1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * net/dccp/ipv4.c 4 * 5 * An implementation of the DCCP protocol 6 * Arnaldo Carvalho de Melo <acme@conectiva.com.br> 7 */ 8 9 #include <linux/dccp.h> 10 #include <linux/icmp.h> 11 #include <linux/slab.h> 12 #include <linux/module.h> 13 #include <linux/skbuff.h> 14 #include <linux/random.h> 15 16 #include <net/icmp.h> 17 #include <net/inet_common.h> 18 #include <net/inet_hashtables.h> 19 #include <net/inet_sock.h> 20 #include <net/protocol.h> 21 #include <net/sock.h> 22 #include <net/timewait_sock.h> 23 #include <net/tcp_states.h> 24 #include <net/xfrm.h> 25 #include <net/secure_seq.h> 26 #include <net/netns/generic.h> 27 28 #include "ackvec.h" 29 #include "ccid.h" 30 #include "dccp.h" 31 #include "feat.h" 32 33 struct dccp_v4_pernet { 34 struct sock *v4_ctl_sk; 35 }; 36 37 static unsigned int dccp_v4_pernet_id __read_mostly; 38 39 /* 40 * The per-net v4_ctl_sk socket is used for responding to 41 * the Out-of-the-blue (OOTB) packets. A control sock will be created 42 * for this socket at the initialization time. 43 */ 44 45 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 46 { 47 const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; 48 struct inet_sock *inet = inet_sk(sk); 49 struct dccp_sock *dp = dccp_sk(sk); 50 __be16 orig_sport, orig_dport; 51 __be32 daddr, nexthop; 52 struct flowi4 *fl4; 53 struct rtable *rt; 54 int err; 55 struct ip_options_rcu *inet_opt; 56 57 dp->dccps_role = DCCP_ROLE_CLIENT; 58 59 if (addr_len < sizeof(struct sockaddr_in)) 60 return -EINVAL; 61 62 if (usin->sin_family != AF_INET) 63 return -EAFNOSUPPORT; 64 65 nexthop = daddr = usin->sin_addr.s_addr; 66 67 inet_opt = rcu_dereference_protected(inet->inet_opt, 68 lockdep_sock_is_held(sk)); 69 if (inet_opt != NULL && inet_opt->opt.srr) { 70 if (daddr == 0) 71 return -EINVAL; 72 nexthop = inet_opt->opt.faddr; 73 } 74 75 orig_sport = inet->inet_sport; 76 orig_dport = usin->sin_port; 77 fl4 = &inet->cork.fl.u.ip4; 78 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr, 79 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, 80 IPPROTO_DCCP, 81 orig_sport, orig_dport, sk); 82 if (IS_ERR(rt)) 83 return PTR_ERR(rt); 84 85 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { 86 ip_rt_put(rt); 87 return -ENETUNREACH; 88 } 89 90 if (inet_opt == NULL || !inet_opt->opt.srr) 91 daddr = fl4->daddr; 92 93 if (inet->inet_saddr == 0) 94 inet->inet_saddr = fl4->saddr; 95 sk_rcv_saddr_set(sk, inet->inet_saddr); 96 inet->inet_dport = usin->sin_port; 97 sk_daddr_set(sk, daddr); 98 99 inet_csk(sk)->icsk_ext_hdr_len = 0; 100 if (inet_opt) 101 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen; 102 /* 103 * Socket identity is still unknown (sport may be zero). 104 * However we set state to DCCP_REQUESTING and not releasing socket 105 * lock select source port, enter ourselves into the hash tables and 106 * complete initialization after this. 107 */ 108 dccp_set_state(sk, DCCP_REQUESTING); 109 err = inet_hash_connect(&dccp_death_row, sk); 110 if (err != 0) 111 goto failure; 112 113 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport, 114 inet->inet_sport, inet->inet_dport, sk); 115 if (IS_ERR(rt)) { 116 err = PTR_ERR(rt); 117 rt = NULL; 118 goto failure; 119 } 120 /* OK, now commit destination to socket. */ 121 sk_setup_caps(sk, &rt->dst); 122 123 dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr, 124 inet->inet_daddr, 125 inet->inet_sport, 126 inet->inet_dport); 127 inet->inet_id = prandom_u32(); 128 129 err = dccp_connect(sk); 130 rt = NULL; 131 if (err != 0) 132 goto failure; 133 out: 134 return err; 135 failure: 136 /* 137 * This unhashes the socket and releases the local port, if necessary. 138 */ 139 dccp_set_state(sk, DCCP_CLOSED); 140 ip_rt_put(rt); 141 sk->sk_route_caps = 0; 142 inet->inet_dport = 0; 143 goto out; 144 } 145 EXPORT_SYMBOL_GPL(dccp_v4_connect); 146 147 /* 148 * This routine does path mtu discovery as defined in RFC1191. 