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"
37 #include "zebra/interface.h"
38 #include "zebra/zserv.h"
39 #include "zebra/debug.h"
40 #include "zebra/kernel_socket.h"
41 #include "zebra/rib.h"
43 extern struct zebra_privs_t zserv_privs
;
46 * Historically, the BSD routing socket has aligned data following a
47 * struct sockaddr to sizeof(long), which was 4 bytes on some
48 * platforms, and 8 bytes on others. NetBSD 6 changed the routing
49 * socket to align to sizeof(uint64_t), which is 8 bytes. OS X
50 * appears to align to sizeof(int), which is 4 bytes.
52 * Alignment of zero-sized sockaddrs is nonsensical, but historically
53 * BSD defines RT_ROUNDUP(0) to be the alignment interval (rather than
54 * 0). We follow this practice without questioning it, but it is a
55 * bug if quagga calls ROUNDUP with 0.
59 * Because of these varying conventions, the only sane approach is for
60 * the <net/route.h> header to define some flavor of ROUNDUP macro.
64 /* SAROUNDUP is the only thing we need, and SA_SIZE provides that */
65 #define SAROUNDUP(a) SA_SIZE(a)
68 #if defined(RT_ROUNDUP)
69 #define ROUNDUP(a) RT_ROUNDUP(a)
70 #endif /* defined(RT_ROUNDUP) */
73 /* Solaris has struct sockaddr_in[6] definitions at 16 / 32 bytes size,
74 * so the whole concept doesn't really apply. */
75 #define ROUNDUP(a) (a)
79 * If ROUNDUP has not yet been defined in terms of platform-provided
80 * defines, attempt to cope with heuristics.
85 * It's a bug for a platform not to define rounding/alignment for
86 * sockaddrs on the routing socket. This warning really is
87 * intentional, to provoke filing bug reports with operating systems
88 * that don't define RT_ROUNDUP or equivalent.
90 #warning "net/route.h does not define RT_ROUNDUP; making unwarranted assumptions!"
92 /* OS X (Xcode as of 2014-12) is known not to define RT_ROUNDUP */
94 #define ROUNDUP_TYPE int
96 #define ROUNDUP_TYPE long
100 ((a) > 0 ? (1 + (((a) - 1) | (sizeof(ROUNDUP_TYPE) - 1))) : sizeof(ROUNDUP_TYPE))
102 #endif /* defined(ROUNDUP) */
105 * Given a pointer (sockaddr or void *), return the number of bytes
106 * taken up by the sockaddr and any padding needed for alignment.
108 #if defined(HAVE_STRUCT_SOCKADDR_SA_LEN)
109 #define SAROUNDUP(X) ROUNDUP(((struct sockaddr *)(X))->sa_len)
110 #elif defined(HAVE_IPV6)
112 * One would hope all fixed-size structure definitions are aligned,
113 * but round them up nonetheless.
115 #define SAROUNDUP(X) \
116 (((struct sockaddr *)(X))->sa_family == AF_INET ? \
117 ROUNDUP(sizeof(struct sockaddr_in)):\
118 (((struct sockaddr *)(X))->sa_family == AF_INET6 ? \
119 ROUNDUP(sizeof(struct sockaddr_in6)) : \
120 (((struct sockaddr *)(X))->sa_family == AF_LINK ? \
121 ROUNDUP(sizeof(struct sockaddr_dl)) : sizeof(struct sockaddr))))
122 #else /* HAVE_IPV6 */
123 #define SAROUNDUP(X) \
124 (((struct sockaddr *)(X))->sa_family == AF_INET ? \
125 ROUNDUP(sizeof(struct sockaddr_in)):\
126 (((struct sockaddr *)(X))->sa_family == AF_LINK ? \
127 ROUNDUP(sizeof(struct sockaddr_dl)) : sizeof(struct sockaddr)))
128 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
130 #endif /* !SA_SIZE */
133 * We use a call to an inline function to copy (PNT) to (DEST)
134 * 1. Calculating the length of the copy requires an #ifdef to determine
135 * if sa_len is a field and can't be used directly inside a #define
136 * 2. So the compiler doesn't complain when DEST is NULL, which is only true
137 * when we are skipping the copy and incrementing to the next SA
140 rta_copy (union sockunion
*dest
, caddr_t src
) {
144 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
145 len
= (((struct sockaddr
*)src
)->sa_len
> sizeof (*dest
)) ?
146 sizeof (*dest
) : ((struct sockaddr
*)src
)->sa_len
;
148 len
= (SAROUNDUP (src
) > sizeof (*dest
)) ?
