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, 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 struct hlist_nulls_head *hlist; 145 struct hlist_nulls_node *hnode; 146 const struct in6_addr *saddr; 147 const struct in6_addr *daddr; 148 struct sock *sk; 149 bool delivered = false; 150 __u8 hash; 151 152 saddr = &ipv6_hdr(skb)->saddr; 153 daddr = saddr + 1; 154 155 hash = nexthdr & (RAW_HTABLE_SIZE - 1); 156 hlist = &raw_v6_hashinfo.ht[hash]; 157 rcu_read_lock(); 158 sk_nulls_for_each(sk, hnode, hlist) { 159 int filtered; 160 161 if (!raw_v6_match(net, sk, nexthdr, daddr, saddr, 162 inet6_iif(skb), inet6_sdif(skb))) 163 continue; 164 delivered = true; 165 switch (nexthdr) { 166 case IPPROTO_ICMPV6: 167 filtered = icmpv6_filter(sk, skb); 168 break; 169 170 #if IS_ENABLED(CONFIG_IPV6_MIP6) 171 case IPPROTO_MH: 172 { 173 /* XXX: To validate MH only once for each packet, 174 * this is placed here. It should be after checking 175 * xfrm policy, however it doesn't. The checking xfrm 176 * policy is placed in rawv6_rcv() because it is 177 * required for each socket. 178 */ 179 mh_filter_t *filter; 180 181 filter = rcu_dereference(mh_filter); 182 filtered = filter ? (*filter)(sk, skb) : 0; 183 break; 184 } 185 #endif 186 default: 187 filtered = 0; 188 break; 189 } 190 191 if (filtered < 0) 192 break; 193 if (filtered == 0) { 194 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC); 195 196 /* Not releasing hash table! */ 197 if (clone) { 198 nf_reset_ct(clone); 199 rawv6_rcv(sk, clone); 200 } 201 } 202 } 203 rcu_read_unlock(); 204 return delivered; 205 } 206 207 bool raw6_local_deliver(struct sk_buff *skb, int nexthdr) 208 { 209 return ipv6_raw_deliver(skb, nexthdr); 210 } 211 212 /* This cleans up af_inet6 a bit. -DaveM */ 213 static int rawv6_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len) 214 { 215 struct inet_sock *inet = inet_sk(sk); 216 struct ipv6_pinfo *np = inet6_sk(sk); 217 struct sockaddr_in6 *addr = (struct sockaddr_in6 *) uaddr; 218 __be32 v4addr = 0; 219 int addr_type; 220 int err; 221 222 if (addr_len < SIN6_LEN_RFC2133) 223 return -EINVAL; 224 225 if (addr->sin6_family != AF_INET6) 226 return -EINVAL; 227 228 addr_type = ipv6_addr_type(&addr->sin6_addr); 229 230 /* Raw sockets are IPv6 only */ 231 if (addr_type == IPV6_ADDR_MAPPED) 232 return -EADDRNOTAVAIL; 233 234 lock_sock(sk); 235 236 err = -EINVAL; 237 if (sk->sk_state != TCP_CLOSE) 238 goto out; 239 240 rcu_read_lock(); 241 /* Check if the address belongs to the host. */ 242 if (addr_type != IPV6_ADDR_ANY) { 243 struct net_device *dev = NULL; 244 245 if (__ipv6_addr_needs_scope_id(addr_type)) { 246 if (addr_len >= sizeof(struct sockaddr_in6) && 247 addr->sin6_scope_id) { 248 /* Override any existing binding, if another 249 * one is supplied by user. 250 */ 251 sk->sk_bound_dev_if = addr->sin6_scope_id; 252 } 253 254 /* Binding to link-local address requires an interface */ 255 if (!sk->sk_bound_dev_if) 256 goto out_unlock; 257 } 258 259 if (sk->sk_bound_dev_if) { 260 err = -ENODEV; 261 dev = dev_get_by_index_rcu(sock_net(sk), 262 sk->sk_bound_dev_if); 263 if (!dev) 264 goto out_unlock; 265 } 266 267 /* ipv4 addr of the socket is invalid. Only the 268 * unspecified and mapped address have a v4 equivalent. 269 */ 270 v4addr = LOOPBACK4_IPV6; 271 if (!(addr_type & IPV6_ADDR_MULTICAST) && 272 !ipv6_can_nonlocal_bind(sock_net(sk), inet)) { 273 err = -EADDRNOTAVAIL; 274 if (!ipv6_chk_addr(sock_net(sk), &addr->sin6_addr, 275 dev, 0)) { 276 goto out_unlock; 277 } 278 } 279 } 280 281 inet->inet_rcv_saddr = inet->inet_saddr = v4addr; 282 sk->sk_v6_rcv_saddr = addr->sin6_addr; 283 if (!