inotify: fix race
[linux-2.6.22.y-op.git] / net / dccp / ipv6.c
blob31737cdf156a5e657760278c557cbb861d2961f7
1 /*
2 * DCCP over IPv6
3 * Linux INET6 implementation
5 * Based on net/dccp6/ipv6.c
7 * Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License
11 * as published by the Free Software Foundation; either version
12 * 2 of the License, or (at your option) any later version.
15 #include <linux/module.h>
16 #include <linux/random.h>
17 #include <linux/xfrm.h>
19 #include <net/addrconf.h>
20 #include <net/inet_common.h>
21 #include <net/inet_hashtables.h>
22 #include <net/inet_sock.h>
23 #include <net/inet6_connection_sock.h>
24 #include <net/inet6_hashtables.h>
25 #include <net/ip6_route.h>
26 #include <net/ipv6.h>
27 #include <net/protocol.h>
28 #include <net/transp_v6.h>
29 #include <net/ip6_checksum.h>
30 #include <net/xfrm.h>
32 #include "dccp.h"
33 #include "ipv6.h"
34 #include "feat.h"
36 /* Socket used for sending RSTs and ACKs */
37 static struct socket *dccp_v6_ctl_socket;
39 static struct inet_connection_sock_af_ops dccp_ipv6_mapped;
40 static struct inet_connection_sock_af_ops dccp_ipv6_af_ops;
42 static int dccp_v6_get_port(struct sock *sk, unsigned short snum)
44 return inet_csk_get_port(&dccp_hashinfo, sk, snum,
45 inet6_csk_bind_conflict);
48 static void dccp_v6_hash(struct sock *sk)
50 if (sk->sk_state != DCCP_CLOSED) {
51 if (inet_csk(sk)->icsk_af_ops == &dccp_ipv6_mapped) {
52 dccp_hash(sk);
53 return;
55 local_bh_disable();
56 __inet6_hash(&dccp_hashinfo, sk);
57 local_bh_enable();
61 /* add pseudo-header to DCCP checksum stored in skb->csum */
62 static inline __sum16 dccp_v6_csum_finish(struct sk_buff *skb,
63 struct in6_addr *saddr,
64 struct in6_addr *daddr)
66 return csum_ipv6_magic(saddr, daddr, skb->len, IPPROTO_DCCP, skb->csum);
69 static inline void dccp_v6_send_check(struct sock *sk, int unused_value,
70 struct sk_buff *skb)
72 struct ipv6_pinfo *np = inet6_sk(sk);
73 struct dccp_hdr *dh = dccp_hdr(skb);
75 dccp_csum_outgoing(skb);
76 dh->dccph_checksum = dccp_v6_csum_finish(skb, &np->saddr, &np->daddr);
79 static inline __u32 secure_dccpv6_sequence_number(__be32 *saddr, __be32 *daddr,
80 __be16 sport, __be16 dport )
82 return secure_tcpv6_sequence_number(saddr, daddr, sport, dport);
85 static inline __u32 dccp_v6_init_sequence(struct sk_buff *skb)
87 return secure_dccpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
88 ipv6_hdr(skb)->saddr.s6_addr32,
89 dccp_hdr(skb)->dccph_dport,
90 dccp_hdr(skb)->dccph_sport );
94 static void dccp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
95 int type, int code, int offset, __be32 info)
97 struct ipv6hdr *hdr = (struct ipv6hdr *)skb->data;
98 const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
99 struct ipv6_pinfo *np;
100 struct sock *sk;
101 int err;
102 __u64 seq;
104 sk = inet6_lookup(&dccp_hashinfo, &hdr->daddr, dh->dccph_dport,
105 &hdr->saddr, dh->dccph_sport, inet6_iif(skb));
107 if (sk == NULL) {
108 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
109 return;
112 if (sk->sk_state == DCCP_TIME_WAIT) {
113 inet_twsk_put(inet_twsk(sk));
114 return;
117 bh_lock_sock(sk);
118 if (sock_owned_by_user(sk))
119 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
121 if (sk->sk_state == DCCP_CLOSED)
122 goto out;
124 np = inet6_sk(sk);
126 if (type == ICMPV6_PKT_TOOBIG) {
127 struct dst_entry *dst = NULL;
129 if (sock_owned_by_user(sk))
130 goto out;
131 if ((1 << sk->sk_state) & (DCCPF_LISTEN | DCCPF_CLOSED))
132 goto out;
134 /* icmp should have updated the destination cache entry */
135 dst = __sk_dst_check(sk, np->dst_cookie);
136 if (dst == NULL) {
137 struct inet_sock *inet = inet_sk(sk);
138 struct flowi fl;
140 /* BUGGG_FUTURE: Again, it is not clear how
141 to handle rthdr case. Ignore this complexity
142 for now.
