1 /* 2 * net/dccp/ipv4.c 3 * 4 * An implementation of the DCCP protocol 5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br> 6 * 7 * This program is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public License 9 * as published by the Free Software Foundation; either version 10 * 2 of the License, or (at your option) any later version. 11 */ 12 13 #include <linux/dccp.h> 14 #include <linux/icmp.h> 15 #include <linux/slab.h> 16 #include <linux/module.h> 17 #include <linux/skbuff.h> 18 #include <linux/random.h> 19 20 #include <net/icmp.h> 21 #include <net/inet_common.h> 22 #include <net/inet_hashtables.h> 23 #include <net/inet_sock.h> 24 #include <net/protocol.h> 25 #include <net/sock.h> 26 #include <net/timewait_sock.h> 27 #include <net/tcp_states.h> 28 #include <net/xfrm.h> 29 #include <net/secure_seq.h> 30 31 #include "ackvec.h" 32 #include "ccid.h" 33 #include "dccp.h" 34 #include "feat.h" 35 36 /* 37 * The per-net dccp.v4_ctl_sk socket is used for responding to 38 * the Out-of-the-blue (OOTB) packets. A control sock will be created 39 * for this socket at the initialization time. 40 */ 41 42 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 43 { 44 const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; 45 struct inet_sock *inet = inet_sk(sk); 46 struct dccp_sock *dp = dccp_sk(sk); 47 __be16 orig_sport, orig_dport; 48 __be32 daddr, nexthop; 49 struct flowi4 *fl4; 50 struct rtable *rt; 51 int err; 52 struct ip_options_rcu *inet_opt; 53 54 dp->dccps_role = DCCP_ROLE_CLIENT; 55 56 if (addr_len < sizeof(struct sockaddr_in)) 57 return -EINVAL; 58 59 if (usin->sin_family != AF_INET) 60 return -EAFNOSUPPORT; 61 62 nexthop = daddr = usin->sin_addr.s_addr; 63 64 inet_opt = rcu_dereference_protected(inet->inet_opt, 65 lockdep_sock_is_held(sk)); 66 if (inet_opt != NULL && inet_opt->opt.srr) { 67 if (daddr == 0) 68 return -EINVAL; 69 nexthop = inet_opt->opt.faddr; 70 } 71 72 orig_sport = inet->inet_sport; 73 orig_dport = usin->sin_port; 74 fl4 = &inet->cork.fl.u.ip4; 75 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr, 76 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, 77 IPPROTO_DCCP, 78 orig_sport, orig_dport, sk); 79 if (IS_ERR(rt)) 80 return PTR_ERR(rt); 81 82 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { 83 ip_rt_put(rt); 84 return -ENETUNREACH; 85 } 86 87 if (inet_opt == NULL || !inet_opt->opt.srr) 88 daddr = fl4->daddr; 89 90 if (inet->inet_saddr == 0) 91 inet->inet_saddr = fl4->saddr; 92 sk_rcv_saddr_set(sk, inet->inet_saddr); 93 inet->inet_dport = usin->sin_port; 94 sk_daddr_set(sk, daddr); 95 96 inet_csk(sk)->icsk_ext_hdr_len = 0; 97 if (inet_opt) 98 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen; 99 /* 100 * Socket identity is still unknown (sport may be zero). 101 * However we set state to DCCP_REQUESTING and not releasing socket 102 * lock select source port, enter ourselves into the hash tables and 103 * complete initialization after this. 104 */ 105 dccp_set_state(sk, DCCP_REQUESTING); 106 err = inet_hash_connect(&dccp_death_row, sk); 107 if (err != 0) 108 goto failure; 109 110 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport, 111 inet->inet_sport, inet->inet_dport, sk); 112 if (IS_ERR(rt)) { 113 err = PTR_ERR(rt); 114 rt = NULL; 115 goto failure; 116 } 117 /* OK, now commit destination to socket. */ 118 sk_setup_caps(sk, &rt->dst); 119 120 dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr, 121 inet->inet_daddr, 122 inet->inet_sport, 123 inet->inet_dport); 124 inet->inet_id = dp->dccps_iss ^ jiffies; 125 126 err = dccp_connect(sk); 127 rt = NULL; 128 if (err != 0) 129 goto failure; 130 out: 131 return err; 132 failure: 133 /* 134 * This unhashes the socket and releases the local port, if necessary. 135 */ 136 dccp_set_state(sk, DCCP_CLOSED); 137 ip_rt_put(rt); 138 sk->sk_route_caps = 0; 139 inet->inet_dport = 0; 140 goto out; 141 } 142 EXPORT_SYMBOL_GPL(dccp_v4_connect); 143 144 /* 145 * This routine does path mtu discovery as defined in RFC1191. 