Maciej Żenczykowski <zenczykow...@gmail.com> wrote: > From: Maciej Żenczykowski <m...@google.com> > > This reverts commit 114aa35d06d4920c537b72f9fa935de5dd205260. > > By my understanding of kmemleak the reasoning for this patch > is incorrect. If kmemleak couldn't handle rcu we'd have it > reporting leaks all over the place. My belief is that this > was instead papering over a real leak.
Perhaps, but note that this is related to nfct->ext, not nfct itself. I think we could remove __krealloc and use krealloc directly with a bit of changes in the nf_conntrack core to make sure we do not access nfct->ext without holding a reference to nfct, and then drop rcu protection of nfct->ext, I don't think its strictly required anymore.
