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 33 #include "dccp.h" 34 #include "ipv6.h" 35 #include "feat.h" 36 37 /* The per-net dccp.v6_ctl_sk is used for sending RSTs and ACKs */ 38 39 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped; 40 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops; 41 42 static void dccp_v6_hash(struct sock *sk) 43 { 44 if (sk->sk_state != DCCP_CLOSED) { 45 if (inet_csk(sk)->icsk_af_ops == &dccp_ipv6_mapped) { 46 inet_hash(sk); 47 return; 48 } 49 local_bh_disable(); 50 __inet6_hash(sk, NULL); 51 local_bh_enable(); 52 } 53 } 54 55 /* add pseudo-header to DCCP checksum stored in skb->csum */ 56 static inline __sum16 dccp_v6_csum_finish(struct sk_buff *skb, 57 const struct in6_addr *saddr, 58 const struct in6_addr *daddr) 59 { 60 return csum_ipv6_magic(saddr, daddr, skb->len, IPPROTO_DCCP, skb->csum); 61 } 62 63 static inline void dccp_v6_send_check(struct sock *sk, struct sk_buff *skb) 64 { 65 struct ipv6_pinfo *np = inet6_sk(sk); 66 struct dccp_hdr *dh = dccp_hdr(skb); 67 68 dccp_csum_outgoing(skb); 69 dh->dccph_checksum = dccp_v6_csum_finish(skb, &np->saddr, &np->daddr); 70 } 71 72 static inline __u32 secure_dccpv6_sequence_number(__be32 *saddr, __be32 *daddr, 73 __be16 sport, __be16 dport ) 74 { 75 return secure_tcpv6_sequence_number(saddr, daddr, sport, dport); 76 } 77 78 static inline __u32 dccp_v6_init_sequence(struct sk_buff *skb) 79 { 80 return secure_dccpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32, 81 ipv6_hdr(skb)->saddr.s6_addr32, 82 dccp_hdr(skb)->dccph_dport, 83 dccp_hdr(skb)->dccph_sport ); 84 85 } 86 87 static void dccp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 88 u8 type, u8 code, int offset, __be32 info) 89 { 90 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 91 const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset); 92 struct dccp_sock *dp; 93 struct ipv6_pinfo *np; 94 struct sock *sk; 95 int err; 96 __u64 seq; 97 struct net *net = dev_net(skb->dev); 98 99 if (skb->len < offset + sizeof(*dh) || 100 skb->len < offset + __dccp_basic_hdr_len(dh)) { 101 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), 102 ICMP6_MIB_INERRORS); 103 return; 104 } 105 106 sk = inet6_lookup(net, &dccp_hashinfo, 107 &hdr->daddr, dh->dccph_dport, 108 &hdr->saddr, dh->dccph_sport, inet6_iif(skb)); 109 110 if (sk == NULL) { 111 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), 112 ICMP6_MIB_INERRORS); 113 return; 114 } 115 116 if (sk->sk_state == DCCP_TIME_WAIT) { 117 inet_twsk_put(inet_twsk(sk)); 118 return; 119 } 120 121 bh_lock_sock(sk); 122 if (sock_owned_by_user(sk)) 123 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS); 124 125 if (sk->sk_state == DCCP_CLOSED) 126 goto out; 127 128 dp = dccp_sk(sk); 129 seq = dccp_hdr_seq(dh); 130 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) && 131 !between48(seq, dp->dccps_awl, dp->dccps_awh)) { 132 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); 133 goto out; 134 } 135 136 np = inet6_sk(sk); 137 138 if (type == ICMPV6_PKT_TOOBIG) { 139 struct dst_entry *dst = NULL; 140 141 if (sock_owned_by_user(sk)) 142 goto out; 143 if ((1 << sk->sk_state) & (DCCPF_LISTEN | DCCPF_CLOSED)) 144 goto out; 145 146 /* icmp should have updated the destination cache entry */ 147 dst = __sk_dst_check(sk, np->dst_cookie); 148 if (dst == NULL) { 149 struct inet_sock *inet = inet_sk(sk); 150 struct flowi6 fl6; 151 152 /* BUGGG_FUTURE: Again, it is not clear how 153 to handle rthdr case. Ignore this complexity 154 for now. 155 */ 156 memset(&fl6, 0, sizeof(fl6)); 157 fl6.flowi6_proto = IPPROTO_DCCP; 158 ipv6_addr_copy(&fl6.daddr, &np->daddr); 159 ipv6_addr_copy(&fl6.saddr, &np->saddr); 160 fl6.flowi6_oif = sk->sk_bound_dev_if; 161 fl6.fl6_dport = inet->inet_dport; 162 fl6.fl6_sport = inet->inet_sport; 163 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 164 165 dst = ip6_dst_lookup_flow(sk, &fl6, NULL, false); 166 if (IS_ERR(dst)) { 167 sk->sk_err_soft = -PTR_ERR(dst); 168 goto out; 169 } 170 } else 171 dst_hold(dst); 172 173 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) { 174 dccp_sync_mss(sk, dst_mtu(dst)); 175 } /* else let the usual retransmit timer handle it */ 176 dst_release(dst); 177 goto out; 178 } 179 180 icmpv6_err_convert(type, code, &err); 181 182 /* Might be for an request_sock */ 