xref: /linux/net/dccp/ipv4.c (revision a54f42722e494c86ad0eeba198a662d68aeabb15)
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 	/* For the first __dccp_basic_hdr_len() check, we only need dh->dccph_x,
258 	 * which is in byte 7 of the dccp header.
259 	 * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
260 	 *
261 	 * Later on, we want to access the sequence number fields, which are
262 	 * beyond 8 bytes, so we have to pskb_may_pull() ourselves.
263 	 */
264 	dh = (struct dccp_hdr *)(skb->data + offset);
265 	if (!pskb_may_pull(skb, offset + __dccp_basic_hdr_len(dh)))
266 		return -EINVAL;
267 	iph = (struct iphdr *)skb->data;
268 	dh = (struct dccp_hdr *)(skb->data + offset);
269 
270 	sk = __inet_lookup_established(net, &dccp_hashinfo,
271 				       iph->daddr, dh->dccph_dport,
272 				       iph->saddr, ntohs(dh->dccph_sport),
273 				       inet_iif(skb), 0);
274 	if (!sk) {
275 		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
276 		return -ENOENT;
277 	}
278 
279 	if (sk->sk_state == DCCP_TIME_WAIT) {
280 		inet_twsk_put(inet_twsk(sk));
281 		return 0;
282 	}
283 	seq = dccp_hdr_seq(dh);
284 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
285 		dccp_req_err(sk, seq);
286 		return 0;
287 	}
288 
289 	bh_lock_sock(sk);
290 	/* If too many ICMPs get dropped on busy
291 	 * servers this needs to be solved differently.
292 	 */
293 	if (sock_owned_by_user(sk))
294 		__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
295 
296 	if (sk->sk_state == DCCP_CLOSED)
297 		goto out;
298 
299 	dp = dccp_sk(sk);
300 	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
301 	    !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
302 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
303 		goto out;
304 	}
305 
306 	switch (type) {
307 	case ICMP_REDIRECT:
308 		if (!sock_owned_by_user(sk))
309 			dccp_do_redirect(skb, sk);
310 		goto out;
311 	case ICMP_SOURCE_QUENCH:
312 		/* Just silently ignore these. */
313 		goto out;
314 	case ICMP_PARAMETERPROB:
315 		err = EPROTO;
316 		break;
317 	case ICMP_DEST_UNREACH:
318 		if (code > NR_ICMP_UNREACH)
319 			goto out;
320 
321 		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
322 			if (!sock_owned_by_user(sk))
323 				dccp_do_pmtu_discovery(sk, iph, info);
324 			goto out;
325 		}
326 
327 		err = icmp_err_convert[code].errno;
328 		break;
329 	case ICMP_TIME_EXCEEDED:
330 		err = EHOSTUNREACH;
331 		break;
332 	default:
333 		goto out;
334 	}
335 
336 	switch (sk->sk_state) {
337 	case DCCP_REQUESTING:
338 	case DCCP_RESPOND:
339 		if (!sock_owned_by_user(sk)) {
340 			__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
341 			sk->sk_err = err;
342 
343 			sk_error_report(sk);
344 
345 			dccp_done(sk);
346 		} else {
347 			WRITE_ONCE(sk->sk_err_soft, err);
348 		}
349 		goto out;
350 	}
351 
352 	/* If we've already connected we will keep trying
353 	 * until we time out, or the user gives up.
354 	 *
355 	 * rfc1122 4.2.3.9 allows to consider as hard errors
356 	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
357 	 * but it is obsoleted by pmtu discovery).
358 	 *
359 	 * Note, that in modern internet, where routing is unreliable
360 	 * and in each dark corner broken firewalls sit, sending random
361 	 * errors ordered by their masters even this two messages finally lose
362 	 * their original sense (even Linux sends invalid PORT_UNREACHs)
363 	 *
364 	 * Now we are in compliance with RFCs.
