1 /* 2 * L2TPv3 IP encapsulation support 3 * 4 * Copyright (c) 2008,2009,2010 Katalix Systems Ltd 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 */ 11 12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 13 14 #include <linux/icmp.h> 15 #include <linux/module.h> 16 #include <linux/skbuff.h> 17 #include <linux/random.h> 18 #include <linux/socket.h> 19 #include <linux/l2tp.h> 20 #include <linux/in.h> 21 #include <net/sock.h> 22 #include <net/ip.h> 23 #include <net/icmp.h> 24 #include <net/udp.h> 25 #include <net/inet_common.h> 26 #include <net/inet_hashtables.h> 27 #include <net/tcp_states.h> 28 #include <net/protocol.h> 29 #include <net/xfrm.h> 30 31 #include "l2tp_core.h" 32 33 struct l2tp_ip_sock { 34 /* inet_sock has to be the first member of l2tp_ip_sock */ 35 struct inet_sock inet; 36 37 u32 conn_id; 38 u32 peer_conn_id; 39 }; 40 41 static DEFINE_RWLOCK(l2tp_ip_lock); 42 static struct hlist_head l2tp_ip_table; 43 static struct hlist_head l2tp_ip_bind_table; 44 45 static inline struct l2tp_ip_sock *l2tp_ip_sk(const struct sock *sk) 46 { 47 return (struct l2tp_ip_sock *)sk; 48 } 49 50 static struct sock *__l2tp_ip_bind_lookup(struct net *net, __be32 laddr, int dif, u32 tunnel_id) 51 { 52 struct hlist_node *node; 53 struct sock *sk; 54 55 sk_for_each_bound(sk, node, &l2tp_ip_bind_table) { 56 struct inet_sock *inet = inet_sk(sk); 57 struct l2tp_ip_sock *l2tp = l2tp_ip_sk(sk); 58 59 if (l2tp == NULL) 60 continue; 61 62 if ((l2tp->conn_id == tunnel_id) && 63 net_eq(sock_net(sk), net) && 64 !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) && 65 !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 66 goto found; 67 } 68 69 sk = NULL; 70 found: 71 return sk; 72 } 73 74 static inline struct sock *l2tp_ip_bind_lookup(struct net *net, __be32 laddr, int dif, u32 tunnel_id) 75 { 76 struct sock *sk = __l2tp_ip_bind_lookup(net, laddr, dif, tunnel_id); 77 if (sk) 78 sock_hold(sk); 79 80 return sk; 81 } 82 83 /* When processing receive frames, there are two cases to 84 * consider. Data frames consist of a non-zero session-id and an 85 * optional cookie. Control frames consist of a regular L2TP header 86 * preceded by 32-bits of zeros. 87 * 88 * L2TPv3 Session Header Over IP 89 * 90 * 0 1 2 3 91 * 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 92 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 93 * | Session ID | 94 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 95 * | Cookie (optional, maximum 64 bits)... 96 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 97 * | 98 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 99 * 100 * L2TPv3 Control Message Header Over IP 101 * 102 * 0 1 2 3 103 * 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 104 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 105 * | (32 bits of zeros) | 106 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 107 * |T|L|x|x|S|x|x|x|x|x|x|x| Ver | Length | 108 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 109 * | Control Connection ID | 110 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 111 * | Ns | Nr | 112 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 113 * 114 * All control frames are passed to userspace. 