1 /* 2 * DCCP over IPv6 3 * Linux INET6 implementation 4 * 5 * Based on net/dccp6/ipv6.c 6 * 7 * Arnaldo Carvalho de Melo <acme@ghostprotocols.net> 8 * 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. 13 */ 14 15 #include <linux/module.h> 16 #include <linux/random.h> 17 #include <linux/slab.h> 18 #include <linux/xfrm.h> 19 20 #include <net/addrconf.h> 21 #include <net/inet_common.h> 22 #include <net/inet_hashtables.h> 23 #include <net/inet_sock.h> 24 #include <net/inet6_connection_sock.h> 25 #include <net/inet6_hashtables.h> 26 #include <net/ip6_route.h> 27 #include <net/ipv6.h> 28 #include <net/protocol.h> 29 #include <net/transp_v6.h> 30 #include <net/ip6_checksum.h> 31 #include <net/xfrm.h> 32 #include <net/secure_seq.h> 33 34 #include "dccp.h" 35 #include "ipv6.h" 36 #include "feat.h" 37 38 /* The per-net dccp.v6_ctl_sk is used for sending RSTs and ACKs */ 39 40 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped; 41 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops; 42 43 static void dccp_v6_hash(struct sock *sk) 44 { 45 if (sk->sk_state != DCCP_CLOSED) { 46 if (inet_csk(sk)->icsk_af_ops == &dccp_ipv6_mapped) { 47 inet_hash(sk); 48 return; 49 } 50 local_bh_disable(); 51 __inet6_hash(sk, NULL); 52 local_bh_enable(); 53 } 54 } 55 56 /* add pseudo-header to DCCP checksum stored in skb->csum */ 57 static inline __sum16 dccp_v6_csum_finish(struct sk_buff *skb, 58 const struct in6_addr *saddr, 59 const struct in6_addr *daddr) 60 { 61 return csum_ipv6_magic(saddr, daddr, skb->len, IPPROTO_DCCP, skb->csum); 62 } 63 64 static inline void dccp_v6_send_check(struct sock *sk, struct sk_buff *skb) 65 { 66 struct ipv6_pinfo *np = inet6_sk(sk); 67 struct dccp_hdr *dh = dccp_hdr(skb); 68 69 dccp_csum_outgoing(skb); 70 dh->dccph_checksum = dccp_v6_csum_finish(skb, &np->saddr, &np->daddr); 71 } 72 73 static inline __u64 dccp_v6_init_sequence(struct sk_buff *skb) 74 { 75 return secure_dccpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32, 76 ipv6_hdr(skb)->saddr.s6_addr32, 77 dccp_hdr(skb)->dccph_dport, 78 dccp_hdr(skb)->dccph_sport ); 79 80 } 81 82 static void dccp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 83 u8 type, u8 code, int offset, __be32 info) 84 { 85 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 86 const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset); 87 struct dccp_sock *dp; 88 struct ipv6_pinfo *np; 89 struct sock *sk; 90 int err; 91 __u64 seq; 92 struct net *net = dev_net(skb->dev); 93 94 if (skb->len < offset + sizeof(*dh) || 95 skb->len < offset + __dccp_basic_hdr_len(dh)) { 96 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), 97 ICMP6_MIB_INERRORS); 98 return; 99 } 100 101 sk = inet6_lookup(net, &dccp_hashinfo, 102 &hdr->daddr, dh->dccph_dport, 103 &hdr->saddr, dh->dccph_sport, inet6_iif(skb)); 104 105 if (sk == NULL) { 106 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), 107 ICMP6_MIB_INERRORS); 108 return; 109 } 110 111 if (sk->sk_state == DCCP_TIME_WAIT) { 112 inet_twsk_put(inet_twsk(sk)); 113 return; 114 } 115 116 bh_lock_sock(sk); 117 if (sock_owned_by_user(sk)) 118 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS); 119 120 if (sk->sk_state == DCCP_CLOSED) 121 goto out; 122 123 dp = dccp_sk(sk); 124 seq = dccp_hdr_seq(dh); 125 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) && 126 !between48(seq, dp->dccps_awl, dp->dccps_awh)) { 127 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); 128 goto out; 129 } 130 131 np = inet6_sk(sk); 132 133 if (type == ICMPV6_PKT_TOOBIG) { 134 struct dst_entry *dst = NULL; 135 136 if (sock_owned_by_user(sk)) 137 goto out; 138 if ((1 << sk->sk_state) & (DCCPF_LISTEN | DCCPF_CLOSED)) 139 goto out; 140 141 /* icmp should have updated the destination cache entry */ 142 dst = __sk_dst_check(sk, np->dst_cookie); 143 if (dst == NULL) { 144 struct inet_sock *inet = inet_sk(sk); 145 struct flowi6 fl6; 146 147 /* BUGGG_FUTURE: Again, it is not clear how 148 to handle rthdr case. Ignore this complexity 149 for now. 150 */ 151 memset(&fl6, 0, sizeof(fl6)); 152 fl6.flowi6_proto = IPPROTO_DCCP; 153 fl6.daddr = np->daddr; 154 fl6.saddr = np->saddr; 155 fl6.flowi6_oif = sk->sk_bound_dev_if; 156 fl6.fl6_dport = inet->inet_dport; 157 fl6.fl6_sport = inet->inet_sport; 158 