xref: /linux/net/dccp/ipv4.c (revision e31e3f6c0ce473f7ce1e70d54ac8e3ed190509f8)
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_dscp.h>
19 #include <net/inet_hashtables.h>
20 #include <net/inet_sock.h>
21 #include <net/protocol.h>
22 #include <net/sock.h>
23 #include <net/timewait_sock.h>
24 #include <net/tcp_states.h>
25 #include <net/xfrm.h>
26 #include <net/secure_seq.h>
27 #include <net/netns/generic.h>
28 #include <net/rstreason.h>
29 
30 #include "ackvec.h"
31 #include "ccid.h"
32 #include "dccp.h"
33 #include "feat.h"
34 
35 struct dccp_v4_pernet {
36 	struct sock *v4_ctl_sk;
37 };
38 
39 static unsigned int dccp_v4_pernet_id __read_mostly;
40 
41 /*
42  * The per-net v4_ctl_sk socket is used for responding to
43  * the Out-of-the-blue (OOTB) packets. A control sock will be created
44  * for this socket at the initialization time.
45  */
46 
47 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
48 {
49 	const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
50 	struct inet_sock *inet = inet_sk(sk);
51 	struct dccp_sock *dp = dccp_sk(sk);
52 	__be16 orig_sport, orig_dport;
53 	__be32 daddr, nexthop;
54 	struct flowi4 *fl4;
55 	struct rtable *rt;
56 	int err;
57 	struct ip_options_rcu *inet_opt;
58 
59 	dp->dccps_role = DCCP_ROLE_CLIENT;
60 
61 	if (addr_len < sizeof(struct sockaddr_in))
62 		return -EINVAL;
63 
64 	if (usin->sin_family != AF_INET)
65 		return -EAFNOSUPPORT;
66 
67 	nexthop = daddr = usin->sin_addr.s_addr;
68 
69 	inet_opt = rcu_dereference_protected(inet->inet_opt,
70 					     lockdep_sock_is_held(sk));
71 	if (inet_opt != NULL && inet_opt->opt.srr) {
72 		if (daddr == 0)
73 			return -EINVAL;
74 		nexthop = inet_opt->opt.faddr;
75 	}
76 
77 	orig_sport = inet->inet_sport;
78 	orig_dport = usin->sin_port;
79 	fl4 = &inet->cork.fl.u.ip4;
80 	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
81 			      sk->sk_bound_dev_if, IPPROTO_DCCP, orig_sport,
82 			      orig_dport, sk);
83 	if (IS_ERR(rt))
84 		return PTR_ERR(rt);
85 
86 	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
87 		ip_rt_put(rt);
88 		return -ENETUNREACH;
89 	}
90 
91 	if (inet_opt == NULL || !inet_opt->opt.srr)
92 		daddr = fl4->daddr;
93 
94 	if (inet->inet_saddr == 0) {
95 		err = inet_bhash2_update_saddr(sk,  &fl4->saddr, AF_INET);
96 		if (err) {
97 			ip_rt_put(rt);
98 			return err;
99 		}
100 	} else {
101 		sk_rcv_saddr_set(sk, inet->inet_saddr);
102 	}
103 
104 	inet->inet_dport = usin->sin_port;
105 	sk_daddr_set(sk, daddr);
106 
107 	inet_csk(sk)->icsk_ext_hdr_len = 0;
108 	if (inet_opt)
109 		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
110 	/*
111 	 * Socket identity is still unknown (sport may be zero).
112 	 * However we set state to DCCP_REQUESTING and not releasing socket
113 	 * lock select source port, enter ourselves into the hash tables and
114 	 * complete initialization after this.
115 	 */
116 	dccp_set_state(sk, DCCP_REQUESTING);
117 	err = inet_hash_connect(&dccp_death_row, sk);
118 	if (err != 0)
119 		goto failure;
120 
121 	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
122 			       inet->inet_sport, inet->inet_dport, sk);
123 	if (IS_ERR(rt)) {
124 		err = PTR_ERR(rt);
125 		rt = NULL;
126 		goto failure;
127 	}
128 	/* OK, now commit destination to socket.  */
129 	sk_setup_caps(sk, &rt->dst);
130 
131 	dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
132 						    inet->inet_daddr,
133 						    inet->inet_sport,
134 						    inet->inet_dport);
135 	atomic_set(&inet->inet_id, get_random_u16());
136 
137 	err = dccp_connect(sk);
138 	rt = NULL;
139 	if (err != 0)
140 		goto failure;
141 out:
142 	return err;
143 failure:
144 	/*
145 	 * This unhashes the socket and releases the local port, if necessary.
