1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * RAW sockets for IPv6 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Adapted from linux/net/ipv4/raw.c 10 * 11 * Fixes: 12 * Hideaki YOSHIFUJI : sin6_scope_id support 13 * YOSHIFUJI,H.@USAGI : raw checksum (RFC2292(bis) compliance) 14 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data 15 */ 16 17 #include <linux/errno.h> 18 #include <linux/types.h> 19 #include <linux/socket.h> 20 #include <linux/slab.h> 21 #include <linux/sockios.h> 22 #include <linux/net.h> 23 #include <linux/in6.h> 24 #include <linux/netdevice.h> 25 #include <linux/if_arp.h> 26 #include <linux/icmpv6.h> 27 #include <linux/netfilter.h> 28 #include <linux/netfilter_ipv6.h> 29 #include <linux/skbuff.h> 30 #include <linux/compat.h> 31 #include <linux/uaccess.h> 32 #include <asm/ioctls.h> 33 34 #include <net/net_namespace.h> 35 #include <net/ip.h> 36 #include <net/sock.h> 37 #include <net/snmp.h> 38 39 #include <net/ipv6.h> 40 #include <net/ndisc.h> 41 #include <net/protocol.h> 42 #include <net/ip6_route.h> 43 #include <net/ip6_checksum.h> 44 #include <net/addrconf.h> 45 #include <net/transp_v6.h> 46 #include <net/udp.h> 47 #include <net/inet_common.h> 48 #include <net/tcp_states.h> 49 #if IS_ENABLED(CONFIG_IPV6_MIP6) 50 #include <net/mip6.h> 51 #endif 52 #include <linux/mroute6.h> 53 54 #include <net/raw.h> 55 #include <net/rawv6.h> 56 #include <net/xfrm.h> 57 58 #include <linux/proc_fs.h> 59 #include <linux/seq_file.h> 60 #include <linux/export.h> 61 62 #define ICMPV6_HDRLEN 4 /* ICMPv6 header, RFC 4443 Section 2.1 */ 63 64 struct raw_hashinfo raw_v6_hashinfo; 65 EXPORT_SYMBOL_GPL(raw_v6_hashinfo); 66 67 bool raw_v6_match(struct net *net, const struct sock *sk, unsigned short num, 68 const struct in6_addr *loc_addr, 69 const struct in6_addr *rmt_addr, int dif, int sdif) 70 { 71 if (inet_sk(sk)->inet_num != num || 72 !net_eq(sock_net(sk), net) || 73 (!ipv6_addr_any(&sk->sk_v6_daddr) && 74 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 75 !raw_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 76 dif, sdif)) 77 return false; 78 79 if (ipv6_addr_any(&sk->sk_v6_rcv_saddr) || 80 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr) || 81 (ipv6_addr_is_multicast(loc_addr) && 82 inet6_mc_check(sk, loc_addr, rmt_addr))) 83 return true; 84 85 return false; 86 } 87 EXPORT_SYMBOL_GPL(raw_v6_match); 88 89 /* 90 * 0 - deliver 91 * 1 - block 92 */ 93 static int icmpv6_filter(const struct sock *sk, struct sk_buff *skb) 94 { 95 const struct icmp6hdr *hdr; 96 const __u32 *data; 97 unsigned int type; 98 99 /* We require only the four bytes of the ICMPv6 header, not any 100 * additional bytes of message body in "struct icmp6hdr". 101 */ 102 if (!pskb_may_pull(skb, ICMPV6_HDRLEN)) 103 return 1; 104 105 hdr = (struct icmp6hdr *)skb->data; 106 type = hdr->icmp6_type; 107 108 data = &raw6_sk(sk)->filter.data[0]; 109 110 return (data[type >> 5] & (1U << (type & 31))) != 0; 111 } 112 113 #if IS_ENABLED(CONFIG_IPV6_MIP6) 114 typedef int mh_filter_t(struct sock *sock, struct sk_buff *skb); 115 116 static mh_filter_t __rcu *mh_filter __read_mostly; 117 118 int rawv6_mh_filter_register(mh_filter_t filter) 119 { 120 rcu_assign_pointer(mh_filter, filter); 121 return 0; 122 } 123 EXPORT_SYMBOL(rawv6_mh_filter_register); 124 125 int rawv6_mh_filter_unregister(mh_filter_t filter) 126 { 127 RCU_INIT_POINTER(mh_filter, NULL); 128 synchronize_rcu(); 129 return 0; 130 } 131 EXPORT_SYMBOL(rawv6_mh_filter_unregister); 132 133 #endif 134 135 /* 136 * demultiplex raw sockets. 137 * (should consider queueing the skb in the sock receive_queue 138 * without calling rawv6.c) 139 * 140 * Caller owns SKB so we must make clones. 