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 #include <time.h>
  25
  26 #include <linux/fib_rules.h>
  27
  28 #include "relayd.h"
  29
  30 static struct uloop_fd rtnl_sock;
  31 static unsigned int rtnl_seq, rtnl_dump_seq;
  32 int route_table = 16800;
  33
  34 static void rtnl_flush(void)
  35 {
  36         int fd;
  37
  38         fd = open("/proc/sys/net/ipv4/route/flush", O_WRONLY);
  39         if (fd < 0)
  40                 return;
  41
  42         write(fd, "-1", 2);
  43         close(fd);
  44 }
  45
  46 enum {
  47         RULE_F_ADD = (1 << 0),
  48         RULE_F_DEFGW_WORKAROUND = (1 << 1),
  49 };
  50
  51 static int get_route_table(struct relayd_interface *rif)
  52 {
  53         if (rif)
  54                 return rif->rt_table;
  55         else
  56                 return local_route_table;
  57 }
  58
  59 static void
  60 rtnl_rule_request(struct relayd_interface *rif, int flags)
  61 {
  62         static struct {
  63                 struct nlmsghdr nl;
  64                 struct rtmsg rt;
  65                 struct {
  66                         struct rtattr rta;
  67                         int table;
  68                 } __packed table;
  69                 struct {
  70                         struct rtattr rta;
  71                         char ifname[IFNAMSIZ + 1];
  72                 } __packed dev;
  73         } __packed req = {
  74                 .rt = {
  75                         .rtm_family = AF_INET,
  76                         .rtm_table = RT_TABLE_UNSPEC,
  77                         .rtm_scope = RT_SCOPE_UNIVERSE,
  78                         .rtm_protocol = RTPROT_BOOT,
  79                 },
  80                 .table.rta = {
  81                         .rta_type = FRA_TABLE,
  82                         .rta_len = sizeof(req.table),
  83                 },
  84         };
  85         const char *ifname = "lo";
  86         int padding = sizeof(req.dev.ifname);
  87
  88         if (rif)
  89                 ifname = rif->ifname;
  90
  91         if (!(flags & RULE_F_DEFGW_WORKAROUND)) {
  92                 req.dev.rta.rta_type = FRA_IFNAME;
  93                 padding -= strlen(ifname) + 1;
  94                 strcpy(req.dev.ifname, ifname);
  95                 req.dev.rta.rta_len = sizeof(req.dev.rta) + strlen(ifname) + 1;
  96         } else {
  97                 uint32_t val = 1;
  98                 req.dev.rta.rta_type = FRA_PRIORITY;
  99                 req.dev.rta.rta_len = sizeof(req.dev.rta) + sizeof(uint32_t);
 100                 padding -= sizeof(uint32_t);
 101                 memcpy(&req.dev.ifname, &val, sizeof(val));
 102         }
 103         req.table.table = get_route_table(rif);
 104         req.nl.nlmsg_len = sizeof(req) - padding;
 105
 106         req.nl.nlmsg_flags = NLM_F_REQUEST;
 107         if (flags & RULE_F_ADD) {
 108                 req.nl.nlmsg_type = RTM_NEWRULE;
 109                 req.nl.nlmsg_flags |= NLM_F_CREATE | NLM_F_EXCL;
 110
 111                 req.rt.rtm_type = RTN_UNICAST;
 112         } else {
 113                 req.nl.nlmsg_type = RTM_DELRULE;
 114                 req.rt.rtm_type = RTN_UNSPEC;
 115         }
 116
 117         send(rtnl_sock.fd, &req, req.nl.nlmsg_len, 0);
 118         rtnl_flush();
 119 }
 120
 121 struct rtnl_addr {
 122         struct rtattr rta;
 123         uint8_t ipaddr[4];
 124 } __packed;
 125
 126 static struct rtnl_addr *
 127 rtnl_add_addr(struct rtnl_addr *addr, int *len, int type, const uint8_t *ipaddr)
 128 {
 129         addr->rta.rta_type = type;
 130         memcpy(addr->ipaddr, ipaddr, 4);
 131         *len += sizeof(*addr);
 132         return addr + 1;
 133 }
 134
 135 static void
 136 rtnl_route_request(struct relayd_interface *rif, struct relayd_host *host,
 137                    struct relayd_route *route, bool add)
 138 {
 139         static struct {
 140                 struct nlmsghdr nl;
 141                 struct rtmsg rt;
 142                 struct {
 143                         struct rtattr rta;
 144                         int table;
 145                 } __packed table;
 146                 struct {
 147                         struct rtattr rta;
 148                         int ifindex;
 149                 } __packed dev;
 150                 struct rtnl_addr addr[3];
 151         } __packed req = {
 152                 .rt = {
 153                         .rtm_family = AF_INET,
 154                         .rtm_dst_len = 32,
 155                         .rtm_table = RT_TABLE_MAIN,
 156                 },
 157                 .table.rta = {
 158                         .rta_type = RTA_TABLE,
 159                         .rta_len = sizeof(req.table),
 160                 },
 161                 .dev.rta = {
 162                         .rta_type = RTA_OIF,
 163                         .rta_len = sizeof(req.dev),
 164                 },
 165                 .addr[0].rta.rta_len = sizeof(struct rtnl_addr),
 166                 .addr[1].rta.rta_len = sizeof(struct rtnl_addr),
 167                 .addr[2].rta.rta_len = sizeof(struct rtnl_addr),
 168         };
 169         int pktlen = sizeof(req) - sizeof(req.addr);
 170         struct rtnl_addr *addr = &req.addr[0];
 171         const char *ifname = "loopback";
 172
 173         req.dev.ifindex = host->rif->sll.sll_ifindex;
 174         req.table.table = get_route_table(rif);
 175
 176         req.nl.nlmsg_flags = NLM_F_REQUEST;
 177         if (add) {
 178                 req.nl.nlmsg_type = RTM_NEWROUTE;
 179                 req.nl.nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
 180
 181                 req.rt.rtm_protocol = RTPROT_BOOT;
 182                 if (route) {
 183                         req.rt.rtm_scope = RT_SCOPE_UNIVERSE;
 184                 } else {
 185                         req.rt.rtm_scope = RT_SCOPE_LINK;
 186                 }
 187                 req.rt.rtm_type = RTN_UNICAST;
 188         } else {
 189                 req.nl.nlmsg_type = RTM_DELROUTE;
 190                 req.rt.rtm_scope = RT_SCOPE_NOWHERE;
 191         }
 192
 193         if (rif)
 194                 ifname = rif->ifname;
 195
 196         if (route) {
 197                 DPRINTF(2, "%s: add route to "IP_FMT"/%d via "IP_FMT" (%s)\n", ifname,
 198                         IP_BUF(route->dest), route->mask, IP_BUF(host->ipaddr),
 199                         host->rif->ifname);
 200
 201                 req.rt.rtm_dst_len = route->mask;
 202                 if (route->mask)
 203                         addr = rtnl_add_addr(addr, &pktlen, RTA_DST, route->dest);
 204                 addr = rtnl_add_addr(addr, &pktlen, RTA_GATEWAY, host->ipaddr);
 205         } else {
 206                 DPRINTF(2, "%s: add host route to "IP_FMT" (%s)\n", ifname,
 207                         IP_BUF(host->ipaddr), host->rif->ifname);
 208                 addr = rtnl_add_addr(addr, &pktlen, RTA_DST, host->ipaddr);
 209                 req.rt.rtm_dst_len = 32;
 210         }
 211
 212         /* local route */
 213         if (!rif)
 214                 addr = rtnl_add_addr(addr, &pktlen, RTA_PREFSRC, local_addr);
 215
 216         req.nl.nlmsg_len = pktlen;
 217         if (route)
 218                 rtnl_rule_request(rif, RULE_F_DEFGW_WORKAROUND | RULE_F_ADD);
 219         send(rtnl_sock.fd, &req, pktlen, 0);
 220         if (route)
 221                 rtnl_rule_request(rif, RULE_F_DEFGW_WORKAROUND);
 222         rtnl_flush();
 223 }
 224
 225 void
 226 rtnl_route_set(struct relayd_host *host, struct relayd_route *route, bool add)
 227 {
 228         struct relayd_interface *rif;
 229
 230         list_for_each_entry(rif, &interfaces, list) {
 231                 if (rif == host->rif)
 232                         continue;
 233
 234                 rtnl_route_request(rif, host, route, add);
 235         }
 236         if (local_route_table)
 237                 rtnl_route_request(NULL, host, route, add);
 238 }
 239
 240 void relayd_add_interface_routes(struct relayd_interface *rif)
 241 {
 242         rif->rt_table = route_table++;
 243         rtnl_rule_request(rif, RULE_F_ADD);
 244 }
 245
 246 void relayd_del_interface_routes(struct relayd_interface *rif)
 247 {
 248         rtnl_rule_request(rif, 0);
 249 }
 250
 251 #ifndef NDA_RTA
 252 #define NDA_RTA(r) \
 253     ((struct rtattr*)(((char*)(r)) + NLMSG_ALIGN(sizeof(struct ndmsg))))
 254 #endif
 255
 256 static void rtnl_parse_newneigh(struct nlmsghdr *h)
 257 {
 258         struct relayd_interface *rif = NULL;
 259         struct ndmsg *r = NLMSG_DATA(h);
 260         const uint8_t *lladdr = NULL;
 261         const uint8_t *ipaddr = NULL;
 262         struct rtattr *rta;
 263         int len;
 264
 265         if (r->ndm_family != AF_INET)
 266                 return;
 267
 268         list_for_each_entry(rif, &interfaces, list) {
 269                 if (rif->sll.sll_ifindex == r->ndm_ifindex)
 270                         goto found_interface;
 271         }
 272         return;
 273
 274 found_interface:
 275         len = h->nlmsg_len - NLMSG_LENGTH(sizeof(*r));
 276         for (rta = NDA_RTA(r); RTA_OK(rta, len); rta = RTA_NEXT(rta, len)) {
 277                 switch(rta->rta_type) {
 278                 case NDA_LLADDR:
 279                         lladdr = RTA_DATA(rta);
 280                         break;
 281                 case NDA_DST:
 282                         ipaddr = RTA_DATA(rta);
 283                         break;
 284                 default:
 285                         break;
 286                 }
 287         }
 288
 289         if (!lladdr || !ipaddr || (r->ndm_state & (NUD_INCOMPLETE|NUD_FAILED)))
 290                 return;
 291
 292         if (!memcmp(lladdr, "\x00\x00\x00\x00\x00\x00", ETH_ALEN))
 293                 return;
 294
 295         DPRINTF(1, "%s: Found ARP cache entry for host "IP_FMT" ("MAC_FMT")\n",
 296                 rif->ifname, IP_BUF(ipaddr), MAC_BUF(lladdr));
 297         relayd_refresh_host(rif, lladdr, ipaddr);
 298 }
 299
 300 static void rtnl_parse_packet(void *data, int len)
 301 {
 302         struct nlmsghdr *h;
 303
 304         for (h = data; NLMSG_OK(h, len); h = NLMSG_NEXT(h, len)) {
 305                 if (h->nlmsg_type == NLMSG_DONE ||
 306                     h->nlmsg_type == NLMSG_ERROR)
 307                         return;
 308
 309                 if (h->nlmsg_seq != rtnl_dump_seq)
 310                         continue;
 311
 312                 if (h->nlmsg_type == RTM_NEWNEIGH)
 313                         rtnl_parse_newneigh(h);
 314         }
 315 }
 316
 317 static void rtnl_cb(struct uloop_fd *fd, unsigned int events)
 318 {
 319         struct sockaddr_nl nladdr;
 320         static uint8_t buf[16384];
 321         struct iovec iov = {
 322                 .iov_base = buf,
 323                 .iov_len = sizeof(buf),
 324         };
 325         struct msghdr msg = {
 326                 .msg_name = &nladdr,
 327                 .msg_namelen = sizeof(nladdr),
 328                 .msg_iov = &iov,
 329                 .msg_iovlen = 1,
 330         };
 331
 332         do {
 333                 int len;
 334
 335                 len = recvmsg(rtnl_sock.fd, &msg, 0);
 336                 if (len < 0) {
 337                         if (errno == EINTR)
 338                                 continue;
 339
 340                         return;
 341                 }
 342
 343                 if (!len)
 344                         break;
 345
 346                 if (nladdr.nl_pid != 0)
 347                         continue;
 348
 349                 rtnl_parse_packet(buf, len);
 350         } while (1);
 351 }
 352
 353 static void rtnl_dump_request(int nlmsg_type)
 354 {
 355         static struct {
 356                 struct nlmsghdr nlh;
 357                 struct rtgenmsg g;
 358         } req = {
 359                 .nlh = {
 360                         .nlmsg_len = sizeof(req),
 361                         .nlmsg_flags = NLM_F_ROOT|NLM_F_MATCH|NLM_F_REQUEST,
 362                         .nlmsg_pid = 0,
 363                 },
 364                 .g.rtgen_family = AF_INET,
 365         };
 366         req.nlh.nlmsg_type = nlmsg_type;
 367         req.nlh.nlmsg_seq = rtnl_seq;
 368         send(rtnl_sock.fd, &req, sizeof(req), 0);
 369         rtnl_seq++;
 370 }
 371
 372 int relayd_rtnl_init(void)
 373 {
 374         struct sockaddr_nl snl_local;
 375
 376         rtnl_sock.fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
 377         if (rtnl_sock.fd < 0) {
 378                 perror("socket(AF_NETLINK)");
 379                 return -1;
 380         }
 381
 382         snl_local.nl_family = AF_NETLINK;
 383
 384         if (bind(rtnl_sock.fd, (struct sockaddr *) &snl_local, sizeof(struct sockaddr_nl)) < 0) {
 385                 perror("bind");
 386                 close(rtnl_sock.fd);
 387                 return -1;
 388         }
 389
 390         rtnl_sock.cb = rtnl_cb;
 391         uloop_fd_add(&rtnl_sock, ULOOP_READ | ULOOP_EDGE_TRIGGER);
 392
 393         rtnl_seq = time(NULL);
 394         rtnl_dump_seq = rtnl_seq;
 395         rtnl_dump_request(RTM_GETNEIGH);
 396         rtnl_rule_request(NULL, RULE_F_ADD);
 397
 398         return 0;
 399 }
 400
 401 void relayd_rtnl_done(void)
 402 {
 403         rtnl_rule_request(NULL, 0);
 404         uloop_fd_delete(&rtnl_sock);
 405         close(rtnl_sock.fd);
 406 }