115 */ 116 static int l2tp_ip_recv(struct sk_buff *skb) 117 { 118 struct net *net = dev_net(skb->dev); 119 struct sock *sk; 120 u32 session_id; 121 u32 tunnel_id; 122 unsigned char *ptr, *optr; 123 struct l2tp_session *session; 124 struct l2tp_tunnel *tunnel = NULL; 125 int length; 126 127 /* Point to L2TP header */ 128 optr = ptr = skb->data; 129 130 if (!pskb_may_pull(skb, 4)) 131 goto discard; 132 133 session_id = ntohl(*((__be32 *) ptr)); 134 ptr += 4; 135 136 /* RFC3931: L2TP/IP packets have the first 4 bytes containing 137 * the session_id. If it is 0, the packet is a L2TP control 138 * frame and the session_id value can be discarded. 139 */ 140 if (session_id == 0) { 141 __skb_pull(skb, 4); 142 goto pass_up; 143 } 144 145 /* Ok, this is a data packet. Lookup the session. */ 146 session = l2tp_session_find(net, NULL, session_id); 147 if (session == NULL) 148 goto discard; 149 150 tunnel = session->tunnel; 151 if (tunnel == NULL) 152 goto discard; 153 154 /* Trace packet contents, if enabled */ 155 if (tunnel->debug & L2TP_MSG_DATA) { 156 length = min(32u, skb->len); 157 if (!pskb_may_pull(skb, length)) 158 goto discard; 159 160 pr_debug("%s: ip recv\n", tunnel->name); 161 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, ptr, length); 162 } 163 164 l2tp_recv_common(session, skb, ptr, optr, 0, skb->len, tunnel->recv_payload_hook); 165 166 return 0; 167 168 pass_up: 169 /* Get the tunnel_id from the L2TP header */ 170 if (!pskb_may_pull(skb, 12)) 171 goto discard; 172 173 if ((skb->data[0] & 0xc0) != 0xc0) 174 goto discard; 175 176 tunnel_id = ntohl(*(__be32 *) &skb->data[4]); 177 tunnel = l2tp_tunnel_find(net, tunnel_id); 178 if (tunnel != NULL) 179 sk = tunnel->sock; 180 else { 181 struct iphdr *iph = (struct iphdr *) skb_network_header(skb); 182 183 read_lock_bh(&l2tp_ip_lock); 184 sk = __l2tp_ip_bind_lookup(net, iph->daddr, 0, tunnel_id); 185 read_unlock_bh(&l2tp_ip_lock); 186 } 187 188 if (sk == NULL) 189 goto discard; 190 191 sock_hold(sk); 192 193 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 194 goto discard_put; 195 196 nf_reset(skb); 197 198 return sk_receive_skb(sk, skb, 1); 199 200 discard_put: 201 sock_put(sk); 202 203 discard: 204 kfree_skb(skb); 205 return 0; 206 } 207 208 static int l2tp_ip_open(struct sock *sk) 209 { 210 /* Prevent autobind. We don't have ports. */ 211 inet_sk(sk)->inet_num = IPPROTO_L2TP; 212 213 write_lock_bh(&l2tp_ip_lock); 214 sk_add_node(sk, &l2tp_ip_table); 215 write_unlock_bh(&l2tp_ip_lock); 216 217 return 0; 218 } 219 220 static void l2tp_ip_close(struct sock *sk, long timeout) 221 { 222 write_lock_bh(&l2tp_ip_lock); 223 hlist_del_init(&sk->sk_bind_node); 224 sk_del_node_init(sk); 225 write_unlock_bh(&l2tp_ip_lock); 226 sk_common_release(sk); 227 } 228 229 static void l2tp_ip_destroy_sock(struct sock *sk) 230 { 231 struct sk_buff *skb; 232 233 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) 234 kfree_skb(skb); 235 236 sk_refcnt_debug_dec(sk); 237 } 238 239 static int l2tp_ip_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len) 240 { 241 struct inet_sock *inet = inet_sk(sk); 242 struct sockaddr_l2tpip *addr = (struct sockaddr_l2tpip *) uaddr; 243 struct net *net = sock_net(sk); 244 int ret; 245 int chk_addr_ret; 246 247 if (!sock_flag(sk, SOCK_ZAPPED)) 248 return -EINVAL; 249 if (addr_len < sizeof(struct sockaddr_l2tpip)) 250 return -EINVAL; 251 if (addr->l2tp_family != AF_INET) 252 return -EINVAL; 253 254 ret = -EADDRINUSE; 255 read_lock_bh(&l2tp_ip_lock); 256 if (__l2tp_ip_bind_lookup(net, addr->l2tp_addr.s_addr, 257 sk->sk_bound_dev_if, addr->l2tp_conn_id)) 258 goto