1 /* Kernel communication using routing socket.
2 * Copyright (C) 1999 Kunihiro Ishiguro
4 * This file is part of GNU Zebra.
6 * GNU Zebra is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; either version 2, or (at your option) any
11 * GNU Zebra is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with GNU Zebra; see the file COPYING. If not, write to the Free
18 * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
26 #include "sockunion.h"
27 #include "connected.h"
36 #include "zebra/interface.h"
37 #include "zebra/zserv.h"
38 #include "zebra/debug.h"
39 #include "zebra/kernel_socket.h"
41 extern struct zebra_privs_t zserv_privs
;
42 extern struct zebra_t zebrad
;
45 * Given a sockaddr length, round it up to include pad bytes following
46 * it. Assumes the kernel pads to sizeof(long).
48 * XXX: why is ROUNDUP(0) sizeof(long)? 0 is an illegal sockaddr
49 * length anyway (< sizeof (struct sockaddr)), so this shouldn't
53 ((a) > 0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
56 * Given a pointer (sockaddr or void *), return the number of bytes
57 * taken up by the sockaddr and any padding needed for alignment.
59 #if defined(HAVE_STRUCT_SOCKADDR_SA_LEN)
60 #define SAROUNDUP(X) ROUNDUP(((struct sockaddr *)(X))->sa_len)
61 #elif defined(HAVE_IPV6)
63 * One would hope all fixed-size structure definitions are aligned,
64 * but round them up nonetheless.
66 #define SAROUNDUP(X) \
67 (((struct sockaddr *)(X))->sa_family == AF_INET ? \
68 ROUNDUP(sizeof(struct sockaddr_in)):\
69 (((struct sockaddr *)(X))->sa_family == AF_INET6 ? \
70 ROUNDUP(sizeof(struct sockaddr_in6)) : \
71 (((struct sockaddr *)(X))->sa_family == AF_LINK ? \
72 ROUNDUP(sizeof(struct sockaddr_dl)) : sizeof(struct sockaddr))))
74 #define SAROUNDUP(X) \
75 (((struct sockaddr *)(X))->sa_family == AF_INET ? \
76 ROUNDUP(sizeof(struct sockaddr_in)):\
77 (((struct sockaddr *)(X))->sa_family == AF_LINK ? \
78 ROUNDUP(sizeof(struct sockaddr_dl)) : sizeof(struct sockaddr)))
79 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
81 /* We use an additional pointer in following, pdest, rather than (DEST)
82 * directly, because gcc will warn if the macro is expanded and DEST is NULL,
83 * complaining that memcpy is being passed a NULL value, despite the fact
84 * the if (NULL) makes it impossible.
86 #define RTA_ADDR_GET(DEST, RTA, RTMADDRS, PNT) \
87 if ((RTMADDRS) & (RTA)) \
89 void *pdest = (DEST); \
90 int len = SAROUNDUP ((PNT)); \
91 if ( ((DEST) != NULL) && \
92 af_check (((struct sockaddr *)(PNT))->sa_family)) \
93 memcpy (pdest, (PNT), len); \
96 #define RTA_ATTR_GET(DEST, RTA, RTMADDRS, PNT) \
97 if ((RTMADDRS) & (RTA)) \
99 void *pdest = (DEST); \
100 int len = SAROUNDUP ((PNT)); \
101 if ((DEST) != NULL) \
102 memcpy (pdest, (PNT), len); \
106 #define RTA_NAME_GET(DEST, RTA, RTMADDRS, PNT, LEN) \
107 if ((RTMADDRS) & (RTA)) \
109 u_char *pdest = (u_char *) (DEST); \
110 int len = SAROUNDUP ((PNT)); \
111 struct sockaddr_dl *sdl = (struct sockaddr_dl *)(PNT); \
112 if (IS_ZEBRA_DEBUG_KERNEL) \
113 zlog_debug ("%s: RTA_SDL_GET nlen %d, alen %d", \
114 __func__, sdl->sdl_nlen, sdl->sdl_alen); \
115 if ( ((DEST) != NULL) && (sdl->sdl_family == AF_LINK) \
116 && (sdl->sdl_nlen < IFNAMSIZ) && (sdl->sdl_nlen <= len) ) \
118 memcpy (pdest, sdl->sdl_data, sdl->sdl_nlen); \
119 pdest[sdl->sdl_nlen] = '\0'; \
120 (LEN) = sdl->sdl_nlen; \
128 /* Routing socket message types. */
129 struct message rtm_type_str
[] =
131 {RTM_ADD
, "RTM_ADD"},
132 {RTM_DELETE
, "RTM_DELETE"},
133 {RTM_CHANGE
, "RTM_CHANGE"},
134 {RTM_GET
, "RTM_GET"},
135 {RTM_LOSING
, "RTM_LOSING"},
136 {RTM_REDIRECT
, "RTM_REDIRECT"},
137 {RTM_MISS
, "RTM_MISS"},
138 {RTM_LOCK
, "RTM_LOCK"},
140 {RTM_OLDADD
, "RTM_OLDADD"},
141 #endif /* RTM_OLDADD */
143 {RTM_OLDDEL
, "RTM_OLDDEL"},
144 #endif /* RTM_OLDDEL */
145 {RTM_RESOLVE
, "RTM_RESOLVE"},