(addr_type & IPV6_ADDR_MULTICAST)) 284 np->saddr = addr->sin6_addr; 285 err = 0; 286 out_unlock: 287 rcu_read_unlock(); 288 out: 289 release_sock(sk); 290 return err; 291 } 292 293 static void rawv6_err(struct sock *sk, struct sk_buff *skb, 294 struct inet6_skb_parm *opt, 295 u8 type, u8 code, int offset, __be32 info) 296 { 297 struct inet_sock *inet = inet_sk(sk); 298 struct ipv6_pinfo *np = inet6_sk(sk); 299 int err; 300 int harderr; 301 302 /* Report error on raw socket, if: 303 1. User requested recverr. 304 2. Socket is connected (otherwise the error indication 305 is useless without recverr and error is hard. 306 */ 307 if (!np->recverr && sk->sk_state != TCP_ESTABLISHED) 308 return; 309 310 harderr = icmpv6_err_convert(type, code, &err); 311 if (type == ICMPV6_PKT_TOOBIG) { 312 ip6_sk_update_pmtu(skb, sk, info); 313 harderr = (np->pmtudisc == IPV6_PMTUDISC_DO); 314 } 315 if (type == NDISC_REDIRECT) { 316 ip6_sk_redirect(skb, sk); 317 return; 318 } 319 if (np->recverr) { 320 u8 *payload = skb->data; 321 if (!inet->hdrincl) 322 payload += offset; 323 ipv6_icmp_error(sk, skb, err, 0, ntohl(info), payload); 324 } 325 326 if (np->recverr || harderr) { 327 sk->sk_err = err; 328 sk_error_report(sk); 329 } 330 } 331 332 void raw6_icmp_error(struct sk_buff *skb, int nexthdr, 333 u8 type, u8 code, int inner_offset, __be32 info) 334 { 335 struct net *net = dev_net(skb->dev); 336 struct hlist_nulls_head *hlist; 337 struct hlist_nulls_node *hnode; 338 struct sock *sk; 339 int hash; 340 341 hash = nexthdr & (RAW_HTABLE_SIZE - 1); 342 hlist = &raw_v6_hashinfo.ht[hash]; 343 rcu_read_lock(); 344 sk_nulls_for_each(sk, hnode, hlist) { 345 /* Note: ipv6_hdr(skb) != skb->data */ 346 const struct ipv6hdr *ip6h = (const struct ipv6hdr *)skb->data; 347 348 if (!raw_v6_match(net, sk, nexthdr, &ip6h->saddr, &ip6h->daddr, 349 inet6_iif(skb), inet6_iif(skb))) 350 continue; 351 rawv6_err(sk, skb, NULL, type, code, inner_offset, info); 352 } 353 rcu_read_unlock(); 354 } 355 356 static inline int rawv6_rcv_skb(struct sock *sk, struct sk_buff *skb) 357 { 358 if ((raw6_sk(sk)->checksum || rcu_access_pointer(sk->sk_filter)) && 359 skb_checksum_complete(skb)) { 360 atomic_inc(&sk->sk_drops); 361 kfree_skb(skb); 362 return NET_RX_DROP; 363 } 364 365 /* Charge it to the socket. */ 366 skb_dst_drop(skb); 367 if (sock_queue_rcv_skb(sk, skb) < 0) { 368 kfree_skb(skb); 369 return NET_RX_DROP; 370 } 371 372 return 0; 373 } 374 375 /* 376 * This is next to useless... 377 * if we demultiplex in network layer we don't need the extra call 378 * just to queue the skb... 379 * maybe we could have the network decide upon a hint if it 380 * should call raw_rcv for demultiplexing 381 */ 382 int rawv6_rcv(struct sock *sk, struct sk_buff *skb) 383 { 384 struct inet_sock *inet = inet_sk(sk); 385 struct raw6_sock *rp = raw6_sk(sk); 386 387 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 388 atomic_inc(&sk->sk_drops); 389 kfree_skb(skb); 390 return NET_RX_DROP; 391 } 392 393 if (!rp->checksum) 394 skb->ip_summed = CHECKSUM_UNNECESSARY; 395 396 if (skb->ip_summed == CHECKSUM_COMPLETE) { 397 skb_postpull_rcsum(skb, skb_network_header(skb), 398 skb_network_header_len(skb)); 399 if (!csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 400 &ipv6_hdr(skb)->daddr, 401 skb->len, inet->inet_num, skb->csum)) 402 skb->ip_summed = CHECKSUM_UNNECESSARY; 403 } 404 if (!skb_csum_unnecessary(skb)) 405 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 406 &ipv6_hdr(skb)->daddr, 407 skb->len, 408 inet->inet_num, 0)); 409 410 if (inet->hdrincl) { 411 if (skb_checksum_complete(skb)) { 412 atomic_inc(&sk->sk_drops); 413 kfree_skb(skb); 414 return NET_RX_DROP; 415 } 416 } 417 418 rawv6_rcv_skb(sk, skb); 419 return 0; 420 } 421 422 423 /* 424 * This should be easy, if there is something there 425 * we return it, otherwise we block. 426 */ 427 428 static int rawv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 429 int flags, int *addr_len) 430 { 431 struct ipv6_pinfo *np = inet6_sk(sk); 432 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 433 struct sk_buff *skb; 434 size_t copied; 435 int err; 436 437 if (flags & MSG_OOB) 438 return -EOPNOTSUPP; 439 440 if (flags & MSG_ERRQUEUE) 441 return ipv6_recv_error(sk, msg, len, addr_len); 442 443 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 444 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 445 446 skb = skb_recv_datagram(sk, flags, &err); 447 if (!skb) 448 goto out; 449 450 copied = skb->len; 451 if (copied > len) { 452 copied = len; 453 msg->msg_flags |= MSG_TRUNC; 454 } 455 456 if (skb_csum_unnecessary(skb)) { 457 err = skb_copy_datagram_msg(skb, 0, msg, copied); 458 } else if (msg->msg_flags&MSG_TRUNC) { 459 if (__skb_checksum_complete(skb)) 460 goto csum_copy_err; 461 err = skb_copy_datagram_msg(skb, 0, msg, copied); 462 } else { 463 err = skb_copy_and_csum_datagram_msg(skb, 0, msg); 464 if (err == -EINVAL) 465 goto csum_copy_err; 466 } 467 if (err) 468 goto out_free; 469 470 /* Copy the address. */ 471 if (sin6) { 472 sin6->sin6_family = AF_INET6; 473 sin6->sin6_port = 0; 474 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 475 sin6->sin6_flowinfo = 0; 476 sin6->sin6_scope_id = ipv6_iface_scope_id(&sin6->sin6_addr, 477 inet6_iif(skb)); 478 *addr_len = sizeof(*sin6); 479 } 480 481 sock_recv_cmsgs(msg, sk, skb); 482 483 if (np->rxopt.all) 484 ip6_datagram_recv_ctl(sk, msg, skb); 485 486 err = copied; 487 if (flags & MSG_TRUNC) 488 err = skb->len; 489 490 out_free: 491 skb_free_datagram(sk, skb); 492 out: 493 return err; 494 495 csum_copy_err: 496 skb_kill_datagram(sk, skb, flags); 497 498 /* Error for blocking case is chosen to masquerade 499 as some normal condition. 500 */ 501 err = (flags&MSG_DONTWAIT) ? -EAGAIN : -EHOSTUNREACH; 502 goto out; 503 } 504 505 static int rawv6_push_pending_frames(struct sock *sk, struct flowi6 *fl6, 506 struct raw6_sock *rp) 507 { 508 struct sk_buff *skb; 509 int err = 0; 510 int offset; 511 int len; 512 int total_len; 513 __wsum tmp_csum; 514 __sum16 csum; 515 516 if (!rp->checksum) 517 goto send; 518 519 skb = skb_peek(&sk->sk_write_queue); 520 if (!skb) 521 goto out; 522 523 offset = rp->offset; 524 total_len = inet_sk(sk)->cork.base.length; 525 if (offset >= total_len - 1) { 526 err = -EINVAL; 527 ip6_flush_pending_frames(sk); 528 goto out; 529 } 530 531 /* should be check HW csum miyazawa */ 532 if (skb_queue_len(&sk->sk_write_queue) == 1) { 533 /* 534 * Only one fragment on the socket. 535 */ 536 tmp_csum = skb->csum; 537 } else { 538 struct sk_buff *csum_skb = NULL; 539 tmp_csum = 0; 540 541 skb_queue_walk(&sk->sk_write_queue, skb) { 542 tmp_csum = csum_add(tmp_csum, skb->csum); 543 544 if (csum_skb) 545 continue; 546 547 len = skb->len - skb_transport_offset(skb); 548 if (offset >= len) { 549 offset -= len; 550 continue; 551 } 552 553 csum_skb = skb; 554 } 555 556 skb = csum_skb; 557 } 558 559 offset += skb_transport_offset(skb); 560 err = skb_copy_bits(skb, offset, &csum, 2); 561 if (err < 0) { 562 ip6_flush_pending_frames(sk); 563 goto out; 564 } 565 566 /* in case cksum was not initialized */ 567 if (unlikely(csum)) 568 tmp_csum = csum_sub(tmp_csum, csum_unfold(csum)); 569 570 csum = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 571 total_len, fl6->flowi6_proto, tmp_csum); 572 573 if (csum == 0 && fl6->flowi6_proto == IPPROTO_UDP) 574 csum = CSUM_MANGLED_0; 575 576 BUG_ON(skb_store_bits(skb, offset, &csum, 2)); 577 578 send: 579 err = ip6_push_pending_frames(sk); 580 out: 581 return err; 582 } 583 584 static int rawv6_send_hdrinc(struct sock *sk, struct msghdr *msg, int length, 585 struct flowi6 *fl6, struct dst_entry **dstp, 586 unsigned int flags, const struct sockcm_cookie *sockc) 587 { 588 struct ipv6_pinfo *np = inet6_sk(sk); 589 struct net *net = sock_net(sk); 590 struct ipv6hdr *iph; 591 struct sk_buff *skb; 592 int err; 593 struct rt6_info *rt = (struct rt6_info *)*dstp; 594 int hlen = LL_RESERVED_SPACE(rt->dst.dev); 595 int tlen = rt->dst.dev->needed_tailroom; 596 597 if (length > rt->dst.dev->mtu) { 598 ipv6_local_error(sk, EMSGSIZE, fl6, rt->dst.dev->mtu); 599 return -EMSGSIZE; 600 } 601 if (length < sizeof(struct ipv6hdr)) 602 return -EINVAL; 603 if (flags&MSG_PROBE) 604 goto out; 605 606 skb = sock_alloc_send_skb(sk, 607 length + hlen + tlen + 15, 608 flags & MSG_DONTWAIT, &err); 609 if (!skb) 610 goto error; 611 skb_reserve(skb, hlen); 612 613 skb->protocol = htons(ETH_P_IPV6); 614 skb->priority = sk->sk_priority; 615 skb->mark = sockc->mark; 616 skb->tstamp = sockc->transmit_time; 617 618 skb_put(skb, length); 619 skb_reset_network_header(skb); 620 iph = ipv6_hdr(skb); 621 622 skb->ip_summed = CHECKSUM_NONE; 623 624 skb_setup_tx_timestamp(skb, sockc->tsflags); 625 626 if (flags & MSG_CONFIRM) 627 skb_set_dst_pending_confirm(skb, 1); 628 629 skb->transport_header = skb->network_header; 630 err = memcpy_from_msg(iph, msg, length); 631 if (err) { 632 err = -EFAULT; 633 kfree_skb(skb); 634 goto error; 635 } 636 637 skb_dst_set(skb, &rt->dst); 638 *dstp = NULL; 639 640 /* if egress device is enslaved to an L3 master device pass the 641 * skb to its handler for processing 642 */ 643 skb = l3mdev_ip6_out(sk, skb); 644 if (unlikely(!skb)) 645 return 0; 646 647 /* Acquire rcu_read_lock() in case we need to use rt->rt6i_idev 648 * in the error path. Since skb has been freed, the dst could 649 * have been queued for deletion. 650 */ 651 rcu_read_lock(); 652 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len); 653 err = NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, skb, 654 NULL, rt->dst.dev, dst_output); 655 if (err > 0) 656 err = net_xmit_errno(err); 657 if (err) { 658 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); 659 rcu_read_unlock(); 660 goto error_check; 661 } 662 rcu_read_unlock(); 663 out: 664 return 0; 665 666 error: 667 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); 668 error_check: 669 if (err == -ENOBUFS && !np->recverr) 670 err = 0; 671 return err; 672 } 673 674 struct raw6_frag_vec { 675 struct msghdr *msg; 676 int hlen; 677 char c[4]; 678 }; 679 680 static int rawv6_probe_proto_opt(struct raw6_frag_vec *rfv, struct flowi6 *fl6) 681 { 682 int err = 0; 683 switch (fl6->flowi6_proto) { 684 case IPPROTO_ICMPV6: 685 rfv->hlen = 2; 686 err = memcpy_from_msg(rfv->c, rfv->msg, rfv->hlen); 687 if (!err) { 688 fl6->fl6_icmp_type = rfv->c[0]; 689 fl6->fl6_icmp_code = rfv->c[1]; 690 } 691 break; 692 case IPPROTO_MH: 693 rfv->hlen = 4; 694 err = memcpy_from_msg(rfv->c, rfv->msg, rfv->hlen); 695 if (!err) 696 fl6->fl6_mh_type = rfv->c[2]; 697 } 698 return err; 699 } 700 701 static int raw6_getfrag(void *from, char *to, int offset, int len, int odd, 702 struct sk_buff *skb) 703 { 704 struct raw6_frag_vec *rfv = from; 705 706 if (offset < rfv->hlen) { 707 int copy = min(rfv->hlen - offset, len); 708 709 if (skb->ip_summed == CHECKSUM_PARTIAL) 710 memcpy(to, rfv->c + offset, copy); 711 else 712 skb->csum = csum_block_add( 713 skb->csum, 714 csum_partial_copy_nocheck(rfv->c + offset, 715 to, copy), 716 odd); 717 718 odd = 0; 719 offset += copy; 720 to += copy; 721 len -= copy; 722 723 if (!len) 724 return 0; 725 } 726 727 offset -= rfv->hlen; 728 729 return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb); 730 } 731 732 static int rawv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 733 { 734 struct ipv6_txoptions *opt_to_free = NULL; 735 struct ipv6_txoptions opt_space; 736 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 737 struct in6_addr *daddr, *final_p, final; 738 struct inet_sock *inet = inet_sk(sk); 739 struct ipv6_pinfo *np = inet6_sk(sk); 740 struct raw6_sock *rp = raw6_sk(sk); 741 struct ipv6_txoptions *opt = NULL; 742 struct ip6_flowlabel *flowlabel = NULL; 743 struct dst_entry *dst = NULL; 744 struct raw6_frag_vec rfv; 745 struct flowi6 fl6; 746 struct ipcm6_cookie ipc6; 747 int addr_len = msg->msg_namelen; 748 int hdrincl; 749 u16 proto; 750 int err; 751 752 /* Rough check on arithmetic overflow, 753 better check is made in ip6_append_data(). 754 */ 755 if (len > INT_MAX) 756 return -EMSGSIZE; 757 758 /* Mirror BSD error message compatibility */ 759 if (msg->msg_flags & MSG_OOB) 760 return -EOPNOTSUPP; 761 762 /* hdrincl should be READ_ONCE(inet->hdrincl) 763 * but READ_ONCE() doesn't work with bit fields. 764 * Doing this indirectly yields the same result. 765 */ 766 hdrincl = inet->hdrincl; 767 hdrincl = READ_ONCE(hdrincl); 768 769 /* 770 * Get and verify the address. 771 */ 772 memset(&fl6, 0, sizeof(fl6)); 773 774 fl6.flowi6_mark = sk->sk_mark; 775 fl6.flowi6_uid = sk->sk_uid; 776 777 ipcm6_init(&ipc6); 778 ipc6.sockc.tsflags = sk->sk_tsflags; 779 ipc6.sockc.mark = sk->sk_mark; 780 781 if (sin6) { 782 if (addr_len < SIN6_LEN_RFC2133) 783 return -EINVAL; 784 785 if (sin6->sin6_family && sin6->sin6_family != AF_INET6) 786 return -EAFNOSUPPORT; 787 788 /* port is the proto value [0..255] carried in nexthdr */ 789 proto = ntohs(sin6->sin6_port); 790 791 if (!proto) 792 proto = inet->inet_num; 793 else if (proto != inet->inet_num) 794 return -EINVAL; 795 796 if (proto > 255) 797 return -EINVAL; 798 799 daddr = &sin6->sin6_addr; 800 if (np->sndflow) { 801 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 802 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 803 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 804 if (IS_ERR(flowlabel)) 805 return -EINVAL; 806 } 807 } 808 809 /* 810 * Otherwise it will be difficult to maintain 811 * sk->sk_dst_cache. 812 */ 813 if (sk->sk_state == TCP_ESTABLISHED && 814 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 815 daddr = &sk->sk_v6_daddr; 816 817 if (addr_len >= sizeof(struct sockaddr_in6) && 818 sin6->sin6_scope_id && 819 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 820 fl6.flowi6_oif = sin6->sin6_scope_id; 821 } else { 822 if (sk->sk_state != TCP_ESTABLISHED) 823 return -EDESTADDRREQ; 824 825 proto = inet->inet_num; 826 daddr = &sk->sk_v6_daddr; 827 fl6.flowlabel = np->flow_label; 828 } 829 830 if (fl6.flowi6_oif == 0) 831 fl6.flowi6_oif = sk->sk_bound_dev_if; 832 833 if (msg->msg_controllen) { 834 opt = &opt_space; 835 memset(opt, 0, sizeof(struct ipv6_txoptions)); 836 opt->tot_len = sizeof(struct ipv6_txoptions); 837 ipc6.opt = opt; 838 839 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6); 840 if (err < 0) { 841 fl6_sock_release(flowlabel); 842 return err; 843 } 844 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 845 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 846 if (IS_ERR(flowlabel)) 847 return -EINVAL; 848 } 849 if (!(opt->opt_nflen|opt->opt_flen)) 850 opt = NULL; 851 } 852 if (!opt) { 853 opt = txopt_get(np); 854 opt_to_free = opt; 855 } 856 if (flowlabel) 857 opt = fl6_merge_options(&opt_space, flowlabel, opt); 858 opt = ipv6_fixup_options(&opt_space, opt); 859 860 fl6.flowi6_proto = proto; 861 fl6.flowi6_mark = ipc6.sockc.mark; 862 863 if (!hdrincl) { 864 rfv.msg = msg; 865 rfv.hlen = 0; 866 err = rawv6_probe_proto_opt(&rfv, &fl6); 867 if (err) 868 goto out; 869 } 870 871 if (!ipv6_addr_any(daddr)) 872 fl6.daddr = *daddr; 873 else 874 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 875 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 876 fl6.saddr = np->saddr; 877 878 final_p = fl6_update_dst(&fl6, opt, &final); 879 880 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) 881 fl6.flowi6_oif = np->mcast_oif; 882 else if (!fl6.flowi6_oif) 883 fl6.flowi6_oif = np->ucast_oif; 884 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6)); 885 886 if (hdrincl) 887 fl6.flowi6_flags |= FLOWI_FLAG_KNOWN_NH; 888 889 if (ipc6.tclass < 0) 890 ipc6.tclass = np->tclass; 891 892 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 893 894 dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p); 895 if (IS_ERR(dst)) { 896 err = PTR_ERR(dst); 897 goto out; 898 } 899 if (ipc6.hlimit < 0) 900 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 901 902 if (ipc6.dontfrag < 0) 903 ipc6.dontfrag = np->dontfrag; 904 905 if (msg->msg_flags&MSG_CONFIRM) 906 goto do_confirm; 907 908 back_from_confirm: 909 if (hdrincl) 910 err = rawv6_send_hdrinc(sk, msg, len, &fl6, &dst, 911 msg->msg_flags, &ipc6.sockc); 912 else { 913 ipc6.opt = opt; 914 lock_sock(sk); 915 err = ip6_append_data(sk, raw6_getfrag, &rfv, 916 len, 0, &ipc6, &fl6, (struct rt6_info *)dst, 917 msg->msg_flags); 918 919 if (err) 920 ip6_flush_pending_frames(sk); 921 else if (!(msg->msg_flags & MSG_MORE)) 922 err = rawv6_push_pending_frames(sk, &fl6, rp); 923 release_sock(sk); 924 } 925 done: 926 dst_release(dst); 927 out: 928 fl6_sock_release(flowlabel); 929 txopt_put(opt_to_free); 930 return err < 0 ? err : len; 931 do_confirm: 932 if (msg->msg_flags & MSG_PROBE) 933 dst_confirm_neigh(dst, &fl6.daddr); 934 if (!(msg->msg_flags & MSG_PROBE) || len) 935 goto back_from_confirm; 936 err = 0; 937 goto done; 938 } 939 940 static int rawv6_seticmpfilter(struct sock *sk, int level, int optname, 941 sockptr_t optval, int optlen) 942 { 943 switch (optname) { 944 case ICMPV6_FILTER: 945 if (optlen > sizeof(struct icmp6_filter)) 946 optlen = sizeof(struct icmp6_filter); 947 if (copy_from_sockptr(&raw6_sk(sk)->filter, optval, optlen)) 948 return -EFAULT; 949 return 0; 950 default: 951 return -ENOPROTOOPT; 952 } 953 954 return 0; 955 } 956 957 static int rawv6_geticmpfilter(struct sock *sk, int level, int optname, 958 char __user *optval, int __user *optlen) 959 { 960 int len; 961 962 switch (optname) { 963 case ICMPV6_FILTER: 964 if (get_user(len, optlen)) 965 return -EFAULT; 966 if (len < 0) 967 return -EINVAL; 968 if (len > sizeof(struct icmp6_filter)) 969 len = sizeof(struct icmp6_filter); 970 if (put_user(len, optlen)) 971 return -EFAULT; 972 if (copy_to_user(optval, &raw6_sk(sk)->filter, len)) 973 return -EFAULT; 974 return 0; 975 default: 976 return -ENOPROTOOPT; 977 } 978 979 return 0; 980 } 981 982 983 static int do_rawv6_setsockopt(struct sock *sk, int level, int optname, 984 sockptr_t optval, unsigned int optlen) 985 { 986 struct raw6_sock *rp = raw6_sk(sk); 987 int val; 988 989 if (optlen < sizeof(val)) 990 return -EINVAL; 991 992 if (copy_from_sockptr(&val, optval, sizeof(val))) 993 return -EFAULT; 994 995 switch (optname) { 996 case IPV6_HDRINCL: 997 if (sk->sk_type != SOCK_RAW) 998 return -EINVAL; 999 inet_sk(sk)->hdrincl = !!val; 1000 return 0; 1001 case IPV6_CHECKSUM: 1002 if (inet_sk(sk)->inet_num == IPPROTO_ICMPV6 && 1003 level == IPPROTO_IPV6) { 1004 /* 1005 * RFC3542 tells that IPV6_CHECKSUM socket 1006 * option in the IPPROTO_IPV6 level is not 1007 * allowed on ICMPv6 sockets. 1008 * If you want to set it, use IPPROTO_RAW 1009 * level IPV6_CHECKSUM socket option 1010 * (Linux extension). 1011 */ 1012 return -EINVAL; 1013 } 1014 1015 /* You may get strange result with a positive odd offset; 1016 RFC2292bis agrees with me. */ 1017 if (val > 0 && (val&1)) 1018 return -EINVAL; 1019 if (val < 0) { 1020 rp->checksum = 0; 1021 } else { 1022 rp->checksum = 1; 1023 rp->offset = val; 1024 } 1025 1026 return 0; 1027 1028 default: 1029 return -ENOPROTOOPT; 1030 } 1031 } 1032 1033 static int rawv6_setsockopt(struct sock *sk, int level, int optname, 1034 sockptr_t optval, unsigned int optlen) 1035 { 1036 switch (level) { 1037 case SOL_RAW: 1038 break; 1039 1040 case SOL_ICMPV6: 1041 if (inet_sk(sk)->inet_num != IPPROTO_ICMPV6) 1042 return -EOPNOTSUPP; 1043 return rawv6_seticmpfilter(sk, level, optname, optval, optlen); 1044 case SOL_IPV6: 1045 if (optname == IPV6_CHECKSUM || 1046 optname == IPV6_HDRINCL) 1047 break; 1048 fallthrough; 1049 default: 1050 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1051 } 1052 1053 return do_rawv6_setsockopt(sk, level, optname, optval, optlen); 1054 } 1055 1056 static int do_rawv6_getsockopt(struct sock *sk, int level, int optname, 1057 char __user *optval, int __user *optlen) 1058 { 1059 struct raw6_sock *rp = raw6_sk(sk); 1060 int val, len; 1061 1062 if (get_user(len, optlen)) 1063 return -EFAULT; 1064 1065 switch (optname) { 1066 case IPV6_HDRINCL: 1067 val = inet_sk(sk)->hdrincl; 1068 break; 1069 case IPV6_CHECKSUM: 1070 /* 1071 * We allow getsockopt() for IPPROTO_IPV6-level 1072 * IPV6_CHECKSUM socket option on ICMPv6 sockets 1073 * since RFC3542 is silent about it. 1074 */ 1075 if (rp->checksum == 0) 1076 val = -1; 1077 else 1078 val = rp->offset; 1079 break; 1080 1081 default: 1082 return -ENOPROTOOPT; 1083 } 1084 1085 len = min_t(unsigned int, sizeof(int), len); 1086 1087 if (put_user(len, optlen)) 1088 return -EFAULT; 1089 if (copy_to_user(optval, &val, len)) 1090 return -EFAULT; 1091 return 0; 1092 } 1093 1094 static int rawv6_getsockopt(struct sock *sk, int level, int optname, 1095 char __user *optval, int __user *optlen) 1096 { 1097 switch (level) { 1098 case SOL_RAW: 1099 break; 1100 1101 case SOL_ICMPV6: 1102 if (inet_sk(sk)->inet_num != IPPROTO_ICMPV6) 1103 return -EOPNOTSUPP; 1104 return rawv6_geticmpfilter(sk, level, optname, optval, optlen); 1105 case SOL_IPV6: 1106 if (optname == IPV6_CHECKSUM || 1107 optname == IPV6_HDRINCL) 1108 break; 1109 fallthrough; 1110 default: 1111 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1112 } 1113 1114 return do_rawv6_getsockopt(sk, level, optname, optval, optlen); 1115 } 1116 1117 static int rawv6_ioctl(struct sock *sk, int cmd, unsigned long arg) 1118 { 1119 switch (cmd) { 1120 case SIOCOUTQ: { 1121 int amount = sk_wmem_alloc_get(sk); 1122 1123 return put_user(amount, (int __user *)arg); 1124 } 1125 case SIOCINQ: { 1126 struct sk_buff *skb; 1127 int amount = 0; 1128 1129 spin_lock_bh(&sk->sk_receive_queue.lock); 1130 skb = skb_peek(&sk->sk_receive_queue); 1131 if (skb) 1132 amount = skb->len; 1133 spin_unlock_bh(&sk->sk_receive_queue.lock); 1134 return put_user(amount, (int __user *)arg); 1135 } 1136 1137 default: 1138 #ifdef CONFIG_IPV6_MROUTE 1139 return ip6mr_ioctl(sk, cmd, (void __user *)arg); 1140 #else 1141 return -ENOIOCTLCMD; 1142 #endif 1143 } 1144 } 1145 1146 #ifdef CONFIG_COMPAT 1147 static int compat_rawv6_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg) 1148 { 1149 switch (cmd) { 1150 case SIOCOUTQ: 1151 case SIOCINQ: 1152 return -ENOIOCTLCMD; 1153 default: 1154 #ifdef CONFIG_IPV6_MROUTE 1155 return ip6mr_compat_ioctl(sk, cmd, compat_ptr(arg)); 1156 #else 1157 return -ENOIOCTLCMD; 1158 #endif 1159 } 1160 } 1161 #endif 1162 1163 static void rawv6_close(struct sock *sk, long timeout) 1164 { 1165 if (inet_sk(sk)->inet_num == IPPROTO_RAW) 1166 ip6_ra_control(sk, -1); 1167 ip6mr_sk_done(sk); 1168 sk_common_release(sk); 1169 } 1170 1171 static void raw6_destroy(struct sock *sk) 1172 { 1173 lock_sock(sk); 1174 ip6_flush_pending_frames(sk); 1175 release_sock(sk); 1176 1177 inet6_destroy_sock(sk); 1178 } 1179 1180 static int rawv6_init_sk(struct sock *sk) 1181 { 1182 struct raw6_sock *rp = raw6_sk(sk); 1183 1184 switch (inet_sk(sk)->inet_num) { 1185 case IPPROTO_ICMPV6: 1186 rp->checksum = 1; 1187 rp->offset = 2; 1188 break; 1189 case IPPROTO_MH: 1190 rp->checksum = 1; 1191 rp->offset = 4; 1192 break; 1193 default: 1194 break; 1195 } 1196 return 0; 1197 } 1198 1199 struct proto rawv6_prot = { 1200 .name = "RAWv6", 1201 .owner = THIS_MODULE, 1202 .close = rawv6_close, 1203 .destroy = raw6_destroy, 1204 .connect = ip6_datagram_connect_v6_only, 1205 .disconnect = __udp_disconnect, 1206 .ioctl = rawv6_ioctl, 1207 .init = rawv6_init_sk, 1208 .setsockopt = rawv6_setsockopt, 1209 .getsockopt = rawv6_getsockopt, 1210 .sendmsg = rawv6_sendmsg, 1211 .recvmsg = rawv6_recvmsg, 1212 .bind = rawv6_bind, 1213 .backlog_rcv = rawv6_rcv_skb, 1214 .hash = raw_hash_sk, 1215 .unhash = raw_unhash_sk, 1216 .obj_size = sizeof(struct raw6_sock), 1217 .useroffset = offsetof(struct raw6_sock, filter), 1218 .usersize = sizeof_field(struct raw6_sock, filter), 1219 .h.raw_hash = &raw_v6_hashinfo, 1220 #ifdef CONFIG_COMPAT 1221 .compat_ioctl = compat_rawv6_ioctl, 1222 #endif 1223 .diag_destroy = raw_abort, 1224 }; 1225 1226 #ifdef CONFIG_PROC_FS 1227 static int raw6_seq_show(struct seq_file *seq, void *v) 1228 { 1229 if (v == SEQ_START_TOKEN) { 1230 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1231 } else { 1232 struct sock *sp = v; 1233 __u16 srcp = inet_sk(sp)->inet_num; 1234 ip6_dgram_sock_seq_show(seq, v, srcp, 0, 1235 raw_seq_private(seq)->bucket); 1236 } 1237 return 0; 1238 } 1239 1240 static const struct seq_operations raw6_seq_ops = { 1241 .start = raw_seq_start, 1242 .next = raw_seq_next, 1243 .stop = raw_seq_stop, 1244 .show = raw6_seq_show, 1245 }; 1246 1247 static int __net_init raw6_init_net(struct net *net) 1248 { 1249 if (!proc_create_net_data("raw6", 0444, net->proc_net, &raw6_seq_ops, 1250 sizeof(struct raw_iter_state), &raw_v6_hashinfo)) 1251 return -ENOMEM; 1252 1253 return 0; 1254 } 1255 1256 static void __net_exit raw6_exit_net(struct net *net) 1257 { 1258 remove_proc_entry("raw6", net->proc_net); 1259 } 1260 1261 static struct pernet_operations raw6_net_ops = { 1262 .init = raw6_init_net, 1263 .exit = raw6_exit_net, 1264 }; 1265 1266 int __init raw6_proc_init(void) 1267 { 1268 return register_pernet_subsys(&raw6_net_ops); 1269 } 1270 1271 void raw6_proc_exit(void) 1272 { 1273 unregister_pernet_subsys(&raw6_net_ops); 1274 } 1275 #endif /* CONFIG_PROC_FS */ 1276 1277 /* Same as inet6_dgram_ops, sans udp_poll. */ 1278 const struct proto_ops inet6_sockraw_ops = { 1279 .family = PF_INET6, 1280 .owner = THIS_MODULE, 1281 .release = inet6_release, 1282 .bind = inet6_bind, 1283 .connect = inet_dgram_connect, /* ok */ 1284 .socketpair = sock_no_socketpair, /* a do nothing */ 1285 .accept = sock_no_accept, /* a do nothing */ 1286 .getname = inet6_getname, 1287 .poll = datagram_poll, /* ok */ 1288 .ioctl = inet6_ioctl, /* must change */ 1289 .gettstamp = sock_gettstamp, 1290 .listen = sock_no_listen, /* ok */ 1291 .shutdown = inet_shutdown, /* ok */ 1292 .setsockopt = sock_common_setsockopt, /* ok */ 1293 .getsockopt = sock_common_getsockopt, /* ok */ 1294 .sendmsg = inet_sendmsg, /* ok */ 1295 .recvmsg = sock_common_recvmsg, /* ok */ 1296 .mmap = sock_no_mmap, 1297 .sendpage = sock_no_sendpage, 1298 #ifdef CONFIG_COMPAT 1299 .compat_ioctl = inet6_compat_ioctl, 1300 #endif 1301 }; 1302 1303 static struct inet_protosw rawv6_protosw = { 1304 .type = SOCK_RAW, 1305 .protocol = IPPROTO_IP, /* wild card */ 1306 .prot = &rawv6_prot, 1307 .ops = &inet6_sockraw_ops, 1308 .flags = INET_PROTOSW_REUSE, 1309 }; 1310 1311 int __init rawv6_init(void) 1312 { 1313 return inet6_register_protosw(&rawv6_protosw); 1314 } 1315 1316 void rawv6_exit(void) 1317 { 1318 inet6_unregister_protosw(&rawv6_protosw); 1319 } 1320