146 */ 147 static inline void dccp_do_pmtu_discovery(struct sock *sk, 148 const struct iphdr *iph, 149 u32 mtu) 150 { 151 struct dst_entry *dst; 152 const struct inet_sock *inet = inet_sk(sk); 153 const struct dccp_sock *dp = dccp_sk(sk); 154 155 /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs 156 * send out by Linux are always < 576bytes so they should go through 157 * unfragmented). 158 */ 159 if (sk->sk_state == DCCP_LISTEN) 160 return; 161 162 dst = inet_csk_update_pmtu(sk, mtu); 163 if (!dst) 164 return; 165 166 /* Something is about to be wrong... Remember soft error 167 * for the case, if this connection will not able to recover. 168 */ 169 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) 170 sk->sk_err_soft = EMSGSIZE; 171 172 mtu = dst_mtu(dst); 173 174 if (inet->pmtudisc != IP_PMTUDISC_DONT && 175 ip_sk_accept_pmtu(sk) && 176 inet_csk(sk)->icsk_pmtu_cookie > mtu) { 177 dccp_sync_mss(sk, mtu); 178 179 /* 180 * From RFC 4340, sec. 14.1: 181 * 182 * DCCP-Sync packets are the best choice for upward 183 * probing, since DCCP-Sync probes do not risk application 184 * data loss. 185 */ 186 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC); 187 } /* else let the usual retransmit timer handle it */ 188 } 189 190 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk) 191 { 192 struct dst_entry *dst = __sk_dst_check(sk, 0); 193 194 if (dst) 195 dst->ops->redirect(dst, sk, skb); 196 } 197 198 void dccp_req_err(struct sock *sk, u64 seq) 199 { 200 struct request_sock *req = inet_reqsk(sk); 201 struct net *net = sock_net(sk); 202 203 /* 204 * ICMPs are not backlogged, hence we cannot get an established 205 * socket here. 206 */ 207 if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) { 208 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); 209 } else { 210 /* 211 * Still in RESPOND, just remove it silently. 212 * There is no good way to pass the error to the newly 213 * created socket, and POSIX does not want network 214 * errors returned from accept(). 215 */ 216 inet_csk_reqsk_queue_drop(req->rsk_listener, req); 217 } 218 reqsk_put(req); 219 } 220 EXPORT_SYMBOL(dccp_req_err); 221 222 /* 223 * This routine is called by the ICMP module when it gets some sort of error 224 * condition. If err < 0 then the socket should be closed and the error 225 * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code. 226 * After adjustment header points to the first 8 bytes of the tcp header. We 227 * need to find the appropriate port. 228 * 229 * The locking strategy used here is very "optimistic". When someone else 230 * accesses the socket the ICMP is just dropped and for some paths there is no 231 * check at all. A more general error queue to queue errors for later handling 232 * is probably better. 233 */ 234 static void dccp_v4_err(struct sk_buff *skb, u32 info) 235 { 236 const struct iphdr *iph = (struct iphdr *)skb->data; 237 const u8 offset = iph->ihl << 2; 238 const struct dccp_hdr *dh; 239 struct dccp_sock *dp; 240 struct inet_sock *inet; 241 const int type = icmp_hdr(skb)->type; 242 const int code = icmp_hdr(skb)->code; 243 struct sock *sk; 244 __u64 seq; 245 int err; 246 struct net *net = dev_net(skb->dev); 247 248 /* Only need dccph_dport & dccph_sport which are the first 249 * 4 bytes in dccp header. 250 * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us. 251 */ 252 BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8); 253 BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8); 254 dh = (struct dccp_hdr *)(skb->data + offset); 255 256 sk = __inet_lookup_established(net, &dccp_hashinfo, 257 iph->daddr, dh->dccph_dport, 258 iph->saddr, ntohs(dh->dccph_sport), 259 inet_iif(skb), 0); 260 if (!sk) { 261 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 262 return; 263 } 264 265 if (sk->sk_state == DCCP_TIME_WAIT) { 266 inet_twsk_put(inet_twsk(sk)); 267 return; 268 } 269 seq = dccp_hdr_seq(dh); 270 if (sk->sk_state == DCCP_NEW_SYN_RECV) 271 return dccp_req_err(sk, seq); 272 273 bh_lock_sock(sk); 274 /* If too many ICMPs get dropped on busy 275 * servers this needs to be solved differently. 276 */ 277 if (sock_owned_by_user(sk)) 278 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS); 279 280 if (sk->sk_state == DCCP_CLOSED) 281 goto out; 282 283 dp = dccp_sk(sk); 284 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) && 285 !between48(seq, dp->dccps_awl, dp->dccps_awh)) { 286 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); 287 goto out; 288 } 289 290 switch (type) { 291 case ICMP_REDIRECT: 292 if (!sock_owned_by_user(sk)) 293 dccp_do_redirect(skb, sk); 294 goto out; 295 case ICMP_SOURCE_QUENCH: 296 /* Just silently ignore these. */ 297 goto out; 298 case ICMP_PARAMETERPROB: 299 err = EPROTO; 300 break; 301 case ICMP_DEST_UNREACH: 302 if (code > NR_ICMP_UNREACH) 303 goto out; 304 305 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ 306 if (!sock_owned_by_user(sk)) 307 dccp_do_pmtu_discovery(sk, iph, info); 308 goto out; 309 } 310 311 err = icmp_err_convert[code].errno; 312 break; 313 case ICMP_TIME_EXCEEDED: 314 err = EHOSTUNREACH; 315 break; 316 default: 317 goto out; 318 } 319 320 switch (sk->sk_state) { 321 case DCCP_REQUESTING: 322 case DCCP_RESPOND: 323 if (!sock_owned_by_user(sk)) { 324 __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS); 325 sk->sk_err = err; 326 327 sk->sk_error_report(sk); 328 329 dccp_done(sk); 330 } else 331 sk->sk_err_soft = err; 332 goto out; 333 } 334 335 /* If we've already connected we will keep trying 336 * until we time out, or the user gives up. 337 * 338 * rfc1122 4.2.3.9 allows to consider as hard errors 339 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, 340 * but it is obsoleted by pmtu discovery). 341 * 342 * Note, that in modern internet, where routing is unreliable 343 * and in each dark corner broken firewalls sit, sending random 344 * errors ordered by their masters even this two messages finally lose 345 * their original sense (even Linux sends invalid PORT_UNREACHs) 346 * 347 * Now we are in compliance with RFCs. 348 * --ANK (980905) 349 */ 350 351 inet = inet_sk(sk); 352 if (!sock_owned_by_user(sk) && inet->recverr) { 353 sk->sk_err = err; 354 sk->sk_error_report(sk); 355 } else /* Only an error on timeout */ 356 sk->sk_err_soft = err; 357 out: 358 bh_unlock_sock(sk); 359 sock_put(sk); 360 } 361 362 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb, 363 __be32 src, __be32 dst) 364 { 365 return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum); 366 } 367 368 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb) 369 { 370 const struct inet_sock *inet = inet_sk(sk); 371 struct dccp_hdr *dh = dccp_hdr(skb); 372 373 dccp_csum_outgoing(skb); 374 dh->dccph_checksum = dccp_v4_csum_finish(skb, 375 inet->inet_saddr, 376 inet->inet_daddr); 377 } 378 EXPORT_SYMBOL_GPL(dccp_v4_send_check); 379 380 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb) 381 { 382 return secure_dccp_sequence_number(ip_hdr(skb)->daddr, 383 ip_hdr(skb)->saddr, 384 dccp_hdr(skb)->dccph_dport, 385 dccp_hdr(skb)->dccph_sport); 386 } 387 388 /* 389 * The three way handshake has completed - we got a valid ACK or DATAACK - 390 * now create the new socket. 391 * 392 * This is the equivalent of TCP's tcp_v4_syn_recv_sock 393 */ 394 struct sock *dccp_v4_request_recv_sock(const struct sock *sk, 395 struct sk_buff *skb, 396 struct request_sock *req, 397 struct dst_entry *dst, 398 struct request_sock *req_unhash, 399 bool *own_req) 400 { 401 struct inet_request_sock *ireq; 402 struct inet_sock *newinet; 403 struct sock *newsk; 404 405 if (sk_acceptq_is_full(sk)) 406 goto exit_overflow; 407 408 newsk = dccp_create_openreq_child(sk, req, skb); 409 if (newsk == NULL) 410 goto exit_nonewsk; 411 412 newinet = inet_sk(newsk); 413 ireq = inet_rsk(req); 414 sk_daddr_set(newsk, ireq->ir_rmt_addr); 415 