4 * An implementation of the DCCP protocol
5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
13 #include <linux/config.h>
14 #include <linux/dccp.h>
15 #include <linux/icmp.h>
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/random.h>
21 #include <net/inet_hashtables.h>
23 #include <net/tcp_states.h>
30 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
31 .lhash_lock = RW_LOCK_UNLOCKED,
32 .lhash_users = ATOMIC_INIT(0),
33 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
36 EXPORT_SYMBOL_GPL(dccp_hashinfo);
38 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
40 return inet_csk_get_port(&dccp_hashinfo, sk, snum,
41 inet_csk_bind_conflict);
44 static void dccp_v4_hash(struct sock *sk)
46 inet_hash(&dccp_hashinfo, sk);
49 static void dccp_v4_unhash(struct sock *sk)
51 inet_unhash(&dccp_hashinfo, sk);
54 /* called with local bh disabled */
55 static int __dccp_v4_check_established(struct sock *sk, const __u16 lport,
56 struct inet_timewait_sock **twp)
58 struct inet_sock *inet = inet_sk(sk);
59 const u32 daddr = inet->rcv_saddr;
60 const u32 saddr = inet->daddr;
61 const int dif = sk->sk_bound_dev_if;
62 INET_ADDR_COOKIE(acookie, saddr, daddr)
63 const __u32 ports = INET_COMBINED_PORTS(inet->dport, lport);
64 unsigned int hash = inet_ehashfn(daddr, lport, saddr, inet->dport);
65 struct inet_ehash_bucket *head = inet_ehash_bucket(&dccp_hashinfo, hash);
66 const struct sock *sk2;
67 const struct hlist_node *node;
68 struct inet_timewait_sock *tw;
70 prefetch(head->chain.first);
71 write_lock(&head->lock);
73 /* Check TIME-WAIT sockets first. */
74 sk_for_each(sk2, node, &(head + dccp_hashinfo.ehash_size)->chain) {
77 if (INET_TW_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif))
82 /* And established part... */
83 sk_for_each(sk2, node, &head->chain) {
84 if (INET_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif))
88 /* Must record num and sport now. Otherwise we will see
89 * in hash table socket with a funny identity. */
91 inet->sport = htons(lport);
93 BUG_TRAP(sk_unhashed(sk));
94 __sk_add_node(sk, &head->chain);
95 sock_prot_inc_use(sk->sk_prot);
96 write_unlock(&head->lock);
100 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
101 } else if (tw != NULL) {
102 /* Silly. Should hash-dance instead... */
103 inet_twsk_deschedule(tw, &dccp_death_row);
104 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
112 write_unlock(&head->lock);
113 return -EADDRNOTAVAIL;
117 * Bind a port for a connect operation and hash it.
119 static int dccp_v4_hash_connect(struct sock *sk)
121 const unsigned short snum = inet_sk(sk)->num;
122 struct inet_bind_hashbucket *head;
123 struct inet_bind_bucket *tb;
127 int low = sysctl_local_port_range[0];
128 int high = sysctl_local_port_range[1];
129 int remaining = (high - low) + 1;
130 int rover = net_random() % (high - low) + low;
131 struct hlist_node *node;
132 struct inet_timewait_sock *tw = NULL;
136 head = &dccp_hashinfo.bhash[inet_bhashfn(rover,
137 dccp_hashinfo.bhash_size)];
138 spin_lock(&head->lock);
140 /* Does not bother with rcv_saddr checks,
141 * because the established check is already
144 inet_bind_bucket_for_each(tb, node, &head->chain) {
145 if (tb->port == rover) {
146 BUG_TRAP(!hlist_empty(&tb->owners));
147 if (tb->fastreuse >= 0)
149 if (!__dccp_v4_check_established(sk,
157 tb = inet_bind_bucket_create(dccp_hashinfo.bind_bucket_cachep,
160 spin_unlock(&head->lock);
167 spin_unlock(&head->lock);
170 } while (--remaining > 0);
174 return -EADDRNOTAVAIL;
177 /* All locks still held and bhs disabled */
