xref: /kernel/linux/linux-5.10/net/dccp/ipv4.c (revision 8c2ecf20)
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
27#include "ackvec.h"
28#include "ccid.h"
29#include "dccp.h"
30#include "feat.h"
31
32/*
33 * The per-net dccp.v4_ctl_sk socket is used for responding to
34 * the Out-of-the-blue (OOTB) packets. A control sock will be created
35 * for this socket at the initialization time.
36 */
37
38int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
39{
40	const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
41	struct inet_sock *inet = inet_sk(sk);
42	struct dccp_sock *dp = dccp_sk(sk);
43	__be16 orig_sport, orig_dport;
44	__be32 daddr, nexthop;
45	struct flowi4 *fl4;
46	struct rtable *rt;
47	int err;
48	struct ip_options_rcu *inet_opt;
49
50	dp->dccps_role = DCCP_ROLE_CLIENT;
51
52	if (addr_len < sizeof(struct sockaddr_in))
53		return -EINVAL;
54
55	if (usin->sin_family != AF_INET)
56		return -EAFNOSUPPORT;
57
58	nexthop = daddr = usin->sin_addr.s_addr;
59
60	inet_opt = rcu_dereference_protected(inet->inet_opt,
61					     lockdep_sock_is_held(sk));
62	if (inet_opt != NULL && inet_opt->opt.srr) {
63		if (daddr == 0)
64			return -EINVAL;
65		nexthop = inet_opt->opt.faddr;
66	}
67
68	orig_sport = inet->inet_sport;
69	orig_dport = usin->sin_port;
70	fl4 = &inet->cork.fl.u.ip4;
71	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
72			      RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
73			      IPPROTO_DCCP,
74			      orig_sport, orig_dport, sk);
75	if (IS_ERR(rt))
76		return PTR_ERR(rt);
77
78	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
79		ip_rt_put(rt);
80		return -ENETUNREACH;
81	}
82
83	if (inet_opt == NULL || !inet_opt->opt.srr)
84		daddr = fl4->daddr;
85
86	if (inet->inet_saddr == 0)
87		inet->inet_saddr = fl4->saddr;
88	sk_rcv_saddr_set(sk, inet->inet_saddr);
89	inet->inet_dport = usin->sin_port;
90	sk_daddr_set(sk, daddr);
91
92	inet_csk(sk)->icsk_ext_hdr_len = 0;
93	if (inet_opt)
94		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
95	/*
96	 * Socket identity is still unknown (sport may be zero).
97	 * However we set state to DCCP_REQUESTING and not releasing socket
98	 * lock select source port, enter ourselves into the hash tables and
99	 * complete initialization after this.
100	 */
101	dccp_set_state(sk, DCCP_REQUESTING);
102	err = inet_hash_connect(&dccp_death_row, sk);
103	if (err != 0)
104		goto failure;
105
106	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
107			       inet->inet_sport, inet->inet_dport, sk);
108	if (IS_ERR(rt)) {
109		err = PTR_ERR(rt);
110		rt = NULL;
111		goto failure;
112	}
113	/* OK, now commit destination to socket.  */
114	sk_setup_caps(sk, &rt->dst);
115
116	dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
117						    inet->inet_daddr,
118						    inet->inet_sport,
119						    inet->inet_dport);
120	inet->inet_id = prandom_u32();
121
122	err = dccp_connect(sk);
123	rt = NULL;
124	if (err != 0)
125		goto failure;
126out:
127	return err;
128failure:
129	/*
130	 * This unhashes the socket and releases the local port, if necessary.
131	 */
132	dccp_set_state(sk, DCCP_CLOSED);
133	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
134		inet_reset_saddr(sk);
135	ip_rt_put(rt);
136	sk->sk_route_caps = 0;
137	inet->inet_dport = 0;
138	goto out;
139}
140EXPORT_SYMBOL_GPL(dccp_v4_connect);
141
142/*
143 * This routine does path mtu discovery as defined in RFC1191.
