	struct sk_buff *skb;
505 	struct dst_entry *dst;
506 	struct flowi4 fl4;
507 
508 	dst = inet_csk_route_req(sk, &fl4, req);
509 	if (dst == NULL)
510 		goto out;
511 
512 	skb = dccp_make_response(sk, dst, req);
513 	if (skb != NULL) {
514 		const struct inet_request_sock *ireq = inet_rsk(req);
515 		struct dccp_hdr *dh = dccp_hdr(skb);
516 
517 		dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
518 							      ireq->ir_rmt_addr);
519 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
520 					    ireq->ir_rmt_addr,
521 					    ireq->opt);
522 		err = net_xmit_eval(err);
523 	}
524 
525 out:
526 	dst_release(dst);
527 	return err;
528 }
529 
530 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
531 {
532 	int err;
533 	const struct iphdr *rxiph;
534 	struct sk_buff *skb;
535 	struct dst_entry *dst;
536 	struct net *net = dev_net(skb_dst(rxskb)->dev);
537 	struct sock *ctl_sk = net->dccp.v4_ctl_sk;
538 
539 	/* Never send a reset in response to a reset. */
540 	if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
541 		return;
542 
543 	if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
544 		return;
545 
546 	dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
547 	if (dst == NULL)
548 		return;
549 
550 	skb = dccp_ctl_make_reset(ctl_sk, rxskb);
551 	if (skb == NULL)
552 		goto out;
553 
554 	rxiph = ip_hdr(rxskb);
555 	dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
556 								 rxiph->daddr);
557 	skb_dst_set(skb, dst_clone(dst));
558 
559 	bh_lock_sock(ctl_sk);
560 	err = ip_build_and_send_pkt(skb, ctl_sk,
561 				    rxiph->daddr, rxiph->saddr, NULL);
562 	bh_unlock_sock(ctl_sk);
563 
564 	if (net_xmit_eval(err) == 0) {
565 		DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
566 		DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
567 	}
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(inet_rsk(req)->opt);
576 }
577 
578 void dccp_syn_ack_timeout(struct sock *sk, 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 	/*
619 	 * Accept backlog is full. If we have already queued enough
620 	 * of warm entries in syn queue, drop request. It is better than
621 	 * clogging syn queue with openreqs with exponentially increasing
622 	 * timeout.
623 	 */
624 	if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
625 		goto drop;
626 
627 	req = inet_reqsk_alloc(&dccp_request_sock_ops);
628 	if (req == NULL)
629 		goto drop;
630 
631 	if (dccp_reqsk_init(req, dccp_sk(sk), skb))
632 		goto drop_and_free;
633 
634 	dreq = dccp_rsk(req);
635 	if (dccp_parse_options(sk, dreq, skb))
636 		goto drop_and_free;
637 
638 	if (security_inet_conn_request(sk, skb, req))
639 		goto drop_and_free;
640 
641 	ireq = inet_rsk(req);
642 	ireq->ir_loc_addr = ip_hdr(skb)->daddr;
643 	ireq->ir_rmt_addr = ip_hdr(skb)->saddr;
644 
645 	/*
646 	 * Step 3: Process LISTEN state
647 	 *
648 	 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
649 	 *
650 	 * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
651 	 */
652 	dreq->dreq_isr	   = dcb->dccpd_seq;
653 	dreq->dreq_gsr	   = dreq->dreq_isr;
654 	dreq->dreq_iss	   = dccp_v4_init_sequence(skb);
655 	dreq->dreq_gss     = dreq->dreq_iss;
656 	dreq->dreq_service = service;
657 
658 	if (dccp_v4_send_response(sk, req))
659 		goto drop_and_free;
660 
661 	inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
662 	return 0;
663 
664 drop_and_free:
665 	reqsk_free(req);
666 drop:
667 	DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
668 	return -1;
669 }
670 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
671 
672 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
673 {
674 	struct dccp_hdr *dh = dccp_hdr(skb);
675 
676 	if (sk->sk_state == DCCP_OPEN) { /* Fast path */
677 		if (dccp_rcv_established(sk, skb, dh, skb->len))
678 			goto reset;
679 		return 0;
680 	}
681 
682 	/*
683 	 *  Step 3: Process LISTEN state
684 	 *	 If P.type == Request or P contains a valid Init Cookie option,
685 	 *	      (* Must scan the packet's options to check for Init
686 	 *		 Cookies.  Only Init Cookies are processed here,
687 	 *		 however; other options are processed in Step 8.  This
688 	 *		 scan need only be performed if the endpoint uses Init
689 	 *		 Cookies *)
690 	 *	      (* Generate a new socket and switch to that socket *)
691 	 *	      Set S := new socket for this port pair
692 	 *	      S.state = RESPOND
