summaryrefslogtreecommitdiffstats
path: root/nhrpd/nhrp_multicast.c
diff options
context:
space:
mode:
Diffstat (limited to 'nhrpd/nhrp_multicast.c')
-rw-r--r--nhrpd/nhrp_multicast.c298
1 files changed, 298 insertions, 0 deletions
diff --git a/nhrpd/nhrp_multicast.c b/nhrpd/nhrp_multicast.c
new file mode 100644
index 0000000..89be13b
--- /dev/null
+++ b/nhrpd/nhrp_multicast.c
@@ -0,0 +1,298 @@
+/* NHRP Multicast Support
+ * Copyright (c) 2020-2021 4RF Limited
+ *
+ * This file is free software: you may copy, redistribute and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 2 of the License, or
+ * (at your option) any later version.
+ */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <fcntl.h>
+#include <net/if.h>
+#include <net/ethernet.h>
+#include <netinet/if_ether.h>
+#include <linux/netlink.h>
+#include <linux/neighbour.h>
+#include <linux/netfilter/nfnetlink_log.h>
+#include <linux/if_packet.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+
+#include "thread.h"
+#include "nhrpd.h"
+#include "netlink.h"
+#include "znl.h"
+#include "os.h"
+
+DEFINE_MTYPE_STATIC(NHRPD, NHRP_MULTICAST, "NHRP Multicast");
+
+int netlink_mcast_nflog_group;
+static int netlink_mcast_log_fd = -1;
+static struct thread *netlink_mcast_log_thread;
+
+struct mcast_ctx {
+ struct interface *ifp;
+ struct zbuf *pkt;
+};
+
+static void nhrp_multicast_send(struct nhrp_peer *p, struct zbuf *zb)
+{
+ size_t addrlen;
+ int ret;
+
+ addrlen = sockunion_get_addrlen(&p->vc->remote.nbma);
+ ret = os_sendmsg(zb->head, zbuf_used(zb), p->ifp->ifindex,
+ sockunion_get_addr(&p->vc->remote.nbma), addrlen,
+ addrlen == 4 ? ETH_P_IP : ETH_P_IPV6);
+
+ debugf(NHRP_DEBUG_COMMON,
+ "Multicast Packet: %pSU -> %pSU, ret = %d, size = %zu, addrlen = %zu",
+ &p->vc->local.nbma, &p->vc->remote.nbma, ret, zbuf_used(zb),
+ addrlen);
+}
+
+static void nhrp_multicast_forward_nbma(union sockunion *nbma_addr,
+ struct interface *ifp, struct zbuf *pkt)
+{
+ struct nhrp_peer *p = nhrp_peer_get(ifp, nbma_addr);
+
+ if (p && p->online) {
+ /* Send packet */
+ nhrp_multicast_send(p, pkt);
+ }
+ nhrp_peer_unref(p);
+}
+
+static void nhrp_multicast_forward_cache(struct nhrp_cache *c, void *pctx)
+{
+ struct mcast_ctx *ctx = (struct mcast_ctx *)pctx;
+
+ if (c->cur.type == NHRP_CACHE_DYNAMIC && c->cur.peer)
+ nhrp_multicast_forward_nbma(&c->cur.peer->vc->remote.nbma,
+ ctx->ifp, ctx->pkt);
+}
+
+static void nhrp_multicast_forward(struct nhrp_multicast *mcast, void *pctx)
+{
+ struct mcast_ctx *ctx = (struct mcast_ctx *)pctx;
+ struct nhrp_interface *nifp = ctx->ifp->info;
+
+ if (!nifp->enabled)
+ return;
+
+ /* dynamic */
+ if (sockunion_family(&mcast->nbma_addr) == AF_UNSPEC) {
+ nhrp_cache_foreach(ctx->ifp, nhrp_multicast_forward_cache,
+ pctx);
+ return;
+ }
+
+ /* Fixed IP Address */
+ nhrp_multicast_forward_nbma(&mcast->nbma_addr, ctx->ifp, ctx->pkt);
+}
+
+static void netlink_mcast_log_handler(struct nlmsghdr *msg, struct zbuf *zb)
+{
+ struct nfgenmsg *nf;
+ struct rtattr *rta;
+ struct zbuf rtapl;
+ uint32_t *out_ndx = NULL;
+ afi_t afi;
+ struct mcast_ctx ctx;
+
+ nf = znl_pull(zb, sizeof(*nf));
+ if (!nf)
+ return;
+
+ ctx.pkt = NULL;
+ while ((rta = znl_rta_pull(zb, &rtapl)) != NULL) {
+ switch (rta->rta_type) {
+ case NFULA_IFINDEX_OUTDEV:
+ out_ndx = znl_pull(&rtapl, sizeof(*out_ndx));
+ break;
+ case NFULA_PAYLOAD:
+ ctx.pkt = &rtapl;
+ break;
+ /* NFULA_HWHDR exists and is supposed to contain source
+ * hardware address. However, for ip_gre it seems to be
+ * the nexthop destination address if the packet matches
+ * route.