146 {RTM_NEWADDR
, "RTM_NEWADDR"},
147 {RTM_DELADDR
, "RTM_DELADDR"},
148 {RTM_IFINFO
, "RTM_IFINFO"},
150 {RTM_OIFINFO
, "RTM_OIFINFO"},
151 #endif /* RTM_OIFINFO */
153 {RTM_NEWMADDR
, "RTM_NEWMADDR"},
154 #endif /* RTM_NEWMADDR */
156 {RTM_DELMADDR
, "RTM_DELMADDR"},
157 #endif /* RTM_DELMADDR */
158 #ifdef RTM_IFANNOUNCE
159 {RTM_IFANNOUNCE
, "RTM_IFANNOUNCE"},
160 #endif /* RTM_IFANNOUNCE */
163 int rtm_type_str_max
= sizeof (rtm_type_str
) / sizeof (struct message
) - 1;
165 struct message rtm_flag_str
[] =
168 {RTF_GATEWAY
, "GATEWAY"},
170 {RTF_REJECT
, "REJECT"},
171 {RTF_DYNAMIC
, "DYNAMIC"},
172 {RTF_MODIFIED
, "MODIFIED"},
176 #endif /* RTF_MASK */
177 {RTF_CLONING
, "CLONING"},
178 {RTF_XRESOLVE
, "XRESOLVE"},
179 {RTF_LLINFO
, "LLINFO"},
180 {RTF_STATIC
, "STATIC"},
181 {RTF_BLACKHOLE
, "BLACKHOLE"},
183 {RTF_PRIVATE
, "PRIVATE"},
184 #endif /* RTF_PRIVATE */
185 {RTF_PROTO1
, "PROTO1"},
186 {RTF_PROTO2
, "PROTO2"},
188 {RTF_PRCLONING
, "PRCLONING"},
189 #endif /* RTF_PRCLONING */
191 {RTF_WASCLONED
, "WASCLONED"},
192 #endif /* RTF_WASCLONED */
194 {RTF_PROTO3
, "PROTO3"},
195 #endif /* RTF_PROTO3 */
197 {RTF_PINNED
, "PINNED"},
198 #endif /* RTF_PINNED */
200 {RTF_LOCAL
, "LOCAL"},
201 #endif /* RTF_LOCAL */
203 {RTF_BROADCAST
, "BROADCAST"},
204 #endif /* RTF_BROADCAST */
206 {RTF_MULTICAST
, "MULTICAST"},
207 #endif /* RTF_MULTICAST */
209 {RTF_MULTIRT
, "MULTIRT"},
210 #endif /* RTF_MULTIRT */
212 {RTF_SETSRC
, "SETSRC"},
213 #endif /* RTF_SETSRC */
217 /* Kernel routing update socket. */
218 int routing_sock
= -1;
220 /* Yes I'm checking ugly routing socket behavior. */
223 /* Supported address family check. */
225 af_check (int family
)
227 if (family
== AF_INET
)
230 if (family
== AF_INET6
)
232 #endif /* HAVE_IPV6 */
236 /* Dump routing table flag for debug purpose. */
238 rtm_flag_dump (int flag
)
241 static char buf
[BUFSIZ
];
244 for (mes
= rtm_flag_str
; mes
->key
!= 0; mes
++)
248 strlcat (buf
, mes
->str
, BUFSIZ
);
249 strlcat (buf
, " ", BUFSIZ
);
252 zlog_debug ("Kernel: %s", buf
);
255 #ifdef RTM_IFANNOUNCE
256 /* Interface adding function */
258 ifan_read (struct if_announcemsghdr
*ifan
)
260 struct interface
*ifp
;
262 ifp
= if_lookup_by_index (ifan
->ifan_index
);
265 assert ( (ifp
->ifindex
== ifan
->ifan_index
)
266 || (ifp
->ifindex
== IFINDEX_INTERNAL
) );
269 || ((ifp
->ifindex
== IFINDEX_INTERNAL
)
270 && (ifan
->ifan_what
== IFAN_ARRIVAL
)) )
272 if (IS_ZEBRA_DEBUG_KERNEL
)
273 zlog_debug ("%s: creating interface for ifindex %d, name %s",
274 __func__
, ifan
->ifan_index
, ifan
->ifan_name
);
276 /* Create Interface */
277 ifp
= if_get_by_name_len(ifan
->ifan_name
,
278 strnlen(ifan
->ifan_name
,
279 sizeof(ifan
->ifan_name
)));
280 ifp
->ifindex
= ifan
->ifan_index
;
284 else if (ifp
!= NULL
&& ifan
->ifan_what
== IFAN_DEPARTURE
)
285 if_delete_update (ifp
);
291 if (IS_ZEBRA_DEBUG_KERNEL
)
292 zlog_debug ("%s: interface %s index %d",
293 __func__
, ifan
->ifan_name
, ifan
->ifan_index
);
297 #endif /* RTM_IFANNOUNCE */
300 * Handle struct if_msghdr obtained from reading routing socket or
301 * sysctl (from interface_list). There may or may not be sockaddrs
302 * present after the header.
305 ifm_read (struct if_msghdr
*ifm
)
307 struct interface
*ifp
= NULL
;
308 char ifname
[IFNAMSIZ
];
312 /* terminate ifname at head (for strnlen) and tail (for safety) */
313 ifname
[IFNAMSIZ
- 1] = '\0';
315 /* paranoia: sanity check structure */
316 if (ifm
->ifm_msglen
< sizeof(struct if_msghdr
))
318 zlog_err ("ifm_read: ifm->ifm_msglen %d too short\n",
324 * Check for a sockaddr_dl following the message. First, point to
325 * where a socakddr might be if one follows the message.
327 cp
= (void *)(ifm
+ 1);
331 * XXX This behavior should be narrowed to only the kernel versions
332 * for which the structures returned do not match the headers.