693 	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
694 	 *	      Initialize S.GAR := S.ISS
695 	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
696 	 *	      Continue with S.state == RESPOND
697 	 *	      (* A Response packet will be generated in Step 11 *)
698 	 *	 Otherwise,
699 	 *	      Generate Reset(No Connection) unless P.type == Reset
700 	 *	      Drop packet and return
701 	 *
702 	 * NOTE: the check for the packet types is done in
703 	 *	 dccp_rcv_state_process
704 	 */
705 	if (sk->sk_state == DCCP_LISTEN) {
706 		struct sock *nsk = dccp_v4_hnd_req(sk, skb);
707 
708 		if (nsk == NULL)
709 			goto discard;
710 
711 		if (nsk != sk) {
712 			if (dccp_child_process(sk, nsk, skb))
713 				goto reset;
714 			return 0;
715 		}
716 	}
717 
718 	if (dccp_rcv_state_process(sk, skb, dh, skb->len))
719 		goto reset;
720 	return 0;
721 
722 reset:
723 	dccp_v4_ctl_send_reset(sk, skb);
724 discard:
725 	kfree_skb(skb);
726 	return 0;
727 }
728 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
729 
730 /**
731  *	dccp_invalid_packet  -  check for malformed packets
732  *	Implements RFC 4340, 8.5:  Step 1: Check header basics
733  *	Packets that fail these checks are ignored and do not receive Resets.
734  */
735 int dccp_invalid_packet(struct sk_buff *skb)
736 {
737 	const struct dccp_hdr *dh;
738 	unsigned int cscov;
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 	if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
761 		DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
762 		return 1;
763 	}
764 	/*
765 	 * If P.Data Offset is too too large for packet, drop packet and return
766 	 */
767 	if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
768 		DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
769 		return 1;
770 	}
771 
772 	/*
773 	 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
774 	 * has short sequence numbers), drop packet and return
775 	 */
776 	if ((dh->dccph_type < DCCP_PKT_DATA    ||
777 	    dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
778 		DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
779 			  dccp_packet_name(dh->dccph_type));
780 		return 1;
781 	}
782 
783 	/*
784 	 * If P.CsCov is too large for the packet size, drop packet and return.
785 	 * This must come _before_ checksumming (not as RFC 4340 suggests).
786 	 */
787 	cscov = dccp_csum_coverage(skb);
788 	if (cscov > skb->len) {
789 		DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
790 			  dh->dccph_cscov, skb->len);
791 		return 1;
792 	}
793 
794 	/* If header checksum is incorrect, drop packet and return.
795 	 * (This step is completed in the AF-dependent functions.) */
796 	skb->csum = skb_checksum(skb, 0, cscov, 0);
797 
798 	return 0;
799 }
800 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
801 
802 /* this is called when real data arrives */
803 static int dccp_v4_rcv(struct sk_buff *skb)
804 {
805 	const struct dccp_hdr *dh;
806 	const struct iphdr *iph;
807 	struct sock *sk;
808 	int min_cov;
809 
810 	/* Step 1: Check header basics */
811 
812 	if (dccp_invalid_packet(skb))
813 		goto discard_it;
814 
815 	iph = ip_hdr(skb);
816 	/* Step 1: If header checksum is incorrect, drop packet and return */
817 	if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
818 		DCCP_WARN("dropped packet with invalid checksum\n");
819 		goto discard_it;
820 	}
821 
822 	dh = dccp_hdr(skb);
823 
824 	DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
825 	DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
826 
827 	dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
828 		      dccp_packet_name(dh->dccph_type),
829 		      &iph->saddr, ntohs(dh->dccph_sport),
830 		      &iph->daddr, ntohs(dh->dccph_dport),
831 		      (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
832 
833 	if (dccp_packet_without_ack(skb)) {
834 		DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
835 		dccp_pr_debug_cat("\n");
836 	} else {
837 		DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
838 		dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
839 				  DCCP_SKB_CB(skb)->dccpd_ack_seq);
840 	}
841 
842 	/* Step 2:
843 	 *	Look up flow ID in table and get corresponding socket */
844 	sk = __inet_lookup_skb(&dccp_hashinfo, skb,
845 			       dh->dccph_sport, dh->dccph_dport);
846 	/*
847 	 * Step 2:
848 	 *	If no socket ...
