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, const struct sk_buff *skb) 94 { 95 struct icmp6hdr _hdr; 96 const struct icmp6hdr *hdr; 97 98 /* We require only the four bytes of the ICMPv6 header, not any 99 * additional bytes of message body in "struct icmp6hdr". 100 */ 101 hdr = skb_header_pointer(skb, skb_transport_offset(skb), 102 ICMPV6_HDRLEN, &_hdr); 103 if (hdr) { 104 const __u32 *data = &raw6_sk(sk)->filter.data[0]; 105 unsigned int type = hdr->icmp6_type; 106 107 return (data[type >> 5] & (1U << (type & 31))) != 0; 108 } 109 return 1; 110 } 111 112 #if IS_ENABLED(CONFIG_IPV6_MIP6) 113 typedef int mh_filter_t(struct sock *sock, struct sk_buff *skb); 114 115 static mh_filter_t __rcu *mh_filter __read_mostly; 116 117 int rawv6_mh_filter_register(mh_filter_t filter) 118 { 119 rcu_assign_pointer(mh_filter, filter); 120 return 0; 121 } 122 EXPORT_SYMBOL(rawv6_mh_filter_register); 123 124 int rawv6_mh_filter_unregister(mh_filter_t filter) 125 { 126 RCU_INIT_POINTER(mh_filter, NULL); 127 synchronize_rcu(); 128 return 0; 129 } 130 EXPORT_SYMBOL(rawv6_mh_filter_unregister); 131 132 #endif 133 134 /* 135 * demultiplex raw sockets. 136 * (should consider queueing the skb in the sock receive_queue 137 * without calling rawv6.c) 138 * 139 * Caller owns SKB so we must make clones. 140 */ 141 static bool ipv6_raw_deliver(struct sk_buff *skb, int nexthdr) 142 { 143 struct net *net = dev_net(skb->dev); 144 const struct in6_addr *saddr; 145 const struct in6_addr *daddr; 146 struct hlist_head *hlist; 147 struct sock *sk; 148 bool delivered = false; 149 __u8 hash; 150 151 saddr = &ipv6_hdr(skb)->saddr; 152 daddr = saddr + 1; 153 154 hash = raw_hashfunc(net, nexthdr); 155 hlist = &raw_v6_hashinfo.ht[hash]; 156 rcu_read_lock(); 157 sk_for_each_rcu(sk, hlist) { 158 int filtered; 159 160 if (!raw_v6_match(net, sk, nexthdr, daddr, saddr, 161 inet6_iif(skb), inet6_sdif(skb))) 162 continue; 163 delivered = true; 164 switch (nexthdr) { 165 case IPPROTO_ICMPV6: 166 filtered = icmpv6_filter(sk, skb); 167 break; 168 169 #if IS_ENABLED(CONFIG_IPV6_MIP6) 170 case IPPROTO_MH: 171 { 172 /* XXX: To validate MH only once for each packet, 173 * this is placed here. It should be after checking 174 * xfrm policy, however it doesn't. The checking xfrm 175 * policy is placed in rawv6_rcv() because it is 176 * required for each socket. 177 */ 178 mh_filter_t *filter; 179 180 filter = rcu_dereference(mh_filter); 181 filtered = filter ? (*filter)(sk, skb) : 0; 182 break; 183 } 184 #endif 185 default: 186 filtered = 0; 187 break; 188 } 189 190 if (filtered < 0) 191 break; 192 if (filtered == 0) { 193 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC); 194 195 /* Not releasing hash table! */ 196 if (clone) 197 rawv6_rcv(sk, clone); 198 } 199 } 200 rcu_read_unlock(); 201 return delivered; 202 } 203 204 bool raw6_local_deliver(struct sk_buff *skb, int nexthdr) 205 { 206 return ipv6_raw_deliver(skb, nexthdr); 207 } 208 209 /* This cleans up af_inet6 a bit. -DaveM */ 210 static int rawv6_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len) 211 { 212 struct inet_sock *inet = inet_sk(sk); 213 struct ipv6_pinfo *np = inet6_sk(sk); 214 struct sockaddr_in6 *addr = (struct sockaddr_in6 *) uaddr; 215 __be32 v4addr = 0; 216 int addr_type; 217 int err; 218 219 if (addr_len < SIN6_LEN_RFC2133) 220 return -EINVAL; 221 222 if (addr->sin6_family != AF_INET6) 223 return -EINVAL; 224 225 addr_type = ipv6_addr_type(&addr->sin6_addr); 226 227 /* Raw sockets are IPv6 only */ 228 if (addr_type == IPV6_ADDR_MAPPED) 229 return -EADDRNOTAVAIL; 230 231 lock_sock(sk); 232 233 err = -EINVAL; 234 if (sk->sk_state != TCP_CLOSE) 235 goto out; 236 237 rcu_read_lock(); 238 /* Check if the address belongs to the host. */ 239 if (addr_type != IPV6_ADDR_ANY) { 240 struct net_device *dev = NULL; 241 242 if (__ipv6_addr_needs_scope_id(addr_type)) { 243 if (addr_len >= sizeof(struct sockaddr_in6) && 244 addr->sin6_scope_id) { 245 /* Override any existing binding, if another 246 * one is supplied by user. 247 */ 248 sk->sk_bound_dev_if = addr->sin6_scope_id; 249 } 250 251 /* Binding to link-local address requires an interface */ 252 if (!sk->sk_bound_dev_if) 253 goto out_unlock; 254 } 255 256 if (sk->sk_bound_dev_if) { 257 err = -ENODEV; 258 dev = dev_get_by_index_rcu(sock_net(sk), 259 sk->sk_bound_dev_if); 260 if (!dev) 261 goto out_unlock; 262 } 263 264 /* ipv4 addr of the socket is invalid. Only the 265 * unspecified and mapped address have a v4 equivalent. 