#include "zebra/interface.h"
#include "zebra/zserv.h"
#include "zebra/debug.h"
+#include "zebra/kernel_socket.h"
extern struct zebra_privs_t zserv_privs;
extern struct zebra_t zebrad;
-/* Socket length roundup function. */
+/*
+ * 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))
-/* And this macro is wrapper for handling sa_len. */
-#if defined(HAVE_SA_LEN)
-#define WRAPUP(X) ROUNDUP(((struct sockaddr *)(X))->sa_len)
+/*
+ * 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_STRUCT_SOCKADDR_SA_LEN)
+#define SAROUNDUP(X) ROUNDUP(((struct sockaddr *)(X))->sa_len)
#elif defined(HAVE_IPV6)
-#define WRAPUP(X) \
- do { \
+/*
+ * 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)))) \
- } while (0)
+ ROUNDUP(sizeof(struct sockaddr_dl)) : sizeof(struct sockaddr))))
#else /* HAVE_IPV6 */
-#define WRAPUP(X) \
+#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 */
+#endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
+/* We use an additional pointer in following, pdest, rather than (DEST)
+ * directly, because gcc will warn if the macro is expanded and DEST is NULL,
+ * complaining that memcpy is being passed a NULL value, despite the fact
+ * the if (NULL) makes it impossible.
+ */
+#define RTA_ADDR_GET(DEST, RTA, RTMADDRS, PNT) \
+ if ((RTMADDRS) & (RTA)) \
+ { \
+ void *pdest = (DEST); \
+ int len = SAROUNDUP ((PNT)); \
+ if ( ((DEST) != NULL) && \
+ af_check (((struct sockaddr *)(PNT))->sa_family)) \
+ memcpy (pdest, (PNT), len); \
+ (PNT) += len; \
+ }
+#define RTA_ATTR_GET(DEST, RTA, RTMADDRS, PNT) \
+ if ((RTMADDRS) & (RTA)) \
+ { \
+ void *pdest = (DEST); \
+ int len = SAROUNDUP ((PNT)); \
+ if ((DEST) != NULL) \
+ memcpy (pdest, (PNT), len); \
+ (PNT) += len; \
+ }
+
+#define RTA_NAME_GET(DEST, RTA, RTMADDRS, PNT, LEN) \
+ if ((RTMADDRS) & (RTA)) \
+ { \
+ u_char *pdest = (u_char *) (DEST); \
+ 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 (pdest, sdl->sdl_data, sdl->sdl_nlen); \
+ pdest[sdl->sdl_nlen] = '\0'; \
+ (LEN) = sdl->sdl_nlen; \
+ } \
+ (PNT) += len; \
+ } \
+ else \
+ { \
+ (LEN) = 0; \
+ }
/* Routing socket message types. */
-struct message rtm_type_str[] =
+const struct message rtm_type_str[] =
{
{RTM_ADD, "RTM_ADD"},
{RTM_DELETE, "RTM_DELETE"},
{RTM_REDIRECT, "RTM_REDIRECT"},
{RTM_MISS, "RTM_MISS"},
{RTM_LOCK, "RTM_LOCK"},
+#ifdef OLDADD
{RTM_OLDADD, "RTM_OLDADD"},
+#endif /* RTM_OLDADD */
+#ifdef RTM_OLDDEL
{RTM_OLDDEL, "RTM_OLDDEL"},
+#endif /* RTM_OLDDEL */
{RTM_RESOLVE, "RTM_RESOLVE"},
{RTM_NEWADDR, "RTM_NEWADDR"},
{RTM_DELADDR, "RTM_DELADDR"},
{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
#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}
};
/* #define DEBUG */
/* Supported address family check. */
-static int
+static int inline
af_check (int family)
{
if (family == AF_INET)
}
\f
/* Dump routing table flag for debug purpose. */
-void
+static void
rtm_flag_dump (int flag)
{
struct message *mes;
static char buf[BUFSIZ];
- buf[0] = '0';
+ buf[0] = '\0';
for (mes = rtm_flag_str; mes->key != 0; mes++)
{
if (mes->key & flag)
