On Wed, Aug 1, 2018 at 3:46 PM, Florian Westphal fw@strlen.de wrote:
Dmitry Vyukov dvyukov@google.com wrote:
If that scenario is possible that a fix would be to make
Looks possible.
__nf_conntrack_find_get ever return NULL iff it got NULL from ____nf_conntrack_find (not if any of the checks has failed).
I don't see why we need to restart on nf_ct_is_dying(), but other than that this seems ok.
Because it can be a wrong entry dying. When we check dying, we don't yet know if we are looking at the right entry or not. If we successfully acquire a reference, then recheck nf_ct_key_equal and _then_ check dying, then we don't need to restart on dying. But with the current check order, we need to restart on dying too.