Re: [oss-drivers] [PATCH AUTOSEL 5.4 326/350] bpf: Switch bpf_map ref counter to atomic64_t so bpf_map_inc() never fails

[Daniel Borkmann]

On Tue, Dec 10, 2019 at 01:28:34PM -0800, Jakub Kicinski wrote:
> On Tue, 10 Dec 2019 16:07:11 -0500, Sasha Levin wrote:
> > From: Andrii Nakryiko <andriin@xxxxxx>
> > 
> > [ Upstream commit 1e0bd5a091e5d9e0f1d5b0e6329b87bb1792f784 ]
> > 
> > 92117d8443bc ("bpf: fix refcnt overflow") turned refcounting of bpf_map into
> > potentially failing operation, when refcount reaches BPF_MAX_REFCNT limit
> > (32k). Due to using 32-bit counter, it's possible in practice to overflow
> > refcounter and make it wrap around to 0, causing erroneous map free, while
> > there are still references to it, causing use-after-free problems.
> 
> I don't think this is a bug fix, the second sentence here is written
> in a quite confusing way, but there is no bug.
> 
> Could you drop? I don't think it's worth the backporting pain since it
> changes bpf_map_inc().

Agree, this is not a bug fix and should not go to stable. (Also agree that
the changelog is super confusing here and should have been done differently
to avoid exactly where we are here. I think I pointed that out in the
original patch, but seems this slipped through the cracks :/)

Thanks,
Daniel