strlcat (buf, " ", BUFSIZ);
}
}
- zlog_info ("Kernel: %s", buf);
+ zlog_debug ("Kernel: %s", buf);
}
#ifdef RTM_IFANNOUNCE
/* Interface adding function */
-int
+static int
ifan_read (struct if_announcemsghdr *ifan)
{
struct interface *ifp;
-
+
ifp = if_lookup_by_index (ifan->ifan_index);
- if (ifp == NULL && ifan->ifan_what == IFAN_ARRIVAL)
+
+ 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 (ifan->ifan_name);
+ 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_delete (ifp);
- }
+ if_delete_update (ifp);
if_get_flags (ifp);
if_get_mtu (ifp);
if_get_metric (ifp);
if (IS_ZEBRA_DEBUG_KERNEL)
- zlog_info ("interface %s index %d", ifp->name, ifp->ifindex);
+ zlog_debug ("%s: interface %s index %d",
+ __func__, ifan->ifan_name, ifan->ifan_index);
return 0;
}
#endif /* RTM_IFANNOUNCE */
-/* Interface adding function called from interface_list. */
+#ifdef HAVE_BSD_LINK_DETECT
+/* BSD link detect translation */
+static void
+bsd_linkdetect_translate (struct if_msghdr *ifm)
+{
+ if ((ifm->ifm_data.ifi_link_state >= LINK_STATE_UP) ||
+ (ifm->ifm_data.ifi_link_state == LINK_STATE_UNKNOWN))
+ SET_FLAG(ifm->ifm_flags, IFF_RUNNING);
+ else
+ UNSET_FLAG(ifm->ifm_flags, IFF_RUNNING);
+}
+#endif /* HAVE_BSD_LINK_DETECT */
+
+/*
+ * 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;
- struct sockaddr_dl *sdl = 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;
+ }
-#ifdef SUNOS_5
- int i;
- struct sockaddr *sa;
- u_char *cp = (u_char *)(ifm + 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, so make adjustment
- * here.
+ * is 12 bytes larger than the 32 bit version.
*/
- sa = (struct sockaddr *)cp;
- if (sa->sa_family == AF_UNSPEC)
+ if (((struct sockaddr *) cp)->sa_family == AF_UNSPEC)
cp = cp + 12;
+#endif
- for (i = 1; i != 0; i <<= 1)
+ 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 )
{
- if (i & ifm->ifm_addrs)
+ /* 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) )
{
- sa = (struct sockaddr *)cp;
- cp += WRAPUP(sa);
- if (i & RTA_IFP)
- {
- sdl = (struct sockaddr_dl *)sa;
- break;
- }
+ if (IS_ZEBRA_DEBUG_KERNEL)
+ zlog_debug ("%s: ifp name %s doesnt match sdl name %s",
+ __func__, ifp->name, ifname);
+ ifp = NULL;
}
}
-#else
- sdl = (struct sockaddr_dl *)(ifm + 1);
-#endif
-
+
/*
- * Check if ifp already exists. If the interface has already been specified
- * in the conf file, but is just getting created, we would have an
- * entry in the iflist with incomplete data (e.g., ifindex == -1),
- * so we lookup on name.
+ * 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 (sdl != NULL)
- {
- memcpy (ifname, sdl->sdl_data, sdl->sdl_nlen);
- ifname[sdl->sdl_nlen] = '\0';
- ifp = if_lookup_by_name (ifname);
- }
+ if ( (ifp == NULL) && ifnlen)
+ ifp = if_lookup_by_name (ifname);
- if ((ifp == NULL) || (ifp->ifindex == -1))
+ /*
+ * 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))
{
- /* Check interface's address.*/
- if (! (ifm->ifm_addrs & RTA_IFP))
+ /*
+ * To create or fill in an interface, a sockaddr_dl (via
+ * RTA_IFP) is required.