144 */
145static inline void dccp_do_pmtu_discovery(struct sock *sk,
146					  const struct iphdr *iph,
147					  u32 mtu)
148{
149	struct dst_entry *dst;
150	const struct inet_sock *inet = inet_sk(sk);
151	const struct dccp_sock *dp = dccp_sk(sk);
152
153	/* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
154	 * send out by Linux are always < 576bytes so they should go through
155	 * unfragmented).
156	 */
157	if (sk->sk_state == DCCP_LISTEN)
158		return;
159
160	dst = inet_csk_update_pmtu(sk, mtu);
161	if (!dst)
162		return;
163
164	/* Something is about to be wrong... Remember soft error
165	 * for the case, if this connection will not able to recover.
166	 */
167	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
168		sk->sk_err_soft = EMSGSIZE;
169
170	mtu = dst_mtu(dst);
171
172	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
173	    ip_sk_accept_pmtu(sk) &&
174	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
175		dccp_sync_mss(sk, mtu);
176
177		/*
178		 * From RFC 4340, sec. 14.1:
179		 *
180		 *	DCCP-Sync packets are the best choice for upward
181		 *	probing, since DCCP-Sync probes do not risk application
182		 *	data loss.
183		 */
184		dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
185	} /* else let the usual retransmit timer handle it */
186}
187
188static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
189{
190	struct dst_entry *dst = __sk_dst_check(sk, 0);
191
192	if (dst)
193		dst->ops->redirect(dst, sk, skb);
194}
195
196void dccp_req_err(struct sock *sk, u64 seq)
197	{
198	struct request_sock *req = inet_reqsk(sk);
199	struct net *net = sock_net(sk);
200
201	/*
202	 * ICMPs are not backlogged, hence we cannot get an established
203	 * socket here.
204	 */
205	if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
206		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
207	} else {
208		/*
209		 * Still in RESPOND, just remove it silently.
210		 * There is no good way to pass the error to the newly
211		 * created socket, and POSIX does not want network
212		 * errors returned from accept().
213		 */
214		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
215	}
216	reqsk_put(req);
217}
218EXPORT_SYMBOL(dccp_req_err);
219
220/*
221 * This routine is called by the ICMP module when it gets some sort of error
222 * condition. If err < 0 then the socket should be closed and the error
223 * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
224 * After adjustment header points to the first 8 bytes of the tcp header. We
225 * need to find the appropriate port.
226 *
227 * The locking strategy used here is very "optimistic". When someone else
228 * accesses the socket the ICMP is just dropped and for some paths there is no
229 * check at all. A more general error queue to queue errors for later handling
230 * is probably better.
231 */
232static int dccp_v4_err(struct sk_buff *skb, u32 info)
233{
234	const struct iphdr *iph = (struct iphdr *)skb->data;
235	const u8 offset = iph->ihl << 2;
236	const struct dccp_hdr *dh;
237	struct dccp_sock *dp;
238	struct inet_sock *inet;
239	const int type = icmp_hdr(skb)->type;
240	const int code = icmp_hdr(skb)->code;
241	struct sock *sk;
242	__u64 seq;
243	int err;
244	struct net *net = dev_net(skb->dev);
245
246	if (!pskb_may_pull(skb, offset + sizeof(*dh)))
247		return -EINVAL;
248	dh = (struct dccp_hdr *)(skb->data + offset);
249	if (!pskb_may_pull(skb, offset + __dccp_basic_hdr_len(dh)))
250		return -EINVAL;
251	iph = (struct iphdr *)skb->data;
252	dh = (struct dccp_hdr *)(skb->data + offset);
253
254	sk = __inet_lookup_established(net, &dccp_hashinfo,
255				       iph->daddr, dh->dccph_dport,
256				       iph->saddr, ntohs(dh->dccph_sport),
257				       inet_iif(skb), 0);
258	if (!sk) {
259		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
260		return -ENOENT;
261	}
262
263	if (sk->sk_state == DCCP_TIME_WAIT) {
264		inet_twsk_put(inet_twsk(sk));
265		return 0;
266	}
267	seq = dccp_hdr_seq(dh);
268	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
269		dccp_req_err(sk, seq);
270		return 0;
271	}
272
273	bh_lock_sock(sk);
274	/* If too many ICMPs get dropped on busy
275	 * servers this needs to be solved differently.