149 */ 150 static inline void dccp_do_pmtu_discovery(struct sock *sk, 151 const struct iphdr *iph, 152 u32 mtu) 153 { 154 struct dst_entry *dst; 155 const struct inet_sock *inet = inet_sk(sk); 156 const struct dccp_sock *dp = dccp_sk(sk); 157 158 /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs 159 * send out by Linux are always < 576bytes so they should go through 160 * unfragmented). 161 */ 162 if (sk->sk_state == DCCP_LISTEN) 163 return; 164 165 dst = inet_csk_update_pmtu(sk, mtu); 166 if (!dst) 167 return; 168 169 /* Something is about to be wrong... Remember soft error 170 * for the case, if this connection will not able to recover. 171 */ 172 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) 173 sk->sk_err_soft = EMSGSIZE; 174 175 mtu = dst_mtu(dst); 176 177 if (inet->pmtudisc != IP_PMTUDISC_DONT && 178 ip_sk_accept_pmtu(sk) && 179 inet_csk(sk)->icsk_pmtu_cookie > mtu) { 180 dccp_sync_mss(sk, mtu); 181 182 /* 183 * From RFC 4340, sec. 14.1: 184 * 185 * DCCP-Sync packets are the best choice for upward 186 * probing, since DCCP-Sync probes do not risk application 187 * data loss. 188 */ 189 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC); 190 } /* else let the usual retransmit timer handle it */ 191 } 192 193 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk) 194 { 195 struct dst_entry *dst = __sk_dst_check(sk, 0); 196 197 if (dst) 198 dst->ops->redirect(dst, sk, skb); 199 } 200 201 void dccp_req_err(struct sock *sk, u64 seq) 202 { 203 struct request_sock *req = inet_reqsk(sk); 204 struct net *net = sock_net(sk); 205 206 /* 207 * ICMPs are not backlogged, hence we cannot get an established 208 * socket here. 209 */ 210 if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) { 211 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); 212 } else { 213 /* 214 * Still in RESPOND, just remove it silently. 215 * There is no good way to pass the error to the newly 216 * created socket, and POSIX does not want network 217 * errors returned from accept(). 218 */ 219 inet_csk_reqsk_queue_drop(req->rsk_listener, req); 220 } 221 reqsk_put(req); 222 } 223 EXPORT_SYMBOL(dccp_req_err); 224 225 /* 226 * This routine is called by the ICMP module when it gets some sort of error 227 * condition. If err < 0 then the socket should be closed and the error 228 * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code. 229 * After adjustment header points to the first 8 bytes of the tcp header. We 230 * need to find the appropriate port. 231 * 232 * The locking strategy used here is very "optimistic". When someone else 233 * accesses the socket the ICMP is just dropped and for some paths there is no 234 * check at all. A more general error queue to queue errors for later handling 235 * is probably better. 236 */ 237 static int dccp_v4_err(struct sk_buff *skb, u32 info) 238 { 239 const struct iphdr *iph = (struct iphdr *)skb->data; 240 const u8 offset = iph->ihl << 2; 241 const struct dccp_hdr *dh; 242 struct dccp_sock *dp; 243 struct inet_sock *inet; 244 const int type = icmp_hdr(skb)->type; 245 const int code = icmp_hdr(skb)->code; 246 struct sock *sk; 247 __u64 seq; 248 int err; 249 struct net *net = dev_net(skb->dev); 250 251 /* Only need dccph_dport & dccph_sport which are the first 252 * 4 bytes in dccp header. 253 * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us. 254 */ 255 BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8); 256 BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8); 257 dh = (struct dccp_hdr *)(skb->data + offset); 258 259 sk = __inet_lookup_established(net, &dccp_hashinfo, 260 iph->daddr, dh->dccph_dport, 261 iph->saddr, ntohs(dh->dccph_sport), 262 inet_iif(skb), 0); 263 if (!sk) { 264 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 265 return -ENOENT; 266 } 267 268 if (sk->sk_state == DCCP_TIME_WAIT) { 269 inet_twsk_put(inet_twsk(sk)); 270 return 0; 271 } 272 seq = dccp_hdr_seq(dh); 273 if (sk->sk_state == DCCP_NEW_SYN_RECV) { 274 dccp_req_err(sk, seq); 275 return 0; 276 } 277 278 bh_lock_sock(sk); 279 /* If too many ICMPs get dropped on busy 280 * servers this needs to be solved differently. 281 */ 282 if (sock_owned_by_user(sk)) 283 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS); 284 285 if (sk->sk_state == DCCP_CLOSED) 286 goto out; 287 288 dp = dccp_sk(sk); 289 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) && 290 !between48(seq, dp->dccps_awl, dp->dccps_awh)) { 291 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); 292 goto out; 293 } 294 295 switch (type) { 296 case ICMP_REDIRECT: 297 if (!sock_owned_by_user(sk)) 298 dccp_do_redirect(skb, sk); 299 goto out; 300 case ICMP_SOURCE_QUENCH: 301 /* Just silently ignore these. */ 302 goto out; 303 case ICMP_PARAMETERPROB: 304 err = EPROTO; 305 break; 306 case ICMP_DEST_UNREACH: 307 if (code > NR_ICMP_UNREACH) 308 goto out; 309 310 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ 311 if (!sock_owned_by_user(sk)) 312 dccp_do_pmtu_discovery(sk, iph, info); 313 goto out; 314 } 315 316 err = icmp_err_convert[code].errno; 317 break; 318 case ICMP_TIME_EXCEEDED: 319 err = EHOSTUNREACH; 320 break; 321 default: 322 goto out; 323 } 324 325 switch (sk->sk_state) { 326 case DCCP_REQUESTING: 327 case DCCP_RESPOND: 328 if (!sock_owned_by_user(sk)) { 329 __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS); 330 sk->sk_err = err; 331 332 sk_error_report(sk); 333 334 dccp_done(sk); 335 } else 336 sk->sk_err_soft = err; 337 goto out; 338 } 339 340 /* If we've already connected we will keep trying 341 * until we time out, or the user gives up. 342 * 343 * rfc1122 4.2.3.9 allows to consider as hard errors 344 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, 345 * but it is obsoleted by pmtu discovery). 346 * 347 * Note, that in modern internet, where routing is unreliable 348 * and in each dark corner broken firewalls sit, sending random 349 * errors ordered by their masters even this two messages finally lose 350 * their original sense (even Linux sends invalid PORT_UNREACHs) 351 * 352 * Now we are in compliance with RFCs. 353 * --ANK (980905) 354 */ 355 356 inet = inet_sk(sk); 357 if (!sock_owned_by_user(sk) && inet->recverr) { 358 sk->sk_err = err; 359 sk_error_report(sk); 360 } else /* Only an error on timeout */ 361 sk->sk_err_soft = err; 362 out: 363 bh_unlock_sock(sk); 364 sock_put(sk); 365 return 0; 366 } 367 368 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb, 369 __be32 src, __be32 dst) 370 { 371 return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum); 372 } 373 374 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb) 375 { 376 const struct inet_sock *inet = inet_sk(sk); 377 struct dccp_hdr *dh = dccp_hdr(skb); 378 379 dccp_csum_outgoing(skb); 380 dh->dccph_checksum = dccp_v4_csum_finish(skb, 381 inet->inet_saddr, 382 inet->inet_daddr); 383 } 384 EXPORT_SYMBOL_GPL(dccp_v4_send_check); 385 386 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb) 387 { 388 return secure_dccp_sequence_number(ip_hdr(skb)->daddr, 389 ip_hdr(skb)->saddr, 390 dccp_hdr(skb)->dccph_dport, 391 dccp_hdr(skb)->dccph_sport); 392 } 393 394 /* 395 * The three way handshake has completed - we got a valid ACK or DATAACK - 396 * now create the new socket. 397 * 398 * This is the equivalent of TCP's tcp_v4_syn_recv_sock 399 */ 400 struct sock *dccp_v4_request_recv_sock(const struct sock *sk, 401 struct sk_buff *skb, 402 struct request_sock *req, 403 struct dst_entry *dst, 404 struct request_sock *req_unhash, 405 bool *own_req) 406 { 407 struct inet_request_sock *ireq; 408 struct inet_sock *newinet; 409 struct sock *newsk; 410 411 if (sk_acceptq_is_full(sk)) 412 goto exit_overflow; 413 414 newsk = dccp_create_openreq_child(sk, req, skb); 415 if (newsk == NULL) 416 goto exit_nonewsk; 417 418 newinet = inet_sk(newsk); 419 ireq = inet_rsk(req); 420 sk_daddr_set(newsk, ireq->ir_rmt_addr); 421 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr); 