Re: [PATCH v2 bpf-next 03/17] bpf: Introduce BPF trampoline

From: Alexei Starovoitov <alexei.starovoitov@gmail.com>
To: Song Liu <songliubraving@fb.com>
Cc: Alexei Starovoitov <ast@kernel.org>,
	"davem@davemloft.net" <davem@davemloft.net>,
	"daniel@iogearbox.net" <daniel@iogearbox.net>,
	"x86@kernel.org" <x86@kernel.org>,
	"netdev@vger.kernel.org" <netdev@vger.kernel.org>,
	"bpf@vger.kernel.org" <bpf@vger.kernel.org>,
	Kernel Team <Kernel-team@fb.com>
Subject: Re: [PATCH v2 bpf-next 03/17] bpf: Introduce BPF trampoline
Date: Thu, 7 Nov 2019 14:55:54 -0800	[thread overview]
Message-ID: <20191107225553.vnnos6nblxlwx24a@ast-mbp.dhcp.thefacebook.com> (raw)
In-Reply-To: <5967F93A-235B-447E-9B70-E7768998B718@fb.com>

On Thu, Nov 07, 2019 at 10:37:19PM +0000, Song Liu wrote:
> 
> 
> > On Nov 6, 2019, at 9:46 PM, Alexei Starovoitov <ast@kernel.org> wrote:
> > 
> 
> [...]
> 
> > 
> > Signed-off-by: Alexei Starovoitov <ast@kernel.org>
> > Acked-by: Andrii Nakryiko <andriin@fb.com>
> > ---
> > arch/x86/net/bpf_jit_comp.c | 227 ++++++++++++++++++++++++++++++--
> > include/linux/bpf.h         |  98 ++++++++++++++
> > include/uapi/linux/bpf.h    |   2 +
> > kernel/bpf/Makefile         |   1 +
> > kernel/bpf/btf.c            |  77 ++++++++++-
> > kernel/bpf/core.c           |   1 +
> > kernel/bpf/syscall.c        |  53 +++++++-
> > kernel/bpf/trampoline.c     | 252 ++++++++++++++++++++++++++++++++++++
> > kernel/bpf/verifier.c       |  39 ++++++
> > 9 files changed, 732 insertions(+), 18 deletions(-)
> > create mode 100644 kernel/bpf/trampoline.c
> > 
> > diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
> > index 8631d3bd637f..44169e8bffc0 100644
> > --- a/arch/x86/net/bpf_jit_comp.c
> > +++ b/arch/x86/net/bpf_jit_comp.c
> > @@ -98,6 +98,7 @@ static int bpf_size_to_x86_bytes(int bpf_size)
> > 
> > /* Pick a register outside of BPF range for JIT internal work */
> > #define AUX_REG (MAX_BPF_JIT_REG + 1)
> > +#define X86_REG_R9 (MAX_BPF_JIT_REG + 2)
> > 
> > /*
> >  * The following table maps BPF registers to x86-64 registers.
> > @@ -123,6 +124,7 @@ static const int reg2hex[] = {
> > 	[BPF_REG_FP] = 5, /* RBP readonly */
> > 	[BPF_REG_AX] = 2, /* R10 temp register */
> > 	[AUX_REG] = 3,    /* R11 temp register */
> > +	[X86_REG_R9] = 1, /* R9 register, 6th function argument */
> 
> We should update the comment above this:
> 
>  * Also x86-64 register R9 is unused. ...

good point. fixed.

> > +	/* One half of the page has active running trampoline.
> > +	 * Another half is an area for next trampoline.
> > +	 * Make sure the trampoline generation logic doesn't overflow.
> > +	 */
> > +	if (WARN_ON_ONCE(prog - (u8 *)image > PAGE_SIZE / 2 - BPF_INSN_SAFETY))
> > +		return -EFAULT;
> 
> Given max number of args, can we catch this error at compile time? 

I don't see how to do that. I was thinking about having fake __init function
that would call it with flags that can generate the longest trampoline, but
it's not fool proof either.
So I've added a test for it instead. See patch 10.

> > +
> > +static int bpf_trampoline_update(struct bpf_prog *prog)
> 
> Seems argument "prog" is not used at all? 

like one below ? ;)

> > +{
> > +	struct bpf_trampoline *tr = prog->aux->trampoline;
> > +	void *old_image = tr->image + ((tr->selector + 1) & 1) * PAGE_SIZE/2;
> > +	void *new_image = tr->image + (tr->selector & 1) * PAGE_SIZE/2;
> > +	if (err)
> > +		goto out;
> > +	tr->selector++;
> 
> Shall we do selector-- for unlink?

It's a bit flip. I think it would be more confusing with --