+ */
+ }
+ }
+
+ if (!out_ndx || !ctx.pkt)
+ return;
+
+ ctx.ifp = if_lookup_by_index(htonl(*out_ndx), VRF_DEFAULT);
+ if (!ctx.ifp)
+ return;
+
+ debugf(NHRP_DEBUG_COMMON,
+ "Intercepted multicast packet leaving %s len %zu",
+ ctx.ifp->name, zbuf_used(ctx.pkt));
+
+ for (afi = 0; afi < AFI_MAX; afi++) {
+ nhrp_multicast_foreach(ctx.ifp, afi, nhrp_multicast_forward,
+ (void *)&ctx);
+ }
+}
+
+static void netlink_mcast_log_recv(struct thread *t)
+{
+ uint8_t buf[65535]; /* Max OSPF Packet size */
+ int fd = THREAD_FD(t);
+ struct zbuf payload, zb;
+ struct nlmsghdr *n;
+
+
+ zbuf_init(&zb, buf, sizeof(buf), 0);
+ while (zbuf_recv(&zb, fd) > 0) {
+ while ((n = znl_nlmsg_pull(&zb, &payload)) != NULL) {
+ debugf(NHRP_DEBUG_COMMON,
+ "Netlink-mcast-log: Received msg_type %u, msg_flags %u",
+ n->nlmsg_type, n->nlmsg_flags);
+ switch (n->nlmsg_type) {
+ case (NFNL_SUBSYS_ULOG << 8) | NFULNL_MSG_PACKET:
+ netlink_mcast_log_handler(n, &payload);
+ break;
+ }
+ }
+ }
+
+ thread_add_read(master, netlink_mcast_log_recv, 0, netlink_mcast_log_fd,
+ &netlink_mcast_log_thread);
+}
+
+static void netlink_mcast_log_register(int fd, int group)
+{
+ struct nlmsghdr *n;
+ struct nfgenmsg *nf;
+ struct nfulnl_msg_config_cmd cmd;
+ struct zbuf *zb = zbuf_alloc(512);
+
+ n = znl_nlmsg_push(zb, (NFNL_SUBSYS_ULOG << 8) | NFULNL_MSG_CONFIG,
+ NLM_F_REQUEST | NLM_F_ACK);
+ nf = znl_push(zb, sizeof(*nf));
+ *nf = (struct nfgenmsg){
+ .nfgen_family = AF_UNSPEC,
+ .version = NFNETLINK_V0,
+ .res_id = htons(group),
+ };
+ cmd.command = NFULNL_CFG_CMD_BIND;
+ znl_rta_push(zb, NFULA_CFG_CMD, &cmd, sizeof(cmd));
+ znl_nlmsg_complete(zb, n);
+
+ zbuf_send(zb, fd);
+ zbuf_free(zb);
+}
+
+void netlink_mcast_set_nflog_group(int nlgroup)
+{
+ if (netlink_mcast_log_fd >= 0) {
+ THREAD_OFF(netlink_mcast_log_thread);
+ close(netlink_mcast_log_fd);
+ netlink_mcast_log_fd = -1;
+ debugf(NHRP_DEBUG_COMMON, "De-register nflog group");
+ }
+ netlink_mcast_nflog_group = nlgroup;
+ if (nlgroup) {
+ netlink_mcast_log_fd = znl_open(NETLINK_NETFILTER, 0);
+ if (netlink_mcast_log_fd < 0)
+ return;
+
+ netlink_mcast_log_register(netlink_mcast_log_fd, nlgroup);
+ thread_add_read(master, netlink_mcast_log_recv, 0,
+ netlink_mcast_log_fd,
+ &netlink_mcast_log_thread);