422 newinet->inet_saddr = ireq->ir_loc_addr; 423 RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt)); 424 newinet->mc_index = inet_iif(skb); 425 newinet->mc_ttl = ip_hdr(skb)->ttl; 426 newinet->inet_id = prandom_u32(); 427 428 if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL) 429 goto put_and_exit; 430 431 sk_setup_caps(newsk, dst); 432 433 dccp_sync_mss(newsk, dst_mtu(dst)); 434 435 if (__inet_inherit_port(sk, newsk) < 0) 436 goto put_and_exit; 437 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL); 438 if (*own_req) 439 ireq->ireq_opt = NULL; 440 else 441 newinet->inet_opt = NULL; 442 return newsk; 443 444 exit_overflow: 445 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 446 exit_nonewsk: 447 dst_release(dst); 448 exit: 449 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 450 return NULL; 451 put_and_exit: 452 newinet->inet_opt = NULL; 453 inet_csk_prepare_forced_close(newsk); 454 dccp_done(newsk); 455 goto exit; 456 } 457 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock); 458 459 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk, 460 struct sk_buff *skb) 461 { 462 struct rtable *rt; 463 const struct iphdr *iph = ip_hdr(skb); 464 struct flowi4 fl4 = { 465 .flowi4_oif = inet_iif(skb), 466 .daddr = iph->saddr, 467 .saddr = iph->daddr, 468 .flowi4_tos = RT_CONN_FLAGS(sk), 469 .flowi4_proto = sk->sk_protocol, 470 .fl4_sport = dccp_hdr(skb)->dccph_dport, 471 .fl4_dport = dccp_hdr(skb)->dccph_sport, 472 }; 473 474 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4)); 475 rt = ip_route_output_flow(net, &fl4, sk); 476 if (IS_ERR(rt)) { 477 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 478 return NULL; 479 } 480 481 return &rt->dst; 482 } 483 484 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req) 485 { 486 int err = -1; 487 struct sk_buff *skb; 488 struct dst_entry *dst; 489 struct flowi4 fl4; 490 491 dst = inet_csk_route_req(sk, &fl4, req); 492 if (dst == NULL) 493 goto out; 494 495 skb = dccp_make_response(sk, dst, req); 496 if (skb != NULL) { 497 const struct inet_request_sock *ireq = inet_rsk(req); 498 struct dccp_hdr *dh = dccp_hdr(skb); 499 500 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr, 501 ireq->ir_rmt_addr); 502 rcu_read_lock(); 503 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr, 504 ireq->ir_rmt_addr, 505 rcu_dereference(ireq->ireq_opt), 506 inet_sk(sk)->tos); 507 rcu_read_unlock(); 508 err = net_xmit_eval(err); 509 } 510 511 out: 512 dst_release(dst); 513 return err; 514 } 515 516 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb) 517 { 518 int err; 519 const struct iphdr *rxiph; 520 struct sk_buff *skb; 521 struct dst_entry *dst; 522 struct net *net = dev_net(skb_dst(rxskb)->dev); 523 struct dccp_v4_pernet *pn; 524 struct sock *ctl_sk; 525 526 /* Never send a reset in response to a reset. */ 527 if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET) 528 return; 529 530 if (skb_rtable(rxskb)->rt_type != RTN_LOCAL) 531 return; 532 533 pn = net_generic(net, dccp_v4_pernet_id); 534 ctl_sk = pn->v4_ctl_sk; 535 dst = dccp_v4_route_skb(net, ctl_sk, rxskb); 536 if (dst == NULL) 537 return; 538 539 skb = dccp_ctl_make_reset(ctl_sk, rxskb); 540 if (skb == NULL) 541 goto out; 542 543 rxiph = ip_hdr(rxskb); 544 dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr, 545 rxiph->daddr); 546 skb_dst_set(skb, dst_clone(dst)); 547 548 local_bh_disable(); 549 bh_lock_sock(ctl_sk); 550 err = ip_build_and_send_pkt(skb, ctl_sk, 551 rxiph->daddr, rxiph->saddr, NULL, 552 inet_sk(ctl_sk)->tos); 553 bh_unlock_sock(ctl_sk); 554 555 if (net_xmit_eval(err) == 0) { 556 __DCCP_INC_STATS(DCCP_MIB_OUTSEGS); 557 __DCCP_INC_STATS(DCCP_MIB_OUTRSTS); 558 } 559 local_bh_enable(); 560 out: 561 dst_release(dst); 562 } 563 564 static void dccp_v4_reqsk_destructor(struct request_sock *req) 565 { 566 dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg); 567 kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1)); 