276	 */
277	if (sock_owned_by_user(sk))
278		__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
279
280	if (sk->sk_state == DCCP_CLOSED)
281		goto out;
282
283	dp = dccp_sk(sk);
284	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
285	    !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
286		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
287		goto out;
288	}
289
290	switch (type) {
291	case ICMP_REDIRECT:
292		if (!sock_owned_by_user(sk))
293			dccp_do_redirect(skb, sk);
294		goto out;
295	case ICMP_SOURCE_QUENCH:
296		/* Just silently ignore these. */
297		goto out;
298	case ICMP_PARAMETERPROB:
299		err = EPROTO;
300		break;
301	case ICMP_DEST_UNREACH:
302		if (code > NR_ICMP_UNREACH)
303			goto out;
304
305		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
306			if (!sock_owned_by_user(sk))
307				dccp_do_pmtu_discovery(sk, iph, info);
308			goto out;
309		}
310
311		err = icmp_err_convert[code].errno;
312		break;
313	case ICMP_TIME_EXCEEDED:
314		err = EHOSTUNREACH;
315		break;
316	default:
317		goto out;
318	}
319
320	switch (sk->sk_state) {
321	case DCCP_REQUESTING:
322	case DCCP_RESPOND:
323		if (!sock_owned_by_user(sk)) {
324			__DCCP_INC_STATS(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;
357out:
358	bh_unlock_sock(sk);
359	sock_put(sk);
360	return 0;
361}
362
363static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
364				      __be32 src, __be32 dst)
365{
366	return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
367}
368
369void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
370{
371	const struct inet_sock *inet = inet_sk(sk);
372	struct dccp_hdr *dh = dccp_hdr(skb);
373
374	dccp_csum_outgoing(skb);
375	dh->dccph_checksum = dccp_v4_csum_finish(skb,
376						 inet->inet_saddr,
377						 inet->inet_daddr);
378}
379EXPORT_SYMBOL_GPL(dccp_v4_send_check);
380
381static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
382{
383	return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
384					   ip_hdr(skb)->saddr,
385					   dccp_hdr(skb)->dccph_dport,
386					   dccp_hdr(skb)->dccph_sport);
387}
388
389/*
390 * The three way handshake has completed - we got a valid ACK or DATAACK -
391 * now create the new socket.
392 *
393 * This is the equivalent of TCP's tcp_v4_syn_recv_sock
394 */
395struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
396				       struct sk_buff *skb,
397				       struct request_sock *req,
398				       struct dst_entry *dst,
399				       struct request_sock *req_unhash,
400				       bool *own_req)
401{
402	struct inet_request_sock *ireq;
403	struct inet_sock *newinet;
404	struct sock *newsk;
405
406	if (sk_acceptq_is_full(sk))
407		goto exit_overflow;
408
409	newsk = dccp_create_openreq_child(sk, req, skb);
410	if (newsk == NULL)
411		goto exit_nonewsk;
412
413	newinet		   = inet_sk(newsk);
414	ireq		   = inet_rsk(req);
415	sk_daddr_set(newsk, ireq->ir_rmt_addr);
416	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
417	newinet->inet_saddr	= ireq->ir_loc_addr;
418	RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
419	newinet->mc_index  = inet_iif(skb);
420	newinet->mc_ttl	   = ip_hdr(skb)->ttl;
421	newinet->inet_id   = prandom_u32();
422
423	if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
424		goto put_and_exit;
425
426	sk_setup_caps(newsk, dst);
427
428	dccp_sync_mss(newsk, dst_mtu(dst));
429
430	if (__inet_inherit_port(sk, newsk) < 0)
431		goto put_and_exit;
432	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
433	if (*own_req)
434		ireq->ireq_opt = NULL;
435	else
436		newinet->inet_opt = NULL;
437	return newsk;
438
439exit_overflow:
440	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
441exit_nonewsk:
442	dst_release(dst);
443exit:
444	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
445	return NULL;
446put_and_exit:
447	newinet->inet_opt = NULL;
448	inet_csk_prepare_forced_close(newsk);
449	dccp_done(newsk);
450	goto exit;
451}
452EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
453
454static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
455					   struct sk_buff *skb)