334 * if_msghdr_t on 64 bit kernels in Solaris 9 and earlier versions
335 * is 12 bytes larger than the 32 bit version.
337 if (((struct sockaddr
*) cp
)->sa_family
== AF_UNSPEC
)
341 RTA_ADDR_GET (NULL
, RTA_DST
, ifm
->ifm_addrs
, cp
);
342 RTA_ADDR_GET (NULL
, RTA_GATEWAY
, ifm
->ifm_addrs
, cp
);
343 RTA_ATTR_GET (NULL
, RTA_NETMASK
, ifm
->ifm_addrs
, cp
);
344 RTA_ADDR_GET (NULL
, RTA_GENMASK
, ifm
->ifm_addrs
, cp
);
345 RTA_NAME_GET (ifname
, RTA_IFP
, ifm
->ifm_addrs
, cp
, ifnlen
);
346 RTA_ADDR_GET (NULL
, RTA_IFA
, ifm
->ifm_addrs
, cp
);
347 RTA_ADDR_GET (NULL
, RTA_AUTHOR
, ifm
->ifm_addrs
, cp
);
348 RTA_ADDR_GET (NULL
, RTA_BRD
, ifm
->ifm_addrs
, cp
);
350 if (IS_ZEBRA_DEBUG_KERNEL
)
351 zlog_debug ("%s: sdl ifname %s", __func__
, (ifnlen
? ifname
: "(nil)"));
354 * Look up on ifindex first, because ifindices are the primary handle for
355 * interfaces across the user/kernel boundary, for most systems. (Some
356 * messages, such as up/down status changes on NetBSD, do not include a
359 if ( (ifp
= if_lookup_by_index (ifm
->ifm_index
)) != NULL
)
361 /* we have an ifp, verify that the name matches as some systems,
362 * eg Solaris, have a 1:many association of ifindex:ifname
363 * if they dont match, we dont have the correct ifp and should
364 * set it back to NULL to let next check do lookup by name
366 if (ifnlen
&& (strncmp (ifp
->name
, ifname
, IFNAMSIZ
) != 0) )
368 if (IS_ZEBRA_DEBUG_KERNEL
)
369 zlog_debug ("%s: ifp name %s doesnt match sdl name %s",
370 __func__
, ifp
->name
, ifname
);
376 * If we dont have an ifp, try looking up by name. Particularly as some
377 * systems (Solaris) have a 1:many mapping of ifindex:ifname - the ifname
378 * is therefore our unique handle to that interface.
380 * Interfaces specified in the configuration file for which the ifindex
381 * has not been determined will have ifindex == IFINDEX_INTERNAL, and such
382 * interfaces are found by this search, and then their ifindex values can
385 if ( (ifp
== NULL
) && ifnlen
)
386 ifp
= if_lookup_by_name (ifname
);
389 * If ifp still does not exist or has an invalid index (IFINDEX_INTERNAL),
390 * create or fill in an interface.
392 if ((ifp
== NULL
) || (ifp
->ifindex
== IFINDEX_INTERNAL
))
395 * To create or fill in an interface, a sockaddr_dl (via
396 * RTA_IFP) is required.
400 zlog_warn ("Interface index %d (new) missing ifname\n",
405 #ifndef RTM_IFANNOUNCE
406 /* Down->Down interface should be ignored here.
407 * See further comment below.
409 if (!CHECK_FLAG (ifm
->ifm_flags
, IFF_UP
))
411 #endif /* !RTM_IFANNOUNCE */
415 /* Interface that zebra was not previously aware of, so create. */
416 ifp
= if_create (ifname
, ifnlen
);
417 if (IS_ZEBRA_DEBUG_KERNEL
)
418 zlog_debug ("%s: creating ifp for ifindex %d",
419 __func__
, ifm
->ifm_index
);
422 if (IS_ZEBRA_DEBUG_KERNEL
)
423 zlog_debug ("%s: updated/created ifp, ifname %s, ifindex %d",
424 __func__
, ifp
->name
, ifp
->ifindex
);
426 * Fill in newly created interface structure, or larval
427 * structure with ifindex IFINDEX_INTERNAL.
429 ifp
->ifindex
= ifm
->ifm_index
;
430 if_flags_update (ifp
, ifm
->ifm_flags
);
431 #if defined(__bsdi__)
432 if_kvm_get_mtu (ifp
);
435 #endif /* __bsdi__ */
442 * Interface structure exists. Adjust stored flags from
443 * notification. If interface has up->down or down->up
444 * transition, call state change routines (to adjust routes,
445 * notify routing daemons, etc.). (Other flag changes are stored
446 * but apparently do not trigger action.)
449 if (ifp
->ifindex
!= ifm
->ifm_index
)
451 zlog_warn ("%s: index mismatch, ifname %s, ifp index %d, "
453 __func__
, ifp
->name
, ifp
->ifindex
, ifm
->ifm_index
);
457 /* update flags and handle operative->inoperative transition, if any */
458 if_flags_update (ifp
, ifm
->ifm_flags
);
460 #ifndef RTM_IFANNOUNCE
463 /* No RTM_IFANNOUNCE on this platform, so we can never
464 * distinguish between ~IFF_UP and delete. We must presume
465 * it has been deleted.
466 * Eg, Solaris will not notify us of unplumb.