365 	 *							--ANK (980905)
366 	 */
367 
368 	if (!sock_owned_by_user(sk) && inet_test_bit(RECVERR, sk)) {
369 		sk->sk_err = err;
370 		sk_error_report(sk);
371 	} else { /* Only an error on timeout */
372 		WRITE_ONCE(sk->sk_err_soft, err);
373 	}
374 out:
375 	bh_unlock_sock(sk);
376 	sock_put(sk);
377 	return 0;
378 }
379 
380 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
381 				      __be32 src, __be32 dst)
382 {
383 	return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
384 }
385 
386 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
387 {
388 	const struct inet_sock *inet = inet_sk(sk);
389 	struct dccp_hdr *dh = dccp_hdr(skb);
390 
391 	dccp_csum_outgoing(skb);
392 	dh->dccph_checksum = dccp_v4_csum_finish(skb,
393 						 inet->inet_saddr,
394 						 inet->inet_daddr);
395 }
396 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
397 
398 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
399 {
400 	return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
401 					   ip_hdr(skb)->saddr,
402 					   dccp_hdr(skb)->dccph_dport,
403 					   dccp_hdr(skb)->dccph_sport);
404 }
405 
406 /*
407  * The three way handshake has completed - we got a valid ACK or DATAACK -
408  * now create the new socket.
409  *
410  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
411  */
412 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
413 				       struct sk_buff *skb,
414 				       struct request_sock *req,
415 				       struct dst_entry *dst,
416 				       struct request_sock *req_unhash,
417 				       bool *own_req)
418 {
419 	struct inet_request_sock *ireq;
420 	struct inet_sock *newinet;
421 	struct sock *newsk;
422 
423 	if (sk_acceptq_is_full(sk))
424 		goto exit_overflow;
425 
426 	newsk = dccp_create_openreq_child(sk, req, skb);
427 	if (newsk == NULL)
428 		goto exit_nonewsk;
429 
430 	newinet		   = inet_sk(newsk);
431 	ireq		   = inet_rsk(req);
432 	sk_daddr_set(newsk, ireq->ir_rmt_addr);
433 	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
434 	newinet->inet_saddr	= ireq->ir_loc_addr;
435 	RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
436 	newinet->mc_index  = inet_iif(skb);
437 	newinet->mc_ttl	   = ip_hdr(skb)->ttl;
438 	atomic_set(&newinet->inet_id, get_random_u16());
439 
440 	if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
441 		goto put_and_exit;
442 
443 	sk_setup_caps(newsk, dst);
444 
445 	dccp_sync_mss(newsk, dst_mtu(dst));
446 
447 	if (__inet_inherit_port(sk, newsk) < 0)
448 		goto put_and_exit;
449 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
450 	if (*own_req)
451 		ireq->ireq_opt = NULL;
452 	else
453 		newinet->inet_opt = NULL;
454 	return newsk;
455 
456 exit_overflow:
457 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
458 exit_nonewsk:
459 	dst_release(dst);
460 exit:
461 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
462 	return NULL;
463 put_and_exit:
464 	newinet->inet_opt = NULL;
465 	inet_csk_prepare_forced_close(newsk);
466 	dccp_done(newsk);
467 	goto exit;
468 }
469 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
470 
471 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
472 					   struct sk_buff *skb)
473 {
474 	struct rtable *rt;
475 	const struct iphdr *iph = ip_hdr(skb);
476 	struct flowi4 fl4 = {
477 		.flowi4_oif = inet_iif(skb),
478 		.daddr = iph->saddr,
479 		.saddr = iph->daddr,
480 		.flowi4_tos = ip_sock_rt_tos(sk),
481 		.flowi4_scope = ip_sock_rt_scope(sk),
482 		.flowi4_proto = sk->sk_protocol,
483 		.fl4_sport = dccp_hdr(skb)->dccph_dport,
484 		.fl4_dport = dccp_hdr(skb)->dccph_sport,
485 	};
486 
487 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