183 switch (sk->sk_state) { 184 struct request_sock *req, **prev; 185 case DCCP_LISTEN: 186 if (sock_owned_by_user(sk)) 187 goto out; 188 189 req = inet6_csk_search_req(sk, &prev, dh->dccph_dport, 190 &hdr->daddr, &hdr->saddr, 191 inet6_iif(skb)); 192 if (req == NULL) 193 goto out; 194 195 /* 196 * ICMPs are not backlogged, hence we cannot get an established 197 * socket here. 198 */ 199 WARN_ON(req->sk != NULL); 200 201 if (seq != dccp_rsk(req)->dreq_iss) { 202 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); 203 goto out; 204 } 205 206 inet_csk_reqsk_queue_drop(sk, req, prev); 207 goto out; 208 209 case DCCP_REQUESTING: 210 case DCCP_RESPOND: /* Cannot happen. 211 It can, it SYNs are crossed. --ANK */ 212 if (!sock_owned_by_user(sk)) { 213 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS); 214 sk->sk_err = err; 215 /* 216 * Wake people up to see the error 217 * (see connect in sock.c) 218 */ 219 sk->sk_error_report(sk); 220 dccp_done(sk); 221 } else 222 sk->sk_err_soft = err; 223 goto out; 224 } 225 226 if (!sock_owned_by_user(sk) && np->recverr) { 227 sk->sk_err = err; 228 sk->sk_error_report(sk); 229 } else 230 sk->sk_err_soft = err; 231 232 out: 233 bh_unlock_sock(sk); 234 sock_put(sk); 235 } 236 237 238 static int dccp_v6_send_response(struct sock *sk, struct request_sock *req, 239 struct request_values *rv_unused) 240 { 241 struct inet6_request_sock *ireq6 = inet6_rsk(req); 242 struct ipv6_pinfo *np = inet6_sk(sk); 243 struct sk_buff *skb; 244 struct ipv6_txoptions *opt = NULL; 245 struct in6_addr *final_p, final; 246 struct flowi6 fl6; 247 int err = -1; 248 struct dst_entry *dst; 249 250 memset(&fl6, 0, sizeof(fl6)); 251 fl6.flowi6_proto = IPPROTO_DCCP; 252 ipv6_addr_copy(&fl6.daddr, &ireq6->rmt_addr); 253 ipv6_addr_copy(&fl6.saddr, &ireq6->loc_addr); 254 fl6.flowlabel = 0; 255 fl6.flowi6_oif = ireq6->iif; 256 fl6.fl6_dport = inet_rsk(req)->rmt_port; 257 fl6.fl6_sport = inet_rsk(req)->loc_port; 258 security_req_classify_flow(req, flowi6_to_flowi(&fl6)); 259 260 opt = np->opt; 261 262 final_p = fl6_update_dst(&fl6, opt, &final); 263 264 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false); 265 if (IS_ERR(dst)) { 266 err = PTR_ERR(dst); 267 dst = NULL; 268 goto done; 269 } 270 271 skb = dccp_make_response(sk, dst, req); 272 if (skb != NULL) { 273 struct dccp_hdr *dh = dccp_hdr(skb); 274 275 dh->dccph_checksum = dccp_v6_csum_finish(skb, 276 &ireq6->loc_addr, 277 &ireq6->rmt_addr); 278 ipv6_addr_copy(&fl6.daddr, &ireq6->rmt_addr); 279 err = ip6_xmit(sk, skb, &fl6, opt); 280 err = net_xmit_eval(err); 281 } 282 283 done: 284 if (opt != NULL && opt != np->opt) 285 sock_kfree_s(sk, opt, opt->tot_len); 286 dst_release(dst); 287 return err; 288 } 289 290 static void dccp_v6_reqsk_destructor(struct request_sock *req) 291 { 292 dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg); 293 if (inet6_rsk(req)->pktopts != NULL) 294 kfree_skb(inet6_rsk(req)->pktopts); 295 } 296 297 static void dccp_v6_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb) 298 { 299 const struct ipv6hdr *rxip6h; 300 struct sk_buff *skb; 301 struct flowi6 fl6; 302 struct net *net = dev_net(skb_dst(rxskb)->dev); 303 struct sock *ctl_sk = net->dccp.v6_ctl_sk; 304 struct dst_entry *dst; 305 306 if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET) 307 return; 308 309 if (!ipv6_unicast_destination(rxskb)) 310 return; 311 312 skb = dccp_ctl_make_reset(ctl_sk, rxskb); 313 if (skb == NULL) 314 return; 315 316 rxip6h = ipv6_hdr(rxskb); 317 dccp_hdr(skb)->dccph_checksum = dccp_v6_csum_finish(skb, &rxip6h->saddr, 318 &rxip6h->daddr); 319 320 memset(&fl6, 0, sizeof(fl6)); 321 ipv6_addr_copy(&fl6.daddr, &rxip6h->saddr); 322 ipv6_addr_copy(&fl6.saddr, &rxip6h->daddr); 323 324 fl6.flowi6_proto = IPPROTO_DCCP; 325 fl6.flowi6_oif = inet6_iif(rxskb); 326 fl6.fl6_dport = dccp_hdr(skb)->dccph_dport; 327 fl6.fl6_sport = dccp_hdr(skb)->dccph_sport; 328 security_skb_classify_flow(rxskb, flowi6_to_flowi(&fl6)); 329 330 /* sk = NULL, but it is safe for now. RST socket required. */ 331 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL, false); 332 if (!IS_ERR(dst)) { 333 skb_dst_set(skb, dst); 334 ip6_xmit(ctl_sk, skb, &fl6, NULL); 335 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS); 336 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS); 