266 */ 267 v4addr = LOOPBACK4_IPV6; 268 if (!(addr_type & IPV6_ADDR_MULTICAST) && 269 !ipv6_can_nonlocal_bind(sock_net(sk), inet)) { 270 err = -EADDRNOTAVAIL; 271 if (!ipv6_chk_addr(sock_net(sk), &addr->sin6_addr, 272 dev, 0)) { 273 goto out_unlock; 274 } 275 } 276 } 277 278 inet->inet_rcv_saddr = inet->inet_saddr = v4addr; 279 sk->sk_v6_rcv_saddr = addr->sin6_addr; 280 if (!(addr_type & IPV6_ADDR_MULTICAST)) 281 np->saddr = addr->sin6_addr; 282 err = 0; 283 out_unlock: 284 rcu_read_unlock(); 285 out: 286 release_sock(sk); 287 return err; 288 } 289 290 static void rawv6_err(struct sock *sk, struct sk_buff *skb, 291 u8 type, u8 code, int offset, __be32 info) 292 { 293 bool recverr = inet6_test_bit(RECVERR6, sk); 294 struct ipv6_pinfo *np = inet6_sk(sk); 295 int err; 296 int harderr; 297 298 /* Report error on raw socket, if: 299 1. User requested recverr. 300 2. Socket is connected (otherwise the error indication 301 is useless without recverr and error is hard. 302 */ 303 if (!recverr && sk->sk_state != TCP_ESTABLISHED) 304 return; 305 306 harderr = icmpv6_err_convert(type, code, &err); 307 if (type == ICMPV6_PKT_TOOBIG) { 308 ip6_sk_update_pmtu(skb, sk, info); 309 harderr = (READ_ONCE(np->pmtudisc) == IPV6_PMTUDISC_DO); 310 } 311 if (type == NDISC_REDIRECT) { 312 ip6_sk_redirect(skb, sk); 313 return; 314 } 315 if (recverr) { 316 u8 *payload = skb->data; 317 if (!inet_test_bit(HDRINCL, sk)) 318 payload += offset; 319 ipv6_icmp_error(sk, skb, err, 0, ntohl(info), payload); 320 } 321 322 if (recverr || harderr) { 323 sk->sk_err = err; 324 sk_error_report(sk); 325 } 326 } 327 328 void raw6_icmp_error(struct sk_buff *skb, int nexthdr, 329 u8 type, u8 code, int inner_offset, __be32 info) 330 { 331 struct net *net = dev_net(skb->dev); 332 struct hlist_head *hlist; 333 struct sock *sk; 334 int hash; 335 336 hash = raw_hashfunc(net, nexthdr); 337 hlist = &raw_v6_hashinfo.ht[hash]; 338 rcu_read_lock(); 339 sk_for_each_rcu(sk, hlist) { 340 /* Note: ipv6_hdr(skb) != skb->data */ 341 const struct ipv6hdr *ip6h = (const struct ipv6hdr *)skb->data; 342 343 if (!raw_v6_match(net, sk, nexthdr, &ip6h->saddr, &ip6h->daddr, 344 inet6_iif(skb), inet6_iif(skb))) 345 continue; 346 rawv6_err(sk, skb, type, code, inner_offset, info); 347 } 348 rcu_read_unlock(); 349 } 350 351 static inline int rawv6_rcv_skb(struct sock *sk, struct sk_buff *skb) 352 { 353 enum skb_drop_reason reason; 354 355 if ((raw6_sk(sk)->checksum || rcu_access_pointer(sk->sk_filter)) && 356 skb_checksum_complete(skb)) { 357 atomic_inc(&sk->sk_drops); 358 kfree_skb_reason(skb, SKB_DROP_REASON_SKB_CSUM); 359 return NET_RX_DROP; 360 } 361 362 /* Charge it to the socket. */ 363 skb_dst_drop(skb); 364 if (sock_queue_rcv_skb_reason(sk, skb, &reason) < 0) { 365 kfree_skb_reason(skb, reason); 366 return NET_RX_DROP; 367 } 368 369 return 0; 370 } 371 372 /* 373 * This is next to useless... 