On Thu, 29 Aug 2019 20:05:49 +0200
Toke Høiland-Jørgensen [off-list ref] wrote:

Alexei Starovoitov [off-list ref] writes:

quoted

On Thu, Aug 29, 2019 at 09:44:18AM +0200, Toke Høiland-Jørgensen wrote:  

quoted

Alexei Starovoitov [off-list ref] writes:
  

quoted

CAP_BPF allows the following BPF operations:
- Loading all types of BPF programs
- Creating all types of BPF maps except:
   - stackmap that needs CAP_TRACING
   - devmap that needs CAP_NET_ADMIN
   - cpumap that needs CAP_SYS_ADMIN  

Why CAP_SYS_ADMIN instead of CAP_NET_ADMIN for cpumap?  

Currently it's cap_sys_admin and I think it should stay this way
because it creates kthreads.  

Ah, right. I can sorta see that makes sense because of the kthreads, but
it also means that you can use all of XDP *except* cpumap with
CAP_NET_ADMIN+CAP_BPF. That is bound to create confusion, isn't it?

 
Hmm... I see 'cpumap' primarily as a network stack feature.  It is about
starting the network stack on a specific CPU, allocating and building
SKBs on that remote CPU.  It can only be used together with XDP_REDIRECT.
I would prefer CAP_NET_ADMIN like the devmap, to keep the XDP
capabilities consistent.

-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer