From: Yonghong Song <yhs@fb.com>
To: Sami Tolvanen <samitolvanen@google.com>
Cc: "Alexei Starovoitov" <ast@kernel.org>,
"Daniel Borkmann" <daniel@iogearbox.net>,
"Kees Cook" <keescook@chromium.org>, "Martin Lau" <kafai@fb.com>,
"Song Liu" <songliubraving@fb.com>,
"netdev@vger.kernel.org" <netdev@vger.kernel.org>,
"bpf@vger.kernel.org" <bpf@vger.kernel.org>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
"Toke Høiland-Jørgensen" <toke@redhat.com>,
"Björn Töpel" <bjorn.topel@intel.com>
Subject: Re: [PATCH] bpf: validate bpf_func when BPF_JIT is enabled
Date: Wed, 11 Sep 2019 07:42:39 +0000 [thread overview]
Message-ID: <c7c7668e-6336-0367-42b3-2f6026c466dd@fb.com> (raw)
In-Reply-To: <20190910172253.GA164966@google.com>
On 9/10/19 6:22 PM, Sami Tolvanen wrote:
> On Tue, Sep 10, 2019 at 08:37:19AM +0000, Yonghong Song wrote:
>> You did not mention BPF_BINARY_HEADER_MAGIC and added member
>> of `magic` in bpf_binary_header. Could you add some details
>> on what is the purpose for this `magic` member?
>
> Sure, I'll add a description to the next version.
>
> The magic is a random number used to identify bpf_binary_header in
> memory. The purpose of this patch is to limit the possible call
> targets for the function pointer and checking for the magic helps
> ensure we are jumping to a page that contains a jited function,
> instead of allowing calls to arbitrary targets.
>
> This is particularly useful when combined with the compiler-based
> Control-Flow Integrity (CFI) mitigation, which Google started shipping
> in Pixel kernels last year. The compiler injects checks to all
> indirect calls, but cannot obviously validate jumps to dynamically
> generated code.
>
>>> +unsigned int bpf_call_func(const struct bpf_prog *prog, const void *ctx)
>>> +{
>>> + const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog);
>>> +
>>> + if (!IS_ENABLED(CONFIG_BPF_JIT_ALWAYS_ON) && !prog->jited)
>>> + return prog->bpf_func(ctx, prog->insnsi);
>>> +
>>> + if (unlikely(hdr->magic != BPF_BINARY_HEADER_MAGIC ||
>>> + !arch_bpf_jit_check_func(prog))) {
>>> + WARN(1, "attempt to jump to an invalid address");
>>> + return 0;
>>> + }
>>> +
>>> + return prog->bpf_func(ctx, prog->insnsi);
>>> +}
>
>> The above can be rewritten as
>> if (IS_ENABLED(CONFIG_BPF_JIT_ALWAYS_ON) || prog->jited ||
>> hdr->magic != BPF_BINARY_HEADER_MAGIC ||
>> !arch_bpf_jit_check_func(prog))) {
>> WARN(1, "attempt to jump to an invalid address");
>> return 0;
>> }
>
> That doesn't look quite equivalent, but yes, this can be rewritten as a
Indeed, I made a mistake. Your below change is correct.
> single if statement like this:
>
> if ((IS_ENABLED(CONFIG_BPF_JIT_ALWAYS_ON) ||
> prog->jited) &&
> (hdr->magic != BPF_BINARY_HEADER_MAGIC ||
> !arch_bpf_jit_check_func(prog)))
>
> I think splitting the interpreter and JIT paths would be more readable,
> but I can certainly change this if you prefer.
How about this:
if (!IS_ENABLED(CONFIG_BPF_JIT_ALWAYS_ON) && !prog->jited)
goto out;
if (unlikely(hdr->magic != BPF_BINARY_HEADER_MAGIC ||
!arch_bpf_jit_check_func(prog))) {
WARN(1, "attempt to jump to an invalid address");
return 0;
}
out:
return prog->bpf_func(ctx, prog->insnsi);
>
>> BPF_PROG_RUN() will be called during xdp fast path.
>> Have you measured how much slowdown the above change could
>> cost for the performance?
>
> I have not measured the overhead, but it shouldn't be significant. Is
> there a particular benchmark you'd like me to run?
I am not an expert in XDP testing. Toke, Björn, could you give some
suggestions what to test for XDP performance here?
>
> Sami
>
next prev parent reply other threads:[~2019-09-11 7:43 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-09-09 22:32 [PATCH] bpf: validate bpf_func when BPF_JIT is enabled Sami Tolvanen
2019-09-10 8:37 ` Yonghong Song
2019-09-10 17:22 ` Sami Tolvanen
2019-09-11 7:42 ` Yonghong Song [this message]
2019-09-11 10:39 ` Björn Töpel
2019-09-11 12:09 ` Toke Høiland-Jørgensen
2019-09-11 21:07 ` Sami Tolvanen
2019-09-12 10:46 ` Toke Høiland-Jørgensen
2019-09-12 22:01 ` Sami Tolvanen
2019-09-13 12:19 ` Toke Høiland-Jørgensen
2019-09-11 20:29 ` Sami Tolvanen
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=c7c7668e-6336-0367-42b3-2f6026c466dd@fb.com \
--to=yhs@fb.com \
--cc=ast@kernel.org \
--cc=bjorn.topel@intel.com \
--cc=bpf@vger.kernel.org \
--cc=daniel@iogearbox.net \
--cc=kafai@fb.com \
--cc=keescook@chromium.org \
--cc=linux-kernel@vger.kernel.org \
--cc=netdev@vger.kernel.org \
--cc=samitolvanen@google.com \
--cc=songliubraving@fb.com \
--cc=toke@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).