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr); 416 newinet->inet_saddr = ireq->ir_loc_addr; 417 RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt)); 418 newinet->mc_index = inet_iif(skb); 419 newinet->mc_ttl = ip_hdr(skb)->ttl; 420 newinet->inet_id = jiffies; 421 422 if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL) 423 goto put_and_exit; 424 425 sk_setup_caps(newsk, dst); 426 427 dccp_sync_mss(newsk, dst_mtu(dst)); 428 429 if (__inet_inherit_port(sk, newsk) < 0) 430 goto put_and_exit; 431 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash)); 432 if (*own_req) 433 ireq->ireq_opt = NULL; 434 else 435 newinet->inet_opt = NULL; 436 return newsk; 437 438 exit_overflow: 439 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 440 exit_nonewsk: 441 dst_release(dst); 442 exit: 443 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 444 return NULL; 445 put_and_exit: 446 newinet->inet_opt = NULL; 447 inet_csk_prepare_forced_close(newsk); 448 dccp_done(newsk); 449 goto exit; 450 } 451 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock); 452 453 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk, 454 struct sk_buff *skb) 455 { 456 struct rtable *rt; 457 const struct iphdr *iph = ip_hdr(skb); 458 struct flowi4 fl4 = { 459 .flowi4_oif = inet_iif(skb), 460 .daddr = iph->saddr, 461 .saddr = iph->daddr, 462 .flowi4_tos = RT_CONN_FLAGS(sk), 463 .flowi4_proto = sk->sk_protocol, 464 .fl4_sport = dccp_hdr(skb)->dccph_dport, 465 .fl4_dport = dccp_hdr(skb)->dccph_sport, 466 }; 467 468 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 469 rt = ip_route_output_flow(net, &fl4, sk); 470 if (IS_ERR(rt)) { 471 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 472 return NULL; 473 } 474 475 return &rt->dst; 476 } 477 478 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req) 479 { 480 int err = -1; 481 struct sk_buff *skb; 482 struct dst_entry *dst; 483 struct flowi4 fl4; 484 485 dst = inet_csk_route_req(sk, &fl4, req); 486 if (dst == NULL) 487 goto out; 488 489 skb = dccp_make_response(sk, dst, req); 490 if (skb != NULL) { 491 const struct inet_request_sock *ireq = inet_rsk(req); 492 struct dccp_hdr *dh = dccp_hdr(skb); 493 494 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr, 495 ireq->ir_rmt_addr); 496 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr, 497 ireq->ir_rmt_addr, 498 ireq_opt_deref(ireq)); 499 err = net_xmit_eval(err); 500 } 501 502 out: 503 dst_release(dst); 504 return err; 505 } 506 507 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb) 508 { 509 int err; 510 const struct iphdr *rxiph; 511 struct sk_buff *skb; 512 struct dst_entry *dst; 513 struct net *net = dev_net(skb_dst(rxskb)->dev); 514 struct sock *ctl_sk = net->dccp.v4_ctl_sk; 515 516 /* Never send a reset in response to a reset. */ 517 if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET) 518 return; 519 520 if (skb_rtable(rxskb)->rt_type != RTN_LOCAL) 521 return; 522 523 dst = dccp_v4_route_skb(net, ctl_sk, rxskb); 524 if (dst == NULL) 525 return; 526 527 skb = dccp_ctl_make_reset(ctl_sk, rxskb); 528 if (skb == NULL) 529 goto out; 530 531 rxiph = ip_hdr(rxskb); 532 dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr, 533 rxiph->daddr); 534 skb_dst_set(skb, dst_clone(dst)); 535 536 local_bh_disable(); 537 bh_lock_sock(ctl_sk); 538 err = ip_build_and_send_pkt(skb, ctl_sk, 539 rxiph->daddr, rxiph->saddr, NULL); 540 bh_unlock_sock(ctl_sk); 541 542 if (net_xmit_eval(err) == 0) { 543 __DCCP_INC_STATS(DCCP_MIB_OUTSEGS); 544 __DCCP_INC_STATS(DCCP_MIB_OUTRSTS); 545 } 546 local_bh_enable(); 547 out: 548 dst_release(dst); 549 } 550 551 static void dccp_v4_reqsk_destructor(struct request_sock *req) 552 { 553 dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg); 554 kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1)); 555 } 556 557 void dccp_syn_ack_timeout(const struct request_sock *req) 558 { 559 } 560 EXPORT_SYMBOL(dccp_syn_ack_timeout); 561 562 static struct request_sock_ops