178 inet_bind_hash(sk, tb, rover);
179 if (sk_unhashed(sk)) {
180 inet_sk(sk)->sport = htons(rover);
181 __inet_hash(&dccp_hashinfo, sk, 0);
183 spin_unlock(&head->lock);
186 inet_twsk_deschedule(tw, &dccp_death_row);
194 head = &dccp_hashinfo.bhash[inet_bhashfn(snum,
195 dccp_hashinfo.bhash_size)];
196 tb = inet_csk(sk)->icsk_bind_hash;
197 spin_lock_bh(&head->lock);
198 if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
199 __inet_hash(&dccp_hashinfo, sk, 0);
200 spin_unlock_bh(&head->lock);
203 spin_unlock(&head->lock);
204 /* No definite answer... Walk to established hash table */
205 ret = __dccp_v4_check_established(sk, snum, NULL);
212 static int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
215 struct inet_sock *inet = inet_sk(sk);
216 struct dccp_sock *dp = dccp_sk(sk);
217 const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
223 dp->dccps_role = DCCP_ROLE_CLIENT;
225 if (dccp_service_not_initialized(sk))
228 if (addr_len < sizeof(struct sockaddr_in))
231 if (usin->sin_family != AF_INET)
232 return -EAFNOSUPPORT;
234 nexthop = daddr = usin->sin_addr.s_addr;
235 if (inet->opt != NULL && inet->opt->srr) {
238 nexthop = inet->opt->faddr;
241 tmp = ip_route_connect(&rt, nexthop, inet->saddr,
242 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
244 inet->sport, usin->sin_port, sk);
248 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
253 if (inet->opt == NULL || !inet->opt->srr)
256 if (inet->saddr == 0)
257 inet->saddr = rt->rt_src;
258 inet->rcv_saddr = inet->saddr;
260 inet->dport = usin->sin_port;
263 dp->dccps_ext_header_len = 0;
264 if (inet->opt != NULL)
265 dp->dccps_ext_header_len = inet->opt->optlen;
267 * Socket identity is still unknown (sport may be zero).
268 * However we set state to DCCP_REQUESTING and not releasing socket
269 * lock select source port, enter ourselves into the hash tables and
270 * complete initialization after this.
272 dccp_set_state(sk, DCCP_REQUESTING);
273 err = dccp_v4_hash_connect(sk);
277 err = ip_route_newports(&rt, inet->sport, inet->dport, sk);
281 /* OK, now commit destination to socket. */
282 sk_setup_caps(sk, &rt->u.dst);
285 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
289 dccp_update_gss(sk, dp->dccps_iss);
292 * SWL and AWL are initially adjusted so that they are not less than
293 * the initial Sequence Numbers received and sent, respectively:
294 * SWL := max(GSR + 1 - floor(W/4), ISR),
295 * AWL := max(GSS - W' + 1, ISS).
296 * These adjustments MUST be applied only at the beginning of the
299 dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
301 inet->id = dp->dccps_iss ^ jiffies;
303 err = dccp_connect(sk);
311 * This unhashes the socket and releases the local port, if necessary.
313 dccp_set_state(sk, DCCP_CLOSED);
315 sk->sk_route_caps = 0;
321 * This routine does path mtu discovery as defined in RFC1191.
323 static inline void dccp_do_pmtu_discovery(struct sock *sk,
324 const struct iphdr *iph,
327 struct dst_entry *dst;
328 const struct inet_sock *inet = inet_sk(sk);
329 const struct dccp_sock *dp = dccp_sk(sk);
331 /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
332 * send out by Linux are always < 576bytes so they should go through
335 if (sk->sk_state == DCCP_LISTEN)
338 /* We don't check in the destentry if pmtu discovery is forbidden
339 * on this route. We just assume that no packet_to_big packets
340 * are send back when pmtu discovery is not active.
341 * There is a small race when the user changes this flag in the
342 * route, but I think that's acceptable.
