Merge branch 'nh-group-refcnt'

Nikolay Aleksandrov says:

====================
net: nexthop: fix refcount issues when replacing groups

This set fixes a refcount bug when replacing nexthop groups and
modifying routes. It is complex because the objects look valid when
debugging memory dumps, but we end up having refcount dependency between
unlinked objects which can never be released, so in turn they cannot
free their resources and refcounts. The problem happens because we can
have stale IPv6 per-cpu dsts in nexthops which were removed from a
group. Even though the IPv6 gen is bumped, the dsts won't be released
until traffic passes through them or the nexthop is freed, that can take
arbitrarily long time, and even worse we can create a scenario[1] where it
can never be released. The fix is to release the IPv6 per-cpu dsts of
replaced nexthops after an RCU grace period so no new ones can be
created. To do that we add a new IPv6 stub - fib6_nh_release_dsts, which
is used by the nexthop code only when necessary. We can further optimize
group replacement, but that is more suited for net-next as these patches
would have to be backported to stable releases.

v2: patch 02: update commit msg
    patch 03: check for mausezahn before testing and make a few comments
              more verbose

[1]
This info is also present in patch 02's commit message.
Initial state:
 $ ip nexthop list
  id 200 via 2002:db8::2 dev bridge.10 scope link onlink
  id 201 via 2002:db8::3 dev bridge scope link onlink
  id 203 group 201/200
 $ ip -6 route
  2001:db8::10 nhid 203 metric 1024 pref medium
     nexthop via 2002:db8::3 dev bridge weight 1 onlink
     nexthop via 2002:db8::2 dev bridge.10 weight 1 onlink

Create rt6_info through one of the multipath legs, e.g.:
 $ taskset -a -c 1  ./pkt_inj 24 bridge.10 2001:db8::10
 (pkt_inj is just a custom packet generator, nothing special)

Then remove that leg from the group by replace (let's assume it is id
200 in this case):
 $ ip nexthop replace id 203 group 201

Now remove the IPv6 route:
 $ ip -6 route del 2001:db8::10/128

The route won't be really deleted due to the stale rt6_info holding 1
refcnt in nexthop id 200.
At this point we have the following reference count dependency:
 (deleted) IPv6 route holds 1 reference over nhid 203
 nh 203 holds 1 ref over id 201
 nh 200 holds 1 ref over the net device and the route due to the stale
 rt6_info

Now to create circular dependency between nh 200 and the IPv6 route, and
also to get a reference over nh 200, restore nhid 200 in the group:
 $ ip nexthop replace id 203 group 201/200

And now we have a permanent circular dependncy because nhid 203 holds a
reference over nh 200 and 201, but the route holds a ref over nh 203 and
is deleted.

To trigger the bug just delete the group (nhid 203):
 $ ip nexthop del id 203

It won't really be deleted due to the IPv6 route dependency, and now we
have 2 unlinked and deleted objects that reference each other: the group
and the IPv6 route. Since the group drops the reference it holds over its
entries at free time (i.e. its own refcount needs to drop to 0) that will
never happen and we get a permanent ref on them, since one of the entries
holds a reference over the IPv6 route it will also never be released.

At this point the dependencies are:
 (deleted, only unlinked) IPv6 route holds reference over group nh 203
 (deleted, only unlinked) group nh 203 holds reference over nh 201 and 200
 nh 200 holds 1 ref over the net device and the route due to the stale
 rt6_info

This is the last point where it can be fixed by running traffic through
nh 200, and specifically through the same CPU so the rt6_info (dst) will
get released due to the IPv6 genid, that in turn will free the IPv6
route, which in turn will free the ref count over the group nh 203.

If nh 200 is deleted at this point, it will never be released due to the
ref from the unlinked group 203, it will only be unlinked:
 $ ip nexthop del id 200
 $ ip nexthop
 $

Now we can never release that stale rt6_info, we have IPv6 route with ref
over group nh 203, group nh 203 with ref over nh 200 and 201, nh 200 with
rt6_info (dst) with ref over the net device and the IPv6 route. All of
these objects are only unlinked, and cannot be released, thus they can't
release their ref counts.

