sysdep/linux/netlink.c

   1 /*
   2  *      BIRD -- Linux Netlink Interface
   3  *
   4  *      (c) 1999--2000 Martin Mares <mj@ucw.cz>
   5  *
   6  *      Can be freely distributed and used under the terms of the GNU GPL.
   7  */
   8
   9 #include <stdio.h>
  10 #include <fcntl.h>
  11 #include <sys/socket.h>
  12 #include <sys/uio.h>
  13 #include <errno.h>
  14
  15 #undef LOCAL_DEBUG
  16
  17 #include "nest/bird.h"
  18 #include "nest/route.h"
  19 #include "nest/protocol.h"
  20 #include "nest/iface.h"
  21 #include "lib/alloca.h"
  22 #include "lib/timer.h"
  23 #include "lib/unix.h"
  24 #include "lib/krt.h"
  25 #include "lib/socket.h"
  26 #include "lib/string.h"
  27 #include "conf/conf.h"
  28
  29 #include <asm/types.h>
  30 #include <linux/if.h>
  31 #include <linux/netlink.h>
  32 #include <linux/rtnetlink.h>
  33
  34 #ifndef MSG_TRUNC                       /* Hack: Several versions of glibc miss this one :( */
  35 #define MSG_TRUNC 0x20
  36 #endif
  37
  38 #ifndef IFF_LOWER_UP
  39 #define IFF_LOWER_UP 0x10000
  40 #endif
  41
  42 /*
  43  *      Synchronous Netlink interface
  44  */
  45
  46 struct nl_sock
  47 {
  48   int fd;
  49   u32 seq;
  50   byte *rx_buffer;                      /* Receive buffer */
  51   struct nlmsghdr *last_hdr;            /* Recently received packet */
  52   unsigned int last_size;
  53 };
  54
  55 #define NL_RX_SIZE 8192
  56
  57 static struct nl_sock nl_scan = {.fd = -1};     /* Netlink socket for synchronous scan */
  58 static struct nl_sock nl_req  = {.fd = -1};     /* Netlink socket for requests */
  59
  60 static void
  61 nl_open_sock(struct nl_sock *nl)
  62 {
  63   if (nl->fd < 0)
  64     {
  65       nl->fd = socket(PF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
  66       if (nl->fd < 0)
  67         die("Unable to open rtnetlink socket: %m");
  68       nl->seq = now;
  69       nl->rx_buffer = xmalloc(NL_RX_SIZE);
  70       nl->last_hdr = NULL;
  71       nl->last_size = 0;
  72     }
  73 }
  74
  75 static void
  76 nl_open(void)
  77 {
  78   nl_open_sock(&nl_scan);
  79   nl_open_sock(&nl_req);
  80 }
  81
  82 static void
  83 nl_send(struct nl_sock *nl, struct nlmsghdr *nh)
  84 {
  85   struct sockaddr_nl sa;
  86
  87   memset(&sa, 0, sizeof(sa));
  88   sa.nl_family = AF_NETLINK;
  89   nh->nlmsg_pid = 0;
  90   nh->nlmsg_seq = ++(nl->seq);
  91   if (sendto(nl->fd, nh, nh->nlmsg_len, 0, (struct sockaddr *)&sa, sizeof(sa)) < 0)
  92     die("rtnetlink sendto: %m");
  93   nl->last_hdr = NULL;
  94 }
  95
  96 static void
  97 nl_request_dump(int cmd)
  98 {
  99   struct {
 100     struct nlmsghdr nh;
 101     struct rtgenmsg g;
 102   } req;
 103   req.nh.nlmsg_type = cmd;
 104   req.nh.nlmsg_len = sizeof(req);
 105   req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
 106   /* Is it important which PF_* is used for link-level interface scan?
 107      It seems that some information is available only when PF_INET is used. */
 108   req.g.rtgen_family = (cmd == RTM_GETLINK) ? PF_INET : BIRD_PF;
 109   nl_send(&nl_scan, &req.nh);
 110 }
 111
 112 static struct nlmsghdr *
 113 nl_get_reply(struct nl_sock *nl)
 114 {
 115   for(;;)
 116     {
 117       if (!nl->last_hdr)
 118         {
 119           struct iovec iov = { nl->rx_buffer, NL_RX_SIZE };
 120           struct sockaddr_nl sa;
 121           struct msghdr m = { (struct sockaddr *) &sa, sizeof(sa), &iov, 1, NULL, 0, 0 };
 122           int x = recvmsg(nl->fd, &m, 0);
 123           if (x < 0)
 124             die("nl_get_reply: %m");
 125           if (sa.nl_pid)                /* It isn't from the kernel */
 126             {
 127               DBG("Non-kernel packet\n");
 128               continue;
 129             }
 130           nl->last_size = x;
 131           nl->last_hdr = (void *) nl->rx_buffer;
 132           if (m.msg_flags & MSG_TRUNC)
 133             bug("nl_get_reply: got truncated reply which should be impossible");
 134         }
 135       if (NLMSG_OK(nl->last_hdr, nl->last_size))
 136         {
 137           struct nlmsghdr *h = nl->last_hdr;
 138           nl->last_hdr = NLMSG_NEXT(h, nl->last_size);
 139           if (h->nlmsg_seq != nl->seq)
 140             {