out_in_use; 259 260 read_unlock_bh(&l2tp_ip_lock); 261 262 lock_sock(sk); 263 if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_l2tpip)) 264 goto out; 265 266 chk_addr_ret = inet_addr_type(net, addr->l2tp_addr.s_addr); 267 ret = -EADDRNOTAVAIL; 268 if (addr->l2tp_addr.s_addr && chk_addr_ret != RTN_LOCAL && 269 chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST) 270 goto out; 271 272 if (addr->l2tp_addr.s_addr) 273 inet->inet_rcv_saddr = inet->inet_saddr = addr->l2tp_addr.s_addr; 274 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 275 inet->inet_saddr = 0; /* Use device */ 276 sk_dst_reset(sk); 277 278 l2tp_ip_sk(sk)->conn_id = addr->l2tp_conn_id; 279 280 write_lock_bh(&l2tp_ip_lock); 281 sk_add_bind_node(sk, &l2tp_ip_bind_table); 282 sk_del_node_init(sk); 283 write_unlock_bh(&l2tp_ip_lock); 284 ret = 0; 285 sock_reset_flag(sk, SOCK_ZAPPED); 286 287 out: 288 release_sock(sk); 289 290 return ret; 291 292 out_in_use: 293 read_unlock_bh(&l2tp_ip_lock); 294 295 return ret; 296 } 297 298 static int l2tp_ip_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 299 { 300 struct sockaddr_l2tpip *lsa = (struct sockaddr_l2tpip *) uaddr; 301 int rc; 302 303 if (sock_flag(sk, SOCK_ZAPPED)) /* Must bind first - autobinding does not work */ 304 return -EINVAL; 305 306 if (addr_len < sizeof(*lsa)) 307 return -EINVAL; 308 309 if (ipv4_is_multicast(lsa->l2tp_addr.s_addr)) 310 return -EINVAL; 311 312 rc = ip4_datagram_connect(sk, uaddr, addr_len); 313 if (rc < 0) 314 return rc; 315 316 lock_sock(sk); 317 318 l2tp_ip_sk(sk)->peer_conn_id = lsa->l2tp_conn_id; 319 320 write_lock_bh(&l2tp_ip_lock); 321 hlist_del_init(&sk->sk_bind_node); 322 sk_add_bind_node(sk, &l2tp_ip_bind_table); 323 write_unlock_bh(&l2tp_ip_lock); 324 325 release_sock(sk); 326 return rc; 327 } 328 329 static int l2tp_ip_disconnect(struct sock *sk, int flags) 330 { 331 if (sock_flag(sk, SOCK_ZAPPED)) 332 return 0; 333 334 return udp_disconnect(sk, flags); 335 } 336 337 static int l2tp_ip_getname(struct socket *sock, struct sockaddr *uaddr, 338 int *uaddr_len, int peer) 339 { 340 struct sock *sk = sock->sk; 341 struct inet_sock *inet = inet_sk(sk); 342 struct l2tp_ip_sock *lsk = l2tp_ip_sk(sk); 343 struct sockaddr_l2tpip *lsa = (struct sockaddr_l2tpip *)uaddr; 344 345 memset(lsa, 0, sizeof(*lsa)); 346 lsa->l2tp_family = AF_INET; 347 if (peer) { 348 if (!inet->inet_dport) 349 return -ENOTCONN; 350 lsa->l2tp_conn_id = lsk->peer_conn_id; 351 lsa->l2tp_addr.s_addr = inet->inet_daddr; 352 } else { 353 __be32 addr = inet->inet_rcv_saddr; 354 if (!addr) 355 addr = inet->inet_saddr; 356 lsa->l2tp_conn_id = lsk->conn_id; 357 lsa->l2tp_addr.s_addr = addr; 358 } 359 *uaddr_len = sizeof(*lsa); 360 return 0; 361 } 362 363 static int l2tp_ip_backlog_recv(struct sock *sk, struct sk_buff *skb) 364 { 365 int rc; 366 367 /* Charge it to the socket, dropping if the queue is full. */ 368 rc = sock_queue_rcv_skb(sk, skb); 369 if (rc < 0) 370 goto drop; 371 372 return 0; 373 374 drop: 375 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_INDISCARDS); 376 kfree_skb(skb); 377 return -1; 378 } 379 380 /* Userspace will call sendmsg() on the tunnel socket to send L2TP 381 * control frames. 