468 * XXX: Fixme - this should be runtime detected
469 * So that a binary compiled on a system with IFANNOUNCE
470 * will still behave correctly if run on a platform without
472 if_delete_update (ifp
);
474 #endif /* RTM_IFANNOUNCE */
477 #ifdef HAVE_NET_RT_IFLIST
478 ifp
->stats
= ifm
->ifm_data
;
479 #endif /* HAVE_NET_RT_IFLIST */
481 if (IS_ZEBRA_DEBUG_KERNEL
)
482 zlog_debug ("%s: interface %s index %d",
483 __func__
, ifp
->name
, ifp
->ifindex
);
488 /* Address read from struct ifa_msghdr. */
490 ifam_read_mesg (struct ifa_msghdr
*ifm
,
491 union sockunion
*addr
,
492 union sockunion
*mask
,
493 union sockunion
*brd
,
499 union sockunion gateway
;
501 pnt
= (caddr_t
)(ifm
+ 1);
502 end
= ((caddr_t
)ifm
) + ifm
->ifam_msglen
;
504 /* Be sure structure is cleared */
505 memset (mask
, 0, sizeof (union sockunion
));
506 memset (addr
, 0, sizeof (union sockunion
));
507 memset (brd
, 0, sizeof (union sockunion
));
508 memset (&dst
, 0, sizeof (union sockunion
));
509 memset (&gateway
, 0, sizeof (union sockunion
));
511 /* We fetch each socket variable into sockunion. */
512 RTA_ADDR_GET (&dst
, RTA_DST
, ifm
->ifam_addrs
, pnt
);
513 RTA_ADDR_GET (&gateway
, RTA_GATEWAY
, ifm
->ifam_addrs
, pnt
);
514 RTA_ATTR_GET (mask
, RTA_NETMASK
, ifm
->ifam_addrs
, pnt
);
515 RTA_ADDR_GET (NULL
, RTA_GENMASK
, ifm
->ifam_addrs
, pnt
);
516 RTA_NAME_GET (ifname
, RTA_IFP
, ifm
->ifam_addrs
, pnt
, *ifnlen
);
517 RTA_ADDR_GET (addr
, RTA_IFA
, ifm
->ifam_addrs
, pnt
);
518 RTA_ADDR_GET (NULL
, RTA_AUTHOR
, ifm
->ifam_addrs
, pnt
);
519 RTA_ADDR_GET (brd
, RTA_BRD
, ifm
->ifam_addrs
, pnt
);
521 if (IS_ZEBRA_DEBUG_KERNEL
)
523 switch (sockunion_family(addr
))
527 char buf
[4][INET_ADDRSTRLEN
];
528 zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x, "
529 "ifam_flags 0x%x, addr %s/%d broad %s dst %s "
531 __func__
, ifm
->ifam_index
,
532 (ifnlen
? ifname
: "(nil)"), ifm
->ifam_addrs
,
534 inet_ntop(AF_INET
,&addr
->sin
.sin_addr
,
535 buf
[0],sizeof(buf
[0])),
536 ip_masklen(mask
->sin
.sin_addr
),
537 inet_ntop(AF_INET
,&brd
->sin
.sin_addr
,
538 buf
[1],sizeof(buf
[1])),
539 inet_ntop(AF_INET
,&dst
.sin
.sin_addr
,
540 buf
[2],sizeof(buf
[2])),
541 inet_ntop(AF_INET
,&gateway
.sin
.sin_addr
,
542 buf
[3],sizeof(buf
[3])));
548 char buf
[4][INET6_ADDRSTRLEN
];
549 zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x, "
550 "ifam_flags 0x%x, addr %s/%d broad %s dst %s "
552 __func__
, ifm
->ifam_index
,
553 (ifnlen
? ifname
: "(nil)"), ifm
->ifam_addrs
,
555 inet_ntop(AF_INET6
,&addr
->sin6
.sin6_addr
,
556 buf
[0],sizeof(buf
[0])),
557 ip6_masklen(mask
->sin6
.sin6_addr
),
558 inet_ntop(AF_INET6
,&brd
->sin6
.sin6_addr
,
559 buf
[1],sizeof(buf
[1])),
560 inet_ntop(AF_INET6
,&dst
.sin6
.sin6_addr
,
561 buf
[2],sizeof(buf
[2])),
562 inet_ntop(AF_INET6
,&gateway
.sin6
.sin6_addr
,
563 buf
[3],sizeof(buf
[3])));
566 #endif /* HAVE_IPV6 */
568 zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x",
569 __func__
, ifm
->ifam_index
,
570 (ifnlen
? ifname
: "(nil)"), ifm
->ifam_addrs
);
575 /* Assert read up end point matches to end point */
577 zlog_warn ("ifam_read() does't read all socket data");
580 /* Interface's address information get. */
582 ifam_read (struct ifa_msghdr
*ifam
)
584 struct interface
*ifp
= NULL
;
585 union sockunion addr
, mask
, brd
;
586 char ifname
[INTERFACE_NAMSIZ
];
591 ifname
[0] = ifname
[INTERFACE_NAMSIZ
- 1] = '\0';
593 /* Allocate and read address information. */
594 ifam_read_mesg (ifam
, &addr
, &mask
, &brd
, ifname
, &ifnlen
);
596 if ((ifp
= if_lookup_by_index(ifam
->ifam_index
)) == NULL
)
598 zlog_warn ("%s: no interface for ifname %s, index %d",
599 __func__
, ifname
, ifam
->ifam_index
);
603 if (ifnlen
&& strncmp (ifp
->name
, ifname
, INTERFACE_NAMSIZ
))
606 /* N.B. The info in ifa_msghdr does not tell us whether the RTA_BRD
607 field contains a broadcast address or a peer address, so we are forced to
608 rely upon the interface type. */
609 if (if_is_pointopoint(ifp
))
610 SET_FLAG(flags
, ZEBRA_IFA_PEER
);
613 /* it might seem cute to grab the interface metric here, however
614 * we're processing an address update message, and so some systems
615 * (e.g. FBSD) dont bother to fill in ifam_metric. Disabled, but left
616 * in deliberately, as comment.
618 ifp
->metric
= ifam
->ifam_metric
;
621 /* Add connected address. */
622 switch (sockunion_family (&addr
))
625 if (ifam
->ifam_type
== RTM_NEWADDR
)
626 connected_add_ipv4 (ifp
, flags
, &addr
.sin
.sin_addr
,
627 ip_masklen (mask
.sin
.sin_addr
),
629 (isalias
? ifname
: NULL
));
631 connected_delete_ipv4 (ifp
, flags
, &addr
.sin
.sin_addr
,
632 ip_masklen (mask
.sin
.sin_addr
),
637 /* Unset interface index from link-local address when IPv6 stack
639 if (IN6_IS_ADDR_LINKLOCAL (&addr
.sin6
.sin6_addr
))
640 SET_IN6_LINKLOCAL_IFINDEX (addr
.sin6
.sin6_addr
, 0);
642 if (ifam
->ifam_type
== RTM_NEWADDR
)
643 connected_add_ipv6 (ifp
, flags
, &addr
.sin6
.sin6_addr
,
644 ip6_masklen (mask
.sin6
.sin6_addr
),
646 (isalias
? ifname
: NULL
));
648 connected_delete_ipv6 (ifp
,
649 &addr
.sin6
.sin6_addr
,
650 ip6_masklen (mask
.sin6
.sin6_addr
),
651 &brd
.sin6
.sin6_addr
);
653 #endif /* HAVE_IPV6 */
655 /* Unsupported family silently ignore... */
659 /* Check interface flag for implicit up of the interface. */
663 /* In addition to lacking IFANNOUNCE, on SUNOS IFF_UP is strange.
664 * See comments for SUNOS_5 in interface.c::if_flags_mangle.
666 * Here we take care of case where the real IFF_UP was previously
667 * unset (as kept in struct zebra_if.primary_state) and the mangled
668 * IFF_UP (ie IFF_UP set || listcount(connected) has now transitioned
669 * to unset due to the lost non-primary address having DELADDR'd.
671 * we must delete the interface, because in between here and next
672 * event for this interface-name the administrator could unplumb
673 * and replumb the interface.
676 if_delete_update (ifp
);
682 /* Interface function for reading kernel routing table information. */
684 rtm_read_mesg (struct rt_msghdr
*rtm
,
685 union sockunion
*dest
,
686 union sockunion
*mask
,
687 union sockunion
*gate
,
693 /* Pnt points out socket data start point. */
694 pnt
= (caddr_t
)(rtm
+ 1);
695 end
= ((caddr_t
)rtm
) + rtm
->rtm_msglen
;
697 /* rt_msghdr version check. */
698 if (rtm
->rtm_version
!= RTM_VERSION
)
699 zlog (NULL
, LOG_WARNING
,
700 "Routing message version different %d should be %d."
701 "This may cause problem\n", rtm
->rtm_version
, RTM_VERSION
);
703 /* Be sure structure is cleared */
704 memset (dest
, 0, sizeof (union sockunion
));
705 memset (gate
, 0, sizeof (union sockunion
));
706 memset (mask
, 0, sizeof (union sockunion
));
708 /* We fetch each socket variable into sockunion. */
709 RTA_ADDR_GET (dest
, RTA_DST
, rtm
->rtm_addrs
, pnt
);
710 RTA_ADDR_GET (gate
, RTA_GATEWAY
, rtm
->rtm_addrs
, pnt
);
711 RTA_ATTR_GET (mask
, RTA_NETMASK
, rtm
->rtm_addrs
, pnt
);
712 RTA_ADDR_GET (NULL
, RTA_GENMASK
, rtm
->rtm_addrs
, pnt
);
713 RTA_NAME_GET (ifname
, RTA_IFP
, rtm
->rtm_addrs
, pnt
, *ifnlen
);
714 RTA_ADDR_GET (NULL
, RTA_IFA
, rtm
->rtm_addrs
, pnt
);
715 RTA_ADDR_GET (NULL
, RTA_AUTHOR
, rtm
->rtm_addrs
, pnt
);
716 RTA_ADDR_GET (NULL
, RTA_BRD
, rtm
->rtm_addrs
, pnt
);
718 /* If there is netmask information set it's family same as
720 if (rtm
->rtm_addrs
& RTA_NETMASK
)
721 mask
->sa
.sa_family
= dest
->sa
.sa_family
;
723 /* Assert read up to the end of pointer. */
725 zlog (NULL
, LOG_WARNING
, "rtm_read() does't read all socket data.");
727 return rtm
->rtm_flags
;
731 rtm_read (struct rt_msghdr
*rtm
)
735 union sockunion dest
, mask
, gate
;
736 char ifname
[INTERFACE_NAMSIZ
+ 1];
741 /* Read destination and netmask and gateway from rtm message
743 flags
= rtm_read_mesg (rtm
, &dest
, &mask
, &gate
, ifname
, &ifnlen
);
744 if (!(flags
& RTF_DONE
))
746 if (IS_ZEBRA_DEBUG_KERNEL
)
747 zlog_debug ("%s: got rtm of type %d (%s)", __func__
, rtm
->rtm_type
,
748 LOOKUP (rtm_type_str
, rtm
->rtm_type
));
750 #ifdef RTF_CLONED /*bsdi, netbsd 1.6*/
751 if (flags
& RTF_CLONED
)
754 #ifdef RTF_WASCLONED /*freebsd*/
755 if (flags
& RTF_WASCLONED
)
759 if ((rtm
->rtm_type
== RTM_ADD
) && ! (flags
& RTF_UP
))
762 /* This is connected route. */
763 if (! (flags
& RTF_GATEWAY
))
766 if (flags
& RTF_PROTO1
)
767 SET_FLAG (zebra_flags
, ZEBRA_FLAG_SELFROUTE
);
769 /* This is persistent route. */
770 if (flags
& RTF_STATIC
)
771 SET_FLAG (zebra_flags
, ZEBRA_FLAG_STATIC
);
773 /* This is a reject or blackhole route */
774 if (flags
& RTF_REJECT
)
775 SET_FLAG (zebra_flags
, ZEBRA_FLAG_REJECT
);
776 if (flags
& RTF_BLACKHOLE
)
777 SET_FLAG (zebra_flags
, ZEBRA_FLAG_BLACKHOLE
);
779 if (dest
.sa
.sa_family
== AF_INET
)
781 struct prefix_ipv4 p
;
784 p
.prefix
= dest
.sin
.sin_addr
;
785 if (flags
& RTF_HOST
)
786 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
788 p
.prefixlen
= ip_masklen (mask
.sin
.sin_addr
);
790 /* Catch self originated messages and match them against our current RIB.
