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 along
17 * with this program; see the file COPYING; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
25 #include <net/if_types.h>
27 #include <netmpls/mpls.h>
32 #include "sockunion.h"
33 #include "connected.h"
35 #include "zebra_memory.h"
42 #include "lib_errors.h"
45 #include "zebra/interface.h"
46 #include "zebra/zserv.h"
47 #include "zebra/debug.h"
48 #include "zebra/kernel_socket.h"
49 #include "zebra/rib.h"
50 #include "zebra/zebra_errors.h"
51 #include "zebra/zebra_ptm.h"
53 extern struct zebra_privs_t zserv_privs
;
56 * Historically, the BSD routing socket has aligned data following a
57 * struct sockaddr to sizeof(long), which was 4 bytes on some
58 * platforms, and 8 bytes on others. NetBSD 6 changed the routing
59 * socket to align to sizeof(uint64_t), which is 8 bytes. OS X
60 * appears to align to sizeof(int), which is 4 bytes.
62 * Alignment of zero-sized sockaddrs is nonsensical, but historically
63 * BSD defines RT_ROUNDUP(0) to be the alignment interval (rather than
64 * 0). We follow this practice without questioning it, but it is a
65 * bug if quagga calls ROUNDUP with 0.
68 #define ROUNDUP_TYPE int
70 #define ROUNDUP_TYPE long
74 * Because of these varying conventions, the only sane approach is for
75 * the <net/route.h> header to define some flavor of ROUNDUP macro.
78 /* OS X (Xcode as of 2014-12) is known not to define RT_ROUNDUP */
79 #if defined(RT_ROUNDUP)
80 #define ROUNDUP(a) RT_ROUNDUP(a)
81 #endif /* defined(RT_ROUNDUP) */
84 /* Solaris has struct sockaddr_in[6] definitions at 16 / 32 bytes size,
85 * so the whole concept doesn't really apply. */
86 #define ROUNDUP(a) (a)
90 * If ROUNDUP has not yet been defined in terms of platform-provided
91 * defines, attempt to cope with heuristics.
96 * If you're porting to a platform that changed RT_ROUNDUP but doesn't
97 * have it in its headers, this will break rather obviously and you'll
98 * have to fix it here.
101 ((a) > 0 ? (1 + (((a)-1) | (sizeof(ROUNDUP_TYPE) - 1))) \
102 : sizeof(ROUNDUP_TYPE))
104 #endif /* defined(ROUNDUP) */
108 /* SAROUNDUP is the only thing we need, and SA_SIZE provides that */
109 #define SAROUNDUP(a) SA_SIZE(a)
112 * Given a pointer (sockaddr or void *), return the number of bytes
113 * taken up by the sockaddr and any padding needed for alignment.
115 #if defined(HAVE_STRUCT_SOCKADDR_SA_LEN)
116 #define SAROUNDUP(X) ROUNDUP(((struct sockaddr *)(X))->sa_len)
119 * One would hope all fixed-size structure definitions are aligned,
120 * but round them up nonetheless.
122 #define SAROUNDUP(X) \
123 (((struct sockaddr *)(X))->sa_family == AF_INET \
124 ? ROUNDUP(sizeof(struct sockaddr_in)) \
125 : (((struct sockaddr *)(X))->sa_family == AF_INET6 \
126 ? ROUNDUP(sizeof(struct sockaddr_in6)) \
127 : (((struct sockaddr *)(X))->sa_family == AF_LINK \
128 ? ROUNDUP(sizeof(struct sockaddr_dl)) \
129 : sizeof(struct sockaddr))))
130 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
132 #endif /* !SA_SIZE */
134 /* Routing socket message types. */
135 const struct message rtm_type_str
[] = {{RTM_ADD
, "RTM_ADD"},
136 {RTM_DELETE
, "RTM_DELETE"},
137 {RTM_CHANGE
, "RTM_CHANGE"},
138 {RTM_GET
, "RTM_GET"},
139 {RTM_LOSING
, "RTM_LOSING"},
140 {RTM_REDIRECT
, "RTM_REDIRECT"},
141 {RTM_MISS
, "RTM_MISS"},
142 {RTM_LOCK
, "RTM_LOCK"},
144 {RTM_OLDADD
, "RTM_OLDADD"},
145 #endif /* RTM_OLDADD */
147 {RTM_OLDDEL
, "RTM_OLDDEL"},
148 #endif /* RTM_OLDDEL */
149 {RTM_RESOLVE
, "RTM_RESOLVE"},
150 {RTM_NEWADDR
, "RTM_NEWADDR"},
151 {RTM_DELADDR
, "RTM_DELADDR"},
152 {RTM_IFINFO
, "RTM_IFINFO"},
154 {RTM_OIFINFO
, "RTM_OIFINFO"},
155 #endif /* RTM_OIFINFO */
157 {RTM_NEWMADDR
, "RTM_NEWMADDR"},
158 #endif /* RTM_NEWMADDR */
160 {RTM_DELMADDR
, "RTM_DELMADDR"},
161 #endif /* RTM_DELMADDR */
162 #ifdef RTM_IFANNOUNCE
163 {RTM_IFANNOUNCE
, "RTM_IFANNOUNCE"},
164 #endif /* RTM_IFANNOUNCE */
167 static const struct message rtm_flag_str
[] = {{RTF_UP
, "UP"},
168 {RTF_GATEWAY
, "GATEWAY"},
170 {RTF_REJECT
, "REJECT"},
171 {RTF_DYNAMIC
, "DYNAMIC"},
172 {RTF_MODIFIED
, "MODIFIED"},
176 #endif /* RTF_MASK */
178 {RTF_CLONING
, "CLONING"},
179 #endif /* RTF_CLONING */
181 {RTF_XRESOLVE
, "XRESOLVE"},
182 #endif /* RTF_XRESOLVE */
184 {RTF_LLINFO
, "LLINFO"},
185 #endif /* RTF_LLINFO */
186 {RTF_STATIC
, "STATIC"},
187 {RTF_BLACKHOLE
, "BLACKHOLE"},
189 {RTF_PRIVATE
, "PRIVATE"},
190 #endif /* RTF_PRIVATE */
191 {RTF_PROTO1
, "PROTO1"},
192 {RTF_PROTO2
, "PROTO2"},
194 {RTF_PRCLONING
, "PRCLONING"},
195 #endif /* RTF_PRCLONING */
197 {RTF_WASCLONED
, "WASCLONED"},
198 #endif /* RTF_WASCLONED */
200 {RTF_PROTO3
, "PROTO3"},
201 #endif /* RTF_PROTO3 */
203 {RTF_PINNED
, "PINNED"},
204 #endif /* RTF_PINNED */
206 {RTF_LOCAL
, "LOCAL"},
207 #endif /* RTF_LOCAL */
209 {RTF_BROADCAST
, "BROADCAST"},
210 #endif /* RTF_BROADCAST */
212 {RTF_MULTICAST
, "MULTICAST"},
213 #endif /* RTF_MULTICAST */
215 {RTF_MULTIRT
, "MULTIRT"},
216 #endif /* RTF_MULTIRT */
218 {RTF_SETSRC
, "SETSRC"},
219 #endif /* RTF_SETSRC */
222 /* Kernel routing update socket. */
223 int routing_sock
= -1;
225 /* Kernel dataplane routing update socket, used in the dataplane pthread
228 int dplane_routing_sock
= -1;
230 /* Yes I'm checking ugly routing socket behavior. */
233 size_t rta_get(caddr_t sap
, void *dest
, size_t destlen
);
234 size_t rta_getsdlname(caddr_t sap
, void *dest
, short *destlen
);
236 /* Supported address family check. */
237 static inline int af_check(int family
)
239 if (family
== AF_INET
)
241 if (family
== AF_INET6
)
246 size_t rta_get(caddr_t sap
, void *destp
, size_t destlen
)
248 struct sockaddr
*sa
= (struct sockaddr
*)sap
;
249 uint8_t *dest
= destp
;
250 size_t tlen
, copylen
;
252 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
253 copylen
= sa
->sa_len
;
254 tlen
= (copylen
== 0) ? sizeof(ROUNDUP_TYPE
) : ROUNDUP(copylen
);
255 #else /* !HAVE_STRUCT_SOCKADDR_SA_LEN */
256 copylen
= tlen
= SAROUNDUP(sap
);
257 #endif /* !HAVE_STRUCT_SOCKADDR_SA_LEN */
259 if (copylen
> 0 && dest
!= NULL
&& af_check(sa
->sa_family
)) {
260 if (copylen
> destlen
) {
261 zlog_warn("%s: destination buffer too small (%lu vs %lu)",
262 __func__
, copylen
, destlen
);
263 memcpy(dest
, sap
, destlen
);
265 memcpy(dest
, sap
, copylen
);
271 size_t rta_getsdlname(caddr_t sap
, void *destp
, short *destlen
)
273 struct sockaddr_dl
*sdl
= (struct sockaddr_dl
*)sap
;
274 struct sockaddr
*sa
= (struct sockaddr
*)sap
;
275 uint8_t *dest
= destp
;
276 size_t tlen
, copylen
;
278 copylen
= sdl
->sdl_nlen
;
279 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
280 tlen
= (sa
->sa_len
== 0) ? sizeof(ROUNDUP_TYPE
) : ROUNDUP(sa
->sa_len
);
281 #else /* !HAVE_STRUCT_SOCKADDR_SA_LEN */
282 tlen
= SAROUNDUP(sap
);
283 #endif /* !HAVE_STRUCT_SOCKADDR_SA_LEN */
285 if (copylen
> 0 && dest
!= NULL
&& sdl
->sdl_family
== AF_LINK
) {
286 if (copylen
> IFNAMSIZ
) {
287 zlog_warn("%s: destination buffer too small (%lu vs %d)",
288 __func__
, copylen
, IFNAMSIZ
);
289 memcpy(dest
, sdl
->sdl_data
, IFNAMSIZ
);
293 memcpy(dest
, sdl
->sdl_data
, copylen
);
303 /* Dump routing table flag for debug purpose. */
304 static void rtm_flag_dump(int flag
)
306 const struct message
*mes
;
307 static char buf
[BUFSIZ
];
310 for (mes
= rtm_flag_str
; mes
->key
!= 0; mes
++) {
311 if (mes
->key
& flag
) {
312 strlcat(buf
, mes
->str
, BUFSIZ
);
313 strlcat(buf
, " ", BUFSIZ
);
316 zlog_debug("Kernel: %s", buf
);
319 #ifdef RTM_IFANNOUNCE
320 /* Interface adding function */
321 static int ifan_read(struct if_announcemsghdr
*ifan
)
323 struct interface
*ifp
;
325 ifp
= if_lookup_by_index(ifan
->ifan_index
, VRF_DEFAULT
);
328 assert((ifp
->ifindex
== ifan
->ifan_index
)
329 || (ifp
->ifindex
== IFINDEX_INTERNAL
));
331 if ((ifp
== NULL
) || ((ifp
->ifindex
== IFINDEX_INTERNAL
)
332 && (ifan
->ifan_what
== IFAN_ARRIVAL
))) {
333 if (IS_ZEBRA_DEBUG_KERNEL
)
335 "%s: creating interface for ifindex %d, name %s",
336 __func__
, ifan
->ifan_index
, ifan
->ifan_name
);
338 /* Create Interface */
339 ifp
= if_get_by_name(ifan
->ifan_name
, VRF_DEFAULT
);
340 if_set_index(ifp
, ifan
->ifan_index
);
344 } else if (ifp
!= NULL
&& ifan
->ifan_what
== IFAN_DEPARTURE
)
345 if_delete_update(ifp
);
351 if (IS_ZEBRA_DEBUG_KERNEL
)
352 zlog_debug("%s: interface %s index %d", __func__
,
353 ifan
->ifan_name
, ifan
->ifan_index
);
357 #endif /* RTM_IFANNOUNCE */
359 #ifdef HAVE_BSD_IFI_LINK_STATE
360 /* BSD link detect translation */
361 static void bsd_linkdetect_translate(struct if_msghdr
*ifm
)
363 if ((ifm
->ifm_data
.ifi_link_state
>= LINK_STATE_UP
)
364 || (ifm
->ifm_data
.ifi_link_state
== LINK_STATE_UNKNOWN
))
365 SET_FLAG(ifm
->ifm_flags
, IFF_RUNNING
);
367 UNSET_FLAG(ifm
->ifm_flags
, IFF_RUNNING
);
369 #endif /* HAVE_BSD_IFI_LINK_STATE */
371 static enum zebra_link_type
sdl_to_zebra_link_type(unsigned int sdlt
)
375 return ZEBRA_LLT_ETHER
;
377 return ZEBRA_LLT_X25
;
379 return ZEBRA_LLT_FDDI
;
381 return ZEBRA_LLT_PPP
;
383 return ZEBRA_LLT_LOOPBACK
;
385 return ZEBRA_LLT_SLIP
;
387 return ZEBRA_LLT_ARCNET
;
389 return ZEBRA_LLT_ATM
;
391 return ZEBRA_LLT_LOCALTLK
;
393 return ZEBRA_LLT_HIPPI
;
396 return ZEBRA_LLT_IEEE1394
;
400 return ZEBRA_LLT_UNKNOWN
;
405 * Handle struct if_msghdr obtained from reading routing socket or
406 * sysctl (from interface_list). There may or may not be sockaddrs
407 * present after the header.
409 int ifm_read(struct if_msghdr
*ifm
)
411 struct interface
*ifp
= NULL
;
412 struct sockaddr_dl
*sdl
;
413 char ifname
[IFNAMSIZ
];
418 /* terminate ifname at head (for strnlen) and tail (for safety) */
419 ifname
[IFNAMSIZ
- 1] = '\0';
421 /* paranoia: sanity check structure */
422 if (ifm
->ifm_msglen
< sizeof(struct if_msghdr
)) {
423 flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR
,
424 "ifm_read: ifm->ifm_msglen %d too short\n",
430 * Check for a sockaddr_dl following the message. First, point to
431 * where a socakddr might be if one follows the message.
433 cp
= (void *)(ifm
+ 1);
437 * XXX This behavior should be narrowed to only the kernel versions
438 * for which the structures returned do not match the headers.
440 * if_msghdr_t on 64 bit kernels in Solaris 9 and earlier versions
441 * is 12 bytes larger than the 32 bit version.
443 if (((struct sockaddr
*)cp
)->sa_family
== AF_UNSPEC
)
447 /* Look up for RTA_IFP and skip others. */
448 for (maskbit
= 1; maskbit
; maskbit
<<= 1) {
449 if ((maskbit
& ifm
->ifm_addrs
) == 0)
451 if (maskbit
!= RTA_IFP
) {
452 cp
+= rta_get(cp
, NULL
, 0);
456 /* Save the pointer to the structure. */
457 sdl
= (struct sockaddr_dl
*)cp
;
458 cp
+= rta_getsdlname(cp
, ifname
, &ifnlen
);
461 if (IS_ZEBRA_DEBUG_KERNEL
)
462 zlog_debug("%s: sdl ifname %s", __func__
,
463 (ifnlen
? ifname
: "(nil)"));
466 * Look up on ifindex first, because ifindices are the primary handle
468 * interfaces across the user/kernel boundary, for most systems. (Some
469 * messages, such as up/down status changes on NetBSD, do not include a
472 if ((ifp
= if_lookup_by_index(ifm
->ifm_index
, VRF_DEFAULT
)) != NULL
) {
473 /* we have an ifp, verify that the name matches as some systems,
474 * eg Solaris, have a 1:many association of ifindex:ifname
475 * if they dont match, we dont have the correct ifp and should
476 * set it back to NULL to let next check do lookup by name
478 if (ifnlen
&& (strncmp(ifp
->name
, ifname
, IFNAMSIZ
) != 0)) {
479 if (IS_ZEBRA_DEBUG_KERNEL
)
481 "%s: ifp name %s doesn't match sdl name %s",
482 __func__
, ifp
->name
, ifname
);
488 * If we dont have an ifp, try looking up by name. Particularly as some
489 * systems (Solaris) have a 1:many mapping of ifindex:ifname - the
491 * is therefore our unique handle to that interface.
493 * Interfaces specified in the configuration file for which the ifindex
494 * has not been determined will have ifindex == IFINDEX_INTERNAL, and
496 * interfaces are found by this search, and then their ifindex values
500 if ((ifp
== NULL
) && ifnlen
)
501 ifp
= if_lookup_by_name(ifname
, VRF_DEFAULT
);
504 * If ifp still does not exist or has an invalid index
505 * (IFINDEX_INTERNAL),
506 * create or fill in an interface.
508 if ((ifp
== NULL
) || (ifp
->ifindex
== IFINDEX_INTERNAL
)) {
510 * To create or fill in an interface, a sockaddr_dl (via
511 * RTA_IFP) is required.
514 zlog_debug("Interface index %d (new) missing ifname\n",
519 #ifndef RTM_IFANNOUNCE
520 /* Down->Down interface should be ignored here.
521 * See further comment below.
523 if (!CHECK_FLAG(ifm
->ifm_flags
, IFF_UP
))
525 #endif /* !RTM_IFANNOUNCE */
528 /* Interface that zebra was not previously aware of, so
530 ifp
= if_create(ifname
, VRF_DEFAULT
);
531 if (IS_ZEBRA_DEBUG_KERNEL
)
532 zlog_debug("%s: creating ifp for ifindex %d",
533 __func__
, ifm
->ifm_index
);
536 if (IS_ZEBRA_DEBUG_KERNEL
)
538 "%s: updated/created ifp, ifname %s, ifindex %d",
539 __func__
, ifp
->name
, ifp
->ifindex
);
541 * Fill in newly created interface structure, or larval
542 * structure with ifindex IFINDEX_INTERNAL.
544 if_set_index(ifp
, ifm
->ifm_index
);
546 #ifdef HAVE_BSD_IFI_LINK_STATE /* translate BSD kernel msg for link-state */
547 bsd_linkdetect_translate(ifm
);
548 #endif /* HAVE_BSD_IFI_LINK_STATE */
550 if_flags_update(ifp
, ifm
->ifm_flags
);
551 #if defined(__bsdi__)
555 #endif /* __bsdi__ */
559 * XXX sockaddr_dl contents can be larger than the structure
560 * definition. There are 2 big families here:
561 * - BSD has sdl_len + sdl_data[16] + overruns sdl_data
562 * we MUST use sdl_len here or we'll truncate data.
563 * - Solaris has no sdl_len, but sdl_data[244]
564 * presumably, it's not going to run past that, so sizeof()
566 * a nonzero ifnlen from rta_getsdlname() means sdl is valid
568 ifp
->ll_type
= ZEBRA_LLT_UNKNOWN
;
569 ifp
->hw_addr_len
= 0;
571 #ifdef HAVE_STRUCT_SOCKADDR_DL_SDL_LEN
572 memcpy(&((struct zebra_if
*)ifp
->info
)->sdl
, sdl
,
575 memcpy(&((struct zebra_if
*)ifp
->info
)->sdl
, sdl
,
576 sizeof(struct sockaddr_dl
));
577 #endif /* HAVE_STRUCT_SOCKADDR_DL_SDL_LEN */
579 ifp
->ll_type
= sdl_to_zebra_link_type(sdl
->sdl_type
);
580 if (sdl
->sdl_alen
<= sizeof(ifp
->hw_addr
)) {
581 memcpy(ifp
->hw_addr
, LLADDR(sdl
),
583 ifp
->hw_addr_len
= sdl
->sdl_alen
;
590 * Interface structure exists. Adjust stored flags from
591 * notification. If interface has up->down or down->up
592 * transition, call state change routines (to adjust routes,
593 * notify routing daemons, etc.). (Other flag changes are stored
594 * but apparently do not trigger action.)
597 if (ifp
->ifindex
!= ifm
->ifm_index
) {
599 "%s: index mismatch, ifname %s, ifp index %d, "
601 __func__
, ifp
->name
, ifp
->ifindex
,
606 #ifdef HAVE_BSD_IFI_LINK_STATE /* translate BSD kernel msg for link-state */
607 bsd_linkdetect_translate(ifm
);
608 #endif /* HAVE_BSD_IFI_LINK_STATE */
610 /* update flags and handle operative->inoperative transition, if
612 if_flags_update(ifp
, ifm
->ifm_flags
);
614 #ifndef RTM_IFANNOUNCE
615 if (!if_is_up(ifp
)) {
616 /* No RTM_IFANNOUNCE on this platform, so we can never
617 * distinguish between ~IFF_UP and delete. We must
619 * it has been deleted.
620 * Eg, Solaris will not notify us of unplumb.
622 * XXX: Fixme - this should be runtime detected
623 * So that a binary compiled on a system with IFANNOUNCE
624 * will still behave correctly if run on a platform
627 if_delete_update(ifp
);
629 #endif /* RTM_IFANNOUNCE */
631 #if defined(__bsdi__)
635 #endif /* __bsdi__ */
640 #ifdef HAVE_NET_RT_IFLIST
641 ifp
->stats
= ifm
->ifm_data
;
642 #endif /* HAVE_NET_RT_IFLIST */
643 ifp
->speed
= ifm
->ifm_data
.ifi_baudrate
/ 1000000;
645 if (IS_ZEBRA_DEBUG_KERNEL
)
646 zlog_debug("%s: interface %s index %d", __func__
, ifp
->name
,
652 /* Address read from struct ifa_msghdr. */
653 static void ifam_read_mesg(struct ifa_msghdr
*ifm
, union sockunion
*addr
,
654 union sockunion
*mask
, union sockunion
*brd
,
655 char *ifname
, short *ifnlen
)
659 union sockunion gateway
;
662 pnt
= (caddr_t
)(ifm
+ 1);
663 end
= ((caddr_t
)ifm
) + ifm
->ifam_msglen
;
665 /* Be sure structure is cleared */
666 memset(mask
, 0, sizeof(union sockunion
));
667 memset(addr
, 0, sizeof(union sockunion
));
668 memset(brd
, 0, sizeof(union sockunion
));
669 memset(&dst
, 0, sizeof(union sockunion
));
670 memset(&gateway
, 0, sizeof(union sockunion
));
672 /* We fetch each socket variable into sockunion. */
673 for (maskbit
= 1; maskbit
; maskbit
<<= 1) {
674 if ((maskbit
& ifm
->ifam_addrs
) == 0)
679 pnt
+= rta_get(pnt
, &dst
, sizeof(dst
));
682 pnt
+= rta_get(pnt
, &gateway
, sizeof(gateway
));
685 pnt
+= rta_get(pnt
, mask
, sizeof(*mask
));
688 pnt
+= rta_getsdlname(pnt
, ifname
, ifnlen
);
691 pnt
+= rta_get(pnt
, addr
, sizeof(*addr
));
694 pnt
+= rta_get(pnt
, brd
, sizeof(*brd
));
698 pnt
+= rta_get(pnt
, NULL
, 0);
703 zlog_warn("%s: overflow detected (pnt:%p end:%p)",
709 if (IS_ZEBRA_DEBUG_KERNEL
) {
710 int family
= sockunion_family(addr
);
714 char buf
[4][INET6_ADDRSTRLEN
];
716 "%s: ifindex %d, ifname %s, ifam_addrs 0x%x, "
717 "ifam_flags 0x%x, addr %s/%d broad %s dst %s "
719 __func__
, ifm
->ifam_index
,
720 (ifnlen
? ifname
: "(nil)"), ifm
->ifam_addrs
,
722 inet_ntop(family
, &addr
->sin
.sin_addr
, buf
[0],
724 ip_masklen(mask
->sin
.sin_addr
),
725 inet_ntop(family
, &brd
->sin
.sin_addr
, buf
[1],
727 inet_ntop(family
, &dst
.sin
.sin_addr
, buf
[2],
729 inet_ntop(family
, &gateway
.sin
.sin_addr
, buf
[3],
733 zlog_debug("%s: ifindex %d, ifname %s, ifam_addrs 0x%x",
734 __func__
, ifm
->ifam_index
,
735 (ifnlen
? ifname
: "(nil)"),
741 /* Assert read up end point matches to end point */
742 pnt
= (caddr_t
)ROUNDUP((size_t)pnt
);
744 zlog_debug("ifam_read() doesn't read all socket data");
747 /* Interface's address information get. */
748 int ifam_read(struct ifa_msghdr
*ifam
)
750 struct interface
*ifp
= NULL
;
751 union sockunion addr
, mask
, brd
;
752 char ifname
[INTERFACE_NAMSIZ
];
757 ifname
[0] = ifname
[INTERFACE_NAMSIZ
- 1] = '\0';
759 /* Allocate and read address information. */
760 ifam_read_mesg(ifam
, &addr
, &mask
, &brd
, ifname
, &ifnlen
);
762 if ((ifp
= if_lookup_by_index(ifam
->ifam_index
, VRF_DEFAULT
)) == NULL
) {
763 flog_warn(EC_ZEBRA_UNKNOWN_INTERFACE
,
764 "%s: no interface for ifname %s, index %d", __func__
,
765 ifname
, ifam
->ifam_index
);
769 if (ifnlen
&& strncmp(ifp
->name
, ifname
, INTERFACE_NAMSIZ
))
772 /* N.B. The info in ifa_msghdr does not tell us whether the RTA_BRD
773 field contains a broadcast address or a peer address, so we are
775 rely upon the interface type. */
776 if (if_is_pointopoint(ifp
))
777 SET_FLAG(flags
, ZEBRA_IFA_PEER
);
780 /* it might seem cute to grab the interface metric here, however
781 * we're processing an address update message, and so some systems
782 * (e.g. FBSD) dont bother to fill in ifam_metric. Disabled, but left
783 * in deliberately, as comment.
785 ifp
->metric
= ifam
->ifam_metric
;
788 /* Add connected address. */
789 switch (sockunion_family(&addr
)) {
791 if (ifam
->ifam_type
== RTM_NEWADDR
)
792 connected_add_ipv4(ifp
, flags
, &addr
.sin
.sin_addr
,
793 ip_masklen(mask
.sin
.sin_addr
),
795 (isalias
? ifname
: NULL
));
797 connected_delete_ipv4(ifp
, flags
, &addr
.sin
.sin_addr
,
798 ip_masklen(mask
.sin
.sin_addr
),
802 /* Unset interface index from link-local address when IPv6 stack
804 if (IN6_IS_ADDR_LINKLOCAL(&addr
.sin6
.sin6_addr
)) {
805 SET_IN6_LINKLOCAL_IFINDEX(addr
.sin6
.sin6_addr
, 0);
808 if (ifam
->ifam_type
== RTM_NEWADDR
)
809 connected_add_ipv6(ifp
, flags
, &addr
.sin6
.sin6_addr
,
811 ip6_masklen(mask
.sin6
.sin6_addr
),
812 (isalias
? ifname
: NULL
));
814 connected_delete_ipv6(ifp
, &addr
.sin6
.sin6_addr
, NULL
,
815 ip6_masklen(mask
.sin6
.sin6_addr
));
818 /* Unsupported family silently ignore... */
822 /* Check interface flag for implicit up of the interface. */
826 /* In addition to lacking IFANNOUNCE, on SUNOS IFF_UP is strange.
827 * See comments for SUNOS_5 in interface.c::if_flags_mangle.
829 * Here we take care of case where the real IFF_UP was previously
830 * unset (as kept in struct zebra_if.primary_state) and the mangled
831 * IFF_UP (ie IFF_UP set || listcount(connected) has now transitioned
832 * to unset due to the lost non-primary address having DELADDR'd.
834 * we must delete the interface, because in between here and next
835 * event for this interface-name the administrator could unplumb
836 * and replumb the interface.
839 if_delete_update(ifp
);
845 /* Interface function for reading kernel routing table information. */
846 static int rtm_read_mesg(struct rt_msghdr
*rtm
, union sockunion
*dest
,
847 union sockunion
*mask
, union sockunion
*gate
,
848 char *ifname
, short *ifnlen
)
853 /* Pnt points out socket data start point. */
854 pnt
= (caddr_t
)(rtm
+ 1);
855 end
= ((caddr_t
)rtm
) + rtm
->rtm_msglen
;
857 /* rt_msghdr version check. */
858 if (rtm
->rtm_version
!= RTM_VERSION
)
859 flog_warn(EC_ZEBRA_RTM_VERSION_MISMATCH
,
860 "Routing message version different %d should be %d."