dccp_request_sock_ops __read_mostly = { 563 .family = PF_INET, 564 .obj_size = sizeof(struct dccp_request_sock), 565 .rtx_syn_ack = dccp_v4_send_response, 566 .send_ack = dccp_reqsk_send_ack, 567 .destructor = dccp_v4_reqsk_destructor, 568 .send_reset = dccp_v4_ctl_send_reset, 569 .syn_ack_timeout = dccp_syn_ack_timeout, 570 }; 571 572 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb) 573 { 574 struct inet_request_sock *ireq; 575 struct request_sock *req; 576 struct dccp_request_sock *dreq; 577 const __be32 service = dccp_hdr_request(skb)->dccph_req_service; 578 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb); 579 580 /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */ 581 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 582 return 0; /* discard, don't send a reset here */ 583 584 if (dccp_bad_service_code(sk, service)) { 585 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE; 586 goto drop; 587 } 588 /* 589 * TW buckets are converted to open requests without 590 * limitations, they conserve resources and peer is 591 * evidently real one. 592 */ 593 dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY; 594 if (inet_csk_reqsk_queue_is_full(sk)) 595 goto drop; 596 597 if (sk_acceptq_is_full(sk)) 598 goto drop; 599 600 req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true); 601 if (req == NULL) 602 goto drop; 603 604 if (dccp_reqsk_init(req, dccp_sk(sk), skb)) 605 goto drop_and_free; 606 607 dreq = dccp_rsk(req); 608 if (dccp_parse_options(sk, dreq, skb)) 609 goto drop_and_free; 610 611 if (security_inet_conn_request(sk, skb, req)) 612 goto drop_and_free; 613 614 ireq = inet_rsk(req); 615 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr); 616 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr); 617 ireq->ir_mark = inet_request_mark(sk, skb); 618 ireq->ireq_family = AF_INET; 619 ireq->ir_iif = sk->sk_bound_dev_if; 620 621 /* 622 * Step 3: Process LISTEN state 623 * 624 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie 625 * 626 * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child(). 627 */ 628 dreq->dreq_isr = dcb->dccpd_seq; 629 dreq->dreq_gsr = dreq->dreq_isr; 630 dreq->dreq_iss = dccp_v4_init_sequence(skb); 631 dreq->dreq_gss = dreq->dreq_iss; 632 dreq->dreq_service = service; 633 634 if (dccp_v4_send_response(sk, req)) 635 goto drop_and_free; 636 637 inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT); 638 reqsk_put(req); 639 return 0; 640 641 drop_and_free: 642 reqsk_free(req); 643 drop: 644 __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS); 645 return -1; 646 } 647 EXPORT_SYMBOL_GPL(dccp_v4_conn_request); 648 649 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) 650 { 651 struct dccp_hdr *dh = dccp_hdr(skb); 652 653 if (sk->sk_state == DCCP_OPEN) { /* Fast path */ 654 if (dccp_rcv_established(sk, skb, dh, skb->len)) 655 goto reset; 656 return 0; 657 } 658 659 /* 660 * Step 3: Process LISTEN state 661 * If P.type == Request or P contains a valid Init Cookie option, 662 * (* Must scan the packet's options to check for Init 663 * Cookies. Only Init Cookies are processed here, 664 * however; other options are processed in Step 8. This 665 * scan need only be performed if the endpoint uses Init 666 * Cookies *) 667 * (* Generate a new socket and switch to that socket *) 668 * Set S := new socket for this port pair 669 * S.state = RESPOND 670 * Choose S.ISS (initial seqno) or set from Init Cookies 671 * Initialize S.GAR := S.ISS 672 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies 673 * Continue with S.state == RESPOND 674 * (* A Response packet will be generated in Step 11 *) 675 * Otherwise, 676 * Generate Reset(No Connection) unless P.type == Reset 677 * Drop packet and return 678 * 679 * NOTE: the check for the packet types is done in 680 * dccp_rcv_state_process 681 */ 682 683 if (dccp_rcv_state_process(sk, skb, dh, skb->len)) 684 goto reset; 685 return 0; 686 687 reset: 688 dccp_v4_ctl_send_reset(sk, skb); 689 kfree_skb(skb); 690 return 0; 691 } 692 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv); 693 694 /** 695 * dccp_invalid_packet - check for malformed packets 696 * Implements RFC 4340, 8.5: Step 1: Check header basics 697 * Packets that fail these checks are ignored and do not receive Resets. 698 */ 699 int dccp_invalid_packet(struct sk_buff *skb) 700 { 701 const struct dccp_hdr *dh; 702 unsigned int cscov; 703 u8 dccph_doff; 704 705 if (skb->pkt_type != PACKET_HOST) 706 return 1; 707 708 /* If the packet is shorter than 12 bytes, drop packet and return */ 709 if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) { 710 DCCP_WARN("pskb_may_pull failed\n"); 711 return 1; 712 } 713 714 dh = dccp_hdr(skb); 715 716 /* If P.type is not understood, drop packet and return */ 717 if (dh->dccph_type >= DCCP_PKT_INVALID) { 718 DCCP_WARN("invalid packet type\n"); 719 return 1; 720 } 721 722 /* 723 * If P.Data Offset is too small for packet type, drop packet and return 724 */ 725 dccph_doff = dh->dccph_doff; 726 if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) { 727 DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff); 728 return 1; 729 } 730 /* 731 * If P.Data Offset is too too large for packet, drop packet and return 732 */ 733 if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) { 734 DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff); 735 return 1; 736 } 737 dh = dccp_hdr(skb); 738 /* 739 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet 740 * has short sequence numbers), drop packet and return 741 */ 742 if ((dh->dccph_type < DCCP_PKT_DATA || 743 dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0) { 744 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n", 745 dccp_packet_name(dh->dccph_type)); 746 return 1; 747 } 748 749 /* 750 * If P.CsCov is too large for the packet size, drop packet and return. 751 * This must come _before_ checksumming (not as RFC 4340 suggests). 752 */ 753 cscov = dccp_csum_coverage(skb); 754 if (cscov > skb->len) { 755 DCCP_WARN("P.CsCov %u exceeds packet length %d\n", 756 dh->dccph_cscov, skb->len); 757 return 1; 758 } 759 760 /* If header checksum is incorrect, drop packet and return. 761 * (This step is completed in the AF-dependent functions.) */ 762 skb->csum = skb_checksum(skb, 0, cscov, 0); 763 764 return 0; 765 } 766 EXPORT_SYMBOL_GPL(dccp_invalid_packet); 767 768 /* this is called when real data arrives */ 769 static int dccp_v4_rcv(struct sk_buff *skb) 770 { 771 const struct dccp_hdr *dh; 772 const struct iphdr *iph; 773 bool refcounted; 774 struct sock *sk; 775 int min_cov; 776 777 /* Step 1: Check header basics */ 778 779 if (dccp_invalid_packet(skb)) 780 goto discard_it; 781 782 iph = ip_hdr(skb); 783 /* Step 1: If header checksum is incorrect, drop packet and return */ 784 if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) { 785 DCCP_WARN("dropped packet with invalid checksum\n"); 786 goto discard_it; 787 } 788 789 dh = dccp_hdr(skb); 790 791 DCCP_SKB_CB(skb)->dccpd_seq = dccp_hdr_seq(dh); 792 DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type; 793 794 dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu", 795 dccp_packet_name(dh->dccph_type), 796 &iph->saddr, ntohs(dh->dccph_sport), 797 &iph->daddr, ntohs(dh->dccph_dport), 798 (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq); 799 800 if (dccp_packet_without_ack(skb)) { 801 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ; 802 dccp_pr_debug_cat("\n"); 803 } else { 804 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb); 805 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long) 806 DCCP_SKB_CB(skb)->dccpd_ack_seq); 807 } 808 809 lookup: 810 sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh), 811 dh->dccph_sport, dh->dccph_dport, 0, &refcounted); 812 if (!sk) { 813 dccp_pr_debug("failed to look up flow ID in table and " 814 "get corresponding socket\n"); 815 goto no_dccp_socket; 