+ */
+ if (!ifnlen)
{
- zlog_warn ("There must be RTA_IFP address for ifindex %d\n",
+ zlog_warn ("Interface index %d (new) missing ifname\n",
ifm->ifm_index);
return -1;
}
+#ifndef RTM_IFANNOUNCE
+ /* Down->Down interface should be ignored here.
+ * See further comment below.
+ */
+ if (!CHECK_FLAG (ifm->ifm_flags, IFF_UP))
+ return 0;
+#endif /* !RTM_IFANNOUNCE */
+
if (ifp == NULL)
- ifp = if_create (sdl->sdl_data, sdl->sdl_nlen);
+ {
+ /* 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;
+
+#ifdef HAVE_BSD_LINK_DETECT /* translate BSD kernel msg for link-state */
+ bsd_linkdetect_translate(ifm);
+#endif /* HAVE_BSD_LINK_DETECT */
+
+ if_flags_update (ifp, ifm->ifm_flags);
#if defined(__bsdi__)
if_kvm_get_mtu (ifp);
#else
#endif /* __bsdi__ */
if_get_metric (ifp);
- /* Fetch hardware address. */
- if (sdl->sdl_family != AF_LINK)
- {
- zlog_warn ("sockaddr_dl->sdl_family is not AF_LINK");
- return -1;
- }
- memcpy (&ifp->sdl, sdl, sizeof (struct sockaddr_dl));
-
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.)
+ */
{
- /* There is a case of promisc, allmulti flag modification. */
+ 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;
+ }
+
+#ifdef HAVE_BSD_LINK_DETECT /* translate BSD kernel msg for link-state */
+ bsd_linkdetect_translate(ifm);
+#endif /* HAVE_BSD_LINK_DETECT */
+
+ /* update flags and handle operative->inoperative transition, if any */
+ if_flags_update (ifp, ifm->ifm_flags);
+
+#ifndef RTM_IFANNOUNCE
+ if (!if_is_up (ifp))
+ {
+ /* No RTM_IFANNOUNCE on this platform, so we can never
+ * distinguish between ~IFF_UP 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 */
if (if_is_up (ifp))
- {
- ifp->flags = ifm->ifm_flags;
- if (! if_is_up (ifp))
- if_down (ifp);
- }
- else
- {
- ifp->flags = ifm->ifm_flags;
- if (if_is_up (ifp))
- if_up (ifp);
- }
+ {
+#if defined(__bsdi__)
+ if_kvm_get_mtu (ifp);
+#else
+ if_get_mtu (ifp);
+#endif /* __bsdi__ */
+ if_get_metric (ifp);
+ }
}
-
+
#ifdef HAVE_NET_RT_IFLIST
ifp->stats = ifm->ifm_data;
#endif /* HAVE_NET_RT_IFLIST */
if (IS_ZEBRA_DEBUG_KERNEL)
- zlog_info ("interface %s index %d", ifp->name, ifp->ifindex);
+ zlog_debug ("%s: interface %s index %d",
+ __func__, ifp->name, ifp->ifindex);
return 0;
}
\f
/* Address read from struct ifa_msghdr. */
-void
+static void
ifam_read_mesg (struct ifa_msghdr *ifm,
union sockunion *addr,
union sockunion *mask,
- union sockunion *dest)
+ union sockunion *brd,
+ char *ifname,
+ short *ifnlen)
{
caddr_t pnt, end;
+ union sockunion dst;
+ union sockunion gateway;
pnt = (caddr_t)(ifm + 1);
end = ((caddr_t)ifm) + ifm->ifam_msglen;
-#define IFAMADDRGET(X,R) \
- if (ifm->ifam_addrs & (R)) \
- { \
- int len = WRAPUP(pnt); \
- if (((X) != NULL) && af_check (((struct sockaddr *)pnt)->sa_family)) \
- memcpy ((caddr_t)(X), pnt, len); \
