netdev - Re: [PATCH v2 2/2] x86/cfi,bpf: Fix BPF JIT call

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: <20231206183713.GA35897@noisy.programming.kicks-ass.net>
Date: Wed, 6 Dec 2023 19:37:13 +0100
From: Peter Zijlstra <peterz@...radead.org>
To: Alexei Starovoitov <alexei.starovoitov@...il.com>
Cc: Jiri Olsa <olsajiri@...il.com>, Song Liu <song@...nel.org>,
	Song Liu <songliubraving@...a.com>,
	Paul Walmsley <paul.walmsley@...ive.com>,
	Palmer Dabbelt <palmer@...belt.com>,
	Albert Ou <aou@...s.berkeley.edu>,
	Thomas Gleixner <tglx@...utronix.de>,
	Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>,
	Dave Hansen <dave.hansen@...ux.intel.com>, X86 ML <x86@...nel.org>,
	"H. Peter Anvin" <hpa@...or.com>,
	"David S. Miller" <davem@...emloft.net>,
	David Ahern <dsahern@...nel.org>,
	Alexei Starovoitov <ast@...nel.org>,
	Daniel Borkmann <daniel@...earbox.net>,
	Andrii Nakryiko <andrii@...nel.org>,
	Martin KaFai Lau <martin.lau@...ux.dev>,
	Yonghong Song <yonghong.song@...ux.dev>,
	John Fastabend <john.fastabend@...il.com>,
	KP Singh <kpsingh@...nel.org>, Stanislav Fomichev <sdf@...gle.com>,
	Hao Luo <haoluo@...gle.com>, Arnd Bergmann <arnd@...db.de>,
	Sami Tolvanen <samitolvanen@...gle.com>,
	Kees Cook <keescook@...omium.org>,
	Nathan Chancellor <nathan@...nel.org>,
	Nick Desaulniers <ndesaulniers@...gle.com>,
	linux-riscv <linux-riscv@...ts.infradead.org>,
	LKML <linux-kernel@...r.kernel.org>,
	Network Development <netdev@...r.kernel.org>,
	bpf <bpf@...r.kernel.org>, linux-arch <linux-arch@...r.kernel.org>,
	clang-built-linux <llvm@...ts.linux.dev>,
	Josh Poimboeuf <jpoimboe@...nel.org>,
	Joao Moreira <joao@...rdrivepizza.com>,
	Mark Rutland <mark.rutland@....com>
Subject: Re: [PATCH v2 2/2] x86/cfi,bpf: Fix BPF JIT call

On Wed, Dec 06, 2023 at 05:38:14PM +0100, Peter Zijlstra wrote:
> On Mon, Dec 04, 2023 at 05:18:31PM -0800, Alexei Starovoitov wrote:
> 
> > [   13.978497]  ? asm_exc_invalid_op+0x1a/0x20
> > [   13.978798]  ? tcp_set_ca_state+0x51/0xd0
> > [   13.979087]  tcp_v6_syn_recv_sock+0x45c/0x6c0
> > [   13.979401]  tcp_check_req+0x497/0x590
> 
> > The stack trace doesn't have any bpf, but it's a bpf issue too.
> > Here tcp_set_ca_state() calls
> > icsk->icsk_ca_ops->set_state(sk, ca_state);
> > which calls bpf prog via bpf trampoline.
> 
> 
> 
> Specifically, I think this is
> tools/testing/selftests/bpf/progs/bpf_cubic.c, which has:
> 
>         .set_state      = (void *)bpf_cubic_state,
> 
> which comes from:
> 
> BPF_STRUCT_OPS(bpf_cubic_state, struct sock *sk, __u8 *new_state)
> 
> which then wraps:
> 
> BPF_PROG()
> 
> which ends up generating:
> 
> static __always_inline ___bpf_cubic_state(unsigned long long *ctx, struct sock *sk, __u8 *new_state)
> {
> 	...
> }
> 
> void bpf_cubic_state(unsigned long long *ctx)
> {
> 	return ____bpf_cubic_state(ctx, ctx[0], ctx[1]);
> }
> 
> 
> I think this then uses arch_prepare_bpf_trampoline(), but I'm entirely
> lost how this all comes together, because the way I understand it the
> whole bpf_trampoline is used to hook into an ftrace __fentry hook.
> 
> And a __fentry hook is very much not a function pointer. Help!?!?

kernel/bpf/bpf_struct_ops.c:bpf_struct_ops_prepare_trampoline()

And yeah, it seems to use the ftrace trampoline for indirect calls here,
*sigh*.

> The other case:
> 
> For tools/testing/selftests/bpf/progs/bloom_filter_bench.c we have:
> 
>         bpf_for_each_map_elem(&array_map, bloom_callback, &data, 0);
> 
> and here bloom callback appears like a normal function:
> 
> static __u64
> bloom_callback(struct bpf_map *map, __u32 *key, void *val,
>                struct callback_ctx *data)
> 
> 
> But what do functions looks like in the JIT? What's the actual address
> that's then passed into the helper function. Given this seems to work
> without kCFI, it should at least have an ENDBR, but there's only 3 of
> those afaict:
> 
>   - emit_prologue() first insn
>   - emit_prologue() tail-call site
>   - arch_preprare_bpf_trampoline()
> 
> If the function passed to the helper is from do_jit()/emit_prologue(),
> then how do I tell what 'function' is being JIT'ed ?
> 
> If it is arch_prepare_bpf_trampoline(), then we're back at the previous
> question and I don't see how a __fentry site becomes a callable function
> pointer.
> 
> 
> Any clues would be much appreciated.

Still not figured out how this one works...