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
23 #include <net/if_types.h>
27 #include "sockunion.h"
28 #include "connected.h"
38 #include "zebra/interface.h"
39 #include "zebra/zserv.h"
40 #include "zebra/debug.h"
41 #include "zebra/kernel_socket.h"
42 #include "zebra/rib.h"
44 extern struct zebra_privs_t zserv_privs;
45 extern struct zebra_t zebrad;
48 * Historically, the BSD routing socket has aligned data following a
49 * struct sockaddr to sizeof(long), which was 4 bytes on some
50 * platforms, and 8 bytes on others. NetBSD 6 changed the routing
51 * socket to align to sizeof(uint64_t), which is 8 bytes. OS X
52 * appears to align to sizeof(int), which is 4 bytes.
54 * Alignment of zero-sized sockaddrs is nonsensical, but historically
55 * BSD defines RT_ROUNDUP(0) to be the alignment interval (rather than
56 * 0). We follow this practice without questioning it, but it is a
57 * bug if quagga calls ROUNDUP with 0.
61 * Because of these varying conventions, the only sane approach is for
62 * the <net/route.h> header to define some flavor of ROUNDUP macro.
66 /* SAROUNDUP is the only thing we need, and SA_SIZE provides that */
67 #define SAROUNDUP(a) SA_SIZE(a)
70 #if defined(RT_ROUNDUP)
71 #define ROUNDUP(a) RT_ROUNDUP(a)
72 #endif /* defined(RT_ROUNDUP) */
75 /* Solaris has struct sockaddr_in[6] definitions at 16 / 32 bytes size,
76 * so the whole concept doesn't really apply. */
77 #define ROUNDUP(a) (a)
81 * If ROUNDUP has not yet been defined in terms of platform-provided
82 * defines, attempt to cope with heuristics.
87 * It's a bug for a platform not to define rounding/alignment for
88 * sockaddrs on the routing socket. This warning really is
89 * intentional, to provoke filing bug reports with operating systems
90 * that don't define RT_ROUNDUP or equivalent.
92 #warning "net/route.h does not define RT_ROUNDUP; making unwarranted assumptions!"
94 /* OS X (Xcode as of 2014-12) is known not to define RT_ROUNDUP */
96 #define ROUNDUP_TYPE int
98 #define ROUNDUP_TYPE long
102 ((a) > 0 ? (1 + (((a) - 1) | (sizeof(ROUNDUP_TYPE) - 1))) : sizeof(ROUNDUP_TYPE))
104 #endif /* defined(ROUNDUP) */
107 * Given a pointer (sockaddr or void *), return the number of bytes
108 * taken up by the sockaddr and any padding needed for alignment.
110 #if defined(HAVE_STRUCT_SOCKADDR_SA_LEN)
111 #define SAROUNDUP(X) ROUNDUP(((struct sockaddr *)(X))->sa_len)
112 #elif defined(HAVE_IPV6)
114 * One would hope all fixed-size structure definitions are aligned,
115 * but round them up nonetheless.
117 #define SAROUNDUP(X) \
118 (((struct sockaddr *)(X))->sa_family == AF_INET ? \
119 ROUNDUP(sizeof(struct sockaddr_in)):\
120 (((struct sockaddr *)(X))->sa_family == AF_INET6 ? \
121 ROUNDUP(sizeof(struct sockaddr_in6)) : \
122 (((struct sockaddr *)(X))->sa_family == AF_LINK ? \
123 ROUNDUP(sizeof(struct sockaddr_dl)) : sizeof(struct sockaddr))))
124 #else /* HAVE_IPV6 */
125 #define SAROUNDUP(X) \
126 (((struct sockaddr *)(X))->sa_family == AF_INET ? \
127 ROUNDUP(sizeof(struct sockaddr_in)):\
128 (((struct sockaddr *)(X))->sa_family == AF_LINK ? \
129 ROUNDUP(sizeof(struct sockaddr_dl)) : sizeof(struct sockaddr)))
130 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
132 #endif /* !SA_SIZE */
135 * We use a call to an inline function to copy (PNT) to (DEST)
136 * 1. Calculating the length of the copy requires an #ifdef to determine
137 * if sa_len is a field and can't be used directly inside a #define
138 * 2. So the compiler doesn't complain when DEST is NULL, which is only true
139 * when we are skipping the copy and incrementing to the next SA
142 rta_copy (union sockunion *dest, caddr_t src) {
146 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
147 len = (((struct sockaddr *)src)->sa_len > sizeof (*dest)) ?
148 sizeof (*dest) : ((struct sockaddr *)src)->sa_len ;
150 len = (SAROUNDUP (src) > sizeof (*dest)) ?
