decrease host timeout, add more ping tries and make them configurable
[project/relayd.git] / route.c
diff --git a/route.c b/route.c
index f436691..cbb624e 100644 (file)
--- a/route.c
+++ b/route.c
 #include <string.h>
 #include <errno.h>
 #include <unistd.h>
+#include <fcntl.h>
+
+#include <linux/fib_rules.h>
 
 #include "relayd.h"
 
 static struct uloop_fd rtnl_sock;
 static unsigned int rtnl_seq, rtnl_dump_seq;
+int route_table = 16800;
+
+static void rtnl_flush(void)
+{
+       int fd;
+
+       fd = open("/proc/sys/net/ipv4/route/flush", O_WRONLY);
+       if (fd < 0)
+               return;
+
+       write(fd, "-1", 2);
+       close(fd);
+}
+
+enum {
+       RULE_F_ADD = (1 << 0),
+       RULE_F_DEFGW_WORKAROUND = (1 << 1),
+};
+
+static int get_route_table(struct relayd_interface *rif)
+{
+       if (rif)
+               return rif->rt_table;
+       else
+               return local_route_table;
+}
 
-static void rtnl_route_set(struct relayd_host *host, bool add)
+static void
+rtnl_rule_request(struct relayd_interface *rif, int flags)
 {
        static struct {
                struct nlmsghdr nl;
                struct rtmsg rt;
                struct {
                        struct rtattr rta;
-                       uint8_t ipaddr[4];
-               } __packed dst;
+                       int table;
+               } __packed table;
                struct {
                        struct rtattr rta;
-                       int ifindex;
+                       char ifname[IFNAMSIZ + 1];
                } __packed dev;
-       } __packed req;
+       } __packed req = {
+               .rt = {
+                       .rtm_family = AF_INET,
+                       .rtm_table = RT_TABLE_UNSPEC,
+                       .rtm_scope = RT_SCOPE_UNIVERSE,
+                       .rtm_protocol = RTPROT_BOOT,
+               },
+               .table.rta = {
+                       .rta_type = FRA_TABLE,
+                       .rta_len = sizeof(req.table),
+               },
+       };
+       const char *ifname = "lo";
+       int padding = sizeof(req.dev.ifname);
 
-       memset(&req, 0, sizeof(req));
+       if (rif)
+               ifname = rif->ifname;
 
-       req.nl.nlmsg_len = sizeof(req);
-       req.rt.rtm_family = AF_INET;
-       req.rt.rtm_dst_len = 32;
+       if (!(flags & RULE_F_DEFGW_WORKAROUND)) {
+               req.dev.rta.rta_type = FRA_IFNAME;
+               padding -= strlen(ifname) + 1;
+               strcpy(req.dev.ifname, ifname);
+               req.dev.rta.rta_len = sizeof(req.dev.rta) + strlen(ifname) + 1;
+       } else {
+               req.dev.rta.rta_type = FRA_PRIORITY;
+               req.dev.rta.rta_len = sizeof(req.dev.rta) + sizeof(uint32_t);
+               padding -= sizeof(uint32_t);
+               *((uint32_t *) &req.dev.ifname) = 1;
+       }
+       req.table.table = get_route_table(rif);
+       req.nl.nlmsg_len = sizeof(req) - padding;
 
-       req.dst.rta.rta_type = RTA_DST;
-       req.dst.rta.rta_len = sizeof(req.dst);
-       memcpy(req.dst.ipaddr, host->ipaddr, sizeof(req.dst.ipaddr));
+       req.nl.nlmsg_flags = NLM_F_REQUEST;
+       if (flags & RULE_F_ADD) {
+               req.nl.nlmsg_type = RTM_NEWRULE;
+               req.nl.nlmsg_flags |= NLM_F_CREATE | NLM_F_EXCL;
+
+               req.rt.rtm_type = RTN_UNICAST;
+       } else {
+               req.nl.nlmsg_type = RTM_DELRULE;
+               req.rt.rtm_type = RTN_UNSPEC;
+       }
+
+       send(rtnl_sock.fd, &req, req.nl.nlmsg_len, 0);
+       rtnl_flush();
+}
+
+struct rtnl_addr {
+       struct rtattr rta;
+       uint8_t ipaddr[4];
+} __packed;
+
+static struct rtnl_addr *
+rtnl_add_addr(struct rtnl_addr *addr, int *len, int type, const uint8_t *ipaddr)
+{
+       addr->rta.rta_type = type;
+       memcpy(addr->ipaddr, ipaddr, 4);
+       *len += sizeof(*addr);
+       return addr + 1;
+}
+
+static void
+rtnl_route_request(struct relayd_interface *rif, struct relayd_host *host,
+                  struct relayd_route *route, bool add)
+{
+       static struct {
+               struct nlmsghdr nl;
+               struct rtmsg rt;
+               struct {
+                       struct rtattr rta;
+                       int table;
+               } __packed table;
+               struct {
+                       struct rtattr rta;
+                       int ifindex;
+               } __packed dev;
+               struct rtnl_addr addr[3];
+       } __packed req = {
+               .rt = {
+                       .rtm_family = AF_INET,
+                       .rtm_dst_len = 32,
+                       .rtm_table = RT_TABLE_MAIN,
+               },
+               .table.rta = {
+                       .rta_type = RTA_TABLE,
+                       .rta_len = sizeof(req.table),
+               },
+               .dev.rta = {
+                       .rta_type = RTA_OIF,
+                       .rta_len = sizeof(req.dev),
+               },
+               .addr[0].rta.rta_len = sizeof(struct rtnl_addr),
+               .addr[1].rta.rta_len = sizeof(struct rtnl_addr),
+               .addr[2].rta.rta_len = sizeof(struct rtnl_addr),
+       };
+       int pktlen = sizeof(req) - sizeof(req.addr);
+       struct rtnl_addr *addr = &req.addr[0];
+       const char *ifname = "loopback";
 