 Message from syslogd@dev at Nov 19 14:04:10 ...
  kernel:[73501.828730] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
 Message from syslogd@dev at Nov 19 14:04:20 ...
  kernel:[73512.068811] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3

====================

Signed-off-by: David S. Miller <davem@davemloft.net>

include/net/ip6_fib.h

@@ -485,6 +485,7 @@ int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
 		 struct fib6_config *cfg, gfp_t gfp_flags,
 		 struct netlink_ext_ack *extack);
 void fib6_nh_release(struct fib6_nh *fib6_nh);
+void fib6_nh_release_dsts(struct fib6_nh *fib6_nh);
 
 int call_fib6_entry_notifiers(struct net *net,
 			      enum fib_event_type event_type,

include/net/ipv6_stubs.h

@@ -47,6 +47,7 @@ struct ipv6_stub {
 			    struct fib6_config *cfg, gfp_t gfp_flags,
 			    struct netlink_ext_ack *extack);
 	void (*fib6_nh_release)(struct fib6_nh *fib6_nh);
+	void (*fib6_nh_release_dsts)(struct fib6_nh *fib6_nh);
 	void (*fib6_update_sernum)(struct net *net, struct fib6_info *rt);
 	int (*ip6_del_rt)(struct net *net, struct fib6_info *rt, bool skip_notify);
 	void (*fib6_rt_update)(struct net *net, struct fib6_info *rt,

net/ipv4/nexthop.c

@@ -1899,15 +1899,36 @@ static void remove_nexthop(struct net *net, struct nexthop *nh,
 /* if any FIB entries reference this nexthop, any dst entries
  * need to be regenerated
  */
-static void nh_rt_cache_flush(struct net *net, struct nexthop *nh)
+static void nh_rt_cache_flush(struct net *net, struct nexthop *nh,
+			      struct nexthop *replaced_nh)
 {
 	struct fib6_info *f6i;
+	struct nh_group *nhg;
+	int i;
 
 	if (!list_empty(&nh->fi_list))
 		rt_cache_flush(net);
 
 	list_for_each_entry(f6i, &nh->f6i_list, nh_list)
 		ipv6_stub->fib6_update_sernum(net, f6i);
+
+	/* if an IPv6 group was replaced, we have to release all old
+	 * dsts to make sure all refcounts are released
+	 */
+	if (!replaced_nh->is_group)
+		return;
+
+	/* new dsts must use only the new nexthop group */
+	synchronize_net();
+
+	nhg = rtnl_dereference(replaced_nh->nh_grp);
+	for (i = 0; i < nhg->num_nh; i++) {
+		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
+		struct nh_info *nhi = rtnl_dereference(nhge->nh->nh_info);
+
+		if (nhi->family == AF_INET6)
+			ipv6_stub->fib6_nh_release_dsts(&nhi->fib6_nh);
+	}
 }
 
 static int replace_nexthop_grp(struct net *net, struct nexthop *old,
@@ -2247,7 +2268,7 @@ static int replace_nexthop(struct net *net, struct nexthop *old,
 		err = replace_nexthop_single(net, old, new, extack);
 
 	if (!err) {
-		nh_rt_cache_flush(net, old);
+		nh_rt_cache_flush(net, old, new);
 
 		__remove_nexthop(net, new, NULL);
 		nexthop_put(new);

net/ipv6/af_inet6.c

@@ -1026,6 +1026,7 @@ static const struct ipv6_stub ipv6_stub_impl = {
 	.ip6_mtu_from_fib6 = ip6_mtu_from_fib6,
 	.fib6_nh_init	   = fib6_nh_init,
 	.fib6_nh_release   = fib6_nh_release,
+	.fib6_nh_release_dsts = fib6_nh_release_dsts,
 	.fib6_update_sernum = fib6_update_sernum_stub,
 	.fib6_rt_update	   = fib6_rt_update,
 	.ip6_del_rt	   = ip6_del_rt,

net/ipv6/route.c

@@ -3680,6 +3680,25 @@ void fib6_nh_release(struct fib6_nh *fib6_nh)
 	fib_nh_common_release(&fib6_nh->nh_common);
 }
 