456{
457	struct rtable *rt;
458	const struct iphdr *iph = ip_hdr(skb);
459	struct flowi4 fl4 = {
460		.flowi4_oif = inet_iif(skb),
461		.daddr = iph->saddr,
462		.saddr = iph->daddr,
463		.flowi4_tos = RT_CONN_FLAGS(sk),
464		.flowi4_proto = sk->sk_protocol,
465		.fl4_sport = dccp_hdr(skb)->dccph_dport,
466		.fl4_dport = dccp_hdr(skb)->dccph_sport,
467	};
468
469	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
470	rt = ip_route_output_flow(net, &fl4, sk);
471	if (IS_ERR(rt)) {
472		IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
473		return NULL;
474	}
475
476	return &rt->dst;
477}
478
479static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
480{
481	int err = -1;
482	struct sk_buff *skb;
483	struct dst_entry *dst;
484	struct flowi4 fl4;
485
486	dst = inet_csk_route_req(sk, &fl4, req);
487	if (dst == NULL)
488		goto out;
489
490	skb = dccp_make_response(sk, dst, req);
491	if (skb != NULL) {
492		const struct inet_request_sock *ireq = inet_rsk(req);
493		struct dccp_hdr *dh = dccp_hdr(skb);
494
495		dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
496							      ireq->ir_rmt_addr);
497		rcu_read_lock();
498		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
499					    ireq->ir_rmt_addr,
500					    rcu_dereference(ireq->ireq_opt),
501					    inet_sk(sk)->tos);
502		rcu_read_unlock();
503		err = net_xmit_eval(err);
504	}
505
506out:
507	dst_release(dst);
508	return err;
509}
510
511static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
512{
513	int err;
514	const struct iphdr *rxiph;
515	struct sk_buff *skb;
516	struct dst_entry *dst;
517	struct net *net = dev_net(skb_dst(rxskb)->dev);
518	struct sock *ctl_sk = net->dccp.v4_ctl_sk;
519
520	/* Never send a reset in response to a reset. */
521	if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
522		return;
523
524	if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
525		return;
526
527	dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
528	if (dst == NULL)
529		return;
530
531	skb = dccp_ctl_make_reset(ctl_sk, rxskb);
532	if (skb == NULL)
533		goto out;
534
535	rxiph = ip_hdr(rxskb);
536	dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
537								 rxiph->daddr);
538	skb_dst_set(skb, dst_clone(dst));
539
540	local_bh_disable();
541	bh_lock_sock(ctl_sk);
542	err = ip_build_and_send_pkt(skb, ctl_sk,
543				    rxiph->daddr, rxiph->saddr, NULL,
544				    inet_sk(ctl_sk)->tos);
545	bh_unlock_sock(ctl_sk);
546
547	if (net_xmit_eval(err) == 0) {
548		__DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
549		__DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
550	}
551	local_bh_enable();
552out:
553	dst_release(dst);
554}
555
556static void dccp_v4_reqsk_destructor(struct request_sock *req)
557{
558	dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
559	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
560}
561
562void dccp_syn_ack_timeout(const struct request_sock *req)
563{
564}
565EXPORT_SYMBOL(dccp_syn_ack_timeout);
566
567static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
568	.family		= PF_INET,
569	.obj_size	= sizeof(struct dccp_request_sock),
570	.rtx_syn_ack	= dccp_v4_send_response,
571	.send_ack	= dccp_reqsk_send_ack,
572	.destructor	= dccp_v4_reqsk_destructor,
573	.send_reset	= dccp_v4_ctl_send_reset,
574	.syn_ack_timeout = dccp_syn_ack_timeout,
575};
576
577int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
578{
579	struct inet_request_sock *ireq;
580	struct request_sock *req;
581	struct dccp_request_sock *dreq;
582	const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
583	struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
584
585	/* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
586	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
587		return 0;	/* discard, don't send a reset here */
588
589	if (dccp_bad_service_code(sk, service)) {
590		dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
591		goto drop;
592	}
593	/*
594	 * TW buckets are converted to open requests without
595	 * limitations, they conserve resources and peer is
596	 * evidently real one.