- pnt += len; \
- }
-#define IFAMMASKGET(X,R) \
- if (ifm->ifam_addrs & (R)) \
- { \
- int len = WRAPUP(pnt); \
- if ((X) != NULL) \
- memcpy ((caddr_t)(X), pnt, len); \
- pnt += len; \
- }
-
/* Be sure structure is cleared */
memset (mask, 0, sizeof (union sockunion));
memset (addr, 0, sizeof (union sockunion));
- memset (dest, 0, sizeof (union sockunion));
+ memset (brd, 0, sizeof (union sockunion));
+ memset (&dst, 0, sizeof (union sockunion));
+ memset (&gateway, 0, sizeof (union sockunion));
/* We fetch each socket variable into sockunion. */
- IFAMADDRGET (NULL, RTA_DST);
- IFAMADDRGET (NULL, RTA_GATEWAY);
- IFAMMASKGET (mask, RTA_NETMASK);
- IFAMADDRGET (NULL, RTA_GENMASK);
- IFAMADDRGET (NULL, RTA_IFP);
- IFAMADDRGET (addr, RTA_IFA);
- IFAMADDRGET (NULL, RTA_AUTHOR);
- IFAMADDRGET (dest, RTA_BRD);
+ RTA_ADDR_GET (&dst, RTA_DST, ifm->ifam_addrs, pnt);
+ RTA_ADDR_GET (&gateway, 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)
+ {
+ switch (sockunion_family(addr))
+ {
+ case AF_INET:
+ {
+ char buf[4][INET_ADDRSTRLEN];
+ zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x, "
+ "ifam_flags 0x%x, addr %s/%d broad %s dst %s "
+ "gateway %s",
+ __func__, ifm->ifam_index,
+ (ifnlen ? ifname : "(nil)"), ifm->ifam_addrs,
+ ifm->ifam_flags,
+ inet_ntop(AF_INET,&addr->sin.sin_addr,
+ buf[0],sizeof(buf[0])),
+ ip_masklen(mask->sin.sin_addr),
+ inet_ntop(AF_INET,&brd->sin.sin_addr,
+ buf[1],sizeof(buf[1])),
+ inet_ntop(AF_INET,&dst.sin.sin_addr,
+ buf[2],sizeof(buf[2])),
+ inet_ntop(AF_INET,&gateway.sin.sin_addr,
+ buf[3],sizeof(buf[3])));
+ }
+ break;
+#ifdef HAVE_IPV6
+ case AF_INET6:
+ {
+ char buf[4][INET6_ADDRSTRLEN];
+ zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x, "
+ "ifam_flags 0x%x, addr %s/%d broad %s dst %s "
+ "gateway %s",
+ __func__, ifm->ifam_index,
+ (ifnlen ? ifname : "(nil)"), ifm->ifam_addrs,
+ ifm->ifam_flags,
+ inet_ntop(AF_INET6,&addr->sin6.sin6_addr,
+ buf[0],sizeof(buf[0])),
+ ip6_masklen(mask->sin6.sin6_addr),
+ inet_ntop(AF_INET6,&brd->sin6.sin6_addr,
+ buf[1],sizeof(buf[1])),
+ inet_ntop(AF_INET6,&dst.sin6.sin6_addr,
+ buf[2],sizeof(buf[2])),
+ inet_ntop(AF_INET6,&gateway.sin6.sin6_addr,
+ buf[3],sizeof(buf[3])));
+ }
+ break;
+#endif /* HAVE_IPV6 */
+ default:
+ zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x",
+ __func__, ifm->ifam_index,
+ (ifnlen ? ifname : "(nil)"), ifm->ifam_addrs);
+ break;
+ }
+ }
/* Assert read up end point matches to end point */
if (pnt != end)
int
ifam_read (struct ifa_msghdr *ifam)
{
- struct interface *ifp;
- union sockunion addr, mask, gate;
-
- /* Check does this interface exist or not. */
- ifp = if_lookup_by_index (ifam->ifam_index);
- if (ifp == NULL)
+ struct interface *ifp = NULL;
+ union sockunion addr, mask, brd;