344 if ((dst = __sk_dst_check(sk, 0)) == NULL)
347 dst->ops->update_pmtu(dst, mtu);
349 /* Something is about to be wrong... Remember soft error
350 * for the case, if this connection will not able to recover.
352 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
353 sk->sk_err_soft = EMSGSIZE;
357 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
358 dp->dccps_pmtu_cookie > mtu) {
359 dccp_sync_mss(sk, mtu);
362 * From: draft-ietf-dccp-spec-11.txt
364 * DCCP-Sync packets are the best choice for upward
365 * probing, since DCCP-Sync probes do not risk application
368 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
369 } /* else let the usual retransmit timer handle it */
372 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
375 struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
376 const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
377 sizeof(struct dccp_hdr_ext) +
378 sizeof(struct dccp_hdr_ack_bits);
381 if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
384 skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
388 /* Reserve space for headers. */
389 skb_reserve(skb, MAX_DCCP_HEADER);
391 skb->dst = dst_clone(rxskb->dst);
393 skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
395 memset(dh, 0, dccp_hdr_ack_len);
397 /* Build DCCP header and checksum it. */
398 dh->dccph_type = DCCP_PKT_ACK;
399 dh->dccph_sport = rxdh->dccph_dport;
400 dh->dccph_dport = rxdh->dccph_sport;
401 dh->dccph_doff = dccp_hdr_ack_len / 4;
404 dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
405 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
406 DCCP_SKB_CB(rxskb)->dccpd_seq);
408 bh_lock_sock(dccp_ctl_socket->sk);
409 err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
410 rxskb->nh.iph->daddr,
411 rxskb->nh.iph->saddr, NULL);
412 bh_unlock_sock(dccp_ctl_socket->sk);
414 if (err == NET_XMIT_CN || err == 0) {
415 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
416 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
420 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
421 struct request_sock *req)
423 dccp_v4_ctl_send_ack(skb);
426 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
427 struct dst_entry *dst)
432 /* First, grab a route. */
434 if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
437 skb = dccp_make_response(sk, dst, req);
439 const struct inet_request_sock *ireq = inet_rsk(req);
441 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
442 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
445 if (err == NET_XMIT_CN)
455 * This routine is called by the ICMP module when it gets some sort of error
456 * condition. If err < 0 then the socket should be closed and the error
457 * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
458 * After adjustment header points to the first 8 bytes of the tcp header. We
459 * need to find the appropriate port.
461 * The locking strategy used here is very "optimistic". When someone else
462 * accesses the socket the ICMP is just dropped and for some paths there is no
463 * check at all. A more general error queue to queue errors for later handling
464 * is probably better.
466 void dccp_v4_err(struct sk_buff *skb, u32 info)
468 const struct iphdr *iph = (struct iphdr *)skb->data;
469 const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
471 struct dccp_sock *dp;
472 struct inet_sock *inet;
473 const int type = skb->h.icmph->type;
474 const int code = skb->h.icmph->code;
479 if (skb->len < (iph->ihl << 2) + 8) {
480 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
484 sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
485 iph->saddr, dh->dccph_sport, inet_iif(skb));
487 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
491 if (sk->sk_state == DCCP_TIME_WAIT) {
492 inet_twsk_put((struct inet_timewait_sock *)sk);
497 /* If too many ICMPs get dropped on busy
498 * servers this needs to be solved differently.
500 if (sock_owned_by_user(sk))
501 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
503 if (sk->sk_state == DCCP_CLOSED)
507 seq = dccp_hdr_seq(skb);
508 if (sk->sk_state != DCCP_LISTEN &&
509 !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
510 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
515 case ICMP_SOURCE_QUENCH:
516 /* Just silently ignore these. */
518 case ICMP_PARAMETERPROB:
521 case ICMP_DEST_UNREACH:
522 if (code > NR_ICMP_UNREACH)
525 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
526 if (!sock_owned_by_user(sk))
527 dccp_do_pmtu_discovery(sk, iph, info);
531 err = icmp_err_convert[code].errno;
533 case ICMP_TIME_EXCEEDED:
540 switch (sk->sk_state) {
541 struct request_sock *req , **prev;
543 if (sock_owned_by_user(sk))
545 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
546 iph->daddr, iph->saddr);
551 * ICMPs are not backlogged, hence we cannot get an established
556 if (seq != dccp_rsk(req)->dreq_iss) {
557 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
561 * Still in RESPOND, just remove it silently.
562 * There is no good way to pass the error to the newly
563 * created socket, and POSIX does not want network
564 * errors returned from accept().
566 inet_csk_reqsk_queue_drop(sk, req, prev);
569 case DCCP_REQUESTING:
571 if (!sock_owned_by_user(sk)) {
572 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
575 sk->sk_error_report(sk);
579 sk->sk_err_soft = err;
583 /* If we've already connected we will keep trying
584 * until we time out, or the user gives up.
586 * rfc1122 4.2.3.9 allows to consider as hard errors
587 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
588 * but it is obsoleted by pmtu discovery).
590 * Note, that in modern internet, where routing is unreliable
591 * and in each dark corner broken firewalls sit, sending random
592 * errors ordered by their masters even this two messages finally lose
593 * their original sense (even Linux sends invalid PORT_UNREACHs)
595 * Now we are in compliance with RFCs.
600 if (!sock_owned_by_user(sk) && inet->recverr) {
602 sk->sk_error_report(sk);
603 } else /* Only an error on timeout */
604 sk->sk_err_soft = err;
610 /* This routine computes an IPv4 DCCP checksum. */
611 static void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
613 const struct inet_sock *inet = inet_sk(sk);
614 struct dccp_hdr *dh = dccp_hdr(skb);
616 dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
619 int dccp_v4_send_reset(struct sock *sk, enum dccp_reset_codes code)
623 * FIXME: what if rebuild_header fails?
624 * Should we be doing a rebuild_header here?
626 int err = inet_sk_rebuild_header(sk);
631 skb = dccp_make_reset(sk, sk->sk_dst_cache, code);
633 const struct inet_sock *inet = inet_sk(sk);
635 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
636 err = ip_build_and_send_pkt(skb, sk,
637 inet->saddr, inet->daddr, NULL);
638 if (err == NET_XMIT_CN)
645 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
646 const struct sk_buff *skb)
648 return secure_dccp_sequence_number(skb->nh.iph->daddr,
650 dccp_hdr(skb)->dccph_dport,
651 dccp_hdr(skb)->dccph_sport);
654 static inline int dccp_bad_service_code(const struct sock *sk,
657 const struct dccp_sock *dp = dccp_sk(sk);
659 if (dp->dccps_service == service)
661 return !dccp_list_has_service(dp->dccps_service_list, service);
664 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
666 struct inet_request_sock *ireq;
668 struct request_sock *req;
669 struct dccp_request_sock *dreq;
670 const __u32 saddr = skb->nh.iph->saddr;
671 const __u32 daddr = skb->nh.iph->daddr;
672 const __u32 service = dccp_hdr_request(skb)->dccph_req_service;
673 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
674 __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
675 struct dst_entry *dst = NULL;
677 /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
678 if (((struct rtable *)skb->dst)->rt_flags &
679 (RTCF_BROADCAST | RTCF_MULTICAST)) {
680 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
684 if (dccp_bad_service_code(sk, service)) {
685 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
689 * TW buckets are converted to open requests without
690 * limitations, they conserve resources and peer is
691 * evidently real one.
693 if (inet_csk_reqsk_queue_is_full(sk))
697 * Accept backlog is full. If we have already queued enough
698 * of warm entries in syn queue, drop request. It is better than
699 * clogging syn queue with openreqs with exponentially increasing
702 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
705 req = reqsk_alloc(sk->sk_prot->rsk_prot);
709 /* FIXME: process options */
711 dccp_openreq_init(req, &dp, skb);
713 ireq = inet_rsk(req);
714 ireq->loc_addr = daddr;
715 ireq->rmt_addr = saddr;
716 /* FIXME: Merge Aristeu's option parsing code when ready */
717 req->rcv_wnd = 100; /* Fake, option parsing will get the
722 * Step 3: Process LISTEN state
724 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
726 * In fact we defer setting S.GSR, S.SWL, S.SWH to
727 * dccp_create_openreq_child.
729 dreq = dccp_rsk(req);
730 dreq->dreq_isr = dcb->dccpd_seq;
731 dreq->dreq_iss = dccp_v4_init_sequence(sk, skb);
732 dreq->dreq_service = service;
734 if (dccp_v4_send_response(sk, req, dst))
737 inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
742 * FIXME: should be reqsk_free after implementing req->rsk_ops
746 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
747 dcb->dccpd_reset_code = reset_code;
752 * The three way handshake has completed - we got a valid ACK or DATAACK -
753 * now create the new socket.
755 * This is the equivalent of TCP's tcp_v4_syn_recv_sock
757 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
758 struct request_sock *req,
759 struct dst_entry *dst)
761 struct inet_request_sock *ireq;
762 struct inet_sock *newinet;
763 struct dccp_sock *newdp;
766 if (sk_acceptq_is_full(sk))
769 if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
772 newsk = dccp_create_openreq_child(sk, req, skb);
776 sk_setup_caps(newsk, dst);
778 newdp = dccp_sk(newsk);
779 newinet = inet_sk(newsk);
780 ireq = inet_rsk(req);
781 newinet->daddr = ireq->rmt_addr;
782 newinet->rcv_saddr = ireq->loc_addr;
783 newinet->saddr = ireq->loc_addr;
784 newinet->opt = ireq->opt;
786 newinet->mc_index = inet_iif(skb);
787 newinet->mc_ttl = skb->nh.iph->ttl;
788 newinet->id = jiffies;
790 dccp_sync_mss(newsk, dst_mtu(dst));
792 __inet_hash(&dccp_hashinfo, newsk, 0);
793 __inet_inherit_port(&dccp_hashinfo, sk, newsk);
798 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
800 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
805 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
807 const struct dccp_hdr *dh = dccp_hdr(skb);
808 const struct iphdr *iph = skb->nh.iph;
810 struct request_sock **prev;
811 /* Find possible connection requests. */
812 struct request_sock *req = inet_csk_search_req(sk, &prev,
814 iph->saddr, iph->daddr);
816 return dccp_check_req(sk, skb, req, prev);
818 nsk = __inet_lookup_established(&dccp_hashinfo,
819 iph->saddr, dh->dccph_sport,
820 iph->daddr, ntohs(dh->dccph_dport),
823 if (nsk->sk_state != DCCP_TIME_WAIT) {
827 inet_twsk_put((struct inet_timewait_sock *)nsk);
834 int dccp_v4_checksum(const struct sk_buff *skb, const u32 saddr,
837 const struct dccp_hdr* dh = dccp_hdr(skb);
841 if (dh->dccph_cscov == 0)
842 checksum_len = skb->len;
844 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
845 checksum_len = checksum_len < skb->len ? checksum_len :
849 tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
850 return csum_tcpudp_magic(saddr, daddr, checksum_len,
854 static int dccp_v4_verify_checksum(struct sk_buff *skb,
855 const u32 saddr, const u32 daddr)
857 struct dccp_hdr *dh = dccp_hdr(skb);
861 if (dh->dccph_cscov == 0)
862 checksum_len = skb->len;
864 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
865 checksum_len = checksum_len < skb->len ? checksum_len :
868 tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
869 return csum_tcpudp_magic(saddr, daddr, checksum_len,
870 IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
873 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
877 struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
879 { .daddr = skb->nh.iph->saddr,
880 .saddr = skb->nh.iph->daddr,
881 .tos = RT_CONN_FLAGS(sk) } },
882 .proto = sk->sk_protocol,
884 { .sport = dccp_hdr(skb)->dccph_dport,
885 .dport = dccp_hdr(skb)->dccph_sport }
889 if (ip_route_output_flow(&rt, &fl, sk, 0)) {
890 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
897 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
900 struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