374 * if we demultiplex in network layer we don't need the extra call 375 * just to queue the skb... 376 * maybe we could have the network decide upon a hint if it 377 * should call raw_rcv for demultiplexing 378 */ 379 int rawv6_rcv(struct sock *sk, struct sk_buff *skb) 380 { 381 struct inet_sock *inet = inet_sk(sk); 382 struct raw6_sock *rp = raw6_sk(sk); 383 384 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 385 atomic_inc(&sk->sk_drops); 386 kfree_skb_reason(skb, SKB_DROP_REASON_XFRM_POLICY); 387 return NET_RX_DROP; 388 } 389 nf_reset_ct(skb); 390 391 if (!rp->checksum) 392 skb->ip_summed = CHECKSUM_UNNECESSARY; 393 394 if (skb->ip_summed == CHECKSUM_COMPLETE) { 395 skb_postpull_rcsum(skb, skb_network_header(skb), 396 skb_network_header_len(skb)); 397 if (!csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 398 &ipv6_hdr(skb)->daddr, 399 skb->len, inet->inet_num, skb->csum)) 400 skb->ip_summed = CHECKSUM_UNNECESSARY; 401 } 402 if (!skb_csum_unnecessary(skb)) 403 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 404 &ipv6_hdr(skb)->daddr, 405 skb->len, 406 inet->inet_num, 0)); 407 408 if (inet_test_bit(HDRINCL, sk)) { 409 if (skb_checksum_complete(skb)) { 410 atomic_inc(&sk->sk_drops); 411 kfree_skb_reason(skb, SKB_DROP_REASON_SKB_CSUM); 412 return NET_RX_DROP; 413 } 414 } 415 416 rawv6_rcv_skb(sk, skb); 417 return 0; 418 } 419 420 421 /* 422 * This should be easy, if there is something there 423 * we return it, otherwise we block. 424 */ 425 426 static int rawv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 427 int flags, int *addr_len) 428 { 429 struct ipv6_pinfo *np = inet6_sk(sk); 430 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 431 struct sk_buff *skb; 432 size_t copied; 433 int err; 434 435 if (flags & MSG_OOB) 436 return -EOPNOTSUPP; 437 438 if (flags & MSG_ERRQUEUE) 439 return ipv6_recv_error(sk, msg, len, addr_len); 440 441 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 442 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 443 444 skb = skb_recv_datagram(sk, flags, &err); 445 if (!skb) 446 goto out; 447 448 copied = skb->len; 449 if (copied > len) { 450 copied = len; 451 msg->msg_flags |= MSG_TRUNC; 452 } 453 454 if (skb_csum_unnecessary(skb)) { 455 err = skb_copy_datagram_msg(skb, 0, msg, copied); 456 } else if (msg->msg_flags&MSG_TRUNC) { 457 if (__skb_checksum_complete(skb)) 458 goto csum_copy_err; 459 err = skb_copy_datagram_msg(skb, 0, msg, copied); 460 } else { 461 err = skb_copy_and_csum_datagram_msg(skb, 0, msg); 462 if (err == -EINVAL) 463 goto csum_copy_err; 464 } 465 if (err) 466 goto out_free; 467 468 /* Copy the address. */ 469 if (sin6) { 470 sin6->sin6_family = AF_INET6; 471 sin6->sin6_port = 0; 472 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 473 sin6->sin6_flowinfo = 0; 474 sin6->sin6_scope_id = ipv6_iface_scope_id(&sin6->sin6_addr, 475 inet6_iif(skb)); 476 *addr_len = sizeof(*sin6); 477 } 478 479 sock_recv_cmsgs(msg, sk, skb); 480 481 if (np->rxopt.all) 482 ip6_datagram_recv_ctl(sk, msg, skb); 483 484 err = copied; 485 if (flags & MSG_TRUNC) 486 err = skb->len; 487 488 out_free: 489 skb_free_datagram(sk, skb); 490 out: 491 return err; 492 493 csum_copy_err: 494 skb_kill_datagram(sk, skb, flags); 495 496 /* Error for blocking case is chosen to masquerade 497 as some normal condition. 498 */ 499 err = (flags&MSG_DONTWAIT) ? -EAGAIN : -EHOSTUNREACH; 500 goto out; 501 } 502 503 static int rawv6_push_pending_frames(struct sock *sk, struct flowi6 *fl6, 504 struct raw6_sock *rp) 505 { 506 struct ipv6_txoptions *opt; 507 struct sk_buff *skb; 508 int err = 0; 509 int offset; 510 int len; 511 int total_len; 512 __wsum tmp_csum; 513 __sum16 csum; 514 515 if (!rp->checksum) 516 goto send; 517 518 skb = skb_peek(&sk->sk_write_queue); 519 if (!skb) 520 goto out; 521 522 offset = rp->offset; 523 total_len = inet_sk(sk)->cork.base.length; 524 opt = inet6_sk(sk)->cork.opt; 525 total_len -= opt ? opt->opt_flen : 0; 526 527 if (offset >= total_len - 1) { 528 err = -EINVAL; 529 ip6_flush_pending_frames(sk); 530 goto out; 531 } 532 533 /* should be check HW csum miyazawa */ 534 if (skb_queue_len(&sk->sk_write_queue) == 1) { 535 /* 536 * Only one fragment on the socket. 