146 	 */
147 	dccp_set_state(sk, DCCP_CLOSED);
148 	inet_bhash2_reset_saddr(sk);
149 	ip_rt_put(rt);
150 	sk->sk_route_caps = 0;
151 	inet->inet_dport = 0;
152 	goto out;
153 }
154 EXPORT_SYMBOL_GPL(dccp_v4_connect);
155 
156 /*
157  * This routine does path mtu discovery as defined in RFC1191.
158  */
159 static inline void dccp_do_pmtu_discovery(struct sock *sk,
160 					  const struct iphdr *iph,
161 					  u32 mtu)
162 {
163 	struct dst_entry *dst;
164 	const struct inet_sock *inet = inet_sk(sk);
165 	const struct dccp_sock *dp = dccp_sk(sk);
166 
167 	/* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
168 	 * send out by Linux are always < 576bytes so they should go through
169 	 * unfragmented).
170 	 */
171 	if (sk->sk_state == DCCP_LISTEN)
172 		return;
173 
174 	dst = inet_csk_update_pmtu(sk, mtu);
175 	if (!dst)
176 		return;
177 
178 	/* Something is about to be wrong... Remember soft error
179 	 * for the case, if this connection will not able to recover.
180 	 */
181 	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
182 		WRITE_ONCE(sk->sk_err_soft, EMSGSIZE);
183 
184 	mtu = dst_mtu(dst);
185 
186 	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
187 	    ip_sk_accept_pmtu(sk) &&
188 	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
189 		dccp_sync_mss(sk, mtu);
190 
191 		/*
192 		 * From RFC 4340, sec. 14.1:
193 		 *
194 		 *	DCCP-Sync packets are the best choice for upward
195 		 *	probing, since DCCP-Sync probes do not risk application
196 		 *	data loss.
197 		 */
198 		dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
199 	} /* else let the usual retransmit timer handle it */
200 }
201 
202 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
203 {
204 	struct dst_entry *dst = __sk_dst_check(sk, 0);
205 
206 	if (dst)
207 		dst->ops->redirect(dst, sk, skb);
208 }
209 
210 void dccp_req_err(struct sock *sk, u64 seq)
211 	{
212 	struct request_sock *req = inet_reqsk(sk);
213 	struct net *net = sock_net(sk);
214 
215 	/*
216 	 * ICMPs are not backlogged, hence we cannot get an established
217 	 * socket here.
218 	 */
219 	if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
220 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
221 	} else {
222 		/*
223 		 * Still in RESPOND, just remove it silently.
224 		 * There is no good way to pass the error to the newly
225 		 * created socket, and POSIX does not want network
226 		 * errors returned from accept().
227 		 */
228 		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
229 	}
230 	reqsk_put(req);
231 }
232 EXPORT_SYMBOL(dccp_req_err);
233 
234 /*
235  * This routine is called by the ICMP module when it gets some sort of error
236  * condition. If err < 0 then the socket should be closed and the error
237  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
238  * After adjustment header points to the first 8 bytes of the tcp header. We
239  * need to find the appropriate port.
240  *
241  * The locking strategy used here is very "optimistic". When someone else
242  * accesses the socket the ICMP is just dropped and for some paths there is no
243  * check at all. A more general error queue to queue errors for later handling
244  * is probably better.
245  */
246 static int dccp_v4_err(struct sk_buff *skb, u32 info)
247 {
248 	const struct iphdr *iph = (struct iphdr *)skb->data;
249 	const u8 offset = iph->ihl << 2;
250 	const struct dccp_hdr *dh;
251 	struct dccp_sock *dp;
252 	const int type = icmp_hdr(skb)->type;
253 	const int code = icmp_hdr(skb)->code;
254 	struct sock *sk;
255 	__u64 seq;
256 	int err;
257 	struct net *net = dev_net(skb->dev);
258 
259 	if (!pskb_may_pull(skb, offset + sizeof(*dh)))
260 		return -EINVAL;
261 	dh = (struct dccp_hdr *)(skb->data + offset);
262 	if (!pskb_may_pull(skb, offset + __dccp_basic_hdr_len(dh)))
263 		return -EINVAL;
264 	iph = (struct iphdr *)skb->data;
265 	dh = (struct dccp_hdr *)(skb->data + offset);
266 
267 	sk = __inet_lookup_established(net, &dccp_hashinfo,
268 				       iph->daddr, dh->dccph_dport,
269 				       iph->saddr, ntohs(dh->dccph_sport),
270 				       inet_iif(skb), 0);
271 	if (!sk) {
272 		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
273 		return -ENOENT;
274 	}
275 
276 	if (sk->sk_state == DCCP_TIME_WAIT) {
277 		inet_twsk_put(inet_twsk(sk));
278 		return 0;
279 	}
280 	seq = dccp_hdr_seq(dh);
281 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
282 		dccp_req_err(sk, seq);
283 		return 0;
284 	}
285 
286 	bh_lock_sock(sk);
287 	/* If too many ICMPs get dropped on busy
288 	 * servers this needs to be solved differently.