901 const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
902 sizeof(struct dccp_hdr_ext) +
903 sizeof(struct dccp_hdr_reset);
905 struct dst_entry *dst;
908 /* Never send a reset in response to a reset. */
909 if (rxdh->dccph_type == DCCP_PKT_RESET)
912 if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
915 dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
919 skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
923 /* Reserve space for headers. */
924 skb_reserve(skb, MAX_DCCP_HEADER);
925 skb->dst = dst_clone(dst);
927 skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
929 memset(dh, 0, dccp_hdr_reset_len);
931 /* Build DCCP header and checksum it. */
932 dh->dccph_type = DCCP_PKT_RESET;
933 dh->dccph_sport = rxdh->dccph_dport;
934 dh->dccph_dport = rxdh->dccph_sport;
935 dh->dccph_doff = dccp_hdr_reset_len / 4;
937 dccp_hdr_reset(skb)->dccph_reset_code =
938 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
940 /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
942 if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
943 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
945 dccp_hdr_set_seq(dh, seqno);
946 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
947 DCCP_SKB_CB(rxskb)->dccpd_seq);
949 dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
950 rxskb->nh.iph->daddr);
952 bh_lock_sock(dccp_ctl_socket->sk);
953 err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
954 rxskb->nh.iph->daddr,
955 rxskb->nh.iph->saddr, NULL);
956 bh_unlock_sock(dccp_ctl_socket->sk);
958 if (err == NET_XMIT_CN || err == 0) {
959 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
960 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
966 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
968 struct dccp_hdr *dh = dccp_hdr(skb);
970 if (sk->sk_state == DCCP_OPEN) { /* Fast path */
971 if (dccp_rcv_established(sk, skb, dh, skb->len))
977 * Step 3: Process LISTEN state
978 * If S.state == LISTEN,
979 * If P.type == Request or P contains a valid Init Cookie
981 * * Must scan the packet's options to check for an Init
982 * Cookie. Only the Init Cookie is processed here,
983 * however; other options are processed in Step 8. This
984 * scan need only be performed if the endpoint uses Init
986 * * Generate a new socket and switch to that socket *
987 * Set S := new socket for this port pair
989 * Choose S.ISS (initial seqno) or set from Init Cookie
990 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
991 * Continue with S.state == RESPOND
992 * * A Response packet will be generated in Step 11 *
994 * Generate Reset(No Connection) unless P.type == Reset
995 * Drop packet and return
997 * NOTE: the check for the packet types is done in
998 * dccp_rcv_state_process
1000 if (sk->sk_state == DCCP_LISTEN) {
1001 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
1007 if (dccp_child_process(sk, nsk, skb))
1013 if (dccp_rcv_state_process(sk, skb, dh, skb->len))
1018 dccp_v4_ctl_send_reset(skb);
1024 static inline int dccp_invalid_packet(struct sk_buff *skb)
1026 const struct dccp_hdr *dh;
1028 if (skb->pkt_type != PACKET_HOST)
1031 if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
1032 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
1038 /* If the packet type is not understood, drop packet and return */
1039 if (dh->dccph_type >= DCCP_PKT_INVALID) {
1040 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
1045 * If P.Data Offset is too small for packet type, or too large for
1046 * packet, drop packet and return
1048 if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
1049 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