 141               log(L_WARN "nl_get_reply: Ignoring out of sequence netlink packet (%x != %x)",
 142                   h->nlmsg_seq, nl->seq);
 143               continue;
 144             }
 145           return h;
 146         }
 147       if (nl->last_size)
 148         log(L_WARN "nl_get_reply: Found packet remnant of size %d", nl->last_size);
 149       nl->last_hdr = NULL;
 150     }
 151 }
 152
 153 static struct rate_limit rl_netlink_err;
 154
 155 static int
 156 nl_error(struct nlmsghdr *h)
 157 {
 158   struct nlmsgerr *e;
 159   int ec;
 160
 161   if (h->nlmsg_len < NLMSG_LENGTH(sizeof(struct nlmsgerr)))
 162     {
 163       log(L_WARN "Netlink: Truncated error message received");
 164       return ENOBUFS;
 165     }
 166   e = (struct nlmsgerr *) NLMSG_DATA(h);
 167   ec = -e->error;
 168   if (ec)
 169     log_rl(&rl_netlink_err, L_WARN "Netlink: %s", strerror(ec));
 170   return ec;
 171 }
 172
 173 static struct nlmsghdr *
 174 nl_get_scan(void)
 175 {
 176   struct nlmsghdr *h = nl_get_reply(&nl_scan);
 177
 178   if (h->nlmsg_type == NLMSG_DONE)
 179     return NULL;
 180   if (h->nlmsg_type == NLMSG_ERROR)
 181     {
 182       nl_error(h);
 183       return NULL;
 184     }
 185   return h;
 186 }
 187
 188 static int
 189 nl_exchange(struct nlmsghdr *pkt)
 190 {
 191   struct nlmsghdr *h;
 192
 193   nl_send(&nl_req, pkt);
 194   for(;;)
 195     {
 196       h = nl_get_reply(&nl_req);
 197       if (h->nlmsg_type == NLMSG_ERROR)
 198         break;
 199       log(L_WARN "nl_exchange: Unexpected reply received");
 200     }
 201   return nl_error(h) ? -1 : 0;
 202 }
 203
 204 /*
 205  *      Netlink attributes
 206  */
 207
 208 static int nl_attr_len;
 209
 210 static void *
 211 nl_checkin(struct nlmsghdr *h, int lsize)
 212 {
 213   nl_attr_len = h->nlmsg_len - NLMSG_LENGTH(lsize);
 214   if (nl_attr_len < 0)
 215     {
 216       log(L_ERR "nl_checkin: underrun by %d bytes", -nl_attr_len);
 217       return NULL;
 218     }
 219   return NLMSG_DATA(h);
 220 }
 221
 222 static int
 223 nl_parse_attrs(struct rtattr *a, struct rtattr **k, int ksize)
 224 {
 225   int max = ksize / sizeof(struct rtattr *);
 226   bzero(k, ksize);
 227   while (RTA_OK(a, nl_attr_len))
 228     {
 229       if (a->rta_type < max)
 230         k[a->rta_type] = a;
 231       a = RTA_NEXT(a, nl_attr_len);
 232     }
 233   if (nl_attr_len)
 234     {
 235       log(L_ERR "nl_parse_attrs: remnant of size %d", nl_attr_len);
 236       return 0;
 237     }
 238   else
 239     return 1;
 240 }
 241
 242 void
 243 nl_add_attr(struct nlmsghdr *h, unsigned bufsize, unsigned code,
 244             void *data, unsigned dlen)
 245 {
 246   unsigned len = RTA_LENGTH(dlen);
 247   unsigned pos = NLMSG_ALIGN(h->nlmsg_len);
 248   struct rtattr *a;
 249
 250   if (pos + len > bufsize)
 251     bug("nl_add_attr: packet buffer overflow");
 252   a = (struct rtattr *)((char *)h + pos);
 253   a->rta_type = code;
 254   a->rta_len = len;
 255   h->nlmsg_len = pos + len;
 256   memcpy(RTA_DATA(a), data, dlen);
 257 }
 258
 259 static inline void
 260 nl_add_attr_u32(struct nlmsghdr *h, unsigned bufsize, int code, u32 data)
 261 {
 262   nl_add_attr(h, bufsize, code, &data, 4);
 263 }
 264
 265 static inline void
 266 nl_add_attr_ipa(struct nlmsghdr *h, unsigned bufsize, int code, ip_addr ipa)
 267 {
 268   ipa_hton(ipa);
 269   nl_add_attr(h, bufsize, code, &ipa, sizeof(ipa));
 270 }
 271
 272 #define RTNH_SIZE (sizeof(struct rtnexthop) + sizeof(struct rtattr) + sizeof(ip_addr))
 273
 274 static inline void
 275 add_mpnexthop(char *buf, ip_addr ipa, unsigned iface, unsigned char weight)
 276 {
 277   struct rtnexthop *nh = (void *) buf;
 278   struct rtattr *rt = (void *) (buf + sizeof(*nh));
 279   nh->rtnh_len = RTNH_SIZE;
 280   nh->rtnh_flags = 0;
 281   nh->rtnh_hops = weight;
 282   nh->rtnh_ifindex = iface;
 283   rt->rta_len = sizeof(*rt) + sizeof(ipa);
 284   rt->rta_type = RTA_GATEWAY;
 285   ipa_hton(ipa);
 286   memcpy(buf + sizeof(*nh) + sizeof(*rt), &ipa, sizeof(ipa));
 287 }
 288
 289
 290 static void
 291 nl_add_multipath(struct nlmsghdr *h, unsigned bufsize, struct mpnh *nh)
 292 {
 293   unsigned len = sizeof(struct rtattr);
 294   unsigned pos = NLMSG_ALIGN(h->nlmsg_len);
 295   char *buf = (char *)h + pos;
 296   struct rtattr *rt = (void *) buf;
 297   buf += len;
 298
 299   for (; nh; nh = nh->next)
 300     {
 301       len += RTNH_SIZE;
 302       if (pos + len > bufsize)
 303         bug("nl_add_multipath: packet buffer overflow");
 304
 305       add_mpnexthop(buf, nh->gw, nh->iface->index, nh->weight);
 306       buf += RTNH_SIZE;
 307     }
 308
 309   rt->rta_type = RTA_MULTIPATH;
 310   rt->rta_len = len;
 311   h->nlmsg_len = pos + len;
 312 }
 313
 314
 315 static struct mpnh *
 316 nl_parse_multipath(struct krt_proto *p, struct rtattr *ra)
 317 {
 318   /* Temporary buffer for multicast nexthops */
 319   static struct mpnh *nh_buffer;
 320   static int nh_buf_size;       /* in number of structures */
 321   static int nh_buf_used;
 322
 323   struct rtattr *a[RTA_CACHEINFO+1];
 324   struct rtnexthop *nh = RTA_DATA(ra);
 325   struct mpnh *rv, *first, **last;
 326   int len = RTA_PAYLOAD(ra);
 327
 328   first = NULL;
 329   last = &first;
 330   nh_buf_used = 0;
 331
 332   while (len)
 333     {
 334       /* Use RTNH_OK(nh,len) ?? */
 335       if ((len < sizeof(*nh)) || (len < nh->rtnh_len))
 336         return NULL;
 337
 338       if (nh_buf_used == nh_buf_size)
 339       {
 340         nh_buf_size = nh_buf_size ? (nh_buf_size * 2) : 4;
 341         nh_buffer = xrealloc(nh_buffer, nh_buf_size * sizeof(struct mpnh));
 342       }
 343       *last = rv = nh_buffer + nh_buf_used++;
 344       rv->next = NULL;
 345       last = &(rv->next);
 346
 347       rv->weight = nh->rtnh_hops;
 348       rv->iface = if_find_by_index(nh->rtnh_ifindex);
 349       if (!rv->iface)
 350         return NULL;
 351
 352       /* Nonexistent RTNH_PAYLOAD ?? */
 353       nl_attr_len = nh->rtnh_len - RTNH_LENGTH(0);
 354       nl_parse_attrs(RTNH_DATA(nh), a, sizeof(a));
 355       if (a[RTA_GATEWAY])
 356         {
 357           if (RTA_PAYLOAD(a[RTA_GATEWAY]) != sizeof(ip_addr))
 358             return NULL;
 359
 360           memcpy(&rv->gw, RTA_DATA(a[RTA_GATEWAY]), sizeof(ip_addr));
 361           ipa_ntoh(rv->gw);
 362
 363           neighbor *ng = neigh_find2(&p->p, &rv->gw, rv->iface,
 364                                      (nh->rtnh_flags & RTNH_F_ONLINK) ? NEF_ONLINK : 0);
 365           if (!ng || (ng->scope == SCOPE_HOST))
 366             return NULL;
 367         }
 368       else
 369         return NULL;
 370
 371       len -= NLMSG_ALIGN(nh->rtnh_len);
 372       nh = RTNH_NEXT(nh);
 373     }
 374
 375   return first;
 376 }
 377
 378
 379 /*
 380  *      Scanning of interfaces
 381  */
 382
 383 static void
 384 nl_parse_link(struct nlmsghdr *h, int scan)
 385 {
 386   struct ifinfomsg *i;
 387   struct rtattr *a[IFLA_WIRELESS+1];
 388   int new = h->nlmsg_type == RTM_NEWLINK;
 389   struct iface f = {};
 390   struct iface *ifi;
 391   char *name;
 392   u32 mtu;
 393   unsigned int fl;
 394
 395   if (!(i = nl_checkin(h, sizeof(*i))) || !nl_parse_attrs(IFLA_RTA(i), a, sizeof(a)))
 396     return;
 397   if (!a[IFLA_IFNAME] || RTA_PAYLOAD(a[IFLA_IFNAME]) < 2 ||
 398       !a[IFLA_MTU] || RTA_PAYLOAD(a[IFLA_MTU]) != 4)
 399     {
 400       if (scan || !a[IFLA_WIRELESS])
 401         log(L_ERR "nl_parse_link: Malformed message received");
 402       return;
 403     }
 404   name = RTA_DATA(a[IFLA_IFNAME]);
 405   memcpy(&mtu, RTA_DATA(a[IFLA_MTU]), sizeof(u32));
 406
 407   ifi = if_find_by_index(i->ifi_index);
 408   if (!new)
 409     {
 410       DBG("KIF: IF%d(%s) goes down\n", i->ifi_index, name);
 411       if (!ifi)
 412         return;
 413
 414       if_delete(ifi);
 415     }
 416   else
 417     {
 418       DBG("KIF: IF%d(%s) goes up (mtu=%d,flg=%x)\n", i->ifi_index, name, mtu, i->ifi_flags);
 419       if (ifi && strncmp(ifi->name, name, sizeof(ifi->name)-1))
 420         if_delete(ifi);
 421
 422       strncpy(f.name, name, sizeof(f.name)-1);
 423       f.index = i->ifi_index;
 424       f.mtu = mtu;
 425
 426       fl = i->ifi_flags;
 427       if (fl & IFF_UP)
 428         f.flags |= IF_ADMIN_UP;
 429       if (fl & IFF_LOWER_UP)
 430         f.flags |= IF_LINK_UP;
 431       if (fl & IFF_LOOPBACK)            /* Loopback */
 432         f.flags |= IF_MULTIACCESS | IF_LOOPBACK | IF_IGNORE;
