Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 23 Feb 2021 12:28:49 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Andy Lutomirski <luto@kernel.org>
Cc:     x86@kernel.org, LKML <linux-kernel@vger.kernel.org>,
        stable@vger.kernel.org
Subject: Re: [PATCH 2/3] x86/entry: Fix entry/exit mismatch on failed fast
 32-bit syscalls
Message-ID: <YDTm8Q/cvBcfzhPn@hirez.programming.kicks-ass.net>
References: <cover.1614059335.git.luto@kernel.org>
 <a0025117242488a30621fb9858918802532f9ee9.1614059335.git.luto@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <a0025117242488a30621fb9858918802532f9ee9.1614059335.git.luto@kernel.org>
Precedence: bulk

On Mon, Feb 22, 2021 at 09:50:28PM -0800, Andy Lutomirski wrote:
> On a 32-bit fast syscall that fails to read its arguments from user
> memory, the kernel currently does syscall exit work but not
> syscall exit work.  This would confuse audit and ptrace.
> 
> This is a minimal fix intended for ease of backporting.  A more
> complete cleanup is coming.
> 
> Cc: stable@vger.kernel.org
> Fixes: 0b085e68f407 ("x86/entry: Consolidate 32/64 bit syscall entry")
> Signed-off-by: Andy Lutomirski <luto@kernel.org>
> ---
>  arch/x86/entry/common.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
> index 0904f5676e4d..cf4dcf346ca8 100644
> --- a/arch/x86/entry/common.c
> +++ b/arch/x86/entry/common.c
> @@ -128,7 +128,8 @@ static noinstr bool __do_fast_syscall_32(struct pt_regs *regs)
>  		regs->ax = -EFAULT;
>  
>  		instrumentation_end();
> -		syscall_exit_to_user_mode(regs);
> +		local_irq_disable();
> +		exit_to_user_mode();
>  		return false;
>  	}

I'm confused, twice. Once by your Changelog, and second by the actual
patch. Shouldn't every return to userspace pass through
exit_to_user_mode_prepare() ? We shouldn't ignore NEED_RESCHED or
NOTIFY_RESUME, both of which can be set I think, even if the SYSCALL
didn't actually do anything.