[mirror_frr.git] / zebra / kernel_socket.c

/* Kernel communication using routing socket.
 * Copyright (C) 1999 Kunihiro Ishiguro
 *
 * This file is part of GNU Zebra.
 *
 * GNU Zebra is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by the
 * Free Software Foundation; either version 2, or (at your option) any
 * later version.
 *
 * GNU Zebra is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with GNU Zebra; see the file COPYING.  If not, write to the Free
 * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
 * 02111-1307, USA.  
 */

#include <zebra.h>

#include "if.h"
#include "prefix.h"
#include "sockunion.h"
#include "connected.h"
#include "memory.h"
#include "ioctl.h"
#include "log.h"
#include "str.h"
#include "table.h"
#include "rib.h"
#include "privs.h"

#include "zebra/interface.h"
#include "zebra/zserv.h"
#include "zebra/debug.h"

extern struct zebra_privs_t zserv_privs;
extern struct zebra_t zebrad;

/*
 * Given a sockaddr length, round it up to include pad bytes following
 * it.  Assumes the kernel pads to sizeof(long).
 *
 * XXX: why is ROUNDUP(0) sizeof(long)?  0 is an illegal sockaddr
 * length anyway (< sizeof (struct sockaddr)), so this shouldn't
 * matter.
 */
#define ROUNDUP(a) \
  ((a) > 0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))

/*
 * Given a pointer (sockaddr or void *), return the number of bytes
 * taken up by the sockaddr and any padding needed for alignment.
 */
#if defined(HAVE_SA_LEN)
#define SAROUNDUP(X)   ROUNDUP(((struct sockaddr *)(X))->sa_len)
#elif defined(HAVE_IPV6)
/*
 * One would hope all fixed-size structure definitions are aligned,
 * but round them up nonetheless.
 */
#define SAROUNDUP(X) \
    (((struct sockaddr *)(X))->sa_family == AF_INET ?   \
      ROUNDUP(sizeof(struct sockaddr_in)):\
      (((struct sockaddr *)(X))->sa_family == AF_INET6 ? \
       ROUNDUP(sizeof(struct sockaddr_in6)) :  \
       (((struct sockaddr *)(X))->sa_family == AF_LINK ? \
         ROUNDUP(sizeof(struct sockaddr_dl)) : sizeof(struct sockaddr))))
#else /* HAVE_IPV6 */ 
#define SAROUNDUP(X) \
      (((struct sockaddr *)(X))->sa_family == AF_INET ?   \
        ROUNDUP(sizeof(struct sockaddr_in)):\
         (((struct sockaddr *)(X))->sa_family == AF_LINK ? \
           ROUNDUP(sizeof(struct sockaddr_dl)) : sizeof(struct sockaddr)))
#endif /* HAVE_SA_LEN */

#define RTA_ADDR_GET(DEST, RTA, RTMADDRS, PNT) \
  if ((RTMADDRS) & (RTA)) \
    { \
      int len = SAROUNDUP ((PNT)); \
      if ( ((DEST) != NULL) && \
           af_check (((struct sockaddr *)(PNT))->sa_family)) \
        memcpy ((DEST), (PNT), len); \
      (PNT) += len; \
    }
#define RTA_ATTR_GET(DEST, RTA, RTMADDRS, PNT) \
  if ((RTMADDRS) & (RTA)) \
    { \
      int len = SAROUNDUP ((PNT)); \
      if ( ((DEST) != NULL) ) \
        memcpy ((DEST), (PNT), len); \
      (PNT) += len; \
    }

#define RTA_NAME_GET(DEST, RTA, RTMADDRS, PNT, LEN) \
  if ((RTMADDRS) & (RTA)) \
    { \
      int len = SAROUNDUP ((PNT)); \
      struct sockaddr_dl *sdl = (struct sockaddr_dl *)(PNT); \
      if (IS_ZEBRA_DEBUG_KERNEL) \
        zlog_debug ("%s: RTA_SDL_GET nlen %d, alen %d", \
                    __func__, sdl->sdl_nlen, sdl->sdl_alen); \
      if ( ((DEST) != NULL) && (sdl->sdl_family == AF_LINK) \
           && (sdl->sdl_nlen < IFNAMSIZ) && (sdl->sdl_nlen <= len) ) \
        { \
          memcpy ((DEST), sdl->sdl_data, sdl->sdl_nlen); \
          (DEST)[sdl->sdl_nlen] = '\0'; \
          (LEN) = sdl->sdl_nlen; \
        } \
      (PNT) += len; \
    } \
  else \
    { \
      (LEN) = 0; \
    }
/* Routing socket message types. */
struct message rtm_type_str[] =
{
  {RTM_ADD,      "RTM_ADD"},
  {RTM_DELETE,   "RTM_DELETE"},
  {RTM_CHANGE,   "RTM_CHANGE"},
  {RTM_GET,      "RTM_GET"},
  {RTM_LOSING,   "RTM_LOSING"},
  {RTM_REDIRECT, "RTM_REDIRECT"},
  {RTM_MISS,     "RTM_MISS"},
  {RTM_LOCK,     "RTM_LOCK"},
  {RTM_OLDADD,   "RTM_OLDADD"},
  {RTM_OLDDEL,   "RTM_OLDDEL"},
  {RTM_RESOLVE,  "RTM_RESOLVE"},
  {RTM_NEWADDR,  "RTM_NEWADDR"},
  {RTM_DELADDR,  "RTM_DELADDR"},
  {RTM_IFINFO,   "RTM_IFINFO"},
#ifdef RTM_OIFINFO
  {RTM_OIFINFO,   "RTM_OIFINFO"},
#endif /* RTM_OIFINFO */
#ifdef RTM_NEWMADDR
  {RTM_NEWMADDR, "RTM_NEWMADDR"},
#endif /* RTM_NEWMADDR */
#ifdef RTM_DELMADDR
  {RTM_DELMADDR, "RTM_DELMADDR"},
#endif /* RTM_DELMADDR */
#ifdef RTM_IFANNOUNCE
  {RTM_IFANNOUNCE, "RTM_IFANNOUNCE"},
#endif /* RTM_IFANNOUNCE */
  {0,            NULL}
};