289 	 */
290 	if (sock_owned_by_user(sk))
291 		__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
292 
293 	if (sk->sk_state == DCCP_CLOSED)
294 		goto out;
295 
296 	dp = dccp_sk(sk);
297 	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
298 	    !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
299 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
300 		goto out;
301 	}
302 
303 	switch (type) {
304 	case ICMP_REDIRECT:
305 		if (!sock_owned_by_user(sk))
306 			dccp_do_redirect(skb, sk);
307 		goto out;
308 	case ICMP_SOURCE_QUENCH:
309 		/* Just silently ignore these. */
310 		goto out;
311 	case ICMP_PARAMETERPROB:
312 		err = EPROTO;
313 		break;
314 	case ICMP_DEST_UNREACH:
315 		if (code > NR_ICMP_UNREACH)
316 			goto out;
317 
318 		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
319 			if (!sock_owned_by_user(sk))
320 				dccp_do_pmtu_discovery(sk, iph, info);
321 			goto out;
322 		}
323 
324 		err = icmp_err_convert[code].errno;
325 		break;
326 	case ICMP_TIME_EXCEEDED:
327 		err = EHOSTUNREACH;
328 		break;
329 	default:
330 		goto out;
331 	}
332 
333 	switch (sk->sk_state) {
334 	case DCCP_REQUESTING:
335 	case DCCP_RESPOND:
336 		if (!sock_owned_by_user(sk)) {
337 			__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
338 			sk->sk_err = err;
339 
340 			sk_error_report(sk);
341 
342 			dccp_done(sk);
343 		} else {
344 			WRITE_ONCE(sk->sk_err_soft, err);
345 		}
346 		goto out;
347 	}
348 
349 	/* If we've already connected we will keep trying
350 	 * until we time out, or the user gives up.
351 	 *
352 	 * rfc1122 4.2.3.9 allows to consider as hard errors
353 	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
354 	 * but it is obsoleted by pmtu discovery).
355 	 *
356 	 * Note, that in modern internet, where routing is unreliable
357 	 * and in each dark corner broken firewalls sit, sending random
358 	 * errors ordered by their masters even this two messages finally lose
359 	 * their original sense (even Linux sends invalid PORT_UNREACHs)
360 	 *
361 	 * Now we are in compliance with RFCs.
362 	 *							--ANK (980905)
363 	 */
364 
365 	if (!sock_owned_by_user(sk) && inet_test_bit(RECVERR, sk)) {
366 		sk->sk_err = err;
367 		sk_error_report(sk);
368 	} else { /* Only an error on timeout */
369 		WRITE_ONCE(sk->sk_err_soft, err);
370 	}
371 out:
372 	bh_unlock_sock(sk);
373 	sock_put(sk);
374 	return 0;
375 }
376 
377 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
378 				      __be32 src, __be32 dst)
379 {
380 	return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
381 }
382 
383 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
384 {
385 	const struct inet_sock *inet = inet_sk(sk);
386 	struct dccp_hdr *dh = dccp_hdr(skb);
387 
388 	dccp_csum_outgoing(skb);
389 	dh->dccph_checksum = dccp_v4_csum_finish(skb,
390 						 inet->inet_saddr,
391 						 inet->inet_daddr);
392 }
393 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
394 
395 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
396 {
397 	return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
398 					   ip_hdr(skb)->saddr,
399 					   dccp_hdr(skb)->dccph_dport,
400 					   dccp_hdr(skb)->dccph_sport);
401 }
402 
403 /*
404  * The three way handshake has completed - we got a valid ACK or DATAACK -
405  * now create the new socket.
406  *
407  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
408  */
409 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
410 				       struct sk_buff *skb,
411 				       struct request_sock *req,
412 				       struct dst_entry *dst,
413 				       struct request_sock *req_unhash,
414 				       bool *own_req)
415 {
416 	struct inet_request_sock *ireq;
417 	struct inet_sock *newinet;
418 	struct sock *newsk;
419 
420 	if (sk_acceptq_is_full(sk))
421 		goto exit_overflow;
422 
423 	newsk = dccp_create_openreq_child(sk, req, skb);
424 	if (newsk == NULL)
425 		goto exit_nonewsk;
426 
427 	newinet		   = inet_sk(newsk);
428 	ireq		   = inet_rsk(req);
429 	sk_daddr_set(newsk, ireq->ir_rmt_addr);
430 	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
431 	newinet->inet_saddr	= ireq->ir_loc_addr;
432 	RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
433 	newinet->mc_index  = inet_iif(skb);
434 	newinet->mc_ttl	   = ip_hdr(skb)->ttl;
435 	atomic_set(&newinet->inet_id, get_random_u16());
436 
437 	if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
438 		goto put_and_exit;
439 
440 	sk_setup_caps(newsk, dst);
441 
442 	dccp_sync_mss(newsk, dst_mtu(dst));
443 
444 	if (__inet_inherit_port(sk, newsk) < 0)
445 		goto put_and_exit;
446 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
447 	if (*own_req)
448 		ireq->ireq_opt = NULL;
449 	else
450 		newinet->inet_opt = NULL;
451 	return newsk;
452 
453 exit_overflow:
454 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
455 exit_nonewsk:
456 	dst_release(dst);
457 exit:
458 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
459 	return NULL;
460 put_and_exit:
461 	newinet->inet_opt = NULL;
462 	inet_csk_prepare_forced_close(newsk);
463 	dccp_done(newsk);
464 	goto exit;
465 }
466 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
467 
468 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
469 					   struct sk_buff *skb)
470 {
471 	struct rtable *rt;
472 	const struct iphdr *iph = ip_hdr(skb);
473 	struct flowi4 fl4 = {
474 		.flowi4_oif = inet_iif(skb),
475 		.daddr = iph->saddr,
476 		.saddr = iph->daddr,
477 		.flowi4_tos = inet_dscp_to_dsfield(inet_sk_dscp(inet_sk(sk))),
478 		.flowi4_scope = ip_sock_rt_scope(sk),
479 		.flowi4_proto = sk->sk_protocol,
480 		.fl4_sport = dccp_hdr(skb)->dccph_dport,
481 		.fl4_dport = dccp_hdr(skb)->dccph_sport,
482 	};
483 
484 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
485 	rt = ip_route_output_flow(net, &fl4, sk);
486 	if (IS_ERR(rt)) {
487 		IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
488 		return NULL;
489 	}
490 
491 	return &rt->dst;
492 }
493 
494 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
495 {
496 	int err = -1;
497 	struct sk_buff *skb;
498 	struct dst_entry *dst;
499 	struct flowi4 fl4;
500 
501 	dst = inet_csk_route_req(sk, &fl4, req);
502 	if (dst == NULL)
503 		goto out;
504 
505 	skb = dccp_make_response(sk, dst, req);
506 	if (skb != NULL) {
507 		const struct inet_request_sock *ireq = inet_rsk(req);
508 		struct dccp_hdr *dh = dccp_hdr(skb);
509 
510 		dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
511 							      ireq->ir_rmt_addr);
512 		rcu_read_lock();
513 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
514 					    ireq->ir_rmt_addr,
515 					    rcu_dereference(ireq->ireq_opt),
516 					    READ_ONCE(inet_sk(sk)->tos));
517 		rcu_read_unlock();
518 		err = net_xmit_eval(err);
519 	}
520 
521 out:
522 	dst_release(dst);
523 	return err;
524 }
525 
526 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb,
527 				   enum sk_rst_reason reason)
528 {
529 	int err;
530 	const struct iphdr *rxiph;
531 	struct sk_buff *skb;
532 	struct dst_entry *dst;
533 	struct net *net = dev_net(skb_dst(rxskb)->dev);
534 	struct dccp_v4_pernet *pn;
535 	struct sock *ctl_sk;
536 
537 	/* Never send a reset in response to a reset. */
538 	if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
539 		return;
540 
541 	if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
542 		return;
543 
544 	pn = net_generic(net, dccp_v4_pernet_id);
545 	ctl_sk = pn->v4_ctl_sk;
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 	local_bh_disable();
560 	bh_lock_sock(ctl_sk);
561 	err = ip_build_and_send_pkt(skb, ctl_sk,
562 				    rxiph->daddr, rxiph->saddr, NULL,
563 				    inet_sk(ctl_sk)->tos);
564 	bh_unlock_sock(ctl_sk);
565 
566 	if (net_xmit_eval(err) == 0) {
567 		__DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
568 		__DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
569 	}
570 	local_bh_enable();
571 out:
572 	dst_release(dst);
573 }
574 
575 static void dccp_v4_reqsk_destructor(struct request_sock *req)
576 {
577 	dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
578 	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
579 }
580 
581 void dccp_syn_ack_timeout(const struct request_sock *req)
582 {
583 }
584 EXPORT_SYMBOL(dccp_syn_ack_timeout);
585 
586 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
587 	.family		= PF_INET,
588 	.obj_size	= sizeof(struct dccp_request_sock),
589 	.rtx_syn_ack	= dccp_v4_send_response,
590 	.send_ack	= dccp_reqsk_send_ack,
591 	.destructor	= dccp_v4_reqsk_destructor,
592 	.send_reset	= dccp_v4_ctl_send_reset,
593 	.syn_ack_timeout = dccp_syn_ack_timeout,
594 };
595 
596 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
597 {
598 	struct inet_request_sock *ireq;
599 	struct request_sock *req;
600 	struct dccp_request_sock *dreq;
601 	const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
602 	struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
603 
604 	/* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
605 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
606 		return 0;	/* discard, don't send a reset here */
607 
608 	if (dccp_bad_service_code(sk, service)) {
609 		dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
610 		goto drop;
611 	}
612 	/*
613 	 * TW buckets are converted to open requests without
614 	 * limitations, they conserve resources and peer is
615 	 * evidently real one.
616 	 */
617 	dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
618 	if (inet_csk_reqsk_queue_is_full(sk))
619 		goto drop;
620 
621 	if (sk_acceptq_is_full(sk))
622 		goto drop;
623 
624 	req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
625 	if (req == NULL)
626 		goto drop;
627 
628 	if (dccp_reqsk_init(req, dccp_sk(sk), skb))
629 		goto drop_and_free;
630 
631 	dreq = dccp_rsk(req);
632 	if (dccp_parse_options(sk, dreq, skb))
633 		goto drop_and_free;
634 
635 	ireq = inet_rsk(req);
636 	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
637 	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
638 	ireq->ir_mark = inet_request_mark(sk, skb);
639 	ireq->ireq_family = AF_INET;
640 	ireq->ir_iif = READ_ONCE(sk->sk_bound_dev_if);
641 
642 	if (security_inet_conn_request(sk, skb, req))
643 		goto drop_and_free;
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 	if (unlikely(!inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT)))
662 		reqsk_free(req);
663 	else
664 		reqsk_put(req);
665 
666 	return 0;
667 
668 drop_and_free:
669 	reqsk_free(req);
670 drop:
671 	__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
672 	return -1;
673 }
674 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
675 
676 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
677 {
678 	struct dccp_hdr *dh = dccp_hdr(skb);
679 
680 	if (sk->sk_state == DCCP_OPEN) { /* Fast path */
681 		if (dccp_rcv_established(sk, skb, dh, skb->len))
682 			goto reset;
683 		return 0;
684 	}
685 
686 	/*
687 	 *  Step 3: Process LISTEN state
688 	 *	 If P.type == Request or P contains a valid Init Cookie option,
689 	 *	      (* Must scan the packet's options to check for Init
690 	 *		 Cookies.  Only Init Cookies are processed here,
691 	 *		 however; other options are processed in Step 8.  This
692 	 *		 scan need only be performed if the endpoint uses Init
693 	 *		 Cookies *)
694 	 *	      (* Generate a new socket and switch to that socket *)
695 	 *	      Set S := new socket for this port pair
696 	 *	      S.state = RESPOND
697 	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
698 	 *	      Initialize S.GAR := S.ISS
699 	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
700 	 *	      Continue with S.state == RESPOND
701 	 *	      (* A Response packet will be generated in Step 11 *)
702 	 *	 Otherwise,
703 	 *	      Generate Reset(No Connection) unless P.type == Reset
704 	 *	      Drop packet and return
705 	 *
706 	 * NOTE: the check for the packet types is done in
707 	 *	 dccp_rcv_state_process
708 	 */
709 
710 	if (dccp_rcv_state_process(sk, skb, dh, skb->len))
711 		goto reset;
712 	return 0;
713 
714 reset:
715 	dccp_v4_ctl_send_reset(sk, skb, SK_RST_REASON_NOT_SPECIFIED);
716 	kfree_skb(skb);
717 	return 0;
718 }
719 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
720 
721 /**
722  *	dccp_invalid_packet  -  check for malformed packets
723  *	@skb: Packet to validate
724  *
725  *	Implements RFC 4340, 8.5:  Step 1: Check header basics
726  *	Packets that fail these checks are ignored and do not receive Resets.