141 */ 142 static bool ipv6_raw_deliver(struct sk_buff *skb, int nexthdr) 143 { 144 struct net *net = dev_net(skb->dev); 145 const struct ipv6hdr *ip6h; 146 struct hlist_head *hlist; 147 bool delivered = false; 148 struct sock *sk; 149 __u8 hash; 150 151 ip6h = ipv6_hdr(skb); 152 153 hash = raw_hashfunc(net, nexthdr); 154 hlist = &raw_v6_hashinfo.ht[hash]; 155 rcu_read_lock(); 156 sk_for_each_rcu(sk, hlist) { 157 int filtered; 158 159 if (!raw_v6_match(net, sk, nexthdr, &ip6h->daddr, &ip6h->saddr, 160 inet6_iif(skb), inet6_sdif(skb))) 161 continue; 162 163 if (atomic_read(&sk->sk_rmem_alloc) >= 164 READ_ONCE(sk->sk_rcvbuf)) { 165 sk_drops_inc(sk); 166 continue; 167 } 168 169 delivered = true; 170 switch (nexthdr) { 171 case IPPROTO_ICMPV6: 172 filtered = icmpv6_filter(sk, skb); 173 ip6h = ipv6_hdr(skb); 174 break; 175 176 #if IS_ENABLED(CONFIG_IPV6_MIP6) 177 case IPPROTO_MH: 178 { 179 /* XXX: To validate MH only once for each packet, 180 * this is placed here. It should be after checking 181 * xfrm policy, however it doesn't. The checking xfrm 182 * policy is placed in rawv6_rcv() because it is 183 * required for each socket. 184 */ 185 mh_filter_t *filter; 186 187 filter = rcu_dereference(mh_filter); 188 filtered = filter ? (*filter)(sk, skb) : 0; 189 break; 190 } 191 #endif 192 default: 193 filtered = 0; 194 break; 195 } 196 197 if (filtered < 0) 198 break; 199 if (filtered == 0) { 200 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC); 201 202 /* Not releasing hash table! */ 203 if (clone) 204 rawv6_rcv(sk, clone); 205 } 206 } 207 rcu_read_unlock(); 208 return delivered; 209 } 210 211 bool raw6_local_deliver(struct sk_buff *skb, int nexthdr) 212 { 213 return ipv6_raw_deliver(skb, nexthdr); 214 } 215 216 /* This cleans up af_inet6 a bit. -DaveM */ 217 static int rawv6_bind(struct sock *sk, struct sockaddr_unsized *uaddr, 218 int addr_len) 219 { 220 struct inet_sock *inet = inet_sk(sk); 221 struct ipv6_pinfo *np = inet6_sk(sk); 222 struct sockaddr_in6 *addr = (struct sockaddr_in6 *) uaddr; 223 __be32 v4addr = 0; 224 int addr_type; 225 int err; 226 227 if (addr_len < SIN6_LEN_RFC2133) 228 return -EINVAL; 229 230 if (addr->sin6_family != AF_INET6) 231 return -EINVAL; 232 233 addr_type = ipv6_addr_type(&addr->sin6_addr); 234 235 /* Raw sockets are IPv6 only */ 236 if (addr_type == IPV6_ADDR_MAPPED) 237 return -EADDRNOTAVAIL; 238 239 lock_sock(sk); 240 241 err = -EINVAL; 242 if (sk->sk_state != TCP_CLOSE) 243 goto out; 244 245 rcu_read_lock(); 246 /* Check if the address belongs to the host. */ 247 if (addr_type != IPV6_ADDR_ANY) { 248 struct net_device *dev = NULL; 249 250 if (__ipv6_addr_needs_scope_id(addr_type)) { 251 if (addr_len >= sizeof(struct sockaddr_in6) && 252 addr->sin6_scope_id) { 253 /* Override any existing binding, if another 254 * one is supplied by user. 255 */ 256 sk->sk_bound_dev_if = addr->sin6_scope_id; 257 } 258 259 /* Binding to link-local address requires an interface */ 260 if (!sk->sk_bound_dev_if) 261 goto out_unlock; 262 } 263 264 if (sk->sk_bound_dev_if) { 265 err = -ENODEV; 266 dev = dev_get_by_index_rcu(sock_net(sk), 267 sk->sk_bound_dev_if); 268 if (!dev) 269 goto out_unlock; 270 } 271 272 /* ipv4 addr of the socket is invalid. Only the 273 * unspecified and mapped address have a v4 equivalent. 274 */ 275 v4addr = LOOPBACK4_IPV6; 276 if (!(addr_type & IPV6_ADDR_MULTICAST) && 277 !ipv6_can_nonlocal_bind(sock_net(sk), inet)) { 278 err = -EADDRNOTAVAIL; 279 if (!ipv6_chk_addr(sock_net(sk), &addr->sin6_addr, 280 dev, 0)) { 281 goto out_unlock; 282 } 283 } 284 } 285 286 inet->inet_rcv_saddr = inet->inet_saddr = v4addr; 287 sk->sk_v6_rcv_saddr = addr->sin6_addr; 288 if (!(addr_type & IPV6_ADDR_MULTICAST)) 289 np->saddr = addr->sin6_addr; 290 err = 0; 291 out_unlock: 292 rcu_read_unlock(); 293 out: 294 release_sock(sk); 295 return err; 296 } 297 298 static void rawv6_err(struct sock *sk, struct sk_buff *skb, 299 u8 type, u8 code, int offset, __be32 info) 300 { 301 bool recverr = inet6_test_bit(RECVERR6, sk); 302 struct ipv6_pinfo *np = inet6_sk(sk); 303 int err; 304 int harderr; 305 306 /* Report error on raw socket, if: 307 1. User requested recverr. 