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 159 160 dst = ip6_dst_lookup_flow(sk, &fl6, NULL, false); 161 if (IS_ERR(dst)) { 162 sk->sk_err_soft = -PTR_ERR(dst); 163 goto out; 164 } 165 } else 166 dst_hold(dst); 167 168 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) { 169 dccp_sync_mss(sk, dst_mtu(dst)); 170 } /* else let the usual retransmit timer handle it */ 171 dst_release(dst); 172 goto out; 173 } 174 175 icmpv6_err_convert(type, code, &err); 176 177 /* 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; 183 184 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; 189 190 /* 191 * ICMPs are not backlogged, hence we cannot get an established 192 * socket here. 193 */ 194 WARN_ON(req->sk != NULL); 195 196 if (seq != dccp_rsk(req)->dreq_iss) { 197 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); 198 goto out; 199 } 200 201 inet_csk_reqsk_queue_drop(sk, req, prev); 202 goto out; 203 204 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; 210 /* 211 * Wake people up to see the error 212 * (see connect in sock.c) 213 */ 214 sk->sk_error_report(sk); 215 dccp_done(sk); 216 } else 217 sk->sk_err_soft = err; 218 goto out; 219 } 220 221 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; 226 227 out: 228 bh_unlock_sock(sk); 229 sock_put(sk); 230 } 231 232 233 static int dccp_v6_send_response(struct sock *sk, struct request_sock *req, 234 struct request_values *rv_unused) 235 { 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, final; 241 struct flowi6 fl6; 242 int err = -1; 243 struct dst_entry *dst; 244 245 memset(&fl6, 0, sizeof(fl6)); 246 fl6.flowi6_proto = IPPROTO_DCCP; 247 fl6.daddr = ireq6->rmt_addr; 248 fl6.saddr = ireq6->loc_addr; 249 fl6.flowlabel = 0; 250 fl6.flowi6_oif = ireq6->iif; 251 fl6.fl6_dport = inet_rsk(req)->rmt_port; 252 fl6.fl6_sport = inet_rsk(req)->loc_port; 253 security_req_classify_flow(req, flowi6_to_flowi(&fl6)); 254 255 opt = np->opt; 256 257 final_p = fl6_update_dst(&fl6, opt, &final); 258 259 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false); 260 if (IS_ERR(dst)) { 261 err = PTR_ERR(dst); 262 dst = NULL; 263 goto done; 264 } 265 266 skb = dccp_make_response(sk, dst, req); 267 if (skb != NULL) { 268 struct dccp_hdr *dh = dccp_hdr(skb); 269 270 dh->dccph_checksum = dccp_v6_csum_finish(skb, 271 &ireq6->loc_addr, 272 &ireq6->rmt_addr); 273 fl6.daddr = ireq6->rmt_addr; 274 err = ip6_xmit(sk, skb, &fl6, opt, np->tclass); 275 err = net_xmit_eval(err); 276 } 277 278 done: 279 if (opt != NULL && opt != np->opt) 280 sock_kfree_s(sk, opt, opt->tot_len); 281 dst_release(dst); 282 return err; 283 } 284 285 static void dccp_v6_reqsk_destructor(struct request_sock *req) 286 { 287 dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg); 288 if (inet6_rsk(req)->pktopts != NULL) 289 kfree_skb(inet6_rsk(req)->pktopts); 290 } 291 292 static void dccp_v6_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb) 293 { 294 const struct ipv6hdr *rxip6h; 295 struct sk_buff *skb; 296 struct flowi6 fl6; 297 struct net *net = dev_net(skb_dst(rxskb)->dev); 298 struct sock *ctl_sk = net->dccp.v6_ctl_sk; 299 struct dst_entry *dst; 300 301 if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET) 302 return; 303 304 if (!ipv6_unicast_destination(rxskb)) 305 return; 306 307 skb = dccp_ctl_make_reset(ctl_sk, rxskb); 308 if (skb == NULL) 309 return; 310 311 rxip6h = ipv6_hdr(rxskb); 312 dccp_hdr(skb)->dccph_checksum = dccp_v6_csum_finish(skb, &rxip6h->saddr, 313 &rxip6h->daddr); 314 315 memset(&fl6, 0, sizeof(fl6)); 316 fl6.daddr = rxip6h->saddr; 317 fl6.saddr = rxip6h->daddr; 318 319 fl6.flowi6_proto = IPPROTO_DCCP; 320 fl6.flowi6_oif = inet6_iif(rxskb); 321 fl6.fl6_dport = dccp_hdr(skb)->dccph_dport; 322 fl6.fl6_sport = dccp_hdr(skb)->dccph_sport; 323 security_skb_classify_flow(rxskb, flowi6_to_flowi(&fl6)); 324 325 /* sk = NULL, but it is safe for now. RST socket required. */ 326 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL, false); 327 if (!IS_ERR(dst)) { 328 skb_dst_set(skb, dst); 329 ip6_xmit(ctl_sk, skb, &fl6, NULL, 0); 330 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS); 331 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS); 332 return; 333 } 334 335 kfree_skb(skb); 336 } 337 338 static struct request_sock_ops