488 	rt = ip_route_output_flow(net, &fl4, sk);
489 	if (IS_ERR(rt)) {
490 		IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
491 		return NULL;
492 	}
493 
494 	return &rt->dst;
495 }
496 
497 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
498 {
499 	int err = -1;
500 	struct sk_buff *skb;
501 	struct dst_entry *dst;
502 	struct flowi4 fl4;
503 
504 	dst = inet_csk_route_req(sk, &fl4, req);
505 	if (dst == NULL)
506 		goto out;
507 
508 	skb = dccp_make_response(sk, dst, req);
509 	if (skb != NULL) {
510 		const struct inet_request_sock *ireq = inet_rsk(req);
511 		struct dccp_hdr *dh = dccp_hdr(skb);
512 
513 		dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
514 							      ireq->ir_rmt_addr);
515 		rcu_read_lock();
516 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
517 					    ireq->ir_rmt_addr,
518 					    rcu_dereference(ireq->ireq_opt),
519 					    inet_sk(sk)->tos);
520 		rcu_read_unlock();
521 		err = net_xmit_eval(err);
522 	}
523 
524 out:
525 	dst_release(dst);
526 	return err;
527 }
528 
529 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
530 {
531 	int err;
532 	const struct iphdr *rxiph;
533 	struct sk_buff *skb;
534 	struct dst_entry *dst;
535 	struct net *net = dev_net(skb_dst(rxskb)->dev);
536 	struct dccp_v4_pernet *pn;
537 	struct sock *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 	pn = net_generic(net, dccp_v4_pernet_id);
547 	ctl_sk = pn->v4_ctl_sk;
548 	dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
549 	if (dst == NULL)
550 		return;
551 
552 	skb = dccp_ctl_make_reset(ctl_sk, rxskb);
553 	if (skb == NULL)
554 		goto out;
555 
556 	rxiph = ip_hdr(rxskb);
557 	dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
558 								 rxiph->daddr);
559 	skb_dst_set(skb, dst_clone(dst));
560 
561 	local_bh_disable();
562 	bh_lock_sock(ctl_sk);
563 	err = ip_build_and_send_pkt(skb, ctl_sk,
564 				    rxiph->daddr, rxiph->saddr, NULL,
565 				    inet_sk(ctl_sk)->tos);
566 	bh_unlock_sock(ctl_sk);
567 
568 	if (net_xmit_eval(err) == 0) {
569 		__DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
570 		__DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
571 	}
572 	local_bh_enable();
573 out:
574 	dst_release(dst);
575 }
576 
577 static void dccp_v4_reqsk_destructor(struct request_sock *req)
578 {
579 	dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
580 	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
581 }
582 
583 void dccp_syn_ack_timeout(const struct request_sock *req)
584 {
585 }
586 EXPORT_SYMBOL(dccp_syn_ack_timeout);
587 
588 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
589 	.family		= PF_INET,
590 	.obj_size	= sizeof(struct dccp_request_sock),
591 	.rtx_syn_ack	= dccp_v4_send_response,
592 	.send_ack	= dccp_reqsk_send_ack,
593 	.destructor	= dccp_v4_reqsk_destructor,
594 	.send_reset	= dccp_v4_ctl_send_reset,
595 	.syn_ack_timeout = dccp_syn_ack_timeout,
596 };
597 
598 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
599 {
600 	struct inet_request_sock *ireq;
601 	struct request_sock *req;
602 	struct dccp_request_sock *dreq;
603 	const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
604 	struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
605 
606 	/* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
607 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
608 		return 0;	/* discard, don't send a reset here */
609 
610 	if (dccp_bad_service_code(sk, service)) {
611 		dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
612 		goto drop;
613 	}
614 	/*
615 	 * TW buckets are converted to open requests without
616 	 * limitations, they conserve resources and peer is
617 	 * evidently real one.
618 	 */
619 	dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
620 	if (inet_csk_reqsk_queue_is_full(sk))
621 		goto drop;
622 
623 	if (sk_acceptq_is_full(sk))
624 		goto drop;
625 
626 	req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
627 	if (req == NULL)
628 		goto drop;
629 
630 	if (dccp_reqsk_init(req, dccp_sk(sk), skb))
631 		goto drop_and_free;
632 
633 	dreq = dccp_rsk(req);
634 	if (dccp_parse_options(sk, dreq, skb))
635 		goto drop_and_free;
636 
637 	if (security_inet_conn_request(sk, skb, req))
638 		goto drop_and_free;
639 
640 	ireq = inet_rsk(req);
641 	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
642 	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
643 	ireq->ir_mark = inet_request_mark(sk, skb);
644 	ireq->ireq_family = AF_INET;
645 	ireq->ir_iif = READ_ONCE(sk->sk_bound_dev_if);
646 
647 	/*
648 	 * Step 3: Process LISTEN state
649 	 *
650 	 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
651 	 *
652 	 * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
653 	 */
654 	dreq->dreq_isr	   = dcb->dccpd_seq;
655 	dreq->dreq_gsr	   = dreq->dreq_isr;
656 	dreq->dreq_iss	   = dccp_v4_init_sequence(skb);
657 	dreq->dreq_gss     = dreq->dreq_iss;
658 	dreq->dreq_service = service;
659 
660 	if (dccp_v4_send_response(sk, req))
661 		goto drop_and_free;
662 
663 	inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
664 	reqsk_put(req);
665 	return 0;
666 
667 drop_and_free:
668 	reqsk_free(req);
669 drop:
670 	__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
671 	return -1;
672 }
673 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
674 
675 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
676 {
677 	struct dccp_hdr *dh = dccp_hdr(skb);
678 
679 	if (sk->sk_state == DCCP_OPEN) { /* Fast path */
680 		if (dccp_rcv_established(sk, skb, dh, skb->len))
681 			goto reset;
682 		return 0;
683 	}
684 
685 	/*
686 	 *  Step 3: Process LISTEN state
687 	 *	 If P.type == Request or P contains a valid Init Cookie option,
688 	 *	      (* Must scan the packet's options to check for Init
689 	 *		 Cookies.  Only Init Cookies are processed here,
690 	 *		 however; other options are processed in Step 8.  This
691 	 *		 scan need only be performed if the endpoint uses Init
692 	 *		 Cookies *)
693 	 *	      (* Generate a new socket and switch to that socket *)
694 	 *	      Set S := new socket for this port pair
695 	 *	      S.state = RESPOND
696 	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
697 	 *	      Initialize S.GAR := S.ISS
698 	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
699 	 *	      Continue with S.state == RESPOND
700 	 *	      (* A Response packet will be generated in Step 11 *)
701 	 *	 Otherwise,
702 	 *	      Generate Reset(No Connection) unless P.type == Reset
703 	 *	      Drop packet and return
704 	 *
705 	 * NOTE: the check for the packet types is done in
706 	 *	 dccp_rcv_state_process
707 	 */
708 
709 	if (dccp_rcv_state_process(sk, skb, dh, skb->len))
710 		goto reset;
711 	return 0;
712 
713 reset:
714 	dccp_v4_ctl_send_reset(sk, skb);
715 	kfree_skb(skb);
716 	return 0;
717 }
718 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
719 
720 /**
721  *	dccp_invalid_packet  -  check for malformed packets
722  *	@skb: Packet to validate
723  *
724  *	Implements RFC 4340, 8.5:  Step 1: Check header basics
725  *	Packets that fail these checks are ignored and do not receive Resets.
726  */
727 int dccp_invalid_packet(struct sk_buff *skb)
728 {
729 	const struct dccp_hdr *dh;
730 	unsigned int cscov;
731 	u8 dccph_doff;
732 
733 	if (skb->pkt_type != PACKET_HOST)
734 		return 1;
735 
736 	/* If the packet is shorter than 12 bytes, drop packet and return */
737 	if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
738 		DCCP_WARN("pskb_may_pull failed\n");
739 		return 1;
740 	}
741 
742 	dh = dccp_hdr(skb);
743 
744 	/* If P.type is not understood, drop packet and return */
745 	if (dh->dccph_type >= DCCP_PKT_INVALID) {
746 		DCCP_WARN("invalid packet type\n");
747 		return 1;
748 	}
749 
750 	/*
751 	 * If P.Data Offset is too small for packet type, drop packet and return
752 	 */
753 	dccph_doff = dh->dccph_doff;
754 	if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
755 		DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
756 		return 1;
757 	}
758 	/*
759 	 * If P.Data Offset is too large for packet, drop packet and return
760 	 */
761 	if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
762 		DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
763 		return 1;
764 	}
765 	dh = dccp_hdr(skb);
766 	/*
767 	 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
768 	 * has short sequence numbers), drop packet and return
769 	 */
770 	if ((dh->dccph_type < DCCP_PKT_DATA    ||
771 	    dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
772 		DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
773 			  dccp_packet_name(dh->dccph_type));
774 		return 1;
775 	}
776 
777 	/*
778 	 * If P.CsCov is too large for the packet size, drop packet and return.