+ char ifname[INTERFACE_NAMSIZ];
+ short ifnlen = 0;
+ char isalias = 0;
+ int flags = 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 ("no interface for index %d", ifam->ifam_index);
+ zlog_warn ("%s: no interface for ifname %s, index %d",
+ __func__, ifname, ifam->ifam_index);
return -1;
}
-
- /* Allocate and read address information. */
- ifam_read_mesg (ifam, &addr, &mask, &gate);
-
- /* Check interface flag for implicit up of the interface. */
- if_refresh (ifp);
+
+ if (ifnlen && strncmp (ifp->name, ifname, INTERFACE_NAMSIZ))
+ isalias = 1;
+
+ /* N.B. The info in ifa_msghdr does not tell us whether the RTA_BRD
+ field contains a broadcast address or a peer address, so we are forced to
+ rely upon the interface type. */
+ if (if_is_pointopoint(ifp))
+ SET_FLAG(flags, ZEBRA_IFA_PEER);
+
+#if 0
+ /* it might seem cute to grab the interface metric here, however
+ * we're processing an address update message, and so some systems
+ * (e.g. FBSD) dont bother to fill in ifam_metric. Disabled, but left
+ * in deliberately, as comment.
+ */
+ ifp->metric = ifam->ifam_metric;
+#endif
/* 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,
+ connected_add_ipv4 (ifp, flags, &addr.sin.sin_addr,
ip_masklen (mask.sin.sin_addr),
- &gate.sin.sin_addr, NULL);
+ &brd.sin.sin_addr,
+ (isalias ? ifname : NULL));
else
- connected_delete_ipv4 (ifp, 0, &addr.sin.sin_addr,
+ connected_delete_ipv4 (ifp, flags, &addr.sin.sin_addr,
ip_masklen (mask.sin.sin_addr),
- &gate.sin.sin_addr, NULL);
+ &brd.sin.sin_addr);
break;
#ifdef HAVE_IPV6
case AF_INET6:
SET_IN6_LINKLOCAL_IFINDEX (addr.sin6.sin6_addr, 0);
if (ifam->ifam_type == RTM_NEWADDR)
- connected_add_ipv6 (ifp,
- &addr.sin6.sin6_addr,
+ connected_add_ipv6 (ifp, flags, &addr.sin6.sin6_addr,
ip6_masklen (mask.sin6.sin6_addr),
- &gate.sin6.sin6_addr);
+ &brd.sin6.sin6_addr,
+ (isalias ? ifname : NULL));
else
connected_delete_ipv6 (ifp,
&addr.sin6.sin6_addr,
ip6_masklen (mask.sin6.sin6_addr),
- &gate.sin6.sin6_addr);
+ &brd.sin6.sin6_addr);
break;
#endif /* HAVE_IPV6 */
default:
/* Unsupported family silently ignore... */
break;
}
+
+ /* Check interface flag for implicit up of the interface. */
+ if_refresh (ifp);
+
+#ifdef SUNOS_5
+ /* In addition to lacking IFANNOUNCE, on SUNOS IFF_UP is strange.
+ * See comments for SUNOS_5 in interface.c::if_flags_mangle.
+ *
+ * Here we take care of case where the real IFF_UP was previously
+ * unset (as kept in struct zebra_if.primary_state) and the mangled
+ * IFF_UP (ie IFF_UP set || listcount(connected) has now transitioned
+ * to unset due to the lost non-primary address having DELADDR'd.
+ *
+ * we must delete the interface, because in between here and next
+ * event for this interface-name the administrator could unplumb
+ * and replumb the interface.