597	 */
598	dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
599	if (inet_csk_reqsk_queue_is_full(sk))
600		goto drop;
601
602	if (sk_acceptq_is_full(sk))
603		goto drop;
604
605	req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
606	if (req == NULL)
607		goto drop;
608
609	if (dccp_reqsk_init(req, dccp_sk(sk), skb))
610		goto drop_and_free;
611
612	dreq = dccp_rsk(req);
613	if (dccp_parse_options(sk, dreq, skb))
614		goto drop_and_free;
615
616	ireq = inet_rsk(req);
617	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
618	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
619	ireq->ir_mark = inet_request_mark(sk, skb);
620	ireq->ireq_family = AF_INET;
621	ireq->ir_iif = sk->sk_bound_dev_if;
622
623	if (security_inet_conn_request(sk, skb, req))
624		goto drop_and_free;
625
626	/*
627	 * Step 3: Process LISTEN state
628	 *
629	 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
630	 *
631	 * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
632	 */
633	dreq->dreq_isr	   = dcb->dccpd_seq;
634	dreq->dreq_gsr	   = dreq->dreq_isr;
635	dreq->dreq_iss	   = dccp_v4_init_sequence(skb);
636	dreq->dreq_gss     = dreq->dreq_iss;
637	dreq->dreq_service = service;
638
639	if (dccp_v4_send_response(sk, req))
640		goto drop_and_free;
641
642	inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
643	reqsk_put(req);
644	return 0;
645
646drop_and_free:
647	reqsk_free(req);
648drop:
649	__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
650	return -1;
651}
652EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
653
654int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
655{
656	struct dccp_hdr *dh = dccp_hdr(skb);
657
658	if (sk->sk_state == DCCP_OPEN) { /* Fast path */
659		if (dccp_rcv_established(sk, skb, dh, skb->len))
660			goto reset;
661		return 0;
662	}
663
664	/*
665	 *  Step 3: Process LISTEN state
666	 *	 If P.type == Request or P contains a valid Init Cookie option,
667	 *	      (* Must scan the packet's options to check for Init
668	 *		 Cookies.  Only Init Cookies are processed here,
669	 *		 however; other options are processed in Step 8.  This
670	 *		 scan need only be performed if the endpoint uses Init
671	 *		 Cookies *)
672	 *	      (* Generate a new socket and switch to that socket *)
673	 *	      Set S := new socket for this port pair
674	 *	      S.state = RESPOND
675	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
676	 *	      Initialize S.GAR := S.ISS
677	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
678	 *	      Continue with S.state == RESPOND
679	 *	      (* A Response packet will be generated in Step 11 *)
680	 *	 Otherwise,
681	 *	      Generate Reset(No Connection) unless P.type == Reset
682	 *	      Drop packet and return
683	 *
684	 * NOTE: the check for the packet types is done in
685	 *	 dccp_rcv_state_process
686	 */
687
688	if (dccp_rcv_state_process(sk, skb, dh, skb->len))
689		goto reset;
690	return 0;
691
692reset:
693	dccp_v4_ctl_send_reset(sk, skb);
694	kfree_skb(skb);
695	return 0;
696}
697EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
698
699/**
700 *	dccp_invalid_packet  -  check for malformed packets
701 *	@skb: Packet to validate
702 *
703 *	Implements RFC 4340, 8.5:  Step 1: Check header basics
704 *	Packets that fail these checks are ignored and do not receive Resets.