1055 if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
1056 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
1065 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
1066 * has short sequence numbers), drop packet and return
1068 if (dh->dccph_x == 0 &&
1069 dh->dccph_type != DCCP_PKT_DATA &&
1070 dh->dccph_type != DCCP_PKT_ACK &&
1071 dh->dccph_type != DCCP_PKT_DATAACK) {
1072 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
1073 "nor DataAck and P.X == 0\n",
1074 dccp_packet_name(dh->dccph_type));
1078 /* If the header checksum is incorrect, drop packet and return */
1079 if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
1080 skb->nh.iph->daddr) < 0) {
1081 LIMIT_NETDEBUG(KERN_WARNING "DCCP: header checksum is "
1089 /* this is called when real data arrives */
1090 int dccp_v4_rcv(struct sk_buff *skb)
1092 const struct dccp_hdr *dh;
1096 /* Step 1: Check header basics: */
1098 if (dccp_invalid_packet(skb))
1103 DCCP_SKB_CB(skb)->dccpd_seq = dccp_hdr_seq(skb);
1104 DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
1106 dccp_pr_debug("%8.8s "
1107 "src=%u.%u.%u.%u@%-5d "
1108 "dst=%u.%u.%u.%u@%-5d seq=%llu",
1109 dccp_packet_name(dh->dccph_type),
1110 NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
1111 NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
1112 (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
1114 if (dccp_packet_without_ack(skb)) {
1115 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
1116 dccp_pr_debug_cat("\n");
1118 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
1119 dccp_pr_debug_cat(", ack=%llu\n",
1120 (unsigned long long)
1121 DCCP_SKB_CB(skb)->dccpd_ack_seq);
1125 * Look up flow ID in table and get corresponding socket */
1126 sk = __inet_lookup(&dccp_hashinfo,
1127 skb->nh.iph->saddr, dh->dccph_sport,
1128 skb->nh.iph->daddr, ntohs(dh->dccph_dport),
1134 * Generate Reset(No Connection) unless P.type == Reset
1135 * Drop packet and return
1138 dccp_pr_debug("failed to look up flow ID in table and "
1139 "get corresponding socket\n");
1140 goto no_dccp_socket;
1145 * ... or S.state == TIMEWAIT,
1146 * Generate Reset(No Connection) unless P.type == Reset
1147 * Drop packet and return
1150 if (sk->sk_state == DCCP_TIME_WAIT) {
1151 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
1156 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
1157 dccp_pr_debug("xfrm4_policy_check failed\n");
1158 goto discard_and_relse;
1161 if (sk_filter(sk, skb, 0)) {
1162 dccp_pr_debug("sk_filter failed\n");
1163 goto discard_and_relse;
1170 if (!sock_owned_by_user(sk))
1171 rc = dccp_v4_do_rcv(sk, skb);
1173 sk_add_backlog(sk, skb);
1180 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1184 * Generate Reset(No Connection) unless P.type == Reset
1185 * Drop packet and return
1187 if (dh->dccph_type != DCCP_PKT_RESET) {
1188 DCCP_SKB_CB(skb)->dccpd_reset_code =
1189 DCCP_RESET_CODE_NO_CONNECTION;
1190 dccp_v4_ctl_send_reset(skb);
1194 /* Discard frame. */
1203 inet_twsk_put((struct inet_timewait_sock *)sk);
1204 goto no_dccp_socket;
1207 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
1208 .queue_xmit = ip_queue_xmit,
1209 .send_check = dccp_v4_send_check,
1210 .rebuild_header = inet_sk_rebuild_header,
1211 .conn_request = dccp_v4_conn_request,
1212 .syn_recv_sock = dccp_v4_request_recv_sock,
1213 .net_header_len = sizeof(struct iphdr),
1214 .setsockopt = ip_setsockopt,
1215 .getsockopt = ip_getsockopt,
1216 .addr2sockaddr = inet_csk_addr2sockaddr,
1217 .sockaddr_len = sizeof(struct sockaddr_in),
1220 static int dccp_v4_init_sock(struct sock *sk)
1222 struct dccp_sock *dp = dccp_sk(sk);
1223 static int dccp_ctl_socket_init = 1;
1225 dccp_options_init(&dp->dccps_options);
1226 do_gettimeofday(&dp->dccps_epoch);
1228 if (dp->dccps_options.dccpo_send_ack_vector) {
1229 dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(DCCP_MAX_ACKVEC_LEN,
1231 if (dp->dccps_hc_rx_ackvec == NULL)
1236 * FIXME: We're hardcoding the CCID, and doing this at this point makes
1237 * the listening (master) sock get CCID control blocks, which is not
1238 * necessary, but for now, to not mess with the test userspace apps,
1239 * lets leave it here, later the real solution is to do this in a
1240 * setsockopt(CCIDs-I-want/accept). -acme
1242 if (likely(!dccp_ctl_socket_init)) {
1243 dp->dccps_hc_rx_ccid = ccid_init(dp->dccps_options.dccpo_rx_ccid,
1245 dp->dccps_hc_tx_ccid = ccid_init(dp->dccps_options.dccpo_tx_ccid,
1247 if (dp->dccps_hc_rx_ccid == NULL ||
1248 dp->dccps_hc_tx_ccid == NULL) {
1249 ccid_exit(dp->dccps_hc_rx_ccid, sk);
1250 ccid_exit(dp->dccps_hc_tx_ccid, sk);
1251 if (dp->dccps_options.dccpo_send_ack_vector) {
1252 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1253 dp->dccps_hc_rx_ackvec = NULL;
1255 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1259 dccp_ctl_socket_init = 0;
1261 dccp_init_xmit_timers(sk);
1262 inet_csk(sk)->icsk_rto = DCCP_TIMEOUT_INIT;
1263 sk->sk_state = DCCP_CLOSED;
1264 sk->sk_write_space = dccp_write_space;
1265 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1266 dp->dccps_mss_cache = 536;
1267 dp->dccps_role = DCCP_ROLE_UNDEFINED;
1268 dp->dccps_service = DCCP_SERVICE_INVALID_VALUE;
1273 static int dccp_v4_destroy_sock(struct sock *sk)
1275 struct dccp_sock *dp = dccp_sk(sk);
1278 * DCCP doesn't use sk_write_queue, just sk_send_head
1279 * for retransmissions
1281 if (sk->sk_send_head != NULL) {
1282 kfree_skb(sk->sk_send_head);
1283 sk->sk_send_head = NULL;
1286 /* Clean up a referenced DCCP bind bucket. */
1287 if (inet_csk(sk)->icsk_bind_hash != NULL)
1288 inet_put_port(&dccp_hashinfo, sk);
1290 kfree(dp->dccps_service_list);
1291 dp->dccps_service_list = NULL;
1293 ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
1294 ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
1295 if (dp->dccps_options.dccpo_send_ack_vector) {
1296 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1297 dp->dccps_hc_rx_ackvec = NULL;
1299 ccid_exit(dp->dccps_hc_rx_ccid, sk);
1300 ccid_exit(dp->dccps_hc_tx_ccid, sk);
1301 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1306 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1308 kfree(inet_rsk(req)->opt);
1311 static struct request_sock_ops dccp_request_sock_ops = {
1313 .obj_size = sizeof(struct dccp_request_sock),
1314 .rtx_syn_ack = dccp_v4_send_response,
1315 .send_ack = dccp_v4_reqsk_send_ack,
1316 .destructor = dccp_v4_reqsk_destructor,
1317 .send_reset = dccp_v4_ctl_send_reset,
1320 struct proto dccp_prot = {
1322 .owner = THIS_MODULE,
1323 .close = dccp_close,
1324 .connect = dccp_v4_connect,
1325 .disconnect = dccp_disconnect,
1326 .ioctl = dccp_ioctl,
1327 .init = dccp_v4_init_sock,
1328 .setsockopt = dccp_setsockopt,
1329 .getsockopt = dccp_getsockopt,
1330 .sendmsg = dccp_sendmsg,
1331 .recvmsg = dccp_recvmsg,
1332 .backlog_rcv = dccp_v4_do_rcv,
1333 .hash = dccp_v4_hash,
1334 .unhash = dccp_v4_unhash,
1335 .accept = inet_csk_accept,
1336 .get_port = dccp_v4_get_port,
1337 .shutdown = dccp_shutdown,
1338 .destroy = dccp_v4_destroy_sock,
1339 .orphan_count = &dccp_orphan_count,
1340 .max_header = MAX_DCCP_HEADER,
1341 .obj_size = sizeof(struct dccp_sock),
1342 .rsk_prot = &dccp_request_sock_ops,
1343 .twsk_obj_size = sizeof(struct inet_timewait_sock),