849 	 */
850 	if (sk == NULL) {
851 		dccp_pr_debug("failed to look up flow ID in table and "
852 			      "get corresponding socket\n");
853 		goto no_dccp_socket;
854 	}
855 
856 	/*
857 	 * Step 2:
858 	 *	... or S.state == TIMEWAIT,
859 	 *		Generate Reset(No Connection) unless P.type == Reset
860 	 *		Drop packet and return
861 	 */
862 	if (sk->sk_state == DCCP_TIME_WAIT) {
863 		dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
864 		inet_twsk_put(inet_twsk(sk));
865 		goto no_dccp_socket;
866 	}
867 
868 	/*
869 	 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
870 	 *	o if MinCsCov = 0, only packets with CsCov = 0 are accepted
871 	 *	o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
872 	 */
873 	min_cov = dccp_sk(sk)->dccps_pcrlen;
874 	if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
875 		dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
876 			      dh->dccph_cscov, min_cov);
877 		/* FIXME: "Such packets SHOULD be reported using Data Dropped
878 		 *         options (Section 11.7) with Drop Code 0, Protocol
879 		 *         Constraints."                                     */
880 		goto discard_and_relse;
881 	}
882 
883 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
884 		goto discard_and_relse;
885 	nf_reset(skb);
886 
887 	return sk_receive_skb(sk, skb, 1);
888 
889 no_dccp_socket:
890 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
891 		goto discard_it;
892 	/*
893 	 * Step 2:
894 	 *	If no socket ...
895 	 *		Generate Reset(No Connection) unless P.type == Reset
896 	 *		Drop packet and return
897 	 */
898 	if (dh->dccph_type != DCCP_PKT_RESET) {
899 		DCCP_SKB_CB(skb)->dccpd_reset_code =
900 					DCCP_RESET_CODE_NO_CONNECTION;
901 		dccp_v4_ctl_send_reset(sk, skb);
902 	}
903 
904 discard_it:
905 	kfree_skb(skb);
906 	return 0;
907 
908 discard_and_relse:
909 	sock_put(sk);
910 	goto discard_it;
911 }
912 
913 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
914 	.queue_xmit	   = ip_queue_xmit,
915 	.send_check	   = dccp_v4_send_check,
916 	.rebuild_header	   = inet_sk_rebuild_header,
917 	.conn_request	   = dccp_v4_conn_request,
918 	.syn_recv_sock	   = dccp_v4_request_recv_sock,
919 	.net_header_len	   = sizeof(struct iphdr),
920 	.setsockopt	   = ip_setsockopt,
921 	.getsockopt	   = ip_getsockopt,
922 	.addr2sockaddr	   = inet_csk_addr2sockaddr,
923 	.sockaddr_len	   = sizeof(struct sockaddr_in),
924 	.bind_conflict	   = inet_csk_bind_conflict,
925 #ifdef CONFIG_COMPAT
926 	.compat_setsockopt = compat_ip_setsockopt,
927 	.compat_getsockopt = compat_ip_getsockopt,
928 #endif
929 };
930 
931 static int dccp_v4_init_sock(struct sock *sk)
932 {
933 	static __u8 dccp_v4_ctl_sock_initialized;
934 	int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
935 
936 	if (err == 0) {
937 		if (unlikely(!dccp_v4_ctl_sock_initialized))
938 			dccp_v4_ctl_sock_initialized = 1;
939 		inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
940 	}
941 
942 	return err;
943 }
944 
945 static struct timewait_sock_ops dccp_timewait_sock_ops = {
946 	.twsk_obj_size	= sizeof(struct inet_timewait_sock),
947 };
948 
949 static struct proto dccp_v4_prot = {
950 	.name			= "DCCP",
951 	.owner			= THIS_MODULE,
952 	.close			= dccp_close,
953 	.connect		= dccp_v4_connect,
954 	.disconnect		= dccp_disconnect,
955 	.ioctl			= dccp_ioctl,
956 	.init			= dccp_v4_init_sock,
957 	.setsockopt		= dccp_setsockopt,
958 	.getsockopt		= dccp_getsockopt,
959 	.sendmsg		= dccp_sendmsg,
960 	.recvmsg		= dccp_recvmsg,
961 	.backlog_rcv		= dccp_v4_do_rcv,
962 	.hash			= inet_hash,
963 	.unhash			= inet_unhash,
964 	.accept			= inet_csk_accept,
965 	.get_port		= inet_csk_get_port,
966 	.shutdown		= dccp_shutdown,
967 	.destroy		= dccp_destroy_sock,
968 	.orphan_count		= &dccp_orphan_count,