dccp6_request_sock_ops = { 339 .family = AF_INET6, 340 .obj_size = sizeof(struct dccp6_request_sock), 341 .rtx_syn_ack = dccp_v6_send_response, 342 .send_ack = dccp_reqsk_send_ack, 343 .destructor = dccp_v6_reqsk_destructor, 344 .send_reset = dccp_v6_ctl_send_reset, 345 }; 346 347 static struct sock *dccp_v6_hnd_req(struct sock *sk,struct sk_buff *skb) 348 { 349 const struct dccp_hdr *dh = dccp_hdr(skb); 350 const struct ipv6hdr *iph = ipv6_hdr(skb); 351 struct sock *nsk; 352 struct request_sock **prev; 353 /* Find possible connection requests. */ 354 struct request_sock *req = inet6_csk_search_req(sk, &prev, 355 dh->dccph_sport, 356 &iph->saddr, 357 &iph->daddr, 358 inet6_iif(skb)); 359 if (req != NULL) 360 return dccp_check_req(sk, skb, req, prev); 361 362 nsk = __inet6_lookup_established(sock_net(sk), &dccp_hashinfo, 363 &iph->saddr, dh->dccph_sport, 364 &iph->daddr, ntohs(dh->dccph_dport), 365 inet6_iif(skb)); 366 if (nsk != NULL) { 367 if (nsk->sk_state != DCCP_TIME_WAIT) { 368 bh_lock_sock(nsk); 369 return nsk; 370 } 371 inet_twsk_put(inet_twsk(nsk)); 372 return NULL; 373 } 374 375 return sk; 376 } 377 378 static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb) 379 { 380 struct request_sock *req; 381 struct dccp_request_sock *dreq; 382 struct inet6_request_sock *ireq6; 383 struct ipv6_pinfo *np = inet6_sk(sk); 384 const __be32 service = dccp_hdr_request(skb)->dccph_req_service; 385 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb); 386 387 if (skb->protocol == htons(ETH_P_IP)) 388 return dccp_v4_conn_request(sk, skb); 389 390 if (!ipv6_unicast_destination(skb)) 391 return 0; /* discard, don't send a reset here */ 392 393 if (dccp_bad_service_code(sk, service)) { 394 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE; 395 goto drop; 396 } 397 /* 398 * There are no SYN attacks on IPv6, yet... 399 */ 400 dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY; 401 if (inet_csk_reqsk_queue_is_full(sk)) 402 goto drop; 403 404 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) 405 goto drop; 406 407 req = inet6_reqsk_alloc(&dccp6_request_sock_ops); 408 if (req == NULL) 409 goto drop; 410 411 if (dccp_reqsk_init(req, dccp_sk(sk), skb)) 412 goto drop_and_free; 413 414 dreq = dccp_rsk(req); 415 if (dccp_parse_options(sk, dreq, skb)) 416 goto drop_and_free; 417 418 if (security_inet_conn_request(sk, skb, req)) 419 goto drop_and_free; 420 421 ireq6 = inet6_rsk(req); 422 ireq6->rmt_addr = ipv6_hdr(skb)->saddr; 423 ireq6->loc_addr = ipv6_hdr(skb)->daddr; 424 425 if (ipv6_opt_accepted(sk, skb) || 426 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo || 427 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) { 428 atomic_inc(&skb->users); 429 ireq6->pktopts = skb; 430 } 431 ireq6->iif = sk->sk_bound_dev_if; 432 433 /* So that link locals have meaning */ 434 if (!sk->sk_bound_dev_if && 435 ipv6_addr_type(&ireq6->rmt_addr) & IPV6_ADDR_LINKLOCAL) 436 ireq6->iif = inet6_iif(skb); 437 438 /* 439 * Step 3: Process LISTEN state 440 * 441 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie 442 * 443 * In fact we defer setting S.GSR, S.SWL, S.SWH to 444 * dccp_create_openreq_child. 445 */ 446 dreq->dreq_isr = dcb->dccpd_seq; 447 dreq->dreq_iss = dccp_v6_init_sequence(skb); 448 dreq->dreq_service = service; 449 450 if (dccp_v6_send_response(sk, req, NULL)) 451 goto drop_and_free; 452 453 inet6_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT); 454 return 0; 455 456 drop_and_free: 457 reqsk_free(req); 458 drop: 459 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS); 460 return -1; 461 } 462 463 static struct sock *dccp_v6_request_recv_sock(struct sock *sk, 464 struct sk_buff *skb, 465 struct request_sock *req, 466 struct dst_entry *dst) 467 { 468 struct inet6_request_sock *ireq6 = inet6_rsk(req); 469 struct ipv6_pinfo *newnp, *np = inet6_sk(sk); 470 struct inet_sock *newinet; 471 struct dccp6_sock *newdp6; 472 struct sock *newsk; 473 struct ipv6_txoptions *opt; 474 475 if (skb->protocol == htons(ETH_P_IP)) { 476 /* 477 * v6 mapped 478 */ 479 newsk = dccp_v4_request_recv_sock(sk, skb, req, dst); 480 if (newsk == NULL) 481 return NULL; 482 483 newdp6 = (struct dccp6_sock *)newsk; 484 newinet = inet_sk(newsk); 485 newinet->pinet6 = &newdp6->inet6; 486 newnp = inet6_sk(newsk); 487 488 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 489 490 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newnp->daddr); 491 492 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr); 493 494 newnp->rcv_saddr = newnp->saddr; 495 496 inet_csk(newsk)->icsk_af_ops = &dccp_ipv6_mapped; 497 newsk->sk_backlog_rcv = dccp_v4_do_rcv; 498 newnp->pktoptions = NULL; 499 newnp->opt = NULL; 500 newnp->mcast_oif = inet6_iif(skb); 501 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; 502 503 /* 504 * No need to charge this sock to the relevant IPv6 refcnt debug socks count 505 * here, dccp_create_openreq_child now does this for us, see the comment in 506 * that function for the gory details. -acme 507 */ 508 509 /* It is tricky place. Until this moment IPv4 tcp 510 worked with IPv6 icsk.icsk_af_ops. 511 Sync it now. 512 */ 513 dccp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie); 514 515 return newsk; 516 } 517 518 opt = np->opt; 519 520 if (sk_acceptq_is_full(sk)) 521 goto out_overflow; 522 523 if (dst == NULL) { 524 struct in6_addr *final_p, final; 525 struct flowi6 fl6; 526 527 memset(&fl6, 0, sizeof(fl6)); 528 fl6.flowi6_proto = IPPROTO_DCCP; 529 fl6.daddr = ireq6->rmt_addr; 530 final_p = fl6_update_dst(&fl6, opt, &final); 531 fl6.saddr = ireq6->loc_addr; 532 fl6.flowi6_oif = sk->sk_bound_dev_if; 533 fl6.fl6_dport = inet_rsk(req)->rmt_port; 534 fl6.fl6_sport = inet_rsk(req)->loc_port; 535 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 536 537 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false); 538 if (IS_ERR(dst)) 539 goto out; 540 } 541 542 newsk = dccp_create_openreq_child(sk, req, skb); 543 if (newsk == NULL) 544 goto out_nonewsk; 545 546 /* 547 * No need to charge this sock to the relevant IPv6 refcnt debug socks 548 * count here, dccp_create_openreq_child now does this for us, see the 549 * comment in that function for the gory details. -acme 550 */ 551 552 __ip6_dst_store(newsk, dst, NULL, NULL); 553 newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM | 554 NETIF_F_TSO); 555 newdp6 = (struct dccp6_sock *)newsk; 556 newinet = inet_sk(newsk); 557 newinet->pinet6 = &newdp6->inet6; 558 newnp = inet6_sk(newsk); 559 560 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 561 562 newnp->daddr = ireq6->rmt_addr; 563 newnp->saddr = ireq6->loc_addr; 564 newnp->rcv_saddr = ireq6->loc_addr; 565 newsk->sk_bound_dev_if = ireq6->iif; 566 567 /* Now IPv6 options... 568 569 First: no IPv4 options. 570 */ 571 newinet->inet_opt = NULL; 572 573 /* Clone RX bits */ 574 newnp->rxopt.all = np->rxopt.all; 575 576 /* Clone pktoptions received with SYN */ 577 newnp->pktoptions = NULL; 578 if (ireq6->pktopts != NULL) { 579 newnp->pktoptions = skb_clone(ireq6->pktopts, GFP_ATOMIC); 580 kfree_skb(ireq6->pktopts); 581 ireq6->pktopts = NULL; 582 if (newnp->pktoptions) 583 skb_set_owner_r(newnp->pktoptions, newsk); 584 } 585 newnp->opt = NULL; 586 newnp->mcast_oif = inet6_iif(skb); 587 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; 588 589 /* 590 * Clone native IPv6 options from listening socket (if any) 591 * 592 * Yes, keeping reference count would be much more clever, but we make 593 * one more one thing there: reattach optmem to newsk. 594 */ 595 if (opt != NULL) { 596 newnp->opt = ipv6_dup_options(newsk, opt); 597 if (opt != np->opt) 598 sock_kfree_s(sk, opt, opt->tot_len); 599 } 600 601 inet_csk(newsk)->icsk_ext_hdr_len = 0; 602 if (newnp->opt != NULL) 603 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen + 604 newnp->opt->opt_flen); 605 606 dccp_sync_mss(newsk, dst_mtu(dst)); 607 608 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6; 609 newinet->inet_rcv_saddr = LOOPBACK4_IPV6; 610 611 if (__inet_inherit_port(sk, newsk) < 0) { 612 sock_put(newsk); 613 goto out; 614 } 615 __inet6_hash(newsk, NULL); 616 617 return newsk; 618 619 out_overflow: 620 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 621 out_nonewsk: 622 dst_release(dst); 623 out: 624 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); 625 if (opt != NULL && opt != np->opt) 626 sock_kfree_s(sk, opt, opt->tot_len); 627 return NULL; 628 } 629 630 /* The socket must have it's spinlock held when we get 631 * here. 632 * 633 * We have a potential double-lock case here, so even when 634 * doing backlog processing we use the BH locking scheme. 