382 */ 383 static int l2tp_ip_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, size_t len) 384 { 385 struct sk_buff *skb; 386 int rc; 387 struct inet_sock *inet = inet_sk(sk); 388 struct rtable *rt = NULL; 389 struct flowi4 *fl4; 390 int connected = 0; 391 __be32 daddr; 392 393 lock_sock(sk); 394 395 rc = -ENOTCONN; 396 if (sock_flag(sk, SOCK_DEAD)) 397 goto out; 398 399 /* Get and verify the address. */ 400 if (msg->msg_name) { 401 struct sockaddr_l2tpip *lip = (struct sockaddr_l2tpip *) msg->msg_name; 402 rc = -EINVAL; 403 if (msg->msg_namelen < sizeof(*lip)) 404 goto out; 405 406 if (lip->l2tp_family != AF_INET) { 407 rc = -EAFNOSUPPORT; 408 if (lip->l2tp_family != AF_UNSPEC) 409 goto out; 410 } 411 412 daddr = lip->l2tp_addr.s_addr; 413 } else { 414 rc = -EDESTADDRREQ; 415 if (sk->sk_state != TCP_ESTABLISHED) 416 goto out; 417 418 daddr = inet->inet_daddr; 419 connected = 1; 420 } 421 422 /* Allocate a socket buffer */ 423 rc = -ENOMEM; 424 skb = sock_wmalloc(sk, 2 + NET_SKB_PAD + sizeof(struct iphdr) + 425 4 + len, 0, GFP_KERNEL); 426 if (!skb) 427 goto error; 428 429 /* Reserve space for headers, putting IP header on 4-byte boundary. */ 430 skb_reserve(skb, 2 + NET_SKB_PAD); 431 skb_reset_network_header(skb); 432 skb_reserve(skb, sizeof(struct iphdr)); 433 skb_reset_transport_header(skb); 434 435 /* Insert 0 session_id */ 436 *((__be32 *) skb_put(skb, 4)) = 0; 437 438 /* Copy user data into skb */ 439 rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len); 440 if (rc < 0) { 441 kfree_skb(skb); 442 goto error; 443 } 444 445 fl4 = &inet->cork.fl.u.ip4; 446 if (connected) 447 rt = (struct rtable *) __sk_dst_check(sk, 0); 448 449 rcu_read_lock(); 450 if (rt == NULL) { 451 const struct ip_options_rcu *inet_opt; 452 453 inet_opt = rcu_dereference(inet->inet_opt); 454 455 /* Use correct destination address if we have options. */ 456 if (inet_opt && inet_opt->opt.srr) 457 daddr = inet_opt->opt.faddr; 458 459 /* If this fails, retransmit mechanism of transport layer will 460 * keep trying until route appears or the connection times 461 * itself out. 462 */ 463 rt = ip_route_output_ports(sock_net(sk), fl4, sk, 464 daddr, inet->inet_saddr, 465 inet->inet_dport, inet->inet_sport, 466 sk->sk_protocol, RT_CONN_FLAGS(sk), 467 sk->sk_bound_dev_if); 468 if (IS_ERR(rt)) 469 goto no_route; 470 if (connected) { 471 sk_setup_caps(sk, &rt->dst); 472 } else { 473 skb_dst_set(skb, &rt->dst); 474 goto xmit; 475 } 476 } 477 478 /* We dont need to clone dst here, it is guaranteed to not disappear. 479 * __dev_xmit_skb() might force a refcount if needed. 480 */ 481 skb_dst_set_noref(skb, &rt->dst); 482 483 xmit: 484 /* Queue the packet to IP for output */ 485 rc = ip_queue_xmit(skb, &inet->cork.fl); 486 rcu_read_unlock(); 487 488 error: 489 if (rc >= 0) 490 rc = len; 491 492 out: 493 release_sock(sk); 494 return rc; 495 496 no_route: 497 rcu_read_unlock(); 498 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); 499 kfree_skb(skb); 500 rc = -EHOSTUNREACH; 501 goto out; 502 } 503 504 static int l2tp_ip_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 505 size_t len, int noblock, int flags, int *addr_len) 506 { 507 struct inet_sock *inet = inet_sk(sk); 508 size_t copied = 0; 509 int err = -EOPNOTSUPP; 510 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 511 struct sk_buff *skb; 512 513 if (flags & MSG_OOB) 514 goto out; 515 516 if (addr_len) 517 *addr_len = sizeof(*sin); 518 519 skb = skb_recv_datagram(sk, flags, noblock, &err); 520 if (!skb) 521 goto out; 522 523 copied = skb->len; 524 if (len < copied) { 525 msg->msg_flags |= MSG_TRUNC; 526 copied = len; 