Skip to contentSkip to conversation

ipv6: fib6_select_path can not use out path for nexthop objects

PATCH Superseded David Ahern 2 participants 2 messages Jul 6, 2020

Subsystems (4)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

David Ahern

author maintainer
#0 lore raw

PATCH 1/1 ipv6: fib6_select_path can not use out path for nexthop objects

Brian reported a crash in IPv6 code when using rpfilter with a setup running FRR and external nexthop objects. The root cause of the crash is fib6_select_path setting fib6_nh in the result to NULL because of an improper check for nexthop objects (fib6_nh in fib6_info is not set for this case).

More specifically, rpfilter invokes ip6_route_lookup with flowi6_oif set causing fib6_select_path to be called with have_oif_match set. fib6_select_path has early check on have_oif_match and jumps to the out label which presumes a builtin fib6_nh. This path is invalid for nexthop objects; for external nexthops fib6_select_path needs to proceed and return after the call to nexthop_path_fib6_result. Update the check on have_oif_match to not bail on external nexthops.

Update selftests for this problem.

Fixes f88d8ea67fbd ("ipv6: Plumb support for nexthop object in a fib6_info")
---
 net/ipv6/route.c                            |  2 +-
 tools/testing/selftests/net/fib_nexthops.sh | 13 +++++++++++++
 2 files changed, 14 insertions(+), 1 deletion(-)
net/ipv6/route.c +1 -1
index 82cbb46a2a4f..6451ba313506 100644
@@ -431,7 +431,7 @@ void fib6_select_path(const struct net *net, struct fib6_result *res,
 	struct fib6_info *sibling, *next_sibling;
 	struct fib6_info *match = res->f6i;
 
-	if ((!match->fib6_nsiblings && !match->nh) || have_oif_match)
+	if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 		goto out;
 
 	/* We might have already computed the hash for ICMPv6 errors. In such
tools/testing/selftests/net/fib_nexthops.sh +13 -0
index dee567f7576a..f6d388dbd881 100755
@@ -747,6 +747,19 @@ ipv6_fcnal_runtime()
 	run_cmd "$IP nexthop add id 86 via 2001:db8:91::2 dev veth1"
 	run_cmd "$IP ro add 2001:db8:101::1/128 nhid 81"
 
+	# rpfilter and default route
+	$IP nexthop flush >/dev/null 2>&1
+	run_cmd "ip netns exec me ip6tables -t mangle -I PREROUTING 1 -m rpfilter --invert -j ACCEPT"
+	run_cmd "$IP nexthop add id 91 via 2001:db8:91::2 dev veth1"
+	run_cmd "$IP nexthop add id 92 via 2001:db8:92::2 dev veth3"
+	run_cmd "$IP nexthop add id 93 group 91/92"
+	run_cmd "$IP -6 ro add default nhid 91"
+	run_cmd "ip netns exec me ping -c1 -w1 2001:db8:101::1"
+	log_test $? 0 "Nexthop with default route and rpfilter"
+	run_cmd "$IP -6 ro replace default nhid 93"
+	run_cmd "ip netns exec me ping -c1 -w1 2001:db8:101::1"
+	log_test $? 0 "Nexthop with multipath default route and rpfilter"
+
 	# TO-DO:
 	# existing route with old nexthop; append route with new nexthop
 	# existing route with old nexthop; replace route with new
Signature · 1 line
2.17.1
#1 lore raw

net/ipv6/route.c void fib6_select_path(const struct net *net, struct fib6_result *res, · line 437

Quoted 13 lines · David Ahern
net/ipv6/route.c +1 -1
index 82cbb46a2a4f..6451ba313506 100644
@@ -431,7 +431,7 @@ void fib6_select_path(const struct net *net, struct fib6_result *res,
 	struct fib6_info *sibling, *next_sibling;
 	struct fib6_info *match = res->f6i;
 
-	if ((!match->fib6_nsiblings && !match->nh) || have_oif_match)
+	if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 		goto out;
 
 	/* We might have already computed the hash for ICMPv6 errors. In such

of course I notice this after sending the patch. This can actually break with multipath nexthop objects since it can overwrite the selected path if done earlier.

DaveM: please drop; will send a v2