149 sizeof (*dest
) : SAROUNDUP (src
) ;
151 memcpy (dest
, src
, len
);
154 #define RTA_ADDR_GET(DEST, RTA, RTMADDRS, PNT) \
155 if ((RTMADDRS) & (RTA)) \
157 int len = SAROUNDUP ((PNT)); \
158 if (af_check (((struct sockaddr *)(PNT))->sa_family)) \
159 rta_copy((DEST), (PNT)); \
162 #define RTA_ATTR_GET(DEST, RTA, RTMADDRS, PNT) \
163 if ((RTMADDRS) & (RTA)) \
165 int len = SAROUNDUP ((PNT)); \
166 rta_copy((DEST), (PNT)); \
170 #define RTA_NAME_GET(DEST, RTA, RTMADDRS, PNT, LEN) \
171 if ((RTMADDRS) & (RTA)) \
173 u_char *pdest = (u_char *) (DEST); \
174 int len = SAROUNDUP ((PNT)); \
175 struct sockaddr_dl *sdl = (struct sockaddr_dl *)(PNT); \
176 if (IS_ZEBRA_DEBUG_KERNEL) \
177 zlog_debug ("%s: RTA_SDL_GET nlen %d, alen %d", \
178 __func__, sdl->sdl_nlen, sdl->sdl_alen); \
179 if ( ((DEST) != NULL) && (sdl->sdl_family == AF_LINK) \
180 && (sdl->sdl_nlen < IFNAMSIZ) && (sdl->sdl_nlen <= len) ) \
182 memcpy (pdest, sdl->sdl_data, sdl->sdl_nlen); \
183 pdest[sdl->sdl_nlen] = '\0'; \
184 (LEN) = sdl->sdl_nlen; \
192 /* Routing socket message types. */
193 const struct message rtm_type_str
[] =
195 {RTM_ADD
, "RTM_ADD"},
196 {RTM_DELETE
, "RTM_DELETE"},
197 {RTM_CHANGE
, "RTM_CHANGE"},
198 {RTM_GET
, "RTM_GET"},
199 {RTM_LOSING
, "RTM_LOSING"},
200 {RTM_REDIRECT
, "RTM_REDIRECT"},
201 {RTM_MISS
, "RTM_MISS"},
202 {RTM_LOCK
, "RTM_LOCK"},
204 {RTM_OLDADD
, "RTM_OLDADD"},
205 #endif /* RTM_OLDADD */
207 {RTM_OLDDEL
, "RTM_OLDDEL"},
208 #endif /* RTM_OLDDEL */
209 {RTM_RESOLVE
, "RTM_RESOLVE"},
210 {RTM_NEWADDR
, "RTM_NEWADDR"},
211 {RTM_DELADDR
, "RTM_DELADDR"},
212 {RTM_IFINFO
, "RTM_IFINFO"},
214 {RTM_OIFINFO
, "RTM_OIFINFO"},
215 #endif /* RTM_OIFINFO */
217 {RTM_NEWMADDR
, "RTM_NEWMADDR"},
218 #endif /* RTM_NEWMADDR */
220 {RTM_DELMADDR
, "RTM_DELMADDR"},
221 #endif /* RTM_DELMADDR */
222 #ifdef RTM_IFANNOUNCE
223 {RTM_IFANNOUNCE
, "RTM_IFANNOUNCE"},
224 #endif /* RTM_IFANNOUNCE */
228 static const struct message rtm_flag_str
[] =
231 {RTF_GATEWAY
, "GATEWAY"},
233 {RTF_REJECT
, "REJECT"},
234 {RTF_DYNAMIC
, "DYNAMIC"},
235 {RTF_MODIFIED
, "MODIFIED"},
239 #endif /* RTF_MASK */
241 {RTF_CLONING
, "CLONING"},
242 #endif /* RTF_CLONING */
243 {RTF_XRESOLVE
, "XRESOLVE"},
245 {RTF_LLINFO
, "LLINFO"},
246 #endif /* RTF_LLINFO */
247 {RTF_STATIC
, "STATIC"},
248 {RTF_BLACKHOLE
, "BLACKHOLE"},
250 {RTF_PRIVATE
, "PRIVATE"},
251 #endif /* RTF_PRIVATE */
252 {RTF_PROTO1
, "PROTO1"},
253 {RTF_PROTO2
, "PROTO2"},
255 {RTF_PRCLONING
, "PRCLONING"},
256 #endif /* RTF_PRCLONING */
258 {RTF_WASCLONED
, "WASCLONED"},
259 #endif /* RTF_WASCLONED */
261 {RTF_PROTO3
, "PROTO3"},
262 #endif /* RTF_PROTO3 */
264 {RTF_PINNED
, "PINNED"},
265 #endif /* RTF_PINNED */
267 {RTF_LOCAL
, "LOCAL"},
268 #endif /* RTF_LOCAL */
270 {RTF_BROADCAST
, "BROADCAST"},
271 #endif /* RTF_BROADCAST */
273 {RTF_MULTICAST
, "MULTICAST"},
274 #endif /* RTF_MULTICAST */
276 {RTF_MULTIRT
, "MULTIRT"},
277 #endif /* RTF_MULTIRT */
279 {RTF_SETSRC
, "SETSRC"},
280 #endif /* RTF_SETSRC */
284 /* Kernel routing update socket. */
285 int routing_sock
= -1;
287 /* Yes I'm checking ugly routing socket behavior. */
290 /* Supported address family check. */
292 af_check (int family
)
294 if (family
== AF_INET
)
297 if (family
== AF_INET6
)
299 #endif /* HAVE_IPV6 */
303 /* Dump routing table flag for debug purpose. */
305 rtm_flag_dump (int flag
)
307 const struct message
*mes
;
308 static char buf
[BUFSIZ
];
311 for (mes
= rtm_flag_str
; mes
->key
!= 0; mes
++)
315 strlcat (buf
, mes
->str
, BUFSIZ
);
316 strlcat (buf
, " ", BUFSIZ
);
319 zlog_debug ("Kernel: %s", buf
);
322 #ifdef RTM_IFANNOUNCE
323 /* Interface adding function */
325 ifan_read (struct if_announcemsghdr
*ifan
)
327 struct interface
*ifp
;
329 ifp
= if_lookup_by_index (ifan
->ifan_index
);
332 assert ( (ifp
->ifindex
== ifan
->ifan_index
)
333 || (ifp
->ifindex
== IFINDEX_INTERNAL
) );
336 || ((ifp
->ifindex
== IFINDEX_INTERNAL
)
337 && (ifan
->ifan_what
== IFAN_ARRIVAL
)) )
339 if (IS_ZEBRA_DEBUG_KERNEL
)
340 zlog_debug ("%s: creating interface for ifindex %d, name %s",
341 __func__
, ifan
->ifan_index
, ifan
->ifan_name
);
343 /* Create Interface */
344 ifp
= if_get_by_name_len(ifan
->ifan_name
,
345 strnlen(ifan
->ifan_name
,
346 sizeof(ifan
->ifan_name
)));
347 ifp
->ifindex
= ifan
->ifan_index
;
352 else if (ifp
!= NULL
&& ifan
->ifan_what
== IFAN_DEPARTURE
)
353 if_delete_update (ifp
);
359 if (IS_ZEBRA_DEBUG_KERNEL
)
360 zlog_debug ("%s: interface %s index %d",
361 __func__
, ifan
->ifan_name
, ifan
->ifan_index
);
365 #endif /* RTM_IFANNOUNCE */
367 #ifdef HAVE_BSD_IFI_LINK_STATE
368 /* BSD link detect translation */
370 bsd_linkdetect_translate (struct if_msghdr
*ifm
)
372 if ((ifm
->ifm_data
.ifi_link_state
>= LINK_STATE_UP
) ||
373 (ifm
->ifm_data
.ifi_link_state
== LINK_STATE_UNKNOWN
))
374 SET_FLAG(ifm
->ifm_flags
, IFF_RUNNING
);
376 UNSET_FLAG(ifm
->ifm_flags
, IFF_RUNNING
);
378 #endif /* HAVE_BSD_IFI_LINK_STATE */
381 * Handle struct if_msghdr obtained from reading routing socket or
382 * sysctl (from interface_list). There may or may not be sockaddrs
383 * present after the header.
