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