npd: rework IPv6 relay logic (FS#396)
[project/odhcpd.git] / src / ndp.c
index d2fbe07..8824879 100644 (file)
--- a/src/ndp.c
+++ b/src/ndp.c
 
 #include <linux/rtnetlink.h>
 #include <linux/filter.h>
+
+#include <netlink/msg.h>
+#include <netlink/socket.h>
+#include <netlink/attr.h>
+
 #include "router.h"
 #include "dhcpv6.h"
 #include "ndp.h"
 
-
+struct event_socket {
+       struct odhcpd_event ev;
+       struct nl_sock *sock;
+};
 
 static void handle_solicit(void *addr, void *data, size_t len,
                struct interface *iface, void *dest);
-static void handle_rtnetlink(void *addr, void *data, size_t len,
-               struct interface *iface, void *dest);
+static void handle_rtnl_event(struct odhcpd_event *ev);
+static int cb_rtnl_valid(struct nl_msg *msg, void *arg);
 static void catch_rtnetlink(int error);
 
-static uint32_t rtnl_seqid = 0;
 static int ping_socket = -1;
-static struct odhcpd_event rtnl_event = {{.fd = -1}, handle_rtnetlink, catch_rtnetlink};
-
+static struct event_socket rtnl_event = {
+       .ev = {
+               .uloop = {.fd = - 1, },
+               .handle_dgram = NULL,
+               .handle_error = catch_rtnetlink,
+               .recv_msgs = handle_rtnl_event,
+       },
+       .sock = NULL,
+};
 
 // Filter ICMPv6 messages of type neighbor soliciation
 static struct sock_filter bpf[] = {
@@ -63,22 +77,26 @@ int init_ndp(void)
 {
        int val = 256 * 1024;
 
-       // Setup netlink socket
-       if ((rtnl_event.uloop.fd = odhcpd_open_rtnl()) < 0)
-               return -1;
+       rtnl_event.sock = odhcpd_create_nl_socket(NETLINK_ROUTE);
+       if (!rtnl_event.sock)
+               goto err;
+
+       rtnl_event.ev.uloop.fd = nl_socket_get_fd(rtnl_event.sock);
+
+       if (nl_socket_set_buffer_size(rtnl_event.sock, val, 0))
+               goto err;
 
-       if (setsockopt(rtnl_event.uloop.fd, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)))
-               setsockopt(rtnl_event.uloop.fd, SOL_SOCKET, SO_RCVBUFFORCE, &val, sizeof(val));
+       nl_socket_disable_seq_check(rtnl_event.sock);
 
-       // Receive netlink neighbor and ip-address events
-       uint32_t group = RTNLGRP_IPV6_IFADDR;
-       setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
-                       NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
-       group = RTNLGRP_IPV6_ROUTE;
-       setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
-                       NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
+       nl_socket_modify_cb(rtnl_event.sock, NL_CB_VALID, NL_CB_CUSTOM,
+                       cb_rtnl_valid, NULL);
 
-       odhcpd_register(&rtnl_event);
+       // Receive IPv6 address, IPv6 routes and neighbor events
+       if (nl_socket_add_memberships(rtnl_event.sock, RTNLGRP_IPV6_IFADDR,
+                               RTNLGRP_IPV6_ROUTE, RTNLGRP_NEIGH, 0))
+               goto err;
+
+       odhcpd_register(&rtnl_event.ev);
 
        // Open ICMPv6 socket
        ping_socket = socket(AF_INET6, SOCK_RAW | SOCK_CLOEXEC, IPPROTO_ICMPV6);
@@ -100,29 +118,54 @@ int init_ndp(void)
        ICMP6_FILTER_SETBLOCKALL(&filt);
        setsockopt(ping_socket, IPPROTO_ICMPV6, ICMP6_FILTER, &filt, sizeof(filt));
 
+       return 0;
 
-       // Netlink socket, continued...
-       group = RTNLGRP_NEIGH;
-       setsockopt(rtnl_event.uloop.fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
+err:
+       if (rtnl_event.sock) {
+               nl_socket_free(rtnl_event.sock);
+               rtnl_event.sock = NULL;
+               rtnl_event.ev.uloop.fd = -1;
+       }
 
-       return 0;
+       return -1;
 }
 
-
-static void dump_neigh_table(bool proxy)
+static void dump_neigh_table(const bool proxy)
 {
-       struct {
-               struct nlmsghdr nh;
-               struct ndmsg ndm;
-       } req = {
-               {sizeof(req), RTM_GETNEIGH, NLM_F_REQUEST | NLM_F_DUMP,
-                               ++rtnl_seqid, 0},
-               {.ndm_family = AF_INET6, .ndm_flags = (proxy) ? NTF_PROXY : 0}
+       struct nl_msg *msg;
+       struct ndmsg ndm = {
+               .ndm_family = AF_INET6,
+               .ndm_flags = proxy ? NTF_PROXY : 0,
        };
-       send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
-       odhcpd_process(&rtnl_event);
+
+       msg = nlmsg_alloc_simple(RTM_GETNEIGH, NLM_F_REQUEST | NLM_F_DUMP);
+       if (!msg)
+               return;
+
+       nlmsg_append(msg, &ndm, sizeof(ndm), 0);
+
+       nl_send_auto_complete(rtnl_event.sock, msg);
+
+       nlmsg_free(msg);
 }
 
+static void dump_addr_table(void)
+{
+       struct nl_msg *msg;
+       struct ifaddrmsg ifa = {
+               .ifa_family = AF_INET6,
+       };
+
+       msg = nlmsg_alloc_simple(RTM_GETADDR, NLM_F_REQUEST | NLM_F_DUMP);
+       if (!msg)
+               return;
+
+       nlmsg_append(msg, &ifa, sizeof(ifa), 0);
+
+       nl_send_auto_complete(rtnl_event.sock, msg);
+
+       nlmsg_free(msg);
+}
 
 int setup_ndp_interface(struct interface *iface, bool enable)
 {
@@ -143,18 +186,8 @@ int setup_ndp_interface(struct interface *iface, bool enable)
        }
 
        if (enable && (iface->ra == RELAYD_SERVER ||
-                       iface->dhcpv6 == RELAYD_SERVER || iface->ndp == RELAYD_RELAY)) {
-               // Synthesize initial address events
-               struct {
-                       struct nlmsghdr nh;
-                       struct ifaddrmsg ifa;
-               } req2 = {
-                       {sizeof(req2), RTM_GETADDR, NLM_F_REQUEST | NLM_F_DUMP,
-                                       ++rtnl_seqid, 0},
-                       {.ifa_family = AF_INET6, .ifa_index = iface->ifindex}
-               };
-               send(rtnl_event.uloop.fd, &req2, sizeof(req2), MSG_DONTWAIT);
-       }
+                       iface->dhcpv6 == RELAYD_SERVER || iface->ndp == RELAYD_RELAY))
+               dump_addr_table();
 
        if (enable && iface->ndp == RELAYD_RELAY) {
                if (write(procfd, "1\n", 2) < 0) {}
@@ -201,9 +234,8 @@ int setup_ndp_interface(struct interface *iface, bool enable)
                        dump_neigh_table(false);
                else
                        dump_neigh = false;
-       } else {
+       } else
                close(procfd);
-       }
 
        if (dump_neigh)
                dump_neigh_table(true);
@@ -217,9 +249,9 @@ int setup_ndp_interface(struct interface *iface, bool enable)
 static void ping6(struct in6_addr *addr,
                const struct interface *iface)
 {
-       struct sockaddr_in6 dest = {AF_INET6, 0, 0, *addr, iface->ifindex};
-       struct icmp6_hdr echo = {.icmp6_type = ICMP6_ECHO_REQUEST};
-       struct iovec iov = {&echo, sizeof(echo)};
+       struct sockaddr_in6 dest = { .sin6_family = AF_INET6, .sin6_addr = *addr, .sin6_scope_id = iface->ifindex, };
+       struct icmp6_hdr echo = { .icmp6_type = ICMP6_ECHO_REQUEST };
+       struct iovec iov = { .iov_base = &echo, .iov_len = sizeof(echo) };
 
        odhcpd_setup_route(addr, 128, iface, NULL, 128, true);
        odhcpd_send(ping_socket, &dest, &iov, 1, iface);
@@ -238,9 +270,10 @@ static void handle_solicit(void *addr, void *data, size_t len,
        // Solicitation is for duplicate address detection
        bool ns_is_dad = IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src);
 
+       // Don't process solicit messages on non relay interfaces
        // Don't forward any non-DAD solicitation for external ifaces
        // TODO: check if we should even forward DADs for them
-       if (iface->external && !ns_is_dad)
+       if (iface->ndp != RELAYD_RELAY || (iface->external && !ns_is_dad))
                return;
 
        if (len < sizeof(*ip6) + sizeof(*req))
@@ -262,7 +295,7 @@ static void handle_solicit(void *addr, void *data, size_t len,
 
        struct interface *c;
        list_for_each_entry(c, &interfaces, head)
-               if (iface->ndp == RELAYD_RELAY && iface != c &&
+               if (iface != c && c->ndp == RELAYD_RELAY &&
                                (ns_is_dad || !c->external))
                        ping6(&req->nd_ns_target, c);
 }
@@ -289,7 +322,7 @@ static int prefixcmp(const void *va, const void *vb)
 }
 
 // Check address update
-static void check_updates(struct interface *iface)
+static void check_addr_updates(struct interface *iface)
 {
        struct odhcpd_ipaddr addr[RELAYD_MAX_ADDRS] = {{IN6ADDR_ANY_INIT, 0, 0, 0, 0}};
        time_t now = odhcpd_time();
@@ -328,193 +361,169 @@ static void check_updates(struct interface *iface)
                dhcpv6_ia_postupdate(iface, now);
 
        if (change) {
-               syslog(LOG_DEBUG, "Raising SIGUSR1 due to address change");
+               syslog(LOG_INFO, "Raising SIGUSR1 due to address change on %s", iface->ifname);
                raise(SIGUSR1);
        }
 }
 
-
-// Handler for neighbor cache entries from the kernel. This is our source
-// to learn and unlearn hosts on interfaces.
-static void handle_rtnetlink(_unused void *addr, void *data, size_t len,
-               _unused struct interface *iface, _unused void *dest)
+void setup_addr_for_relaying(struct in6_addr *addr, struct interface *iface, bool add)
 {
-       bool dump_neigh = false;
-       struct in6_addr last_solicited = IN6ADDR_ANY_INIT;
-
-       for (struct nlmsghdr *nh = data; NLMSG_OK(nh, len);
-                       nh = NLMSG_NEXT(nh, len)) {
-               struct ndmsg *ndm = NLMSG_DATA(nh);
-               struct rtmsg *rtm = NLMSG_DATA(nh);
-
-               bool is_addr = (nh->nlmsg_type == RTM_NEWADDR
-                               || nh->nlmsg_type == RTM_DELADDR);
-               bool is_route = (nh->nlmsg_type == RTM_NEWROUTE
-                               || nh->nlmsg_type == RTM_DELROUTE);
-               bool is_neigh = (nh->nlmsg_type == RTM_NEWNEIGH
-                               || nh->nlmsg_type == RTM_DELNEIGH);
-
-               // Family and ifindex are on the same offset for NEIGH and ADDR
-               if ((!is_addr && !is_route && !is_neigh)
-                               || NLMSG_PAYLOAD(nh, 0) < sizeof(*ndm)
-                               || ndm->ndm_family != AF_INET6)
+       struct interface *c;
+
+       list_for_each_entry(c, &interfaces, head) {
+               if (iface == c || (c->ndp != RELAYD_RELAY && !add))
                        continue;
 
-               if (is_route) {
-                       // Inform about a change in default route
-                       if (rtm->rtm_dst_len == 0) {
-                               syslog(LOG_DEBUG, "Raising SIGUSR1 due to default route change");
-                               raise(SIGUSR1);
-                       }
+               odhcpd_setup_proxy_neigh(addr, c, c->ndp == RELAYD_RELAY ? add : false);
+       }
+}
 
+void setup_ping6(struct in6_addr *addr, struct interface *iface)
+{
+       struct interface *c;
+
+       list_for_each_entry(c, &interfaces, head) {
+               if (iface == c || c->ndp != RELAYD_RELAY ||
+                               c->external == true)
                        continue;
-               }
 
-               // Data to retrieve
-               size_t rta_offset = (is_addr) ? sizeof(struct ifaddrmsg) : sizeof(*ndm);
-               uint16_t atype = (is_addr) ? IFA_ADDRESS : NDA_DST;
-               ssize_t alen = NLMSG_PAYLOAD(nh, rta_offset);
-               struct in6_addr *addr = NULL;
-
-               for (struct rtattr *rta = (void*)(((uint8_t*)ndm) + rta_offset);
-                               RTA_OK(rta, alen); rta = RTA_NEXT(rta, alen)) {
-                       if (rta->rta_type == atype &&
-                                       RTA_PAYLOAD(rta) >= sizeof(*addr)) {
-                               addr = RTA_DATA(rta);
-                       }
+               ping6(addr, c);
+       }
+}
+
+static struct in6_addr last_solicited;
+
+static void handle_rtnl_event(struct odhcpd_event *e)
+{
+       struct event_socket *ev_sock = container_of(e, struct event_socket, ev);
+
+       nl_recvmsgs_default(ev_sock->sock);
+}
+
+
+// Handler for neighbor cache entries from the kernel. This is our source
+// to learn and unlearn hosts on interfaces.
+static int cb_rtnl_valid(struct nl_msg *msg, _unused void *arg)
+{
+       struct nlmsghdr *hdr = nlmsg_hdr(msg);
+       struct in6_addr *addr = NULL;
+       struct interface *iface = NULL;
+       bool add = false;
+
+       switch (hdr->nlmsg_type) {
+       case RTM_NEWROUTE:
+       case RTM_DELROUTE: {
+               struct rtmsg *rtm = nlmsg_data(hdr);
+
+               if (!nlmsg_valid_hdr(hdr, sizeof(*rtm)) ||
+                               rtm->rtm_family != AF_INET6)
+                       return NL_SKIP;
+
+               if (rtm->rtm_dst_len == 0) {
+                       syslog(LOG_INFO, "Raising SIGUSR1 due to default route change");
+                       raise(SIGUSR1);
                }
+               return NL_OK;
+       }
+
+       case RTM_NEWADDR:
+               add = true;
+       case RTM_DELADDR: {
+               struct ifaddrmsg *ifa = nlmsg_data(hdr);
+               struct nlattr *nla[__IFA_MAX];
+
+               if (!nlmsg_valid_hdr(hdr, sizeof(*ifa)) ||
+                               ifa->ifa_family != AF_INET6)
+                       return NL_SKIP;
 
-               // Lookup interface
-               struct interface *iface = odhcpd_get_interface_by_index(ndm->ndm_ifindex);
+               iface = odhcpd_get_interface_by_index(ifa->ifa_index);
                if (!iface)
-                       continue;
+                       return NL_SKIP;
+
+               nlmsg_parse(hdr, sizeof(*ifa), nla, __IFA_MAX - 1, NULL);
+               if (!nla[IFA_ADDRESS])
+                       return NL_SKIP;
 
-               // Address not specified or unrelated
+               addr = nla_data(nla[IFA_ADDRESS]);
                if (!addr || IN6_IS_ADDR_LINKLOCAL(addr) ||
                                IN6_IS_ADDR_MULTICAST(addr))
-                       continue;
+                       return NL_SKIP;
 
-               // Check for states
-               bool add;
-               if (is_addr)
-                       add = (nh->nlmsg_type == RTM_NEWADDR);
-               else
-                       add = (nh->nlmsg_type == RTM_NEWNEIGH && (ndm->ndm_state &
-                               (NUD_REACHABLE | NUD_STALE | NUD_DELAY | NUD_PROBE
-                                               | NUD_PERMANENT | NUD_NOARP)));
-
-               if (iface->ndp == RELAYD_RELAY) {
-                       // Replay change to all neighbor cache
-                       struct {
-                               struct nlmsghdr nh;
-                               struct ndmsg ndm;
-                               struct nlattr nla_dst;
-                               struct in6_addr dst;
-                       } req = {
-                               {sizeof(req), RTM_DELNEIGH, NLM_F_REQUEST,
-                                               ++rtnl_seqid, 0},
-                               {.ndm_family = AF_INET6, .ndm_flags = NTF_PROXY},
-                               {sizeof(struct nlattr) + sizeof(struct in6_addr), NDA_DST},
-                               *addr
-                       };
-
-                       if (ndm->ndm_flags & NTF_PROXY) {
-                               // Dump & flush proxy entries
-                               if (nh->nlmsg_type == RTM_NEWNEIGH) {
-                                       req.ndm.ndm_ifindex = iface->ifindex;
-                                       send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
-                                       setup_route(addr, iface, false);
-                                       dump_neigh = true;
-                               }
-                       } else if (add) {
-                               struct interface *c;
-                               list_for_each_entry(c, &interfaces, head) {
-                                       if (iface == c)
-                                               continue;
-
-                                       if (c->ndp == RELAYD_RELAY) {
-                                               req.nh.nlmsg_type = RTM_NEWNEIGH;
-                                               req.nh.nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
-
-                                               req.ndm.ndm_ifindex = c->ifindex;
-                                               send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
-                                       } else { // Delete NDP cache from interfaces without relay
-                                               req.nh.nlmsg_type = RTM_DELNEIGH;
-                                               req.nh.nlmsg_flags &= ~(NLM_F_CREATE | NLM_F_REPLACE);
-
-                                               req.ndm.ndm_ifindex = c->ifindex;
-                                               send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
-                                       }
-                               }
-
-                               setup_route(addr, iface, true);
-                       } else {
-                               if (nh->nlmsg_type == RTM_NEWNEIGH) {
-                                       // might be locally originating
-                                       if (!IN6_ARE_ADDR_EQUAL(&last_solicited, addr)) {
-                                               last_solicited = *addr;
-
-                                               struct interface *c;
-                                               list_for_each_entry(c, &interfaces, head)
-                                                       if (iface->ndp == RELAYD_RELAY && iface != c &&
-                                                                       !c->external == false)
-                                                               ping6(addr, c);
-                                       }
-                               } else {
-                                       struct interface *c;
-                                       list_for_each_entry(c, &interfaces, head) {
-                                               if (c->ndp == RELAYD_RELAY && iface != c) {
-                                                       req.ndm.ndm_ifindex = c->ifindex;
-                                                       send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
-                                               }
-                                       }
-                                       setup_route(addr, iface, false);
-
-                                       // also: dump to add proxies back in case it moved elsewhere
-                                       dump_neigh = true;
-                               }
-                       }
-               }
+               check_addr_updates(iface);
+
+               if (iface->ndp != RELAYD_RELAY)
+                       break;
+
+               /* handle the relay logic below */
+               setup_addr_for_relaying(addr, iface, add);
+
+               if (!add)
+                       dump_neigh_table(false);
+               break;
+       }
 
-               if (is_addr) {
-                       check_updates(iface);
+       case RTM_NEWNEIGH:
+               add = true;
+       case RTM_DELNEIGH: {
+               struct ndmsg *ndm = nlmsg_data(hdr);
+               struct nlattr *nla[__NDA_MAX];
 
-                       if (iface->dhcpv6 == RELAYD_SERVER)
-                               iface->ia_reconf = true;
+               if (!nlmsg_valid_hdr(hdr, sizeof(*ndm)) ||
+                               ndm->ndm_family != AF_INET6)
+                       return NL_SKIP;
 
-                       if (iface->ndp == RELAYD_RELAY && iface->master) {
-                               // Replay address changes on all slave interfaces
-                               nh->nlmsg_flags = NLM_F_REQUEST;
+               iface = odhcpd_get_interface_by_index(ndm->ndm_ifindex);
+               if (!iface || iface->ndp != RELAYD_RELAY)
+                       return (iface ? NL_OK : NL_SKIP);
 
-                               if (nh->nlmsg_type == RTM_NEWADDR)
-                                       nh->nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
+               nlmsg_parse(hdr, sizeof(*ndm), nla, __NDA_MAX - 1, NULL);
+               if (!nla[NDA_DST])
+                       return NL_SKIP;
 
-                               struct interface *c;
-                               list_for_each_entry(c, &interfaces, head) {
-                                       if (c->ndp == RELAYD_RELAY && !c->master) {
-                                               ndm->ndm_ifindex = c->ifindex;
-                                               send(rtnl_event.uloop.fd, nh, nh->nlmsg_len, MSG_DONTWAIT);
-                                       }
-                               }
+               addr = nla_data(nla[NDA_DST]);
+               if (!addr || IN6_IS_ADDR_LINKLOCAL(addr) ||
+                               IN6_IS_ADDR_MULTICAST(addr))
+                       return NL_SKIP;
+
+               if (ndm->ndm_flags & NTF_PROXY) {
+                       /* Dump and flush proxy entries */
+                       if (hdr->nlmsg_type == RTM_NEWNEIGH) {
+                               odhcpd_setup_proxy_neigh(addr, iface, false);
+                               setup_route(addr, iface, false);
+                               dump_neigh_table(false);
                        }
+
+                       return NL_OK;
                }
+
+               if (add && !(ndm->ndm_state &
+                               (NUD_REACHABLE | NUD_STALE | NUD_DELAY | NUD_PROBE |
+                                NUD_PERMANENT | NUD_NOARP))) {
+                       if (!IN6_ARE_ADDR_EQUAL(&last_solicited, addr)) {
+                               last_solicited = *addr;
+                               setup_ping6(addr, iface);
+                       }
+
+                       return NL_OK;
+               }
+
+               setup_addr_for_relaying(addr, iface, add);
+               setup_route(addr, iface, add);
+
+               if (!add)
+                       dump_neigh_table(false);
+               break;
        }
 
-       if (dump_neigh)
-               dump_neigh_table(false);
+       default:
+               return NL_SKIP;
+       }
+
+       return NL_OK;
 }
 
 static void catch_rtnetlink(int error)
 {
-       if (error == ENOBUFS) {
-               struct {
-                       struct nlmsghdr nh;
-                       struct ifaddrmsg ifa;
-               } req2 = {
-                       {sizeof(req2), RTM_GETADDR, NLM_F_REQUEST | NLM_F_DUMP,
-                                       ++rtnl_seqid, 0},
-                       {.ifa_family = AF_INET6}
-               };
-               send(rtnl_event.uloop.fd, &req2, sizeof(req2), MSG_DONTWAIT);
-       }
+       if (error == ENOBUFS)
+               dump_addr_table();
 }