791 * At the same time, ignore unconfirmed messages, they should be tracked
792 * by rtm_write() and kernel_rtm_ipv4().
794 if (rtm
->rtm_type
!= RTM_GET
795 && (rtm
->rtm_pid
== pid
|| rtm
->rtm_pid
== old_pid
))
797 char buf
[INET_ADDRSTRLEN
], gate_buf
[INET_ADDRSTRLEN
];
799 if (! IS_ZEBRA_DEBUG_RIB
)
801 ret
= rib_lookup_ipv4_route (&p
, &gate
);
802 inet_ntop (AF_INET
, &p
.prefix
, buf
, INET_ADDRSTRLEN
);
803 switch (rtm
->rtm_type
)
808 /* The kernel notifies us about a new route in FIB created by us.
809 Do we have a correspondent entry in our RIB? */
812 case ZEBRA_RIB_NOTFOUND
:
813 zlog_debug ("%s: %s %s/%d: desync: RR isn't yet in RIB, while already in FIB",
814 __func__
, LOOKUP (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
);
816 case ZEBRA_RIB_FOUND_CONNECTED
:
817 case ZEBRA_RIB_FOUND_NOGATE
:
818 inet_ntop (AF_INET
, &gate
.sin
.sin_addr
, gate_buf
, INET_ADDRSTRLEN
);
819 zlog_debug ("%s: %s %s/%d: desync: RR is in RIB, but gate differs (ours is %s)",
820 __func__
, LOOKUP (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
, gate_buf
);
822 case ZEBRA_RIB_FOUND_EXACT
: /* RIB RR == FIB RR */
823 zlog_debug ("%s: %s %s/%d: done Ok",
824 __func__
, LOOKUP (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
);
825 rib_lookup_and_dump (&p
);
831 /* The kernel notifies us about a route deleted by us. Do we still
832 have it in the RIB? Do we have anything instead? */
835 case ZEBRA_RIB_FOUND_EXACT
:
836 zlog_debug ("%s: %s %s/%d: desync: RR is still in RIB, while already not in FIB",
837 __func__
, LOOKUP (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
);
838 rib_lookup_and_dump (&p
);
840 case ZEBRA_RIB_FOUND_CONNECTED
:
841 case ZEBRA_RIB_FOUND_NOGATE
:
842 zlog_debug ("%s: %s %s/%d: desync: RR is still in RIB, plus gate differs",
843 __func__
, LOOKUP (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
);
844 rib_lookup_and_dump (&p
);
846 case ZEBRA_RIB_NOTFOUND
: /* RIB RR == FIB RR */
847 zlog_debug ("%s: %s %s/%d: done Ok",
848 __func__
, LOOKUP (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
);
849 rib_lookup_and_dump (&p
);
855 zlog_debug ("%s: %s/%d: warning: loopback RTM of type %s received",
856 __func__
, buf
, p
.prefixlen
, LOOKUP (rtm_type_str
, rtm
->rtm_type
));
861 /* Change, delete the old prefix, we have no further information
862 * to specify the route really
864 if (rtm
->rtm_type
== RTM_CHANGE
)
865 rib_delete_ipv4 (ZEBRA_ROUTE_KERNEL
, zebra_flags
, &p
,
868 if (rtm
->rtm_type
== RTM_GET
869 || rtm
->rtm_type
== RTM_ADD
870 || rtm
->rtm_type
== RTM_CHANGE
)
871 rib_add_ipv4 (ZEBRA_ROUTE_KERNEL
, zebra_flags
,
872 &p
, &gate
.sin
.sin_addr
, NULL
, 0, 0, 0, 0);
874 rib_delete_ipv4 (ZEBRA_ROUTE_KERNEL
, zebra_flags
,
875 &p
, &gate
.sin
.sin_addr
, 0, 0);
878 if (dest
.sa
.sa_family
== AF_INET6
)
880 struct prefix_ipv6 p
;
881 unsigned int ifindex
= 0;
884 p
.prefix
= dest
.sin6
.sin6_addr
;
885 if (flags
& RTF_HOST
)
886 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
888 p
.prefixlen
= ip6_masklen (mask
.sin6
.sin6_addr
);
891 if (IN6_IS_ADDR_LINKLOCAL (&gate
.sin6
.sin6_addr
))
893 ifindex
= IN6_LINKLOCAL_IFINDEX (gate
.sin6
.sin6_addr
);
894 SET_IN6_LINKLOCAL_IFINDEX (gate
.sin6
.sin6_addr
, 0);
898 /* CHANGE: delete the old prefix, we have no further information
899 * to specify the route really
901 if (rtm
->rtm_type
== RTM_CHANGE
)
902 rib_delete_ipv6 (ZEBRA_ROUTE_KERNEL
, zebra_flags
, &p
,
905 if (rtm
->rtm_type
== RTM_GET
906 || rtm
->rtm_type
== RTM_ADD
907 || rtm
->rtm_type
== RTM_CHANGE
)
908 rib_add_ipv6 (ZEBRA_ROUTE_KERNEL
, zebra_flags
,
909 &p
, &gate
.sin6
.sin6_addr
, ifindex
, 0, 0, 0);
911 rib_delete_ipv6 (ZEBRA_ROUTE_KERNEL
, zebra_flags
,
912 &p
, &gate
.sin6
.sin6_addr
, ifindex
, 0);
914 #endif /* HAVE_IPV6 */
917 /* Interface function for the kernel routing table updates. Support
918 * for RTM_CHANGE will be needed.
