netdev - Re: [PATCH] bpf,x64: pad NOPs to make images converge more easily

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <X9b0LHBvj4UDGIwe@GaryWorkstation>
Date:   Mon, 14 Dec 2020 13:12:12 +0800
From:   Gary Lin <glin@...e.com>
To:     Alexei Starovoitov <alexei.starovoitov@...il.com>
Cc:     Daniel Borkmann <daniel@...earbox.net>,
        Andrii Nakryiko <andrii.nakryiko@...il.com>,
        Networking <netdev@...r.kernel.org>, bpf <bpf@...r.kernel.org>,
        Alexei Starovoitov <ast@...nel.org>,
        Eric Dumazet <eric.dumazet@...il.com>,
        andreas.taschner@...e.com
Subject: Re: [PATCH] bpf,x64: pad NOPs to make images converge more easily

On Fri, Dec 11, 2020 at 06:24:47PM -0800, Alexei Starovoitov wrote:
> On Fri, Dec 11, 2020 at 1:13 PM Daniel Borkmann <daniel@...earbox.net> wrote:
> > >> +                       }
> > >>   emit_jmp:
> > >>                          if (is_imm8(jmp_offset)) {
> > >> +                               if (jmp_padding)
> > >> +                                       cnt += emit_nops(&prog, INSN_SZ_DIFF - 2);
> 
> Could you describe all possible numbers of bytes in padding?
> Is it 0, 2, 4 ?
> Would be good to add warn_on_once to make sure the number
> of nops is expected.
> 
For the conditional jumps, it could be 0 or 4. As for nop jumps, it may be
0, 2, or 5. For the pure jumps, 0 or 3. Will add the warning in the next
version.

> > >>   struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
> > >>   {
> > >>          struct bpf_binary_header *header = NULL;
> > >> @@ -1981,6 +1997,7 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
> > >>          struct jit_context ctx = {};
> > >>          bool tmp_blinded = false;
> > >>          bool extra_pass = false;
> > >> +       bool padding = prog->padded;
> > >
> > > can this ever be true on assignment? I.e., can the program be jitted twice?
> >
> > Yes, progs can be passed into the JIT twice, see also jit_subprogs(). In one of
> > the earlier patches it would still potentially change the image size a second
> > time which would break subprogs aka bpf2bpf calls.
> 
> Right. I think memorized padded flag shouldn't be in sticky bits
> of the prog structure.
> It's only needed between the last pass and extra pass for bpf2bpf calls.
> I think it would be cleaner to keep it in struct x64_jit_data *jit_data.
> 
Okay, jit_data is surely a better place for the flag.

> As others have said the selftests are must have.
> Especially for bpf2bpf calls where one subprog is padded.
> 
Will try to craft some test cases for this patch in v2.

Gary Lin