635 * This is because we cannot sleep with the original spinlock 636 * held. 637 */ 638 static int dccp_v6_do_rcv(struct sock *sk, struct sk_buff *skb) 639 { 640 struct ipv6_pinfo *np = inet6_sk(sk); 641 struct sk_buff *opt_skb = NULL; 642 643 /* Imagine: socket is IPv6. IPv4 packet arrives, 644 goes to IPv4 receive handler and backlogged. 645 From backlog it always goes here. Kerboom... 646 Fortunately, dccp_rcv_established and rcv_established 647 handle them correctly, but it is not case with 648 dccp_v6_hnd_req and dccp_v6_ctl_send_reset(). --ANK 649 */ 650 651 if (skb->protocol == htons(ETH_P_IP)) 652 return dccp_v4_do_rcv(sk, skb); 653 654 if (sk_filter(sk, skb)) 655 goto discard; 656 657 /* 658 * socket locking is here for SMP purposes as backlog rcv is currently 659 * called with bh processing disabled. 660 */ 661 662 /* Do Stevens' IPV6_PKTOPTIONS. 663 664 Yes, guys, it is the only place in our code, where we 665 may make it not affecting IPv4. 666 The rest of code is protocol independent, 667 and I do not like idea to uglify IPv4. 668 669 Actually, all the idea behind IPV6_PKTOPTIONS 670 looks not very well thought. For now we latch 671 options, received in the last packet, enqueued 672 by tcp. Feel free to propose better solution. 673 --ANK (980728) 674 */ 675 if (np->rxopt.all) 676 /* 677 * FIXME: Add handling of IPV6_PKTOPTIONS skb. See the comments below 678 * (wrt ipv6_pktopions) and net/ipv6/tcp_ipv6.c for an example. 679 */ 680 opt_skb = skb_clone(skb, GFP_ATOMIC); 681 682 if (sk->sk_state == DCCP_OPEN) { /* Fast path */ 683 if (dccp_rcv_established(sk, skb, dccp_hdr(skb), skb->len)) 684 goto reset; 685 if (opt_skb) { 686 /* XXX This is where we would goto ipv6_pktoptions. */ 687 __kfree_skb(opt_skb); 688 } 689 return 0; 690 } 691 692 /* 693 * Step 3: Process LISTEN state 694 * If S.state == LISTEN, 695 * If P.type == Request or P contains a valid Init Cookie option, 696 * (* Must scan the packet's options to check for Init 697 * Cookies. Only Init Cookies are processed here, 698 * however; other options are processed in Step 8. This 699 * scan need only be performed if the endpoint uses Init 700 * Cookies *) 701 * (* Generate a new socket and switch to that socket *) 702 * Set S := new socket for this port pair 703 * S.state = RESPOND 704 * Choose S.ISS (initial seqno) or set from Init Cookies 705 * Initialize S.GAR := S.ISS 706 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies 707 * Continue with S.state == RESPOND 708 * (* A Response packet will be generated in Step 11 *) 709 * Otherwise, 710 * Generate Reset(No Connection) unless P.type == Reset 711 * Drop packet and return 712 * 713 * NOTE: the check for the packet types is done in 714 * dccp_rcv_state_process 715 */ 716 if (sk->sk_state == DCCP_LISTEN) { 717 struct sock *nsk = dccp_v6_hnd_req(sk, skb); 718 719 if (nsk == NULL) 720 goto discard; 721 /* 722 * Queue it on the new socket if the new socket is active, 723 * otherwise we just shortcircuit this and continue with 724 * the new socket.. 725 */ 726 if (nsk != sk) { 727 if (dccp_child_process(sk, nsk, skb)) 728 goto reset; 729 if (opt_skb != NULL) 730 __kfree_skb(opt_skb); 731 return 0; 732 } 733 } 734 735 if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len)) 736 goto reset; 737 if (opt_skb) { 738 /* XXX This is where we would goto ipv6_pktoptions. */ 739 __kfree_skb(opt_skb); 740 } 741 return 0; 742 743 reset: 744 dccp_v6_ctl_send_reset(sk, skb); 745 discard: 746 if (opt_skb != NULL) 747 __kfree_skb(opt_skb); 748 kfree_skb(skb); 749 return 0; 750 } 751 752 static int dccp_v6_rcv(struct sk_buff *skb) 753 { 754 const struct dccp_hdr *dh; 755 struct sock *sk; 756 int min_cov; 757 758 /* Step 1: Check header basics */ 759 760 if (dccp_invalid_packet(skb)) 761 goto discard_it; 762 763 /* Step 1: If header checksum is incorrect, drop packet and return. */ 764 if (dccp_v6_csum_finish(skb, &ipv6_hdr(skb)->saddr, 765 &ipv6_hdr(skb)->daddr)) { 766 DCCP_WARN("dropped packet with invalid checksum\n"); 767 goto discard_it; 768 } 769 770 dh = dccp_hdr(skb); 771 772 DCCP_SKB_CB(skb)->dccpd_seq = dccp_hdr_seq(dh); 773 DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type; 774 775 if (dccp_packet_without_ack(skb)) 776 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ; 777 else 778 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb); 779 780 /* Step 2: 781 * Look up flow ID in table and get corresponding socket */ 782 sk = __inet6_lookup_skb(&dccp_hashinfo, skb, 783 dh->dccph_sport, dh->dccph_dport); 784 /* 785 * Step 2: 786 * If no socket ... 