919 * Exported only for rt_socket.c
922 rtm_write (int message
,
923 union sockunion
*dest
,
924 union sockunion
*mask
,
925 union sockunion
*gate
,
932 struct interface
*ifp
;
934 /* Sequencial number of routing message. */
935 static int msg_seq
= 0;
937 /* Struct of rt_msghdr and buffer for storing socket's data. */
940 struct rt_msghdr rtm
;
944 if (routing_sock
< 0)
945 return ZEBRA_ERR_EPERM
;
947 /* Clear and set rt_msghdr values */
948 memset (&msg
, 0, sizeof (struct rt_msghdr
));
949 msg
.rtm
.rtm_version
= RTM_VERSION
;
950 msg
.rtm
.rtm_type
= message
;
951 msg
.rtm
.rtm_seq
= msg_seq
++;
952 msg
.rtm
.rtm_addrs
= RTA_DST
;
953 msg
.rtm
.rtm_addrs
|= RTA_GATEWAY
;
954 msg
.rtm
.rtm_flags
= RTF_UP
;
955 msg
.rtm
.rtm_index
= index
;
959 msg
.rtm
.rtm_rmx
.rmx_hopcount
= metric
;
960 msg
.rtm
.rtm_inits
|= RTV_HOPCOUNT
;
963 ifp
= if_lookup_by_index (index
);
965 if (gate
&& message
== RTM_ADD
)
966 msg
.rtm
.rtm_flags
|= RTF_GATEWAY
;
968 if (! gate
&& message
== RTM_ADD
&& ifp
&&
969 (ifp
->flags
& IFF_POINTOPOINT
) == 0)
970 msg
.rtm
.rtm_flags
|= RTF_CLONING
;
972 /* If no protocol specific gateway is specified, use link
973 address for gateway. */
978 char dest_buf
[INET_ADDRSTRLEN
] = "NULL", mask_buf
[INET_ADDRSTRLEN
] = "255.255.255.255";
980 inet_ntop (AF_INET
, &dest
->sin
.sin_addr
, dest_buf
, INET_ADDRSTRLEN
);
982 inet_ntop (AF_INET
, &mask
->sin
.sin_addr
, mask_buf
, INET_ADDRSTRLEN
);
983 zlog_warn ("%s: %s/%s: gate == NULL and no gateway found for ifindex %d",
984 __func__
, dest_buf
, mask_buf
, index
);
987 gate
= (union sockunion
*) & ifp
->sdl
;
991 msg
.rtm
.rtm_addrs
|= RTA_NETMASK
;
992 else if (message
== RTM_ADD
)
993 msg
.rtm
.rtm_flags
|= RTF_HOST
;
995 /* Tagging route with flags */
996 msg
.rtm
.rtm_flags
|= (RTF_PROTO1
);
998 /* Additional flags. */
999 if (zebra_flags
& ZEBRA_FLAG_BLACKHOLE
)
1000 msg
.rtm
.rtm_flags
|= RTF_BLACKHOLE
;
1001 if (zebra_flags
& ZEBRA_FLAG_REJECT
)
1002 msg
.rtm
.rtm_flags
|= RTF_REJECT
;
1005 #ifdef HAVE_STRUCT_SOCKADDR_IN_SIN_LEN
1006 #define SOCKADDRSET(X,R) \
1007 if (msg.rtm.rtm_addrs & (R)) \
1009 int len = ROUNDUP ((X)->sa.sa_len); \
1010 memcpy (pnt, (caddr_t)(X), len); \
1014 #define SOCKADDRSET(X,R) \
1015 if (msg.rtm.rtm_addrs & (R)) \
1017 int len = SAROUNDUP (X); \
1018 memcpy (pnt, (caddr_t)(X), len); \
1021 #endif /* HAVE_STRUCT_SOCKADDR_IN_SIN_LEN */
1023 pnt
= (caddr_t
) msg
.buf
;
1025 /* Write each socket data into rtm message buffer */
1026 SOCKADDRSET (dest
, RTA_DST
);
1027 SOCKADDRSET (gate
, RTA_GATEWAY
);
1028 SOCKADDRSET (mask
, RTA_NETMASK
);
1030 msg
.rtm
.rtm_msglen
= pnt
- (caddr_t
) &msg
;
1032 ret
= write (routing_sock
, &msg
, msg
.rtm
.rtm_msglen
);
1034 if (ret
!= msg
.rtm
.rtm_msglen
)
1036 if (errno
== EEXIST
)
1037 return ZEBRA_ERR_RTEXIST
;
1038 if (errno
== ENETUNREACH
)
1039 return ZEBRA_ERR_RTUNREACH
;
1041 return ZEBRA_ERR_RTNOEXIST
;
1043 zlog_warn ("%s: write : %s (%d)", __func__
, safe_strerror (errno
), errno
);
1044 return ZEBRA_ERR_KERNEL
;
1046 return ZEBRA_ERR_NOERROR
;
1051 #include "zebra/zserv.h"
1053 /* For debug purpose. */
1055 rtmsg_debug (struct rt_msghdr
*rtm
)
1057 zlog_debug ("Kernel: Len: %d Type: %s", rtm
->rtm_msglen
, LOOKUP (rtm_type_str
, rtm
->rtm_type
));
1058 rtm_flag_dump (rtm
->rtm_flags
);
1059 zlog_debug ("Kernel: message seq %d", rtm
->rtm_seq