144 memset(&fl, 0, sizeof(fl));
145 fl.proto = IPPROTO_DCCP;
146 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
147 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
148 fl.oif = sk->sk_bound_dev_if;
149 fl.fl_ip_dport = inet->dport;
150 fl.fl_ip_sport = inet->sport;
151 security_sk_classify_flow(sk, &fl);
153 err = ip6_dst_lookup(sk, &dst, &fl);
154 if (err) {
155 sk->sk_err_soft = -err;
156 goto out;
159 err = xfrm_lookup(&dst, &fl, sk, 0);
160 if (err < 0) {
161 sk->sk_err_soft = -err;
162 goto out;
164 } else
165 dst_hold(dst);
167 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
168 dccp_sync_mss(sk, dst_mtu(dst));
169 } /* else let the usual retransmit timer handle it */
170 dst_release(dst);
171 goto out;
174 icmpv6_err_convert(type, code, &err);
176 seq = DCCP_SKB_CB(skb)->dccpd_seq;
177 /* Might be for an request_sock */
178 switch (sk->sk_state) {
179 struct request_sock *req, **prev;
180 case DCCP_LISTEN:
181 if (sock_owned_by_user(sk))
182 goto out;
184 req = inet6_csk_search_req(sk, &prev, dh->dccph_dport,
185 &hdr->daddr, &hdr->saddr,
186 inet6_iif(skb));
187 if (req == NULL)
188 goto out;
191 * ICMPs are not backlogged, hence we cannot get an established
192 * socket here.
194 BUG_TRAP(req->sk == NULL);
196 if (seq != dccp_rsk(req)->dreq_iss) {
197 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
198 goto out;
201 inet_csk_reqsk_queue_drop(sk, req, prev);
202 goto out;
204 case DCCP_REQUESTING:
205 case DCCP_RESPOND: /* Cannot happen.
206 It can, it SYNs are crossed. --ANK */
207 if (!sock_owned_by_user(sk)) {
208 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
209 sk->sk_err = err;
211 * Wake people up to see the error
212 * (see connect in sock.c)
214 sk->sk_error_report(sk);
215 dccp_done(sk);
216 } else
217 sk->sk_err_soft = err;
218 goto out;
221 if (!sock_owned_by_user(sk) && np->recverr) {
222 sk->sk_err = err;
223 sk->sk_error_report(sk);
224 } else
225 sk->sk_err_soft = err;
227 out:
228 bh_unlock_sock(sk);
229 sock_put(sk);
233 static int dccp_v6_send_response(struct sock *sk, struct request_sock *req,
234 struct dst_entry *dst)
236 struct inet6_request_sock *ireq6 = inet6_rsk(req);
237 struct ipv6_pinfo *np = inet6_sk(sk);
238 struct sk_buff *skb;
239 struct ipv6_txoptions *opt = NULL;
240 struct in6_addr *final_p = NULL, final;
241 struct flowi fl;
242 int err = -1;
244 memset(&fl, 0, sizeof(fl));
245 fl.proto = IPPROTO_DCCP;
246 ipv6_addr_copy(&fl.fl6_dst, &ireq6->rmt_addr);
247 ipv6_addr_copy(&fl.fl6_src, &ireq6->loc_addr);
248 fl.fl6_flowlabel = 0;
249 fl.oif = ireq6->iif;
250 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
251 fl.fl_ip_sport = inet_sk(sk)->sport;
252 security_req_classify_flow(req, &fl);
254 if (dst == NULL) {
255 opt = np->opt;
256 if (opt == NULL &&
257 np->rxopt.bits.osrcrt == 2 &&
258 ireq6->pktopts) {
259 struct sk_buff *pktopts = ireq6->pktopts;
260 struct inet6_skb_parm *rxopt = IP6CB(pktopts);
262 if (rxopt->srcrt)
263 opt = ipv6_invert_rthdr(sk,
264 (struct ipv6_rt_hdr *)(skb_network_header(pktopts) +
265 rxopt->srcrt));
268 if (opt != NULL && opt->srcrt != NULL) {
269 const struct rt0_hdr *rt0 = (struct rt0_hdr *)opt->srcrt;
271 ipv6_addr_copy(&final, &fl.fl6_dst);
272 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
273 final_p = &final;
276 err = ip6_dst_lookup(sk, &dst, &fl);
277 if (err)
278 goto done;
280 if (final_p)
281 ipv6_addr_copy(&fl.fl6_dst, final_p);
283 err = xfrm_lookup(&dst, &fl, sk, 0);
284 if (err < 0)
285 goto done;
288 skb = dccp_make_response(sk, dst, req);
289 if (skb != NULL) {
290 struct dccp_hdr *dh = dccp_hdr(skb);
292 dh->dccph_checksum = dccp_v6_csum_finish(skb,
293 &ireq6->loc_addr,
294 &ireq6->rmt_addr);
295 ipv6_addr_copy(&fl.fl6_dst, &ireq6->rmt_addr);
296 err = ip6_xmit(sk, skb, &fl, opt, 0);
297 err = net_xmit_eval(err);
300 done:
301 if (opt != NULL && opt != np->opt)
302 sock_kfree_s(sk, opt, opt->tot_len);
303 dst_release(dst);
304 return err;
307 static void dccp_v6_reqsk_destructor(struct request_sock *req)
309 if (inet6_rsk(req)->pktopts != NULL)
310 kfree_skb(inet6_rsk(req)->pktopts);
313 static void dccp_v6_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
315 struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
316 struct ipv6hdr *rxip6h;
317 const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
318 sizeof(struct dccp_hdr_ext) +
319 sizeof(struct dccp_hdr_reset);
320 struct sk_buff *skb;
321 struct flowi fl;
322 u64 seqno = 0;
324 if (rxdh->dccph_type == DCCP_PKT_RESET)
325 return;
327 if (!ipv6_unicast_destination(rxskb))
328 return;
330 skb = alloc_skb(dccp_v6_ctl_socket->sk->sk_prot->max_header,
331 GFP_ATOMIC);
332 if (skb == NULL)
333 return;
335 skb_reserve(skb, dccp_v6_ctl_socket->sk->sk_prot->max_header);
337 dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
339 /* Swap the send and the receive. */
340 dh->dccph_type = DCCP_PKT_RESET;
341 dh->dccph_sport = rxdh->dccph_dport;
342 dh->dccph_dport = rxdh->dccph_sport;
343 dh->dccph_doff = dccp_hdr_reset_len / 4;
344 dh->dccph_x = 1;
345 dccp_hdr_reset(skb)->dccph_reset_code =
346 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
348 /* See "8.3.1. Abnormal Termination" in RFC 4340 */
349 if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
350 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
352 dccp_hdr_set_seq(dh, seqno);
353 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), DCCP_SKB_CB(rxskb)->dccpd_seq);
355 dccp_csum_outgoing(skb);
356 rxip6h = ipv6_hdr(rxskb);
357 dh->dccph_checksum = dccp_v6_csum_finish(skb, &rxip6h->saddr,
358 &rxip6h->daddr);
360 memset(&fl, 0, sizeof(fl));
361 ipv6_addr_copy(&fl.fl6_dst, &rxip6h->saddr);
362 ipv6_addr_copy(&fl.fl6_src, &rxip6h->daddr);
364 fl.proto = IPPROTO_DCCP;
365 fl.oif = inet6_iif(rxskb);
366 fl.fl_ip_dport = dh->dccph_dport;
367 fl.fl_ip_sport = dh->dccph_sport;
368 security_skb_classify_flow(rxskb, &fl);
370 /* sk = NULL, but it is safe for now. RST socket required. */
371 if (!ip6_dst_lookup(NULL, &skb->dst, &fl)) {
372 if (xfrm_lookup(&skb->dst, &fl, NULL, 0) >= 0) {
373 ip6_xmit(dccp_v6_ctl_socket->sk, skb, &fl, NULL, 0);
374 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
375 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
376 return;
380 kfree_skb(skb);
383 static struct request_sock_ops dccp6_request_sock_ops = {
384 .family = AF_INET6,
385 .obj_size = sizeof(struct dccp6_request_sock),
386 .rtx_syn_ack = dccp_v6_send_response,
387 .send_ack = dccp_reqsk_send_ack,
388 .destructor = dccp_v6_reqsk_destructor,
389 .send_reset = dccp_v6_ctl_send_reset,
392 static struct sock *dccp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
394 const struct dccp_hdr *dh = dccp_hdr(skb);
395 const struct ipv6hdr *iph = ipv6_hdr(skb);
396 struct sock *nsk;
397 struct request_sock **prev;
398 /* Find possible connection requests. */
399 struct request_sock *req = inet6_csk_search_req(sk, &prev,
400 dh->dccph_sport,
401 &iph->saddr,
402 &iph->daddr,
403 inet6_iif(skb));
404 if (req != NULL)
405 return dccp_check_req(sk, skb, req, prev);
407 nsk = __inet6_lookup_established(&dccp_hashinfo,
408 &iph->saddr, dh->dccph_sport,
409 &iph->daddr, ntohs(dh->dccph_dport),
410 inet6_iif(skb));
411 if (nsk != NULL) {
412 if (nsk->sk_state != DCCP_TIME_WAIT) {
413 bh_lock_sock(nsk);
414 return nsk;
416 inet_twsk_put(inet_twsk(nsk));
417 return NULL;
420 return sk;
423 static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
425 struct request_sock *req;
426 struct dccp_request_sock *dreq;
427 struct inet6_request_sock *ireq6;
428 struct ipv6_pinfo *np = inet6_sk(sk);
429 const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
430 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
431 __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
433 if (skb->protocol == htons(ETH_P_IP))
434 return dccp_v4_conn_request(sk, skb);
436 if (!ipv6_unicast_destination(skb))
437 goto drop;
439 if (dccp_bad_service_code(sk, service)) {
440 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
441 goto drop;
444 * There are no SYN attacks on IPv6, yet...
446 if (inet_csk_reqsk_queue_is_full(sk))
447 goto drop;
449 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
450 goto drop;
452 req = inet6_reqsk_alloc(&dccp6_request_sock_ops);
453 if (req == NULL)
454 goto drop;
456 if (dccp_parse_options(sk, skb))
457 goto drop_and_free;
459 dccp_reqsk_init(req, skb);
461 if (security_inet_conn_request(sk, skb, req))
462 goto drop_and_free;
464 ireq6 = inet6_rsk(req);
465 ipv6_addr_copy(&ireq6->rmt_addr, &ipv6_hdr(skb)->saddr);
466 ipv6_addr_copy(&ireq6->loc_addr, &ipv6_hdr(skb)->daddr);
467 ireq6->pktopts = NULL;
469 if (ipv6_opt_accepted(sk, skb) ||
470 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
471 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
472 atomic_inc(&skb->users);
473 ireq6->pktopts = skb;
475 ireq6->iif = sk->sk_bound_dev_if;
477 /* So that link locals have meaning */
478 if (!sk->sk_bound_dev_if &&
479 ipv6_addr_type(&ireq6->rmt_addr) & IPV6_ADDR_LINKLOCAL)
480 ireq6->iif = inet6_iif(skb);
483 * Step 3: Process LISTEN state
485 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
487 * In fact we defer setting S.GSR, S.SWL, S.SWH to
488 * dccp_create_openreq_child.
490 dreq = dccp_rsk(req);
491 dreq->dreq_isr = dcb->dccpd_seq;
492 dreq->dreq_iss = dccp_v6_init_sequence(skb);
493 dreq->dreq_service = service;
495 if (dccp_v6_send_response(sk, req, NULL))
496 goto drop_and_free;
498 inet6_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
499 return 0;
501 drop_and_free:
502 reqsk_free(req);
503 drop:
504 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
505 dcb->dccpd_reset_code = reset_code;
506 return -1;
509 static struct sock *dccp_v6_request_recv_sock(struct sock *sk,
510 struct sk_buff *skb,
511 struct request_sock *req,
512 struct dst_entry *dst)
514 struct inet6_request_sock *ireq6 = inet6_rsk(req);
515 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
516 struct inet_sock *newinet;
517 struct dccp_sock *newdp;
518 struct dccp6_sock *newdp6;
519 struct sock *newsk;
520 struct ipv6_txoptions *opt;
522 if (skb->protocol == htons(ETH_P_IP)) {
524 * v6 mapped
526 newsk = dccp_v4_request_recv_sock(sk, skb, req, dst);
527 if (newsk == NULL)
528 return NULL;
530 newdp6 = (struct dccp6_sock *)newsk;
531 newdp = dccp_sk(newsk);
532 newinet = inet_sk(newsk);
533 newinet->pinet6 = &newdp6->inet6;
534 newnp = inet6_sk(newsk);
536 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
538 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
539 newinet->daddr);
541 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
542 newinet->saddr);
544 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
546 inet_csk(newsk)->icsk_af_ops = &dccp_ipv6_mapped;
547 newsk->sk_backlog_rcv = dccp_v4_do_rcv;
548 newnp->pktoptions = NULL;
549 newnp->opt = NULL;
550 newnp->mcast_oif = inet6_iif(skb);
551 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
554 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
555 * here, dccp_create_openreq_child now does this for us, see the comment in
556 * that function for the gory details. -acme
559 /* It is tricky place. Until this moment IPv4 tcp
560 worked with IPv6 icsk.icsk_af_ops.
561 Sync it now.
563 dccp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
565 return newsk;
568 opt = np->opt;
570 if (sk_acceptq_is_full(sk))
571 goto out_overflow;
573 if (np->rxopt.bits.osrcrt == 2 && opt == NULL && ireq6->pktopts) {
574 const struct inet6_skb_parm *rxopt = IP6CB(ireq6->pktopts);
576 if (rxopt->srcrt)
577 opt = ipv6_invert_rthdr(sk,
578 (struct ipv6_rt_hdr *)(skb_network_header(ireq6->pktopts) +
579 rxopt->srcrt));
582 if (dst == NULL) {
583 struct in6_addr *final_p = NULL, final;
584 struct flowi fl;
586 memset(&fl, 0, sizeof(fl));
587 fl.proto = IPPROTO_DCCP;
588 ipv6_addr_copy(&fl.fl6_dst, &ireq6->rmt_addr);
589 if (opt != NULL && opt->srcrt != NULL) {
590 const struct rt0_hdr *rt0 = (struct rt0_hdr *)opt->srcrt;
592 ipv6_addr_copy(&final, &fl.fl6_dst);
593 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
594 final_p = &final;
596 ipv6_addr_copy(&fl.fl6_src, &ireq6->loc_addr);
597 fl.oif = sk->sk_bound_dev_if;
598 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
599 fl.fl_ip_sport = inet_sk(sk)->sport;
600 security_sk_classify_flow(sk, &fl);
602 if (ip6_dst_lookup(sk, &dst, &fl))
603 goto out;
605 if (final_p)
606 ipv6_addr_copy(&fl.fl6_dst, final_p);
608 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
609 goto out;
612 newsk = dccp_create_openreq_child(sk, req, skb);
613 if (newsk == NULL)
614 goto out;
617 * No need to charge this sock to the relevant IPv6 refcnt debug socks
618 * count here, dccp_create_openreq_child now does this for us, see the
619 * comment in that function for the gory details. -acme
622 __ip6_dst_store(newsk, dst, NULL, NULL);
623 newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM |
624 NETIF_F_TSO);
625 newdp6 = (struct dccp6_sock *)newsk;
626 newinet = inet_sk(newsk);
627 newinet->pinet6 = &newdp6->inet6;
628 newdp = dccp_sk(newsk);
629 newnp = inet6_sk(newsk);
631 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
633 ipv6_addr_copy(&newnp->daddr, &ireq6->rmt_addr);
634 ipv6_addr_copy(&newnp->saddr, &ireq6->loc_addr);
635 ipv6_addr_copy(&newnp->rcv_saddr, &ireq6->loc_addr);
636 newsk->sk_bound_dev_if = ireq6->iif;
638 /* Now IPv6 options...
640 First: no IPv4 options.
642 newinet->opt = NULL;
644 /* Clone RX bits */
645 newnp->rxopt.all = np->rxopt.all;
647 /* Clone pktoptions received with SYN */
648 newnp->pktoptions = NULL;
649 if (ireq6->pktopts != NULL) {
650 newnp->pktoptions = skb_clone(ireq6->pktopts, GFP_ATOMIC);
651 kfree_skb(ireq6->pktopts);
652 ireq6->pktopts = NULL;
653 if (newnp->pktoptions)
654 skb_set_owner_r(newnp->pktoptions, newsk);
656 newnp->opt = NULL;
657 newnp->mcast_oif = inet6_iif(skb);
658 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
661 * Clone native IPv6 options from listening socket (if any)
663 * Yes, keeping reference count would be much more clever, but we make
664 * one more one thing there: reattach optmem to newsk.
666 if (opt != NULL) {
667 newnp->opt = ipv6_dup_options(newsk, opt);
668 if (opt != np->opt)
669 sock_kfree_s(sk, opt, opt->tot_len);
672 inet_csk(newsk)->icsk_ext_hdr_len = 0;
673 if (newnp->opt != NULL)
674 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
675 newnp->opt->opt_flen);
677 dccp_sync_mss(newsk, dst_mtu(dst));
679 newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
681 __inet6_hash(&dccp_hashinfo, newsk);
682 inet_inherit_port(&dccp_hashinfo, sk, newsk);
684 return newsk;
686 out_overflow:
687 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
688 out:
689 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
690 if (opt != NULL && opt != np->opt)
691 sock_kfree_s(sk, opt, opt->tot_len);
692 dst_release(dst);
693 return NULL;
696 /* The socket must have it's spinlock held when we get
697 * here.
699 * We have a potential double-lock case here, so even when
700 * doing backlog processing we use the BH locking scheme.
701 * This is because we cannot sleep with the original spinlock
702 * held.
704 static int dccp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
706 struct ipv6_pinfo *np = inet6_sk(sk);
707 struct sk_buff *opt_skb = NULL;
709 /* Imagine: socket is IPv6. IPv4 packet arrives,
710 goes to IPv4 receive handler and backlogged.
711 From backlog it always goes here. Kerboom...
712 Fortunately, dccp_rcv_established and rcv_established
713 handle them correctly, but it is not case with
714 dccp_v6_hnd_req and dccp_v6_ctl_send_reset(). --ANK
717 if (skb->protocol == htons(ETH_P_IP))
718 return dccp_v4_do_rcv(sk, skb);
720 if (sk_filter(sk, skb))
721 goto discard;
724 * socket locking is here for SMP purposes as backlog rcv is currently
725 * called with bh processing disabled.
728 /* Do Stevens' IPV6_PKTOPTIONS.
730 Yes, guys, it is the only place in our code, where we
731 may make it not affecting IPv4.
732 The rest of code is protocol independent,
733 and I do not like idea to uglify IPv4.
735 Actually, all the idea behind IPV6_PKTOPTIONS
736 looks not very well thought. For now we latch
737 options, received in the last packet, enqueued
738 by tcp. Feel free to propose better solution.
739 --ANK (980728)
741 if (np->rxopt.all)
743 * FIXME: Add handling of IPV6_PKTOPTIONS skb. See the comments below
744 * (wrt ipv6_pktopions) and net/ipv6/tcp_ipv6.c for an example.
746 opt_skb = skb_clone(skb, GFP_ATOMIC);
748 if (sk->sk_state == DCCP_OPEN) { /* Fast path */
749 if (dccp_rcv_established(sk, skb, dccp_hdr(skb), skb->len))
750 goto reset;
751 if (opt_skb) {
752 /* XXX This is where we would goto ipv6_pktoptions. */
753 __kfree_skb(opt_skb);
755 return 0;
759 * Step 3: Process LISTEN state
760 * If S.state == LISTEN,
761 * If P.type == Request or P contains a valid Init Cookie option,
762 * (* Must scan the packet's options to check for Init
763 * Cookies. Only Init Cookies are processed here,
764 * however; other options are processed in Step 8. This
765 * scan need only be performed if the endpoint uses Init
766 * Cookies *)
767 * (* Generate a new socket and switch to that socket *)
768 * Set S := new socket for this port pair
769 * S.state = RESPOND
770 * Choose S.ISS (initial seqno) or set from Init Cookies
771 * Initialize S.GAR := S.ISS
772 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
773 * Continue with S.state == RESPOND
774 * (* A Response packet will be generated in Step 11 *)
775 * Otherwise,
776 * Generate Reset(No Connection) unless P.type == Reset
777 * Drop packet and return
779 * NOTE: the check for the packet types is done in
780 * dccp_rcv_state_process
782 if (sk->sk_state == DCCP_LISTEN) {
783 struct sock *nsk = dccp_v6_hnd_req(sk, skb);
785 if (nsk == NULL)
786 goto discard;
788 * Queue it on the new socket if the new socket is active,
789 * otherwise we just shortcircuit this and continue with
790 * the new socket..
