route.c

   1 /*
   2  *   Copyright (C) 2010 Felix Fietkau <nbd@openwrt.org>
   3  *
   4  *   This program is free software; you can redistribute it and/or modify
   5  *   it under the terms of the GNU General Public License v2 as published by
   6  *   the Free Software Foundation.
   7  *
   8  *   This program is distributed in the hope that it will be useful,
   9  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  *   GNU General Public License for more details.
  12  *
  13  *   You should have received a copy of the GNU General Public License
  14  *   along with this program; if not, write to the Free Software
  15  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307, USA.
  16  */
  17
  18 #include <sys/socket.h>
  19 #include <stdio.h>
  20 #include <string.h>
  21 #include <errno.h>
  22 #include <unistd.h>
  23 #include <fcntl.h>
  24
  25 #include <linux/fib_rules.h>
  26
  27 #include "relayd.h"
  28
  29 static struct uloop_fd rtnl_sock;
  30 static unsigned int rtnl_seq, rtnl_dump_seq;
  31 int route_table = 16800;
  32
  33 static void rtnl_flush(void)
  34 {
  35         int fd;
  36
  37         fd = open("/proc/sys/net/ipv4/route/flush", O_WRONLY);
  38         if (fd < 0)
  39                 return;
  40
  41         write(fd, "-1", 2);
  42         close(fd);
  43 }
  44
  45 static void
  46 rtnl_rule_request(struct relayd_interface *rif, bool add, bool defgw_workaround)
  47 {
  48         static struct {
  49                 struct nlmsghdr nl;
  50                 struct rtmsg rt;
  51                 struct {
  52                         struct rtattr rta;
  53                         int table;
  54                 } __packed table;
  55                 struct {
  56                         struct rtattr rta;
  57                         char ifname[IFNAMSIZ + 1];
  58                 } __packed dev;
  59         } __packed req = {
  60                 .rt = {
  61                         .rtm_family = AF_INET,
  62                         .rtm_table = RT_TABLE_UNSPEC,
  63                         .rtm_scope = RT_SCOPE_UNIVERSE,
  64                         .rtm_protocol = RTPROT_BOOT,
  65                 },
  66                 .table.rta = {
  67                         .rta_type = FRA_TABLE,
  68                         .rta_len = sizeof(req.table),
  69                 },
  70         };
  71
  72         int padding = sizeof(req.dev.ifname);
  73
  74         if (!defgw_workaround) {
  75                 req.dev.rta.rta_type = FRA_IFNAME;
  76                 padding -= strlen(rif->ifname) + 1;
  77                 strcpy(req.dev.ifname, rif->ifname);
  78                 req.dev.rta.rta_len = sizeof(req.dev.rta) + strlen(rif->ifname) + 1;
  79         } else {
  80                 req.dev.rta.rta_type = FRA_PRIORITY;
  81                 req.dev.rta.rta_len = sizeof(req.dev.rta) + sizeof(uint32_t);
  82                 padding -= sizeof(uint32_t);
  83                 *((uint32_t *) &req.dev.ifname) = 1;
  84         }
  85         req.table.table = rif->rt_table;
  86         req.nl.nlmsg_len = sizeof(req) - padding;
  87
  88         req.nl.nlmsg_flags = NLM_F_REQUEST;
  89         if (add) {
  90                 req.nl.nlmsg_type = RTM_NEWRULE;
  91                 req.nl.nlmsg_flags |= NLM_F_CREATE | NLM_F_EXCL;
  92
  93                 req.rt.rtm_type = RTN_UNICAST;
  94         } else {
  95                 req.nl.nlmsg_type = RTM_DELRULE;
  96                 req.rt.rtm_type = RTN_UNSPEC;
  97         }
  98
  99         send(rtnl_sock.fd, &req, req.nl.nlmsg_len, 0);
 100         rtnl_flush();
 101 }
 102
 103
 104 static void
 105 rtnl_route_request(struct relayd_interface *rif, struct relayd_host *host,
 106                    struct relayd_route *route, bool add)
 107 {
 108         static struct {
 109                 struct nlmsghdr nl;
 110                 struct rtmsg rt;
 111                 struct {
 112                         struct rtattr rta;
 113                         uint8_t ipaddr[4];
 114                 } __packed dst;
 115                 struct {
 116                         struct rtattr rta;
 117                         int table;
 118                 } __packed table;
 119                 struct {
 120                         struct rtattr rta;
 121                         int ifindex;
 122                 } __packed dev;
 123                 struct {
 124                         struct rtattr rta;
 125                         uint8_t ipaddr[4];
 126                 } __packed gw;
 127         } __packed req = {
 128                 .rt = {
 129                         .rtm_family = AF_INET,
 130                         .rtm_dst_len = 32,
 131                         .rtm_table = RT_TABLE_MAIN,
 132                 },
 133                 .table.rta = {
 134                         .rta_type = RTA_TABLE,
 135                         .rta_len = sizeof(req.table),
 136                 },
 137                 .dst.rta = {
 138                         .rta_len = sizeof(req.dst),
 139                 },
 140                 .dev.rta = {
 141                         .rta_type = RTA_OIF,
 142                         .rta_len = sizeof(req.dev),
 143                 },
 144                 .gw.rta = {
 145                         .rta_type = RTA_GATEWAY,
 146                         .rta_len = sizeof(req.gw),
 147                 },
 148         };
 149         int pktlen = sizeof(req);
 150
 151         req.dev.ifindex = host->rif->sll.sll_ifindex;
 152         req.table.table = rif->rt_table;
 153
 154         req.nl.nlmsg_flags = NLM_F_REQUEST;
 155         if (add) {
 156                 req.nl.nlmsg_type = RTM_NEWROUTE;
 157                 req.nl.nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
 158
 159                 req.rt.rtm_protocol = RTPROT_BOOT;
 160                 if (route) {
 161                         req.rt.rtm_scope = RT_SCOPE_UNIVERSE;
 162                 } else {
 163                         req.rt.rtm_scope = RT_SCOPE_LINK;
 164                 }
 165                 req.rt.rtm_type = RTN_UNICAST;
 166         } else {
 167                 req.nl.nlmsg_type = RTM_DELROUTE;
 168                 req.rt.rtm_scope = RT_SCOPE_NOWHERE;
 169         }
 170
 171         req.dst.rta.rta_type = RTA_DST;
 172         if (route) {
 173                 DPRINTF(2, "%s: add route to "IP_FMT"/%d via "IP_FMT"\n",
 174                         host->rif->ifname, IP_BUF(route->dest), route->mask, IP_BUF(host->ipaddr));
 175                 if (!route->mask) {
 176                         /* No DST -> reuse the DST attribute as the gateway attribute */
 177                         memcpy(req.dst.ipaddr, host->ipaddr, sizeof(req.dst.ipaddr));
 178                         req.dst.rta.rta_type = RTA_GATEWAY;
 179                         pktlen -= sizeof(req.gw);
 180                         req.rt.rtm_dst_len = 0;
 181                 } else {
 182                         memcpy(req.gw.ipaddr, host->ipaddr, sizeof(req.gw.ipaddr));
 183                         memcpy(req.dst.ipaddr, route->dest, sizeof(req.dst.ipaddr));
 184                         req.rt.rtm_dst_len = route->mask;
 185                 }
 186         } else {
 187                 pktlen -= sizeof(req.gw);
 188                 req.rt.rtm_dst_len = 32;
 189                 memcpy(req.dst.ipaddr, host->ipaddr, sizeof(req.dst.ipaddr));
 190         }
 191
 192         req.nl.nlmsg_len = pktlen;
 193         if (route)
 194                 rtnl_rule_request(rif, true, true);
 195         send(rtnl_sock.fd, &req, pktlen, 0);
 196         if (route)
 197                 rtnl_rule_request(rif, false, true);
 198         rtnl_flush();
 199 }
 200
 201 void
 202 rtnl_route_set(struct relayd_host *host, struct relayd_route *route, bool add)
 203 {
 204         struct relayd_interface *rif;
 205
 206         list_for_each_entry(rif, &interfaces, list) {
 207                 if (rif == host->rif)
 208                         continue;
 209
 210                 rtnl_route_request(rif, host, route, add);
 211         }
 212 }
 213
 214 void relayd_add_interface_routes(struct relayd_interface *rif)
 215 {
 216         rif->rt_table = route_table++;
 217         rtnl_rule_request(rif, true, false);
 218 }
 219
 220 void relayd_del_interface_routes(struct relayd_interface *rif)
 221 {
 222         rtnl_rule_request(rif, false, false);
 223 }
 224
 225 #ifndef NDA_RTA
 226 #define NDA_RTA(r) \
 227     ((struct rtattr*)(((char*)(r)) + NLMSG_ALIGN(sizeof(struct ndmsg))))
 228 #endif
 229
 230 static void rtnl_parse_newneigh(struct nlmsghdr *h)
 231 {
 232         struct relayd_interface *rif = NULL;
 233         struct ndmsg *r = NLMSG_DATA(h);
 234         const uint8_t *lladdr = NULL;
 235         const uint8_t *ipaddr = NULL;
 236         struct rtattr *rta;
 237         int len;
 238
 239         if (r->ndm_family != AF_INET)
 240                 return;
 241
 242         list_for_each_entry(rif, &interfaces, list) {
 243                 if (rif->sll.sll_ifindex == r->ndm_ifindex)
 244                         goto found_interface;
 245         }
 246         return;
 247
 248 found_interface:
 249         len = h->nlmsg_len - NLMSG_LENGTH(sizeof(*r));
 250         for (rta = NDA_RTA(r); RTA_OK(rta, len); rta = RTA_NEXT(rta, len)) {
 251                 switch(rta->rta_type) {
 252                 case NDA_LLADDR:
 253                         lladdr = RTA_DATA(rta);
 254                         break;
 255                 case NDA_DST:
 256                         ipaddr = RTA_DATA(rta);
 257                         break;
 258                 default:
 259                         break;
 260                 }
 261         }
 262
 263         if (!lladdr || !ipaddr || (r->ndm_state & (NUD_INCOMPLETE|NUD_FAILED)))
 264                 return;
 265
 266         if (!memcmp(lladdr, "\x00\x00\x00\x00\x00\x00", ETH_ALEN))
 267                 return;
 268
 269         DPRINTF(1, "%s: Found ARP cache entry for host "IP_FMT" ("MAC_FMT")\n",
 270                 rif->ifname, IP_BUF(ipaddr), MAC_BUF(lladdr));
 271         relayd_refresh_host(rif, lladdr, ipaddr);
 272 }
 273
 274 static void rtnl_parse_packet(void *data, int len)
 275 {
 276         struct nlmsghdr *h;
 277
 278         for (h = data; NLMSG_OK(h, len); h = NLMSG_NEXT(h, len)) {
 279                 if (h->nlmsg_type == NLMSG_DONE ||
 280                     h->nlmsg_type == NLMSG_ERROR)
 281                         return;
 282
 283                 if (h->nlmsg_seq != rtnl_dump_seq)
 284                         continue;
 285
 286                 if (h->nlmsg_type == RTM_NEWNEIGH)
 287                         rtnl_parse_newneigh(h);
 288         }
 289 }
 290
 291 static void rtnl_cb(struct uloop_fd *fd, unsigned int events)
 292 {
 293         struct sockaddr_nl nladdr;
 294         static uint8_t buf[16384];
 295         struct iovec iov = {
 296                 .iov_base = buf,
 297                 .iov_len = sizeof(buf),
 298         };
 299         struct msghdr msg = {
 300                 .msg_name = &nladdr,
 301                 .msg_namelen = sizeof(nladdr),
 302                 .msg_iov = &iov,
 303                 .msg_iovlen = 1,
 304         };
 305
 306         do {
 307                 int len;
 308
 309                 len = recvmsg(rtnl_sock.fd, &msg, 0);
 310                 if (len < 0) {
 311                         if (errno == EINTR)
 312                                 continue;
 313
 314                         return;
 315                 }
 316
 317                 if (!len)
 318                         break;
 319
 320                 if (nladdr.nl_pid != 0)
 321                         continue;
 322
 323                 rtnl_parse_packet(buf, len);
 324         } while (1);
 325 }
 326
 327 int relayd_rtnl_init(void)
 328 {
 329         struct sockaddr_nl snl_local;
 330         static struct {
 331                 struct nlmsghdr nlh;
 332                 struct rtgenmsg g;
 333         } req = {
 334                 .nlh = {
 335                         .nlmsg_len = sizeof(req),
 336                         .nlmsg_type = RTM_GETNEIGH,
 337                         .nlmsg_flags = NLM_F_ROOT|NLM_F_MATCH|NLM_F_REQUEST,
 338                         .nlmsg_pid = 0,
 339                 },
 340                 .g.rtgen_family = AF_INET,
 341         };
 342
 343         rtnl_sock.fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
 344         if (rtnl_sock.fd < 0) {
 345                 perror("socket(AF_NETLINK)");
 346                 return -1;
 347         }
 348
 349         snl_local.nl_family = AF_NETLINK;
 350
 351         if (bind(rtnl_sock.fd, (struct sockaddr *) &snl_local, sizeof(struct sockaddr_nl)) < 0) {
 352                 perror("bind");
 353                 close(rtnl_sock.fd);
 354                 return -1;
 355         }
 356
 357         rtnl_sock.cb = rtnl_cb;
 358         uloop_fd_add(&rtnl_sock, ULOOP_READ | ULOOP_EDGE_TRIGGER);
 359
 360         rtnl_seq = time(NULL);
 361         rtnl_dump_seq = rtnl_seq;
 362         req.nlh.nlmsg_seq = rtnl_seq;
 363         send(rtnl_sock.fd, &req, sizeof(req), 0);
 364
 365         return 0;
 366 }
 367
 368 void relayd_rtnl_done(void)
 369 {
 370         uloop_fd_delete(&rtnl_sock);
 371         close(rtnl_sock.fd);
 372 }