568 } 569 570 void dccp_syn_ack_timeout(const struct request_sock *req) 571 { 572 } 573 EXPORT_SYMBOL(dccp_syn_ack_timeout); 574 575 static struct request_sock_ops dccp_request_sock_ops __read_mostly = { 576 .family = PF_INET, 577 .obj_size = sizeof(struct dccp_request_sock), 578 .rtx_syn_ack = dccp_v4_send_response, 579 .send_ack = dccp_reqsk_send_ack, 580 .destructor = dccp_v4_reqsk_destructor, 581 .send_reset = dccp_v4_ctl_send_reset, 582 .syn_ack_timeout = dccp_syn_ack_timeout, 583 }; 584 585 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb) 586 { 587 struct inet_request_sock *ireq; 588 struct request_sock *req; 589 struct dccp_request_sock *dreq; 590 const __be32 service = dccp_hdr_request(skb)->dccph_req_service; 591 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb); 592 593 /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */ 594 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 595 return 0; /* discard, don't send a reset here */ 596 597 if (dccp_bad_service_code(sk, service)) { 598 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE; 599 goto drop; 600 } 601 /* 602 * TW buckets are converted to open requests without 603 * limitations, they conserve resources and peer is 604 * evidently real one. 605 */ 606 dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY; 607 if (inet_csk_reqsk_queue_is_full(sk)) 608 goto drop; 609 610 if (sk_acceptq_is_full(sk)) 611 goto drop; 612 613 req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true); 614 if (req == NULL) 615 goto drop; 616 617 if (dccp_reqsk_init(req, dccp_sk(sk), skb)) 618 goto drop_and_free; 619 620 dreq = dccp_rsk(req); 621 if (dccp_parse_options(sk, dreq, skb)) 622 goto drop_and_free; 623 624 if (security_inet_conn_request(sk, skb, req)) 625 goto drop_and_free; 626 627 ireq = inet_rsk(req); 628 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr); 629 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr); 630 ireq->ir_mark = inet_request_mark(sk, skb); 631 ireq->ireq_family = AF_INET; 632 ireq->ir_iif = sk->sk_bound_dev_if; 633 634 /* 635 * Step 3: Process LISTEN state 636 * 637 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie 638 * 639 * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child(). 640 */ 641 dreq->dreq_isr = dcb->dccpd_seq; 642 dreq->dreq_gsr = dreq->dreq_isr; 643 dreq->dreq_iss = dccp_v4_init_sequence(skb); 644 dreq->dreq_gss = dreq->dreq_iss; 645 dreq->dreq_service = service; 646 647 if (dccp_v4_send_response(sk, req)) 648 goto drop_and_free; 649 650 inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT); 651 reqsk_put(req); 652 return 0; 653 654 drop_and_free: 655 reqsk_free(req); 656 drop: 657 __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS); 658 return -1; 659 } 660 EXPORT_SYMBOL_GPL(dccp_v4_conn_request); 661 662 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) 663 { 664 struct dccp_hdr *dh = dccp_hdr(skb); 665 666 if (sk->sk_state == DCCP_OPEN) { /* Fast path */ 667 if (dccp_rcv_established(sk, skb, dh, skb->len)) 668 goto reset; 669 return 0; 670 } 671 672 /* 673 * Step 3: Process LISTEN state 674 * If P.type == Request or P contains a valid Init Cookie option, 675 * (* Must scan the packet's options to check for Init 676 * Cookies. Only Init Cookies are processed here, 677 * however; other options are processed in Step 8. This 678 * scan need only be performed if the endpoint uses Init 679 * Cookies *) 680 * (* Generate a new socket and switch to that socket *) 681 * Set S := new socket for this port pair 682 * S.state = RESPOND 683 * Choose S.ISS (initial seqno) or set from Init Cookies 684 * Initialize S.GAR := S.ISS 685 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies 686 * Continue with S.state == RESPOND 687 * (* A Response packet will be generated in Step 11 *) 688 * Otherwise, 689 * Generate Reset(No Connection) unless P.type == Reset 690 * Drop packet and return 691 * 692 * NOTE: the check for the packet types is done in 693 * dccp_rcv_state_process 694 */ 695 696 if (dccp_rcv_state_process(sk, skb, dh, skb->len)) 697 goto reset; 698 return 0; 699 700 reset: 701 dccp_v4_ctl_send_reset(sk, skb); 702 kfree_skb(skb); 703 return 0; 704 } 705 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv); 706 707 /** 708 * dccp_invalid_packet - check for malformed packets 709 * @skb: Packet to validate 710 * 711 * Implements RFC 4340, 8.5: Step 1: Check header basics 712 * Packets that fail these checks are ignored and do not receive Resets. 