727  */
728 int dccp_invalid_packet(struct sk_buff *skb)
729 {
730 	const struct dccp_hdr *dh;
731 	unsigned int cscov;
732 	u8 dccph_doff;
733 
734 	if (skb->pkt_type != PACKET_HOST)
735 		return 1;
736 
737 	/* If the packet is shorter than 12 bytes, drop packet and return */
738 	if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
739 		DCCP_WARN("pskb_may_pull failed\n");
740 		return 1;
741 	}
742 
743 	dh = dccp_hdr(skb);
744 
745 	/* If P.type is not understood, drop packet and return */
746 	if (dh->dccph_type >= DCCP_PKT_INVALID) {
747 		DCCP_WARN("invalid packet type\n");
748 		return 1;
749 	}
750 
751 	/*
752 	 * If P.Data Offset is too small for packet type, drop packet and return
753 	 */
754 	dccph_doff = dh->dccph_doff;
755 	if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
756 		DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
757 		return 1;
758 	}
759 	/*
760 	 * If P.Data Offset is too large for packet, drop packet and return
761 	 */
762 	if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
763 		DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
764 		return 1;
765 	}
766 	dh = dccp_hdr(skb);
767 	/*
768 	 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
769 	 * has short sequence numbers), drop packet and return
770 	 */
771 	if ((dh->dccph_type < DCCP_PKT_DATA    ||
772 	    dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
773 		DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
774 			  dccp_packet_name(dh->dccph_type));
775 		return 1;
776 	}
777 
778 	/*
779 	 * If P.CsCov is too large for the packet size, drop packet and return.
780 	 * This must come _before_ checksumming (not as RFC 4340 suggests).
781 	 */
782 	cscov = dccp_csum_coverage(skb);
783 	if (cscov > skb->len) {
784 		DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
785 			  dh->dccph_cscov, skb->len);
786 		return 1;
787 	}
788 
789 	/* If header checksum is incorrect, drop packet and return.
790 	 * (This step is completed in the AF-dependent functions.) */
791 	skb->csum = skb_checksum(skb, 0, cscov, 0);
792 
793 	return 0;
794 }
795 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
796 
797 /* this is called when real data arrives */
798 static int dccp_v4_rcv(struct sk_buff *skb)
799 {
800 	const struct dccp_hdr *dh;
801 	const struct iphdr *iph;
802 	bool refcounted;
803 	struct sock *sk;
804 	int min_cov;
805 
806 	/* Step 1: Check header basics */
807 
808 	if (dccp_invalid_packet(skb))
809 		goto discard_it;
810 
811 	iph = ip_hdr(skb);
812 	/* Step 1: If header checksum is incorrect, drop packet and return */
813 	if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
814 		DCCP_WARN("dropped packet with invalid checksum\n");
815 		goto discard_it;
816 	}
817 
818 	dh = dccp_hdr(skb);
819 
820 	DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
821 	DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
822 
823 	dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
824 		      dccp_packet_name(dh->dccph_type),
825 		      &iph->saddr, ntohs(dh->dccph_sport),
826 		      &iph->daddr, ntohs(dh->dccph_dport),
827 		      (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
828 
829 	if (dccp_packet_without_ack(skb)) {
830 		DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
831 		dccp_pr_debug_cat("\n");
832 	} else {
833 		DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
834 		dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
835 				  DCCP_SKB_CB(skb)->dccpd_ack_seq);
836 	}
837 
838 lookup:
839 	sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
840 			       dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
841 	if (!sk) {
842 		dccp_pr_debug("failed to look up flow ID in table and "
843 			      "get corresponding socket\n");
844 		goto no_dccp_socket;
845 	}
846 
847 	/*
848 	 * Step 2:
849 	 *	... or S.state == TIMEWAIT,
850 	 *		Generate Reset(No Connection) unless P.type == Reset
851 	 *		Drop packet and return
852 	 */
