linux-kernel - Re: [PATCH 1/2] powerpc/signal: Fix handling of SA

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] [day] [month] [year] [list]

Message-ID: <381e3dc3-350c-c373-bc45-8dafd72ec011@csgroup.eu>
Date:   Fri, 4 Feb 2022 11:00:32 +0000
From:   Christophe Leroy <christophe.leroy@...roup.eu>
To:     Michael Ellerman <mpe@...erman.id.au>
CC:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "linuxppc-dev@...ts.ozlabs.org" <linuxppc-dev@...ts.ozlabs.org>,
        Tulio Magno Quites Machado Filho <tuliom@...ux.ibm.com>
Subject: Re: [PATCH 1/2] powerpc/signal: Fix handling of SA_RESTORER sigaction
 flag



Le 04/02/2022 à 11:22, Michael Ellerman a écrit :
> Christophe Leroy <christophe.leroy@...roup.eu> writes:
>> powerpc advertises support of SA_RESTORER sigaction flag.
>>
>> Make it the truth.
>>
>> Cc: stable@...r.kernel.org
>> Signed-off-by: Christophe Leroy <christophe.leroy@...roup.eu>
>> ---
>>   arch/powerpc/kernel/signal_32.c | 8 ++++++--
>>   arch/powerpc/kernel/signal_64.c | 4 +++-
>>   2 files changed, 9 insertions(+), 3 deletions(-)
> 
> Hi Christophe,
> 
> I dug into the history a bit on this.
> 
> The 32-bit port originally did not define SA_RESTORER in
> include/asm-ppc/signal.h, but it was added in 2.1.79.
> 
>    https://github.com/mpe/linux-fullhistory/commit/4e7e9c0d54ff5725a73d2210a950f8bc0f225073
> 
> That commit added SA_RESTORER to the header, added code to get/set it in
> sys_sigaction(), but didn't add any code to use it for signal delivery.
> 
> 
> The 64-bit port was merged with SA_RESTORER already defined in
> include/asm-ppc64/signal.h:
> 
>    https://github.com/mpe/linux-fullhistory/commit/c3aa9878533e724f639852c3d951e6a169e04081
>    
> Similarly there was code to set/get it in sys_sigaction(), but no code
> to use it for signal delivery.
> 
> 
> Later the two ports were merged, and the headers were moved and
> disintegrated into uapi, so we end up today with SA_RESTORER defined in
> arch/powerpc/include/uapi/asm/signal.h, but no code to use it.
> 
> So essentially we've had SA_RESTORER defined since ancient kernels, but
> never actually supported using it for anything.
> 
> 
> One problem with enabling it now is there's no way for userspace to
> determine if it's on a fixed kernel or not. That makes it unusable for
> userspace, unless it does version checks, or is happy to break on all
> old kernels (not likely). We could solve that by defining
> SA_RESTORER_FIXED or something, but that's slightly gross.
> 
> It's also described in the man page as "Not intended for application
> use", ie. it's intended for use by libc. I'm not sure there's any value
> in adding support for it to the kernel unless we know there's interest
> from glibc/musl in using it.
> 
> So my inclination is that we should *not* add support for it, rather we
> should leave it unimplemented and remove SA_RESTORER from the header.
> There's precedent in riscv for not supporting it at all.
> 

Nowadays, stacks are mapped noexec, so the fallback stack trampoline 
cannot work anymore. If a process doesn't want exec stack and doesn't 
want to map the VDSO, the SA_RESTORER is the only alternative to get 
signal working.

On several architectures including arm64 and s390 only VDSO and 
SA_RESTORER are supported, on stack signal trampoline is not supported 
anymore.

So my idea was to first implement SA_RESTORER and then as a second step 
to retire the on stack signal trampoline which has become useless with 
noexec stacks.

See 
https://elixir.bootlin.com/linux/v5.17-rc1/source/arch/arm64/kernel/signal.c#L761

or 
https://elixir.bootlin.com/linux/v5.17-rc1/source/arch/s390/kernel/signal.c#L337

Christophe