);
1060 zlog_debug ("Kernel: pid %d, rtm_addrs 0x%x", rtm
->rtm_pid
, rtm
->rtm_addrs
);
1063 /* This is pretty gross, better suggestions welcome -- mhandler */
1066 #define RTAX_MAX RTA_NUMBITS
1069 #endif /* RTA_NUMBITS */
1070 #endif /* RTAX_MAX */
1072 /* Kernel routing table and interface updates via routing socket. */
1074 kernel_read (struct thread
*thread
)
1078 struct rt_msghdr
*rtm
;
1081 * This must be big enough for any message the kernel might send.
1082 * Rather than determining how many sockaddrs of what size might be
1083 * in each particular message, just use RTAX_MAX of sockaddr_storage
1084 * for each. Note that the sockaddrs must be after each message
1085 * definition, or rather after whichever happens to be the largest,
1086 * since the buffer needs to be big enough for a message and the
1087 * sockaddrs together.
1091 /* Routing information. */
1094 struct rt_msghdr rtm
;
1095 struct sockaddr_storage addr
[RTAX_MAX
];
1098 /* Interface information. */
1101 struct if_msghdr ifm
;
1102 struct sockaddr_storage addr
[RTAX_MAX
];
1105 /* Interface address information. */
1108 struct ifa_msghdr ifa
;
1109 struct sockaddr_storage addr
[RTAX_MAX
];
1112 #ifdef RTM_IFANNOUNCE
1113 /* Interface arrival/departure */
1116 struct if_announcemsghdr ifan
;
1117 struct sockaddr_storage addr
[RTAX_MAX
];
1119 #endif /* RTM_IFANNOUNCE */
1123 /* Fetch routing socket. */
1124 sock
= THREAD_FD (thread
);
1126 nbytes
= read (sock
, &buf
, sizeof buf
);
1130 if (nbytes
< 0 && errno
!= EWOULDBLOCK
&& errno
!= EAGAIN
)
1131 zlog_warn ("routing socket error: %s", safe_strerror (errno
));
1135 thread_add_read (zebrad
.master
, kernel_read
, NULL
, sock
);
1137 if (IS_ZEBRA_DEBUG_KERNEL
)
1138 rtmsg_debug (&buf
.r
.rtm
);
1143 * Ensure that we didn't drop any data, so that processing routines
1144 * can assume they have the whole message.
1146 if (rtm
->rtm_msglen
!= nbytes
)
1148 zlog_warn ("kernel_read: rtm->rtm_msglen %d, nbytes %d, type %d\n",
1149 rtm
->rtm_msglen
, nbytes
, rtm
->rtm_type
);
1153 switch (rtm
->rtm_type
)
1161 ifm_read (&buf
.im
.ifm
);
1165 ifam_read (&buf
.ia
.ifa
);
1167 #ifdef RTM_IFANNOUNCE
1168 case RTM_IFANNOUNCE
:
1169 ifan_read (&buf
.ian
.ifan
);
1171 #endif /* RTM_IFANNOUNCE */
1173 if (IS_ZEBRA_DEBUG_KERNEL
)
1174 zlog_debug("Unprocessed RTM_type: %d", rtm
->rtm_type
);
1180 /* Make routing socket. */
1182 routing_socket (void)
1184 if ( zserv_privs
.change (ZPRIVS_RAISE
) )
1185 zlog_err ("routing_socket: Can't raise privileges");
1187 routing_sock
= socket (AF_ROUTE
, SOCK_RAW
, 0);
1189 if (routing_sock
< 0)
1191 if ( zserv_privs
.change (ZPRIVS_LOWER
) )
1192 zlog_err ("routing_socket: Can't lower privileges");
1193 zlog_warn ("Can't init kernel routing socket");
1197 /* XXX: Socket should be NONBLOCK, however as we currently
1198 * discard failed writes, this will lead to inconsistencies.
1199 * For now, socket must be blocking.
1201 /*if (fcntl (routing_sock, F_SETFL, O_NONBLOCK) < 0)
1202 zlog_warn ("Can't set O_NONBLOCK to routing socket");*/
1204 if ( zserv_privs
.change (ZPRIVS_LOWER
) )
1205 zlog_err ("routing_socket: Can't lower privileges");
1207 /* kernel_read needs rewrite. */
1208 thread_add_read (zebrad
.master
, kernel_read
, NULL
, routing_sock
);
1211 /* Exported interface function. This function simply calls
1212 routing_socket (). */