Re: [PATCH RFC v2 net-next 10/16] bpf: add eBPF verifier

From: Alexei Starovoitov
Date: Tue Aug 12 2014 - 16:00:58 EST


On Tue, Aug 12, 2014 at 12:32 PM, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
> On Thu, Jul 24, 2014 at 12:25 PM, Alexei Starovoitov <ast@xxxxxxxxxxxx> wrote:
>> On Thu, Jul 24, 2014 at 11:25 AM, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
>>> On Thu, Jul 17, 2014 at 9:20 PM, Alexei Starovoitov <ast@xxxxxxxxxxxx> wrote:
>>>> Safety of eBPF programs is statically determined by the verifier, which detects:
>>>> - loops
>>>> - out of range jumps
>>>> - unreachable instructions
>>>> - invalid instructions
>>>> - uninitialized register access
>>>> - uninitialized stack access
>>>> - misaligned stack access
>>>> - out of range stack access
>>>> - invalid calling convention
>>>
>>> Is there something that documents exactly what conditions an eBPF
>>> program must satisfy in order to be considered valid?
>>
>> I did a writeup in the past on things that verifiers checks and gave it
>> to internal folks to review. Guys have said that now they understand very
>> well how it works, but in reality it didn't help at all to write valid programs.
>> What worked is 'verification trace' = the instruction by instruction dump
>> of verifier state while it's analyzing the program.
>> I gave few simple examples of it in
>> 'Understanding eBPF verifier messages' section:
>> https://git.kernel.org/cgit/linux/kernel/git/ast/bpf.git/diff/Documentation/networking/filter.txt?id=b22459133b9f52d2176c8c0f8b5eb036478a40c9
>> Every example there is what "program must satisfy to be valid"...
>>
>> Therefore I'm addressing two things:
>> 1. how verifier works and what it checks for.
>> that is described in 'eBPF verifier' section of the doc and
>> in 200 lines of comments inside verifier.c
>
> That doc is pretty good. I'll try to read it carefully soon. Sorry
> for the huge delay here -- I've been on vacation.

I've been sitting on v4 for few weeks, since it's a merge window.
So please hold on a careful review. I'll post v4 later today.
Mainly I've split the verifier into several patches to make it
easier to read.
Thanks!
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/