+ */
+ if (!if_is_up (ifp))
+ if_delete_update (ifp);
+#endif /* SUNOS_5 */
+
return 0;
}
\f
/* Interface function for reading kernel routing table information. */
-int
+static int
rtm_read_mesg (struct rt_msghdr *rtm,
union sockunion *dest,
union sockunion *mask,
- union sockunion *gate)
+ union sockunion *gate,
+ char *ifname,
+ short *ifnlen)
{
caddr_t pnt, end;
zlog (NULL, LOG_WARNING,
"Routing message version different %d should be %d."
"This may cause problem\n", rtm->rtm_version, RTM_VERSION);
-
-#define RTMADDRGET(X,R) \
- if (rtm->rtm_addrs & (R)) \
- { \
- int len = WRAPUP (pnt); \
- if (((X) != NULL) && af_check (((struct sockaddr *)pnt)->sa_family)) \
- memcpy ((caddr_t)(X), pnt, len); \
- pnt += len; \
- }
-#define RTMMASKGET(X,R) \
- if (rtm->rtm_addrs & (R)) \
- { \
- int len = WRAPUP (pnt); \
- if ((X) != NULL) \
- memcpy ((caddr_t)(X), pnt, len); \
- pnt += len; \
- }
-
+
/* 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. */
- RTMADDRGET (dest, RTA_DST);
- RTMADDRGET (gate, RTA_GATEWAY);
- RTMMASKGET (mask, RTA_NETMASK);
- RTMADDRGET (NULL, RTA_GENMASK);
- RTMADDRGET (NULL, RTA_IFP);
- RTMADDRGET (NULL, RTA_IFA);
- RTMADDRGET (NULL, RTA_AUTHOR);
- RTMADDRGET (NULL, RTA_BRD);
+ 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*/
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);
+ flags = rtm_read_mesg (rtm, &dest, &mask, &gate, ifname, &ifnlen);
+ if (!(flags & RTF_DONE))
+ return;
+ if (IS_ZEBRA_DEBUG_KERNEL)
+ zlog_debug ("%s: got rtm of type %d (%s)", __func__, rtm->rtm_type,
+ lookup (rtm_type_str, rtm->rtm_type));
#ifdef RTF_CLONED /*bsdi, netbsd 1.6*/
if (flags & RTF_CLONED)
p.prefixlen = IPV4_MAX_PREFIXLEN;
else
p.prefixlen = ip_masklen (mask.sin.sin_addr);
+
+ /* Catch self originated messages and match them against our current RIB.
+ * At the same time, ignore unconfirmed messages, they should be tracked
+ * by rtm_write() and kernel_rtm_ipv4().
+ */
+ if (rtm->rtm_type != RTM_GET && rtm->rtm_pid == pid)
+ {
+ char buf[INET_ADDRSTRLEN], gate_buf[INET_ADDRSTRLEN];
+ int ret;
+ if (! IS_ZEBRA_DEBUG_RIB)
+ return;
+ ret = rib_lookup_ipv4_route (&p, &gate);
+ inet_ntop (AF_INET, &p.prefix, buf, INET_ADDRSTRLEN);
+ switch (rtm->rtm_type)
+ {
+ case RTM_ADD:
+ case RTM_GET:
+ case RTM_CHANGE:
+ /* The kernel notifies us about a new route in FIB created by us.