787 */ 788 if (sk == NULL) { 789 dccp_pr_debug("failed to look up flow ID in table and " 790 "get corresponding socket\n"); 791 goto no_dccp_socket; 792 } 793 794 /* 795 * Step 2: 796 * ... or S.state == TIMEWAIT, 797 * Generate Reset(No Connection) unless P.type == Reset 798 * Drop packet and return 799 */ 800 if (sk->sk_state == DCCP_TIME_WAIT) { 801 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n"); 802 inet_twsk_put(inet_twsk(sk)); 803 goto no_dccp_socket; 804 } 805 806 /* 807 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage 808 * o if MinCsCov = 0, only packets with CsCov = 0 are accepted 809 * o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov 810 */ 811 min_cov = dccp_sk(sk)->dccps_pcrlen; 812 if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov)) { 813 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n", 814 dh->dccph_cscov, min_cov); 815 /* FIXME: send Data Dropped option (see also dccp_v4_rcv) */ 816 goto discard_and_relse; 817 } 818 819 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 820 goto discard_and_relse; 821 822 return sk_receive_skb(sk, skb, 1) ? -1 : 0; 823 824 no_dccp_socket: 825 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 826 goto discard_it; 827 /* 828 * Step 2: 829 * If no socket ... 830 * Generate Reset(No Connection) unless P.type == Reset 831 * Drop packet and return 832 */ 833 if (dh->dccph_type != DCCP_PKT_RESET) { 834 DCCP_SKB_CB(skb)->dccpd_reset_code = 835 DCCP_RESET_CODE_NO_CONNECTION; 836 dccp_v6_ctl_send_reset(sk, skb); 837 } 838 839 discard_it: 840 kfree_skb(skb); 841 return 0; 842 843 discard_and_relse: 844 sock_put(sk); 845 goto discard_it; 846 } 847 848 static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr, 849 int addr_len) 850 { 851 struct sockaddr_in6 *usin = (struct sockaddr_in6 *)uaddr; 852 struct inet_connection_sock *icsk = inet_csk(sk); 853 struct inet_sock *inet = inet_sk(sk); 854 struct ipv6_pinfo *np = inet6_sk(sk); 855 struct dccp_sock *dp = dccp_sk(sk); 856 struct in6_addr *saddr = NULL, *final_p, final; 857 struct flowi6 fl6; 858 struct dst_entry *dst; 859 int addr_type; 860 int err; 861 862 dp->dccps_role = DCCP_ROLE_CLIENT; 863 864 if (addr_len < SIN6_LEN_RFC2133) 865 return -EINVAL; 866 867 if (usin->sin6_family != AF_INET6) 868 return -EAFNOSUPPORT; 869 870 memset(&fl6, 0, sizeof(fl6)); 871 872 if (np->sndflow) { 873 fl6.flowlabel = usin->sin6_flowinfo & IPV6_FLOWINFO_MASK; 874 IP6_ECN_flow_init(fl6.flowlabel); 875 if (fl6.flowlabel & IPV6_FLOWLABEL_MASK) { 876 struct ip6_flowlabel *flowlabel; 877 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 878 if (flowlabel == NULL) 879 return -EINVAL; 880 usin->sin6_addr = flowlabel->dst; 881 fl6_sock_release(flowlabel); 882 } 883 } 884 /* 885 * connect() to INADDR_ANY means loopback (BSD'ism). 886 */ 887 if (ipv6_addr_any(&usin->sin6_addr)) 888 usin->sin6_addr.s6_addr[15] = 1; 889 890 addr_type = ipv6_addr_type(&usin->sin6_addr); 891 892 if (addr_type & IPV6_ADDR_MULTICAST) 893 return -ENETUNREACH; 894 895 if (addr_type & IPV6_ADDR_LINKLOCAL) { 896 if (addr_len >= sizeof(struct sockaddr_in6) && 897 usin->sin6_scope_id) { 898 /* If interface is set while binding, indices 899 * must coincide. 900 */ 901 if (sk->sk_bound_dev_if && 902 sk->sk_bound_dev_if != usin->sin6_scope_id) 903 return -EINVAL; 904 905 sk->sk_bound_dev_if = usin->sin6_scope_id; 906 } 907 908 /* Connect to link-local address requires an interface */ 909 if (!sk->sk_bound_dev_if) 910 return -EINVAL; 911 } 912 913 np->daddr = usin->sin6_addr; 914 np->flow_label = fl6.flowlabel; 915 916 /* 917 * DCCP over IPv4 918 */ 919 if (addr_type == IPV6_ADDR_MAPPED) { 920 u32 exthdrlen = icsk->icsk_ext_hdr_len; 921 struct sockaddr_in sin; 922 923 SOCK_DEBUG(sk, "connect: ipv4 mapped\n"); 924 925 if (__ipv6_only_sock(sk)) 926 return -ENETUNREACH; 927 928 sin.sin_family = AF_INET; 929 sin.sin_port = usin->sin6_port; 930 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3]; 931 932 icsk->icsk_af_ops = &dccp_ipv6_mapped; 933 sk->sk_backlog_rcv = dccp_v4_do_rcv; 934 935 err = dccp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin)); 936 if (err) { 937 icsk->icsk_ext_hdr_len = exthdrlen; 938 icsk->icsk_af_ops = &dccp_ipv6_af_ops; 939 sk->sk_backlog_rcv = dccp_v6_do_rcv; 940 goto failure; 941 } 942 ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr); 943 ipv6_addr_set_v4mapped(inet->inet_rcv_saddr, &np->rcv_saddr); 944 945 return err; 946 } 947 948 if (!ipv6_addr_any(&np->rcv_saddr)) 949 saddr = &np->rcv_saddr; 950 951 fl6.flowi6_proto = IPPROTO_DCCP; 952 fl6.daddr = np->daddr; 953 fl6.saddr = saddr ? *saddr : np->saddr; 954 fl6.flowi6_oif = sk->sk_bound_dev_if; 955 fl6.fl6_dport = usin->sin6_port; 956 fl6.fl6_sport = inet->inet_sport; 957 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 958 959 final_p = fl6_update_dst(&fl6, np->opt, &final); 960 961 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, true); 962 if (IS_ERR(dst)) { 963 err = PTR_ERR(dst); 964 goto failure; 965 } 966 967 if (saddr == NULL) { 968 saddr = &fl6.saddr; 969 np->rcv_saddr = *saddr; 970 } 971 972 /* set the source address */ 973 np->saddr = *saddr; 974 inet->inet_rcv_saddr = LOOPBACK4_IPV6; 975 976 __ip6_dst_store(sk, dst, NULL, NULL); 977 978 icsk->icsk_ext_hdr_len = 0; 979 if (np->opt != NULL) 980 icsk->icsk_ext_hdr_len = (np->opt->opt_flen + 981 np->opt->opt_nflen); 982 983 inet->inet_dport = usin->sin6_port; 984 985 dccp_set_state(sk, DCCP_REQUESTING); 986 err = inet6_hash_connect(&dccp_death_row, sk); 987 if (err) 988 goto late_failure; 989 990 dp->dccps_iss = secure_dccpv6_sequence_number(np->saddr.s6_addr32, 991 np->daddr.s6_addr32, 992 inet->inet_sport, 993 inet->inet_dport); 994 err = dccp_connect(sk); 995 if (err) 996 goto late_failure; 997 998 return 0; 999 1000 late_failure: 1001 dccp_set_state(sk, DCCP_CLOSED); 1002 __sk_dst_reset(sk); 1003 failure: 1004 inet->inet_dport = 0; 1005 sk->sk_route_caps = 0; 1006 return err; 1007 } 1008 1009 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops = { 1010 .queue_xmit = inet6_csk_xmit, 1011 .send_check = dccp_v6_send_check, 1012 .rebuild_header = inet6_sk_rebuild_header, 1013 .conn_request = dccp_v6_conn_request, 1014 .syn_recv_sock = dccp_v6_request_recv_sock, 1015 .net_header_len = sizeof(struct ipv6hdr), 1016 .setsockopt = ipv6_setsockopt, 1017 .getsockopt = ipv6_getsockopt, 1018 .addr2sockaddr = inet6_csk_addr2sockaddr, 1019 .sockaddr_len = sizeof(struct sockaddr_in6), 1020 .bind_conflict = inet6_csk_bind_conflict, 1021 #ifdef CONFIG_COMPAT 1022 .compat_setsockopt = compat_ipv6_setsockopt, 1023 .compat_getsockopt = compat_ipv6_getsockopt, 1024 #endif 1025 }; 1026 1027 /* 1028 * DCCP over IPv4 via INET6 API 1029 */ 1030 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped = { 1031 .queue_xmit = ip_queue_xmit, 1032 .send_check = dccp_v4_send_check, 1033 .rebuild_header = inet_sk_rebuild_header, 1034 .conn_request = dccp_v6_conn_request, 1035 .syn_recv_sock = dccp_v6_request_recv_sock, 1036 .net_header_len = sizeof(struct iphdr), 1037 .setsockopt = ipv6_setsockopt, 1038 .getsockopt = ipv6_getsockopt, 1039 .addr2sockaddr = inet6_csk_addr2sockaddr, 1040 .sockaddr_len = sizeof(struct sockaddr_in6), 1041 #ifdef CONFIG_COMPAT 1042 .compat_setsockopt = compat_ipv6_setsockopt, 1043 .compat_getsockopt = compat_ipv6_getsockopt, 1044 #endif 1045 }; 1046 1047 /* NOTE: A lot of things set to zero explicitly by call to 1048 * sk_alloc() so need not be done here. 1049 */ 1050 static int dccp_v6_init_sock(struct sock *sk) 1051 { 1052 static __u8 dccp_v6_ctl_sock_initialized; 1053 int err = dccp_init_sock(sk, dccp_v6_ctl_sock_initialized); 1054 1055 if (err == 0) { 1056 if (unlikely(!dccp_v6_ctl_sock_initialized)) 1057 dccp_v6_ctl_sock_initialized = 1; 1058 inet_csk(sk)->icsk_af_ops = &dccp_ipv6_af_ops; 1059 } 1060 1061 return err; 1062 } 1063 1064 static void dccp_v6_destroy_sock(struct sock *sk) 1065 { 1066 dccp_destroy_sock(sk); 1067 inet6_destroy_sock(sk); 1068 } 1069 1070 static struct timewait_sock_ops dccp6_timewait_sock_ops = { 1071 .twsk_obj_size = sizeof(struct dccp6_timewait_sock), 1072 }; 1073 1074 