537 */ 538 tmp_csum = skb->csum; 539 } else { 540 struct sk_buff *csum_skb = NULL; 541 tmp_csum = 0; 542 543 skb_queue_walk(&sk->sk_write_queue, skb) { 544 tmp_csum = csum_add(tmp_csum, skb->csum); 545 546 if (csum_skb) 547 continue; 548 549 len = skb->len - skb_transport_offset(skb); 550 if (offset >= len) { 551 offset -= len; 552 continue; 553 } 554 555 csum_skb = skb; 556 } 557 558 skb = csum_skb; 559 } 560 561 offset += skb_transport_offset(skb); 562 err = skb_copy_bits(skb, offset, &csum, 2); 563 if (err < 0) { 564 ip6_flush_pending_frames(sk); 565 goto out; 566 } 567 568 /* in case cksum was not initialized */ 569 if (unlikely(csum)) 570 tmp_csum = csum_sub(tmp_csum, csum_unfold(csum)); 571 572 csum = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 573 total_len, fl6->flowi6_proto, tmp_csum); 574 575 if (csum == 0 && fl6->flowi6_proto == IPPROTO_UDP) 576 csum = CSUM_MANGLED_0; 577 578 BUG_ON(skb_store_bits(skb, offset, &csum, 2)); 579 580 send: 581 err = ip6_push_pending_frames(sk); 582 out: 583 return err; 584 } 585 586 static int rawv6_send_hdrinc(struct sock *sk, struct msghdr *msg, int length, 587 struct flowi6 *fl6, struct dst_entry **dstp, 588 unsigned int flags, const struct sockcm_cookie *sockc) 589 { 590 struct net *net = sock_net(sk); 591 struct ipv6hdr *iph; 592 struct sk_buff *skb; 593 int err; 594 struct rt6_info *rt = (struct rt6_info *)*dstp; 595 int hlen = LL_RESERVED_SPACE(rt->dst.dev); 596 int tlen = rt->dst.dev->needed_tailroom; 597 598 if (length > rt->dst.dev->mtu) { 599 ipv6_local_error(sk, EMSGSIZE, fl6, rt->dst.dev->mtu); 600 return -EMSGSIZE; 601 } 602 if (length < sizeof(struct ipv6hdr)) 603 return -EINVAL; 604 if (flags&MSG_PROBE) 605 goto out; 606 607 skb = sock_alloc_send_skb(sk, 608 length + hlen + tlen + 15, 609 flags & MSG_DONTWAIT, &err); 610 if (!skb) 611 goto error; 612 skb_reserve(skb, hlen); 613 614 skb->protocol = htons(ETH_P_IPV6); 615 skb->priority = READ_ONCE(sk->sk_priority); 616 skb->mark = sockc->mark; 617 skb->tstamp = sockc->transmit_time; 618 skb->mono_delivery_time = !!skb->tstamp; 619 skb_put(skb, length); 620 skb_reset_network_header(skb); 621 iph = ipv6_hdr(skb); 622 623 skb->ip_summed = CHECKSUM_NONE; 624 625 skb_setup_tx_timestamp(skb, sockc->tsflags); 626 627 if (flags & MSG_CONFIRM) 628 skb_set_dst_pending_confirm(skb, 1); 629 630 skb->transport_header = skb->network_header; 631 err = memcpy_from_msg(iph, msg, length); 632 if (err) { 633 err = -EFAULT; 634 kfree_skb(skb); 635 goto error; 636 } 637 638 skb_dst_set(skb, &rt->dst); 639 *dstp = NULL; 640 641 /* if egress device is enslaved to an L3 master device pass the 642 * skb to its handler for processing 643 */ 644 skb = l3mdev_ip6_out(sk, skb); 645 if (unlikely(!skb)) 646 return 0; 647 648 /* Acquire rcu_read_lock() in case we need to use rt->rt6i_idev 649 * in the error path. Since skb has been freed, the dst could 650 * have been queued for deletion. 651 */ 652 rcu_read_lock(); 653 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS); 654 err = NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, skb, 655 NULL, rt->dst.dev, dst_output); 656 if (err > 0) 657 err = net_xmit_errno(err); 658 if (err) { 659 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); 660 rcu_read_unlock(); 661 goto error_check; 662 } 663 rcu_read_unlock(); 664 out: 665 return 0; 666 667 error: 668 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); 669 error_check: 670 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) 671 err = 0; 672 return err; 673 } 674 675 struct raw6_frag_vec { 676 struct msghdr *msg; 677 int hlen; 678 char c[4]; 679 }; 680 681 static int rawv6_probe_proto_opt(struct raw6_frag_vec *rfv, struct flowi6 *fl6) 682 { 683 int err = 0; 684 switch (fl6->flowi6_proto) { 685 case IPPROTO_ICMPV6: 686 rfv->hlen = 2; 687 err = memcpy_from_msg(rfv->c, rfv->msg, rfv->hlen); 688 if (!err) { 689 fl6->fl6_icmp_type = rfv->c[0]; 690 fl6->fl6_icmp_code = rfv->c[1]; 691 } 692 break; 693 case IPPROTO_MH: 694 rfv->hlen = 4; 695 err = memcpy_from_msg(rfv->c, rfv->msg, rfv->hlen); 696 if (!err) 697 fl6->fl6_mh_type = rfv->c[2]; 698 } 699 return err; 700 } 701 