527 } 528 529 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 530 if (err) 531 goto done; 532 533 sock_recv_timestamp(msg, sk, skb); 534 535 /* Copy the address. */ 536 if (sin) { 537 sin->sin_family = AF_INET; 538 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 539 sin->sin_port = 0; 540 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero)); 541 } 542 if (inet->cmsg_flags) 543 ip_cmsg_recv(msg, skb); 544 if (flags & MSG_TRUNC) 545 copied = skb->len; 546 done: 547 skb_free_datagram(sk, skb); 548 out: 549 return err ? err : copied; 550 } 551 552 static struct proto l2tp_ip_prot = { 553 .name = "L2TP/IP", 554 .owner = THIS_MODULE, 555 .init = l2tp_ip_open, 556 .close = l2tp_ip_close, 557 .bind = l2tp_ip_bind, 558 .connect = l2tp_ip_connect, 559 .disconnect = l2tp_ip_disconnect, 560 .ioctl = udp_ioctl, 561 .destroy = l2tp_ip_destroy_sock, 562 .setsockopt = ip_setsockopt, 563 .getsockopt = ip_getsockopt, 564 .sendmsg = l2tp_ip_sendmsg, 565 .recvmsg = l2tp_ip_recvmsg, 566 .backlog_rcv = l2tp_ip_backlog_recv, 567 .hash = inet_hash, 568 .unhash = inet_unhash, 569 .obj_size = sizeof(struct l2tp_ip_sock), 570 #ifdef CONFIG_COMPAT 571 .compat_setsockopt = compat_ip_setsockopt, 572 .compat_getsockopt = compat_ip_getsockopt, 573 #endif 574 }; 575 576 static const struct proto_ops l2tp_ip_ops = { 577 .family = PF_INET, 578 .owner = THIS_MODULE, 579 .release = inet_release, 580 .bind = inet_bind, 581 .connect = inet_dgram_connect, 582 .socketpair = sock_no_socketpair, 583 .accept = sock_no_accept, 584 .getname = l2tp_ip_getname, 585 .poll = datagram_poll, 586 .ioctl = inet_ioctl, 587 .listen = sock_no_listen, 588 .shutdown = inet_shutdown, 589 .setsockopt = sock_common_setsockopt, 590 .getsockopt = sock_common_getsockopt, 591 .sendmsg = inet_sendmsg, 592 .recvmsg = sock_common_recvmsg, 593 .mmap = sock_no_mmap, 594 .sendpage = sock_no_sendpage, 595 #ifdef CONFIG_COMPAT 596 .compat_setsockopt = compat_sock_common_setsockopt, 597 .compat_getsockopt = compat_sock_common_getsockopt, 598 #endif 599 }; 600 601 static struct inet_protosw l2tp_ip_protosw = { 602 .type = SOCK_DGRAM, 603 .protocol = IPPROTO_L2TP, 604 .prot = &l2tp_ip_prot, 605 .ops = &l2tp_ip_ops, 606 .no_check = 0, 607 }; 608 609 static struct net_protocol l2tp_ip_protocol __read_mostly = { 610 .handler = l2tp_ip_recv, 611 .netns_ok = 1, 612 }; 613 614 static int __init l2tp_ip_init(void) 615 { 616 int err; 617 618 pr_info("L2TP IP encapsulation support (L2TPv3)\n"); 619 620 err = proto_register(&l2tp_ip_prot, 1); 621 if (err != 0) 622 goto out; 623 624 err = inet_add_protocol(&l2tp_ip_protocol, IPPROTO_L2TP); 625 if (err) 626 goto out1; 627 628 inet_register_protosw(&l2tp_ip_protosw); 629 return 0; 630 631 out1: 632 proto_unregister(&l2tp_ip_prot); 633 out: 634 return err; 635 } 636 637 static void __exit l2tp_ip_exit(void) 638 { 639 inet_unregister_protosw(&l2tp_ip_protosw); 640 inet_del_protocol(&l2tp_ip_protocol, IPPROTO_L2TP); 641 proto_unregister(&l2tp_ip_prot); 642 } 643 644 module_init(l2tp_ip_init); 645 module_exit(l2tp_ip_exit); 646 647 MODULE_LICENSE("GPL"); 648 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>"); 649 MODULE_DESCRIPTION("L2TP over IP"); 650 MODULE_VERSION("1.0"); 651 652 /* Use the value of SOCK_DGRAM (2) directory, because __stringify doesn't like 653 * enums 654 */ 655 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 2, IPPROTO_L2TP); 656