386 ifm_read (struct if_msghdr
*ifm
)
388 struct interface
*ifp
= NULL
;
389 struct sockaddr_dl
*sdl
;
390 char ifname
[IFNAMSIZ
];
394 /* terminate ifname at head (for strnlen) and tail (for safety) */
395 ifname
[IFNAMSIZ
- 1] = '\0';
397 /* paranoia: sanity check structure */
398 if (ifm
->ifm_msglen
< sizeof(struct if_msghdr
))
400 zlog_err ("ifm_read: ifm->ifm_msglen %d too short\n",
406 * Check for a sockaddr_dl following the message. First, point to
407 * where a socakddr might be if one follows the message.
409 cp
= (void *)(ifm
+ 1);
413 * XXX This behavior should be narrowed to only the kernel versions
414 * for which the structures returned do not match the headers.
416 * if_msghdr_t on 64 bit kernels in Solaris 9 and earlier versions
417 * is 12 bytes larger than the 32 bit version.
419 if (((struct sockaddr
*) cp
)->sa_family
== AF_UNSPEC
)
423 RTA_ADDR_GET (NULL
, RTA_DST
, ifm
->ifm_addrs
, cp
);
424 RTA_ADDR_GET (NULL
, RTA_GATEWAY
, ifm
->ifm_addrs
, cp
);
425 RTA_ATTR_GET (NULL
, RTA_NETMASK
, ifm
->ifm_addrs
, cp
);
426 RTA_ADDR_GET (NULL
, RTA_GENMASK
, ifm
->ifm_addrs
, cp
);
427 sdl
= (struct sockaddr_dl
*)cp
;
428 RTA_NAME_GET (ifname
, RTA_IFP
, ifm
->ifm_addrs
, cp
, ifnlen
);
429 RTA_ADDR_GET (NULL
, RTA_IFA
, ifm
->ifm_addrs
, cp
);
430 RTA_ADDR_GET (NULL
, RTA_AUTHOR
, ifm
->ifm_addrs
, cp
);
431 RTA_ADDR_GET (NULL
, RTA_BRD
, ifm
->ifm_addrs
, cp
);
433 if (IS_ZEBRA_DEBUG_KERNEL
)
434 zlog_debug ("%s: sdl ifname %s", __func__
, (ifnlen
? ifname
: "(nil)"));
437 * Look up on ifindex first, because ifindices are the primary handle for
438 * interfaces across the user/kernel boundary, for most systems. (Some
439 * messages, such as up/down status changes on NetBSD, do not include a
442 if ( (ifp
= if_lookup_by_index (ifm
->ifm_index
)) != NULL
)
444 /* we have an ifp, verify that the name matches as some systems,
445 * eg Solaris, have a 1:many association of ifindex:ifname
446 * if they dont match, we dont have the correct ifp and should
447 * set it back to NULL to let next check do lookup by name
449 if (ifnlen
&& (strncmp (ifp
->name
, ifname
, IFNAMSIZ
) != 0) )
451 if (IS_ZEBRA_DEBUG_KERNEL
)
452 zlog_debug ("%s: ifp name %s doesnt match sdl name %s",
453 __func__
, ifp
->name
, ifname
);
459 * If we dont have an ifp, try looking up by name. Particularly as some
460 * systems (Solaris) have a 1:many mapping of ifindex:ifname - the ifname
461 * is therefore our unique handle to that interface.
463 * Interfaces specified in the configuration file for which the ifindex
464 * has not been determined will have ifindex == IFINDEX_INTERNAL, and such
465 * interfaces are found by this search, and then their ifindex values can
468 if ( (ifp
== NULL
) && ifnlen
)
469 ifp
= if_lookup_by_name (ifname
);
472 * If ifp still does not exist or has an invalid index (IFINDEX_INTERNAL),
473 * create or fill in an interface.
475 if ((ifp
== NULL
) || (ifp
->ifindex
== IFINDEX_INTERNAL
))
478 * To create or fill in an interface, a sockaddr_dl (via
479 * RTA_IFP) is required.
483 zlog_warn ("Interface index %d (new) missing ifname\n",
488 #ifndef RTM_IFANNOUNCE
489 /* Down->Down interface should be ignored here.
490 * See further comment below.
492 if (!CHECK_FLAG (ifm
->ifm_flags
, IFF_UP
))
494 #endif /* !RTM_IFANNOUNCE */
498 /* Interface that zebra was not previously aware of, so create. */
499 ifp
= if_create (ifname
, ifnlen
);
500 if (IS_ZEBRA_DEBUG_KERNEL
)
501 zlog_debug ("%s: creating ifp for ifindex %d",
502 __func__
, ifm
->ifm_index
);
505 if (IS_ZEBRA_DEBUG_KERNEL
)
506 zlog_debug ("%s: updated/created ifp, ifname %s, ifindex %d",
507 __func__
, ifp
->name
, ifp
->ifindex
);
509 * Fill in newly created interface structure, or larval
510 * structure with ifindex IFINDEX_INTERNAL.
512 ifp
->ifindex
= ifm
->ifm_index
;
514 #ifdef HAVE_BSD_IFI_LINK_STATE /* translate BSD kernel msg for link-state */
515 bsd_linkdetect_translate(ifm
);
516 #endif /* HAVE_BSD_IFI_LINK_STATE */
518 if_flags_update (ifp
, ifm
->ifm_flags
);
519 #if defined(__bsdi__)
520 if_kvm_get_mtu (ifp
);
523 #endif /* __bsdi__ */
527 * XXX sockaddr_dl contents can be larger than the structure
528 * definition. There are 2 big families here:
529 * - BSD has sdl_len + sdl_data[16] + overruns sdl_data
530 * we MUST use sdl_len here or we'll truncate data.
531 * - Solaris has no sdl_len, but sdl_data[244]
532 * presumably, it's not going to run past that, so sizeof()
534 * a nonzero ifnlen from RTA_NAME_GET() means sdl is valid
538 #ifdef HAVE_STRUCT_SOCKADDR_DL_SDL_LEN
539 memcpy (&ifp
->sdl
, sdl
, sdl
->sdl_len
);
541 memcpy (&ifp
->sdl
, sdl
, sizeof (struct sockaddr_dl
));
542 #endif /* HAVE_STRUCT_SOCKADDR_DL_SDL_LEN */
549 * Interface structure exists. Adjust stored flags from
550 * notification. If interface has up->down or down->up
551 * transition, call state change routines (to adjust routes,
552 * notify routing daemons, etc.). (Other flag changes are stored
553 * but apparently do not trigger action.)
556 if (ifp
->ifindex
!= ifm
->ifm_index
)
558 zlog_warn ("%s: index mismatch, ifname %s, ifp index %d, "
560 __func__
, ifp
->name
, ifp
->ifindex
, ifm
->ifm_index
);
564 #ifdef HAVE_BSD_IFI_LINK_STATE /* translate BSD kernel msg for link-state */
565 bsd_linkdetect_translate(ifm
);
566 #endif /* HAVE_BSD_IFI_LINK_STATE */
568 /* update flags and handle operative->inoperative transition, if any */
569 if_flags_update (ifp
, ifm
->ifm_flags
);
571 #ifndef RTM_IFANNOUNCE
574 /* No RTM_IFANNOUNCE on this platform, so we can never
575 * distinguish between ~IFF_UP and delete. We must presume
576 * it has been deleted.
577 * Eg, Solaris will not notify us of unplumb.
579 * XXX: Fixme - this should be runtime detected
580 * So that a binary compiled on a system with IFANNOUNCE
581 * will still behave correctly if run on a platform without
583 if_delete_update (ifp
);
585 #endif /* RTM_IFANNOUNCE */
588 #if defined(__bsdi__)
589 if_kvm_get_mtu (ifp
);
592 #endif /* __bsdi__ */
597 #ifdef HAVE_NET_RT_IFLIST
598 ifp
->stats
= ifm
->ifm_data
;
599 #endif /* HAVE_NET_RT_IFLIST */
601 if (IS_ZEBRA_DEBUG_KERNEL
)
602 zlog_debug ("%s: interface %s index %d",
603 __func__
, ifp
->name
, ifp
->ifindex
);
608 /* Address read from struct ifa_msghdr. */
610 ifam_read_mesg (struct ifa_msghdr
*ifm
,
611 union sockunion
*addr
,
612 union sockunion
*mask
,
613 union sockunion
*brd
,
619 union sockunion gateway
;
621 pnt
= (caddr_t
)(ifm
+ 1);
622 end
= ((caddr_t
)ifm
) + ifm
->ifam_msglen
;
624 /* Be sure structure is cleared */
625 memset (mask
, 0, sizeof (union sockunion
));
626 memset (addr
, 0, sizeof (union sockunion
));
627 memset (brd
, 0, sizeof (union sockunion
));
628 memset (&dst
, 0, sizeof (union sockunion
));
629 memset (&gateway
, 0, sizeof (union sockunion
));
631 /* We fetch each socket variable into sockunion. */
632 RTA_ADDR_GET (&dst
, RTA_DST
, ifm
->ifam_addrs
, pnt
);
633 RTA_ADDR_GET (&gateway
, RTA_GATEWAY
, ifm
->ifam_addrs
, pnt
);
634 RTA_ATTR_GET (mask
, RTA_NETMASK
, ifm
->ifam_addrs
, pnt
);
635 RTA_ADDR_GET (NULL
, RTA_GENMASK
, ifm
->ifam_addrs
, pnt
);
636 RTA_NAME_GET (ifname
, RTA_IFP
, ifm
->ifam_addrs
, pnt
, *ifnlen
);
637 RTA_ADDR_GET (addr
, RTA_IFA
, ifm
->ifam_addrs
, pnt
);
638 RTA_ADDR_GET (NULL
, RTA_AUTHOR
, ifm
->ifam_addrs
, pnt
);
639 RTA_ADDR_GET (brd
, RTA_BRD
, ifm
->ifam_addrs
, pnt
);
641 if (IS_ZEBRA_DEBUG_KERNEL
)
643 switch (sockunion_family(addr
))
647 char buf
[4][INET_ADDRSTRLEN
];
648 zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x, "
649 "ifam_flags 0x%x, addr %s/%d broad %s dst %s "
651 __func__
, ifm
->ifam_index
,
652 (ifnlen
? ifname
: "(nil)"), ifm
->ifam_addrs
,
654 inet_ntop(AF_INET
,&addr
->sin
.sin_addr
,
655 buf
[0],sizeof(buf
[0])),
656 ip_masklen(mask
->sin
.sin_addr
),
657 inet_ntop(AF_INET
,&brd
->sin
.sin_addr
,
658 buf
[1],sizeof(buf
[1])),
659 inet_ntop(AF_INET
,&dst
.sin
.sin_addr
,
660 buf
[2],sizeof(buf
[2])),
661 inet_ntop(AF_INET
,&gateway
.sin
.sin_addr
,
662 buf
[3],sizeof(buf
[3])));
668 char buf
[4][INET6_ADDRSTRLEN
];
669 zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x, "
670 "ifam_flags 0x%x, addr %s/%d broad %s dst %s "
672 __func__
, ifm
->ifam_index
,
673 (ifnlen
? ifname
: "(nil)"), ifm
->ifam_addrs
,
675 inet_ntop(AF_INET6
,&addr
->sin6
.sin6_addr
,
676 buf
[0],sizeof(buf
[0])),
677 ip6_masklen(mask
->sin6
.sin6_addr
),
678 inet_ntop(AF_INET6
,&brd
->sin6
.sin6_addr
,
679 buf
[1],sizeof(buf
[1])),
680 inet_ntop(AF_INET6
,&dst
.sin6
.sin6_addr
,
681 buf
[2],sizeof(buf
[2])),
682 inet_ntop(AF_INET6
,&gateway
.sin6
.sin6_addr
,
683 buf
[3],sizeof(buf
[3])));
686 #endif /* HAVE_IPV6 */
688 zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x",
689 __func__
, ifm
->ifam_index
,
690 (ifnlen
? ifname
: "(nil)"), ifm
->ifam_addrs
);
695 /* Assert read up end point matches to end point */
697 zlog_warn ("ifam_read() doesn't read all socket data");
700 /* Interface's address information get. */
702 ifam_read (struct ifa_msghdr
*ifam
)
704 struct interface
*ifp
= NULL
;
705 union sockunion addr
, mask
, brd
;
706 char ifname
[INTERFACE_NAMSIZ
];
711 ifname
[0] = ifname
[INTERFACE_NAMSIZ
- 1] = '\0';
713 /* Allocate and read address information. */
714 ifam_read_mesg (ifam
, &addr
, &mask
, &brd
, ifname
, &ifnlen
);
716 if ((ifp
= if_lookup_by_index(ifam
->ifam_index
)) == NULL
)
718 zlog_warn ("%s: no interface for ifname %s, index %d",
719 __func__
, ifname
, ifam
->ifam_index
);
723 if (ifnlen
&& strncmp (ifp
->name
, ifname
, INTERFACE_NAMSIZ
))
726 /* N.B. The info in ifa_msghdr does not tell us whether the RTA_BRD
727 field contains a broadcast address or a peer address, so we are forced to
728 rely upon the interface type. */
729 if (if_is_pointopoint(ifp
))
730 SET_FLAG(flags
, ZEBRA_IFA_PEER
);
733 /* it might seem cute to grab the interface metric here, however
734 * we're processing an address update message, and so some systems
735 * (e.g. FBSD) dont bother to fill in ifam_metric. Disabled, but left
736 * in deliberately, as comment.
738 ifp
->metric
= ifam
->ifam_metric
;
741 /* Add connected address. */
742 switch (sockunion_family (&addr
))
745 if (ifam
->ifam_type
== RTM_NEWADDR
)
746 connected_add_ipv4 (ifp
, flags
, &addr
.sin
.sin_addr
,
747 ip_masklen (mask
.sin
.sin_addr
),
749 (isalias
? ifname
: NULL
));
751 connected_delete_ipv4 (ifp
, flags
, &addr
.sin
.sin_addr
,
752 ip_masklen (mask
.sin
.sin_addr
),
757 /* Unset interface index from link-local address when IPv6 stack
759 if (IN6_IS_ADDR_LINKLOCAL (&addr
.sin6
.sin6_addr
))
761 SET_IN6_LINKLOCAL_IFINDEX (addr
.sin6
.sin6_addr
, 0);
764 if (ifam
->ifam_type
== RTM_NEWADDR
)
765 connected_add_ipv6 (ifp
, flags
, &addr
.sin6
.sin6_addr
,
766 ip6_masklen (mask
.sin6
.sin6_addr
),
768 (isalias
? ifname
: NULL
));
770 connected_delete_ipv6 (ifp
,
771 &addr
.sin6
.sin6_addr
,
772 ip6_masklen (mask
.sin6
.sin6_addr
),
773 &brd
.sin6
.sin6_addr
);
775 #endif /* HAVE_IPV6 */
777 /* Unsupported family silently ignore... */
781 /* Check interface flag for implicit up of the interface. */
785 /* In addition to lacking IFANNOUNCE, on SUNOS IFF_UP is strange.
786 * See comments for SUNOS_5 in interface.c::if_flags_mangle.
788 * Here we take care of case where the real IFF_UP was previously
789 * unset (as kept in struct zebra_if.primary_state) and the mangled
790 * IFF_UP (ie IFF_UP set || listcount(connected) has now transitioned
791 * to unset due to the lost non-primary address having DELADDR'd.
793 * we must delete the interface, because in between here and next
794 * event for this interface-name the administrator could unplumb
795 * and replumb the interface.
798 if_delete_update (ifp
);
804 /* Interface function for reading kernel routing table information. */
806 rtm_read_mesg (struct rt_msghdr
*rtm
,
807 union sockunion
*dest
,
808 union sockunion
*mask
,
809 union sockunion
*gate
,
815 /* Pnt points out socket data start point. */
816 pnt
= (caddr_t
)(rtm
+ 1);
817 end
= ((caddr_t
)rtm
) + rtm
->rtm_msglen
;
819 /* rt_msghdr version check. */
820 if (rtm
->rtm_version
!= RTM_VERSION
)
821 zlog (NULL
, LOG_WARNING
,
822 "Routing message version different %d should be %d."
823 "This may cause problem\n", rtm
->rtm_version
, RTM_VERSION
);
825 /* Be sure structure is cleared */
826 memset (dest
, 0, sizeof (union sockunion
));
827 memset (gate
, 0, sizeof (union sockunion
));
828 memset (mask
, 0, sizeof (union sockunion
));
830 /* We fetch each socket variable into sockunion. */
831 RTA_ADDR_GET (dest
, RTA_DST
, rtm
->rtm_addrs
, pnt
);
832 RTA_ADDR_GET (gate
, RTA_GATEWAY
, rtm
->rtm_addrs
, pnt
);
833 RTA_ATTR_GET (mask
, RTA_NETMASK
, rtm
->rtm_addrs
, pnt
);
834 RTA_ADDR_GET (NULL
, RTA_GENMASK
, rtm
->rtm_addrs
, pnt
);
835 RTA_NAME_GET (ifname
, RTA_IFP
, rtm
->rtm_addrs
, pnt
, *ifnlen
);
836 RTA_ADDR_GET (NULL
, RTA_IFA
, rtm
->rtm_addrs
, pnt
);
837 RTA_ADDR_GET (NULL
, RTA_AUTHOR
, rtm
->rtm_addrs
, pnt
);
838 RTA_ADDR_GET (NULL
, RTA_BRD
, rtm
->rtm_addrs
, pnt
);
840 /* If there is netmask information set it's family same as
842 if (rtm
->rtm_addrs
& RTA_NETMASK
)
843 mask
->sa
.sa_family
= dest
->sa
.sa_family
;
845 /* Assert read up to the end of pointer. */
847 zlog (NULL
, LOG_WARNING
, "rtm_read() doesn't read all socket data.");
849 return rtm
->rtm_flags
;
853 rtm_read (struct rt_msghdr
*rtm
)
857 union sockunion dest
, mask
, gate
;
858 char ifname
[INTERFACE_NAMSIZ
+ 1];
863 /* Read destination and netmask and gateway from rtm message
865 flags
= rtm_read_mesg (rtm
, &dest
, &mask
, &gate
, ifname
, &ifnlen
);
866 if (!(flags
& RTF_DONE
))
868 if (IS_ZEBRA_DEBUG_KERNEL
)
869 zlog_debug ("%s: got rtm of type %d (%s)", __func__
, rtm
->rtm_type
,
870 lookup (rtm_type_str
, rtm
->rtm_type
));
872 #ifdef RTF_CLONED /*bsdi, netbsd 1.6*/
873 if (flags
& RTF_CLONED
)
876 #ifdef RTF_WASCLONED /*freebsd*/
877 if (flags
& RTF_WASCLONED
)
881 if ((rtm
->rtm_type
== RTM_ADD
) && ! (flags
& RTF_UP
))
884 /* This is connected route. */
885 if (! (flags
& RTF_GATEWAY
))
888 if (flags
& RTF_PROTO1
)
889 SET_FLAG (zebra_flags
, ZEBRA_FLAG_SELFROUTE
);
891 /* This is persistent route. */
892 if (flags
& RTF_STATIC
)
893 SET_FLAG (zebra_flags
, ZEBRA_FLAG_STATIC
);
895 /* This is a reject or blackhole route */
896 if (flags
& RTF_REJECT
)
897 SET_FLAG (zebra_flags
, ZEBRA_FLAG_REJECT
);
898 if (flags
& RTF_BLACKHOLE
)
899 SET_FLAG (zebra_flags
, ZEBRA_FLAG_BLACKHOLE
);
901 if (dest
.sa
.sa_family
== AF_INET
)
903 struct prefix_ipv4 p
;
906 p
.prefix
= dest
.sin
.sin_addr
;
907 if (flags
& RTF_HOST
)
908 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
910 p
.prefixlen
= ip_masklen (mask
.sin
.sin_addr
);
912 /* Catch self originated messages and match them against our current RIB.
913 * At the same time, ignore unconfirmed messages, they should be tracked
914 * by rtm_write() and kernel_rtm_ipv4().
916 if (rtm
->rtm_type
!= RTM_GET
&& rtm
->rtm_pid
== pid
)
918 char buf
[PREFIX2STR_BUFFER
], gate_buf
[INET_ADDRSTRLEN
];
920 if (! IS_ZEBRA_DEBUG_RIB
)
922 ret
= rib_lookup_ipv4_route (&p
, &gate
, VRF_DEFAULT
);
923 prefix2str (&p
, buf
, sizeof(buf
));
924 switch (rtm
->rtm_type
)
929 /* The kernel notifies us about a new route in FIB created by us.
930 Do we have a correspondent entry in our RIB? */
933 case ZEBRA_RIB_NOTFOUND
:
934 zlog_debug ("%s: %s %s/%d: desync: RR isn't yet in RIB, while already in FIB",
935 __func__
, lookup (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
);
937 case ZEBRA_RIB_FOUND_CONNECTED
:
938 case ZEBRA_RIB_FOUND_NOGATE
:
939 inet_ntop (AF_INET
, &gate
.sin
.sin_addr
, gate_buf
, INET_ADDRSTRLEN
);
940 zlog_debug ("%s: %s %s/%d: desync: RR is in RIB, but gate differs (ours is %s)",
941 __func__
, lookup (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
, gate_buf
);
943 case ZEBRA_RIB_FOUND_EXACT
: /* RIB RR == FIB RR */
944 zlog_debug ("%s: %s %s/%d: done Ok",
945 __func__
, lookup (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
);
946 rib_lookup_and_dump (&p
);
952 /* The kernel notifies us about a route deleted by us. Do we still
953 have it in the RIB? Do we have anything instead? */
956 case ZEBRA_RIB_FOUND_EXACT
:
957 zlog_debug ("%s: %s %s/%d: desync: RR is still in RIB, while already not in FIB",
958 __func__
, lookup (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
);
959 rib_lookup_and_dump (&p
);
961 case ZEBRA_RIB_FOUND_CONNECTED
:
962 case ZEBRA_RIB_FOUND_NOGATE
:
963 zlog_debug ("%s: %s %s/%d: desync: RR is still in RIB, plus gate differs",
964 __func__
, lookup (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
);
965 rib_lookup_and_dump (&p
);
967 case ZEBRA_RIB_NOTFOUND
: /* RIB RR == FIB RR */
968 zlog_debug ("%s: %s %s/%d: done Ok",
969 __func__
, lookup (rtm_type_str
, rtm
->rtm_type
), buf
, p
.prefixlen
);
970 rib_lookup_and_dump (&p
);
976 zlog_debug ("%s: %s/%d: warning: loopback RTM of type %s received",
977 __func__
, buf
, p
.prefixlen
, lookup (rtm_type_str
, rtm
->rtm_type
));
982 /* Change, delete the old prefix, we have no further information
983 * to specify the route really
985 if (rtm
->rtm_type
== RTM_CHANGE
)
986 rib_delete_ipv4 (ZEBRA_ROUTE_KERNEL
, 0, zebra_flags
, &p
,
987 NULL
, 0, VRF_DEFAULT
, SAFI_UNICAST
);
989 if (rtm
->rtm_type
== RTM_GET
990 || rtm
->rtm_type
== RTM_ADD
991 || rtm
->rtm_type
== RTM_CHANGE
)
992 rib_add_ipv4 (ZEBRA_ROUTE_KERNEL
, 0, zebra_flags
,
993 &p
, &gate
.sin
.sin_addr
, NULL
, 0, VRF_DEFAULT
, 0, 0, SAFI_UNICAST
);
995 rib_delete_ipv4 (ZEBRA_ROUTE_KERNEL
, 0 zebra_flags
,
996 &p
, &gate
.sin
.sin_addr
, 0, VRF_DEFAULT
, SAFI_UNICAST
);
999 if (dest
.sa
.sa_family
== AF_INET6
)
1001 /* One day we might have a debug section here like one in the
1002 * IPv4 case above. Just ignore own messages at the moment.
1004 if (rtm
->rtm_type
!= RTM_GET
&& rtm
->rtm_pid
== pid
)
1006 struct prefix_ipv6 p
;
1007 unsigned int ifindex
= 0;
1009 p
.family
= AF_INET6
;
1010 p
.prefix
= dest
.sin6
.sin6_addr
;
1011 if (flags
& RTF_HOST
)
1012 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
1014 p
.prefixlen
= ip6_masklen (mask
.sin6
.sin6_addr
);
1017 if (IN6_IS_ADDR_LINKLOCAL (&gate
.sin6
.sin6_addr
))
1019 ifindex
= IN6_LINKLOCAL_IFINDEX (gate
.sin6
.sin6_addr
);
1020 SET_IN6_LINKLOCAL_IFINDEX (gate
.sin6
.sin6_addr
, 0);
1024 /* CHANGE: delete the old prefix, we have no further information
1025 * to specify the route really
1027 if (rtm
->rtm_type
== RTM_CHANGE
)
1028 rib_delete_ipv6 (ZEBRA_ROUTE_KERNEL
, 0, zebra_flags
, &p
,
1029 NULL
, 0, VRF_DEFAULT
, SAFI_UNICAST
);
1031 if (rtm
->rtm_type
== RTM_GET
1032 || rtm
->rtm_type
== RTM_ADD
1033 || rtm
->rtm_type
== RTM_CHANGE
)
1034 rib_add_ipv6 (ZEBRA_ROUTE_KERNEL
, 0, zebra_flags
,
1035 &p
, &gate
.sin6
.sin6_addr
, ifindex
, VRF_DEFAULT
, 0, 0, SAFI_UNICAST
);
1037 rib_delete_ipv6 (ZEBRA_ROUTE_KERNEL
, 0, zebra_flags
,
1038 &p
, &gate
.sin6
.sin6_addr
, ifindex
, VRF_DEFAULT
, SAFI_UNICAST
);
1040 #endif /* HAVE_IPV6 */
1043 /* Interface function for the kernel routing table updates. Support
1044 * for RTM_CHANGE will be needed.
1045 * Exported only for rt_socket.c
1048 rtm_write (int message
,
1049 union sockunion
*dest
,
1050 union sockunion
*mask
,
1051 union sockunion
*gate
,
1058 struct interface
*ifp
;
1060 /* Sequencial number of routing message. */
1061 static int msg_seq
= 0;
1063 /* Struct of rt_msghdr and buffer for storing socket's data. */
1066 struct rt_msghdr rtm
;
1070 if (routing_sock
< 0)
1071 return ZEBRA_ERR_EPERM
;
1073 /* Clear and set rt_msghdr values */
1074 memset (&msg
, 0, sizeof (struct rt_msghdr
));
1075 msg
.rtm
.rtm_version
= RTM_VERSION
;
1076 msg
.rtm
.rtm_type
= message
;
1077 msg
.rtm
.rtm_seq
= msg_seq
++;
1078 msg
.rtm
.rtm_addrs
= RTA_DST
;
1079 msg
.rtm
.rtm_addrs
|= RTA_GATEWAY
;
1080 msg
.rtm
.rtm_flags
= RTF_UP
;
1081 msg
.rtm
.rtm_index
= index
;
1085 msg
.rtm
.rtm_rmx
.rmx_hopcount
= metric
;
1086 msg
.rtm
.rtm_inits
|= RTV_HOPCOUNT
;
1089 ifp
= if_lookup_by_index (index
);
1091 if (gate
&& message
== RTM_ADD
)
1092 msg
.rtm
.rtm_flags
|= RTF_GATEWAY
;
1094 /* When RTF_CLONING is unavailable on BSD, should we set some
1095 * other flag instead?
1098 if (! gate
&& message
== RTM_ADD
&& ifp
&&
1099 (ifp
->flags
& IFF_POINTOPOINT
) == 0)
1100 msg
.rtm
.rtm_flags
|= RTF_CLONING
;
1101 #endif /* RTF_CLONING */
1103 /* If no protocol specific gateway is specified, use link
1104 address for gateway. */
1109 char dest_buf
[INET_ADDRSTRLEN
] = "NULL", mask_buf
[INET_ADDRSTRLEN
] = "255.255.255.255";
1111 inet_ntop (AF_INET
, &dest
->sin
.sin_addr
, dest_buf
, INET_ADDRSTRLEN
);
1113 inet_ntop (AF_INET
, &mask
->sin
.sin_addr
, mask_buf
, INET_ADDRSTRLEN
);
1114 zlog_warn ("%s: %s/%s: gate == NULL and no gateway found for ifindex %d",
1115 __func__
, dest_buf
, mask_buf
, index
);
1118 gate
= (union sockunion
*) & ifp
->sdl
;
1122 msg
.rtm
.rtm_addrs
|= RTA_NETMASK
;
1123 else if (message
== RTM_ADD
)
1124 msg
.rtm
.rtm_flags
|= RTF_HOST
;
1126 /* Tagging route with flags */
1127 msg
.rtm
.rtm_flags
|= (RTF_PROTO1
);
1129 /* Additional flags. */
1130 if (zebra_flags
& ZEBRA_FLAG_BLACKHOLE
)
1131 msg
.rtm
.rtm_flags
|= RTF_BLACKHOLE
;
1132 if (zebra_flags
& ZEBRA_FLAG_REJECT
)
1133 msg
.rtm
.rtm_flags
|= RTF_REJECT
;
1136 #define SOCKADDRSET(X,R) \
1137 if (msg.rtm.rtm_addrs & (R)) \
1139 int len = SAROUNDUP (X); \
1140 memcpy (pnt, (caddr_t)(X), len); \
1144 pnt
= (caddr_t
) msg
.buf
;
1146 /* Write each socket data into rtm message buffer */
1147 SOCKADDRSET (dest
, RTA_DST
);
1148 SOCKADDRSET (gate
, RTA_GATEWAY
);
1149 SOCKADDRSET (mask
, RTA_NETMASK
);
1151 msg
.rtm
.rtm_msglen
= pnt
- (caddr_t
) &msg
;
1153 ret
= write (routing_sock
, &msg
, msg
.rtm
.rtm_msglen
);
1155 if (ret
!= msg
.rtm
.rtm_msglen
)
1157 if (errno
== EEXIST
)
1158 return ZEBRA_ERR_RTEXIST
;
1159 if (errno
== ENETUNREACH
)
1160 return ZEBRA_ERR_RTUNREACH
;
1162 return ZEBRA_ERR_RTNOEXIST
;
1164 zlog_warn ("%s: write : %s (%d)", __func__
, safe_strerror (errno
), errno
);
1165 return ZEBRA_ERR_KERNEL
;
1167 return ZEBRA_ERR_NOERROR
;
1172 #include "zebra/zserv.h"
1174 /* For debug purpose. */
1176 rtmsg_debug (struct rt_msghdr
*rtm
)
1178 zlog_debug ("Kernel: Len: %d Type: %s", rtm
->rtm_msglen
, lookup (rtm_type_str
, rtm
->rtm_type
));
1179 rtm_flag_dump (rtm
->rtm_flags
);
1180 zlog_debug ("Kernel: message seq %d", rtm
->rtm_seq
);
1181 zlog_debug ("Kernel: pid %lld, rtm_addrs 0x%x",
1182 (long long)rtm
->rtm_pid
, rtm
->rtm_addrs
);
1185 /* This is pretty gross, better suggestions welcome -- mhandler */
1188 #define RTAX_MAX RTA_NUMBITS
1191 #endif /* RTA_NUMBITS */
1192 #endif /* RTAX_MAX */
1194 /* Kernel routing table and interface updates via routing socket. */
1196 kernel_read (struct thread
*thread
)
1200 struct rt_msghdr
*rtm
;
1203 * This must be big enough for any message the kernel might send.
1204 * Rather than determining how many sockaddrs of what size might be
1205 * in each particular message, just use RTAX_MAX of sockaddr_storage
1206 * for each. Note that the sockaddrs must be after each message
1207 * definition, or rather after whichever happens to be the largest,
1208 * since the buffer needs to be big enough for a message and the
1209 * sockaddrs together.
1213 /* Routing information. */
1216 struct rt_msghdr rtm
;
1217 struct sockaddr_storage addr
[RTAX_MAX
];
1220 /* Interface information. */
1223 struct if_msghdr ifm
;
1224 struct sockaddr_storage addr
[RTAX_MAX
];
1227 /* Interface address information. */
1230 struct ifa_msghdr ifa
;
1231 struct sockaddr_storage addr
[RTAX_MAX
];
1234 #ifdef RTM_IFANNOUNCE
1235 /* Interface arrival/departure */
1238 struct if_announcemsghdr ifan
;
1239 struct sockaddr_storage addr
[RTAX_MAX
];
1241 #endif /* RTM_IFANNOUNCE */
1245 /* Fetch routing socket. */
1246 sock
= THREAD_FD (thread
);
1248 nbytes
= read (sock
, &buf
, sizeof buf
);
1252 if (nbytes
< 0 && errno
!= EWOULDBLOCK
&& errno
!= EAGAIN
)
1253 zlog_warn ("routing socket error: %s", safe_strerror (errno
));
1257 thread_add_read (zebrad
.master
, kernel_read
, NULL
, sock
);
1259 if (IS_ZEBRA_DEBUG_KERNEL
)
1260 rtmsg_debug (&buf
.r
.rtm
);
1265 * Ensure that we didn't drop any data, so that processing routines
1266 * can assume they have the whole message.
1268 if (rtm
->rtm_msglen
!= nbytes
)
1270 zlog_warn ("kernel_read: rtm->rtm_msglen %d, nbytes %d, type %d\n",
1271 rtm
->rtm_msglen
, nbytes
, rtm
->rtm_type
);
1275 switch (rtm
->rtm_type
)
1283 ifm_read (&buf
.im
.ifm
);
1287 ifam_read (&buf
.ia
.ifa
);
1289 #ifdef RTM_IFANNOUNCE
1290 case RTM_IFANNOUNCE
:
1291 ifan_read (&buf
.ian
.ifan
);
1293 #endif /* RTM_IFANNOUNCE */
1295 if (IS_ZEBRA_DEBUG_KERNEL
)
1296 zlog_debug("Unprocessed RTM_type: %d", rtm
->rtm_type
);
1302 /* Make routing socket. */
1304 routing_socket (struct zebra_ns
*zns
)
1306 if ( zserv_privs
.change (ZPRIVS_RAISE
) )
1307 zlog_err ("routing_socket: Can't raise privileges");
1309 routing_sock
= socket (AF_ROUTE
, SOCK_RAW
, 0);
1311 if (routing_sock
< 0)
1313 if ( zserv_privs
.change (ZPRIVS_LOWER
) )
1314 zlog_err ("routing_socket: Can't lower privileges");
1315 zlog_warn ("Can't init kernel routing socket");
1319 /* XXX: Socket should be NONBLOCK, however as we currently
1320 * discard failed writes, this will lead to inconsistencies.
1321 * For now, socket must be blocking.
1323 /*if (fcntl (routing_sock, F_SETFL, O_NONBLOCK) < 0)
1324 zlog_warn ("Can't set O_NONBLOCK to routing socket");*/
1326 if ( zserv_privs
.change (ZPRIVS_LOWER
) )
1327 zlog_err ("routing_socket: Can't lower privileges");
1329 /* kernel_read needs rewrite. */
1330 thread_add_read (zebrad
.master
, kernel_read
, NULL
, routing_sock
);
1333 /* Exported interface function. This function simply calls
1334 routing_socket (). */
1336 kernel_init (struct zebra_ns
*zns
)
1338 routing_socket (zns
);
1342 kernel_terminate (struct zebra_ns
*zns
)