308 2. Socket is connected (otherwise the error indication 309 is useless without recverr and error is hard. 310 */ 311 if (!recverr && sk->sk_state != TCP_ESTABLISHED) 312 return; 313 314 harderr = icmpv6_err_convert(type, code, &err); 315 if (type == ICMPV6_PKT_TOOBIG) { 316 ip6_sk_update_pmtu(skb, sk, info); 317 harderr = (READ_ONCE(np->pmtudisc) == IPV6_PMTUDISC_DO); 318 } 319 if (type == NDISC_REDIRECT) { 320 ip6_sk_redirect(skb, sk); 321 return; 322 } 323 if (recverr) { 324 u8 *payload = skb->data; 325 if (!inet_test_bit(HDRINCL, sk)) 326 payload += offset; 327 ipv6_icmp_error(sk, skb, err, 0, ntohl(info), payload); 328 } 329 330 if (recverr || harderr) { 331 sk->sk_err = err; 332 sk_error_report(sk); 333 } 334 } 335 336 void raw6_icmp_error(struct sk_buff *skb, int nexthdr, 337 u8 type, u8 code, int inner_offset, __be32 info) 338 { 339 struct net *net = dev_net(skb->dev); 340 struct hlist_head *hlist; 341 struct sock *sk; 342 int hash; 343 344 hash = raw_hashfunc(net, nexthdr); 345 hlist = &raw_v6_hashinfo.ht[hash]; 346 rcu_read_lock(); 347 sk_for_each_rcu(sk, hlist) { 348 /* Note: ipv6_hdr(skb) != skb->data */ 349 const struct ipv6hdr *ip6h = (const struct ipv6hdr *)skb->data; 350 351 if (!raw_v6_match(net, sk, nexthdr, &ip6h->saddr, &ip6h->daddr, 352 inet6_iif(skb), inet6_iif(skb))) 353 continue; 354 rawv6_err(sk, skb, type, code, inner_offset, info); 355 } 356 rcu_read_unlock(); 357 } 358 359 static inline int rawv6_rcv_skb(struct sock *sk, struct sk_buff *skb) 360 { 361 enum skb_drop_reason reason; 362 363 if ((raw6_sk(sk)->checksum || rcu_access_pointer(sk->sk_filter)) && 364 skb_checksum_complete(skb)) { 365 sk_drops_inc(sk); 366 sk_skb_reason_drop(sk, skb, SKB_DROP_REASON_SKB_CSUM); 367 return NET_RX_DROP; 368 } 369 370 /* Charge it to the socket. */ 371 skb_dst_drop(skb); 372 reason = sock_queue_rcv_skb_reason(sk, skb); 373 if (reason) { 374 sk_skb_reason_drop(sk, skb, reason); 375 return NET_RX_DROP; 376 } 377 378 return 0; 379 } 380 381 /* 382 * This is next to useless... 383 * if we demultiplex in network layer we don't need the extra call 384 * just to queue the skb... 385 * maybe we could have the network decide upon a hint if it 386 * should call raw_rcv for demultiplexing 387 */ 388 int rawv6_rcv(struct sock *sk, struct sk_buff *skb) 389 { 390 struct inet_sock *inet = inet_sk(sk); 391 struct raw6_sock *rp = raw6_sk(sk); 392 393 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 394 sk_drops_inc(sk); 395 sk_skb_reason_drop(sk, skb, SKB_DROP_REASON_XFRM_POLICY); 396 return NET_RX_DROP; 397 } 398 nf_reset_ct(skb); 399 400 if (!rp->checksum) 401 skb->ip_summed = CHECKSUM_UNNECESSARY; 402 403 if (skb->ip_summed == CHECKSUM_COMPLETE) { 404 skb_postpull_rcsum(skb, skb_network_header(skb), 405 skb_network_header_len(skb)); 406 if (!csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 407 &ipv6_hdr(skb)->daddr, 408 skb->len, inet->inet_num, skb->csum)) 409 skb->ip_summed = CHECKSUM_UNNECESSARY; 410 } 411 if (!skb_csum_unnecessary(skb)) 412 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 413 &ipv6_hdr(skb)->daddr, 414 skb->len, 415 inet->inet_num, 0)); 416 417 if (inet_test_bit(HDRINCL, sk)) { 418 if (skb_checksum_complete(skb)) { 419 sk_drops_inc(sk); 420 sk_skb_reason_drop(sk, skb, SKB_DROP_REASON_SKB_CSUM); 421 return NET_RX_DROP; 422 } 423 } 424 425 rawv6_rcv_skb(sk, skb); 426 return 0; 427 } 428 429 430 /* 431 * This should be easy, if there is something there 432 * we return it, otherwise we block. 433 */ 434 435 static int rawv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 436 int flags) 437 { 438 struct ipv6_pinfo *np = inet6_sk(sk); 439 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 440 struct sk_buff *skb; 441 size_t copied; 442 int err; 443 444 if (flags & MSG_OOB) 445 return -EOPNOTSUPP; 446 447 if (flags & MSG_ERRQUEUE) 448 return ipv6_recv_error(sk, msg, len); 449 450 if (np->rxopt.bits.rxpmtu && READ_ONCE(np->rxpmtu)) 451 return ipv6_recv_rxpmtu(sk, msg, len); 452 453 skb = skb_recv_datagram(sk, flags, &err); 454 if (!skb) 455 goto out; 456 457 copied = skb->len; 458 if (copied > len) { 459 copied = len; 460 msg->msg_flags |= MSG_TRUNC; 461 } 462 463 if (skb_csum_unnecessary(skb)) { 464 err = skb_copy_datagram_msg(skb, 0, msg, copied); 465 } else if (msg->msg_flags&MSG_TRUNC) { 466 if (__skb_checksum_complete(skb)) 467 goto csum_copy_err; 468 err = skb_copy_datagram_msg(skb, 0, msg, copied); 469 } else { 470 err = skb_copy_and_csum_datagram_msg(skb, 0, msg); 471 if (err == -EINVAL) 472 goto csum_copy_err; 473 } 474 if (err) 475 goto out_free; 476 477 /* Copy the address. */ 478 if (sin6) { 479 sin6->sin6_family = AF_INET6; 480 sin6->sin6_port = 0; 481 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 482 sin6->sin6_flowinfo = 0; 483 sin6->sin6_scope_id = ipv6_iface_scope_id(&sin6->sin6_addr, 484 inet6_iif(skb)); 485 msg->msg_namelen = sizeof(*sin6); 486 } 487 488 sock_recv_cmsgs(msg, sk, skb); 489 490 if (np->rxopt.all) 491 ip6_datagram_recv_ctl(sk, msg, skb); 492 493 err = copied; 494 if (flags & MSG_TRUNC) 495 err = skb->len; 496 497 out_free: 498 skb_free_datagram(sk, skb); 499 out: 500 return err; 501 502 csum_copy_err: 503 skb_kill_datagram(sk, skb, flags); 504 505 /* Error for blocking case is chosen to masquerade 506 as some normal condition. 