853 	if (sk->sk_state == DCCP_TIME_WAIT) {
854 		dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
855 		inet_twsk_put(inet_twsk(sk));
856 		goto no_dccp_socket;
857 	}
858 
859 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
860 		struct request_sock *req = inet_reqsk(sk);
861 		struct sock *nsk;
862 
863 		sk = req->rsk_listener;
864 		if (unlikely(sk->sk_state != DCCP_LISTEN)) {
865 			inet_csk_reqsk_queue_drop_and_put(sk, req);
866 			goto lookup;
867 		}
868 		sock_hold(sk);
869 		refcounted = true;
870 		nsk = dccp_check_req(sk, skb, req);
871 		if (!nsk) {
872 			reqsk_put(req);
873 			goto discard_and_relse;
874 		}
875 		if (nsk == sk) {
876 			reqsk_put(req);
877 		} else if (dccp_child_process(sk, nsk, skb)) {
878 			dccp_v4_ctl_send_reset(sk, skb, SK_RST_REASON_NOT_SPECIFIED);
879 			goto discard_and_relse;
880 		} else {
881 			sock_put(sk);
882 			return 0;
883 		}
884 	}
885 	/*
886 	 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
887 	 *	o if MinCsCov = 0, only packets with CsCov = 0 are accepted
888 	 *	o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
889 	 */
890 	min_cov = dccp_sk(sk)->dccps_pcrlen;
891 	if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
892 		dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
893 			      dh->dccph_cscov, min_cov);
894 		/* FIXME: "Such packets SHOULD be reported using Data Dropped
895 		 *         options (Section 11.7) with Drop Code 0, Protocol
896 		 *         Constraints."                                     */
897 		goto discard_and_relse;
898 	}
899 
900 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
901 		goto discard_and_relse;
902 	nf_reset_ct(skb);
903 
904 	return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
905 
906 no_dccp_socket:
907 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
908 		goto discard_it;
909 	/*
910 	 * Step 2:
911 	 *	If no socket ...
912 	 *		Generate Reset(No Connection) unless P.type == Reset
913 	 *		Drop packet and return
914 	 */
915 	if (dh->dccph_type != DCCP_PKT_RESET) {
916 		DCCP_SKB_CB(skb)->dccpd_reset_code =
917 					DCCP_RESET_CODE_NO_CONNECTION;
918 		dccp_v4_ctl_send_reset(sk, skb, SK_RST_REASON_NOT_SPECIFIED);
919 	}
920 
921 discard_it:
922 	kfree_skb(skb);
923 	return 0;
924 
925 discard_and_relse:
926 	if (refcounted)
927 		sock_put(sk);
928 	goto discard_it;
929 }
930 
931 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
932 	.queue_xmit	   = ip_queue_xmit,
933 	.send_check	   = dccp_v4_send_check,
934 	.rebuild_header	   = inet_sk_rebuild_header,
935 	.conn_request	   = dccp_v4_conn_request,
936 	.syn_recv_sock	   = dccp_v4_request_recv_sock,
937 	.net_header_len	   = sizeof(struct iphdr),
938 	.setsockopt	   = ip_setsockopt,
939 	.getsockopt	   = ip_getsockopt,
940 	.addr2sockaddr	   = inet_csk_addr2sockaddr,
941 	.sockaddr_len	   = sizeof(struct sockaddr_in),
942 };
943 
944 static int dccp_v4_init_sock(struct sock *sk)
945 {
946 	static __u8 dccp_v4_ctl_sock_initialized;
947 	int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
948 
949 	if (err == 0) {
950 		if (unlikely(!dccp_v4_ctl_sock_initialized))
951 			dccp_v4_ctl_sock_initialized = 1;
952 		inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
953 	}
954 
955 	return err;
956 }
957 
958 static struct timewait_sock_ops dccp_timewait_sock_ops = {
959 	.twsk_obj_size	= sizeof(struct inet_timewait_sock),
960 };
961 
962 static struct proto dccp_v4_prot = {
963 	.name			= "DCCP",
964 	.owner			= THIS_MODULE,
965 	.close			= dccp_close,
966 	.connect		= dccp_v4_connect,
967 	.disconnect		= dccp_disconnect,
968 	.ioctl			= dccp_ioctl,
969 	.init			= dccp_v4_init_sock,
970 	.setsockopt		= dccp_setsockopt,
971 	.getsockopt		= dccp_getsockopt,
972 	.sendmsg		= dccp_sendmsg,
973 	.recvmsg		= dccp_recvmsg,
974 	.backlog_rcv		= dccp_v4_do_rcv,
975 	.hash			= inet_hash,
976 	.unhash			= inet_unhash,
977 	.accept			= inet_csk_accept,
978 	.get_port		= inet_csk_get_port,
979 	.shutdown		= dccp_shutdown,
980 	.destroy		= dccp_destroy_sock,
981 	.orphan_count		= &dccp_orphan_count,
982 	.max_header		= MAX_DCCP_HEADER,
983 	.obj_size		= sizeof(struct dccp_sock),
984 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
985 	.rsk_prot		= &dccp_request_sock_ops,
986 	.twsk_prot		= &dccp_timewait_sock_ops,
987 	.h.hashinfo		= &dccp_hashinfo,
988 };
989 
990 static const struct net_protocol dccp_v4_protocol = {
991 	.handler	= dccp_v4_rcv,
992 	.err_handler	= dccp_v4_err,
993 	.no_policy	= 1,
994 	.icmp_strict_tag_validation = 1,
995 };
996 
997 static const struct proto_ops inet_dccp_ops = {
998 	.family		   = PF_INET,
999 	.owner		   = THIS_MODULE,
1000 	.release	   = inet_release,
1001 	.bind		   = inet_bind,
1002 	.connect	   = inet_stream_connect,
1003 	.socketpair	   = sock_no_socketpair,
1004 	.accept		   = inet_accept,
1005 	.getname	   = inet_getname,
1006 	/* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1007 	.poll		   = dccp_poll,
1008 	.ioctl		   = inet_ioctl,
1009 	.gettstamp	   = sock_gettstamp,
1010 	/* FIXME: work on inet_listen to rename it to sock_common_listen */
1011 	.listen		   = inet_dccp_listen,
1012 	.shutdown	   = inet_shutdown,
1013 	.setsockopt	   = sock_common_setsockopt,
1014 	.getsockopt	   = sock_common_getsockopt,
1015 	.sendmsg	   = inet_sendmsg,
1016 	.recvmsg	   = sock_common_recvmsg,
1017 	.mmap		   = sock_no_mmap,
1018 };
1019 
1020 static struct inet_protosw dccp_v4_protosw = {
1021 	.type		= SOCK_DCCP,
1022 	.protocol	= IPPROTO_DCCP,
1023 	.prot		= &dccp_v4_prot,
1024 	.ops		= &inet_dccp_ops,
1025 	.flags		= INET_PROTOSW_ICSK,
1026 };
1027 
1028 static int __net_init dccp_v4_init_net(struct net *net)
1029 {
1030 	struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1031 
1032 	if (dccp_hashinfo.bhash == NULL)
1033 		return -ESOCKTNOSUPPORT;
1034 
1035 	return inet_ctl_sock_create(&pn->v4_ctl_sk, PF_INET,
1036 				    SOCK_DCCP, IPPROTO_DCCP, net);
1037 }
1038 
1039 static void __net_exit dccp_v4_exit_net(struct net *net)
1040 {
1041 	struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1042 
1043 	inet_ctl_sock_destroy(pn->v4_ctl_sk);
1044 }
1045 
1046 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1047 {
1048 	inet_twsk_purge(&dccp_hashinfo);
1049 }
1050 
1051 static struct pernet_operations dccp_v4_ops = {
1052 	.init	= dccp_v4_init_net,
1053 	.exit	= dccp_v4_exit_net,
1054 	.exit_batch = dccp_v4_exit_batch,
1055 	.id	= &dccp_v4_pernet_id,
1056 	.size   = sizeof(struct dccp_v4_pernet),
1057 };
1058 
1059 static int __init dccp_v4_init(void)
1060 {
1061 	int err = proto_register(&dccp_v4_prot, 1);
1062 
1063 	if (err)
1064 		goto out;
1065 
1066 	inet_register_protosw(&dccp_v4_protosw);
1067 
1068 	err = register_pernet_subsys(&dccp_v4_ops);
1069 	if (err)
1070 		goto out_destroy_ctl_sock;
1071 
1072 	err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1073 	if (err)
1074 		goto out_proto_unregister;
1075 
1076 out:
1077 	return err;
1078 out_proto_unregister:
1079 	unregister_pernet_subsys(&dccp_v4_ops);
1080 out_destroy_ctl_sock:
1081 	inet_unregister_protosw(&dccp_v4_protosw);
1082 	proto_unregister(&dccp_v4_prot);
1083 	goto out;
1084 }
1085 
1086 static void __exit dccp_v4_exit(void)
1087 {
1088 	inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1089 	unregister_pernet_subsys(&dccp_v4_ops);
1090 	inet_unregister_protosw(&dccp_v4_protosw);
1091 	proto_unregister(&dccp_v4_prot);
1092 }
1093 
1094 module_init(dccp_v4_init);
1095 module_exit(dccp_v4_exit);
1096 
1097 /*
1098  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1099  * values directly, Also cover the case where the protocol is not specified,
1100  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1101  */
1102 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1103 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1104 MODULE_LICENSE("GPL");
1105 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1106 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1107