705 */
706int dccp_invalid_packet(struct sk_buff *skb)
707{
708	const struct dccp_hdr *dh;
709	unsigned int cscov;
710	u8 dccph_doff;
711
712	if (skb->pkt_type != PACKET_HOST)
713		return 1;
714
715	/* If the packet is shorter than 12 bytes, drop packet and return */
716	if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
717		DCCP_WARN("pskb_may_pull failed\n");
718		return 1;
719	}
720
721	dh = dccp_hdr(skb);
722
723	/* If P.type is not understood, drop packet and return */
724	if (dh->dccph_type >= DCCP_PKT_INVALID) {
725		DCCP_WARN("invalid packet type\n");
726		return 1;
727	}
728
729	/*
730	 * If P.Data Offset is too small for packet type, drop packet and return
731	 */
732	dccph_doff = dh->dccph_doff;
733	if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
734		DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
735		return 1;
736	}
737	/*
738	 * If P.Data Offset is too large for packet, drop packet and return
739	 */
740	if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
741		DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
742		return 1;
743	}
744	dh = dccp_hdr(skb);
745	/*
746	 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
747	 * has short sequence numbers), drop packet and return
748	 */
749	if ((dh->dccph_type < DCCP_PKT_DATA    ||
750	    dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
751		DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
752			  dccp_packet_name(dh->dccph_type));
753		return 1;
754	}
755
756	/*
757	 * If P.CsCov is too large for the packet size, drop packet and return.
758	 * This must come _before_ checksumming (not as RFC 4340 suggests).
759	 */
760	cscov = dccp_csum_coverage(skb);
761	if (cscov > skb->len) {
762		DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
763			  dh->dccph_cscov, skb->len);
764		return 1;
765	}
766
767	/* If header checksum is incorrect, drop packet and return.
768	 * (This step is completed in the AF-dependent functions.) */
769	skb->csum = skb_checksum(skb, 0, cscov, 0);
770
771	return 0;
772}
773EXPORT_SYMBOL_GPL(dccp_invalid_packet);
774
775/* this is called when real data arrives */
776static int dccp_v4_rcv(struct sk_buff *skb)
777{
778	const struct dccp_hdr *dh;
779	const struct iphdr *iph;
780	bool refcounted;
781	struct sock *sk;
782	int min_cov;
783
784	/* Step 1: Check header basics */
785
786	if (dccp_invalid_packet(skb))
787		goto discard_it;
788
789	iph = ip_hdr(skb);
790	/* Step 1: If header checksum is incorrect, drop packet and return */
791	if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
792		DCCP_WARN("dropped packet with invalid checksum\n");
793		goto discard_it;
794	}
795
796	dh = dccp_hdr(skb);
797
798	DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
799	DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
800
801	dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
802		      dccp_packet_name(dh->dccph_type),
803		      &iph->saddr, ntohs(dh->dccph_sport),
804		      &iph->daddr, ntohs(dh->dccph_dport),
805		      (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
806
807	if (dccp_packet_without_ack(skb)) {
808		DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
809		dccp_pr_debug_cat("\n");
810	} else {
811		DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
812		dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
813				  DCCP_SKB_CB(skb)->dccpd_ack_seq);
814	}
815
816lookup:
817	sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
818			       dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
819	if (!sk) {
820		dccp_pr_debug("failed to look up flow ID in table and "
821			      "get corresponding socket\n");
822		goto no_dccp_socket;
823	}
824
825	/*
826	 * Step 2:
827	 *	... or S.state == TIMEWAIT,
828	 *		Generate Reset(No Connection) unless P.type == Reset
829	 *		Drop packet and return
830	 */
831	if (sk->sk_state == DCCP_TIME_WAIT) {
832		dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
833		inet_twsk_put(inet_twsk(sk));
834		goto no_dccp_socket;
835	}
836
837	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
838		struct request_sock *req = inet_reqsk(sk);
839		struct sock *nsk;
840
841		sk = req->rsk_listener;
842		if (unlikely(sk->sk_state != DCCP_LISTEN)) {
843			inet_csk_reqsk_queue_drop_and_put(sk, req);
844			goto lookup;
845		}
846		sock_hold(sk);
847		refcounted = true;
848		nsk = dccp_check_req(sk, skb, req);
849		if (!nsk) {
850			reqsk_put(req);
851			goto discard_and_relse;
852		}
853		if (nsk == sk) {
854			reqsk_put(req);
855		} else if (dccp_child_process(sk, nsk, skb)) {
856			dccp_v4_ctl_send_reset(sk, skb);
857			goto discard_and_relse;
858		} else {
859			sock_put(sk);
860			return 0;
861		}
862	}
863	/*
864	 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
865	 *	o if MinCsCov = 0, only packets with CsCov = 0 are accepted
866	 *	o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
867	 */
868	min_cov = dccp_sk(sk)->dccps_pcrlen;
869	if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
870		dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
871			      dh->dccph_cscov, min_cov);
872		/* FIXME: "Such packets SHOULD be reported using Data Dropped
873		 *         options (Section 11.7) with Drop Code 0, Protocol
874		 *         Constraints."                                     */
875		goto discard_and_relse;
876	}
877
878	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
879		goto discard_and_relse;
880	nf_reset_ct(skb);
881
882	return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
883
884no_dccp_socket:
885	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
886		goto discard_it;
887	/*
888	 * Step 2:
889	 *	If no socket ...