static struct proto dccp_v6_prot = { 1075 .name = "DCCPv6", 1076 .owner = THIS_MODULE, 1077 .close = dccp_close, 1078 .connect = dccp_v6_connect, 1079 .disconnect = dccp_disconnect, 1080 .ioctl = dccp_ioctl, 1081 .init = dccp_v6_init_sock, 1082 .setsockopt = dccp_setsockopt, 1083 .getsockopt = dccp_getsockopt, 1084 .sendmsg = dccp_sendmsg, 1085 .recvmsg = dccp_recvmsg, 1086 .backlog_rcv = dccp_v6_do_rcv, 1087 .hash = dccp_v6_hash, 1088 .unhash = inet_unhash, 1089 .accept = inet_csk_accept, 1090 .get_port = inet_csk_get_port, 1091 .shutdown = dccp_shutdown, 1092 .destroy = dccp_v6_destroy_sock, 1093 .orphan_count = &dccp_orphan_count, 1094 .max_header = MAX_DCCP_HEADER, 1095 .obj_size = sizeof(struct dccp6_sock), 1096 .slab_flags = SLAB_DESTROY_BY_RCU, 1097 .rsk_prot = &dccp6_request_sock_ops, 1098 .twsk_prot = &dccp6_timewait_sock_ops, 1099 .h.hashinfo = &dccp_hashinfo, 1100 #ifdef CONFIG_COMPAT 1101 .compat_setsockopt = compat_dccp_setsockopt, 1102 .compat_getsockopt = compat_dccp_getsockopt, 1103 #endif 1104 }; 1105 1106 static const struct inet6_protocol dccp_v6_protocol = { 1107 .handler = dccp_v6_rcv, 1108 .err_handler = dccp_v6_err, 1109 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL, 1110 }; 1111 1112 static const struct proto_ops inet6_dccp_ops = { 1113 .family = PF_INET6, 1114 .owner = THIS_MODULE, 1115 .release = inet6_release, 1116 .bind = inet6_bind, 1117 .connect = inet_stream_connect, 1118 .socketpair = sock_no_socketpair, 1119 .accept = inet_accept, 1120 .getname = inet6_getname, 1121 .poll = dccp_poll, 1122 .ioctl = inet6_ioctl, 1123 .listen = inet_dccp_listen, 1124 .shutdown = inet_shutdown, 1125 .setsockopt = sock_common_setsockopt, 1126 .getsockopt = sock_common_getsockopt, 1127 .sendmsg = inet_sendmsg, 1128 .recvmsg = sock_common_recvmsg, 1129 .mmap = sock_no_mmap, 1130 .sendpage = sock_no_sendpage, 1131 #ifdef CONFIG_COMPAT 1132 .compat_setsockopt = compat_sock_common_setsockopt, 1133 .compat_getsockopt = compat_sock_common_getsockopt, 1134 #endif 1135 }; 1136 1137 static struct inet_protosw dccp_v6_protosw = { 1138 .type = SOCK_DCCP, 1139 .protocol = IPPROTO_DCCP, 1140 .prot = &dccp_v6_prot, 1141 .ops = &inet6_dccp_ops, 1142 .flags = INET_PROTOSW_ICSK, 1143 }; 1144 1145 static int __net_init dccp_v6_init_net(struct net *net) 1146 { 1147 if (dccp_hashinfo.bhash == NULL) 1148 return -ESOCKTNOSUPPORT; 1149 1150 return inet_ctl_sock_create(&net->dccp.v6_ctl_sk, PF_INET6, 1151 SOCK_DCCP, IPPROTO_DCCP, net); 1152 } 1153 1154 static void __net_exit dccp_v6_exit_net(struct net *net) 1155 { 1156 inet_ctl_sock_destroy(net->dccp.v6_ctl_sk); 1157 } 1158 1159 static struct pernet_operations dccp_v6_ops = { 1160 .init = dccp_v6_init_net, 1161 .exit = dccp_v6_exit_net, 1162 }; 1163 1164 static int __init dccp_v6_init(void) 1165 { 1166 int err = proto_register(&dccp_v6_prot, 1); 1167 1168 if (err != 0) 1169 goto out; 1170 1171 err = inet6_add_protocol(&dccp_v6_protocol, IPPROTO_DCCP); 1172 if (err != 0) 1173 goto out_unregister_proto; 1174 1175 inet6_register_protosw(&dccp_v6_protosw); 1176 1177 err = register_pernet_subsys(&dccp_v6_ops); 1178 if (err != 0) 1179 goto out_destroy_ctl_sock; 1180 out: 1181 return err; 1182 1183 out_destroy_ctl_sock: 1184 inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP); 1185 inet6_unregister_protosw(&dccp_v6_protosw); 1186 out_unregister_proto: 1187 proto_unregister(&dccp_v6_prot); 1188 goto out; 1189 } 1190 1191 static void __exit dccp_v6_exit(void) 1192 { 1193 unregister_pernet_subsys(&dccp_v6_ops); 1194 inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP); 1195 inet6_unregister_protosw(&dccp_v6_protosw); 1196 proto_unregister(&dccp_v6_prot); 1197 } 1198 1199 module_init(dccp_v6_init); 1200 module_exit(dccp_v6_exit); 1201 1202 /* 1203 * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33) 1204 * values directly, Also cover the case where the protocol is not specified, 1205 * i.e. net-pf-PF_INET6-proto-0-type-SOCK_DCCP 1206 */ 1207 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 33, 6); 1208 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 0, 6); 1209 MODULE_LICENSE("GPL"); 1210 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>"); 1211 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol"); 1212