816 } 817 818 /* 819 * Step 2: 820 * ... or S.state == TIMEWAIT, 821 * Generate Reset(No Connection) unless P.type == Reset 822 * Drop packet and return 823 */ 824 if (sk->sk_state == DCCP_TIME_WAIT) { 825 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n"); 826 inet_twsk_put(inet_twsk(sk)); 827 goto no_dccp_socket; 828 } 829 830 if (sk->sk_state == DCCP_NEW_SYN_RECV) { 831 struct request_sock *req = inet_reqsk(sk); 832 struct sock *nsk; 833 834 sk = req->rsk_listener; 835 if (unlikely(sk->sk_state != DCCP_LISTEN)) { 836 inet_csk_reqsk_queue_drop_and_put(sk, req); 837 goto lookup; 838 } 839 sock_hold(sk); 840 refcounted = true; 841 nsk = dccp_check_req(sk, skb, req); 842 if (!nsk) { 843 reqsk_put(req); 844 goto discard_and_relse; 845 } 846 if (nsk == sk) { 847 reqsk_put(req); 848 } else if (dccp_child_process(sk, nsk, skb)) { 849 dccp_v4_ctl_send_reset(sk, skb); 850 goto discard_and_relse; 851 } else { 852 sock_put(sk); 853 return 0; 854 } 855 } 856 /* 857 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage 858 * o if MinCsCov = 0, only packets with CsCov = 0 are accepted 859 * o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov 860 */ 861 min_cov = dccp_sk(sk)->dccps_pcrlen; 862 if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov)) { 863 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n", 864 dh->dccph_cscov, min_cov); 865 /* FIXME: "Such packets SHOULD be reported using Data Dropped 866 * options (Section 11.7) with Drop Code 0, Protocol 867 * Constraints." */ 868 goto discard_and_relse; 869 } 870 871 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 872 goto discard_and_relse; 873 nf_reset(skb); 874 875 return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted); 876 877 no_dccp_socket: 878 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 879 goto discard_it; 880 /* 881 * Step 2: 882 * If no socket ... 883 * Generate Reset(No Connection) unless P.type == Reset 884 * Drop packet and return 885 */ 886 if (dh->dccph_type != DCCP_PKT_RESET) { 887 DCCP_SKB_CB(skb)->dccpd_reset_code = 888 DCCP_RESET_CODE_NO_CONNECTION; 889 dccp_v4_ctl_send_reset(sk, skb); 890 } 891 892 discard_it: 893 kfree_skb(skb); 894 return 0; 895 896 discard_and_relse: 897 if (refcounted) 898 sock_put(sk); 899 goto discard_it; 900 } 901 902 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = { 903 .queue_xmit = ip_queue_xmit, 904 .send_check = dccp_v4_send_check, 905 .rebuild_header = inet_sk_rebuild_header, 906 .conn_request = dccp_v4_conn_request, 907 .syn_recv_sock = dccp_v4_request_recv_sock, 908 .net_header_len = sizeof(struct iphdr), 909 .setsockopt = ip_setsockopt, 910 .getsockopt = ip_getsockopt, 911 .addr2sockaddr = inet_csk_addr2sockaddr, 912 .sockaddr_len = sizeof(struct sockaddr_in), 913 #ifdef CONFIG_COMPAT 914 .compat_setsockopt = compat_ip_setsockopt, 915 .compat_getsockopt = compat_ip_getsockopt, 916 #endif 917 }; 918 919 static int dccp_v4_init_sock(struct sock *sk) 920 { 921 static __u8 dccp_v4_ctl_sock_initialized; 922 int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized); 923 924 if (err == 0) { 925 if (unlikely(!dccp_v4_ctl_sock_initialized)) 926 dccp_v4_ctl_sock_initialized = 1; 927 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops; 928 } 929 930 return err; 931 } 932 933 static struct timewait_sock_ops dccp_timewait_sock_ops = { 934 .twsk_obj_size = sizeof(struct inet_timewait_sock), 935 }; 936 937 static struct proto dccp_v4_prot = { 938 .name = "DCCP", 939 .owner = THIS_MODULE, 940 .close = dccp_close, 941 .connect = dccp_v4_connect, 942 .disconnect = dccp_disconnect, 943 .ioctl = dccp_ioctl, 944 .init = dccp_v4_init_sock, 945 .setsockopt = dccp_setsockopt, 946 .getsockopt = dccp_getsockopt, 947 .sendmsg = dccp_sendmsg, 948 .recvmsg = dccp_recvmsg, 949 .backlog_rcv = dccp_v4_do_rcv, 950 .hash = inet_hash, 951 .unhash = inet_unhash, 952 .accept = inet_csk_accept, 953 .get_port = inet_csk_get_port, 954 .shutdown = dccp_shutdown, 955 .destroy = dccp_destroy_sock, 956 .orphan_count = &dccp_orphan_count, 957 .max_header = MAX_DCCP_HEADER, 958 .obj_size = sizeof(struct dccp_sock), 959 .slab_flags = SLAB_TYPESAFE_BY_RCU, 960 .rsk_prot = &dccp_request_sock_ops, 961 .twsk_prot = &dccp_timewait_sock_ops, 962 .h.hashinfo = &dccp_hashinfo, 963 #ifdef CONFIG_COMPAT 964 .compat_setsockopt = compat_dccp_setsockopt, 965 .compat_getsockopt = compat_dccp_getsockopt, 966 #endif 967 }; 968 969 static const struct net_protocol dccp_v4_protocol = { 970 .handler = dccp_v4_rcv, 971 .err_handler = dccp_v4_err, 972 .no_policy = 1, 973 .netns_ok = 1, 974 .icmp_strict_tag_validation = 1, 975 }; 976 977 static const struct proto_ops inet_dccp_ops = { 978 .family = PF_INET, 979 .owner = THIS_MODULE, 980 .release = inet_release, 981 .bind = inet_bind, 982 .connect = inet_stream_connect, 983 .socketpair = sock_no_socketpair, 984 .accept = inet_accept, 985 .getname = inet_getname, 986 /* FIXME: work on tcp_poll to rename it to inet_csk_poll */ 987 .poll_mask = dccp_poll_mask, 988 .ioctl = inet_ioctl, 989 /* FIXME: work on inet_listen to rename it to sock_common_listen */ 990 .listen = inet_dccp_listen, 991 .shutdown = inet_shutdown, 992 .setsockopt = sock_common_setsockopt, 993 .getsockopt = sock_common_getsockopt, 994 .sendmsg = inet_sendmsg, 995 .recvmsg = sock_common_recvmsg, 996 .mmap = sock_no_mmap, 997 .sendpage = sock_no_sendpage, 998 #ifdef CONFIG_COMPAT 999 .compat_setsockopt = compat_sock_common_setsockopt, 1000 .compat_getsockopt = compat_sock_common_getsockopt, 1001 #endif 1002 }; 1003 1004 static struct inet_protosw dccp_v4_protosw = { 1005 .type = SOCK_DCCP, 1006 .protocol = IPPROTO_DCCP, 1007 .prot = &dccp_v4_prot, 1008 .ops = &inet_dccp_ops, 1009 .flags = INET_PROTOSW_ICSK, 1010 }; 1011 1012 static int __net_init dccp_v4_init_net(struct net *net) 1013 { 1014 if (dccp_hashinfo.bhash == NULL) 1015 return -ESOCKTNOSUPPORT; 1016 1017 return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET, 1018 SOCK_DCCP, IPPROTO_DCCP, net); 1019 } 1020 1021 static void __net_exit dccp_v4_exit_net(struct net *net) 1022 { 1023 inet_ctl_sock_destroy(net->dccp.v4_ctl_sk); 1024 } 1025 1026 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list) 1027 { 1028 inet_twsk_purge(&dccp_hashinfo, AF_INET); 1029 } 1030 1031 static struct pernet_operations dccp_v4_ops = { 1032 .init = dccp_v4_init_net, 1033 .exit = dccp_v4_exit_net, 1034 .exit_batch = dccp_v4_exit_batch, 1035 }; 1036 1037 static int __init dccp_v4_init(void) 1038 { 1039 int err = proto_register(&dccp_v4_prot, 1); 1040 1041 if (err) 1042 goto out; 1043 1044 inet_register_protosw(&dccp_v4_protosw); 1045 1046 err = register_pernet_subsys(&dccp_v4_ops); 1047 if (err) 1048 goto out_destroy_ctl_sock; 1049 1050 err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP); 1051 if (err) 1052 goto out_proto_unregister; 1053 1054 out: 1055 return err; 1056 out_proto_unregister: 1057 unregister_pernet_subsys(&dccp_v4_ops); 1058 out_destroy_ctl_sock: 1059 inet_unregister_protosw(&dccp_v4_protosw); 1060 proto_unregister(&dccp_v4_prot); 1061 goto out; 1062 } 1063 1064 static void __exit dccp_v4_exit(void) 1065 { 1066 inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP); 1067 unregister_pernet_subsys(&dccp_v4_ops); 1068 inet_unregister_protosw(&dccp_v4_protosw); 1069 proto_unregister(&dccp_v4_prot); 1070 } 1071 1072 module_init(dccp_v4_init); 1073 module_exit(dccp_v4_exit); 1074 1075 /* 1076 * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33) 1077 * values directly, Also cover the case where the protocol is not specified, 1078 * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP 1079 */ 1080 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6); 1081 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6); 1082 MODULE_LICENSE("GPL"); 1083 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>"); 1084 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol"); 1085