Hello,
On Mon, 4 Jul 2016, Vegard Nossum wrote:
> struct fib_nh->nh_dev can be NULL, so we should check it before calling
> __in_dev_get_rcu on it.
>
> Multiple places seem to want this (and check the return value), so we can
> add a convenience wrapper for this.
>
> This fixes a crash in AF_NETLINK sendmsg().
>
> Please double check that I caught all the callers that need the NULL
> guard.
What kind of configuration causes such crash?
> Fixes: 0eeb075fad73 ("net: ipv4 sysctl option to ignore routes when nexthop
> link is down")
> Cc: Andy Gospodarek <go...@cumulusnetworks.com>
> Cc: Dinesh Dutt <dd...@cumulusnetworks.com>
> Cc: Scott Feldman <sfel...@gmail.com>
> Cc: David S. Miller <da...@davemloft.net>
> Signed-off-by: Vegard Nossum <vegard.nos...@oracle.com>
> ---
> include/linux/inetdevice.h | 7 +++++++
> net/ipv4/fib_semantics.c | 8 ++++----
> 2 files changed, 11 insertions(+), 4 deletions(-)
>
> diff --git a/include/linux/inetdevice.h b/include/linux/inetdevice.h
> index ee971f3..2d8bea1 100644
> --- a/include/linux/inetdevice.h
> +++ b/include/linux/inetdevice.h
> @@ -222,6 +222,13 @@ static inline struct in_device *__in_dev_get_rtnl(const
> struct net_device *dev)
> return rtnl_dereference(dev->ip_ptr);
> }
>
> +static inline struct in_device *in_dev_get_rtnl(const struct net_device *dev)
> +{
> + if (!dev)
> + return NULL;
> + return __in_dev_get_rtnl(dev);
> +}
> +
> static inline struct neigh_parms *__in_dev_arp_parms_get_rcu(const struct
> net_device *dev)
> {
> struct in_device *in_dev = __in_dev_get_rcu(dev);
> diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c
> index d09173b..8289799 100644
> --- a/net/ipv4/fib_semantics.c
> +++ b/net/ipv4/fib_semantics.c
> @@ -545,7 +545,7 @@ static void fib_rebalance(struct fib_info *fi)
> if (nh->nh_flags & RTNH_F_DEAD)
> continue;
>
> - in_dev = __in_dev_get_rtnl(nh->nh_dev);
> + in_dev = in_dev_get_rtnl(nh->nh_dev);
fib_rebalance can not crash because for multipath
routes (fib_nhs > 1) all nexthops should have valid nh_dev.
>
> if (in_dev &&
> IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
> @@ -559,7 +559,7 @@ static void fib_rebalance(struct fib_info *fi)
> change_nexthops(fi) {
> int upper_bound;
>
> - in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
> + in_dev = in_dev_get_rtnl(nexthop_nh->nh_dev);
>
> if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
> upper_bound = -1;
> @@ -1261,7 +1261,7 @@ int fib_dump_info(struct sk_buff *skb, u32 portid, u32
> seq, int event,
> nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
> goto nla_put_failure;
> if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
> - in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
> + in_dev = in_dev_get_rtnl(fi->fib_nh->nh_dev);
Looks like this is the only place that can crash,
adding extra cycles to the other places is not good. And this
can happen only because fib_create_info() allows RTNH_F_LINKDOWN
to come for routes with error. May be fc_flags should be masked
there? Or there is another place that sets the flag when nh_dev
is NULL?
> if (in_dev &&
> IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
> rtm->rtm_flags |= RTNH_F_DEAD;
> @@ -1292,7 +1292,7 @@ int fib_dump_info(struct sk_buff *skb, u32 portid, u32
> seq, int event,
>
> rtnh->rtnh_flags = nh->nh_flags & 0xFF;
> if (nh->nh_flags & RTNH_F_LINKDOWN) {
> - in_dev = __in_dev_get_rtnl(nh->nh_dev);
> + in_dev = in_dev_get_rtnl(nh->nh_dev);
Not needed because fib_nhs > 1
> if (in_dev &&
> IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
> rtnh->rtnh_flags |= RTNH_F_DEAD;
> --
Regards
--
Julian Anastasov <j...@ssi.bg>