 433       else if (fl & IFF_POINTOPOINT)    /* PtP */
 434         f.flags |= IF_MULTICAST;
 435       else if (fl & IFF_BROADCAST)      /* Broadcast */
 436         f.flags |= IF_MULTIACCESS | IF_BROADCAST | IF_MULTICAST;
 437       else
 438         f.flags |= IF_MULTIACCESS;      /* NBMA */
 439       if_update(&f);
 440     }
 441 }
 442
 443 static void
 444 nl_parse_addr(struct nlmsghdr *h)
 445 {
 446   struct ifaddrmsg *i;
 447   struct rtattr *a[IFA_ANYCAST+1];
 448   int new = h->nlmsg_type == RTM_NEWADDR;
 449   struct ifa ifa;
 450   struct iface *ifi;
 451   int scope;
 452
 453   if (!(i = nl_checkin(h, sizeof(*i))) || !nl_parse_attrs(IFA_RTA(i), a, sizeof(a)))
 454     return;
 455   if (i->ifa_family != BIRD_AF)
 456     return;
 457   if (!a[IFA_ADDRESS] || RTA_PAYLOAD(a[IFA_ADDRESS]) != sizeof(ip_addr)
 458 #ifdef IPV6
 459       || a[IFA_LOCAL] && RTA_PAYLOAD(a[IFA_LOCAL]) != sizeof(ip_addr)
 460 #else
 461       || !a[IFA_LOCAL] || RTA_PAYLOAD(a[IFA_LOCAL]) != sizeof(ip_addr)
 462       || (a[IFA_BROADCAST] && RTA_PAYLOAD(a[IFA_BROADCAST]) != sizeof(ip_addr))
 463 #endif
 464       )
 465     {
 466       log(L_ERR "nl_parse_addr: Malformed message received");
 467       return;
 468     }
 469
 470   ifi = if_find_by_index(i->ifa_index);
 471   if (!ifi)
 472     {
 473       log(L_ERR "KIF: Received address message for unknown interface %d", i->ifa_index);
 474       return;
 475     }
 476
 477   bzero(&ifa, sizeof(ifa));
 478   ifa.iface = ifi;
 479   if (i->ifa_flags & IFA_F_SECONDARY)
 480     ifa.flags |= IA_SECONDARY;
 481
 482   /* IFA_LOCAL can be unset for IPv6 interfaces */
 483   memcpy(&ifa.ip, RTA_DATA(a[IFA_LOCAL] ? : a[IFA_ADDRESS]), sizeof(ifa.ip));
 484   ipa_ntoh(ifa.ip);
 485   ifa.pxlen = i->ifa_prefixlen;
 486   if (i->ifa_prefixlen > BITS_PER_IP_ADDRESS)
 487     {
 488       log(L_ERR "KIF: Invalid prefix length for interface %s: %d", ifi->name, i->ifa_prefixlen);
 489       new = 0;
 490     }
 491   if (i->ifa_prefixlen == BITS_PER_IP_ADDRESS)
 492     {
 493       ip_addr addr;
 494       memcpy(&addr, RTA_DATA(a[IFA_ADDRESS]), sizeof(addr));
 495       ipa_ntoh(addr);
 496       ifa.prefix = ifa.brd = addr;
 497
 498       /* It is either a host address or a peer address */
 499       if (ipa_equal(ifa.ip, addr))
 500         ifa.flags |= IA_HOST;
 501       else
 502         {
 503           ifa.flags |= IA_PEER;
 504           ifa.opposite = addr;
 505         }
 506     }
 507   else
 508     {
 509       ip_addr netmask = ipa_mkmask(ifa.pxlen);
 510       ifa.prefix = ipa_and(ifa.ip, netmask);
 511       ifa.brd = ipa_or(ifa.ip, ipa_not(netmask));
 512       if (i->ifa_prefixlen == BITS_PER_IP_ADDRESS - 1)
 513         ifa.opposite = ipa_opposite_m1(ifa.ip);
 514
 515 #ifndef IPV6
 516       if (i->ifa_prefixlen == BITS_PER_IP_ADDRESS - 2)
 517         ifa.opposite = ipa_opposite_m2(ifa.ip);
 518
 519       if ((ifi->flags & IF_BROADCAST) && a[IFA_BROADCAST])
 520         {
 521           ip_addr xbrd;
 522           memcpy(&xbrd, RTA_DATA(a[IFA_BROADCAST]), sizeof(xbrd));
 523           ipa_ntoh(xbrd);
 524           if (ipa_equal(xbrd, ifa.prefix) || ipa_equal(xbrd, ifa.brd))
 525             ifa.brd = xbrd;
 526           else if (ifi->flags & IF_TMP_DOWN) /* Complain only during the first scan */
 527             log(L_ERR "KIF: Invalid broadcast address %I for %s", xbrd, ifi->name);
 528         }
 529 #endif
 530     }
 531
 532   scope = ipa_classify(ifa.ip);
 533   if (scope < 0)
 534     {
 535       log(L_ERR "KIF: Invalid interface address %I for %s", ifa.ip, ifi->name);
 536       return;
 537     }
 538   ifa.scope = scope & IADDR_SCOPE_MASK;
 539
 540   DBG("KIF: IF%d(%s): %s IPA %I, flg %x, net %I/%d, brd %I, opp %I\n",
 541       ifi->index, ifi->name,
 542       new ? "added" : "removed",
 543       ifa.ip, ifa.flags, ifa.prefix, ifa.pxlen, ifa.brd, ifa.opposite);
 544   if (new)
 545     ifa_update(&ifa);
 546   else
 547     ifa_delete(&ifa);
 548 }
 549
 550 void
 551 kif_do_scan(struct kif_proto *p UNUSED)
 552 {
 553   struct nlmsghdr *h;
 554
 555   if_start_update();
 556
 557   nl_request_dump(RTM_GETLINK);
 558   while (h = nl_get_scan())
 559     if (h->nlmsg_type == RTM_NEWLINK || h->nlmsg_type == RTM_DELLINK)
 560       nl_parse_link(h, 1);
 561     else
 562       log(L_DEBUG "nl_scan_ifaces: Unknown packet received (type=%d)", h->nlmsg_type);
 563
 564   nl_request_dump(RTM_GETADDR);
 565   while (h = nl_get_scan())
 566     if (h->nlmsg_type == RTM_NEWADDR || h->nlmsg_type == RTM_DELADDR)
 567       nl_parse_addr(h);
 568     else
 569       log(L_DEBUG "nl_scan_ifaces: Unknown packet received (type=%d)", h->nlmsg_type);
 570
 571   if_end_update();
 572 }
 573
 574 /*
 575  *      Routes
 576  */
 577
 578 static struct krt_proto *nl_table_map[NL_NUM_TABLES];
 579
 580 int
 581 krt_capable(rte *e)
 582 {
 583   rta *a = e->attrs;
 584
 585   if (a->cast != RTC_UNICAST)
 586     return 0;
 587
 588   switch (a->dest)
 589     {
 590     case RTD_ROUTER:
 591     case RTD_DEVICE:
 592       if (a->iface == NULL)
 593         return 0;
 594     case RTD_BLACKHOLE:
 595     case RTD_UNREACHABLE:
 596     case RTD_PROHIBIT:
 597     case RTD_MULTIPATH:
 598       break;
 599     default:
 600       return 0;
 601     }
 602   return 1;
 603 }
 604
 605 static inline int
 606 nh_bufsize(struct mpnh *nh)
 607 {
 608   int rv = 0;
 609   for (; nh != NULL; nh = nh->next)
 610     rv += RTNH_SIZE;
 611   return rv;
 612 }
 613
 614 static int
 615 nl_send_route(struct krt_proto *p, rte *e, struct ea_list *eattrs, int new)
 616 {
 617   eattr *ea;
 618   net *net = e->net;
 619   rta *a = e->attrs;
 620   struct {
 621     struct nlmsghdr h;
 622     struct rtmsg r;
 623     char buf[128 + nh_bufsize(a->nexthops)];
 624   } r;
 625
 626   DBG("nl_send_route(%I/%d,new=%d)\n", net->n.prefix, net->n.pxlen, new);
 627
 628   bzero(&r.h, sizeof(r.h));
 629   bzero(&r.r, sizeof(r.r));
 630   r.h.nlmsg_type = new ? RTM_NEWROUTE : RTM_DELROUTE;
 631   r.h.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg));
 632   r.h.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | (new ? NLM_F_CREATE|NLM_F_EXCL : 0);
 633
 634   r.r.rtm_family = BIRD_AF;
 635   r.r.rtm_dst_len = net->n.pxlen;
 636   r.r.rtm_tos = 0;
 637   r.r.rtm_table = KRT_CF->sys.table_id;
 638   r.r.rtm_protocol = RTPROT_BIRD;
 639   r.r.rtm_scope = RT_SCOPE_UNIVERSE;
 640   nl_add_attr_ipa(&r.h, sizeof(r), RTA_DST, net->n.prefix);
 641
 642   u32 metric = 0;
 643   if (new && e->attrs->source == RTS_INHERIT)
 644     metric = e->u.krt.metric;
 645   if (ea = ea_find(eattrs, EA_KRT_METRIC))
 646     metric = ea->u.data;
 647   if (metric != 0)
 648     nl_add_attr_u32(&r.h, sizeof(r), RTA_PRIORITY, metric);
 649
 650   if (ea = ea_find(eattrs, EA_KRT_PREFSRC))
 651     nl_add_attr_ipa(&r.h, sizeof(r), RTA_PREFSRC, *(ip_addr *)ea->u.ptr->data);
 652
 653   if (ea = ea_find(eattrs, EA_KRT_REALM))
 654     nl_add_attr_u32(&r.h, sizeof(r), RTA_FLOW, ea->u.data);
 655
 656   /* a->iface != NULL checked in krt_capable() for router and device routes */
 657
 658   switch (a->dest)
 659     {
 660     case RTD_ROUTER:
 661       r.r.rtm_type = RTN_UNICAST;
 662       nl_add_attr_u32(&r.h, sizeof(r), RTA_OIF, a->iface->index);
 663       nl_add_attr_ipa(&r.h, sizeof(r), RTA_GATEWAY, a->gw);
 664       break;
 665     case RTD_DEVICE:
 666       r.r.rtm_type = RTN_UNICAST;
 667       nl_add_attr_u32(&r.h, sizeof(r), RTA_OIF, a->iface->index);
 668       break;
 669     case RTD_BLACKHOLE:
 670       r.r.rtm_type = RTN_BLACKHOLE;
 671       break;
 672     case RTD_UNREACHABLE:
 673       r.r.rtm_type = RTN_UNREACHABLE;
 674       break;
 675     case RTD_PROHIBIT:
 676       r.r.rtm_type = RTN_PROHIBIT;
 677       break;
 678     case RTD_MULTIPATH:
 679       r.r.rtm_type = RTN_UNICAST;
 680       nl_add_multipath(&r.h, sizeof(r), a->nexthops);
 681       break;
 682     default:
 683       bug("krt_capable inconsistent with nl_send_route");
 684     }
 685
 686   return nl_exchange(&r.h);
 687 }
 688
 689 void
 690 krt_replace_rte(struct krt_proto *p, net *n, rte *new, rte *old, struct ea_list *eattrs)
 691 {
 692   int err = 0;
 693
 694   /*
 695    * NULL for eattr of the old route is a little hack, but we don't
 696    * get proper eattrs for old in rt_notify() anyway. NULL means no
 697    * extended route attributes and therefore matches if the kernel
 698    * route has any of them.
 699    */
 700
 701   if (old)
 702     nl_send_route(p, old, NULL, 0);
 703
 704   if (new)
 705     err = nl_send_route(p, new, eattrs, 1);
 706
 707   if (err < 0)
 708     n->n.flags |= KRF_SYNC_ERROR;
 709   else
 710     n->n.flags &= ~KRF_SYNC_ERROR;
 711 }
 712
 713
 714 #define SKIP(ARG...) do { DBG("KRT: Ignoring route - " ARG); return; } while(0)
 715
 716 static void
 717 nl_parse_route(struct nlmsghdr *h, int scan)
 718 {
 719   struct krt_proto *p;
 720   struct rtmsg *i;
 721   struct rtattr *a[RTA_CACHEINFO+1];
 722   int new = h->nlmsg_type == RTM_NEWROUTE;
 723
 724   ip_addr dst = IPA_NONE;
 725   u32 oif = ~0;
 726   int src;
 727
 728   if (!(i = nl_checkin(h, sizeof(*i))) || !nl_parse_attrs(RTM_RTA(i), a, sizeof(a)))
 729     return;
 730   if (i->rtm_family != BIRD_AF)
 731     return;
 732   if ((a[RTA_DST] && RTA_PAYLOAD(a[RTA_DST]) != sizeof(ip_addr)) ||
 733 #ifdef IPV6
 734       (a[RTA_IIF] && RTA_PAYLOAD(a[RTA_IIF]) != 4) ||
 735 #endif
 736       (a[RTA_OIF] && RTA_PAYLOAD(a[RTA_OIF]) != 4) ||
 737       (a[RTA_GATEWAY] && RTA_PAYLOAD(a[RTA_GATEWAY]) != sizeof(ip_addr)) ||
 738       (a[RTA_PRIORITY] && RTA_PAYLOAD(a[RTA_PRIORITY]) != 4) ||
 739       (a[RTA_PREFSRC] && RTA_PAYLOAD(a[RTA_PREFSRC]) != sizeof(ip_addr)) ||
 740       (a[RTA_FLOW] && RTA_PAYLOAD(a[RTA_FLOW]) != 4))
 741     {
 742       log(L_ERR "KRT: Malformed message received");
 743       return;
 744     }
 745
 746   if (a[RTA_DST])
 747     {
 748       memcpy(&dst, RTA_DATA(a[RTA_DST]), sizeof(dst));
 749       ipa_ntoh(dst);
 750     }
 751
 752   if (a[RTA_OIF])
 753     memcpy(&oif, RTA_DATA(a[RTA_OIF]), sizeof(oif));
 754
 755   p = nl_table_map[i->rtm_table];       /* Do we know this table? */
 756   DBG("KRT: Got %I/%d, type=%d, oif=%d, table=%d, prid=%d, proto=%s\n", dst, i->rtm_dst_len, i->rtm_type, oif, i->rtm_table, i->rtm_protocol, p ? p->p.name : "(none)");
 757   if (!p)
 758     SKIP("unknown table %d\n", i->rtm_table);
 759
 760
 761 #ifdef IPV6
 762   if (a[RTA_IIF])
 763     SKIP("IIF set\n");
 764 #else
 765   if (i->rtm_tos != 0)                  /* We don't support TOS */
 766     SKIP("TOS %02x\n", i->rtm_tos);
 767 #endif
 768
 769   if (scan && !new)
 770     SKIP("RTM_DELROUTE in scan\n");
 771
 772   int c = ipa_classify_net(dst);
 773   if ((c < 0) || !(c & IADDR_HOST) || ((c & IADDR_SCOPE_MASK) <= SCOPE_LINK))
 774     SKIP("strange class/scope\n");
 775
 776   // ignore rtm_scope, it is not a real scope
 777   // if (i->rtm_scope != RT_SCOPE_UNIVERSE)
 778   //   SKIP("scope %u\n", i->rtm_scope);
 779
 780   switch (i->rtm_protocol)
 781     {
 782     case RTPROT_UNSPEC:
 783       SKIP("proto unspec\n");
 784
 785     case RTPROT_REDIRECT:
 786       src = KRT_SRC_REDIRECT;
 787       break;
 788
 789     case RTPROT_KERNEL:
 790       src = KRT_SRC_KERNEL;
 791       return;
 792
 793     case RTPROT_BIRD:
 794       if (!scan)
 795         SKIP("echo\n");
 796       src = KRT_SRC_BIRD;
 797       break;
 798
 799     case RTPROT_BOOT:
 800     default:
 801       src = KRT_SRC_ALIEN;
 802     }
 803
 804   net *net = net_get(p->p.table, dst, i->rtm_dst_len);
 805
 806   rta ra = {
 807     .src= p->p.main_source,
 808     .source = RTS_INHERIT,
 809     .scope = SCOPE_UNIVERSE,
 810     .cast = RTC_UNICAST
 811   };
 812
 813   switch (i->rtm_type)
 814     {
 815     case RTN_UNICAST:
 816
 817       if (a[RTA_MULTIPATH])
 818         {
 819           ra.dest = RTD_MULTIPATH;
 820           ra.nexthops = nl_parse_multipath(p, a[RTA_MULTIPATH]);
 821           if (!ra.nexthops)
 822             {
 823               log(L_ERR "KRT: Received strange multipath route %I/%d",
 824                   net->n.prefix, net->n.pxlen);
 825               return;
 826             }
 827
 828           break;
 829         }
 830
 831       ra.iface = if_find_by_index(oif);
 832       if (!ra.iface)
 833         {
 834           log(L_ERR "KRT: Received route %I/%d with unknown ifindex %u",
 835               net->n.prefix, net->n.pxlen, oif);
 836           return;
 837         }
 838
 839       if (a[RTA_GATEWAY])
 840         {
 841           neighbor *ng;
 842           ra.dest = RTD_ROUTER;
 843           memcpy(&ra.gw, RTA_DATA(a[RTA_GATEWAY]), sizeof(ra.gw));
 844           ipa_ntoh(ra.gw);
 845
 846           /* Silently skip strange 6to4 routes */
 847           if (ipa_in_net(ra.gw, IPA_NONE, 96))
 848             return;
 849
 850           ng = neigh_find2(&p->p, &ra.gw, ra.iface,
 851                            (i->rtm_flags & RTNH_F_ONLINK) ? NEF_ONLINK : 0);
 852           if (!ng || (ng->scope == SCOPE_HOST))
 853             {
 854               log(L_ERR "KRT: Received route %I/%d with strange next-hop %I",
 855                   net->n.prefix, net->n.pxlen, ra.gw);
 856               return;
 857             }
 858         }
 859       else
 860         {
 861           ra.dest = RTD_DEVICE;
 862
 863           /*
 864            * In Linux IPv6, 'native' device routes have proto
 865            * RTPROT_BOOT and not RTPROT_KERNEL (which they have in
 866            * IPv4 and which is expected). We cannot distinguish
 867            * 'native' and user defined device routes, so we ignore all
 868            * such device routes and for consistency, we have the same
 869            * behavior in IPv4. Anyway, users should use RTPROT_STATIC
 870            * for their 'alien' routes.
 871            */
 872
 873           if (i->rtm_protocol == RTPROT_BOOT)
 874             src = KRT_SRC_KERNEL;
 875         }
 876
 877       break;
 878     case RTN_BLACKHOLE:
 879       ra.dest = RTD_BLACKHOLE;
 880       break;
 881     case RTN_UNREACHABLE:
 882       ra.dest = RTD_UNREACHABLE;
 883       break;
 884     case RTN_PROHIBIT:
 885       ra.dest = RTD_PROHIBIT;
 886       break;
 887     /* FIXME: What about RTN_THROW? */
 888     default:
 889       SKIP("type %d\n", i->rtm_type);
 890       return;
 891     }
 892
 893   rte *e = rte_get_temp(&ra);
 894   e->net = net;
 895   e->u.krt.src = src;
 896   e->u.krt.proto = i->rtm_protocol;
 897   e->u.krt.type = i->rtm_type;
 898
 899   if (a[RTA_PRIORITY])
 900     memcpy(&e->u.krt.metric, RTA_DATA(a[RTA_PRIORITY]), sizeof(e->u.krt.metric));
 901   else
 902     e->u.krt.metric = 0;
 903
 904   if (a[RTA_PREFSRC])
 905     {
 906       ip_addr ps;
 907       memcpy(&ps, RTA_DATA(a[RTA_PREFSRC]), sizeof(ps));
 908       ipa_ntoh(ps);
 909
 910       ea_list *ea = alloca(sizeof(ea_list) + sizeof(eattr));
 911       ea->next = ra.eattrs;
 912       ra.eattrs = ea;
 913       ea->flags = EALF_SORTED;
 914       ea->count = 1;
 915       ea->attrs[0].id = EA_KRT_PREFSRC;
 916       ea->attrs[0].flags = 0;
 917       ea->attrs[0].type = EAF_TYPE_IP_ADDRESS;
 918       ea->attrs[0].u.ptr = alloca(sizeof(struct adata) + sizeof(ps));
 919       ea->attrs[0].u.ptr->length = sizeof(ps);
 920       memcpy(ea->attrs[0].u.ptr->data, &ps, sizeof(ps));
 921     }
 922
 923   if (a[RTA_FLOW])
 924     {
 925       ea_list *ea = alloca(sizeof(ea_list) + sizeof(eattr));
 926       ea->next = ra.eattrs;
 927       ra.eattrs = ea;
 928       ea->flags = EALF_SORTED;
 929       ea->count = 1;
 930       ea->attrs[0].id = EA_KRT_REALM;
 931       ea->attrs[0].flags = 0;
 932       ea->attrs[0].type = EAF_TYPE_INT;
 933       memcpy(&ea->attrs[0].u.data, RTA_DATA(a[RTA_FLOW]), 4);
 934     }
 935
 936   if (scan)
 937     krt_got_route(p, e);
 938   else
 939     krt_got_route_async(p, e, new);
 940 }
 941
 942 void
 943 krt_do_scan(struct krt_proto *p UNUSED) /* CONFIG_ALL_TABLES_AT_ONCE => p is NULL */
 944 {
 945   struct nlmsghdr *h;
 946
 947   nl_request_dump(RTM_GETROUTE);
 948   while (h = nl_get_scan())
 949     if (h->nlmsg_type == RTM_NEWROUTE || h->nlmsg_type == RTM_DELROUTE)
 950       nl_parse_route(h, 1);
 951     else
 952       log(L_DEBUG "nl_scan_fire: Unknown packet received (type=%d)", h->nlmsg_type);
 953 }
 954
 955 /*
 956  *      Asynchronous Netlink interface
 957  */
 958
 959 static sock *nl_async_sk;               /* BIRD socket for asynchronous notifications */
 960 static byte *nl_async_rx_buffer;        /* Receive buffer */
 961
 962 static void
 963 nl_async_msg(struct nlmsghdr *h)
 964 {
 965   switch (h->nlmsg_type)
 966     {
 967     case RTM_NEWROUTE:
 968     case RTM_DELROUTE:
 969       DBG("KRT: Received async route notification (%d)\n", h->nlmsg_type);
 970       nl_parse_route(h, 0);
 971       break;
 972     case RTM_NEWLINK:
 973     case RTM_DELLINK:
 974       DBG("KRT: Received async link notification (%d)\n", h->nlmsg_type);
 975       nl_parse_link(h, 0);
 976       break;
 977     case RTM_NEWADDR:
 978     case RTM_DELADDR:
 979       DBG("KRT: Received async address notification (%d)\n", h->nlmsg_type);
 980       nl_parse_addr(h);
 981       break;
 982     default:
 983       DBG("KRT: Received unknown async notification (%d)\n", h->nlmsg_type);
 984     }
 985 }
 986
 987 static int
 988 nl_async_hook(sock *sk, int size UNUSED)
 989 {
 990   struct iovec iov = { nl_async_rx_buffer, NL_RX_SIZE };
 991   struct sockaddr_nl sa;
 992   struct msghdr m = { (struct sockaddr *) &sa, sizeof(sa), &iov, 1, NULL, 0, 0 };
 993   struct nlmsghdr *h;
 994   int x;
 995   unsigned int len;
 996
 997   x = recvmsg(sk->fd, &m, 0);
 998   if (x < 0)
 999     {
1000       if (errno == ENOBUFS)
1001         {
1002           /*
1003            *  Netlink reports some packets have been thrown away.
1004            *  One day we might react to it by asking for route table
1005            *  scan in near future.
1006            */
1007           return 1;     /* More data are likely to be ready */
1008         }
1009       else if (errno != EWOULDBLOCK)
1010         log(L_ERR "Netlink recvmsg: %m");
1011       return 0;
1012     }
1013   if (sa.nl_pid)                /* It isn't from the kernel */
1014     {
1015       DBG("Non-kernel packet\n");
1016       return 1;
1017     }
1018   h = (void *) nl_async_rx_buffer;
1019   len = x;
1020   if (m.msg_flags & MSG_TRUNC)
1021     {
1022       log(L_WARN "Netlink got truncated asynchronous message");
1023       return 1;
1024     }
1025   while (NLMSG_OK(h, len))
1026     {
1027       nl_async_msg(h);
1028       h = NLMSG_NEXT(h, len);
1029     }
1030   if (len)
1031     log(L_WARN "nl_async_hook: Found packet remnant of size %d", len);
1032   return 1;
1033 }
1034
1035 static void
1036 nl_open_async(void)
1037 {
1038   sock *sk;
1039   struct sockaddr_nl sa;
1040   int fd;
1041   static int nl_open_tried = 0;
1042
1043   if (nl_open_tried)
1044     return;
1045   nl_open_tried = 1;
1046
1047   DBG("KRT: Opening async netlink socket\n");
1048
1049   fd = socket(PF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
1050   if (fd < 0)
1051     {
1052       log(L_ERR "Unable to open asynchronous rtnetlink socket: %m");
1053       return;
1054     }
1055
1056   bzero(&sa, sizeof(sa));
1057   sa.nl_family = AF_NETLINK;
1058 #ifdef IPV6
1059   sa.nl_groups = RTMGRP_LINK | RTMGRP_IPV6_IFADDR | RTMGRP_IPV6_ROUTE;
1060 #else
1061   sa.nl_groups = RTMGRP_LINK | RTMGRP_IPV4_IFADDR | RTMGRP_IPV4_ROUTE;
1062 #endif
1063   if (bind(fd, (struct sockaddr *) &sa, sizeof(sa)) < 0)
1064     {
1065       log(L_ERR "Unable to bind asynchronous rtnetlink socket: %m");
1066       return;
1067     }
1068
1069   sk = nl_async_sk = sk_new(krt_pool);
1070   sk->type = SK_MAGIC;
1071   sk->rx_hook = nl_async_hook;
1072   sk->fd = fd;
1073   if (sk_open(sk))
1074     bug("Netlink: sk_open failed");
1075
1076   if (!nl_async_rx_buffer)
1077     nl_async_rx_buffer = xmalloc(NL_RX_SIZE);
1078 }
1079
1080 /*
1081  *      Interface to the UNIX krt module
1082  */
1083
1084 static u8 nl_cf_table[(NL_NUM_TABLES+7) / 8];
1085
1086 void
1087 krt_sys_start(struct krt_proto *p, int first)
1088 {
1089   nl_table_map[KRT_CF->sys.table_id] = p;
1090   if (first)
1091     {
1092       nl_open();
1093       nl_open_async();
1094     }
1095 }
1096
1097 void
1098 krt_sys_shutdown(struct krt_proto *p UNUSED, int last UNUSED)
1099 {
1100 }
1101
1102 int
1103 krt_sys_reconfigure(struct krt_proto *p UNUSED, struct krt_config *n, struct krt_config *o)
1104 {
1105   return n->sys.table_id == o->sys.table_id;
1106 }
1107
1108
1109 void
1110 krt_sys_preconfig(struct config *c UNUSED)
1111 {
1112   bzero(&nl_cf_table, sizeof(nl_cf_table));
1113 }
1114
1115 void
1116 krt_sys_postconfig(struct krt_config *x)
1117 {
1118   int id = x->sys.table_id;
1119
1120   if (nl_cf_table[id/8] & (1 << (id%8)))
1121     cf_error("Multiple kernel syncers defined for table #%d", id);
1122   nl_cf_table[id/8] |= (1 << (id%8));
1123 }
1124
1125 void
1126 krt_sys_init_config(struct krt_config *cf)
1127 {
1128   cf->sys.table_id = RT_TABLE_MAIN;
1129 }
1130
1131 void
1132 krt_sys_copy_config(struct krt_config *d, struct krt_config *s)
1133 {
1134   d->sys.table_id = s->sys.table_id;
1135 }
1136
1137
1138
1139 void
1140 kif_sys_start(struct kif_proto *p UNUSED)
1141 {
1142   nl_open();
1143   nl_open_async();
1144 }
1145
1146 void
1147 kif_sys_shutdown(struct kif_proto *p UNUSED)
1148 {
1149 }