struct message rtm_flag_str[] =
{
  {RTF_UP,        "UP"},
  {RTF_GATEWAY,   "GATEWAY"},
  {RTF_HOST,      "HOST"},
  {RTF_REJECT,    "REJECT"},
  {RTF_DYNAMIC,   "DYNAMIC"},
  {RTF_MODIFIED,  "MODIFIED"},
  {RTF_DONE,      "DONE"},
#ifdef RTF_MASK
  {RTF_MASK,      "MASK"},
#endif /* RTF_MASK */
  {RTF_CLONING,   "CLONING"},
  {RTF_XRESOLVE,  "XRESOLVE"},
  {RTF_LLINFO,    "LLINFO"},
  {RTF_STATIC,    "STATIC"},
  {RTF_BLACKHOLE, "BLACKHOLE"},
#ifdef RTF_PRIVATE
  {RTF_PRIVATE,	  "PRIVATE"},
#endif /* RTF_PRIVATE */
  {RTF_PROTO1,    "PROTO1"},
  {RTF_PROTO2,    "PROTO2"},
#ifdef RTF_PRCLONING
  {RTF_PRCLONING, "PRCLONING"},
#endif /* RTF_PRCLONING */
#ifdef RTF_WASCLONED
  {RTF_WASCLONED, "WASCLONED"},
#endif /* RTF_WASCLONED */
#ifdef RTF_PROTO3
  {RTF_PROTO3,    "PROTO3"},
#endif /* RTF_PROTO3 */
#ifdef RTF_PINNED
  {RTF_PINNED,    "PINNED"},
#endif /* RTF_PINNED */
#ifdef RTF_LOCAL
  {RTF_LOCAL,    "LOCAL"},
#endif /* RTF_LOCAL */
#ifdef RTF_BROADCAST
  {RTF_BROADCAST, "BROADCAST"},
#endif /* RTF_BROADCAST */
#ifdef RTF_MULTICAST
  {RTF_MULTICAST, "MULTICAST"},
#endif /* RTF_MULTICAST */
#ifdef RTF_MULTIRT
  {RTF_MULTIRT,   "MULTIRT"},
#endif /* RTF_MULTIRT */
#ifdef RTF_SETSRC
  {RTF_SETSRC,    "SETSRC"},
#endif /* RTF_SETSRC */
  {0,             NULL}
};

/* Kernel routing update socket. */
int routing_sock = -1;

/* Yes I'm checking ugly routing socket behavior. */
/* #define DEBUG */

/* Supported address family check. */
static int inline
af_check (int family)
{
  if (family == AF_INET)
    return 1;
#ifdef HAVE_IPV6
  if (family == AF_INET6)
    return 1;
#endif /* HAVE_IPV6 */
  return 0;
}
\f
/* Dump routing table flag for debug purpose. */
static void
rtm_flag_dump (int flag)
{
  struct message *mes;
  static char buf[BUFSIZ];

  buf[0] = '\0';
  for (mes = rtm_flag_str; mes->key != 0; mes++)
    {
      if (mes->key & flag)
	{
	  strlcat (buf, mes->str, BUFSIZ);
	  strlcat (buf, " ", BUFSIZ);
	}
    }
  zlog_debug ("Kernel: %s", buf);
}

#ifdef RTM_IFANNOUNCE
/* Interface adding function */
int
ifan_read (struct if_announcemsghdr *ifan)
{
  struct interface *ifp;
  
  ifp = if_lookup_by_index (ifan->ifan_index);
  
  if (ifp)
    assert ( (ifp->ifindex == ifan->ifan_index) 
             || (ifp->ifindex == IFINDEX_INTERNAL) );

  if ( (ifp == NULL || (ifp->ifindex == IFINDEX_INTERNAL)
      && ifan->ifan_what == IFAN_ARRIVAL)
    {
      if (IS_ZEBRA_DEBUG_KERNEL)
        zlog_debug ("%s: creating interface for ifindex %d, name %s",
                    __func__, ifan->ifan_index, ifan->ifan_name);
      
      /* Create Interface */
      ifp = if_get_by_name_len(ifan->ifan_name,
			       strnlen(ifan->ifan_name,
				       sizeof(ifan->ifan_name)));
      ifp->ifindex = ifan->ifan_index;

      if_add_update (ifp);
    }
  else if (ifp != NULL && ifan->ifan_what == IFAN_DEPARTURE)
    if_delete_update (ifp);

  if_get_flags (ifp);
  if_get_mtu (ifp);
  if_get_metric (ifp);

  if (IS_ZEBRA_DEBUG_KERNEL)
    zlog_debug ("%s: interface %s index %d", 
                __func__, ifan->ifan_name, ifan->ifan_index);

  return 0;
}
#endif /* RTM_IFANNOUNCE */

/*
 * Handle struct if_msghdr obtained from reading routing socket or
 * sysctl (from interface_list).  There may or may not be sockaddrs
 * present after the header.
 */
int
ifm_read (struct if_msghdr *ifm)
{
  struct interface *ifp = NULL;
  char ifname[IFNAMSIZ];
  short ifnlen = 0;
  caddr_t *cp;
  
  /* terminate ifname at head (for strnlen) and tail (for safety) */
  ifname[IFNAMSIZ - 1] = '\0';
  
  /* paranoia: sanity check structure */
  if (ifm->ifm_msglen < sizeof(struct if_msghdr))
    {
      zlog_err ("ifm_read: ifm->ifm_msglen %d too short\n",
		ifm->ifm_msglen);
      return -1;
    }

  /*
   * Check for a sockaddr_dl following the message.  First, point to
   * where a socakddr might be if one follows the message.
   */
  cp = (void *)(ifm + 1);

#ifdef SUNOS_5
  /* 
   * XXX This behavior should be narrowed to only the kernel versions
   * for which the structures returned do not match the headers.
   *
   * if_msghdr_t on 64 bit kernels in Solaris 9 and earlier versions
   * is 12 bytes larger than the 32 bit version.
   */
  if (((struct sockaddr *) cp)->sa_family == AF_UNSPEC)
  	cp = cp + 12;
#endif

  RTA_ADDR_GET (NULL, RTA_DST, ifm->ifm_addrs, cp);
  RTA_ADDR_GET (NULL, RTA_GATEWAY, ifm->ifm_addrs, cp);
  RTA_ATTR_GET (NULL, RTA_NETMASK, ifm->ifm_addrs, cp);
  RTA_ADDR_GET (NULL, RTA_GENMASK, ifm->ifm_addrs, cp);
  RTA_NAME_GET (ifname, RTA_IFP, ifm->ifm_addrs, cp, ifnlen);
  RTA_ADDR_GET (NULL, RTA_IFA, ifm->ifm_addrs, cp);
  RTA_ADDR_GET (NULL, RTA_AUTHOR, ifm->ifm_addrs, cp);
  RTA_ADDR_GET (NULL, RTA_BRD, ifm->ifm_addrs, cp);
  
  if (IS_ZEBRA_DEBUG_KERNEL)
    zlog_debug ("%s: sdl ifname %s", __func__, (ifnlen ? ifname : "(nil)"));
  
  /* 
   * Look up on ifindex first, because ifindices are the primary handle for
   * interfaces across the user/kernel boundary, for most systems.  (Some
   * messages, such as up/down status changes on NetBSD, do not include a
   * sockaddr_dl).
   */
  if ( (ifp = if_lookup_by_index (ifm->ifm_index)) != NULL )
    {
      /* we have an ifp, verify that the name matches as some systems,
       * eg Solaris, have a 1:many association of ifindex:ifname
       * if they dont match, we dont have the correct ifp and should
       * set it back to NULL to let next check do lookup by name
       */
      if (ifnlen && (strncmp (ifp->name, ifname, IFNAMSIZ) != 0) )
        {
          if (IS_ZEBRA_DEBUG_KERNEL)
            zlog_debug ("%s: ifp name %s doesnt match sdl name %s",
                        __func__, ifp->name, ifname);
          ifp = NULL;
        }
    }
  
  /* 
   * If we dont have an ifp, try looking up by name.  Particularly as some
   * systems (Solaris) have a 1:many mapping of ifindex:ifname - the ifname
   * is therefore our unique handle to that interface.
   *
   * Interfaces specified in the configuration file for which the ifindex
   * has not been determined will have ifindex == IFINDEX_INTERNAL, and such
   * interfaces are found by this search, and then their ifindex values can
   * be filled in.
   */
  if ( (ifp == NULL) && ifnlen)
    ifp = if_lookup_by_name (ifname);

  /*
   * If ifp still does not exist or has an invalid index (IFINDEX_INTERNAL),
   * create or fill in an interface.
   */
  if ((ifp == NULL) || (ifp->ifindex == IFINDEX_INTERNAL))
    {
      /*
       * To create or fill in an interface, a sockaddr_dl (via
       * RTA_IFP) is required.
       */
      if (!ifnlen)
	{
	  zlog_warn ("Interface index %d (new) missing ifname\n",
		     ifm->ifm_index);
	  return -1;
	}
      
      if (ifp == NULL)
        {
	  /* Interface that zebra was not previously aware of, so create. */ 
	  ifp = if_create (ifname, ifnlen);
	  if (IS_ZEBRA_DEBUG_KERNEL)
	    zlog_debug ("%s: creating ifp for ifindex %d", 
	                __func__, ifm->ifm_index);
        }

      if (IS_ZEBRA_DEBUG_KERNEL)
        zlog_debug ("%s: updated/created ifp, ifname %s, ifindex %d",
                    __func__, ifp->name, ifp->ifindex);
      /* 
       * Fill in newly created interface structure, or larval
       * structure with ifindex IFINDEX_INTERNAL.
       */
      ifp->ifindex = ifm->ifm_index;
      ifp->flags = ifm->ifm_flags;
#if defined(__bsdi__)
      if_kvm_get_mtu (ifp);
#else
      if_get_mtu (ifp);
#endif /* __bsdi__ */
      if_get_metric (ifp);

      if_add_update (ifp);
    }
  else
    /*
     * Interface structure exists.  Adjust stored flags from
     * notification.  If interface has up->down or down->up
     * transition, call state change routines (to adjust routes,
     * notify routing daemons, etc.).  (Other flag changes are stored
     * but apparently do not trigger action.)
     */
    {
      if (ifp->ifindex != ifm->ifm_index)
        {
          zlog_warn ("%s: index mismatch, ifname %s, ifp index %d, "
                     "ifm index %d", 
                     __func__, ifp->name, ifp->ifindex, ifm->ifm_index);
          return -1;
        }
      
      if (if_is_up (ifp))
	{
	  ifp->flags = ifm->ifm_flags;
	  if (! if_is_up (ifp))
	    {
	      if_down (ifp);
#ifndef RTM_IFANNOUNCE
              /* No RTM_IFANNOUNCE on this platform, so we can never
               * distinguish between down and delete. We must presume
               * it has been deleted.
               * Eg, Solaris will not notify us of unplumb.
               *
               * XXX: Fixme - this should be runtime detected
               * So that a binary compiled on a system with IFANNOUNCE
               * will still behave correctly if run on a platform without
               */
              if_delete_update (ifp);
#endif /* RTM_IFANNOUNCE */
            }
	}
      else
	{
	  ifp->flags = ifm->ifm_flags;
	  if (if_is_up (ifp))
	    if_up (ifp);
	}
    }
  
#ifdef HAVE_NET_RT_IFLIST
  ifp->stats = ifm->ifm_data;
#endif /* HAVE_NET_RT_IFLIST */

  if (IS_ZEBRA_DEBUG_KERNEL)
    zlog_debug ("%s: interface %s index %d", 
                __func__, ifp->name, ifp->ifindex);

  return 0;
}
\f
/* Address read from struct ifa_msghdr. */
void
ifam_read_mesg (struct ifa_msghdr *ifm,
		union sockunion *addr,
		union sockunion *mask,
		union sockunion *brd,
		char *ifname,
		short *ifnlen)
{
  caddr_t pnt, end;

  pnt = (caddr_t)(ifm + 1);
  end = ((caddr_t)ifm) + ifm->ifam_msglen;

  /* Be sure structure is cleared */
  memset (mask, 0, sizeof (union sockunion));
  memset (addr, 0, sizeof (union sockunion));
  memset (dest, 0, sizeof (union sockunion));

  /* We fetch each socket variable into sockunion. */
  RTA_ADDR_GET (NULL, RTA_DST, ifm->ifam_addrs, pnt);
  RTA_ADDR_GET (NULL, RTA_GATEWAY, ifm->ifam_addrs, pnt);
  RTA_ATTR_GET (mask, RTA_NETMASK, ifm->ifam_addrs, pnt);
  RTA_ADDR_GET (NULL, RTA_GENMASK, ifm->ifam_addrs, pnt);
  RTA_NAME_GET (ifname, RTA_IFP, ifm->ifam_addrs, pnt, *ifnlen);
  RTA_ADDR_GET (addr, RTA_IFA, ifm->ifam_addrs, pnt);
  RTA_ADDR_GET (NULL, RTA_AUTHOR, ifm->ifam_addrs, pnt);
  RTA_ADDR_GET (brd, RTA_BRD, ifm->ifam_addrs, pnt);

  if (IS_ZEBRA_DEBUG_KERNEL)
      zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x", 
                  __func__, ifm->ifam_index, 
                  (ifnlen ? ifname : "(nil)"), ifm->ifam_addrs);
  
  /* Assert read up end point matches to end point */
  if (pnt != end)
    zlog_warn ("ifam_read() does't read all socket data");
}

/* Interface's address information get. */
int
ifam_read (struct ifa_msghdr *ifam)
{
  struct interface *ifp = NULL;
  union sockunion addr, mask, brd;
  char ifname[INTERFACE_NAMSIZ];
  short ifnlen = 0;
  char isalias = 0;
  
  ifname[0] = ifname[INTERFACE_NAMSIZ - 1] = '\0';
  
  /* Allocate and read address information. */
  ifam_read_mesg (ifam, &addr, &mask, &brd, ifname, &ifnlen);
  
  if ((ifp = if_lookup_by_index(ifam->ifam_index)) == NULL)
    {
      zlog_warn ("%s: no interface for ifname %s, index %d", 
                 __func__, ifname, ifam->ifam_index);
      return -1;
    }
  
  if (ifnlen && strncmp (ifp->name, ifname, INTERFACE_NAMSIZ))
    isalias = 1;
  
  /* Check interface flag for implicit up of the interface. */
  if_refresh (ifp);

  /* Add connected address. */
  switch (sockunion_family (&addr))
    {
    case AF_INET:
      if (ifam->ifam_type == RTM_NEWADDR)
	connected_add_ipv4 (ifp, 0, &addr.sin.sin_addr, 
			    ip_masklen (mask.sin.sin_addr),
			    &brd.sin.sin_addr, NULL);
      else
	connected_delete_ipv4 (ifp, 0, &addr.sin.sin_addr, 
			       ip_masklen (mask.sin.sin_addr),
			       &brd.sin.sin_addr);
      break;
#ifdef HAVE_IPV6
    case AF_INET6:
      /* Unset interface index from link-local address when IPv6 stack
	 is KAME. */
      if (IN6_IS_ADDR_LINKLOCAL (&addr.sin6.sin6_addr))
	SET_IN6_LINKLOCAL_IFINDEX (addr.sin6.sin6_addr, 0);

      if (ifam->ifam_type == RTM_NEWADDR)
	connected_add_ipv6 (ifp,
			    &addr.sin6.sin6_addr, 
			    ip6_masklen (mask.sin6.sin6_addr),
			    &brd.sin6.sin6_addr, NULL);
      else
	connected_delete_ipv6 (ifp,
			       &addr.sin6.sin6_addr, 
			       ip6_masklen (mask.sin6.sin6_addr),
			       &brd.sin6.sin6_addr);
      break;
#endif /* HAVE_IPV6 */
    default:
      /* Unsupported family silently ignore... */
      break;
    }
  return 0;
}
\f
/* Interface function for reading kernel routing table information. */
int
rtm_read_mesg (struct rt_msghdr *rtm,
	       union sockunion *dest,
	       union sockunion *mask,
	       union sockunion *gate,
	       char *ifname,
	       short *ifnlen)
{
  caddr_t pnt, end;

  /* Pnt points out socket data start point. */
  pnt = (caddr_t)(rtm + 1);
  end = ((caddr_t)rtm) + rtm->rtm_msglen;

  /* rt_msghdr version check. */
  if (rtm->rtm_version != RTM_VERSION) 
      zlog (NULL, LOG_WARNING,
	      "Routing message version different %d should be %d."
	      "This may cause problem\n", rtm->rtm_version, RTM_VERSION);
  
  /* Be sure structure is cleared */
  memset (dest, 0, sizeof (union sockunion));
  memset (gate, 0, sizeof (union sockunion));
  memset (mask, 0, sizeof (union sockunion));

  /* We fetch each socket variable into sockunion. */
  RTA_ADDR_GET (dest, RTA_DST, rtm->rtm_addrs, pnt);
  RTA_ADDR_GET (gate, RTA_GATEWAY, rtm->rtm_addrs, pnt);
  RTA_ATTR_GET (mask, RTA_NETMASK, rtm->rtm_addrs, pnt);
  RTA_ADDR_GET (NULL, RTA_GENMASK, rtm->rtm_addrs, pnt);
  RTA_NAME_GET (ifname, RTA_IFP, rtm->rtm_addrs, pnt, *ifnlen);
  RTA_ADDR_GET (NULL, RTA_IFA, rtm->rtm_addrs, pnt);
  RTA_ADDR_GET (NULL, RTA_AUTHOR, rtm->rtm_addrs, pnt);
  RTA_ADDR_GET (NULL, RTA_BRD, rtm->rtm_addrs, pnt);

  /* If there is netmask information set it's family same as
     destination family*/
  if (rtm->rtm_addrs & RTA_NETMASK)
    mask->sa.sa_family = dest->sa.sa_family;

  /* Assert read up to the end of pointer. */
  if (pnt != end) 
      zlog (NULL, LOG_WARNING, "rtm_read() does't read all socket data.");

  return rtm->rtm_flags;
}

void
rtm_read (struct rt_msghdr *rtm)
{
  int flags;
  u_char zebra_flags;
  union sockunion dest, mask, gate;
  char ifname[INTERFACE_NAMSIZ + 1];
  short ifnlen = 0;

  zebra_flags = 0;

  /* Discard self send message. */
  if (rtm->rtm_type != RTM_GET 
      && (rtm->rtm_pid == pid || rtm->rtm_pid == old_pid))
    return;

  /* Read destination and netmask and gateway from rtm message
     structure. */
  flags = rtm_read_mesg (rtm, &dest, &mask, &gate, ifname, &ifnlen);

#ifdef RTF_CLONED	/*bsdi, netbsd 1.6*/
  if (flags & RTF_CLONED)
    return;
#endif
#ifdef RTF_WASCLONED	/*freebsd*/
  if (flags & RTF_WASCLONED)
    return;
#endif

  if ((rtm->rtm_type == RTM_ADD) && ! (flags & RTF_UP))
    return;

  /* This is connected route. */
  if (! (flags & RTF_GATEWAY))
      return;

  if (flags & RTF_PROTO1)
    SET_FLAG (zebra_flags, ZEBRA_FLAG_SELFROUTE);

  /* This is persistent route. */
  if (flags & RTF_STATIC)
    SET_FLAG (zebra_flags, ZEBRA_FLAG_STATIC);

  /* This is a reject or blackhole route */
  if (flags & RTF_REJECT)
    SET_FLAG (zebra_flags, ZEBRA_FLAG_REJECT);
  if (flags & RTF_BLACKHOLE)
    SET_FLAG (zebra_flags, ZEBRA_FLAG_BLACKHOLE);

  if (dest.sa.sa_family == AF_INET)
    {
      struct prefix_ipv4 p;

      p.family = AF_INET;
      p.prefix = dest.sin.sin_addr;
      if (flags & RTF_HOST)
	p.prefixlen = IPV4_MAX_PREFIXLEN;
      else
	p.prefixlen = ip_masklen (mask.sin.sin_addr);
      
      /* Change, delete the old prefix, we have no further information
       * to specify the route really
       */
      if (rtm->rtm_type == RTM_CHANGE)
        rib_delete_ipv4 (ZEBRA_ROUTE_KERNEL, zebra_flags, &p,
                         NULL, 0, 0);
      
      if (rtm->rtm_type == RTM_GET 
          || rtm->rtm_type == RTM_ADD
          || rtm->rtm_type == RTM_CHANGE)
	rib_add_ipv4 (ZEBRA_ROUTE_KERNEL, zebra_flags, 
		      &p, &gate.sin.sin_addr, 0, 0, 0, 0);
      else
	rib_delete_ipv4 (ZEBRA_ROUTE_KERNEL, zebra_flags, 
		      &p, &gate.sin.sin_addr, 0, 0);
    }
#ifdef HAVE_IPV6
  if (dest.sa.sa_family == AF_INET6)
    {
      struct prefix_ipv6 p;
      unsigned int ifindex = 0;

      p.family = AF_INET6;
      p.prefix = dest.sin6.sin6_addr;
      if (flags & RTF_HOST)
	p.prefixlen = IPV6_MAX_PREFIXLEN;
      else
	p.prefixlen = ip6_masklen (mask.sin6.sin6_addr);

#ifdef KAME
      if (IN6_IS_ADDR_LINKLOCAL (&gate.sin6.sin6_addr))
	{
	  ifindex = IN6_LINKLOCAL_IFINDEX (gate.sin6.sin6_addr);
	  SET_IN6_LINKLOCAL_IFINDEX (gate.sin6.sin6_addr, 0);
	}
#endif /* KAME */

      /* CHANGE: delete the old prefix, we have no further information
       * to specify the route really
       */
      if (rtm->rtm_type == RTM_CHANGE)
        rib_delete_ipv4 (ZEBRA_ROUTE_KERNEL, zebra_flags, &p,
                         NULL, 0, 0);
      
      if (rtm->rtm_type == RTM_GET 
          || rtm->rtm_type == RTM_ADD
          || rtm->rtm_type == RTM_CHANGE)
	rib_add_ipv6 (ZEBRA_ROUTE_KERNEL, zebra_flags,
		      &p, &gate.sin6.sin6_addr, ifindex, 0, 0, 0);
      else
	rib_delete_ipv6 (ZEBRA_ROUTE_KERNEL, zebra_flags,
			 &p, &gate.sin6.sin6_addr, ifindex, 0);
    }
#endif /* HAVE_IPV6 */
}

/* Interface function for the kernel routing table updates.  Support
   for RTM_CHANGE will be needed. */
int
rtm_write (int message,
	   union sockunion *dest,
	   union sockunion *mask,
	   union sockunion *gate,
	   unsigned int index,
	   int zebra_flags,
	   int metric)
{
  int ret;
  caddr_t pnt;
  struct interface *ifp;
  struct sockaddr_in tmp_gate;
#ifdef HAVE_IPV6
  struct sockaddr_in6 tmp_gate6;
#endif /* HAVE_IPV6 */

  /* Sequencial number of routing message. */
  static int msg_seq = 0;

  /* Struct of rt_msghdr and buffer for storing socket's data. */
  struct 
  {
    struct rt_msghdr rtm;
    char buf[512];
  } msg;
  
  memset (&tmp_gate, 0, sizeof (struct sockaddr_in));
  tmp_gate.sin_family = AF_INET;
#ifdef HAVE_SIN_LEN
  tmp_gate.sin_len = sizeof (struct sockaddr_in);
#endif /* HAVE_SIN_LEN */

#ifdef HAVE_IPV6
  memset (&tmp_gate6, 0, sizeof (struct sockaddr_in6));
  tmp_gate6.sin6_family = AF_INET6;
#ifdef SIN6_LEN
  tmp_gate6.sin6_len = sizeof (struct sockaddr_in6);
#endif /* SIN6_LEN */
#endif /* HAVE_IPV6 */

  if (routing_sock < 0)
    return ZEBRA_ERR_EPERM;

  /* Clear and set rt_msghdr values */
  memset (&msg, 0, sizeof (struct rt_msghdr));
  msg.rtm.rtm_version = RTM_VERSION;
  msg.rtm.rtm_type = message;
  msg.rtm.rtm_seq = msg_seq++;
  msg.rtm.rtm_addrs = RTA_DST;
  msg.rtm.rtm_addrs |= RTA_GATEWAY;
  msg.rtm.rtm_flags = RTF_UP;
  msg.rtm.rtm_index = index;

  if (metric != 0)
    {
      msg.rtm.rtm_rmx.rmx_hopcount = metric;
      msg.rtm.rtm_inits |= RTV_HOPCOUNT;
    }

  ifp = if_lookup_by_index (index);

  if (gate && message == RTM_ADD)
    msg.rtm.rtm_flags |= RTF_GATEWAY;

  if (! gate && message == RTM_ADD && ifp &&
      (ifp->flags & IFF_POINTOPOINT) == 0)
    msg.rtm.rtm_flags |= RTF_CLONING;

  /* If no protocol specific gateway is specified, use link
     address for gateway. */
  if (! gate)
    {
      if (!ifp)
        {
          zlog_warn ("no gateway found for interface index %d", index);
          return -1;
        }
      gate = (union sockunion *) & ifp->sdl;
    }

  if (mask)
    msg.rtm.rtm_addrs |= RTA_NETMASK;
  else if (message == RTM_ADD) 
    msg.rtm.rtm_flags |= RTF_HOST;

  /* Tagging route with flags */
  msg.rtm.rtm_flags |= (RTF_PROTO1);

  /* Additional flags. */
  if (zebra_flags & ZEBRA_FLAG_BLACKHOLE)
    msg.rtm.rtm_flags |= RTF_BLACKHOLE;
  if (zebra_flags & ZEBRA_FLAG_REJECT)
    msg.rtm.rtm_flags |= RTF_REJECT;


#ifdef HAVE_SIN_LEN
#define SOCKADDRSET(X,R) \
  if (msg.rtm.rtm_addrs & (R)) \
    { \
      int len = ROUNDUP ((X)->sa.sa_len); \
      memcpy (pnt, (caddr_t)(X), len); \
      pnt += len; \
    }
#else 
#define SOCKADDRSET(X,R) \
  if (msg.rtm.rtm_addrs & (R)) \
    { \
      int len = SAROUNDUP (X); \
      memcpy (pnt, (caddr_t)(X), len); \
      pnt += len; \
    }
#endif /* HAVE_SIN_LEN */

  pnt = (caddr_t) msg.buf;

  /* Write each socket data into rtm message buffer */
  SOCKADDRSET (dest, RTA_DST);
  SOCKADDRSET (gate, RTA_GATEWAY);
  SOCKADDRSET (mask, RTA_NETMASK);

  msg.rtm.rtm_msglen = pnt - (caddr_t) &msg;

  ret = write (routing_sock, &msg, msg.rtm.rtm_msglen);

  if (ret != msg.rtm.rtm_msglen) 
    {
      if (errno == EEXIST) 
	return ZEBRA_ERR_RTEXIST;
      if (errno == ENETUNREACH)
	return ZEBRA_ERR_RTUNREACH;
      
      zlog_warn ("write : %s (%d)", safe_strerror (errno), errno);
      return -1;
    }
  return 0;
}

\f
#include "thread.h"
#include "zebra/zserv.h"

/* For debug purpose. */
static void
rtmsg_debug (struct rt_msghdr *rtm)
{
  const char *type = "Unknown";
  struct message *mes;

  for (mes = rtm_type_str; mes->str; mes++)
    if (mes->key == rtm->rtm_type)
      {
	type = mes->str;
	break;
      }

  zlog_debug ("Kernel: Len: %d Type: %s", rtm->rtm_msglen, type);
  rtm_flag_dump (rtm->rtm_flags);
  zlog_debug ("Kernel: message seq %d", rtm->rtm_seq);
  zlog_debug ("Kernel: pid %d, rtm_addrs 0x%x", rtm->rtm_pid, rtm->rtm_addrs);
}

/* This is pretty gross, better suggestions welcome -- mhandler */
#ifndef RTAX_MAX
#ifdef RTA_NUMBITS
#define RTAX_MAX	RTA_NUMBITS
#else
#define RTAX_MAX	8
#endif /* RTA_NUMBITS */
#endif /* RTAX_MAX */

/* Kernel routing table and interface updates via routing socket. */
int
kernel_read (struct thread *thread)
{
  int sock;
  int nbytes;
  struct rt_msghdr *rtm;

  /*
   * This must be big enough for any message the kernel might send.
   * Rather than determining how many sockaddrs of what size might be
   * in each particular message, just use RTAX_MAX of sockaddr_storage
   * for each.  Note that the sockaddrs must be after each message
   * definition, or rather after whichever happens to be the largest,
   * since the buffer needs to be big enough for a message and the
   * sockaddrs together.
   */
  union 
  {
    /* Routing information. */
    struct 
    {
      struct rt_msghdr rtm;
      struct sockaddr_storage addr[RTAX_MAX];
    } r;

    /* Interface information. */
    struct
    {
      struct if_msghdr ifm;
      struct sockaddr_storage addr[RTAX_MAX];
    } im;

    /* Interface address information. */
    struct
    {
      struct ifa_msghdr ifa;
      struct sockaddr_storage addr[RTAX_MAX];
    } ia;

#ifdef RTM_IFANNOUNCE
    /* Interface arrival/departure */
    struct
    {
      struct if_announcemsghdr ifan;
      struct sockaddr_storage addr[RTAX_MAX];
    } ian;
#endif /* RTM_IFANNOUNCE */

  } buf;

  /* Fetch routing socket. */
  sock = THREAD_FD (thread);

  nbytes= read (sock, &buf, sizeof buf);

  if (nbytes <= 0)
    {
      if (nbytes < 0 && errno != EWOULDBLOCK && errno != EAGAIN)
	zlog_warn ("routing socket error: %s", safe_strerror (errno));
      return 0;
    }

  thread_add_read (zebrad.master, kernel_read, NULL, sock);

  if (IS_ZEBRA_DEBUG_KERNEL)
    rtmsg_debug (&buf.r.rtm);

  rtm = &buf.r.rtm;

  /*
   * Ensure that we didn't drop any data, so that processing routines
   * can assume they have the whole message.
   */
  if (rtm->rtm_msglen != nbytes)
    {
      zlog_warn ("kernel_read: rtm->rtm_msglen %d, nbytes %d, type %d\n",
		 rtm->rtm_msglen, nbytes, rtm->rtm_type);
      return -1;
    }

  switch (rtm->rtm_type)
    {
    case RTM_ADD:
    case RTM_DELETE:
    case RTM_CHANGE:
      rtm_read (rtm);
      break;
    case RTM_IFINFO:
      ifm_read (&buf.im.ifm);
      break;
    case RTM_NEWADDR:
    case RTM_DELADDR:
      ifam_read (&buf.ia.ifa);
      break;
#ifdef RTM_IFANNOUNCE
    case RTM_IFANNOUNCE:
      ifan_read (&buf.ian.ifan);
      break;
#endif /* RTM_IFANNOUNCE */
    default:
      if (IS_ZEBRA_DEBUG_KERNEL)
        zlog_debug("Unprocessed RTM_type: %d", rtm->rtm_type);
      break;
    }
  return 0;
}

/* Make routing socket. */
void
routing_socket ()
{
  if ( zserv_privs.change (ZPRIVS_RAISE) )
    zlog_err ("routing_socket: Can't raise privileges");

  routing_sock = socket (AF_ROUTE, SOCK_RAW, 0);

  if (routing_sock < 0) 
    {
      if ( zserv_privs.change (ZPRIVS_LOWER) )
        zlog_err ("routing_socket: Can't lower privileges");
      zlog_warn ("Can't init kernel routing socket");
      return;
    }

  /* XXX: Socket should be NONBLOCK, however as we currently 
   * discard failed writes, this will lead to inconsistencies.
   * For now, socket must be blocking.
   */
  /*if (fcntl (routing_sock, F_SETFL, O_NONBLOCK) < 0) 
    zlog_warn ("Can't set O_NONBLOCK to routing socket");*/
    
  if ( zserv_privs.change (ZPRIVS_LOWER) )
    zlog_err ("routing_socket: Can't lower privileges");

  /* kernel_read needs rewrite. */
  thread_add_read (zebrad.master, kernel_read, NULL, routing_sock);
}

/* Exported interface function.  This function simply calls
   routing_socket (). */
void
kernel_init ()
{
  routing_socket ();
}