337 return; 338 } 339 340 kfree_skb(skb); 341 } 342 343 static struct request_sock_ops dccp6_request_sock_ops = { 344 .family = AF_INET6, 345 .obj_size = sizeof(struct dccp6_request_sock), 346 .rtx_syn_ack = dccp_v6_send_response, 347 .send_ack = dccp_reqsk_send_ack, 348 .destructor = dccp_v6_reqsk_destructor, 349 .send_reset = dccp_v6_ctl_send_reset, 350 }; 351 352 static struct sock *dccp_v6_hnd_req(struct sock *sk,struct sk_buff *skb) 353 { 354 const struct dccp_hdr *dh = dccp_hdr(skb); 355 const struct ipv6hdr *iph = ipv6_hdr(skb); 356 struct sock *nsk; 357 struct request_sock **prev; 358 /* Find possible connection requests. */ 359 struct request_sock *req = inet6_csk_search_req(sk, &prev, 360 dh->dccph_sport, 361 &iph->saddr, 362 &iph->daddr, 363 inet6_iif(skb)); 364 if (req != NULL) 365 return dccp_check_req(sk, skb, req, prev); 366 367 nsk = __inet6_lookup_established(sock_net(sk), &dccp_hashinfo, 368 &iph->saddr, dh->dccph_sport, 369 &iph->daddr, ntohs(dh->dccph_dport), 370 inet6_iif(skb)); 371 if (nsk != NULL) { 372 if (nsk->sk_state != DCCP_TIME_WAIT) { 373 bh_lock_sock(nsk); 374 return nsk; 375 } 376 inet_twsk_put(inet_twsk(nsk)); 377 return NULL; 378 } 379 380 return sk; 381 } 382 383 static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb) 384 { 385 struct request_sock *req; 386 struct dccp_request_sock *dreq; 387 struct inet6_request_sock *ireq6; 388 struct ipv6_pinfo *np = inet6_sk(sk); 389 const __be32 service = dccp_hdr_request(skb)->dccph_req_service; 390 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb); 391 392 if (skb->protocol == htons(ETH_P_IP)) 393 return dccp_v4_conn_request(sk, skb); 394 395 if (!ipv6_unicast_destination(skb)) 396 return 0; /* discard, don't send a reset here */ 397 398 if (dccp_bad_service_code(sk, service)) { 399 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE; 400 goto drop; 401 } 402 /* 403 * There are no SYN attacks on IPv6, yet... 404 */ 405 dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY; 406 if (inet_csk_reqsk_queue_is_full(sk)) 407 goto drop; 408 409 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) 410 goto drop; 411 412 req = inet6_reqsk_alloc(&dccp6_request_sock_ops); 413 if (req == NULL) 414 goto drop; 415 416 if (dccp_reqsk_init(req, dccp_sk(sk), skb)) 417 goto drop_and_free; 418 419 dreq = dccp_rsk(req); 420 if (dccp_parse_options(sk, dreq, skb)) 421 goto drop_and_free; 422 423 if (security_inet_conn_request(sk, skb, req)) 424 goto drop_and_free; 425 426 ireq6 = inet6_rsk(req); 427 ipv6_addr_copy(&ireq6->rmt_addr, &ipv6_hdr(skb)->saddr); 428 ipv6_addr_copy(&ireq6->loc_addr, &ipv6_hdr(skb)->daddr); 429 430 if (ipv6_opt_accepted(sk, skb) || 431 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo || 432 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) { 433 atomic_inc(&skb->users); 434 ireq6->pktopts = skb; 435 } 436 ireq6->iif = sk->sk_bound_dev_if; 437 438 /* So that link locals have meaning */ 439 if (!sk->sk_bound_dev_if && 440 ipv6_addr_type(&ireq6->rmt_addr) & IPV6_ADDR_LINKLOCAL) 441 ireq6->iif = inet6_iif(skb); 442 443 /* 444 * Step 3: Process LISTEN state 445 * 446 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie 447 * 448 * In fact we defer setting S.GSR, S.SWL, S.SWH to 449 * dccp_create_openreq_child. 450 */ 451 dreq->dreq_isr = dcb->dccpd_seq; 452 dreq->dreq_iss = dccp_v6_init_sequence(skb); 453 dreq->dreq_service = service; 454 455 if (dccp_v6_send_response(sk, req, NULL)) 456 goto drop_and_free; 457 458 inet6_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT); 459 return 0; 460 461 drop_and_free: 462 reqsk_free(req); 463 drop: 464 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS); 465 return -1; 466 } 467 468 static struct sock *dccp_v6_request_recv_sock(struct sock *sk, 469 struct sk_buff *skb, 470 struct request_sock *req, 471 struct dst_entry *dst) 472 { 473 struct inet6_request_sock *ireq6 = inet6_rsk(req); 474 struct ipv6_pinfo *newnp, *np = inet6_sk(sk); 475 struct inet_sock *newinet; 476 struct dccp6_sock *newdp6; 477 struct sock *newsk; 478 struct ipv6_txoptions *opt; 479 480 if (skb->protocol == htons(ETH_P_IP)) { 481 /* 482 * v6 mapped 483 */ 484 newsk = dccp_v4_request_recv_sock(sk, skb, req, dst); 485 if (newsk == NULL) 486 return NULL; 487 488 newdp6 = (struct dccp6_sock *)newsk; 489 newinet = inet_sk(newsk); 490 newinet->pinet6 = &newdp6->inet6; 491 newnp = inet6_sk(newsk); 492 493 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 494 495 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newnp->daddr); 496 497 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr); 498 499 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr); 500 501 inet_csk(newsk)->icsk_af_ops = &dccp_ipv6_mapped; 502 newsk->sk_backlog_rcv = dccp_v4_do_rcv; 503 newnp->pktoptions = NULL; 504 newnp->opt = NULL; 505 newnp->mcast_oif = inet6_iif(skb); 506 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; 507 508 /* 509 * No need to charge this sock to the relevant IPv6 refcnt debug socks count 510 * here, dccp_create_openreq_child now does this for us, see the comment in 511 * that function for the gory details. -acme 512 */ 513 514 /* It is tricky place. Until this moment IPv4 tcp 515 worked with IPv6 icsk.icsk_af_ops. 516 Sync it now. 517 */ 518 dccp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie); 519 520 return newsk; 521 } 522 523 opt = np->opt; 524 525 if (sk_acceptq_is_full(sk)) 526 goto out_overflow; 527 528 if (dst == NULL) { 529 struct in6_addr *final_p, final; 530 struct flowi6 fl6; 531 532 memset(&fl6, 0, sizeof(fl6)); 533 fl6.flowi6_proto = IPPROTO_DCCP; 534 ipv6_addr_copy(&fl6.daddr, &ireq6->rmt_addr); 535 final_p = fl6_update_dst(&fl6, opt, &final); 536 ipv6_addr_copy(&fl6.saddr, &ireq6->loc_addr); 537 fl6.flowi6_oif = sk->sk_bound_dev_if; 538 fl6.fl6_dport = inet_rsk(req)->rmt_port; 539 fl6.fl6_sport = inet_rsk(req)->loc_port; 540 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 541 542 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false); 543 if (IS_ERR(dst)) 544 goto out; 545 } 546 547 newsk = dccp_create_openreq_child(sk, req, skb); 548 if (newsk == NULL) 549 goto out_nonewsk; 550 551 /* 552 * No need to charge this sock to the relevant IPv6 refcnt debug socks 553 * count here, dccp_create_openreq_child now does this for us, see the 554 * comment in that function for the gory details. -acme 555 */ 556 557 __ip6_dst_store(newsk, dst, NULL, NULL); 558 newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM | 559 NETIF_F_TSO); 560 newdp6 = (struct dccp6_sock *)newsk; 561 newinet = inet_sk(newsk); 562 newinet->pinet6 = &newdp6->inet6; 563 newnp = inet6_sk(newsk); 564 565 memcpy(newnp, np, sizeof(struct ipv6_pinfo)); 566 567 ipv6_addr_copy(&newnp->daddr, &ireq6->rmt_addr); 568 ipv6_addr_copy(&newnp->saddr, &ireq6->loc_addr); 569 ipv6_addr_copy(&newnp->rcv_saddr, &ireq6->loc_addr); 570 newsk->sk_bound_dev_if = ireq6->iif; 571 572 /* Now IPv6 options... 573 574 First: no IPv4 options. 575 */ 576 newinet->inet_opt = NULL; 577 578 /* Clone RX bits */ 579 newnp->rxopt.all = np->rxopt.all; 580 581 /* Clone pktoptions received with SYN */ 582 newnp->pktoptions = NULL; 583 if (ireq6->pktopts != NULL) { 584 newnp->pktoptions = skb_clone(ireq6->pktopts, GFP_ATOMIC); 585 kfree_skb(ireq6->pktopts); 586 ireq6->pktopts = NULL; 587 if (newnp->pktoptions) 588 skb_set_owner_r(newnp->pktoptions, newsk); 589 } 590 newnp->opt = NULL; 591 newnp->mcast_oif = inet6_iif(skb); 592 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; 593 594 /* 595 * Clone native IPv6 options from listening socket (if any) 596 * 597 * Yes, keeping reference count would be much more clever, but we make 598 * one more one thing there: reattach optmem to newsk. 599 */ 600 if (opt != NULL) { 601 newnp->opt = ipv6_dup_options(newsk, opt); 602 if (opt != np->opt) 603 sock_kfree_s(sk, opt, opt->tot_len); 604 } 605 606 inet_csk(newsk)->icsk_ext_hdr_len = 0; 607 if (newnp->opt != NULL) 608 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen + 609 newnp->opt->opt_flen); 610 611 dccp_sync_mss(newsk, dst_mtu(dst)); 612 613 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6; 614 newinet->inet_rcv_saddr = LOOPBACK4_IPV6; 615 616 if (__inet_inherit_port(sk, newsk) < 0) { 617 sock_put(newsk); 618 goto out; 619 } 620 __inet6_hash(newsk, NULL); 621 622 return newsk; 623 624 out_overflow: 625 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 626 out_nonewsk: 627 dst_release(dst); 628 out: 629 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); 630 if (opt != NULL && opt != np->opt) 631 sock_kfree_s(sk, opt, opt->tot_len); 632 return NULL; 633 } 634 635 /* The socket must have it's spinlock held when we get 636 * here. 637 * 638 * We have a potential double-lock case here, so even when 639 * doing backlog processing we use the BH locking scheme. 