702 static int raw6_getfrag(void *from, char *to, int offset, int len, int odd, 703 struct sk_buff *skb) 704 { 705 struct raw6_frag_vec *rfv = from; 706 707 if (offset < rfv->hlen) { 708 int copy = min(rfv->hlen - offset, len); 709 710 if (skb->ip_summed == CHECKSUM_PARTIAL) 711 memcpy(to, rfv->c + offset, copy); 712 else 713 skb->csum = csum_block_add( 714 skb->csum, 715 csum_partial_copy_nocheck(rfv->c + offset, 716 to, copy), 717 odd); 718 719 odd = 0; 720 offset += copy; 721 to += copy; 722 len -= copy; 723 724 if (!len) 725 return 0; 726 } 727 728 offset -= rfv->hlen; 729 730 return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb); 731 } 732 733 static int rawv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 734 { 735 struct ipv6_txoptions *opt_to_free = NULL; 736 struct ipv6_txoptions opt_space; 737 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 738 struct in6_addr *daddr, *final_p, final; 739 struct inet_sock *inet = inet_sk(sk); 740 struct ipv6_pinfo *np = inet6_sk(sk); 741 struct raw6_sock *rp = raw6_sk(sk); 742 struct ipv6_txoptions *opt = NULL; 743 struct ip6_flowlabel *flowlabel = NULL; 744 struct dst_entry *dst = NULL; 745 struct raw6_frag_vec rfv; 746 struct flowi6 fl6; 747 struct ipcm6_cookie ipc6; 748 int addr_len = msg->msg_namelen; 749 int hdrincl; 750 u16 proto; 751 int err; 752 753 /* Rough check on arithmetic overflow, 754 better check is made in ip6_append_data(). 755 */ 756 if (len > INT_MAX) 757 return -EMSGSIZE; 758 759 /* Mirror BSD error message compatibility */ 760 if (msg->msg_flags & MSG_OOB) 761 return -EOPNOTSUPP; 762 763 hdrincl = inet_test_bit(HDRINCL, sk); 764 765 /* 766 * Get and verify the address. 767 */ 768 memset(&fl6, 0, sizeof(fl6)); 769 770 fl6.flowi6_mark = READ_ONCE(sk->sk_mark); 771 fl6.flowi6_uid = sk->sk_uid; 772 773 ipcm6_init(&ipc6); 774 ipc6.sockc.tsflags = READ_ONCE(sk->sk_tsflags); 775 ipc6.sockc.mark = fl6.flowi6_mark; 776 777 if (sin6) { 778 if (addr_len < SIN6_LEN_RFC2133) 779 return -EINVAL; 780 781 if (sin6->sin6_family && sin6->sin6_family != AF_INET6) 782 return -EAFNOSUPPORT; 783 784 /* port is the proto value [0..255] carried in nexthdr */ 785 proto = ntohs(sin6->sin6_port); 786 787 if (!proto) 788 proto = inet->inet_num; 789 else if (proto != inet->inet_num && 790 inet->inet_num != IPPROTO_RAW) 791 return -EINVAL; 792 793 if (proto > 255) 794 return -EINVAL; 795 796 daddr = &sin6->sin6_addr; 797 if (inet6_test_bit(SNDFLOW, sk)) { 798 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 799 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 800 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 801 if (IS_ERR(flowlabel)) 802 return -EINVAL; 803 } 804 } 805 806 /* 807 * Otherwise it will be difficult to maintain 808 * sk->sk_dst_cache. 809 */ 810 if (sk->sk_state == TCP_ESTABLISHED && 811 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 812 daddr = &sk->sk_v6_daddr; 813 814 if (addr_len >= sizeof(struct sockaddr_in6) && 815 sin6->sin6_scope_id && 816 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 817 fl6.flowi6_oif = sin6->sin6_scope_id; 818 } else { 819 if (sk->sk_state != TCP_ESTABLISHED) 820 return -EDESTADDRREQ; 821 822 proto = inet->inet_num; 823 daddr = &sk->sk_v6_daddr; 824 fl6.flowlabel = np->flow_label; 825 } 826 827 if (fl6.flowi6_oif == 0) 828 fl6.flowi6_oif = sk->sk_bound_dev_if; 829 830 if (msg->msg_controllen) { 831 opt = &opt_space; 832 memset(opt, 0, sizeof(struct ipv6_txoptions)); 833 opt->tot_len = sizeof(struct ipv6_txoptions); 834 ipc6.opt = opt; 835 836 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6); 837 if (err < 0) { 838 fl6_sock_release(flowlabel); 839 return err; 840 } 841 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 842 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 843 if (IS_ERR(flowlabel)) 844 return -EINVAL; 845 } 846 if (!