779 	 * This must come _before_ checksumming (not as RFC 4340 suggests).
780 	 */
781 	cscov = dccp_csum_coverage(skb);
782 	if (cscov > skb->len) {
783 		DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
784 			  dh->dccph_cscov, skb->len);
785 		return 1;
786 	}
787 
788 	/* If header checksum is incorrect, drop packet and return.
789 	 * (This step is completed in the AF-dependent functions.) */
790 	skb->csum = skb_checksum(skb, 0, cscov, 0);
791 
792 	return 0;
793 }
794 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
795 
796 /* this is called when real data arrives */
797 static int dccp_v4_rcv(struct sk_buff *skb)
798 {
799 	const struct dccp_hdr *dh;
800 	const struct iphdr *iph;
801 	bool refcounted;
802 	struct sock *sk;
803 	int min_cov;
804 
805 	/* Step 1: Check header basics */
806 
807 	if (dccp_invalid_packet(skb))
808 		goto discard_it;
809 
810 	iph = ip_hdr(skb);
811 	/* Step 1: If header checksum is incorrect, drop packet and return */
812 	if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
813 		DCCP_WARN("dropped packet with invalid checksum\n");
814 		goto discard_it;
815 	}
816 
817 	dh = dccp_hdr(skb);
818 
819 	DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
820 	DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
821 
822 	dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
823 		      dccp_packet_name(dh->dccph_type),
824 		      &iph->saddr, ntohs(dh->dccph_sport),
825 		      &iph->daddr, ntohs(dh->dccph_dport),
826 		      (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
827 
828 	if (dccp_packet_without_ack(skb)) {
829 		DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
830 		dccp_pr_debug_cat("\n");
831 	} else {
832 		DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
833 		dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
834 				  DCCP_SKB_CB(skb)->dccpd_ack_seq);
835 	}
836 
837 lookup:
838 	sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
839 			       dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
840 	if (!sk) {
841 		dccp_pr_debug("failed to look up flow ID in table and "
842 			      "get corresponding socket\n");
843 		goto no_dccp_socket;
844 	}
845 
846 	/*
847 	 * Step 2:
848 	 *	... or S.state == TIMEWAIT,
849 	 *		Generate Reset(No Connection) unless P.type == Reset
850 	 *		Drop packet and return
851 	 */
852 	if (sk->sk_state == DCCP_TIME_WAIT) {
853 		dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
854 		inet_twsk_put(inet_twsk(sk));
855 		goto no_dccp_socket;
856 	}
857 
858 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
859 		struct request_sock *req = inet_reqsk(sk);
860 		struct sock *nsk;
861 
862 		sk = req->rsk_listener;
863 		if (unlikely(sk->sk_state != DCCP_LISTEN)) {
864 			inet_csk_reqsk_queue_drop_and_put(sk, req);
865 			goto lookup;
866 		}
867 		sock_hold(sk);
868 		refcounted = true;
869 		nsk = dccp_check_req(sk, skb, req);
870 		if (!nsk) {
871 			reqsk_put(req);
872 			goto discard_and_relse;
873 		}
874 		if (nsk == sk) {
875 			reqsk_put(req);
876 		} else if (dccp_child_process(sk, nsk, skb)) {
877 			dccp_v4_ctl_send_reset(sk, skb);
878 			goto discard_and_relse;
879 		} else {
880 			sock_put(sk);
881 			return 0;
882 		}
883 	}
884 	/*
885 	 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
886 	 *	o if MinCsCov = 0, only packets with CsCov = 0 are accepted
887 	 *	o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
888 	 */
889 	min_cov = dccp_sk(sk)->dccps_pcrlen;
890 	if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
891 		dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
892 			      dh->dccph_cscov, min_cov);
893 		/* FIXME: "Such packets SHOULD be reported using Data Dropped
894 		 *         options (Section 11.7) with Drop Code 0, Protocol
895 		 *         Constraints."                                     */
896 		goto discard_and_relse;
897 	}
898 
899 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
900 		goto discard_and_relse;
901 	nf_reset_ct(skb);
902 
903 	return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
904 
905 no_dccp_socket:
906 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
907 		goto discard_it;
908 	/*
909 	 * Step 2:
910 	 *	If no socket ...