640 * This is because we cannot sleep with the original spinlock 641 * held. 642 */ 643 static int dccp_v6_do_rcv(struct sock *sk, struct sk_buff *skb) 644 { 645 struct ipv6_pinfo *np = inet6_sk(sk); 646 struct sk_buff *opt_skb = NULL; 647 648 /* Imagine: socket is IPv6. IPv4 packet arrives, 649 goes to IPv4 receive handler and backlogged. 650 From backlog it always goes here. Kerboom... 651 Fortunately, dccp_rcv_established and rcv_established 652 handle them correctly, but it is not case with 653 dccp_v6_hnd_req and dccp_v6_ctl_send_reset(). --ANK 654 */ 655 656 if (skb->protocol == htons(ETH_P_IP)) 657 return dccp_v4_do_rcv(sk, skb); 658 659 if (sk_filter(sk, skb)) 660 goto discard; 661 662 /* 663 * socket locking is here for SMP purposes as backlog rcv is currently 664 * called with bh processing disabled. 665 */ 666 667 /* Do Stevens' IPV6_PKTOPTIONS. 668 669 Yes, guys, it is the only place in our code, where we 670 may make it not affecting IPv4. 671 The rest of code is protocol independent, 672 and I do not like idea to uglify IPv4. 673 674 Actually, all the idea behind IPV6_PKTOPTIONS 675 looks not very well thought. For now we latch 676 options, received in the last packet, enqueued 677 by tcp. Feel free to propose better solution. 678 --ANK (980728) 679 */ 680 if (np->rxopt.all) 681 /* 682 * FIXME: Add handling of IPV6_PKTOPTIONS skb. See the comments below 683 * (wrt ipv6_pktopions) and net/ipv6/tcp_ipv6.c for an example. 684 */ 685 opt_skb = skb_clone(skb, GFP_ATOMIC); 686 687 if (sk->sk_state == DCCP_OPEN) { /* Fast path */ 688 if (dccp_rcv_established(sk, skb, dccp_hdr(skb), skb->len)) 689 goto reset; 690 if (opt_skb) { 691 /* XXX This is where we would goto ipv6_pktoptions. */ 692 __kfree_skb(opt_skb); 693 } 694 return 0; 695 } 696 697 /* 698 * Step 3: Process LISTEN state 699 * If S.state == LISTEN, 700 * If P.type == Request or P contains a valid Init Cookie option, 701 * (* Must scan the packet's options to check for Init 702 * Cookies. Only Init Cookies are processed here, 703 * however; other options are processed in Step 8. This 704 * scan need only be performed if the endpoint uses Init 705 * Cookies *) 706 * (* Generate a new socket and switch to that socket *) 707 * Set S := new socket for this port pair 708 * S.state = RESPOND 709 * Choose S.ISS (initial seqno) or set from Init Cookies 710 * Initialize S.GAR := S.ISS 711 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies 712 * Continue with S.state == RESPOND 713 * (* A Response packet will be generated in Step 11 *) 714 * Otherwise, 715 * Generate Reset(No Connection) unless P.type == Reset 716 * Drop packet and return 717 * 718 * NOTE: the check for the packet types is done in 719 * dccp_rcv_state_process 720 */ 721 if (sk->sk_state == DCCP_LISTEN) { 722 struct sock *nsk = dccp_v6_hnd_req(sk, skb); 723 724 if (nsk == NULL) 725 goto discard; 726 /* 727 * Queue it on the new socket if the new socket is active, 728 * otherwise we just shortcircuit this and continue with 729 * the new socket.. 730 */ 731 if (nsk != sk) { 732 if (dccp_child_process(sk, nsk, skb)) 733 goto reset; 734 if (opt_skb != NULL) 735 __kfree_skb(opt_skb); 736 return 0; 737 } 738 } 739 740 if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len)) 741 goto reset; 742 if (opt_skb) { 743 /* XXX This is where we would goto ipv6_pktoptions. */ 744 __kfree_skb(opt_skb); 745 } 746 return 0; 747 748 reset: 749 dccp_v6_ctl_send_reset(sk, skb); 750 discard: 751 if (opt_skb != NULL) 752 __kfree_skb(opt_skb); 753 kfree_skb(skb); 754 return 0; 755 } 756 757 static int dccp_v6_rcv(struct sk_buff *skb) 758 { 759 const struct dccp_hdr *dh; 760 struct sock *sk; 761 int min_cov; 762 763 /* Step 1: Check header basics */ 764 765 if (dccp_invalid_packet(skb)) 766 goto discard_it; 767 768 /* Step 1: If header checksum is incorrect, drop packet and return. */ 769 if (dccp_v6_csum_finish(skb, &ipv6_hdr(skb)->saddr, 770 &ipv6_hdr(skb)->daddr)) { 771 DCCP_WARN("dropped packet with invalid checksum\n"); 772 goto discard_it; 773 } 774 775 dh = dccp_hdr(skb); 776 777 DCCP_SKB_CB(skb)->dccpd_seq = dccp_hdr_seq(dh); 778 DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type; 779 780 if (dccp_packet_without_ack(skb)) 781 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ; 782 else 783 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb); 784 785 /* Step 2: 786 * Look up flow ID in table and get corresponding socket */ 787 sk = __inet6_lookup_skb(&dccp_hashinfo, skb, 788 dh->dccph_sport, dh->dccph_dport); 789 /* 790 * Step 2: 791 * If no socket ... 