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