861 "This may cause problem\n",
862 rtm
->rtm_version
, RTM_VERSION
);
864 /* Be sure structure is cleared */
865 memset(dest
, 0, sizeof(union sockunion
));
866 memset(gate
, 0, sizeof(union sockunion
));
867 memset(mask
, 0, sizeof(union sockunion
));
869 /* We fetch each socket variable into sockunion. */
870 /* We fetch each socket variable into sockunion. */
871 for (maskbit
= 1; maskbit
; maskbit
<<= 1) {
872 if ((maskbit
& rtm
->rtm_addrs
) == 0)
877 pnt
+= rta_get(pnt
, dest
, sizeof(*dest
));
880 pnt
+= rta_get(pnt
, gate
, sizeof(*gate
));
883 pnt
+= rta_get(pnt
, mask
, sizeof(*mask
));
886 pnt
+= rta_getsdlname(pnt
, ifname
, ifnlen
);
890 pnt
+= rta_get(pnt
, NULL
, 0);
895 zlog_warn("%s: overflow detected (pnt:%p end:%p)",
901 /* If there is netmask information set it's family same as
903 if (rtm
->rtm_addrs
& RTA_NETMASK
)
904 mask
->sa
.sa_family
= dest
->sa
.sa_family
;
906 /* Assert read up to the end of pointer. */
908 zlog_debug("rtm_read() doesn't read all socket data.");
910 return rtm
->rtm_flags
;
913 void rtm_read(struct rt_msghdr
*rtm
)
917 union sockunion dest
, mask
, gate
;
918 char ifname
[INTERFACE_NAMSIZ
+ 1];
922 ifindex_t ifindex
= 0;
927 /* Read destination and netmask and gateway from rtm message
929 flags
= rtm_read_mesg(rtm
, &dest
, &mask
, &gate
, ifname
, &ifnlen
);
930 if (!(flags
& RTF_DONE
))
932 if (IS_ZEBRA_DEBUG_KERNEL
)
933 zlog_debug("%s: got rtm of type %d (%s)", __func__
,
935 lookup_msg(rtm_type_str
, rtm
->rtm_type
, NULL
));
937 #ifdef RTF_CLONED /*bsdi, netbsd 1.6*/
938 if (flags
& RTF_CLONED
)
941 #ifdef RTF_WASCLONED /*freebsd*/
942 if (flags
& RTF_WASCLONED
)
946 if ((rtm
->rtm_type
== RTM_ADD
|| rtm
->rtm_type
== RTM_CHANGE
)
947 && !(flags
& RTF_UP
))
950 /* This is connected route. */
951 if (!(flags
& RTF_GATEWAY
))
954 if (flags
& RTF_PROTO1
)
955 SET_FLAG(zebra_flags
, ZEBRA_FLAG_SELFROUTE
);
957 memset(&nh
, 0, sizeof(nh
));
959 nh
.vrf_id
= VRF_DEFAULT
;
960 /* This is a reject or blackhole route */
961 if (flags
& RTF_REJECT
) {
962 nh
.type
= NEXTHOP_TYPE_BLACKHOLE
;
963 nh
.bh_type
= BLACKHOLE_REJECT
;
964 } else if (flags
& RTF_BLACKHOLE
) {
965 nh
.type
= NEXTHOP_TYPE_BLACKHOLE
;
966 nh
.bh_type
= BLACKHOLE_NULL
;
970 * Ignore our own messages.
972 if (rtm
->rtm_type
!= RTM_GET
&& rtm
->rtm_pid
== pid
)
975 if (dest
.sa
.sa_family
== AF_INET
) {
978 p
.u
.prefix4
= dest
.sin
.sin_addr
;
979 if (flags
& RTF_HOST
)
980 p
.prefixlen
= IPV4_MAX_PREFIXLEN
;
982 p
.prefixlen
= ip_masklen(mask
.sin
.sin_addr
);
985 nh
.type
= NEXTHOP_TYPE_IPV4
;
986 nh
.gate
.ipv4
= gate
.sin
.sin_addr
;
988 } else if (dest
.sa
.sa_family
== AF_INET6
) {
991 p
.u
.prefix6
= dest
.sin6
.sin6_addr
;
992 if (flags
& RTF_HOST
)
993 p
.prefixlen
= IPV6_MAX_PREFIXLEN
;
995 p
.prefixlen
= ip6_masklen(mask
.sin6
.sin6_addr
);
998 if (IN6_IS_ADDR_LINKLOCAL(&gate
.sin6
.sin6_addr
)) {
999 ifindex
= IN6_LINKLOCAL_IFINDEX(gate
.sin6
.sin6_addr
);
1000 SET_IN6_LINKLOCAL_IFINDEX(gate
.sin6
.sin6_addr
, 0);
1005 nh
.type
= ifindex
? NEXTHOP_TYPE_IPV6_IFINDEX
1006 : NEXTHOP_TYPE_IPV6
;
1007 nh
.gate
.ipv6
= gate
.sin6
.sin6_addr
;
1008 nh
.ifindex
= ifindex
;
1014 * CHANGE: delete the old prefix, we have no further information
1015 * to specify the route really
1017 if (rtm
->rtm_type
== RTM_CHANGE
)
1018 rib_delete(afi
, SAFI_UNICAST
, VRF_DEFAULT
, ZEBRA_ROUTE_KERNEL
,
1019 0, zebra_flags
, &p
, NULL
, NULL
, 0, 0, 0, true);
1020 if (rtm
->rtm_type
== RTM_GET
|| rtm
->rtm_type
== RTM_ADD
1021 || rtm
->rtm_type
== RTM_CHANGE
)
1022 rib_add(afi
, SAFI_UNICAST
, VRF_DEFAULT
, ZEBRA_ROUTE_KERNEL
, 0,
1023 zebra_flags
, &p
, NULL
, &nh
, 0, 0, 0, 0, 0);
1025 rib_delete(afi
, SAFI_UNICAST
, VRF_DEFAULT
, ZEBRA_ROUTE_KERNEL
,
1026 0, zebra_flags
, &p
, NULL
, &nh
, 0, 0, 0, true);
1029 /* Interface function for the kernel routing table updates. Support
1030 * for RTM_CHANGE will be needed.
1031 * Exported only for rt_socket.c
1033 int rtm_write(int message
, union sockunion
*dest
, union sockunion
*mask
,
1034 union sockunion
*gate
, union sockunion
*mpls
, unsigned int index
,
1035 enum blackhole_type bh_type
, int metric
)
1039 struct interface
*ifp
;
1041 /* Sequencial number of routing message. */
1042 static int msg_seq
= 0;
1044 /* Struct of rt_msghdr and buffer for storing socket's data. */
1046 struct rt_msghdr rtm
;
1050 if (dplane_routing_sock
< 0)
1051 return ZEBRA_ERR_EPERM
;
1053 /* Clear and set rt_msghdr values */
1054 memset(&msg
, 0, sizeof(struct rt_msghdr
));
1055 msg
.rtm
.rtm_version
= RTM_VERSION
;
1056 msg
.rtm
.rtm_type
= message
;
1057 msg
.rtm
.rtm_seq
= msg_seq
++;
1058 msg
.rtm
.rtm_addrs
= RTA_DST
;
1059 msg
.rtm
.rtm_addrs
|= RTA_GATEWAY
;
1060 msg
.rtm
.rtm_flags
= RTF_UP
;
1062 msg
.rtm
.rtm_flags
|= RTF_MPATH
;
1063 msg
.rtm
.rtm_fmask
= RTF_MPLS
;
1065 msg
.rtm
.rtm_index
= index
;
1068 msg
.rtm
.rtm_rmx
.rmx_hopcount
= metric
;
1069 msg
.rtm
.rtm_inits
|= RTV_HOPCOUNT
;
1072 ifp
= if_lookup_by_index(index
, VRF_DEFAULT
);
1074 if (gate
&& (message
== RTM_ADD
|| message
== RTM_CHANGE
))
1075 msg
.rtm
.rtm_flags
|= RTF_GATEWAY
;
1077 /* When RTF_CLONING is unavailable on BSD, should we set some
1078 * other flag instead?
1081 if (!gate
&& (message
== RTM_ADD
|| message
== RTM_CHANGE
) && ifp
1082 && (ifp
->flags
& IFF_POINTOPOINT
) == 0)
1083 msg
.rtm
.rtm_flags
|= RTF_CLONING
;
1084 #endif /* RTF_CLONING */
1086 /* If no protocol specific gateway is specified, use link
1087 address for gateway. */
1090 char dest_buf
[INET_ADDRSTRLEN
] = "NULL",
1091 mask_buf
[INET_ADDRSTRLEN
] = "255.255.255.255";
1093 inet_ntop(AF_INET
, &dest
->sin
.sin_addr
,
1094 dest_buf
, INET_ADDRSTRLEN
);
1096 inet_ntop(AF_INET
, &mask
->sin
.sin_addr
,
1097 mask_buf
, INET_ADDRSTRLEN
);
1099 EC_ZEBRA_RTM_NO_GATEWAY
,
1100 "%s: %s/%s: gate == NULL and no gateway found for ifindex %d",
1101 __func__
, dest_buf
, mask_buf
, index
);
1104 gate
= (union sockunion
*)&((struct zebra_if
*)ifp
->info
)->sdl
;
1108 msg
.rtm
.rtm_addrs
|= RTA_NETMASK
;
1109 else if (message
== RTM_ADD
|| message
== RTM_CHANGE
)
1110 msg
.rtm
.rtm_flags
|= RTF_HOST
;
1114 msg
.rtm
.rtm_addrs
|= RTA_SRC
;
1115 msg
.rtm
.rtm_flags
|= RTF_MPLS
;
1117 if (mpls
->smpls
.smpls_label
1118 != htonl(MPLS_LABEL_IMPLICIT_NULL
<< MPLS_LABEL_OFFSET
))
1119 msg
.rtm
.rtm_mpls
= MPLS_OP_PUSH
;
1123 /* Tagging route with flags */
1124 msg
.rtm
.rtm_flags
|= (RTF_PROTO1
);
1127 case BLACKHOLE_UNSPEC
:
1129 case BLACKHOLE_REJECT
:
1130 msg
.rtm
.rtm_flags
|= RTF_REJECT
;
1133 msg
.rtm
.rtm_flags
|= RTF_BLACKHOLE
;
1138 #define SOCKADDRSET(X, R) \
1139 if (msg.rtm.rtm_addrs & (R)) { \
1140 int len = SAROUNDUP(X); \
1141 memcpy(pnt, (caddr_t)(X), len); \
1145 pnt
= (caddr_t
)msg
.buf
;
1147 /* Write each socket data into rtm message buffer */
1148 SOCKADDRSET(dest
, RTA_DST
);
1149 SOCKADDRSET(gate
, RTA_GATEWAY
);
1150 SOCKADDRSET(mask
, RTA_NETMASK
);
1152 SOCKADDRSET(mpls
, RTA_SRC
);
1155 msg
.rtm
.rtm_msglen
= pnt
- (caddr_t
)&msg
;
1157 ret
= write(dplane_routing_sock
, &msg
, msg
.rtm
.rtm_msglen
);
1159 if (ret
!= msg
.rtm
.rtm_msglen
) {
1160 if (errno
== EEXIST
)
1161 return ZEBRA_ERR_RTEXIST
;
1162 if (errno
== ENETUNREACH
)
1163 return ZEBRA_ERR_RTUNREACH
;
1165 return ZEBRA_ERR_RTNOEXIST
;
1167 flog_err_sys(EC_LIB_SOCKET
, "%s: write : %s (%d)", __func__
,
1168 safe_strerror(errno
), errno
);
1169 return ZEBRA_ERR_KERNEL
;
1171 return ZEBRA_ERR_NOERROR
;
1176 #include "zebra/zserv.h"
1178 /* For debug purpose. */
1179 static void rtmsg_debug(struct rt_msghdr
*rtm
)
1181 zlog_debug("Kernel: Len: %d Type: %s", rtm
->rtm_msglen
,
1182 lookup_msg(rtm_type_str
, rtm
->rtm_type
, NULL
));
1183 rtm_flag_dump(rtm
->rtm_flags
);
1184 zlog_debug("Kernel: message seq %d", rtm
->rtm_seq
);
1185 zlog_debug("Kernel: pid %lld, rtm_addrs 0x%x", (long long)rtm
->rtm_pid
,
1189 /* This is pretty gross, better suggestions welcome -- mhandler */
1192 #define RTAX_MAX RTA_NUMBITS
1195 #endif /* RTA_NUMBITS */
1196 #endif /* RTAX_MAX */
1198 /* Kernel routing table and interface updates via routing socket. */
1199 static int kernel_read(struct thread
*thread
)
1203 struct rt_msghdr
*rtm
;
1206 * This must be big enough for any message the kernel might send.