+ debugf(NHRP_DEBUG_COMMON, "Register nflog group: %d",
+ netlink_mcast_nflog_group);
+ }
+}
+
+static int nhrp_multicast_free(struct interface *ifp,
+ struct nhrp_multicast *mcast)
+{
+ struct nhrp_interface *nifp = ifp->info;
+
+ nhrp_mcastlist_del(&nifp->afi[mcast->afi].mcastlist_head, mcast);
+ XFREE(MTYPE_NHRP_MULTICAST, mcast);
+ return 0;
+}
+
+int nhrp_multicast_add(struct interface *ifp, afi_t afi,
+ union sockunion *nbma_addr)
+{
+ struct nhrp_interface *nifp = ifp->info;
+ struct nhrp_multicast *mcast;
+
+ frr_each (nhrp_mcastlist, &nifp->afi[afi].mcastlist_head, mcast) {
+ if (sockunion_same(&mcast->nbma_addr, nbma_addr))
+ return NHRP_ERR_ENTRY_EXISTS;
+ }
+
+ mcast = XMALLOC(MTYPE_NHRP_MULTICAST, sizeof(struct nhrp_multicast));
+
+ *mcast = (struct nhrp_multicast){
+ .afi = afi, .ifp = ifp, .nbma_addr = *nbma_addr,
+ };
+ nhrp_mcastlist_add_tail(&nifp->afi[afi].mcastlist_head, mcast);
+
+ debugf(NHRP_DEBUG_COMMON, "Adding multicast entry (%pSU)", nbma_addr);
+
+ return NHRP_OK;
+}
+
+int nhrp_multicast_del(struct interface *ifp, afi_t afi,
+ union sockunion *nbma_addr)
+{
+ struct nhrp_interface *nifp = ifp->info;
+ struct nhrp_multicast *mcast;
+
+ frr_each_safe (nhrp_mcastlist, &nifp->afi[afi].mcastlist_head, mcast) {
+ if (!sockunion_same(&mcast->nbma_addr, nbma_addr))
+ continue;
+
+ debugf(NHRP_DEBUG_COMMON, "Deleting multicast entry (%pSU)",
+ nbma_addr);
+
+ nhrp_multicast_free(ifp, mcast);
+
+ return NHRP_OK;
+ }
+
+ return NHRP_ERR_ENTRY_NOT_FOUND;
+}
+
+void nhrp_multicast_interface_del(struct interface *ifp)
+{
+ struct nhrp_interface *nifp = ifp->info;
+ struct nhrp_multicast *mcast;
+ afi_t afi;
+
+ for (afi = 0; afi < AFI_MAX; afi++) {
+ debugf(NHRP_DEBUG_COMMON, "Cleaning up multicast entries (%zu)",
+ nhrp_mcastlist_count(&nifp->afi[afi].mcastlist_head));
+
+ frr_each_safe (nhrp_mcastlist, &nifp->afi[afi].mcastlist_head,
+ mcast) {
+ nhrp_multicast_free(ifp, mcast);
+ }
+ }
+}
+
+void nhrp_multicast_foreach(struct interface *ifp, afi_t afi,
+ void (*cb)(struct nhrp_multicast *, void *),
+ void *ctx)
+{
+ struct nhrp_interface *nifp = ifp->info;
+ struct nhrp_multicast *mcast;
+
+ frr_each (nhrp_mcastlist, &nifp->afi[afi].mcastlist_head, mcast) {
+ cb(mcast, ctx);
+ }
+}