Received-SPF: pass (google.com: domain of linux-crypto-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Fri, 21 Jan 2022 11:54:34 -0800
From:   Eric Biggers <ebiggers@kernel.org>
To:     "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc:     Miles Chen <miles.chen@mediatek.com>,
        Ard Biesheuvel <ardb@kernel.org>,
        Linux ARM <linux-arm-kernel@lists.infradead.org>,
        Linux Crypto Mailing List <linux-crypto@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        linux-mediatek@lists.infradead.org,
        Nick Desaulniers <ndesaulniers@google.com>,
        Sami Tolvanen <samitolvanen@google.com>
Subject: Re: [PATCH] lib/crypto: blake2s: avoid indirect calls to compression
 function for Clang CFI
Message-ID: <YesPesOhftSzp2ft@sol.localdomain>
References: <CAHmME9qVMomgb53rABKsucCoEhwsk+=KzDdEcGKtecOXuahTZw@mail.gmail.com>
 <20220119135450.564115-1-Jason@zx2c4.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220119135450.564115-1-Jason@zx2c4.com>
Precedence: bulk

On Wed, Jan 19, 2022 at 02:54:50PM +0100, Jason A. Donenfeld wrote:
> blake2s_compress_generic is weakly aliased to blake2s_generic. The
> current harness for function selection uses a function pointer, which is
> ordinarily inlined and resolved at compile time. But when Clang's CFI is
> enabled, CFI still triggers when making an indirect call via a weak
> symbol. This seems like a bug in Clang's CFI, as though it's bucketing
> weak symbols and strong symbols differently. It also only seems to
> trigger when "full LTO" mode is used, rather than "thin LTO".
> 
> [    0.000000][    T0] Kernel panic - not syncing: CFI failure (target: blake2s_compress_generic+0x0/0x1444)
> [    0.000000][    T0] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.16.0-mainline-06981-g076c855b846e #1
> [    0.000000][    T0] Hardware name: MT6873 (DT)
> [    0.000000][    T0] Call trace:
> [    0.000000][    T0]  dump_backtrace+0xfc/0x1dc
> [    0.000000][    T0]  dump_stack_lvl+0xa8/0x11c
> [    0.000000][    T0]  panic+0x194/0x464
> [    0.000000][    T0]  __cfi_check_fail+0x54/0x58
> [    0.000000][    T0]  __cfi_slowpath_diag+0x354/0x4b0
> [    0.000000][    T0]  blake2s_update+0x14c/0x178
> [    0.000000][    T0]  _extract_entropy+0xf4/0x29c
> [    0.000000][    T0]  crng_initialize_primary+0x24/0x94
> [    0.000000][    T0]  rand_initialize+0x2c/0x6c
> [    0.000000][    T0]  start_kernel+0x2f8/0x65c
> [    0.000000][    T0]  __primary_switched+0xc4/0x7be4
> [    0.000000][    T0] Rebooting in 5 seconds..
> 
> Nonetheless, the function pointer method isn't so terrific anyway, so
> this patch replaces it with a simple boolean, which also gets inlined
> away. This successfully works around the Clang bug.
> 
> In general, I'm not too keen on all of the indirection involved here; it
> clearly does more harm than good. Hopefully the whole thing can get
> cleaned up down the road when lib/crypto is overhauled more
> comprehensively. But for now, we go with a simple bandaid.
> 
> Fixes: 6048fdcc5f26 ("lib/crypto: blake2s: include as built-in")
> Reported-by: Miles Chen <miles.chen@mediatek.com>
> Cc: Nick Desaulniers <ndesaulniers@google.com>
> Cc: Sami Tolvanen <samitolvanen@google.com>
> Cc: Ard Biesheuvel <ardb@kernel.org>
> Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
> ---
>  arch/arm/crypto/blake2s-shash.c   |  4 ++--
>  arch/x86/crypto/blake2s-shash.c   |  4 ++--
>  crypto/blake2s_generic.c          |  4 ++--
>  include/crypto/internal/blake2s.h | 36 ++++++++++++++++++-------------
>  lib/crypto/blake2s.c              |  4 ++--
>  5 files changed, 29 insertions(+), 23 deletions(-)

There are some lines over 80 columns in this patch.

Otherwise this looks fine.  It would be really nice to fix this in clang,
though.

- Eric