507 */ 508 err = (flags&MSG_DONTWAIT) ? -EAGAIN : -EHOSTUNREACH; 509 goto out; 510 } 511 512 static int rawv6_push_pending_frames(struct sock *sk, struct flowi6 *fl6, 513 struct raw6_sock *rp) 514 { 515 struct ipv6_txoptions *opt; 516 struct sk_buff *skb; 517 int err = 0; 518 int offset; 519 int len; 520 int total_len; 521 __wsum tmp_csum; 522 __sum16 csum; 523 524 if (!rp->checksum) 525 goto send; 526 527 skb = skb_peek(&sk->sk_write_queue); 528 if (!skb) 529 goto out; 530 531 offset = rp->offset; 532 total_len = inet_sk(sk)->cork.base.length; 533 opt = inet_sk(sk)->cork.base6.opt; 534 total_len -= opt ? opt->opt_flen : 0; 535 536 if (offset >= total_len - 1) { 537 err = -EINVAL; 538 ip6_flush_pending_frames(sk); 539 goto out; 540 } 541 542 /* should be check HW csum miyazawa */ 543 if (skb_queue_len(&sk->sk_write_queue) == 1) { 544 /* 545 * Only one fragment on the socket. 546 */ 547 tmp_csum = skb->csum; 548 } else { 549 struct sk_buff *csum_skb = NULL; 550 tmp_csum = 0; 551 552 skb_queue_walk(&sk->sk_write_queue, skb) { 553 tmp_csum = csum_add(tmp_csum, skb->csum); 554 555 if (csum_skb) 556 continue; 557 558 len = skb->len - skb_transport_offset(skb); 559 if (offset >= len) { 560 offset -= len; 561 continue; 562 } 563 564 csum_skb = skb; 565 } 566 567 skb = csum_skb; 568 } 569 570 offset += skb_transport_offset(skb); 571 err = skb_copy_bits(skb, offset, &csum, 2); 572 if (err < 0) { 573 ip6_flush_pending_frames(sk); 574 goto out; 575 } 576 577 /* in case cksum was not initialized */ 578 if (unlikely(csum)) 579 tmp_csum = csum_sub(tmp_csum, csum_unfold(csum)); 580 581 csum = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 582 total_len, fl6->flowi6_proto, tmp_csum); 583 584 if (csum == 0 && fl6->flowi6_proto == IPPROTO_UDP) 585 csum = CSUM_MANGLED_0; 586 587 BUG_ON(skb_store_bits(skb, offset, &csum, 2)); 588 589 send: 590 err = ip6_push_pending_frames(sk); 591 out: 592 return err; 593 } 594 595 static int rawv6_send_hdrinc(struct sock *sk, struct msghdr *msg, int length, 596 struct flowi6 *fl6, struct dst_entry **dstp, 597 unsigned int flags, const struct sockcm_cookie *sockc) 598 { 599 struct net *net = sock_net(sk); 600 struct ipv6hdr *iph; 601 struct sk_buff *skb; 602 int err; 603 struct rt6_info *rt = dst_rt6_info(*dstp); 604 int hlen = LL_RESERVED_SPACE(rt->dst.dev); 605 int tlen = rt->dst.dev->needed_tailroom; 606 607 if (length > rt->dst.dev->mtu) { 608 ipv6_local_error(sk, EMSGSIZE, fl6, rt->dst.dev->mtu); 609 return -EMSGSIZE; 610 } 611 if (length < sizeof(struct ipv6hdr)) 612 return -EINVAL; 613 if (flags&MSG_PROBE) 614 goto out; 615 616 skb = sock_alloc_send_skb(sk, 617 length + hlen + tlen + 15, 618 flags & MSG_DONTWAIT, &err); 619 if (!skb) 620 goto error; 621 skb_reserve(skb, hlen); 622 623 skb->protocol = htons(ETH_P_IPV6); 624 skb->priority = sockc->priority; 625 skb->mark = sockc->mark; 626 skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, sk->sk_clockid); 627 628 skb_put(skb, length); 629 skb_reset_network_header(skb); 630 iph = ipv6_hdr(skb); 631 632 skb->ip_summed = CHECKSUM_NONE; 633 634 skb_setup_tx_timestamp(skb, sockc); 635 636 if (flags & MSG_CONFIRM) 637 skb_set_dst_pending_confirm(skb, 1); 638 639 skb->transport_header = skb->network_header; 640 err = memcpy_from_msg(iph, msg, length); 641 if (err) { 642 err = -EFAULT; 643 kfree_skb(skb); 644 goto error; 645 } 646 647 skb_dst_set(skb, &rt->dst); 648 *dstp = NULL; 649 650 /* if egress device is enslaved to an L3 master device pass the 651 * skb to its handler for processing 652 */ 653 skb = l3mdev_ip6_out(sk, skb); 654 if (unlikely(!skb)) 655 return 0; 656 657 /* Acquire rcu_read_lock() in case we need to use rt->rt6i_idev 658 * in the error path. Since skb has been freed, the dst could 659 * have been queued for deletion. 