+ Do we have a correspondent entry in our RIB? */
+ switch (ret)
+ {
+ case ZEBRA_RIB_NOTFOUND:
+ zlog_debug ("%s: %s %s/%d: desync: RR isn't yet in RIB, while already in FIB",
+ __func__, lookup (rtm_type_str, rtm->rtm_type), buf, p.prefixlen);
+ break;
+ case ZEBRA_RIB_FOUND_CONNECTED:
+ case ZEBRA_RIB_FOUND_NOGATE:
+ inet_ntop (AF_INET, &gate.sin.sin_addr, gate_buf, INET_ADDRSTRLEN);
+ zlog_debug ("%s: %s %s/%d: desync: RR is in RIB, but gate differs (ours is %s)",
+ __func__, lookup (rtm_type_str, rtm->rtm_type), buf, p.prefixlen, gate_buf);
+ break;
+ case ZEBRA_RIB_FOUND_EXACT: /* RIB RR == FIB RR */
+ zlog_debug ("%s: %s %s/%d: done Ok",
+ __func__, lookup (rtm_type_str, rtm->rtm_type), buf, p.prefixlen);
+ rib_lookup_and_dump (&p);
+ return;
+ break;
+ }
+ break;
+ case RTM_DELETE:
+ /* The kernel notifies us about a route deleted by us. Do we still
+ have it in the RIB? Do we have anything instead? */
+ switch (ret)
+ {
+ case ZEBRA_RIB_FOUND_EXACT:
+ zlog_debug ("%s: %s %s/%d: desync: RR is still in RIB, while already not in FIB",
+ __func__, lookup (rtm_type_str, rtm->rtm_type), buf, p.prefixlen);
+ rib_lookup_and_dump (&p);
+ break;
+ case ZEBRA_RIB_FOUND_CONNECTED:
+ case ZEBRA_RIB_FOUND_NOGATE:
+ zlog_debug ("%s: %s %s/%d: desync: RR is still in RIB, plus gate differs",
+ __func__, lookup (rtm_type_str, rtm->rtm_type), buf, p.prefixlen);
+ rib_lookup_and_dump (&p);
+ break;
+ case ZEBRA_RIB_NOTFOUND: /* RIB RR == FIB RR */
+ zlog_debug ("%s: %s %s/%d: done Ok",
+ __func__, lookup (rtm_type_str, rtm->rtm_type), buf, p.prefixlen);
+ rib_lookup_and_dump (&p);
+ return;
+ break;
+ }
+ break;
+ default:
+ zlog_debug ("%s: %s/%d: warning: loopback RTM of type %s received",
+ __func__, buf, p.prefixlen, lookup (rtm_type_str, rtm->rtm_type));
+ }
+ return;
+ }
- if (rtm->rtm_type == RTM_GET || rtm->rtm_type == RTM_ADD)
+ /* 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);
+ &p, &gate.sin.sin_addr, NULL, 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)
{
+ /* One day we might have a debug section here like one in the
+ * IPv4 case above. Just ignore own messages at the moment.
+ */
+ if (rtm->rtm_type != RTM_GET && rtm->rtm_pid == pid)
+ return;
struct prefix_ipv6 p;
unsigned int ifindex = 0;
}
#endif /* KAME */
- if (rtm->rtm_type == RTM_GET || rtm->rtm_type == RTM_ADD)
+ /* CHANGE: delete the old prefix, we have no further information
+ * to specify the route really
+ */
+ if (rtm->rtm_type == RTM_CHANGE)
+ rib_delete_ipv6 (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);
+ &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);
}
/* Interface function for the kernel routing table updates. Support
- for RTM_CHANGE will be needed. */
+ * for RTM_CHANGE will be needed.