911 	 *		Generate Reset(No Connection) unless P.type == Reset
912 	 *		Drop packet and return
913 	 */
914 	if (dh->dccph_type != DCCP_PKT_RESET) {
915 		DCCP_SKB_CB(skb)->dccpd_reset_code =
916 					DCCP_RESET_CODE_NO_CONNECTION;
917 		dccp_v4_ctl_send_reset(sk, skb);
918 	}
919 
920 discard_it:
921 	kfree_skb(skb);
922 	return 0;
923 
924 discard_and_relse:
925 	if (refcounted)
926 		sock_put(sk);
927 	goto discard_it;
928 }
929 
930 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
931 	.queue_xmit	   = ip_queue_xmit,
932 	.send_check	   = dccp_v4_send_check,
933 	.rebuild_header	   = inet_sk_rebuild_header,
934 	.conn_request	   = dccp_v4_conn_request,
935 	.syn_recv_sock	   = dccp_v4_request_recv_sock,
936 	.net_header_len	   = sizeof(struct iphdr),
937 	.setsockopt	   = ip_setsockopt,
938 	.getsockopt	   = ip_getsockopt,
939 	.addr2sockaddr	   = inet_csk_addr2sockaddr,
940 	.sockaddr_len	   = sizeof(struct sockaddr_in),
941 };
942 
943 static int dccp_v4_init_sock(struct sock *sk)
944 {
945 	static __u8 dccp_v4_ctl_sock_initialized;
946 	int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
947 
948 	if (err == 0) {
949 		if (unlikely(!dccp_v4_ctl_sock_initialized))
950 			dccp_v4_ctl_sock_initialized = 1;
951 		inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
952 	}
953 
954 	return err;
955 }
956 
957 static struct timewait_sock_ops dccp_timewait_sock_ops = {
958 	.twsk_obj_size	= sizeof(struct inet_timewait_sock),
959 };
960 
961 static struct proto dccp_v4_prot = {
962 	.name			= "DCCP",
963 	.owner			= THIS_MODULE,
964 	.close			= dccp_close,
965 	.connect		= dccp_v4_connect,
966 	.disconnect		= dccp_disconnect,
967 	.ioctl			= dccp_ioctl,
968 	.init			= dccp_v4_init_sock,
969 	.setsockopt		= dccp_setsockopt,
970 	.getsockopt		= dccp_getsockopt,
971 	.sendmsg		= dccp_sendmsg,
972 	.recvmsg		= dccp_recvmsg,
973 	.backlog_rcv		= dccp_v4_do_rcv,
974 	.hash			= inet_hash,
975 	.unhash			= inet_unhash,
976 	.accept			= inet_csk_accept,
977 	.get_port		= inet_csk_get_port,
978 	.shutdown		= dccp_shutdown,
979 	.destroy		= dccp_destroy_sock,
980 	.orphan_count		= &dccp_orphan_count,
981 	.max_header		= MAX_DCCP_HEADER,
982 	.obj_size		= sizeof(struct dccp_sock),
983 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
984 	.rsk_prot		= &dccp_request_sock_ops,
985 	.twsk_prot		= &dccp_timewait_sock_ops,
986 	.h.hashinfo		= &dccp_hashinfo,
987 };
988 
989 static const struct net_protocol dccp_v4_protocol = {
990 	.handler	= dccp_v4_rcv,
991 	.err_handler	= dccp_v4_err,
992 	.no_policy	= 1,
993 	.icmp_strict_tag_validation = 1,
994 };
995 
996 static const struct proto_ops inet_dccp_ops = {
997 	.family		   = PF_INET,
998 	.owner		   = THIS_MODULE,
999 	.release	   = inet_release,
1000 	.bind		   = inet_bind,
1001 	.connect	   = inet_stream_connect,
1002 	.socketpair	   = sock_no_socketpair,
1003 	.accept		   = inet_accept,
1004 	.getname	   = inet_getname,
1005 	/* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1006 	.poll		   = dccp_poll,
