Message-ID: <F854F567-567C-401D-8A80-B25F4F5987C6@zytor.com>
Date:   Mon, 01 Jun 2020 10:48:36 -0700
From:   hpa@...or.com
To:     Andy Lutomirski <luto@...capital.net>,
        Billy Laws <blaws05@...il.com>
CC:     krisman@...labora.com, gofmanp@...il.com, keescook@...omium.org,
        kernel@...labora.com, linux-kernel@...r.kernel.org,
        linux-mm@...ck.org, tglx@...utronix.de, wad@...omium.org
Subject: Re: [PATCH RFC] seccomp: Implement syscall isolation based on memory areas

On June 1, 2020 6:59:26 AM PDT, Andy Lutomirski <luto@...capital.net> wrote:
>
>
>> On Jun 1, 2020, at 2:23 AM, Billy Laws <blaws05@...il.com> wrote:
>> 
>> 
>>> 
>>> On May 30, 2020, at 5:26 PM, Gabriel Krisman Bertazi
><krisman@...labora.com> wrote:
>>> 
>>> Andy Lutomirski <luto@...capital.net> writes:
>>> 
>>>>>>> On May 29, 2020, at 11:00 PM, Gabriel Krisman Bertazi
><krisman@...labora.com> wrote:
>>>>>> 
>>>>>> Modern Windows applications are executing system call
>instructions
>>>>>> directly from the application's code without going through the
>WinAPI.
>>>>>> This breaks Wine emulation, because it doesn't have a chance to
>>>>>> intercept and emulate these syscalls before they are submitted to
>Linux.
>>>>>> 
>>>>>> In addition, we cannot simply trap every system call of the
>application
>>>>>> to userspace using PTRACE_SYSEMU, because performance would
>suffer,
>>>>>> since our main use case is to run Windows games over Linux. 
>Therefore,
>>>>>> we need some in-kernel filtering to decide whether the syscall
>was
>>>>>> issued by the wine code or by the windows application.
>>>> 
>>>> Do you really need in-kernel filtering?  What if you could have
>>>> efficient userspace filtering instead?  That is, set something up
>so
>>>> that all syscalls, except those from a special address, are
>translated
>>>> to CALL thunk where the thunk is configured per task.  Then the
>thunk
>>>> can do whatever emulation is needed.
>>> 
>>> Hi,
>>> 
>>> I suggested something similar to my customer, by using
>>> libsyscall-intercept.  The idea would be overwritting the syscall
>>> instruction with a call to the entry point.  I'm not a specialist on
>the
>>> specifics of Windows games, (cc'ed Paul Gofman, who can provide more
>>> details on that side), but as far as I understand, the reason why
>that
>>> is not feasible is that the anti-cheat protection in games will
>abort
>>> execution if the binary region was modified either on-disk or
>in-memory.
>>> 
>>> Is there some mechanism to do that without modiyfing the
>application?
>> 
>> Hi,
>> 
>> I work on an emulator for the Nintendo Switch that uses a similar
>technique,
>> in our testing it works very well and is much more performant than
>even
>> PTRACE_SYSEMU.
>> 
>> To work around DRM reading the memory contents I think mprotect could
>> be used, after patching the syscall a copy of the original code could
>be
>> kept somewhere in memory and the patched region mapped --X.
>> With this, any time the DRM attempts to read to the patched region
>and
>> perform integrity checks it will cause a segfault and a branch to the
>> signal handler. This handler can then return the contents of the
>original,
>> unpatched region to satisfy them checks.
>> 
>> Are memory contents checked by DRM solutions too often for this to be
>> performant?
>
>A bigger issue is that hardware support for —X is quite spotty. There
>is no x86 CPU that can do it cleanly in a bare metal setup, and client
>CPUs that can do it at all without hypervisor help may be nonexistent.
>I don’t know if the ARM situation is much better.
>
>> --
>> Billy Laws
>>> 
>>>> Getting the details and especially the interaction with any seccomp
>>>> filters that may be installed right could be tricky, but the
>performance
>>>> should be decent, at least on non-PTI systems.
>>>> 
>>>> (If we go this route, I suspect that the correct interaction with
>>>> seccomp is that this type of redirection takes precedence over
>seccomp
>>>> and seccomp filters are not invoked for redirected syscalls. After
>all,
>>>> a redirected syscall is, functionally, not a syscall at all.)
>>>> 
>>> 
>>> 
>>> --
>>> Gabriel Krisman Bertazi

Running these things in a minimal VM container would allow this kind of filtering/trapping to be done in the VMM, too. I don't know how many layers deep you invoke native Linux libraries, and so if the option would exist to use out-of-range system call numbers for the Linux system numbers?
-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.

Powered by blists - more mailing lists