713 */ 714 int dccp_invalid_packet(struct sk_buff *skb) 715 { 716 const struct dccp_hdr *dh; 717 unsigned int cscov; 718 u8 dccph_doff; 719 720 if (skb->pkt_type != PACKET_HOST) 721 return 1; 722 723 /* If the packet is shorter than 12 bytes, drop packet and return */ 724 if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) { 725 DCCP_WARN("pskb_may_pull failed\n"); 726 return 1; 727 } 728 729 dh = dccp_hdr(skb); 730 731 /* If P.type is not understood, drop packet and return */ 732 if (dh->dccph_type >= DCCP_PKT_INVALID) { 733 DCCP_WARN("invalid packet type\n"); 734 return 1; 735 } 736 737 /* 738 * If P.Data Offset is too small for packet type, drop packet and return 739 */ 740 dccph_doff = dh->dccph_doff; 741 if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) { 742 DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff); 743 return 1; 744 } 745 /* 746 * If P.Data Offset is too large for packet, drop packet and return 747 */ 748 if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) { 749 DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff); 750 return 1; 751 } 752 dh = dccp_hdr(skb); 753 /* 754 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet 755 * has short sequence numbers), drop packet and return 756 */ 757 if ((dh->dccph_type < DCCP_PKT_DATA || 758 dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0) { 759 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n", 760 dccp_packet_name(dh->dccph_type)); 761 return 1; 762 } 763 764 /* 765 * If P.CsCov is too large for the packet size, drop packet and return. 766 * This must come _before_ checksumming (not as RFC 4340 suggests). 767 */ 768 cscov = dccp_csum_coverage(skb); 769 if (cscov > skb->len) { 770 DCCP_WARN("P.CsCov %u exceeds packet length %d\n", 771 dh->dccph_cscov, skb->len); 772 return 1; 773 } 774 775 /* If header checksum is incorrect, drop packet and return. 776 * (This step is completed in the AF-dependent functions.) */ 777 skb->csum = skb_checksum(skb, 0, cscov, 0); 778 779 return 0; 780 } 781 EXPORT_SYMBOL_GPL(dccp_invalid_packet); 782 783 /* this is called when real data arrives */ 784 static int dccp_v4_rcv(struct sk_buff *skb) 785 { 786 const struct dccp_hdr *dh; 787 const struct iphdr *iph; 788 bool refcounted; 789 struct sock *sk; 790 int min_cov; 791 792 /* Step 1: Check header basics */ 793 794 if (dccp_invalid_packet(skb)) 795 goto discard_it; 796 797 iph = ip_hdr(skb); 798 /* Step 1: If header checksum is incorrect, drop packet and return */ 799 if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) { 800 DCCP_WARN("dropped packet with invalid checksum\n"); 801 goto discard_it; 802 } 803 804 dh = dccp_hdr(skb); 805 806 DCCP_SKB_CB(skb)->dccpd_seq = dccp_hdr_seq(dh); 807 DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type; 808 809 dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu", 810 dccp_packet_name(dh->dccph_type), 811 &iph->saddr, ntohs(dh->dccph_sport), 812 &iph->daddr, ntohs(dh->dccph_dport), 813 (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq); 814 815 if (dccp_packet_without_ack(skb)) { 816 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ; 817 dccp_pr_debug_cat("\n"); 818 } else { 819 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb); 820 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long) 821 DCCP_SKB_CB(skb)->dccpd_ack_seq); 822 } 823 824 lookup: 825 sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh), 