(opt->opt_nflen|opt->opt_flen)) 847 opt = NULL; 848 } 849 if (!opt) { 850 opt = txopt_get(np); 851 opt_to_free = opt; 852 } 853 if (flowlabel) 854 opt = fl6_merge_options(&opt_space, flowlabel, opt); 855 opt = ipv6_fixup_options(&opt_space, opt); 856 857 fl6.flowi6_proto = proto; 858 fl6.flowi6_mark = ipc6.sockc.mark; 859 860 if (!hdrincl) { 861 rfv.msg = msg; 862 rfv.hlen = 0; 863 err = rawv6_probe_proto_opt(&rfv, &fl6); 864 if (err) 865 goto out; 866 } 867 868 if (!ipv6_addr_any(daddr)) 869 fl6.daddr = *daddr; 870 else 871 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 872 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 873 fl6.saddr = np->saddr; 874 875 final_p = fl6_update_dst(&fl6, opt, &final); 876 877 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) 878 fl6.flowi6_oif = READ_ONCE(np->mcast_oif); 879 else if (!fl6.flowi6_oif) 880 fl6.flowi6_oif = READ_ONCE(np->ucast_oif); 881 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6)); 882 883 if (hdrincl) 884 fl6.flowi6_flags |= FLOWI_FLAG_KNOWN_NH; 885 886 if (ipc6.tclass < 0) 887 ipc6.tclass = np->tclass; 888 889 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 890 891 dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p); 892 if (IS_ERR(dst)) { 893 err = PTR_ERR(dst); 894 goto out; 895 } 896 if (ipc6.hlimit < 0) 897 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 898 899 if (ipc6.dontfrag < 0) 900 ipc6.dontfrag = inet6_test_bit(DONTFRAG, sk); 901 902 if (msg->msg_flags&MSG_CONFIRM) 903 goto do_confirm; 904 905 back_from_confirm: 906 if (hdrincl) 907 err = rawv6_send_hdrinc(sk, msg, len, &fl6, &dst, 908 msg->msg_flags, &ipc6.sockc); 909 else { 910 ipc6.opt = opt; 911 lock_sock(sk); 912 err = ip6_append_data(sk, raw6_getfrag, &rfv, 913 len, 0, &ipc6, &fl6, (struct rt6_info *)dst, 914 msg->msg_flags); 915 916 if (err) 917 ip6_flush_pending_frames(sk); 918 else if (!(msg->msg_flags & MSG_MORE)) 919 err = rawv6_push_pending_frames(sk, &fl6, rp); 920 release_sock(sk); 921 } 922 done: 923 dst_release(dst); 924 out: 925 fl6_sock_release(flowlabel); 926 txopt_put(opt_to_free); 927 return err < 0 ? err : len; 928 do_confirm: 929 if (msg->msg_flags & MSG_PROBE) 930 dst_confirm_neigh(dst, &fl6.daddr); 931 if (!(msg->msg_flags & MSG_PROBE) || len) 932 goto back_from_confirm; 933 err = 0; 934 goto done; 935 } 936 937 static int rawv6_seticmpfilter(struct sock *sk, int optname, 938 sockptr_t optval, int optlen) 939 { 940 switch (optname) { 941 case ICMPV6_FILTER: 942 if (optlen > sizeof(struct icmp6_filter)) 943 optlen = sizeof(struct icmp6_filter); 944 if (copy_from_sockptr(&raw6_sk(sk)->filter, optval, optlen)) 945 return -EFAULT; 946 return 0; 947 default: 948 return -ENOPROTOOPT; 949 } 950 951 return 0; 952 } 953 954 static int rawv6_geticmpfilter(struct sock *sk, int optname, 955 char __user *optval, int __user *optlen) 956 { 957 int len; 958 959 switch (optname) { 960 case ICMPV6_FILTER: 961 if (get_user(len, optlen)) 962 return -EFAULT; 963 if (len < 0) 964 return -EINVAL; 965 if (len > sizeof(struct icmp6_filter)) 966 len = sizeof(struct icmp6_filter); 967 if (put_user(len, optlen)) 968 return -EFAULT; 969 if (copy_to_user(optval, &raw6_sk(sk)->filter, len)) 970 return -EFAULT; 971 return 0; 972 default: 973 return -ENOPROTOOPT; 974 } 975 976 return 0; 977 } 978 979 980 static int do_rawv6_setsockopt(struct sock *sk, int level, int optname, 981 sockptr_t optval, unsigned int optlen) 982 { 983 struct raw6_sock *rp = raw6_sk(sk); 984 int val; 985 986 if (optlen < sizeof(val)) 987 return -EINVAL; 988 989 if (copy_from_sockptr(&val, optval, sizeof(val))) 990 return -EFAULT; 991 992 switch (optname) { 993 case IPV6_HDRINCL: 994 if (sk->sk_type != SOCK_RAW) 995 return -EINVAL; 996 inet_assign_bit(HDRINCL, sk, val); 997 return 0; 998 case IPV6_CHECKSUM: 999 if (inet_sk(sk)->inet_num == IPPROTO_ICMPV6 && 1000 level == IPPROTO_IPV6) { 1001 /* 1002 * RFC3542 tells that IPV6_CHECKSUM socket 1003 * option in the IPPROTO_IPV6 level is not 1004 * allowed on ICMPv6 sockets. 