staticd: Fix crash because registering unknown vrf

With recent commit:

c1adc8f1d6795124df022a36388df173d217a34e staticd has started to crash
aproximately 1/10 of the tine in the static_vrf topotest

(gdb) bt
0  __pthread_kill_implementation (no_tid=0, signo=6, threadid=140400982256064) at ./nptl/pthread_kill.c:44
1  __pthread_kill_internal (signo=6, threadid=140400982256064) at ./nptl/pthread_kill.c:78
2  __GI___pthread_kill (threadid=140400982256064, signo=signo@entry=6) at ./nptl/pthread_kill.c:89
3  0x00007fb1a6442476 in __GI_raise (sig=6) at ../sysdeps/posix/raise.c:26
4  0x00007fb1a6950823 in core_handler (signo=6, siginfo=0x7ffd6d832ff0, context=0x7ffd6d832ec0) at lib/sigevent.c:268
5  <signal handler called>
6  __pthread_kill_implementation (no_tid=0, signo=6, threadid=140400982256064) at ./nptl/pthread_kill.c:44
7  __pthread_kill_internal (signo=6, threadid=140400982256064) at ./nptl/pthread_kill.c:78
8  __GI___pthread_kill (threadid=140400982256064, signo=signo@entry=6) at ./nptl/pthread_kill.c:89
9  0x00007fb1a6442476 in __GI_raise (sig=sig@entry=6) at ../sysdeps/posix/raise.c:26
10 0x00007fb1a64287f3 in __GI_abort () at ./stdlib/abort.c:79
11 0x00007fb1a699a422 in _zlog_assert_failed (xref=0x55f7dfd3dac0 <_xref.117>,
   extra=0x55f7dfd30c30 "BUG: NH %pFX registered but not in hashtable") at lib/zlog.c:789
12 0x000055f7dfd1201f in static_zebra_nht_register (nh=0x55f7fd2ecd80, reg=true) at staticd/static_zebra.c:333
13 0x000055f7dfd29c9d in static_install_nexthop (nh=0x55f7fd2ecd80) at staticd/static_routes.c:299
14 0x000055f7dfd2a126 in static_fixup_vrf (vrf=0x55f7fd2333a0, stable=0x55f7fd271030, afi=AFI_IP, safi=SAFI_UNICAST)
   at staticd/static_routes.c:441
15 0x000055f7dfd2a2be in static_fixup_vrf_ids (vrf=0x55f7fd2333a0) at staticd/static_routes.c:494
16 0x000055f7dfd15b53 in static_vrf_enable (vrf=0x55f7fd2333a0) at staticd/static_vrf.c:124
17 0x00007fb1a696ffa5 in vrf_enable (vrf=0x55f7fd2333a0) at lib/vrf.c:325
18 0x00007fb1a6991c87 in zclient_vrf_add (cmd=33, zclient=0x55f7fd29f740, length=76, vrf_id=8) at lib/zclient.c:2701
19 0x00007fb1a6996cba in zclient_read (thread=0x7ffd6d834230) at lib/zclient.c:4764
20 0x00007fb1a696bd9b in event_call (thread=0x7ffd6d834230) at lib/event.c:2019
21 0x00007fb1a68e1a3a in frr_run (master=0x55f7fd102e10) at lib/libfrr.c:1246
22 0x000055f7dfd1081e in main (argc=7, argv=0x7ffd6d834478, envp=0x7ffd6d8344b8) at staticd/static_main.c:193

Tracking this down, the crash is because the nh believes that is already
registered but lookup fails, causing this assert.  Looking at the code
static_fixup_vrf is changing the vrf_id.  I put a zlog_debug right
before the change of the nh vrf_id and noticed that the vrf id was
UNKNOWN.  So, the code is attempting to register into zebra the nexthop
with a vrf unknown( which will be ignored ).

Modify the code in the registration process to notice that the nh is
still UNKNOWN and as such nothing should be done.

Signed-off-by: Donald Sharp <sharpd@nvidia.com>

diff --git a/staticd/static_zebra.c b/staticd/static_zebra.c
index 552dd3ee1f74822e9d66e9d96766fa2993a740f5..3faeb3d37ab39fd570c10348b45529707358892a 100644 (file)
--- a/staticd/static_zebra.c
+++ b/staticd/static_zebra.c
@@ -323,6 +323,10 @@ void static_zebra_nht_register(struct static_nexthop *nh, bool reg)
 
        if (!static_zebra_nht_get_prefix(nh, &lookup.nh))
                return;
+
+       if (nh->nh_vrf_id == VRF_UNKNOWN)
+               return;
+
        lookup.nh_vrf_id = nh->nh_vrf_id;
        lookup.safi = si->safi;