1007 	.ioctl		   = inet_ioctl,
1008 	.gettstamp	   = sock_gettstamp,
1009 	/* FIXME: work on inet_listen to rename it to sock_common_listen */
1010 	.listen		   = inet_dccp_listen,
1011 	.shutdown	   = inet_shutdown,
1012 	.setsockopt	   = sock_common_setsockopt,
1013 	.getsockopt	   = sock_common_getsockopt,
1014 	.sendmsg	   = inet_sendmsg,
1015 	.recvmsg	   = sock_common_recvmsg,
1016 	.mmap		   = sock_no_mmap,
1017 };
1018 
1019 static struct inet_protosw dccp_v4_protosw = {
1020 	.type		= SOCK_DCCP,
1021 	.protocol	= IPPROTO_DCCP,
1022 	.prot		= &dccp_v4_prot,
1023 	.ops		= &inet_dccp_ops,
1024 	.flags		= INET_PROTOSW_ICSK,
1025 };
1026 
1027 static int __net_init dccp_v4_init_net(struct net *net)
1028 {
1029 	struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1030 
1031 	if (dccp_hashinfo.bhash == NULL)
1032 		return -ESOCKTNOSUPPORT;
1033 
1034 	return inet_ctl_sock_create(&pn->v4_ctl_sk, PF_INET,
1035 				    SOCK_DCCP, IPPROTO_DCCP, net);
1036 }
1037 
1038 static void __net_exit dccp_v4_exit_net(struct net *net)
1039 {
1040 	struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1041 
1042 	inet_ctl_sock_destroy(pn->v4_ctl_sk);
1043 }
1044 
1045 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1046 {
1047 	inet_twsk_purge(&dccp_hashinfo, AF_INET);
1048 }
1049 
1050 static struct pernet_operations dccp_v4_ops = {
1051 	.init	= dccp_v4_init_net,
1052 	.exit	= dccp_v4_exit_net,
1053 	.exit_batch = dccp_v4_exit_batch,
1054 	.id	= &dccp_v4_pernet_id,
1055 	.size   = sizeof(struct dccp_v4_pernet),
1056 };
1057 
1058 static int __init dccp_v4_init(void)
1059 {
1060 	int err = proto_register(&dccp_v4_prot, 1);
1061 
1062 	if (err)
1063 		goto out;
1064 
1065 	inet_register_protosw(&dccp_v4_protosw);
1066 
1067 	err = register_pernet_subsys(&dccp_v4_ops);
1068 	if (err)
1069 		goto out_destroy_ctl_sock;
1070 
1071 	err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1072 	if (err)
1073 		goto out_proto_unregister;
1074 
1075 out:
1076 	return err;
1077 out_proto_unregister:
1078 	unregister_pernet_subsys(&dccp_v4_ops);
1079 out_destroy_ctl_sock:
1080 	inet_unregister_protosw(&dccp_v4_protosw);
1081 	proto_unregister(&dccp_v4_prot);
1082 	goto out;
1083 }
1084 
1085 static void __exit dccp_v4_exit(void)
1086 {
1087 	inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1088 	unregister_pernet_subsys(&dccp_v4_ops);
1089 	inet_unregister_protosw(&dccp_v4_protosw);
1090 	proto_unregister(&dccp_v4_prot);
1091 }
1092 
1093 module_init(dccp_v4_init);
1094 module_exit(dccp_v4_exit);
1095 
1096 /*
1097  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1098  * values directly, Also cover the case where the protocol is not specified,
1099  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1100  */
1101 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1102 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1103 MODULE_LICENSE("GPL");
1104 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1105 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1106