/** @file * @brief IPv6 MLD related functions */ /* * Copyright (c) 2018 Intel Corporation * * SPDX-License-Identifier: Apache-2.0 */ #include LOG_MODULE_DECLARE(net_ipv6, CONFIG_NET_IPV6_LOG_LEVEL); #include #include #include #include #include #include #include "net_private.h" #include "connection.h" #include "icmpv6.h" #include "udp_internal.h" #include "tcp_internal.h" #include "ipv6.h" #include "nbr.h" #include "6lo.h" #include "route.h" #include "net_stats.h" /* Timeout for various buffer allocations in this file. */ #define PKT_WAIT_TIME K_MSEC(50) #define MLDv2_MCAST_RECORD_LEN sizeof(struct net_icmpv6_mld_mcast_record) #define IPV6_OPT_HDR_ROUTER_ALERT_LEN 8 #define MLDv2_LEN (MLDv2_MCAST_RECORD_LEN + sizeof(struct in6_addr)) static int mld_create(struct net_pkt *pkt, const struct in6_addr *addr, u8_t record_type, u16_t num_sources) { NET_PKT_DATA_ACCESS_DEFINE(mld_access, struct net_icmpv6_mld_mcast_record); struct net_icmpv6_mld_mcast_record *mld; mld = (struct net_icmpv6_mld_mcast_record *) net_pkt_get_data(pkt, &mld_access); if (!mld) { return -ENOBUFS; } mld->record_type = record_type; mld->aux_data_len = 0U; mld->num_sources = htons(num_sources); net_ipaddr_copy(&mld->mcast_address, addr); if (net_pkt_set_data(pkt, &mld_access)) { return -ENOBUFS; } if (num_sources > 0) { /* All source addresses, RFC 3810 ch 3 */ if (net_pkt_write(pkt, net_ipv6_unspecified_address()->s6_addr, sizeof(struct in6_addr))) { return -ENOBUFS; } } return 0; } static int mld_create_packet(struct net_pkt *pkt, u16_t count) { struct in6_addr dst; /* Sent to all MLDv2-capable routers */ net_ipv6_addr_create(&dst, 0xff02, 0, 0, 0, 0, 0, 0, 0x0016); net_pkt_set_ipv6_hop_limit(pkt, 1); /* RFC 3810 ch 7.4 */ if (net_ipv6_create(pkt, net_if_ipv6_select_src_addr( net_pkt_iface(pkt), &dst), &dst)) { return -ENOBUFS; } /* Add hop-by-hop option and router alert option, RFC 3810 ch 5. */ if (net_pkt_write_u8(pkt, IPPROTO_ICMPV6) || net_pkt_write_u8(pkt, 0)) { return -ENOBUFS; } /* IPv6 router alert option is described in RFC 2711. * - 0x0502 RFC 2711 ch 2.1 * - MLD (value 0) * - 2 bytes of padding */ if (net_pkt_write_be16(pkt, 0x0502) || net_pkt_write_be16(pkt, 0) || net_pkt_write_be16(pkt, 0)) { return -ENOBUFS; } net_pkt_set_ipv6_ext_len(pkt, IPV6_OPT_HDR_ROUTER_ALERT_LEN); /* ICMPv6 header + reserved space + count. * MLDv6 stuff will come right after */ if (net_icmpv6_create(pkt, NET_ICMPV6_MLDv2, 0) || net_pkt_write_be16(pkt, 0) || net_pkt_write_be16(pkt, count)) { return -ENOBUFS; } net_pkt_set_ipv6_next_hdr(pkt, NET_IPV6_NEXTHDR_HBHO); return 0; } static int mld_send(struct net_pkt *pkt) { net_pkt_cursor_init(pkt); net_ipv6_finalize(pkt, IPPROTO_ICMPV6); if (net_send_data(pkt) < 0) { net_stats_update_icmp_drop(net_pkt_iface(pkt)); net_stats_update_ipv6_mld_drop(net_pkt_iface(pkt)); net_pkt_unref(pkt); return -1; } net_stats_update_icmp_sent(net_pkt_iface(pkt)); net_stats_update_ipv6_mld_sent(net_pkt_iface(pkt)); return 0; } static int mld_send_generic(struct net_if *iface, const struct in6_addr *addr, u8_t mode) { struct net_pkt *pkt; int ret; pkt = net_pkt_alloc_with_buffer(iface, IPV6_OPT_HDR_ROUTER_ALERT_LEN + NET_ICMPV6_UNUSED_LEN + MLDv2_MCAST_RECORD_LEN + sizeof(struct in6_addr), AF_INET6, IPPROTO_ICMPV6, PKT_WAIT_TIME); if (!pkt) { return -ENOMEM; } if (mld_create_packet(pkt, 1) || mld_create(pkt, addr, mode, 1)) { ret = -ENOBUFS; goto drop; } ret = mld_send(pkt); if (ret) { goto drop; } return 0; drop: net_pkt_unref(pkt); return ret; } int net_ipv6_mld_join(struct net_if *iface, const struct in6_addr *addr) { struct net_if_mcast_addr *maddr; int ret; maddr = net_if_ipv6_maddr_lookup(addr, &iface); if (maddr && net_if_ipv6_maddr_is_joined(maddr)) { return -EALREADY; } if (!maddr) { maddr = net_if_ipv6_maddr_add(iface, addr); if (!maddr) { return -ENOMEM; } } ret = mld_send_generic(iface, addr, NET_IPV6_MLDv2_MODE_IS_EXCLUDE); if (ret < 0) { return ret; } net_if_ipv6_maddr_join(maddr); net_if_mcast_monitor(iface, addr, true); net_mgmt_event_notify_with_info(NET_EVENT_IPV6_MCAST_JOIN, iface, &maddr->address.in6_addr, sizeof(struct in6_addr)); return ret; } int net_ipv6_mld_leave(struct net_if *iface, const struct in6_addr *addr) { struct net_if_mcast_addr *maddr; int ret; maddr = net_if_ipv6_maddr_lookup(addr, &iface); if (!maddr) { return -ENOENT; } if (!net_if_ipv6_maddr_rm(iface, addr)) { return -EINVAL; } ret = mld_send_generic(iface, addr, NET_IPV6_MLDv2_MODE_IS_INCLUDE); if (ret < 0) { return ret; } net_if_mcast_monitor(iface, addr, false); net_mgmt_event_notify_with_info(NET_EVENT_IPV6_MCAST_LEAVE, iface, &maddr->address.in6_addr, sizeof(struct in6_addr)); return ret; } static void send_mld_report(struct net_if *iface) { struct net_if_ipv6 *ipv6 = iface->config.ip.ipv6; struct net_pkt *pkt; int i, count = 0; NET_ASSERT(ipv6); for (i = 0; i < NET_IF_MAX_IPV6_MADDR; i++) { if (!ipv6->mcast[i].is_used || !ipv6->mcast[i].is_joined) { continue; } count++; } pkt = net_pkt_alloc_with_buffer(iface, IPV6_OPT_HDR_ROUTER_ALERT_LEN + NET_ICMPV6_UNUSED_LEN + count * MLDv2_MCAST_RECORD_LEN, AF_INET6, IPPROTO_ICMPV6, PKT_WAIT_TIME); if (!pkt) { return; } if (mld_create_packet(pkt, count)) { goto drop; } for (i = 0; i < NET_IF_MAX_IPV6_MADDR; i++) { if (!ipv6->mcast[i].is_used || !ipv6->mcast[i].is_joined) { continue; } if (!mld_create(pkt, &ipv6->mcast[i].address.in6_addr, NET_IPV6_MLDv2_MODE_IS_EXCLUDE, 0)) { goto drop; } } if (!mld_send(pkt)) { return; } drop: net_pkt_unref(pkt); } #define dbg_addr(action, pkt_str, src, dst) \ do { \ NET_DBG("%s %s from %s to %s", action, pkt_str, \ log_strdup(net_sprint_ipv6_addr(src)), \ log_strdup(net_sprint_ipv6_addr(dst))); \ } while (0) #define dbg_addr_recv(pkt_str, src, dst) \ dbg_addr("Received", pkt_str, src, dst) static enum net_verdict handle_mld_query(struct net_pkt *pkt, struct net_ipv6_hdr *ip_hdr, struct net_icmp_hdr *icmp_hdr) { NET_PKT_DATA_ACCESS_CONTIGUOUS_DEFINE(mld_access, struct net_icmpv6_mld_query); u16_t length = net_pkt_get_len(pkt); struct net_icmpv6_mld_query *mld_query; u16_t pkt_len; mld_query = (struct net_icmpv6_mld_query *) net_pkt_get_data(pkt, &mld_access); if (!mld_query) { NET_DBG("DROP: NULL MLD query"); goto drop; } net_pkt_acknowledge_data(pkt, &mld_access); dbg_addr_recv("Multicast Listener Query", &ip_hdr->src, &ip_hdr->dst); net_stats_update_ipv6_mld_recv(net_pkt_iface(pkt)); mld_query->num_sources = ntohs(mld_query->num_sources); pkt_len = sizeof(struct net_ipv6_hdr) + net_pkt_ipv6_ext_len(pkt) + sizeof(struct net_icmp_hdr) + sizeof(struct net_icmpv6_mld_query) + sizeof(struct in6_addr) * mld_query->num_sources; if (length < pkt_len || pkt_len > NET_IPV6_MTU || ip_hdr->hop_limit != 1U || icmp_hdr->code != 0U) { goto drop; } /* Currently we only support an unspecified address query. */ if (!net_ipv6_addr_cmp(&mld_query->mcast_address, net_ipv6_unspecified_address())) { NET_DBG("DROP: only supporting unspecified address query"); goto drop; } send_mld_report(net_pkt_iface(pkt)); net_pkt_unref(pkt); return NET_OK; drop: net_stats_update_ipv6_mld_drop(net_pkt_iface(pkt)); return NET_DROP; } static struct net_icmpv6_handler mld_query_input_handler = { .type = NET_ICMPV6_MLD_QUERY, .code = 0, .handler = handle_mld_query, }; void net_ipv6_mld_init(void) { net_icmpv6_register_handler(&mld_query_input_handler); }