792 */ 793 if (sk == NULL) { 794 dccp_pr_debug("failed to look up flow ID in table and " 795 "get corresponding socket\n"); 796 goto no_dccp_socket; 797 } 798 799 /* 800 * Step 2: 801 * ... or S.state == TIMEWAIT, 802 * Generate Reset(No Connection) unless P.type == Reset 803 * Drop packet and return 804 */ 805 if (sk->sk_state == DCCP_TIME_WAIT) { 806 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n"); 807 inet_twsk_put(inet_twsk(sk)); 808 goto no_dccp_socket; 809 } 810 811 /* 812 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage 813 * o if MinCsCov = 0, only packets with CsCov = 0 are accepted 814 * o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov 815 */ 816 min_cov = dccp_sk(sk)->dccps_pcrlen; 817 if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov)) { 818 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n", 819 dh->dccph_cscov, min_cov); 820 /* FIXME: send Data Dropped option (see also dccp_v4_rcv) */ 821 goto discard_and_relse; 822 } 823 824 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 825 goto discard_and_relse; 826 827 return sk_receive_skb(sk, skb, 1) ? -1 : 0; 828 829 no_dccp_socket: 830 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 831 goto discard_it; 832 /* 833 * Step 2: 834 * If no socket ... 835 * Generate Reset(No Connection) unless P.type == Reset 836 * Drop packet and return 837 */ 838 if (dh->dccph_type != DCCP_PKT_RESET) { 839 DCCP_SKB_CB(skb)->dccpd_reset_code = 840 DCCP_RESET_CODE_NO_CONNECTION; 841 dccp_v6_ctl_send_reset(sk, skb); 842 } 843 844 discard_it: 845 kfree_skb(skb); 846 return 0; 847 848 discard_and_relse: 849 sock_put(sk); 850 goto discard_it; 851 } 852 853 static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr, 854 int addr_len) 855 { 856 struct sockaddr_in6 *usin = (struct sockaddr_in6 *)uaddr; 857 struct inet_connection_sock *icsk = inet_csk(sk); 858 struct inet_sock *inet = inet_sk(sk); 859 struct ipv6_pinfo *np = inet6_sk(sk); 860 struct dccp_sock *dp = dccp_sk(sk); 861 struct in6_addr *saddr = NULL, *final_p, final; 862 struct flowi6 fl6; 863 struct dst_entry *dst; 864 int addr_type; 865 int err; 866 867 dp->dccps_role = DCCP_ROLE_CLIENT; 868 869 if (addr_len < SIN6_LEN_RFC2133) 870 return -EINVAL; 871 872 if (usin->sin6_family != AF_INET6) 873 return -EAFNOSUPPORT; 874 875 memset(&fl6, 0, sizeof(fl6)); 876 877 if (np->sndflow) { 878 fl6.flowlabel = usin->sin6_flowinfo & IPV6_FLOWINFO_MASK; 879 IP6_ECN_flow_init(fl6.flowlabel); 880 if (fl6.flowlabel & IPV6_FLOWLABEL_MASK) { 881 struct ip6_flowlabel *flowlabel; 882 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 883 if (flowlabel == NULL) 884 return -EINVAL; 885 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst); 886 fl6_sock_release(flowlabel); 887 } 888 } 889 /* 890 * connect() to INADDR_ANY means loopback (BSD'ism). 891 */ 892 if (ipv6_addr_any(&usin->sin6_addr)) 893 usin->sin6_addr.s6_addr[15] = 1; 894 895 addr_type = ipv6_addr_type(&usin->sin6_addr); 896 897 if (addr_type & IPV6_ADDR_MULTICAST) 898 return -ENETUNREACH; 899 900 if (addr_type & IPV6_ADDR_LINKLOCAL) { 901 if (addr_len >= sizeof(struct sockaddr_in6) && 902 usin->sin6_scope_id) { 903 /* If interface is set while binding, indices 904 * must coincide. 