826 dh->dccph_sport, dh->dccph_dport, 0, &refcounted); 827 if (!sk) { 828 dccp_pr_debug("failed to look up flow ID in table and " 829 "get corresponding socket\n"); 830 goto no_dccp_socket; 831 } 832 833 /* 834 * Step 2: 835 * ... or S.state == TIMEWAIT, 836 * Generate Reset(No Connection) unless P.type == Reset 837 * Drop packet and return 838 */ 839 if (sk->sk_state == DCCP_TIME_WAIT) { 840 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n"); 841 inet_twsk_put(inet_twsk(sk)); 842 goto no_dccp_socket; 843 } 844 845 if (sk->sk_state == DCCP_NEW_SYN_RECV) { 846 struct request_sock *req = inet_reqsk(sk); 847 struct sock *nsk; 848 849 sk = req->rsk_listener; 850 if (unlikely(sk->sk_state != DCCP_LISTEN)) { 851 inet_csk_reqsk_queue_drop_and_put(sk, req); 852 goto lookup; 853 } 854 sock_hold(sk); 855 refcounted = true; 856 nsk = dccp_check_req(sk, skb, req); 857 if (!nsk) { 858 reqsk_put(req); 859 goto discard_and_relse; 860 } 861 if (nsk == sk) { 862 reqsk_put(req); 863 } else if (dccp_child_process(sk, nsk, skb)) { 864 dccp_v4_ctl_send_reset(sk, skb); 865 goto discard_and_relse; 866 } else { 867 sock_put(sk); 868 return 0; 869 } 870 } 871 /* 872 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage 873 * o if MinCsCov = 0, only packets with CsCov = 0 are accepted 874 * o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov 875 */ 876 min_cov = dccp_sk(sk)->dccps_pcrlen; 877 if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov)) { 878 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n", 879 dh->dccph_cscov, min_cov); 880 /* FIXME: "Such packets SHOULD be reported using Data Dropped 881 * options (Section 11.7) with Drop Code 0, Protocol 882 * Constraints." */ 883 goto discard_and_relse; 884 } 885 886 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 887 goto discard_and_relse; 888 nf_reset_ct(skb); 889 890 return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted); 891 892 no_dccp_socket: 893 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 894 goto discard_it; 895 /* 896 * Step 2: 897 * If no socket ... 898 * Generate Reset(No Connection) unless P.type == Reset 899 * Drop packet and return 900 */ 901 if (dh->dccph_type != DCCP_PKT_RESET) { 902 DCCP_SKB_CB(skb)->dccpd_reset_code = 903 DCCP_RESET_CODE_NO_CONNECTION; 904 dccp_v4_ctl_send_reset(sk, skb); 905 } 906 907 discard_it: 908 kfree_skb(skb); 909 return 0; 910 911 discard_and_relse: 912 if (refcounted) 913 sock_put(sk); 914 goto discard_it; 915 } 916 917 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = { 918 .queue_xmit = ip_queue_xmit, 919 .send_check = dccp_v4_send_check, 920 .rebuild_header = inet_sk_rebuild_header, 921 .conn_request = dccp_v4_conn_request, 922 .syn_recv_sock = dccp_v4_request_recv_sock, 923 .net_header_len = sizeof(struct iphdr), 924 .setsockopt = ip_setsockopt, 925 .getsockopt = ip_getsockopt, 926 .addr2sockaddr = inet_csk_addr2sockaddr, 927 .sockaddr_len = sizeof(struct sockaddr_in), 928 }; 929 930 static int dccp_v4_init_sock(struct sock *sk) 931 { 932 static __u8 dccp_v4_ctl_sock_initialized; 933 int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized); 934 935 if (err == 0) { 936 if (unlikely(!dccp_v4_ctl_sock_initialized)) 937 dccp_v4_ctl_sock_initialized = 1; 938 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops; 939 } 940 941 return err; 942 } 943 944 static struct timewait_sock_ops dccp_timewait_sock_ops = { 945 .twsk_obj_size = sizeof(struct inet_timewait_sock), 946 }; 947 948 static struct proto dccp_v4_prot = { 949 .name = "DCCP", 950 .owner = THIS_MODULE, 951 .close = dccp_close, 952 .connect = dccp_v4_connect, 953 .disconnect = dccp_disconnect, 954 .ioctl = dccp_ioctl, 955 .init = dccp_v4_init_sock, 956 .setsockopt = dccp_setsockopt, 957 .getsockopt = dccp_getsockopt, 958 .sendmsg = dccp_sendmsg, 959 .recvmsg = dccp_recvmsg, 960 .backlog_rcv = dccp_v4_do_rcv, 961 .hash = inet_hash, 962 .unhash = inet_unhash, 963 .accept = inet_csk_accept, 964 .get_port = inet_csk_get_port, 965 .shutdown = dccp_shutdown, 966 .destroy = dccp_destroy_sock, 967 .orphan_count = &dccp_orphan_count, 968 .max_header = MAX_DCCP_HEADER, 969 .obj_size = sizeof(struct dccp_sock), 970 .slab_flags = SLAB_TYPESAFE_BY_RCU, 971 .rsk_prot = &dccp_request_sock_ops, 972 .twsk_prot = &dccp_timewait_sock_ops, 973 .h.hashinfo = &dccp_hashinfo, 974 }; 975 976 static const struct net_protocol dccp_v4_protocol = { 977 .handler = dccp_v4_rcv, 978 .err_handler = dccp_v4_err, 979 .no_policy = 1, 980 .icmp_strict_tag_validation = 1, 981 }; 982 983 static const struct proto_ops inet_dccp_ops = { 984 .family = PF_INET, 985 .owner = THIS_MODULE, 986 .release = inet_release, 987 .bind = inet_bind, 988 .connect = inet_stream_connect, 989 .socketpair = sock_no_socketpair, 990 .accept = inet_accept, 991 .getname = inet_getname, 992 /* FIXME: work on tcp_poll to rename it to inet_csk_poll */ 993 .poll = dccp_poll, 994 .ioctl = inet_ioctl, 995 .gettstamp = sock_gettstamp, 996 /* FIXME: work on inet_listen to rename it to sock_common_listen */ 997 .listen = inet_dccp_listen, 998 .shutdown = inet_shutdown, 999 .setsockopt = sock_common_setsockopt, 1000 .getsockopt = sock_common_getsockopt, 1001 .sendmsg = inet_sendmsg, 1002 .recvmsg = sock_common_recvmsg, 1003 .mmap = sock_no_mmap, 1004 .sendpage = sock_no_sendpage, 1005 }; 1006 1007 static struct inet_protosw dccp_v4_protosw = { 1008 .type = SOCK_DCCP, 1009 .protocol = IPPROTO_DCCP, 1010 .prot = &dccp_v4_prot, 1011 .ops = &inet_dccp_ops, 1012 .flags = INET_PROTOSW_ICSK, 1013 }; 1014 1015 static int __net_init dccp_v4_init_net(struct net *net) 1016 { 1017 struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id); 1018 1019 if (dccp_hashinfo.bhash == NULL) 1020 return -ESOCKTNOSUPPORT; 1021 1022 return inet_ctl_sock_create(&pn->v4_ctl_sk, PF_INET, 1023 SOCK_DCCP, IPPROTO_DCCP, net); 1024 } 1025 1026 static void __net_exit dccp_v4_exit_net(struct net *net) 1027 { 1028 struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id); 1029 1030 inet_ctl_sock_destroy(pn->v4_ctl_sk); 1031 } 1032 1033 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list) 1034 { 1035 inet_twsk_purge(&dccp_hashinfo, AF_INET); 1036 } 1037 1038 static struct pernet_operations dccp_v4_ops = { 1039 .init = dccp_v4_init_net, 1040 .exit = dccp_v4_exit_net, 1041 .exit_batch = dccp_v4_exit_batch, 1042 .id = &dccp_v4_pernet_id, 1043 .size = sizeof(struct dccp_v4_pernet), 1044 }; 1045 1046 static int __init dccp_v4_init(void) 1047 { 1048 int err = proto_register(&dccp_v4_prot, 1); 1049 1050 if (err) 1051 goto out; 1052 1053 inet_register_protosw(&dccp_v4_protosw); 1054 1055 err = register_pernet_subsys(&dccp_v4_ops); 1056 if (err) 1057 goto out_destroy_ctl_sock; 1058 1059 err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP); 1060 if (err) 1061 goto out_proto_unregister; 1062 1063 out: 1064 return err; 1065 out_proto_unregister: 1066 unregister_pernet_subsys(&dccp_v4_ops); 1067 out_destroy_ctl_sock: 1068 inet_unregister_protosw(&dccp_v4_protosw); 1069 proto_unregister(&dccp_v4_prot); 1070 goto out; 1071 } 1072 1073 static void __exit dccp_v4_exit(void) 1074 { 1075 inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP); 1076 unregister_pernet_subsys(&dccp_v4_ops); 1077 inet_unregister_protosw(&dccp_v4_protosw); 1078 proto_unregister(&dccp_v4_prot); 1079 } 1080 1081 module_init(dccp_v4_init); 1082 module_exit(dccp_v4_exit); 1083 1084 /* 1085 * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33) 1086 * values directly, Also cover the case where the protocol is not specified, 1087 * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP 1088 */ 1089 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6); 1090 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6); 1091 MODULE_LICENSE("GPL"); 1092 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>"); 1093 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol"); 1094