969 	.max_header		= MAX_DCCP_HEADER,
970 	.obj_size		= sizeof(struct dccp_sock),
971 	.slab_flags		= SLAB_DESTROY_BY_RCU,
972 	.rsk_prot		= &dccp_request_sock_ops,
973 	.twsk_prot		= &dccp_timewait_sock_ops,
974 	.h.hashinfo		= &dccp_hashinfo,
975 #ifdef CONFIG_COMPAT
976 	.compat_setsockopt	= compat_dccp_setsockopt,
977 	.compat_getsockopt	= compat_dccp_getsockopt,
978 #endif
979 };
980 
981 static const struct net_protocol dccp_v4_protocol = {
982 	.handler	= dccp_v4_rcv,
983 	.err_handler	= dccp_v4_err,
984 	.no_policy	= 1,
985 	.netns_ok	= 1,
986 	.icmp_strict_tag_validation = 1,
987 };
988 
989 static const struct proto_ops inet_dccp_ops = {
990 	.family		   = PF_INET,
991 	.owner		   = THIS_MODULE,
992 	.release	   = inet_release,
993 	.bind		   = inet_bind,
994 	.connect	   = inet_stream_connect,
995 	.socketpair	   = sock_no_socketpair,
996 	.accept		   = inet_accept,
997 	.getname	   = inet_getname,
998 	/* FIXME: work on tcp_poll to rename it to inet_csk_poll */
999 	.poll		   = dccp_poll,
1000 	.ioctl		   = inet_ioctl,
1001 	/* FIXME: work on inet_listen to rename it to sock_common_listen */
1002 	.listen		   = inet_dccp_listen,
1003 	.shutdown	   = inet_shutdown,
1004 	.setsockopt	   = sock_common_setsockopt,
1005 	.getsockopt	   = sock_common_getsockopt,
1006 	.sendmsg	   = inet_sendmsg,
1007 	.recvmsg	   = sock_common_recvmsg,
1008 	.mmap		   = sock_no_mmap,
1009 	.sendpage	   = sock_no_sendpage,
1010 #ifdef CONFIG_COMPAT
1011 	.compat_setsockopt = compat_sock_common_setsockopt,
1012 	.compat_getsockopt = compat_sock_common_getsockopt,
1013 #endif
1014 };
1015 
1016 static struct inet_protosw dccp_v4_protosw = {
1017 	.type		= SOCK_DCCP,
1018 	.protocol	= IPPROTO_DCCP,
1019 	.prot		= &dccp_v4_prot,
1020 	.ops		= &inet_dccp_ops,
1021 	.flags		= INET_PROTOSW_ICSK,
1022 };
1023 
1024 static int __net_init dccp_v4_init_net(struct net *net)
1025 {
1026 	if (dccp_hashinfo.bhash == NULL)
1027 		return -ESOCKTNOSUPPORT;
1028 
1029 	return inet_ctl_sock_create(&net->dccp.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 	inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1036 }
1037 
1038 static struct pernet_operations dccp_v4_ops = {
1039 	.init	= dccp_v4_init_net,
1040 	.exit	= dccp_v4_exit_net,
1041 };
1042 
1043 static int __init dccp_v4_init(void)
1044 {
1045 	int err = proto_register(&dccp_v4_prot, 1);
1046 
1047 	if (err != 0)
1048 		goto out;
1049 
1050 	err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1051 	if (err != 0)
1052 		goto out_proto_unregister;
1053 
1054 	inet_register_protosw(&dccp_v4_protosw);
1055 
1056 	err = register_pernet_subsys(&dccp_v4_ops);
1057 	if (err)
1058 		goto out_destroy_ctl_sock;
1059 out:
1060 	return err;
1061 out_destroy_ctl_sock:
1062 	inet_unregister_protosw(&dccp_v4_protosw);
1063 	inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1064 out_proto_unregister:
1065 	proto_unregister(&dccp_v4_prot);
1066 	goto out;
1067 }
1068 
1069 static void __exit dccp_v4_exit(void)
1070 {
1071 	unregister_pernet_subsys(&dccp_v4_ops);
1072 	inet_unregister_protosw(&dccp_v4_protosw);
1073 	inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1074 	proto_unregister(&dccp_v4_prot);
1075 }
1076 
1077 module_init(dccp_v4_init);
1078 module_exit(dccp_v4_exit);
1079 
1080 /*
1081  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1082  * values directly, Also cover the case where the protocol is not specified,
1083  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1084  */
1085 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1086 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1087 MODULE_LICENSE("GPL");
1088 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1089 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1090