-       req.dev.rta.rta_type = RTA_OIF;
-       req.dev.rta.rta_len = sizeof(req.dev);
        req.dev.ifindex = host->rif->sll.sll_ifindex;
+       req.table.table = get_route_table(rif);
 
        req.nl.nlmsg_flags = NLM_F_REQUEST;
-       req.rt.rtm_table = RT_TABLE_MAIN;
        if (add) {
                req.nl.nlmsg_type = RTM_NEWROUTE;
                req.nl.nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
 
                req.rt.rtm_protocol = RTPROT_BOOT;
-               req.rt.rtm_scope = RT_SCOPE_LINK;
+               if (route) {
+                       req.rt.rtm_scope = RT_SCOPE_UNIVERSE;
+               } else {
+                       req.rt.rtm_scope = RT_SCOPE_LINK;
+               }
                req.rt.rtm_type = RTN_UNICAST;
        } else {
                req.nl.nlmsg_type = RTM_DELROUTE;
                req.rt.rtm_scope = RT_SCOPE_NOWHERE;
        }
 
-       send(rtnl_sock.fd, &req, sizeof(req), 0);
+       if (rif)
+               ifname = rif->ifname;
+
+       if (route) {
+               DPRINTF(2, "%s: add route to "IP_FMT"/%d via "IP_FMT" (%s)\n", ifname,
+                       IP_BUF(route->dest), route->mask, IP_BUF(host->ipaddr),
+                       host->rif->ifname);
+
+               req.rt.rtm_dst_len = route->mask;
+               if (route->mask)
+                       addr = rtnl_add_addr(addr, &pktlen, RTA_DST, route->dest);
+               addr = rtnl_add_addr(addr, &pktlen, RTA_GATEWAY, host->ipaddr);
+       } else {
+               DPRINTF(2, "%s: add host route to "IP_FMT" (%s)\n", ifname,
+                       IP_BUF(host->ipaddr), host->rif->ifname);
+               addr = rtnl_add_addr(addr, &pktlen, RTA_DST, host->ipaddr);
+               req.rt.rtm_dst_len = 32;
+       }
+
+       /* local route */
+       if (!rif)
+               addr = rtnl_add_addr(addr, &pktlen, RTA_PREFSRC, local_addr);
+
+       req.nl.nlmsg_len = pktlen;
+       if (route)
+               rtnl_rule_request(rif, RULE_F_DEFGW_WORKAROUND | RULE_F_ADD);
+       send(rtnl_sock.fd, &req, pktlen, 0);
+       if (route)
+               rtnl_rule_request(rif, RULE_F_DEFGW_WORKAROUND);
+       rtnl_flush();
+}
+
+void
+rtnl_route_set(struct relayd_host *host, struct relayd_route *route, bool add)
+{
+       struct relayd_interface *rif;
+
+       list_for_each_entry(rif, &interfaces, list) {
+               if (rif == host->rif)
+                       continue;
+
+               rtnl_route_request(rif, host, route, add);
+       }
+       if (local_route_table)
+               rtnl_route_request(NULL, host, route, add);
 }
 
-void relayd_add_route(struct relayd_host *host)
+void relayd_add_interface_routes(struct relayd_interface *rif)
 {
-       rtnl_route_set(host, true);
+       rif->rt_table = route_table++;
+       rtnl_rule_request(rif, RULE_F_ADD);
 }
 
-void relayd_del_route(struct relayd_host *host)
+void relayd_del_interface_routes(struct relayd_interface *rif)
 {
-       rtnl_route_set(host, false);
+       rtnl_rule_request(rif, 0);
 }
 
 #ifndef NDA_RTA
@@ -184,21 +348,28 @@ static void rtnl_cb(struct uloop_fd *fd, unsigned int events)
        } while (1);
 }
 
-int relayd_rtnl_init(void)
+static void rtnl_dump_request(int nlmsg_type)
 {
-       struct sockaddr_nl snl_local;
        static struct {
                struct nlmsghdr nlh;
                struct rtgenmsg g;
        } req = {
                .nlh = {
                        .nlmsg_len = sizeof(req),
-                       .nlmsg_type = RTM_GETNEIGH,
                        .nlmsg_flags = NLM_F_ROOT|NLM_F_MATCH|NLM_F_REQUEST,
                        .nlmsg_pid = 0,
                },
                .g.rtgen_family = AF_INET,
        };
+       req.nlh.nlmsg_type = nlmsg_type;
+       req.nlh.nlmsg_seq = rtnl_seq;
+       send(rtnl_sock.fd, &req, sizeof(req), 0);
+       rtnl_seq++;
+}
+
+int relayd_rtnl_init(void)
+{
+       struct sockaddr_nl snl_local;
 
        rtnl_sock.fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
        if (rtnl_sock.fd < 0) {
@@ -219,14 +390,15 @@ int relayd_rtnl_init(void)
 
        rtnl_seq = time(NULL);
        rtnl_dump_seq = rtnl_seq;
-       req.nlh.nlmsg_seq = rtnl_seq;
-       send(rtnl_sock.fd, &req, sizeof(req), 0);
+       rtnl_dump_request(RTM_GETNEIGH);
+       rtnl_rule_request(NULL, RULE_F_ADD);
 
        return 0;
 }
 
 void relayd_rtnl_done(void)
 {
+       rtnl_rule_request(NULL, 0);
        uloop_fd_delete(&rtnl_sock);
        close(rtnl_sock.fd);
 }