Re: [PATCH v2 1/2 net] bitfield.h: don't compile-time validate _val in FIELD_FIT

From: Alex Elder <elder@linaro.org>
To: Nick Desaulniers <ndesaulniers@google.com>,
	"David S . Miller" <davem@davemloft.net>
Cc: Alexei Starovoitov <ast@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Martin KaFai Lau <kafai@fb.com>, Song Liu <songliubraving@fb.com>,
	Yonghong Song <yhs@fb.com>, Andrii Nakryiko <andriin@fb.com>,
	John Fastabend <john.fastabend@gmail.com>,
	KP Singh <kpsingh@chromium.org>,
	oss-drivers@netronome.com, netdev@vger.kernel.org,
	linux-kernel@vger.kernel.org, Jakub Kicinski <kuba@kernel.org>,
	stable@vger.kernel.org, Masahiro Yamada <masahiroy@kernel.org>,
	Sami Tolvanen <samitolvanen@google.com>
Subject: Re: [PATCH v2 1/2 net] bitfield.h: don't compile-time validate _val in FIELD_FIT
Date: Wed, 8 Jul 2020 20:48:56 -0500	[thread overview]
Message-ID: <af1c02b8-3856-dac9-67e9-af27803c2eee@linaro.org> (raw)
In-Reply-To: <20200708230402.1644819-2-ndesaulniers@google.com>

On 7/8/20 6:04 PM, Nick Desaulniers wrote:
> From: Jakub Kicinski <kuba@kernel.org>
> 
> When ur_load_imm_any() is inlined into jeq_imm(), it's possible for the
> compiler to deduce a case where _val can only have the value of -1 at
> compile time. Specifically,
> 
> /* struct bpf_insn: _s32 imm */
> u64 imm = insn->imm; /* sign extend */
> if (imm >> 32) { /* non-zero only if insn->imm is negative */
>   /* inlined from ur_load_imm_any */
>   u32 __imm = imm >> 32; /* therefore, always 0xffffffff */
>   if (__builtin_constant_p(__imm) && __imm > 255)
>     compiletime_assert_XXX()
> 
> This can result in tripping a BUILD_BUG_ON() in __BF_FIELD_CHECK() that
> checks that a given value is representable in one byte (interpreted as
> unsigned).

Looking at the 12 other callers of FIELD_FIT(), it's hard to
tell but it appears most of them don't supply constant _val
to the macro.  So maybe relaxing this check is no great loss.

I feel like I need to look much more deeply into this to call
it a review, so:

Acked-by: Alex Elder <elder@linaro.org>

> FIELD_FIT() should return true or false at runtime for whether a value
> can fit for not. Don't break the build over a value that's too large for
> the mask. We'd prefer to keep the inlining and compiler optimizations
> though we know this case will always return false.
> 
> Cc: stable@vger.kernel.org
> Link: https://lore.kernel.org/kernel-hardening/CAK7LNASvb0UDJ0U5wkYYRzTAdnEs64HjXpEUL7d=V0CXiAXcNw@mail.gmail.com/
> Reported-by: Masahiro Yamada <masahiroy@kernel.org>
> Debugged-by: Sami Tolvanen <samitolvanen@google.com>
> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
> Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
> ---
> Changes V1->V2:
> * None
> 
>  include/linux/bitfield.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/include/linux/bitfield.h b/include/linux/bitfield.h
> index 48ea093ff04c..4e035aca6f7e 100644
> --- a/include/linux/bitfield.h
> +++ b/include/linux/bitfield.h
> @@ -77,7 +77,7 @@
>   */
>  #define FIELD_FIT(_mask, _val)						\
>  	({								\
> -		__BF_FIELD_CHECK(_mask, 0ULL, _val, "FIELD_FIT: ");	\
> +		__BF_FIELD_CHECK(_mask, 0ULL, 0ULL, "FIELD_FIT: ");	\
>  		!((((typeof(_mask))_val) << __bf_shf(_mask)) & ~(_mask)); \
>  	})
>  
> 