On Tue, Oct 15, 2019 at 03:19:13PM +0200, Florian Westphal wrote: > conntrack extensions are free'd via kfree_rcu, but there appears to be > no need for this anymore. > > Lookup doesn't access ct->ext. All other accesses i found occur > after taking either the hash bucket lock, the dying list lock, > or a ct reference count. > > Only exception was ctnetlink, where we could potentially see a > ct->ext that is about to be free'd via krealloc on other cpu. > Since that only affects unconfirmed conntracks, just skip dumping > extensions for those. > Applied.