890	 *		Generate Reset(No Connection) unless P.type == Reset
891	 *		Drop packet and return
892	 */
893	if (dh->dccph_type != DCCP_PKT_RESET) {
894		DCCP_SKB_CB(skb)->dccpd_reset_code =
895					DCCP_RESET_CODE_NO_CONNECTION;
896		dccp_v4_ctl_send_reset(sk, skb);
897	}
898
899discard_it:
900	kfree_skb(skb);
901	return 0;
902
903discard_and_relse:
904	if (refcounted)
905		sock_put(sk);
906	goto discard_it;
907}
908
909static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
910	.queue_xmit	   = ip_queue_xmit,
911	.send_check	   = dccp_v4_send_check,
912	.rebuild_header	   = inet_sk_rebuild_header,
913	.conn_request	   = dccp_v4_conn_request,
914	.syn_recv_sock	   = dccp_v4_request_recv_sock,
915	.net_header_len	   = sizeof(struct iphdr),
916	.setsockopt	   = ip_setsockopt,
917	.getsockopt	   = ip_getsockopt,
918	.addr2sockaddr	   = inet_csk_addr2sockaddr,
919	.sockaddr_len	   = sizeof(struct sockaddr_in),
920};
921
922static int dccp_v4_init_sock(struct sock *sk)
923{
924	static __u8 dccp_v4_ctl_sock_initialized;
925	int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
926
927	if (err == 0) {
928		if (unlikely(!dccp_v4_ctl_sock_initialized))
929			dccp_v4_ctl_sock_initialized = 1;
930		inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
931	}
932
933	return err;
934}
935
936static struct timewait_sock_ops dccp_timewait_sock_ops = {
937	.twsk_obj_size	= sizeof(struct inet_timewait_sock),
938};
939
940static struct proto dccp_v4_prot = {
941	.name			= "DCCP",
942	.owner			= THIS_MODULE,
943	.close			= dccp_close,
944	.connect		= dccp_v4_connect,
945	.disconnect		= dccp_disconnect,
946	.ioctl			= dccp_ioctl,
947	.init			= dccp_v4_init_sock,
948	.setsockopt		= dccp_setsockopt,
949	.getsockopt		= dccp_getsockopt,
950	.sendmsg		= dccp_sendmsg,
951	.recvmsg		= dccp_recvmsg,
952	.backlog_rcv		= dccp_v4_do_rcv,
953	.hash			= inet_hash,
954	.unhash			= inet_unhash,
955	.accept			= inet_csk_accept,
956	.get_port		= inet_csk_get_port,
957	.shutdown		= dccp_shutdown,
958	.destroy		= dccp_destroy_sock,
959	.orphan_count		= &dccp_orphan_count,
960	.max_header		= MAX_DCCP_HEADER,
961	.obj_size		= sizeof(struct dccp_sock),
962	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
963	.rsk_prot		= &dccp_request_sock_ops,
964	.twsk_prot		= &dccp_timewait_sock_ops,
965	.h.hashinfo		= &dccp_hashinfo,
966};
967
968static const struct net_protocol dccp_v4_protocol = {
969	.handler	= dccp_v4_rcv,
970	.err_handler	= dccp_v4_err,
971	.no_policy	= 1,
972	.netns_ok	= 1,
973	.icmp_strict_tag_validation = 1,
974};
975
976static const struct proto_ops inet_dccp_ops = {
977	.family		   = PF_INET,
978	.owner		   = THIS_MODULE,
979	.release	   = inet_release,
980	.bind		   = inet_bind,
981	.connect	   = inet_stream_connect,
982	.socketpair	   = sock_no_socketpair,
983	.accept		   = inet_accept,
984	.getname	   = inet_getname,