1005 * If you want to set it, use IPPROTO_RAW 1006 * level IPV6_CHECKSUM socket option 1007 * (Linux extension). 1008 */ 1009 return -EINVAL; 1010 } 1011 1012 /* You may get strange result with a positive odd offset; 1013 RFC2292bis agrees with me. */ 1014 if (val > 0 && (val&1)) 1015 return -EINVAL; 1016 if (val < 0) { 1017 rp->checksum = 0; 1018 } else { 1019 rp->checksum = 1; 1020 rp->offset = val; 1021 } 1022 1023 return 0; 1024 1025 default: 1026 return -ENOPROTOOPT; 1027 } 1028 } 1029 1030 static int rawv6_setsockopt(struct sock *sk, int level, int optname, 1031 sockptr_t optval, unsigned int optlen) 1032 { 1033 switch (level) { 1034 case SOL_RAW: 1035 break; 1036 1037 case SOL_ICMPV6: 1038 if (inet_sk(sk)->inet_num != IPPROTO_ICMPV6) 1039 return -EOPNOTSUPP; 1040 return rawv6_seticmpfilter(sk, optname, optval, optlen); 1041 case SOL_IPV6: 1042 if (optname == IPV6_CHECKSUM || 1043 optname == IPV6_HDRINCL) 1044 break; 1045 fallthrough; 1046 default: 1047 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1048 } 1049 1050 return do_rawv6_setsockopt(sk, level, optname, optval, optlen); 1051 } 1052 1053 static int do_rawv6_getsockopt(struct sock *sk, int level, int optname, 1054 char __user *optval, int __user *optlen) 1055 { 1056 struct raw6_sock *rp = raw6_sk(sk); 1057 int val, len; 1058 1059 if (get_user(len, optlen)) 1060 return -EFAULT; 1061 1062 switch (optname) { 1063 case IPV6_HDRINCL: 1064 val = inet_test_bit(HDRINCL, sk); 1065 break; 1066 case IPV6_CHECKSUM: 1067 /* 1068 * We allow getsockopt() for IPPROTO_IPV6-level 1069 * IPV6_CHECKSUM socket option on ICMPv6 sockets 1070 * since RFC3542 is silent about it. 1071 */ 1072 if (rp->checksum == 0) 1073 val = -1; 1074 else 1075 val = rp->offset; 1076 break; 1077 1078 default: 1079 return -ENOPROTOOPT; 1080 } 1081 1082 len = min_t(unsigned int, sizeof(int), len); 1083 1084 if (put_user(len, optlen)) 1085 return -EFAULT; 1086 if (copy_to_user(optval, &val, len)) 1087 return -EFAULT; 1088 return 0; 1089 } 1090 1091 static int rawv6_getsockopt(struct sock *sk, int level, int optname, 1092 char __user *optval, int __user *optlen) 1093 { 1094 switch (level) { 1095 case SOL_RAW: 1096 break; 1097 1098 case SOL_ICMPV6: 1099 if (inet_sk(sk)->inet_num != IPPROTO_ICMPV6) 1100 return -EOPNOTSUPP; 1101 return rawv6_geticmpfilter(sk, optname, optval, optlen); 1102 case SOL_IPV6: 1103 if (optname == IPV6_CHECKSUM || 1104 optname == IPV6_HDRINCL) 1105 break; 1106 fallthrough; 1107 default: 1108 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1109 } 1110 1111 return do_rawv6_getsockopt(sk, level, optname, optval, optlen); 1112 } 1113 1114 static int rawv6_ioctl(struct sock *sk, int cmd, int *karg) 1115 { 1116 switch (cmd) { 1117 case SIOCOUTQ: { 1118 *karg = sk_wmem_alloc_get(sk); 1119 return 0; 1120 } 1121 case SIOCINQ: { 1122 struct sk_buff *skb; 1123 1124 spin_lock_bh(&sk->sk_receive_queue.lock); 1125 skb = skb_peek(&sk->sk_receive_queue); 1126 if (skb) 1127 *karg = skb->len; 1128 else 1129 *karg = 0; 1130 spin_unlock_bh(&sk->sk_receive_queue.lock); 1131 return 0; 1132 } 1133 1134 default: 1135 #ifdef CONFIG_IPV6_MROUTE 1136 return ip6mr_ioctl(sk, cmd, karg); 1137 #else 1138 return -ENOIOCTLCMD; 1139 #endif 1140 } 1141 } 1142 1143 #ifdef CONFIG_COMPAT 1144 static int compat_rawv6_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg) 1145 { 1146 switch (cmd) { 1147 case SIOCOUTQ: 1148 case SIOCINQ: 1149 return -ENOIOCTLCMD; 1150 default: 1151 #ifdef CONFIG_IPV6_MROUTE 1152 return ip6mr_compat_ioctl(sk, cmd, compat_ptr(arg)); 1153 #else 1154 return -ENOIOCTLCMD; 1155 #endif 1156 } 1157 } 1158 #endif 1159 1160 static void rawv6_close(struct sock *sk, long timeout) 1161 { 1162 if (inet_sk(sk)->inet_num == IPPROTO_RAW) 1163 ip6_ra_control(sk, -1); 1164 ip6mr_sk_done(sk); 1165 sk_common_release(sk); 1166 } 1167 1168 static void raw6_destroy(struct sock *sk) 1169 { 1170 lock_sock(sk); 1171 