905 */ 906 if (sk->sk_bound_dev_if && 907 sk->sk_bound_dev_if != usin->sin6_scope_id) 908 return -EINVAL; 909 910 sk->sk_bound_dev_if = usin->sin6_scope_id; 911 } 912 913 /* Connect to link-local address requires an interface */ 914 if (!sk->sk_bound_dev_if) 915 return -EINVAL; 916 } 917 918 ipv6_addr_copy(&np->daddr, &usin->sin6_addr); 919 np->flow_label = fl6.flowlabel; 920 921 /* 922 * DCCP over IPv4 923 */ 924 if (addr_type == IPV6_ADDR_MAPPED) { 925 u32 exthdrlen = icsk->icsk_ext_hdr_len; 926 struct sockaddr_in sin; 927 928 SOCK_DEBUG(sk, "connect: ipv4 mapped\n"); 929 930 if (__ipv6_only_sock(sk)) 931 return -ENETUNREACH; 932 933 sin.sin_family = AF_INET; 934 sin.sin_port = usin->sin6_port; 935 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3]; 936 937 icsk->icsk_af_ops = &dccp_ipv6_mapped; 938 sk->sk_backlog_rcv = dccp_v4_do_rcv; 939 940 err = dccp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin)); 941 if (err) { 942 icsk->icsk_ext_hdr_len = exthdrlen; 943 icsk->icsk_af_ops = &dccp_ipv6_af_ops; 944 sk->sk_backlog_rcv = dccp_v6_do_rcv; 945 goto failure; 946 } 947 ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr); 948 ipv6_addr_set_v4mapped(inet->inet_rcv_saddr, &np->rcv_saddr); 949 950 return err; 951 } 952 953 if (!ipv6_addr_any(&np->rcv_saddr)) 954 saddr = &np->rcv_saddr; 955 956 fl6.flowi6_proto = IPPROTO_DCCP; 957 ipv6_addr_copy(&fl6.daddr, &np->daddr); 958 ipv6_addr_copy(&fl6.saddr, saddr ? saddr : &np->saddr); 959 fl6.flowi6_oif = sk->sk_bound_dev_if; 960 fl6.fl6_dport = usin->sin6_port; 961 fl6.fl6_sport = inet->inet_sport; 962 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 963 964 final_p = fl6_update_dst(&fl6, np->opt, &final); 965 966 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, true); 967 if (IS_ERR(dst)) { 968 err = PTR_ERR(dst); 969 goto failure; 970 } 971 972 if (saddr == NULL) { 973 saddr = &fl6.saddr; 974 ipv6_addr_copy(&np->rcv_saddr, saddr); 975 } 976 977 /* set the source address */ 978 ipv6_addr_copy(&np->saddr, saddr); 979 inet->inet_rcv_saddr = LOOPBACK4_IPV6; 980 981 __ip6_dst_store(sk, dst, NULL, NULL); 982 983 icsk->icsk_ext_hdr_len = 0; 984 if (np->opt != NULL) 985 icsk->icsk_ext_hdr_len = (np->opt->opt_flen + 986 np->opt->opt_nflen); 987 988 inet->inet_dport = usin->sin6_port; 989 990 dccp_set_state(sk, DCCP_REQUESTING); 991 err = inet6_hash_connect(&dccp_death_row, sk); 992 if (err) 993 goto late_failure; 994 995 dp->dccps_iss = secure_dccpv6_sequence_number(np->saddr.s6_addr32, 996 np->daddr.s6_addr32, 997 inet->inet_sport, 998 inet->inet_dport); 999 err = dccp_connect(sk); 1000 if (err) 1001 goto late_failure; 1002 1003 return 0; 1004 1005 late_failure: 1006 dccp_set_state(sk, DCCP_CLOSED); 1007 __sk_dst_reset(sk); 1008 failure: 1009 inet->inet_dport = 0; 1010 sk->sk_route_caps = 0; 1011 return err; 1012 } 1013 1014 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops = { 1015 .queue_xmit = inet6_csk_xmit, 1016 .send_check = dccp_v6_send_check, 1017 .rebuild_header = inet6_sk_rebuild_header, 1018 .conn_request = dccp_v6_conn_request, 1019 .syn_recv_sock = dccp_v6_request_recv_sock, 1020 .net_header_len = sizeof(struct ipv6hdr), 1021 .setsockopt = ipv6_setsockopt, 1022 .getsockopt = ipv6_getsockopt, 1023 .addr2sockaddr = inet6_csk_addr2sockaddr, 1024 .sockaddr_len = sizeof(struct sockaddr_in6), 1025 .bind_conflict = inet6_csk_bind_conflict, 1026 #ifdef CONFIG_COMPAT 1027 .compat_setsockopt = compat_ipv6_setsockopt, 1028 .compat_getsockopt = compat_ipv6_getsockopt, 1029 #endif 1030 }; 1031 1032 /* 1033 * DCCP over IPv4 via INET6 API 1034 */ 1035 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped = { 1036 .queue_xmit = ip_queue_xmit, 1037 .send_check = dccp_v4_send_check, 1038 .rebuild_header = inet_sk_rebuild_header, 1039 .conn_request = dccp_v6_conn_request, 1040 .syn_recv_sock = dccp_v6_request_recv_sock, 1041 .net_header_len = sizeof(struct iphdr), 1042 .setsockopt = ipv6_setsockopt, 1043 .getsockopt = ipv6_getsockopt, 1044 .addr2sockaddr = inet6_csk_addr2sockaddr, 1045 .sockaddr_len = sizeof(struct sockaddr_in6), 1046 #ifdef CONFIG_COMPAT 1047 .compat_setsockopt = compat_ipv6_setsockopt, 1048 .compat_getsockopt = compat_ipv6_getsockopt, 1049 #endif 1050 }; 1051 1052 /* NOTE: A lot of things set to zero explicitly by call to 1053 * sk_alloc() so need not be done here. 1054 */ 1055 static int dccp_v6_init_sock(struct sock *sk) 1056 { 1057 static __u8 dccp_v6_ctl_sock_initialized; 1058 int err = dccp_init_sock(sk, dccp_v6_ctl_sock_initialized); 1059 1060 if (err == 0) { 1061 if (unlikely(!dccp_v6_ctl_sock_initialized)) 1062 dccp_v6_ctl_sock_initialized = 1; 1063 inet_csk(sk)->icsk_af_ops = &dccp_ipv6_af_ops; 1064 } 1065 1066 return err; 1067 } 1068 1069 static void dccp_v6_destroy_sock(struct sock *sk) 1070 { 1071 dccp_destroy_sock(sk); 1072 inet6_destroy_sock(sk); 1073 } 1074 1075 static struct timewait_sock_ops dccp6_timewait_sock_ops = { 1076 .twsk_obj_size = sizeof(struct dccp6_timewait_sock), 1077 }; 1078 1079 static struct proto dccp_v6_prot = { 1080 .name = "DCCPv6", 1081 .owner = THIS_MODULE, 1082 .close = dccp_close, 1083 .connect = dccp_v6_connect, 1084 .disconnect = dccp_disconnect, 1085 .ioctl = dccp_ioctl, 1086 .init = dccp_v6_init_sock, 1087 .setsockopt = dccp_setsockopt, 1088 .getsockopt = dccp_getsockopt, 1089 .sendmsg = dccp_sendmsg, 1090 .recvmsg = dccp_recvmsg, 1091 .backlog_rcv = dccp_v6_do_rcv, 1092 .hash = dccp_v6_hash, 1093 .unhash = inet_unhash, 1094 .accept = inet_csk_accept, 1095 .get_port = inet_csk_get_port, 1096 .shutdown = dccp_shutdown, 1097 .destroy = dccp_v6_destroy_sock, 1098 .orphan_count = &dccp_orphan_count, 1099 .max_header = MAX_DCCP_HEADER, 1100 .obj_size = sizeof(struct dccp6_sock), 1101 .slab_flags = SLAB_DESTROY_BY_RCU, 1102 .rsk_prot = &dccp6_request_sock_ops, 1103 .twsk_prot = &dccp6_timewait_sock_ops, 1104 .h.hashinfo = &dccp_hashinfo, 1105 #ifdef CONFIG_COMPAT 1106 .compat_setsockopt = compat_dccp_setsockopt, 1107 .compat_getsockopt = compat_dccp_getsockopt, 1108 #endif 1109 }; 1110 1111 static const struct inet6_protocol dccp_v6_protocol = { 1112 .handler = dccp_v6_rcv, 1113 .err_handler = dccp_v6_err, 1114 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL, 1115 }; 1116 1117 static const struct proto_ops inet6_dccp_ops = { 1118 .family = PF_INET6, 1119 .owner = THIS_MODULE, 1120 .release = inet6_release, 1121 .bind = inet6_bind, 1122 .connect = inet_stream_connect, 1123 .socketpair = sock_no_socketpair, 1124 .accept = inet_accept, 1125 .getname = inet6_getname, 1126 .poll = dccp_poll, 1127 .ioctl = inet6_ioctl, 1128 .listen = inet_dccp_listen, 1129 .shutdown = inet_shutdown, 1130 .setsockopt = sock_common_setsockopt, 1131 .getsockopt = sock_common_getsockopt, 1132 .sendmsg = inet_sendmsg, 1133 .recvmsg = sock_common_recvmsg, 1134 .mmap = sock_no_mmap, 1135 .sendpage = sock_no_sendpage, 1136 #ifdef CONFIG_COMPAT 1137 .compat_setsockopt = compat_sock_common_setsockopt, 1138 .compat_getsockopt = compat_sock_common_getsockopt, 1139 #endif 1140 }; 1141 1142 static struct inet_protosw dccp_v6_protosw = { 1143 .type = SOCK_DCCP, 1144 .protocol = IPPROTO_DCCP, 1145 .prot = &dccp_v6_prot, 1146 .ops = &inet6_dccp_ops, 1147 .flags = INET_PROTOSW_ICSK, 1148 }; 1149 1150 static int __net_init dccp_v6_init_net(struct net *net) 1151 { 1152 if (dccp_hashinfo.bhash == NULL) 1153 return -ESOCKTNOSUPPORT; 1154 1155 return inet_ctl_sock_create(&net->dccp.v6_ctl_sk, PF_INET6, 1156 SOCK_DCCP, IPPROTO_DCCP, net); 1157 } 1158 1159 static void __net_exit dccp_v6_exit_net(struct net *net) 1160 { 1161 inet_ctl_sock_destroy(net->dccp.v6_ctl_sk); 1162 } 1163 1164 static struct pernet_operations dccp_v6_ops = { 1165 .init = dccp_v6_init_net, 1166 .exit = dccp_v6_exit_net, 1167 }; 1168 1169 static int __init dccp_v6_init(void) 1170 { 1171 int err = proto_register(&dccp_v6_prot, 1); 1172 1173 if (err != 0) 1174 goto out; 1175 1176 err = inet6_add_protocol(&dccp_v6_protocol, IPPROTO_DCCP); 1177 if (err != 0) 1178 goto out_unregister_proto; 1179 1180 inet6_register_protosw(&dccp_v6_protosw); 1181 1182 err = register_pernet_subsys(&dccp_v6_ops); 1183 if (err != 0) 1184 goto out_destroy_ctl_sock; 1185 out: 1186 return err; 1187 1188 out_destroy_ctl_sock: 1189 inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP); 1190 inet6_unregister_protosw(&dccp_v6_protosw); 1191 out_unregister_proto: 1192 proto_unregister(&dccp_v6_prot); 1193 goto out; 1194 } 1195 1196 static void __exit dccp_v6_exit(void) 1197 { 1198 unregister_pernet_subsys(&dccp_v6_ops); 1199 inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP); 1200 inet6_unregister_protosw(&dccp_v6_protosw); 1201 proto_unregister(&dccp_v6_prot); 1202 } 1203 1204 module_init(dccp_v6_init); 1205 module_exit(dccp_v6_exit); 1206 1207 /* 1208 * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33) 1209 * values directly, Also cover the case where the protocol is not specified, 1210 * i.e. net-pf-PF_INET6-proto-0-type-SOCK_DCCP 1211 */ 1212 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 33, 6); 1213 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 0, 6); 1214 MODULE_LICENSE("GPL"); 1215 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>"); 1216 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol"); 1217