1207 * Rather than determining how many sockaddrs of what size might be
1208 * in each particular message, just use RTAX_MAX of sockaddr_storage
1209 * for each. Note that the sockaddrs must be after each message
1210 * definition, or rather after whichever happens to be the largest,
1211 * since the buffer needs to be big enough for a message and the
1212 * sockaddrs together.
1215 /* Routing information. */
1217 struct rt_msghdr rtm
;
1218 struct sockaddr_storage addr
[RTAX_MAX
];
1221 /* Interface information. */
1223 struct if_msghdr ifm
;
1224 struct sockaddr_storage addr
[RTAX_MAX
];
1227 /* Interface address information. */
1229 struct ifa_msghdr ifa
;
1230 struct sockaddr_storage addr
[RTAX_MAX
];
1233 #ifdef RTM_IFANNOUNCE
1234 /* Interface arrival/departure */
1236 struct if_announcemsghdr ifan
;
1237 struct sockaddr_storage addr
[RTAX_MAX
];
1239 #endif /* RTM_IFANNOUNCE */
1243 /* Fetch routing socket. */
1244 sock
= THREAD_FD(thread
);
1246 nbytes
= read(sock
, &buf
, sizeof buf
);
1249 if (nbytes
< 0 && errno
!= EWOULDBLOCK
&& errno
!= EAGAIN
)
1250 flog_err_sys(EC_LIB_SOCKET
, "routing socket error: %s",
1251 safe_strerror(errno
));
1255 thread_add_read(zebrad
.master
, kernel_read
, NULL
, sock
, NULL
);
1257 if (IS_ZEBRA_DEBUG_KERNEL
)
1258 rtmsg_debug(&buf
.r
.rtm
);
1263 * Ensure that we didn't drop any data, so that processing routines
1264 * can assume they have the whole message.
1266 if (rtm
->rtm_msglen
!= nbytes
) {
1268 "kernel_read: rtm->rtm_msglen %d, nbytes %d, type %d\n",
1269 rtm
->rtm_msglen
, nbytes
, rtm
->rtm_type
);
1273 switch (rtm
->rtm_type
) {
1280 ifm_read(&buf
.im
.ifm
);
1284 ifam_read(&buf
.ia
.ifa
);
1286 #ifdef RTM_IFANNOUNCE
1287 case RTM_IFANNOUNCE
:
1288 ifan_read(&buf
.ian
.ifan
);
1290 #endif /* RTM_IFANNOUNCE */
1292 if (IS_ZEBRA_DEBUG_KERNEL
)
1293 zlog_debug("Unprocessed RTM_type: %d", rtm
->rtm_type
);
1299 /* Make routing socket. */
1300 static void routing_socket(struct zebra_ns
*zns
)
1302 frr_elevate_privs(&zserv_privs
) {
1303 routing_sock
= ns_socket(AF_ROUTE
, SOCK_RAW
, 0, zns
->ns_id
);
1305 dplane_routing_sock
=
1306 ns_socket(AF_ROUTE
, SOCK_RAW
, 0, zns
->ns_id
);
1309 if (routing_sock
< 0) {
1310 flog_err_sys(EC_LIB_SOCKET
, "Can't init kernel routing socket");
1314 if (dplane_routing_sock
< 0) {
1315 flog_err_sys(EC_LIB_SOCKET
,
1316 "Can't init kernel dataplane routing socket");
1320 /* XXX: Socket should be NONBLOCK, however as we currently
1321 * discard failed writes, this will lead to inconsistencies.
1322 * For now, socket must be blocking.
1324 /*if (fcntl (routing_sock, F_SETFL, O_NONBLOCK) < 0)
1325 zlog_warn ("Can't set O_NONBLOCK to routing socket");*/
1327 /* kernel_read needs rewrite. */
1328 thread_add_read(zebrad
.master
, kernel_read
, NULL
, routing_sock
, NULL
);
1331 /* Exported interface function. This function simply calls
1332 routing_socket (). */
1333 void kernel_init(struct zebra_ns
*zns
)
1335 routing_socket(zns
);
1338 void kernel_terminate(struct zebra_ns
*zns
, bool complete
)
1343 #endif /* !HAVE_NETLINK */