ip6_flush_pending_frames(sk); 1172 release_sock(sk); 1173 } 1174 1175 static int rawv6_init_sk(struct sock *sk) 1176 { 1177 struct raw6_sock *rp = raw6_sk(sk); 1178 1179 switch (inet_sk(sk)->inet_num) { 1180 case IPPROTO_ICMPV6: 1181 rp->checksum = 1; 1182 rp->offset = 2; 1183 break; 1184 case IPPROTO_MH: 1185 rp->checksum = 1; 1186 rp->offset = 4; 1187 break; 1188 default: 1189 break; 1190 } 1191 return 0; 1192 } 1193 1194 struct proto rawv6_prot = { 1195 .name = "RAWv6", 1196 .owner = THIS_MODULE, 1197 .close = rawv6_close, 1198 .destroy = raw6_destroy, 1199 .connect = ip6_datagram_connect_v6_only, 1200 .disconnect = __udp_disconnect, 1201 .ioctl = rawv6_ioctl, 1202 .init = rawv6_init_sk, 1203 .setsockopt = rawv6_setsockopt, 1204 .getsockopt = rawv6_getsockopt, 1205 .sendmsg = rawv6_sendmsg, 1206 .recvmsg = rawv6_recvmsg, 1207 .bind = rawv6_bind, 1208 .backlog_rcv = rawv6_rcv_skb, 1209 .hash = raw_hash_sk, 1210 .unhash = raw_unhash_sk, 1211 .obj_size = sizeof(struct raw6_sock), 1212 .ipv6_pinfo_offset = offsetof(struct raw6_sock, inet6), 1213 .useroffset = offsetof(struct raw6_sock, filter), 1214 .usersize = sizeof_field(struct raw6_sock, filter), 1215 .h.raw_hash = &raw_v6_hashinfo, 1216 #ifdef CONFIG_COMPAT 1217 .compat_ioctl = compat_rawv6_ioctl, 1218 #endif 1219 .diag_destroy = raw_abort, 1220 }; 1221 1222 #ifdef CONFIG_PROC_FS 1223 static int raw6_seq_show(struct seq_file *seq, void *v) 1224 { 1225 if (v == SEQ_START_TOKEN) { 1226 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1227 } else { 1228 struct sock *sp = v; 1229 __u16 srcp = inet_sk(sp)->inet_num; 1230 ip6_dgram_sock_seq_show(seq, v, srcp, 0, 1231 raw_seq_private(seq)->bucket); 1232 } 1233 return 0; 1234 } 1235 1236 static const struct seq_operations raw6_seq_ops = { 1237 .start = raw_seq_start, 1238 .next = raw_seq_next, 1239 .stop = raw_seq_stop, 1240 .show = raw6_seq_show, 1241 }; 1242 1243 static int __net_init raw6_init_net(struct net *net) 1244 { 1245 if (!proc_create_net_data("raw6", 0444, net->proc_net, &raw6_seq_ops, 1246 sizeof(struct raw_iter_state), &raw_v6_hashinfo)) 1247 return -ENOMEM; 1248 1249 return 0; 1250 } 1251 1252 static void __net_exit raw6_exit_net(struct net *net) 1253 { 1254 remove_proc_entry("raw6", net->proc_net); 1255 } 1256 1257 static struct pernet_operations raw6_net_ops = { 1258 .init = raw6_init_net, 1259 .exit = raw6_exit_net, 1260 }; 1261 1262 int __init raw6_proc_init(void) 1263 { 1264 return register_pernet_subsys(&raw6_net_ops); 1265 } 1266 1267 void raw6_proc_exit(void) 1268 { 1269 unregister_pernet_subsys(&raw6_net_ops); 1270 } 1271 #endif /* CONFIG_PROC_FS */ 1272 1273 /* Same as inet6_dgram_ops, sans udp_poll. */ 1274 const struct proto_ops inet6_sockraw_ops = { 1275 .family = PF_INET6, 1276 .owner = THIS_MODULE, 1277 .release = inet6_release, 1278 .bind = inet6_bind, 1279 .connect = inet_dgram_connect, /* ok */ 1280 .socketpair = sock_no_socketpair, /* a do nothing */ 1281 .accept = sock_no_accept, /* a do nothing */ 1282 .getname = inet6_getname, 1283 .poll = datagram_poll, /* ok */ 1284 .ioctl = inet6_ioctl, /* must change */ 1285 .gettstamp = sock_gettstamp, 1286 .listen = sock_no_listen, /* ok */ 1287 .shutdown = inet_shutdown, /* ok */ 1288 .setsockopt = sock_common_setsockopt, /* ok */ 1289 .getsockopt = sock_common_getsockopt, /* ok */ 1290 .sendmsg = inet_sendmsg, /* ok */ 1291 .recvmsg = sock_common_recvmsg, /* ok */ 1292 .mmap = sock_no_mmap, 1293 #ifdef CONFIG_COMPAT 1294 .compat_ioctl = inet6_compat_ioctl, 1295 #endif 1296 }; 1297 1298 static struct inet_protosw rawv6_protosw = { 1299 .type = SOCK_RAW, 1300 .protocol = IPPROTO_IP, /* wild card */ 1301 .prot = &rawv6_prot, 1302 .ops = &inet6_sockraw_ops, 1303 .flags = INET_PROTOSW_REUSE, 1304 }; 1305 1306 int __init rawv6_init(void) 1307 { 1308 return inet6_register_protosw(&rawv6_protosw); 1309 } 1310 1311 void rawv6_exit(void) 1312 { 1313 inet6_unregister_protosw(&rawv6_protosw); 1314 } 1315