792 if (nsk != sk) {
793 if (dccp_child_process(sk, nsk, skb))
794 goto reset;
795 if (opt_skb != NULL)
796 __kfree_skb(opt_skb);
797 return 0;
801 if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len))
802 goto reset;
803 if (opt_skb) {
804 /* XXX This is where we would goto ipv6_pktoptions. */
805 __kfree_skb(opt_skb);
807 return 0;
809 reset:
810 dccp_v6_ctl_send_reset(sk, skb);
811 discard:
812 if (opt_skb != NULL)
813 __kfree_skb(opt_skb);
814 kfree_skb(skb);
815 return 0;
818 static int dccp_v6_rcv(struct sk_buff **pskb)
820 const struct dccp_hdr *dh;
821 struct sk_buff *skb = *pskb;
822 struct sock *sk;
823 int min_cov;
825 /* Step 1: Check header basics */
827 if (dccp_invalid_packet(skb))
828 goto discard_it;
830 /* Step 1: If header checksum is incorrect, drop packet and return. */
831 if (dccp_v6_csum_finish(skb, &ipv6_hdr(skb)->saddr,
832 &ipv6_hdr(skb)->daddr)) {
833 DCCP_WARN("dropped packet with invalid checksum\n");
834 goto discard_it;
837 dh = dccp_hdr(skb);
839 DCCP_SKB_CB(skb)->dccpd_seq = dccp_hdr_seq(skb);
840 DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
842 if (dccp_packet_without_ack(skb))
843 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
844 else
845 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
847 /* Step 2:
848 * Look up flow ID in table and get corresponding socket */
849 sk = __inet6_lookup(&dccp_hashinfo, &ipv6_hdr(skb)->saddr,
850 dh->dccph_sport,
851 &ipv6_hdr(skb)->daddr, ntohs(dh->dccph_dport),
852 inet6_iif(skb));
854 * Step 2:
855 * If no socket ...
857 if (sk == NULL) {
858 dccp_pr_debug("failed to look up flow ID in table and "
859 "get corresponding socket\n");
860 goto no_dccp_socket;
864 * Step 2:
865 * ... or S.state == TIMEWAIT,
866 * Generate Reset(No Connection) unless P.type == Reset
867 * Drop packet and return
869 if (sk->sk_state == DCCP_TIME_WAIT) {
870 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
871 inet_twsk_put(inet_twsk(sk));
872 goto no_dccp_socket;
876 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
877 * o if MinCsCov = 0, only packets with CsCov = 0 are accepted
878 * o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
880 min_cov = dccp_sk(sk)->dccps_pcrlen;
881 if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov)) {
882 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
883 dh->dccph_cscov, min_cov);
884 /* FIXME: send Data Dropped option (see also dccp_v4_rcv) */
885 goto discard_and_relse;
888 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
889 goto discard_and_relse;
891 return sk_receive_skb(sk, skb, 1) ? -1 : 0;
893 no_dccp_socket:
894 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
895 goto discard_it;
897 * Step 2:
898 * If no socket ...
899 * Generate Reset(No Connection) unless P.type == Reset
900 * Drop packet and return
902 if (dh->dccph_type != DCCP_PKT_RESET) {
903 DCCP_SKB_CB(skb)->dccpd_reset_code =
904 DCCP_RESET_CODE_NO_CONNECTION;
905 dccp_v6_ctl_send_reset(sk, skb);
908 discard_it:
909 kfree_skb(skb);
910 return 0;
912 discard_and_relse:
913 sock_put(sk);
914 goto discard_it;
917 static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
918 int addr_len)
920 struct sockaddr_in6 *usin = (struct sockaddr_in6 *)uaddr;
921 struct inet_connection_sock *icsk = inet_csk(sk);
922 struct inet_sock *inet = inet_sk(sk);
923 struct ipv6_pinfo *np = inet6_sk(sk);
924 struct dccp_sock *dp = dccp_sk(sk);
925 struct in6_addr *saddr = NULL, *final_p = NULL, final;
926 struct flowi fl;
927 struct dst_entry *dst;
928 int addr_type;
929 int err;
931 dp->dccps_role = DCCP_ROLE_CLIENT;
933 if (addr_len < SIN6_LEN_RFC2133)
934 return -EINVAL;
936 if (usin->sin6_family != AF_INET6)
937 return -EAFNOSUPPORT;
939 memset(&fl, 0, sizeof(fl));
941 if (np->sndflow) {
942 fl.fl6_flowlabel = usin->sin6_flowinfo & IPV6_FLOWINFO_MASK;
943 IP6_ECN_flow_init(fl.fl6_flowlabel);
944 if (fl.fl6_flowlabel & IPV6_FLOWLABEL_MASK) {
945 struct ip6_flowlabel *flowlabel;
946 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
947 if (flowlabel == NULL)
948 return -EINVAL;
949 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
950 fl6_sock_release(flowlabel);
954 * connect() to INADDR_ANY means loopback (BSD'ism).
956 if (ipv6_addr_any(&usin->sin6_addr))
957 usin->sin6_addr.s6_addr[15] = 1;
959 addr_type = ipv6_addr_type(&usin->sin6_addr);
961 if (addr_type & IPV6_ADDR_MULTICAST)
962 return -ENETUNREACH;
964 if (addr_type & IPV6_ADDR_LINKLOCAL) {
965 if (addr_len >= sizeof(struct sockaddr_in6) &&
966 usin->sin6_scope_id) {
967 /* If interface is set while binding, indices
968 * must coincide.