985	/* FIXME: work on tcp_poll to rename it to inet_csk_poll */
986	.poll		   = dccp_poll,
987	.ioctl		   = inet_ioctl,
988	.gettstamp	   = sock_gettstamp,
989	/* FIXME: work on inet_listen to rename it to sock_common_listen */
990	.listen		   = inet_dccp_listen,
991	.shutdown	   = inet_shutdown,
992	.setsockopt	   = sock_common_setsockopt,
993	.getsockopt	   = sock_common_getsockopt,
994	.sendmsg	   = inet_sendmsg,
995	.recvmsg	   = sock_common_recvmsg,
996	.mmap		   = sock_no_mmap,
997	.sendpage	   = sock_no_sendpage,
998};
999
1000static struct inet_protosw dccp_v4_protosw = {
1001	.type		= SOCK_DCCP,
1002	.protocol	= IPPROTO_DCCP,
1003	.prot		= &dccp_v4_prot,
1004	.ops		= &inet_dccp_ops,
1005	.flags		= INET_PROTOSW_ICSK,
1006};
1007
1008static int __net_init dccp_v4_init_net(struct net *net)
1009{
1010	if (dccp_hashinfo.bhash == NULL)
1011		return -ESOCKTNOSUPPORT;
1012
1013	return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1014				    SOCK_DCCP, IPPROTO_DCCP, net);
1015}
1016
1017static void __net_exit dccp_v4_exit_net(struct net *net)
1018{
1019	inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1020}
1021
1022static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1023{
1024	inet_twsk_purge(&dccp_hashinfo, AF_INET);
1025}
1026
1027static struct pernet_operations dccp_v4_ops = {
1028	.init	= dccp_v4_init_net,
1029	.exit	= dccp_v4_exit_net,
1030	.exit_batch = dccp_v4_exit_batch,
1031};
1032
1033static int __init dccp_v4_init(void)
1034{
1035	int err = proto_register(&dccp_v4_prot, 1);
1036
1037	if (err)
1038		goto out;
1039
1040	inet_register_protosw(&dccp_v4_protosw);
1041
1042	err = register_pernet_subsys(&dccp_v4_ops);
1043	if (err)
1044		goto out_destroy_ctl_sock;
1045
1046	err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1047	if (err)
1048		goto out_proto_unregister;
1049
1050out:
1051	return err;
1052out_proto_unregister:
1053	unregister_pernet_subsys(&dccp_v4_ops);
1054out_destroy_ctl_sock:
1055	inet_unregister_protosw(&dccp_v4_protosw);
1056	proto_unregister(&dccp_v4_prot);
1057	goto out;
1058}
1059
1060static void __exit dccp_v4_exit(void)
1061{
1062	inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1063	unregister_pernet_subsys(&dccp_v4_ops);
1064	inet_unregister_protosw(&dccp_v4_protosw);
1065	proto_unregister(&dccp_v4_prot);
1066}
1067
1068module_init(dccp_v4_init);
1069module_exit(dccp_v4_exit);
1070
1071/*
1072 * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1073 * values directly, Also cover the case where the protocol is not specified,
1074 * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1075 */
1076MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1077MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1078MODULE_LICENSE("GPL");
1079MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1080MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1081