151 sizeof (*dest) : SAROUNDUP (src) ;
153 memcpy (dest, src, len);
156 #define RTA_ADDR_GET(DEST, RTA, RTMADDRS, PNT) \
157 if ((RTMADDRS) & (RTA)) \
159 int len = SAROUNDUP ((PNT)); \
160 if (af_check (((struct sockaddr *)(PNT))->sa_family)) \
161 rta_copy((DEST), (PNT)); \
164 #define RTA_ATTR_GET(DEST, RTA, RTMADDRS, PNT) \
165 if ((RTMADDRS) & (RTA)) \
167 int len = SAROUNDUP ((PNT)); \
168 rta_copy((DEST), (PNT)); \
172 #define RTA_NAME_GET(DEST, RTA, RTMADDRS, PNT, LEN) \
173 if ((RTMADDRS) & (RTA)) \
175 u_char *pdest = (u_char *) (DEST); \
176 int len = SAROUNDUP ((PNT)); \
177 struct sockaddr_dl *sdl = (struct sockaddr_dl *)(PNT); \
178 if (IS_ZEBRA_DEBUG_KERNEL) \
179 zlog_debug ("%s: RTA_SDL_GET nlen %d, alen %d", \
180 __func__, sdl->sdl_nlen, sdl->sdl_alen); \
181 if ( ((DEST) != NULL) && (sdl->sdl_family == AF_LINK) \
182 && (sdl->sdl_nlen < IFNAMSIZ) && (sdl->sdl_nlen <= len) ) \
184 memcpy (pdest, sdl->sdl_data, sdl->sdl_nlen); \
185 pdest[sdl->sdl_nlen] = '\0'; \
186 (LEN) = sdl->sdl_nlen; \
194 /* Routing socket message types. */
195 const struct message rtm_type_str[] =
197 {RTM_ADD, "RTM_ADD"},
198 {RTM_DELETE, "RTM_DELETE"},
199 {RTM_CHANGE, "RTM_CHANGE"},
200 {RTM_GET, "RTM_GET"},
201 {RTM_LOSING, "RTM_LOSING"},
202 {RTM_REDIRECT, "RTM_REDIRECT"},
203 {RTM_MISS, "RTM_MISS"},
204 {RTM_LOCK, "RTM_LOCK"},
206 {RTM_OLDADD, "RTM_OLDADD"},
207 #endif /* RTM_OLDADD */
209 {RTM_OLDDEL, "RTM_OLDDEL"},
210 #endif /* RTM_OLDDEL */
211 {RTM_RESOLVE, "RTM_RESOLVE"},
212 {RTM_NEWADDR, "RTM_NEWADDR"},
213 {RTM_DELADDR, "RTM_DELADDR"},
214 {RTM_IFINFO, "RTM_IFINFO"},
216 {RTM_OIFINFO, "RTM_OIFINFO"},
217 #endif /* RTM_OIFINFO */
219 {RTM_NEWMADDR, "RTM_NEWMADDR"},
220 #endif /* RTM_NEWMADDR */
222 {RTM_DELMADDR, "RTM_DELMADDR"},
223 #endif /* RTM_DELMADDR */
224 #ifdef RTM_IFANNOUNCE
225 {RTM_IFANNOUNCE, "RTM_IFANNOUNCE"},
226 #endif /* RTM_IFANNOUNCE */
230 static const struct message rtm_flag_str[] =
233 {RTF_GATEWAY, "GATEWAY"},
235 {RTF_REJECT, "REJECT"},
236 {RTF_DYNAMIC, "DYNAMIC"},
237 {RTF_MODIFIED, "MODIFIED"},
241 #endif /* RTF_MASK */
243 {RTF_CLONING, "CLONING"},
244 #endif /* RTF_CLONING */
246 {RTF_XRESOLVE, "XRESOLVE"},
247 #endif /* RTF_XRESOLVE */
249 {RTF_LLINFO, "LLINFO"},
250 #endif /* RTF_LLINFO */
251 {RTF_STATIC, "STATIC"},
252 {RTF_BLACKHOLE, "BLACKHOLE"},
254 {RTF_PRIVATE, "PRIVATE"},
255 #endif /* RTF_PRIVATE */
256 {RTF_PROTO1, "PROTO1"},
257 {RTF_PROTO2, "PROTO2"},
259 {RTF_PRCLONING, "PRCLONING"},
260 #endif /* RTF_PRCLONING */
262 {RTF_WASCLONED, "WASCLONED"},
263 #endif /* RTF_WASCLONED */
265 {RTF_PROTO3, "PROTO3"},
266 #endif /* RTF_PROTO3 */
268 {RTF_PINNED, "PINNED"},
269 #endif /* RTF_PINNED */
271 {RTF_LOCAL, "LOCAL"},
272 #endif /* RTF_LOCAL */
274 {RTF_BROADCAST, "BROADCAST"},
275 #endif /* RTF_BROADCAST */
277 {RTF_MULTICAST, "MULTICAST"},
278 #endif /* RTF_MULTICAST */
280 {RTF_MULTIRT, "MULTIRT"},
281 #endif /* RTF_MULTIRT */
283 {RTF_SETSRC, "SETSRC"},
284 #endif /* RTF_SETSRC */
288 /* Kernel routing update socket. */
289 int routing_sock = -1;
291 /* Yes I'm checking ugly routing socket behavior. */
294 /* Supported address family check. */
296 af_check (int family)
298 if (family == AF_INET)
301 if (family == AF_INET6)
303 #endif /* HAVE_IPV6 */
307 /* Dump routing table flag for debug purpose. */
309 rtm_flag_dump (int flag)
311 const struct message *mes;
312 static char buf[BUFSIZ];
315 for (mes = rtm_flag_str; mes->key != 0; mes++)
319 strlcat (buf, mes->str, BUFSIZ);
320 strlcat (buf, " ", BUFSIZ);
323 zlog_debug ("Kernel: %s", buf);
326 #ifdef RTM_IFANNOUNCE
327 /* Interface adding function */
329 ifan_read (struct if_announcemsghdr *ifan)
331 struct interface *ifp;
333 ifp = if_lookup_by_index (ifan->ifan_index);
336 assert ( (ifp->ifindex == ifan->ifan_index)
337 || (ifp->ifindex == IFINDEX_INTERNAL) );
340 || ((ifp->ifindex == IFINDEX_INTERNAL)
341 && (ifan->ifan_what == IFAN_ARRIVAL)) )
343 if (IS_ZEBRA_DEBUG_KERNEL)
344 zlog_debug ("%s: creating interface for ifindex %d, name %s",
345 __func__, ifan->ifan_index, ifan->ifan_name);
347 /* Create Interface */
348 ifp = if_get_by_name_len(ifan->ifan_name,
349 strnlen(ifan->ifan_name,
350 sizeof(ifan->ifan_name)));
351 ifp->ifindex = ifan->ifan_index;
356 else if (ifp != NULL && ifan->ifan_what == IFAN_DEPARTURE)
357 if_delete_update (ifp);
363 if (IS_ZEBRA_DEBUG_KERNEL)
364 zlog_debug ("%s: interface %s index %d",
365 __func__, ifan->ifan_name, ifan->ifan_index);
369 #endif /* RTM_IFANNOUNCE */
371 #ifdef HAVE_BSD_IFI_LINK_STATE
372 /* BSD link detect translation */
374 bsd_linkdetect_translate (struct if_msghdr *ifm)
376 if ((ifm->ifm_data.ifi_link_state >= LINK_STATE_UP) ||
377 (ifm->ifm_data.ifi_link_state == LINK_STATE_UNKNOWN))
378 SET_FLAG(ifm->ifm_flags, IFF_RUNNING);
380 UNSET_FLAG(ifm->ifm_flags, IFF_RUNNING);
382 #endif /* HAVE_BSD_IFI_LINK_STATE */
384 static enum zebra_link_type
385 sdl_to_zebra_link_type (unsigned int sdlt)
389 case IFT_ETHER: return ZEBRA_LLT_ETHER;
390 case IFT_X25: return ZEBRA_LLT_X25;
391 case IFT_FDDI: return ZEBRA_LLT_FDDI;
392 case IFT_PPP: return ZEBRA_LLT_PPP;
393 case IFT_LOOP: return ZEBRA_LLT_LOOPBACK;
394 case IFT_SLIP: return ZEBRA_LLT_SLIP;
395 case IFT_ARCNET: return ZEBRA_LLT_ARCNET;
396 case IFT_ATM: return ZEBRA_LLT_ATM;
397 case IFT_LOCALTALK: return ZEBRA_LLT_LOCALTLK;
398 case IFT_HIPPI: return ZEBRA_LLT_HIPPI;
400 case IFT_IEEE1394: return ZEBRA_LLT_IEEE1394;
403 default: return ZEBRA_LLT_UNKNOWN;
408 * Handle struct if_msghdr obtained from reading routing socket or
409 * sysctl (from interface_list). There may or may not be sockaddrs
410 * present after the header.
413 ifm_read (struct if_msghdr *ifm)
415 struct interface *ifp = NULL;
416 struct sockaddr_dl *sdl;
417 char ifname[IFNAMSIZ];
421 /* terminate ifname at head (for strnlen) and tail (for safety) */
422 ifname[IFNAMSIZ - 1] = '\0';
424 /* paranoia: sanity check structure */
425 if (ifm->ifm_msglen < sizeof(struct if_msghdr))
427 zlog_err ("ifm_read: ifm->ifm_msglen %d too short\n",
433 * Check for a sockaddr_dl following the message. First, point to
434 * where a socakddr might be if one follows the message.
436 cp = (void *)(ifm + 1);
440 * XXX This behavior should be narrowed to only the kernel versions
441 * for which the structures returned do not match the headers.
443 * if_msghdr_t on 64 bit kernels in Solaris 9 and earlier versions
444 * is 12 bytes larger than the 32 bit version.
446 if (((struct sockaddr *) cp)->sa_family == AF_UNSPEC)
450 RTA_ADDR_GET (NULL, RTA_DST, ifm->ifm_addrs, cp);
451 RTA_ADDR_GET (NULL, RTA_GATEWAY, ifm->ifm_addrs, cp);
452 RTA_ATTR_GET (NULL, RTA_NETMASK, ifm->ifm_addrs, cp);
453 RTA_ADDR_GET (NULL, RTA_GENMASK, ifm->ifm_addrs, cp);
454 sdl = (struct sockaddr_dl *)cp;
455 RTA_NAME_GET (ifname, RTA_IFP, ifm->ifm_addrs, cp, ifnlen);
456 RTA_ADDR_GET (NULL, RTA_IFA, ifm->ifm_addrs, cp);
457 RTA_ADDR_GET (NULL, RTA_AUTHOR, ifm->ifm_addrs, cp);
458 RTA_ADDR_GET (NULL, RTA_BRD, ifm->ifm_addrs, cp);
460 if (IS_ZEBRA_DEBUG_KERNEL)
461 zlog_debug ("%s: sdl ifname %s", __func__, (ifnlen ? ifname : "(nil)"));
464 * Look up on ifindex first, because ifindices are the primary handle for
465 * interfaces across the user/kernel boundary, for most systems. (Some
466 * messages, such as up/down status changes on NetBSD, do not include a
469 if ( (ifp = if_lookup_by_index (ifm->ifm_index)) != NULL )
471 /* we have an ifp, verify that the name matches as some systems,
472 * eg Solaris, have a 1:many association of ifindex:ifname
473 * if they dont match, we dont have the correct ifp and should
474 * set it back to NULL to let next check do lookup by name
476 if (ifnlen && (strncmp (ifp->name, ifname, IFNAMSIZ) != 0) )
478 if (IS_ZEBRA_DEBUG_KERNEL)
479 zlog_debug ("%s: ifp name %s doesn't match sdl name %s",
480 __func__, ifp->name, ifname);
486 * If we dont have an ifp, try looking up by name. Particularly as some
487 * systems (Solaris) have a 1:many mapping of ifindex:ifname - the ifname
488 * is therefore our unique handle to that interface.
490 * Interfaces specified in the configuration file for which the ifindex
491 * has not been determined will have ifindex == IFINDEX_INTERNAL, and such
492 * interfaces are found by this search, and then their ifindex values can
495 if ( (ifp == NULL) && ifnlen)
496 ifp = if_lookup_by_name (ifname);
499 * If ifp still does not exist or has an invalid index (IFINDEX_INTERNAL),
500 * create or fill in an interface.
502 if ((ifp == NULL) || (ifp->ifindex == IFINDEX_INTERNAL))
505 * To create or fill in an interface, a sockaddr_dl (via
506 * RTA_IFP) is required.
510 zlog_warn ("Interface index %d (new) missing ifname\n",
515 #ifndef RTM_IFANNOUNCE
516 /* Down->Down interface should be ignored here.
517 * See further comment below.
519 if (!CHECK_FLAG (ifm->ifm_flags, IFF_UP))
521 #endif /* !RTM_IFANNOUNCE */
525 /* Interface that zebra was not previously aware of, so create. */
526 ifp = if_create (ifname, ifnlen);
527 if (IS_ZEBRA_DEBUG_KERNEL)
528 zlog_debug ("%s: creating ifp for ifindex %d",
529 __func__, ifm->ifm_index);
532 if (IS_ZEBRA_DEBUG_KERNEL)
533 zlog_debug ("%s: updated/created ifp, ifname %s, ifindex %d",
534 __func__, ifp->name, ifp->ifindex);
536 * Fill in newly created interface structure, or larval
537 * structure with ifindex IFINDEX_INTERNAL.
539 ifp->ifindex = ifm->ifm_index;
541 #ifdef HAVE_BSD_IFI_LINK_STATE /* translate BSD kernel msg for link-state */
542 bsd_linkdetect_translate(ifm);
543 #endif /* HAVE_BSD_IFI_LINK_STATE */
545 if_flags_update (ifp, ifm->ifm_flags);
546 #if defined(__bsdi__)
547 if_kvm_get_mtu (ifp);
550 #endif /* __bsdi__ */
554 * XXX sockaddr_dl contents can be larger than the structure
555 * definition. There are 2 big families here:
556 * - BSD has sdl_len + sdl_data[16] + overruns sdl_data
557 * we MUST use sdl_len here or we'll truncate data.
558 * - Solaris has no sdl_len, but sdl_data[244]
559 * presumably, it's not going to run past that, so sizeof()
561 * a nonzero ifnlen from RTA_NAME_GET() means sdl is valid
563 ifp->ll_type = ZEBRA_LLT_UNKNOWN;
564 ifp->hw_addr_len = 0;
567 #ifdef HAVE_STRUCT_SOCKADDR_DL_SDL_LEN
568 memcpy (&((struct zebra_if *)ifp->info)->sdl, sdl, sdl->sdl_len);
570 memcpy (&((struct zebra_if *)ifp->info)->sdl, sdl, sizeof (struct sockaddr_dl));
571 #endif /* HAVE_STRUCT_SOCKADDR_DL_SDL_LEN */
573 ifp->ll_type = sdl_to_zebra_link_type (sdl->sdl_type);
574 if (sdl->sdl_alen <= sizeof(ifp->hw_addr))
576 memcpy (ifp->hw_addr, LLADDR(sdl), sdl->sdl_alen);
577 ifp->hw_addr_len = sdl->sdl_alen;
585 * Interface structure exists. Adjust stored flags from
586 * notification. If interface has up->down or down->up
587 * transition, call state change routines (to adjust routes,
588 * notify routing daemons, etc.). (Other flag changes are stored
589 * but apparently do not trigger action.)
592 if (ifp->ifindex != ifm->ifm_index)
594 zlog_warn ("%s: index mismatch, ifname %s, ifp index %d, "
596 __func__, ifp->name, ifp->ifindex, ifm->ifm_index);
600 #ifdef HAVE_BSD_IFI_LINK_STATE /* translate BSD kernel msg for link-state */
601 bsd_linkdetect_translate(ifm);
602 #endif /* HAVE_BSD_IFI_LINK_STATE */
604 /* update flags and handle operative->inoperative transition, if any */
605 if_flags_update (ifp, ifm->ifm_flags);
607 #ifndef RTM_IFANNOUNCE
610 /* No RTM_IFANNOUNCE on this platform, so we can never
611 * distinguish between ~IFF_UP and delete. We must presume
612 * it has been deleted.
613 * Eg, Solaris will not notify us of unplumb.
615 * XXX: Fixme - this should be runtime detected
616 * So that a binary compiled on a system with IFANNOUNCE
617 * will still behave correctly if run on a platform without
619 if_delete_update (ifp);
621 #endif /* RTM_IFANNOUNCE */
624 #if defined(__bsdi__)
625 if_kvm_get_mtu (ifp);
628 #endif /* __bsdi__ */
633 #ifdef HAVE_NET_RT_IFLIST
634 ifp->stats = ifm->ifm_data;
635 #endif /* HAVE_NET_RT_IFLIST */
637 if (IS_ZEBRA_DEBUG_KERNEL)
638 zlog_debug ("%s: interface %s index %d",
639 __func__, ifp->name, ifp->ifindex);
644 /* Address read from struct ifa_msghdr. */
646 ifam_read_mesg (struct ifa_msghdr *ifm,
647 union sockunion *addr,
648 union sockunion *mask,
649 union sockunion *brd,
655 union sockunion gateway;
657 pnt = (caddr_t)(ifm + 1);
658 end = ((caddr_t)ifm) + ifm->ifam_msglen;
660 /* Be sure structure is cleared */
661 memset (mask, 0, sizeof (union sockunion));
662 memset (addr, 0, sizeof (union sockunion));
663 memset (brd, 0, sizeof (union sockunion));
664 memset (&dst, 0, sizeof (union sockunion));
665 memset (&gateway, 0, sizeof (union sockunion));
667 /* We fetch each socket variable into sockunion. */
668 RTA_ADDR_GET (&dst, RTA_DST, ifm->ifam_addrs, pnt);
669 RTA_ADDR_GET (&gateway, RTA_GATEWAY, ifm->ifam_addrs, pnt);
670 RTA_ATTR_GET (mask, RTA_NETMASK, ifm->ifam_addrs, pnt);
671 RTA_ADDR_GET (NULL, RTA_GENMASK, ifm->ifam_addrs, pnt);
672 RTA_NAME_GET (ifname, RTA_IFP, ifm->ifam_addrs, pnt, *ifnlen);
673 RTA_ADDR_GET (addr, RTA_IFA, ifm->ifam_addrs, pnt);
674 RTA_ADDR_GET (NULL, RTA_AUTHOR, ifm->ifam_addrs, pnt);
675 RTA_ADDR_GET (brd, RTA_BRD, ifm->ifam_addrs, pnt);
677 if (IS_ZEBRA_DEBUG_KERNEL)
679 int family = sockunion_family(addr);
687 char buf[4][INET6_ADDRSTRLEN];
688 zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x, "
689 "ifam_flags 0x%x, addr %s/%d broad %s dst %s "
691 __func__, ifm->ifam_index,
692 (ifnlen ? ifname : "(nil)"), ifm->ifam_addrs,
694 inet_ntop(family,&addr->sin.sin_addr,
695 buf[0],sizeof(buf[0])),
696 ip_masklen(mask->sin.sin_addr),
697 inet_ntop(family,&brd->sin.sin_addr,
698 buf[1],sizeof(buf[1])),
699 inet_ntop(family,&dst.sin.sin_addr,
700 buf[2],sizeof(buf[2])),
701 inet_ntop(family,&gateway.sin.sin_addr,
702 buf[3],sizeof(buf[3])));
706 zlog_debug ("%s: ifindex %d, ifname %s, ifam_addrs 0x%x",
707 __func__, ifm->ifam_index,
708 (ifnlen ? ifname : "(nil)"), ifm->ifam_addrs);
713 /* Assert read up end point matches to end point */
715 zlog_warn ("ifam_read() doesn't read all socket data");
718 /* Interface's address information get. */
720 ifam_read (struct ifa_msghdr *ifam)
722 struct interface *ifp = NULL;
723 union sockunion addr, mask, brd;
724 char ifname[INTERFACE_NAMSIZ];
729 ifname[0] = ifname[INTERFACE_NAMSIZ - 1] = '\0';
731 /* Allocate and read address information. */
732 ifam_read_mesg (ifam, &addr, &mask, &brd, ifname, &ifnlen);
734 if ((ifp = if_lookup_by_index(ifam->ifam_index)) == NULL)
736 zlog_warn ("%s: no interface for ifname %s, index %d",
737 __func__, ifname, ifam->ifam_index);
741 if (ifnlen && strncmp (ifp->name, ifname, INTERFACE_NAMSIZ))
744 /* N.B. The info in ifa_msghdr does not tell us whether the RTA_BRD
745 field contains a broadcast address or a peer address, so we are forced to
746 rely upon the interface type. */
747 if (if_is_pointopoint(ifp))
748 SET_FLAG(flags, ZEBRA_IFA_PEER);
751 /* it might seem cute to grab the interface metric here, however
752 * we're processing an address update message, and so some systems
753 * (e.g. FBSD) dont bother to fill in ifam_metric. Disabled, but left
754 * in deliberately, as comment.
756 ifp->metric = ifam->ifam_metric;
759 /* Add connected address. */
760 switch (sockunion_family (&addr))
763 if (ifam->ifam_type == RTM_NEWADDR)
764 connected_add_ipv4 (ifp, flags, &addr.sin.sin_addr,
765 ip_masklen (mask.sin.sin_addr),
767 (isalias ? ifname : NULL));
769 connected_delete_ipv4 (ifp, flags, &addr.sin.sin_addr,
770 ip_masklen (mask.sin.sin_addr),
775 /* Unset interface index from link-local address when IPv6 stack
777 if (IN6_IS_ADDR_LINKLOCAL (&addr.sin6.sin6_addr))
779 SET_IN6_LINKLOCAL_IFINDEX (addr.sin6.sin6_addr, 0);
782 if (ifam->ifam_type == RTM_NEWADDR)
783 connected_add_ipv6 (ifp, flags, &addr.sin6.sin6_addr,
784 ip6_masklen (mask.sin6.sin6_addr),
786 (isalias ? ifname : NULL));
788 connected_delete_ipv6 (ifp,
789 &addr.sin6.sin6_addr,
790 ip6_masklen (mask.sin6.sin6_addr),
791 &brd.sin6.sin6_addr);
793 #endif /* HAVE_IPV6 */
795 /* Unsupported family silently ignore... */
799 /* Check interface flag for implicit up of the interface. */
803 /* In addition to lacking IFANNOUNCE, on SUNOS IFF_UP is strange.
804 * See comments for SUNOS_5 in interface.c::if_flags_mangle.
806 * Here we take care of case where the real IFF_UP was previously
807 * unset (as kept in struct zebra_if.primary_state) and the mangled
808 * IFF_UP (ie IFF_UP set || listcount(connected) has now transitioned
809 * to unset due to the lost non-primary address having DELADDR'd.
811 * we must delete the interface, because in between here and next
812 * event for this interface-name the administrator could unplumb
813 * and replumb the interface.
816 if_delete_update (ifp);
822 /* Interface function for reading kernel routing table information. */
824 rtm_read_mesg (struct rt_msghdr *rtm,
825 union sockunion *dest,
826 union sockunion *mask,
827 union sockunion *gate,
833 /* Pnt points out socket data start point. */
834 pnt = (caddr_t)(rtm + 1);
835 end = ((caddr_t)rtm) + rtm->rtm_msglen;
837 /* rt_msghdr version check. */
838 if (rtm->rtm_version != RTM_VERSION)
839 zlog (NULL, LOG_WARNING,
840 "Routing message version different %d should be %d."
841 "This may cause problem\n", rtm->rtm_version, RTM_VERSION);
843 /* Be sure structure is cleared */
844 memset (dest, 0, sizeof (union sockunion));
845 memset (gate, 0, sizeof (union sockunion));
846 memset (mask, 0, sizeof (union sockunion));
848 /* We fetch each socket variable into sockunion. */
849 RTA_ADDR_GET (dest, RTA_DST, rtm->rtm_addrs, pnt);
850 RTA_ADDR_GET (gate, RTA_GATEWAY, rtm->rtm_addrs, pnt);
851 RTA_ATTR_GET (mask, RTA_NETMASK, rtm->rtm_addrs, pnt);
852 RTA_ADDR_GET (NULL, RTA_GENMASK, rtm->rtm_addrs, pnt);
853 RTA_NAME_GET (ifname, RTA_IFP, rtm->rtm_addrs, pnt, *ifnlen);
854 RTA_ADDR_GET (NULL, RTA_IFA, rtm->rtm_addrs, pnt);
855 RTA_ADDR_GET (NULL, RTA_AUTHOR, rtm->rtm_addrs, pnt);
856 RTA_ADDR_GET (NULL, RTA_BRD, rtm->rtm_addrs, pnt);
858 /* If there is netmask information set it's family same as
860 if (rtm->rtm_addrs & RTA_NETMASK)
861 mask->sa.sa_family = dest->sa.sa_family;
863 /* Assert read up to the end of pointer. */
865 zlog (NULL, LOG_WARNING, "rtm_read() doesn't read all socket data.");
867 return rtm->rtm_flags;
871 rtm_read (struct rt_msghdr *rtm)
875 union sockunion dest, mask, gate;
876 char ifname[INTERFACE_NAMSIZ + 1];
881 /* Read destination and netmask and gateway from rtm message
883 flags = rtm_read_mesg (rtm, &dest, &mask, &gate, ifname, &ifnlen);
884 if (!(flags & RTF_DONE))
886 if (IS_ZEBRA_DEBUG_KERNEL)
887 zlog_debug ("%s: got rtm of type %d (%s)", __func__, rtm->rtm_type,
888 lookup (rtm_type_str, rtm->rtm_type));
890 #ifdef RTF_CLONED /*bsdi, netbsd 1.6*/
891 if (flags & RTF_CLONED)
894 #ifdef RTF_WASCLONED /*freebsd*/
895 if (flags & RTF_WASCLONED)
899 if ((rtm->rtm_type == RTM_ADD || rtm->rtm_type == RTM_CHANGE) && ! (flags & RTF_UP))
902 /* This is connected route. */
903 if (! (flags & RTF_GATEWAY))
906 if (flags & RTF_PROTO1)
907 SET_FLAG (zebra_flags, ZEBRA_FLAG_SELFROUTE);
909 /* This is persistent route. */
910 if (flags & RTF_STATIC)
911 SET_FLAG (zebra_flags, ZEBRA_FLAG_STATIC);
913 /* This is a reject or blackhole route */
914 if (flags & RTF_REJECT)
915 SET_FLAG (zebra_flags, ZEBRA_FLAG_REJECT);
916 if (flags & RTF_BLACKHOLE)
917 SET_FLAG (zebra_flags, ZEBRA_FLAG_BLACKHOLE);
919 if (dest.sa.sa_family == AF_INET)
921 struct prefix_ipv4 p;
924 p.prefix = dest.sin.sin_addr;
925 if (flags & RTF_HOST)
926 p.prefixlen = IPV4_MAX_PREFIXLEN;
928 p.prefixlen = ip_masklen (mask.sin.sin_addr);
930 /* Catch self originated messages and match them against our current RIB.
931 * At the same time, ignore unconfirmed messages, they should be tracked
932 * by rtm_write() and kernel_rtm_ipv4().
934 if (rtm->rtm_type != RTM_GET && rtm->rtm_pid == pid)
936 char buf[PREFIX_STRLEN], gate_buf[INET_ADDRSTRLEN];
938 if (! IS_ZEBRA_DEBUG_RIB)
940 ret = rib_lookup_ipv4_route (&p, &gate, VRF_DEFAULT);
941 prefix2str (&p, buf, sizeof(buf));
942 switch (rtm->rtm_type)
947 /* The kernel notifies us about a new route in FIB created by us.
948 Do we have a correspondent entry in our RIB? */
951 case ZEBRA_RIB_NOTFOUND:
952 zlog_debug ("%s: %s %s: desync: RR isn't yet in RIB, while already in FIB",
953 __func__, lookup (rtm_type_str, rtm->rtm_type), buf);
955 case ZEBRA_RIB_FOUND_CONNECTED:
956 case ZEBRA_RIB_FOUND_NOGATE:
957 inet_ntop (AF_INET, &gate.sin.sin_addr, gate_buf, INET_ADDRSTRLEN);
958 zlog_debug ("%s: %s %s: desync: RR is in RIB, but gate differs (ours is %s)",
959 __func__, lookup (rtm_type_str, rtm->rtm_type), buf, gate_buf);
961 case ZEBRA_RIB_FOUND_EXACT: /* RIB RR == FIB RR */
962 zlog_debug ("%s: %s %s: done Ok",
963 __func__, lookup (rtm_type_str, rtm->rtm_type), buf);
964 rib_lookup_and_dump (&p);
970 /* The kernel notifies us about a route deleted by us. Do we still
971 have it in the RIB? Do we have anything instead? */
974 case ZEBRA_RIB_FOUND_EXACT:
975 zlog_debug ("%s: %s %s: desync: RR is still in RIB, while already not in FIB",
976 __func__, lookup (rtm_type_str, rtm->rtm_type), buf);
977 rib_lookup_and_dump (&p);
979 case ZEBRA_RIB_FOUND_CONNECTED:
980 case ZEBRA_RIB_FOUND_NOGATE:
981 zlog_debug ("%s: %s %s: desync: RR is still in RIB, plus gate differs",
982 __func__, lookup (rtm_type_str, rtm->rtm_type), buf);
983 rib_lookup_and_dump (&p);
985 case ZEBRA_RIB_NOTFOUND: /* RIB RR == FIB RR */
986 zlog_debug ("%s: %s %s: done Ok",
987 __func__, lookup (rtm_type_str, rtm->rtm_type), buf);
988 rib_lookup_and_dump (&p);
994 zlog_debug ("%s: %s: warning: loopback RTM of type %s received",
995 __func__, buf, lookup (rtm_type_str, rtm->rtm_type));
1000 /* Change, delete the old prefix, we have no further information
1001 * to specify the route really
1003 if (rtm->rtm_type == RTM_CHANGE)
1004 rib_delete_ipv4 (ZEBRA_ROUTE_KERNEL, zebra_flags, &p,
1005 NULL, 0, VRF_DEFAULT, SAFI_UNICAST);
1007 if (rtm->rtm_type == RTM_GET
1008 || rtm->rtm_type == RTM_ADD
1009 || rtm->rtm_type == RTM_CHANGE)
1010 rib_add_ipv4 (ZEBRA_ROUTE_KERNEL, zebra_flags, &p, &gate.sin.sin_addr,
1011 NULL, 0, VRF_DEFAULT, 0, 0, 0, 0, SAFI_UNICAST);
1013 rib_delete_ipv4 (ZEBRA_ROUTE_KERNEL, zebra_flags, &p,
1014 &gate.sin.sin_addr, 0, VRF_DEFAULT, SAFI_UNICAST);
1017 if (dest.sa.sa_family == AF_INET6)
1019 /* One day we might have a debug section here like one in the
1020 * IPv4 case above. Just ignore own messages at the moment.
1022 if (rtm->rtm_type != RTM_GET && rtm->rtm_pid == pid)
1024 struct prefix_ipv6 p;
1025 ifindex_t ifindex = 0;
1027 p.family = AF_INET6;
1028 p.prefix = dest.sin6.sin6_addr;
1029 if (flags & RTF_HOST)
1030 p.prefixlen = IPV6_MAX_PREFIXLEN;
1032 p.prefixlen = ip6_masklen (mask.sin6.sin6_addr);
1035 if (IN6_IS_ADDR_LINKLOCAL (&gate.sin6.sin6_addr))
1037 ifindex = IN6_LINKLOCAL_IFINDEX (gate.sin6.sin6_addr);
1038 SET_IN6_LINKLOCAL_IFINDEX (gate.sin6.sin6_addr, 0);
1042 /* CHANGE: delete the old prefix, we have no further information
1043 * to specify the route really
1045 if (rtm->rtm_type == RTM_CHANGE)
1046 rib_delete_ipv6 (ZEBRA_ROUTE_KERNEL, zebra_flags, &p,
1047 NULL, 0, VRF_DEFAULT, SAFI_UNICAST);
1049 if (rtm->rtm_type == RTM_GET
1050 || rtm->rtm_type == RTM_ADD
1051 || rtm->rtm_type == RTM_CHANGE)
1052 rib_add_ipv6 (ZEBRA_ROUTE_KERNEL, zebra_flags, &p, &gate.sin6.sin6_addr,
1053 ifindex, VRF_DEFAULT, RT_TABLE_MAIN, 0, 0, 0, SAFI_UNICAST);
1055 rib_delete_ipv6 (ZEBRA_ROUTE_KERNEL, zebra_flags, &p,
1056 &gate.sin6.sin6_addr, ifindex,
1057 VRF_DEFAULT, SAFI_UNICAST);
1059 #endif /* HAVE_IPV6 */
1062 /* Interface function for the kernel routing table updates. Support
1063 * for RTM_CHANGE will be needed.
1064 * Exported only for rt_socket.c
1067 rtm_write (int message,
1068 union sockunion *dest,
1069 union sockunion *mask,
1070 union sockunion *gate,
1077 struct interface *ifp;
1079 /* Sequencial number of routing message. */
1080 static int msg_seq = 0;
1082 /* Struct of rt_msghdr and buffer for storing socket's data. */
1085 struct rt_msghdr rtm;
1089 if (routing_sock < 0)
1090 return ZEBRA_ERR_EPERM;
1092 /* Clear and set rt_msghdr values */
1093 memset (&msg, 0, sizeof (struct rt_msghdr));
1094 msg.rtm.rtm_version = RTM_VERSION;
1095 msg.rtm.rtm_type = message;
1096 msg.rtm.rtm_seq = msg_seq++;
1097 msg.rtm.rtm_addrs = RTA_DST;
1098 msg.rtm.rtm_addrs |= RTA_GATEWAY;
1099 msg.rtm.rtm_flags = RTF_UP;
1100 msg.rtm.rtm_index = index;
1104 msg.rtm.rtm_rmx.rmx_hopcount = metric;
1105 msg.rtm.rtm_inits |= RTV_HOPCOUNT;
1108 ifp = if_lookup_by_index (index);
1110 if (gate && (message == RTM_ADD || message == RTM_CHANGE))
1111 msg.rtm.rtm_flags |= RTF_GATEWAY;
1113 /* When RTF_CLONING is unavailable on BSD, should we set some
1114 * other flag instead?
1117 if (! gate && (message == RTM_ADD || message == RTM_CHANGE) && ifp &&
1118 (ifp->flags & IFF_POINTOPOINT) == 0)
1119 msg.rtm.rtm_flags |= RTF_CLONING;
1120 #endif /* RTF_CLONING */
1122 /* If no protocol specific gateway is specified, use link
1123 address for gateway. */
1128 char dest_buf[INET_ADDRSTRLEN] = "NULL", mask_buf[INET_ADDRSTRLEN] = "255.255.255.255";
1130 inet_ntop (AF_INET, &dest->sin.sin_addr, dest_buf, INET_ADDRSTRLEN);
1132 inet_ntop (AF_INET, &mask->sin.sin_addr, mask_buf, INET_ADDRSTRLEN);
1133 zlog_warn ("%s: %s/%s: gate == NULL and no gateway found for ifindex %d",
1134 __func__, dest_buf, mask_buf, index);
1137 gate = (union sockunion *) &((struct zebra_if *)ifp->info)->sdl;
1141 msg.rtm.rtm_addrs |= RTA_NETMASK;
1142 else if (message == RTM_ADD || message == RTM_CHANGE)
1143 msg.rtm.rtm_flags |= RTF_HOST;
1145 /* Tagging route with flags */
1146 msg.rtm.rtm_flags |= (RTF_PROTO1);
1148 /* Additional flags. */
1149 if (zebra_flags & ZEBRA_FLAG_BLACKHOLE)
1150 msg.rtm.rtm_flags |= RTF_BLACKHOLE;
1151 if (zebra_flags & ZEBRA_FLAG_REJECT)
1152 msg.rtm.rtm_flags |= RTF_REJECT;
1155 #define SOCKADDRSET(X,R) \
1156 if (msg.rtm.rtm_addrs & (R)) \
1158 int len = SAROUNDUP (X); \
1159 memcpy (pnt, (caddr_t)(X), len); \
1163 pnt = (caddr_t) msg.buf;
1165 /* Write each socket data into rtm message buffer */
1166 SOCKADDRSET (dest, RTA_DST);
1167 SOCKADDRSET (gate, RTA_GATEWAY);
1168 SOCKADDRSET (mask, RTA_NETMASK);
1170 msg.rtm.rtm_msglen = pnt - (caddr_t) &msg;
1172 ret = write (routing_sock, &msg, msg.rtm.rtm_msglen);
1174 if (ret != msg.rtm.rtm_msglen)
1176 if (errno == EEXIST)
1177 return ZEBRA_ERR_RTEXIST;
1178 if (errno == ENETUNREACH)
1179 return ZEBRA_ERR_RTUNREACH;
1181 return ZEBRA_ERR_RTNOEXIST;
1183 zlog_warn ("%s: write : %s (%d)", __func__, safe_strerror (errno), errno);
1184 return ZEBRA_ERR_KERNEL;
1186 return ZEBRA_ERR_NOERROR;
1191 #include "zebra/zserv.h"
1193 /* For debug purpose. */
1195 rtmsg_debug (struct rt_msghdr *rtm)
1197 zlog_debug ("Kernel: Len: %d Type: %s", rtm->rtm_msglen, lookup (rtm_type_str, rtm->rtm_type));
1198 rtm_flag_dump (rtm->rtm_flags);
1199 zlog_debug ("Kernel: message seq %d", rtm->rtm_seq);
1200 zlog_debug ("Kernel: pid %lld, rtm_addrs 0x%x",
1201 (long long)rtm->rtm_pid, rtm->rtm_addrs);
1204 /* This is pretty gross, better suggestions welcome -- mhandler */
1207 #define RTAX_MAX RTA_NUMBITS
1210 #endif /* RTA_NUMBITS */
1211 #endif /* RTAX_MAX */
1213 /* Kernel routing table and interface updates via routing socket. */
1215 kernel_read (struct thread *thread)
1219 struct rt_msghdr *rtm;
1222 * This must be big enough for any message the kernel might send.
1223 * Rather than determining how many sockaddrs of what size might be
1224 * in each particular message, just use RTAX_MAX of sockaddr_storage
1225 * for each. Note that the sockaddrs must be after each message
1226 * definition, or rather after whichever happens to be the largest,
1227 * since the buffer needs to be big enough for a message and the
1228 * sockaddrs together.
1232 /* Routing information. */
1235 struct rt_msghdr rtm;
1236 struct sockaddr_storage addr[RTAX_MAX];
1239 /* Interface information. */
1242 struct if_msghdr ifm;
1243 struct sockaddr_storage addr[RTAX_MAX];
1246 /* Interface address information. */
1249 struct ifa_msghdr ifa;
1250 struct sockaddr_storage addr[RTAX_MAX];
1253 #ifdef RTM_IFANNOUNCE
1254 /* Interface arrival/departure */
1257 struct if_announcemsghdr ifan;
1258 struct sockaddr_storage addr[RTAX_MAX];
1260 #endif /* RTM_IFANNOUNCE */
1264 /* Fetch routing socket. */
1265 sock = THREAD_FD (thread);
1267 nbytes= read (sock, &buf, sizeof buf);
1271 if (nbytes < 0 && errno != EWOULDBLOCK && errno != EAGAIN)
1272 zlog_warn ("routing socket error: %s", safe_strerror (errno));
1276 thread_add_read (zebrad.master, kernel_read, NULL, sock);
1278 if (IS_ZEBRA_DEBUG_KERNEL)
1279 rtmsg_debug (&buf.r.rtm);
1284 * Ensure that we didn't drop any data, so that processing routines
1285 * can assume they have the whole message.
1287 if (rtm->rtm_msglen != nbytes)
1289 zlog_warn ("kernel_read: rtm->rtm_msglen %d, nbytes %d, type %d\n",
1290 rtm->rtm_msglen, nbytes, rtm->rtm_type);
1294 switch (rtm->rtm_type)
1302 ifm_read (&buf.im.ifm);
1306 ifam_read (&buf.ia.ifa);
1308 #ifdef RTM_IFANNOUNCE
1309 case RTM_IFANNOUNCE:
1310 ifan_read (&buf.ian.ifan);
1312 #endif /* RTM_IFANNOUNCE */
1314 if (IS_ZEBRA_DEBUG_KERNEL)
1315 zlog_debug("Unprocessed RTM_type: %d", rtm->rtm_type);
1321 /* Make routing socket. */
1323 routing_socket (struct zebra_vrf *zvrf)
1325 if (zvrf->vrf_id != VRF_DEFAULT)
1328 if ( zserv_privs.change (ZPRIVS_RAISE) )
1329 zlog_err ("routing_socket: Can't raise privileges");
1331 routing_sock = socket (AF_ROUTE, SOCK_RAW, 0);
1333 if (routing_sock < 0)
1335 if ( zserv_privs.change (ZPRIVS_LOWER) )
1336 zlog_err ("routing_socket: Can't lower privileges");
1337 zlog_warn ("Can't init kernel routing socket");
1341 /* XXX: Socket should be NONBLOCK, however as we currently
1342 * discard failed writes, this will lead to inconsistencies.
1343 * For now, socket must be blocking.
1345 /*if (fcntl (routing_sock, F_SETFL, O_NONBLOCK) < 0)
1346 zlog_warn ("Can't set O_NONBLOCK to routing socket");*/
1348 if ( zserv_privs.change (ZPRIVS_LOWER) )
1349 zlog_err ("routing_socket: Can't lower privileges");
1351 /* kernel_read needs rewrite. */
1352 thread_add_read (zebrad.master, kernel_read, NULL, routing_sock);
1355 /* Exported interface function. This function simply calls
1356 routing_socket (). */
1358 kernel_init (struct zebra_vrf *zvrf)
1360 routing_socket (zvrf);
1364 kernel_terminate (struct zebra_vrf *zvrf)