+ * Exported only for rt_socket.c
+ */
int
rtm_write (int message,
union sockunion *dest,
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;
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;
{
if (!ifp)
{
- zlog_warn ("no gateway found for interface index %d", index);
+ char dest_buf[INET_ADDRSTRLEN] = "NULL", mask_buf[INET_ADDRSTRLEN] = "255.255.255.255";
+ if (dest)
+ inet_ntop (AF_INET, &dest->sin.sin_addr, dest_buf, INET_ADDRSTRLEN);
+ if (mask)
+ inet_ntop (AF_INET, &mask->sin.sin_addr, mask_buf, INET_ADDRSTRLEN);
+ zlog_warn ("%s: %s/%s: gate == NULL and no gateway found for ifindex %d",
+ __func__, dest_buf, mask_buf, index);
return -1;
}
gate = (union sockunion *) & ifp->sdl;
msg.rtm.rtm_flags |= RTF_REJECT;
-#ifdef HAVE_SIN_LEN
+#ifdef HAVE_STRUCT_SOCKADDR_IN_SIN_LEN
#define SOCKADDRSET(X,R) \
if (msg.rtm.rtm_addrs & (R)) \
{ \
#define SOCKADDRSET(X,R) \
if (msg.rtm.rtm_addrs & (R)) \
{ \
- int len = ROUNDUP (sizeof((X)->sa)); \
+ int len = SAROUNDUP (X); \
memcpy (pnt, (caddr_t)(X), len); \
pnt += len; \
}
-#endif /* HAVE_SIN_LEN */
+#endif /* HAVE_STRUCT_SOCKADDR_IN_SIN_LEN */
pnt = (caddr_t) msg.buf;
return ZEBRA_ERR_RTEXIST;
if (errno == ENETUNREACH)
return ZEBRA_ERR_RTUNREACH;
+ if (errno == ESRCH)
+ return ZEBRA_ERR_RTNOEXIST;
- zlog_warn ("write : %s (%d)", strerror (errno), errno);
- return -1;
+ zlog_warn ("%s: write : %s (%d)", __func__, safe_strerror (errno), errno);
+ return ZEBRA_ERR_KERNEL;
}
- return 0;
+ return ZEBRA_ERR_NOERROR;
}
\f
#include "zebra/zserv.h"
/* For debug purpose. */
-void
+static void
rtmsg_debug (struct rt_msghdr *rtm)
{
- 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_info ("Kernel: Len: %d Type: %s", rtm->rtm_msglen, type);
+ zlog_debug ("Kernel: Len: %d Type: %s", rtm->rtm_msglen, lookup (rtm_type_str, rtm->rtm_type));
rtm_flag_dump (rtm->rtm_flags);
- zlog_info ("Kernel: message seq %d", rtm->rtm_seq);
- zlog_info ("Kernel: pid %d", rtm->rtm_pid);
+ 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 */
#endif /* RTAX_MAX */
/* Kernel routing table and interface updates via routing socket. */
-int
+static 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 addr[RTAX_MAX];
+ struct sockaddr_storage addr[RTAX_MAX];
} r;
/* Interface information. */
struct
{
struct if_msghdr ifm;
- struct sockaddr addr[RTAX_MAX];
+ struct sockaddr_storage addr[RTAX_MAX];
} im;
/* Interface address information. */
struct
{
struct ifa_msghdr ifa;
- struct sockaddr addr[RTAX_MAX];
+ struct sockaddr_storage addr[RTAX_MAX];
} ia;
#ifdef RTM_IFANNOUNCE
struct
{
struct if_announcemsghdr ifan;
- struct sockaddr addr[RTAX_MAX];
+ struct sockaddr_storage addr[RTAX_MAX];
} ian;
#endif /* RTM_IFANNOUNCE */
if (nbytes <= 0)
{
if (nbytes < 0 && errno != EWOULDBLOCK && errno != EAGAIN)
- zlog_warn ("routing socket error: %s", strerror (errno));
+ zlog_warn ("routing socket error: %s", safe_strerror (errno));
return 0;
}
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:
#endif /* RTM_IFANNOUNCE */
default:
if (IS_ZEBRA_DEBUG_KERNEL)
- zlog_info("Unprocessed RTM_type: %d", rtm->rtm_type);
+ zlog_debug("Unprocessed RTM_type: %d", rtm->rtm_type);
break;
}
return 0;
}
/* Make routing socket. */
-void
-routing_socket ()
+static void
+routing_socket (void)
{
if ( zserv_privs.change (ZPRIVS_RAISE) )
zlog_err ("routing_socket: Can't raise privileges");
return;
}
- if (fcntl (routing_sock, F_SETFL, O_NONBLOCK) < 0)
- zlog_warn ("Can't set O_NONBLOCK to routing socket");
+ /* 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");
/* Exported interface function. This function simply calls
routing_socket (). */
void
-kernel_init ()
+kernel_init (void)
{
routing_socket ();
}