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:   Wed, 23 Dec 2020 18:54:43 -0800 (PST)
Subject:     Re: [PATCH] riscv: return -ENOSYS for syscall -1
In-Reply-To: <20201223082404.GB24581@infradead.org>
CC:     tycho@tycho.pizza, schwab@suse.de, david.abdurachmanov@sifive.com,
        linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org,
        Paul Walmsley <paul.walmsley@sifive.com>
From:   Palmer Dabbelt <palmer@dabbelt.com>
To:     Christoph Hellwig <hch@infradead.org>
Message-ID: <mhng-c15e4fab-6487-4875-a392-85669982b8b1@palmerdabbelt-glaptop>
Mime-Version: 1.0 (MHng)
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On Wed, 23 Dec 2020 00:24:04 PST (-0800), Christoph Hellwig wrote:
> On Tue, Dec 22, 2020 at 09:22:19AM -0700, Tycho Andersen wrote:
>> On Mon, Dec 21, 2020 at 11:52:00PM +0100, Andreas Schwab wrote:
>> > Properly return -ENOSYS for syscall -1 instead of leaving the return value
>> > uninitialized.  This fixes the strace teststuite.
>> >
>> > Fixes: 5340627e3fe0 ("riscv: add support for SECCOMP and SECCOMP_FILTER")
>> > Signed-off-by: Andreas Schwab <schwab@suse.de>
>> > ---
>> >  arch/riscv/kernel/entry.S | 9 +--------
>> >  1 file changed, 1 insertion(+), 8 deletions(-)
>> >
>> > diff --git a/arch/riscv/kernel/entry.S b/arch/riscv/kernel/entry.S
>> > index 524d918f3601..d07763001eb0 100644
>> > --- a/arch/riscv/kernel/entry.S
>> > +++ b/arch/riscv/kernel/entry.S
>> > @@ -186,14 +186,7 @@ check_syscall_nr:
>> >  	 * Syscall number held in a7.
>> >  	 * If syscall number is above allowed value, redirect to ni_syscall.
>> >  	 */
>> > -	bge a7, t0, 1f
>> > -	/*
>> > -	 * Check if syscall is rejected by tracer, i.e., a7 == -1.
>> > -	 * If yes, we pretend it was executed.
>> > -	 */
>> > -	li t1, -1
>> > -	beq a7, t1, ret_from_syscall_rejected
>> > -	blt a7, t1, 1f
>> > +	bgeu a7, t0, 1f
>>
>> IIUC, this is all dead code anyway for the path where seccomp actually
>> rejects the syscall, since it should do the rejection directly in
>> handle_syscall_trace_enter(), which is called above this hunk. So it
>> seems good to me.
>
> That change really needs to be documented in the commit log, or even
> better split into a separate patch (still documented of course!).

Unless I'm missing something, this is already how it works already?
handle_syscall_trace_enter is checking the result of do_syscall_trace_enter(),
which checks secure_computing().  When secure_computing() rejects the syscall
we already ended up rejecting the syscall, so this code wasn't doing anything
for the case it was supposed to handle.

It was, however, intercepting syscall number -1 when we weren't rejecting the
syscall and directly exiting rather than calling sys_ni_syscall.  That would,
at a bare minimum, result in an uninitialized return value.  It also breaks the
pairing of trace_sys_enter() and trace_sys_exit(), which doesn't smell like a
good idea.