970 if (sk->sk_bound_dev_if &&
971 sk->sk_bound_dev_if != usin->sin6_scope_id)
972 return -EINVAL;
974 sk->sk_bound_dev_if = usin->sin6_scope_id;
977 /* Connect to link-local address requires an interface */
978 if (!sk->sk_bound_dev_if)
979 return -EINVAL;
982 ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
983 np->flow_label = fl.fl6_flowlabel;
986 * DCCP over IPv4
988 if (addr_type == IPV6_ADDR_MAPPED) {
989 u32 exthdrlen = icsk->icsk_ext_hdr_len;
990 struct sockaddr_in sin;
992 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
994 if (__ipv6_only_sock(sk))
995 return -ENETUNREACH;
997 sin.sin_family = AF_INET;
998 sin.sin_port = usin->sin6_port;
999 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
1001 icsk->icsk_af_ops = &dccp_ipv6_mapped;
1002 sk->sk_backlog_rcv = dccp_v4_do_rcv;
1004 err = dccp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
1005 if (err) {
1006 icsk->icsk_ext_hdr_len = exthdrlen;
1007 icsk->icsk_af_ops = &dccp_ipv6_af_ops;
1008 sk->sk_backlog_rcv = dccp_v6_do_rcv;
1009 goto failure;
1010 } else {
1011 ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
1012 inet->saddr);
1013 ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
1014 inet->rcv_saddr);
1017 return err;
1020 if (!ipv6_addr_any(&np->rcv_saddr))
1021 saddr = &np->rcv_saddr;
1023 fl.proto = IPPROTO_DCCP;
1024 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1025 ipv6_addr_copy(&fl.fl6_src, saddr ? saddr : &np->saddr);
1026 fl.oif = sk->sk_bound_dev_if;
1027 fl.fl_ip_dport = usin->sin6_port;
1028 fl.fl_ip_sport = inet->sport;
1029 security_sk_classify_flow(sk, &fl);
1031 if (np->opt != NULL && np->opt->srcrt != NULL) {
1032 const struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
1034 ipv6_addr_copy(&final, &fl.fl6_dst);
1035 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1036 final_p = &final;
1039 err = ip6_dst_lookup(sk, &dst, &fl);
1040 if (err)
1041 goto failure;
1043 if (final_p)
1044 ipv6_addr_copy(&fl.fl6_dst, final_p);
1046 err = __xfrm_lookup(&dst, &fl, sk, 1);
1047 if (err < 0) {
1048 if (err == -EREMOTE)
1049 err = ip6_dst_blackhole(sk, &dst, &fl);
1050 if (err < 0)
1051 goto failure;
1054 if (saddr == NULL) {
1055 saddr = &fl.fl6_src;
1056 ipv6_addr_copy(&np->rcv_saddr, saddr);
1059 /* set the source address */
1060 ipv6_addr_copy(&np->saddr, saddr);
1061 inet->rcv_saddr = LOOPBACK4_IPV6;
1063 __ip6_dst_store(sk, dst, NULL, NULL);
1065 icsk->icsk_ext_hdr_len = 0;
1066 if (np->opt != NULL)
1067 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
1068 np->opt->opt_nflen);
1070 inet->dport = usin->sin6_port;
1072 dccp_set_state(sk, DCCP_REQUESTING);
1073 err = inet6_hash_connect(&dccp_death_row, sk);
1074 if (err)
1075 goto late_failure;
1077 dp->dccps_iss = secure_dccpv6_sequence_number(np->saddr.s6_addr32,
1078 np->daddr.s6_addr32,
1079 inet->sport, inet->dport);
1080 err = dccp_connect(sk);
1081 if (err)
1082 goto late_failure;
1084 return 0;
1086 late_failure:
1087 dccp_set_state(sk, DCCP_CLOSED);
1088 __sk_dst_reset(sk);
1089 failure:
1090 inet->dport = 0;
1091 sk->sk_route_caps = 0;
1092 return err;
1095 static struct inet_connection_sock_af_ops dccp_ipv6_af_ops = {
1096 .queue_xmit = inet6_csk_xmit,
1097 .send_check = dccp_v6_send_check,
1098 .rebuild_header = inet6_sk_rebuild_header,
1099 .conn_request = dccp_v6_conn_request,
1100 .syn_recv_sock = dccp_v6_request_recv_sock,
1101 .net_header_len = sizeof(struct ipv6hdr),
1102 .setsockopt = ipv6_setsockopt,
1103 .getsockopt = ipv6_getsockopt,
1104 .addr2sockaddr = inet6_csk_addr2sockaddr,
1105 .sockaddr_len = sizeof(struct sockaddr_in6),
1106 #ifdef CONFIG_COMPAT
1107 .compat_setsockopt = compat_ipv6_setsockopt,
1108 .compat_getsockopt = compat_ipv6_getsockopt,
1109 #endif
1113 * DCCP over IPv4 via INET6 API
1115 static struct inet_connection_sock_af_ops dccp_ipv6_mapped = {
1116 .queue_xmit = ip_queue_xmit,
1117 .send_check = dccp_v4_send_check,
1118 .rebuild_header = inet_sk_rebuild_header,
1119 .conn_request = dccp_v6_conn_request,
1120 .syn_recv_sock = dccp_v6_request_recv_sock,
1121 .net_header_len = sizeof(struct iphdr),
1122 .setsockopt = ipv6_setsockopt,
1123 .getsockopt = ipv6_getsockopt,
1124 .addr2sockaddr = inet6_csk_addr2sockaddr,
1125 .sockaddr_len = sizeof(struct sockaddr_in6),
1126 #ifdef CONFIG_COMPAT
1127 .compat_setsockopt = compat_ipv6_setsockopt,
1128 .compat_getsockopt = compat_ipv6_getsockopt,
1129 #endif
1132 /* NOTE: A lot of things set to zero explicitly by call to
1133 * sk_alloc() so need not be done here.
