From: Martin KaFai Lau <ka...@fb.com> Date: Mon, 11 Apr 2016 15:29:33 -0700
> v2: > ~ Protect __sk_dst_get() operations with rcu_read_lock in > release_cb() because another thread may do ip6_dst_store() > for a udp sk without taking the sk lock (e.g. in sendmsg). > ~ Do a ipv6_addr_v4mapped(&sk->sk_v6_daddr) check before > calling ip6_datagram_dst_update() in patch 3 and 4. It is > similar to how __ip6_datagram_connect handles it. > ~ One fix in ip6_datagram_dst_update() in patch 2. It needs > to check (np->flow_label & IPV6_FLOWLABEL_MASK) before > doing fl6_sock_lookup. I was confused with the naming > of IPV6_FLOWLABEL_MASK and IPV6_FLOWINFO_MASK. > ~ Check dst->obsolete just on the safe side, although I think it > should at least have DST_OBSOLETE_FORCE_CHK by now. > ~ Add Fixes tag to patch 3 and 4 > ~ Add some points from the previous discussion about holding > sk lock to the commit message in patch 3. > > v1: > There is a case in connected UDP socket such that > getsockopt(IPV6_MTU) will return a stale MTU value. The reproducible > sequence could be the following: > 1. Create a connected UDP socket > 2. Send some datagrams out > 3. Receive a ICMPV6_PKT_TOOBIG > 4. No new outgoing datagrams to trigger the sk_dst_check() > logic to update the sk->sk_dst_cache. > 5. getsockopt(IPV6_MTU) returns the mtu from the invalid > sk->sk_dst_cache instead of the newly created RTF_CACHE clone. > > Patch 1 and 2 are the prep work. > Patch 3 and 4 are the fixes. Series applied, thanks Martin.
