Re: [PATCH net] samples/bpf: bpf_load.c order of prog_fd[] should correspond with ELF order

[Jesper Dangaard Brouer <brouer@redhat.com>]

On Tue, 30 May 2017 09:34:12 -0700
Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:

> On Tue, May 30, 2017 at 02:37:51PM +0200, Jesper Dangaard Brouer wrote:
> > An eBPF ELF file generated with LLVM can contain several program
> > section, which can be used for bpf tail calls.  The bpf prog file
> > descriptors are accessible via array prog_fd[].
> > 
> > At-least XDP samples assume ordering, and uses prog_fd[0] is the main
> > XDP program to attach.  The actual order of array prog_fd[] depend on
> > whether or not a bpf program section is referencing any maps or not.
> > Not using a map result in being loaded/processed after all other
> > prog section.  Thus, this can lead to some very strange and hard to
> > debug situation, as the user can only see a FD and cannot correlated
> > that with the ELF section name.
> > 
> > The fix is rather simple, and even removes duplicate memcmp code.
> > Simply load program sections as the last step, instead of
> > load_and_attach while processing the relocation section.
> > 
> > When working with tail calls, it become even more essential that the
> > order of prog_fd[] is consistant, like the current dependency of the
> > map_fd[] order.
> > 
> > Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>  
> 
> Looks fine, but imo net-next is better, since it's not a bugfix
> to anything in the net tree, but a general improvement.

Okay, I'm fine with this going into net-next. I'm running with own
fix[1] in my git-tree, so I can get my tail call example[2] working.

[1] https://github.com/netoptimizer/prototype-kernel/commit/e785522d84d5bf
[2] https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/samples/bpf

> Acked-by: Alexei Starovoitov <ast@kernel.org>

Thanks!

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