660 */ 661 rcu_read_lock(); 662 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS); 663 err = NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, skb, 664 NULL, rt->dst.dev, dst_output); 665 if (err > 0) 666 err = net_xmit_errno(err); 667 if (err) { 668 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); 669 rcu_read_unlock(); 670 goto error_check; 671 } 672 rcu_read_unlock(); 673 out: 674 return 0; 675 676 error: 677 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); 678 error_check: 679 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) 680 err = 0; 681 return err; 682 } 683 684 struct raw6_frag_vec { 685 struct msghdr *msg; 686 int hlen; 687 char c[4]; 688 }; 689 690 static int rawv6_probe_proto_opt(struct raw6_frag_vec *rfv, struct flowi6 *fl6) 691 { 692 int err = 0; 693 switch (fl6->flowi6_proto) { 694 case IPPROTO_ICMPV6: 695 rfv->hlen = 2; 696 err = memcpy_from_msg(rfv->c, rfv->msg, rfv->hlen); 697 if (!err) { 698 fl6->fl6_icmp_type = rfv->c[0]; 699 fl6->fl6_icmp_code = rfv->c[1]; 700 } 701 break; 702 case IPPROTO_MH: 703 rfv->hlen = 4; 704 err = memcpy_from_msg(rfv->c, rfv->msg, rfv->hlen); 705 if (!err) 706 fl6->fl6_mh_type = rfv->c[2]; 707 } 708 return err; 709 } 710 711 static int raw6_getfrag(void *from, char *to, int offset, int len, int odd, 712 struct sk_buff *skb) 713 { 714 struct raw6_frag_vec *rfv = from; 715 716 if (offset < rfv->hlen) { 717 int copy = min(rfv->hlen - offset, len); 718 719 if (skb->ip_summed == CHECKSUM_PARTIAL) 720 memcpy(to, rfv->c + offset, copy); 721 else 722 skb->csum = csum_block_add( 723 skb->csum, 724 csum_partial_copy_nocheck(rfv->c + offset, 725 to, copy), 726 odd); 727 728 odd = 0; 729 offset += copy; 730 to += copy; 731 len -= copy; 732 733 if (!len) 734 return 0; 735 } 736 737 offset -= rfv->hlen; 738 739 return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb); 740 } 741 742 static int rawv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 743 { 744 struct ipv6_txoptions *opt_to_free = NULL; 745 struct ipv6_txoptions opt_space; 746 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 747 struct in6_addr *daddr, *final_p, final; 748 struct inet_sock *inet = inet_sk(sk); 749 struct ipv6_pinfo *np = inet6_sk(sk); 750 struct raw6_sock *rp = raw6_sk(sk); 751 struct ipv6_txoptions *opt = NULL; 752 struct ip6_flowlabel *flowlabel = NULL; 753 struct dst_entry *dst = NULL; 754 struct raw6_frag_vec rfv; 755 struct flowi6 fl6; 756 struct ipcm6_cookie ipc6; 757 int addr_len = msg->msg_namelen; 758 int hdrincl; 759 u16 proto; 760 int err; 761 762 /* Rough check on arithmetic overflow, 763 better check is made in ip6_append_data(). 764 */ 765 if (len > INT_MAX) 766 return -EMSGSIZE; 767 768 /* Mirror BSD error message compatibility */ 769 if (msg->msg_flags & MSG_OOB) 770 return -EOPNOTSUPP; 771 772 hdrincl = inet_test_bit(HDRINCL, sk); 773 774 ipcm6_init_sk(&ipc6, sk); 775 776 /* 777 * Get and verify the address. 778 */ 779 memset(&fl6, 0, sizeof(fl6)); 780 781 fl6.flowi6_mark = ipc6.sockc.mark; 782 fl6.flowi6_uid = sk_uid(sk); 783 784 if (sin6) { 785 if (addr_len < SIN6_LEN_RFC2133) 786 return -EINVAL; 787 788 if (sin6->sin6_family && sin6->sin6_family != AF_INET6) 789 return -EAFNOSUPPORT; 790 791 /* port is the proto value [0..255] carried in nexthdr */ 792 proto = ntohs(sin6->sin6_port); 793 794 if (!proto) 795 proto = inet->inet_num; 796 else if (proto != inet->inet_num && 797 inet->inet_num != IPPROTO_RAW) 798 return -EINVAL; 799 800 if (proto > 255) 801 return -EINVAL; 802 803 daddr = &sin6->sin6_addr; 804 if (inet6_test_bit(SNDFLOW, sk)) { 805 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 806 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 807 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 808 if (IS_ERR(flowlabel)) 809 return -EINVAL; 810 } 811 } 812 813 /* 814 * Otherwise it will be difficult to maintain 815 * sk->sk_dst_cache. 816 */ 817 if (sk->sk_state == TCP_ESTABLISHED && 818 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 819 daddr = &sk->sk_v6_daddr; 820 821 if (addr_len >= sizeof(struct sockaddr_in6) && 822 sin6->sin6_scope_id && 823 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 824 fl6.flowi6_oif = sin6->sin6_scope_id; 825 } else { 826 if (sk->sk_state != TCP_ESTABLISHED) 827 return -EDESTADDRREQ; 828 829 proto = inet->inet_num; 830 daddr = &sk->sk_v6_daddr; 831 fl6.flowlabel = np->flow_label; 832 } 833 834 if (fl6.flowi6_oif == 0) 835 fl6.flowi6_oif = sk->sk_bound_dev_if; 836 837 if (msg->msg_controllen) { 838 opt = &opt_space; 839 memset(opt, 0, sizeof(struct ipv6_txoptions)); 840 opt->tot_len = sizeof(struct ipv6_txoptions); 841 ipc6.opt = opt; 842 843 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6); 844 if (err < 0) { 845 fl6_sock_release(flowlabel); 846 return err; 847 } 848 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 849 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 850 if (IS_ERR(flowlabel)) 851 return -EINVAL; 852 } 853 if (!(opt->opt_nflen|opt->opt_flen)) 854 opt = NULL; 855 } 856 if (!opt) { 857 opt = txopt_get(np); 858 opt_to_free = opt; 859 } 860 if (flowlabel) 861 opt = fl6_merge_options(&opt_space, flowlabel, opt); 862 opt = ipv6_fixup_options(&opt_space, opt); 863 864 fl6.flowi6_proto = proto; 865 fl6.flowi6_mark = ipc6.sockc.mark; 866 867 if (!hdrincl) { 868 rfv.msg = msg; 869 rfv.hlen = 0; 870 err = rawv6_probe_proto_opt(&rfv, &fl6); 871 if (err) 872 goto out; 873 } 874 875 if (!ipv6_addr_any(daddr)) 876 fl6.daddr = *daddr; 877 else 878 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 879 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 880 fl6.saddr = np->saddr; 881 882 final_p = fl6_update_dst(&fl6, opt, &final); 883 884 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) 885 fl6.flowi6_oif = READ_ONCE(np->mcast_oif); 886 else if (!fl6.flowi6_oif) 887 fl6.flowi6_oif = READ_ONCE(np->ucast_oif); 888 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6)); 889 890 if (hdrincl) 891 fl6.flowi6_flags |= FLOWI_FLAG_KNOWN_NH; 892 893 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 894 895 dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p); 896 if (IS_ERR(dst)) { 897 err = PTR_ERR(dst); 898 goto out; 899 } 900 if (ipc6.hlimit < 0) 901 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 902 903 if (msg->msg_flags&MSG_CONFIRM) 904 goto do_confirm; 905 906 back_from_confirm: 907 if (hdrincl) 908 err = rawv6_send_hdrinc(sk, msg, len, &fl6, &dst, 909 msg->msg_flags, &ipc6.sockc); 910 else { 911 ipc6.opt = opt; 912 lock_sock(sk); 913 err = ip6_append_data(sk, raw6_getfrag, &rfv, 914 len, 0, &ipc6, &fl6, dst_rt6_info(dst), 915 msg->msg_flags); 916 917 if (err) 918 ip6_flush_pending_frames(sk); 919 else if (!(msg->msg_flags & MSG_MORE)) 920 err = rawv6_push_pending_frames(sk, &fl6, rp); 921 release_sock(sk); 922 } 923 done: 924 dst_release(dst); 925 out: 926 fl6_sock_release(flowlabel); 927 txopt_put(opt_to_free); 928 return err < 0 ? err : len; 929 do_confirm: 930 if (msg->msg_flags & MSG_PROBE) 931 dst_confirm_neigh(dst, &fl6.daddr); 932 if (!(msg->msg_flags & MSG_PROBE) || len) 933 goto back_from_confirm; 934 err = 0; 935 goto done; 936 } 937 938 static int rawv6_seticmpfilter(struct sock *sk, int optname, 939 sockptr_t optval, int optlen) 940 { 941 switch (optname) { 942 case ICMPV6_FILTER: 943 if (optlen > sizeof(struct icmp6_filter)) 944 optlen = sizeof(struct icmp6_filter); 945 if (copy_from_sockptr(&raw6_sk(sk)->filter, optval, optlen)) 946 return -EFAULT; 947 return 0; 948 default: 949 return -ENOPROTOOPT; 950 } 951 952 return 0; 953 } 954 955 static int rawv6_geticmpfilter(struct sock *sk, int optname, 956 char __user *optval, int __user *optlen) 957 { 958 int len; 959 960 switch (optname) { 961 case ICMPV6_FILTER: 962 if (get_user(len, optlen)) 963 return -EFAULT; 964 if (len < 0) 965 return -EINVAL; 966 if (len > sizeof(struct icmp6_filter)) 967 len = sizeof(struct icmp6_filter); 968 if (put_user(len, optlen)) 969 return -EFAULT; 970 if (copy_to_user(optval, &raw6_sk(sk)->filter, len)) 971 return -EFAULT; 972 return 0; 973 default: 974 return -ENOPROTOOPT; 975 } 976 977 return 0; 978 } 979 980 981 static int do_rawv6_setsockopt(struct sock *sk, int level, int optname, 982 sockptr_t optval, unsigned int optlen) 983 { 984 struct raw6_sock *rp = raw6_sk(sk); 985 int val; 986 987 if (optlen < sizeof(val)) 988 return -EINVAL; 989 990 if (copy_from_sockptr(&val, optval, sizeof(val))) 991 return -EFAULT; 992 993 switch (optname) { 994 case IPV6_HDRINCL: 995 if (sk->sk_type != SOCK_RAW) 996 return -EINVAL; 997 inet_assign_bit(HDRINCL, sk, val); 998 return 0; 999 case IPV6_CHECKSUM: 1000 if (inet_sk(sk)->inet_num == IPPROTO_ICMPV6 && 1001 level == IPPROTO_IPV6) { 1002 /* 1003 * RFC3542 tells that IPV6_CHECKSUM socket 1004 * option in the IPPROTO_IPV6 level is not 1005 * allowed on ICMPv6 sockets. 1006 * If you want to set it, use IPPROTO_RAW 1007 * level IPV6_CHECKSUM socket option 1008 * (Linux extension). 1009 */ 1010 return -EINVAL; 1011 } 1012 1013 /* You may get strange result with a positive odd offset; 1014 RFC2292bis agrees with me. */ 1015 if (val > 0 && (val&1)) 1016 return -EINVAL; 1017 if (val < 0) { 1018 rp->checksum = 0; 1019 } else { 1020 rp->checksum = 1; 1021 rp->offset = val; 1022 } 1023 1024 return 0; 1025 1026 default: 1027 return -ENOPROTOOPT; 1028 } 1029 } 1030 1031 static int rawv6_setsockopt(struct sock *sk, int level, int optname, 1032 sockptr_t optval, unsigned int optlen) 1033 { 1034 switch (level) { 1035 case SOL_RAW: 1036 break; 1037 1038 case SOL_ICMPV6: 1039 if (inet_sk(sk)->inet_num != IPPROTO_ICMPV6) 1040 return -EOPNOTSUPP; 1041 return rawv6_seticmpfilter(sk, optname, optval, optlen); 1042 case SOL_IPV6: 1043 if (optname == IPV6_CHECKSUM || 1044 optname == IPV6_HDRINCL) 1045 break; 1046 fallthrough; 1047 default: 1048 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1049 } 1050 1051 return do_rawv6_setsockopt(sk, level, optname, optval, optlen); 1052 } 1053 1054 static int do_rawv6_getsockopt(struct sock *sk, int level, int optname, 1055 char __user *optval, int __user *optlen) 1056 { 1057 struct raw6_sock *rp = raw6_sk(sk); 1058 int val, len; 1059 1060 if (get_user(len, optlen)) 1061 return -EFAULT; 1062 1063 switch (optname) { 1064 case IPV6_HDRINCL: 1065 val = inet_test_bit(HDRINCL, sk); 1066 break; 1067 case IPV6_CHECKSUM: 1068 /* 1069 * We allow getsockopt() for IPPROTO_IPV6-level 1070 * IPV6_CHECKSUM socket option on ICMPv6 sockets 1071 * since RFC3542 is silent about it. 1072 */ 1073 if (rp->checksum == 0) 1074 val = -1; 1075 else 1076 val = rp->offset; 1077 break; 1078 1079 default: 1080 return -ENOPROTOOPT; 1081 } 1082 1083 len = min_t(unsigned int, sizeof(int), len); 1084 1085 if (put_user(len, optlen)) 1086 return -EFAULT; 1087 if (copy_to_user(optval, &val, len)) 1088 return -EFAULT; 1089 return 0; 1090 } 1091 1092 static int rawv6_getsockopt(struct sock *sk, int level, int optname, 1093 char __user *optval, int __user *optlen) 1094 { 1095 switch (level) { 1096 case SOL_RAW: 1097 break; 1098 1099 case SOL_ICMPV6: 1100 if (inet_sk(sk)->inet_num != IPPROTO_ICMPV6) 1101 return -EOPNOTSUPP; 1102 return rawv6_geticmpfilter(sk, optname, optval, optlen); 1103 case SOL_IPV6: 1104 if (optname == IPV6_CHECKSUM || 1105 optname == IPV6_HDRINCL) 1106 break; 1107 fallthrough; 1108 default: 1109 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1110 } 1111 1112 return do_rawv6_getsockopt(sk, level, optname, optval, optlen); 1113 } 1114 1115 static int rawv6_ioctl(struct sock *sk, int cmd, int *karg) 1116 { 1117 switch (cmd) { 1118 case SIOCOUTQ: { 1119 *karg = sk_wmem_alloc_get(sk); 1120 return 0; 1121 } 1122 case SIOCINQ: { 1123 struct sk_buff *skb; 1124 1125 spin_lock_bh(&sk->sk_receive_queue.lock); 1126 skb = skb_peek(&sk->sk_receive_queue); 1127 if (skb) 1128 *karg = skb->len; 1129 else 1130 *karg = 0; 1131 spin_unlock_bh(&sk->sk_receive_queue.lock); 1132 return 0; 1133 } 1134 1135 default: 1136 #ifdef CONFIG_IPV6_MROUTE 1137 return ip6mr_ioctl(sk, cmd, karg); 1138 #else 1139 return -ENOIOCTLCMD; 1140 #endif 1141 } 1142 } 1143 1144 #ifdef CONFIG_COMPAT 1145 static int compat_rawv6_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg) 1146 { 1147 switch (cmd) { 1148 case SIOCOUTQ: 1149 case SIOCINQ: 1150 return -ENOIOCTLCMD; 1151 default: 1152 #ifdef CONFIG_IPV6_MROUTE 1153 return ip6mr_compat_ioctl(sk, cmd, compat_ptr(arg)); 1154 #else 1155 return -ENOIOCTLCMD; 1156 #endif 1157 } 1158 } 1159 #endif 1160 1161 static void rawv6_close(struct sock *sk, long timeout) 1162 { 1163 if (inet_sk(sk)->inet_num == IPPROTO_RAW) 1164 ip6_ra_control(sk, -1); 1165 ip6mr_sk_done(sk); 1166 sk_common_release(sk); 1167 } 1168 1169 static void raw6_destroy(struct sock *sk) 1170 { 1171 lock_sock(sk); 1172 ip6_flush_pending_frames(sk); 1173 release_sock(sk); 1174 } 1175 1176 static int rawv6_init_sk(struct sock *sk) 1177 { 1178 struct raw6_sock *rp = raw6_sk(sk); 1179 1180 sk->sk_drop_counters = &rp->drop_counters; 1181 switch (inet_sk(sk)->inet_num) { 1182 case IPPROTO_ICMPV6: 1183 rp->checksum = 1; 1184 rp->offset = 2; 1185 break; 1186 case IPPROTO_MH: 1187 rp->checksum = 1; 1188 rp->offset = 4; 1189 break; 1190 default: 1191 break; 1192 } 1193 return 0; 1194 } 1195 1196 struct proto rawv6_prot = { 1197 .name = "RAWv6", 1198 .owner = THIS_MODULE, 1199 .close = rawv6_close, 1200 .destroy = raw6_destroy, 1201 .connect = ip6_datagram_connect_v6_only, 1202 .disconnect = __udp_disconnect, 1203 .ioctl = rawv6_ioctl, 1204 .init = rawv6_init_sk, 1205 .setsockopt = rawv6_setsockopt, 1206 .getsockopt = rawv6_getsockopt, 1207 .sendmsg = rawv6_sendmsg, 1208 .recvmsg = rawv6_recvmsg, 1209 .bind = rawv6_bind, 1210 .backlog_rcv = rawv6_rcv_skb, 1211 .hash = raw_hash_sk, 1212 .unhash = raw_unhash_sk, 1213 .obj_size = sizeof(struct raw6_sock), 1214 .ipv6_pinfo_offset = offsetof(struct raw6_sock, inet6), 1215 .useroffset = offsetof(struct raw6_sock, filter), 1216 .usersize = sizeof_field(struct raw6_sock, filter), 1217 .h.raw_hash = &raw_v6_hashinfo, 1218 #ifdef CONFIG_COMPAT 1219 .compat_ioctl = compat_rawv6_ioctl, 1220 #endif 1221 .diag_destroy = raw_abort, 1222 }; 1223 1224 #ifdef CONFIG_PROC_FS 1225 static int raw6_seq_show(struct seq_file *seq, void *v) 1226 { 1227 if (v == SEQ_START_TOKEN) { 1228 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1229 } else { 1230 struct sock *sp = v; 1231 __u16 srcp = inet_sk(sp)->inet_num; 1232 ip6_dgram_sock_seq_show(seq, v, srcp, 0, 1233 raw_seq_private(seq)->bucket); 1234 } 1235 return 0; 1236 } 1237 1238 static const struct seq_operations raw6_seq_ops = { 1239 .start = raw_seq_start, 1240 .next = raw_seq_next, 1241 .stop = raw_seq_stop, 1242 .show = raw6_seq_show, 1243 }; 1244 1245 static int __net_init raw6_init_net(struct net *net) 1246 { 1247 if (!proc_create_net_data("raw6", 0444, net->proc_net, &raw6_seq_ops, 1248 sizeof(struct raw_iter_state), &raw_v6_hashinfo)) 1249 return -ENOMEM; 1250 1251 return 0; 1252 } 1253 1254 static void __net_exit raw6_exit_net(struct net *net) 1255 { 1256 remove_proc_entry("raw6", net->proc_net); 1257 } 1258 1259 static struct pernet_operations raw6_net_ops = { 1260 .init = raw6_init_net, 1261 .exit = raw6_exit_net, 1262 }; 1263 1264 int __init raw6_proc_init(void) 1265 { 1266 return register_pernet_subsys(&raw6_net_ops); 1267 } 1268 1269 void raw6_proc_exit(void) 1270 { 1271 unregister_pernet_subsys(&raw6_net_ops); 1272 } 1273 #endif /* CONFIG_PROC_FS */ 1274 1275 /* Same as inet6_dgram_ops, sans udp_poll. */ 1276 const struct proto_ops inet6_sockraw_ops = { 1277 .family = PF_INET6, 1278 .owner = THIS_MODULE, 1279 .release = inet6_release, 1280 .bind = inet6_bind, 1281 .connect = inet_dgram_connect, /* ok */ 1282 .socketpair = sock_no_socketpair, /* a do nothing */ 1283 .accept = sock_no_accept, /* a do nothing */ 1284 .getname = inet6_getname, 1285 .poll = datagram_poll, /* ok */ 1286 .ioctl = inet6_ioctl, /* must change */ 1287 .gettstamp = sock_gettstamp, 1288 .listen = sock_no_listen, /* ok */ 1289 .shutdown = inet_shutdown, /* ok */ 1290 .setsockopt = sock_common_setsockopt, /* ok */ 1291 .getsockopt = sock_common_getsockopt, /* ok */ 1292 .sendmsg = inet_sendmsg, /* ok */ 1293 .recvmsg = sock_common_recvmsg, /* ok */ 1294 .mmap = sock_no_mmap, 1295 #ifdef CONFIG_COMPAT 1296 .compat_ioctl = inet6_compat_ioctl, 1297 #endif 1298 }; 1299 1300 static struct inet_protosw rawv6_protosw = { 1301 .type = SOCK_RAW, 1302 .protocol = IPPROTO_IP, /* wild card */ 1303 .prot = &rawv6_prot, 1304 .ops = &inet6_sockraw_ops, 1305 .flags = INET_PROTOSW_REUSE, 1306 }; 1307 1308 int __init rawv6_init(void) 1309 { 1310 return inet6_register_protosw(&rawv6_protosw); 1311 } 1312 1313 void rawv6_exit(void) 1314 { 1315 inet6_unregister_protosw(&rawv6_protosw); 1316 } 1317