+void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
+{
+	int cpu;
+
+	if (!fib6_nh->rt6i_pcpu)
+		return;
+
+	for_each_possible_cpu(cpu) {
+		struct rt6_info *pcpu_rt, **ppcpu_rt;
+
+		ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
+		pcpu_rt = xchg(ppcpu_rt, NULL);
+		if (pcpu_rt) {
+			dst_dev_put(&pcpu_rt->dst);
+			dst_release(&pcpu_rt->dst);
+		}
+	}
+}
+
 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
 					      gfp_t gfp_flags,
 					      struct netlink_ext_ack *extack)

tools/testing/selftests/net/fib_nexthops.sh

@@ -629,6 +629,66 @@ ipv6_fcnal()
 	log_test $? 0 "Nexthops removed on admin down"
 }
 
+ipv6_grp_refs()
+{
+	if [ ! -x "$(command -v mausezahn)" ]; then
+		echo "SKIP: Could not run test; need mausezahn tool"
+		return
+	fi
+
+	run_cmd "$IP link set dev veth1 up"
+	run_cmd "$IP link add veth1.10 link veth1 up type vlan id 10"
+	run_cmd "$IP link add veth1.20 link veth1 up type vlan id 20"
+	run_cmd "$IP -6 addr add 2001:db8:91::1/64 dev veth1.10"
+	run_cmd "$IP -6 addr add 2001:db8:92::1/64 dev veth1.20"
+	run_cmd "$IP -6 neigh add 2001:db8:91::2 lladdr 00:11:22:33:44:55 dev veth1.10"
+	run_cmd "$IP -6 neigh add 2001:db8:92::2 lladdr 00:11:22:33:44:55 dev veth1.20"
+	run_cmd "$IP nexthop add id 100 via 2001:db8:91::2 dev veth1.10"
+	run_cmd "$IP nexthop add id 101 via 2001:db8:92::2 dev veth1.20"
+	run_cmd "$IP nexthop add id 102 group 100"
+	run_cmd "$IP route add 2001:db8:101::1/128 nhid 102"
+
+	# create per-cpu dsts through nh 100
+	run_cmd "ip netns exec me mausezahn -6 veth1.10 -B 2001:db8:101::1 -A 2001:db8:91::1 -c 5 -t tcp "dp=1-1023, flags=syn" >/dev/null 2>&1"
+
+	# remove nh 100 from the group to delete the route potentially leaving
+	# a stale per-cpu dst which holds a reference to the nexthop's net
+	# device and to the IPv6 route
+	run_cmd "$IP nexthop replace id 102 group 101"
+	run_cmd "$IP route del 2001:db8:101::1/128"
+
+	# add both nexthops to the group so a reference is taken on them
+	run_cmd "$IP nexthop replace id 102 group 100/101"
+
+	# if the bug described in commit "net: nexthop: release IPv6 per-cpu
+	# dsts when replacing a nexthop group" exists at this point we have
+	# an unlinked IPv6 route (but not freed due to stale dst) with a
+	# reference over the group so we delete the group which will again
+	# only unlink it due to the route reference
+	run_cmd "$IP nexthop del id 102"
+
+	# delete the nexthop with stale dst, since we have an unlinked
+	# group with a ref to it and an unlinked IPv6 route with ref to the
+	# group, the nh will only be unlinked and not freed so the stale dst
+	# remains forever and we get a net device refcount imbalance
+	run_cmd "$IP nexthop del id 100"
+
+	# if a reference was lost this command will hang because the net device
+	# cannot be removed
+	timeout -s KILL 5 ip netns exec me ip link del veth1.10 >/dev/null 2>&1
+
+	# we can't cleanup if the command is hung trying to delete the netdev
+	if [ $? -eq 137 ]; then
+		return 1
+	fi
+
+	# cleanup
+	run_cmd "$IP link del veth1.20"
+	run_cmd "$IP nexthop flush"
+
+	return 0
+}
+
 ipv6_grp_fcnal()
 {
 	local rc
@@ -734,6 +794,9 @@ ipv6_grp_fcnal()
 
 	run_cmd "$IP nexthop add id 108 group 31/24"
 	log_test $? 2 "Nexthop group can not have a blackhole and another nexthop"
+
+	ipv6_grp_refs
+	log_test $? 0 "Nexthop group replace refcounts"
 }
 
 ipv6_res_grp_fcnal()