1135 static int dccp_v6_init_sock(struct sock *sk)
1137 static __u8 dccp_v6_ctl_sock_initialized;
1138 int err = dccp_init_sock(sk, dccp_v6_ctl_sock_initialized);
1140 if (err == 0) {
1141 if (unlikely(!dccp_v6_ctl_sock_initialized))
1142 dccp_v6_ctl_sock_initialized = 1;
1143 inet_csk(sk)->icsk_af_ops = &dccp_ipv6_af_ops;
1146 return err;
1149 static int dccp_v6_destroy_sock(struct sock *sk)
1151 dccp_destroy_sock(sk);
1152 return inet6_destroy_sock(sk);
1155 static struct timewait_sock_ops dccp6_timewait_sock_ops = {
1156 .twsk_obj_size = sizeof(struct dccp6_timewait_sock),
1159 static struct proto dccp_v6_prot = {
1160 .name = "DCCPv6",
1161 .owner = THIS_MODULE,
1162 .close = dccp_close,
1163 .connect = dccp_v6_connect,
1164 .disconnect = dccp_disconnect,
1165 .ioctl = dccp_ioctl,
1166 .init = dccp_v6_init_sock,
1167 .setsockopt = dccp_setsockopt,
1168 .getsockopt = dccp_getsockopt,
1169 .sendmsg = dccp_sendmsg,
1170 .recvmsg = dccp_recvmsg,
1171 .backlog_rcv = dccp_v6_do_rcv,
1172 .hash = dccp_v6_hash,
1173 .unhash = dccp_unhash,
1174 .accept = inet_csk_accept,
1175 .get_port = dccp_v6_get_port,
1176 .shutdown = dccp_shutdown,
1177 .destroy = dccp_v6_destroy_sock,
1178 .orphan_count = &dccp_orphan_count,
1179 .max_header = MAX_DCCP_HEADER,
1180 .obj_size = sizeof(struct dccp6_sock),
1181 .rsk_prot = &dccp6_request_sock_ops,
1182 .twsk_prot = &dccp6_timewait_sock_ops,
1183 #ifdef CONFIG_COMPAT
1184 .compat_setsockopt = compat_dccp_setsockopt,
1185 .compat_getsockopt = compat_dccp_getsockopt,
1186 #endif
1189 static struct inet6_protocol dccp_v6_protocol = {
1190 .handler = dccp_v6_rcv,
1191 .err_handler = dccp_v6_err,
1192 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1195 static struct proto_ops inet6_dccp_ops = {
1196 .family = PF_INET6,
1197 .owner = THIS_MODULE,
1198 .release = inet6_release,
1199 .bind = inet6_bind,
1200 .connect = inet_stream_connect,
1201 .socketpair = sock_no_socketpair,
1202 .accept = inet_accept,
1203 .getname = inet6_getname,
1204 .poll = dccp_poll,
1205 .ioctl = inet6_ioctl,
1206 .listen = inet_dccp_listen,
1207 .shutdown = inet_shutdown,
1208 .setsockopt = sock_common_setsockopt,
1209 .getsockopt = sock_common_getsockopt,
1210 .sendmsg = inet_sendmsg,
1211 .recvmsg = sock_common_recvmsg,
1212 .mmap = sock_no_mmap,
1213 .sendpage = sock_no_sendpage,
1214 #ifdef CONFIG_COMPAT
1215 .compat_setsockopt = compat_sock_common_setsockopt,
1216 .compat_getsockopt = compat_sock_common_getsockopt,
1217 #endif
1220 static struct inet_protosw dccp_v6_protosw = {
1221 .type = SOCK_DCCP,
1222 .protocol = IPPROTO_DCCP,
1223 .prot = &dccp_v6_prot,
1224 .ops = &inet6_dccp_ops,
1225 .capability = -1,
1226 .flags = INET_PROTOSW_ICSK,
1229 static int __init dccp_v6_init(void)
1231 int err = proto_register(&dccp_v6_prot, 1);
1233 if (err != 0)
1234 goto out;
1236 err = inet6_add_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1237 if (err != 0)
1238 goto out_unregister_proto;
1240 inet6_register_protosw(&dccp_v6_protosw);
1242 err = inet_csk_ctl_sock_create(&dccp_v6_ctl_socket, PF_INET6,
1243 SOCK_DCCP, IPPROTO_DCCP);
1244 if (err != 0)
1245 goto out_unregister_protosw;
1246 out:
1247 return err;
1248 out_unregister_protosw:
1249 inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1250 inet6_unregister_protosw(&dccp_v6_protosw);
1251 out_unregister_proto:
1252 proto_unregister(&dccp_v6_prot);
1253 goto out;
1256 static void __exit dccp_v6_exit(void)
1258 inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1259 inet6_unregister_protosw(&dccp_v6_protosw);
1260 proto_unregister(&dccp_v6_prot);
1263 module_init(dccp_v6_init);
1264 module_exit(dccp_v6_exit);
1267 * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1268 * values directly, Also cover the case where the protocol is not specified,
1269 * i.e. net-pf-PF_INET6-proto-0-type-SOCK_DCCP
1271 MODULE_ALIAS("net-pf-" __stringify(PF_INET6) "-proto-33-type-6");
1272 MODULE_ALIAS("net-pf-" __stringify(PF_INET6) "-proto-0-type-6");
1273 MODULE_LICENSE("GPL");
1274 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1275 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol");