Re: [PATCH v3 1/3] x86: Implement arch_prctl(ARCH_VSYSCALL_CONTROL) to disable vsyscall

[Andy Lutomirski]

On 1/5/22 08:02, Florian Weimer wrote:
> Distributions struggle with changing the default for vsyscall
> emulation because it is a clear break of userspace ABI, something
> that should not happen.
>
> The legacy vsyscall interface is supposed to be used by libcs only,
> not by applications.  This commit adds a new arch_prctl request,
> ARCH_VSYSCALL_CONTROL, with one argument.  If the argument is 0,
> executing vsyscalls will cause the process to terminate.  Argument 1
> turns vsyscall back on (this is mostly for a largely theoretical
> CRIU use case).
>
> Newer libcs can use a zero ARCH_VSYSCALL_CONTROL at startup to disable
> vsyscall for the process.  Legacy libcs do not perform this call, so
> vsyscall remains enabled for them.  This approach should achieves
> backwards compatibility (perfect compatibility if the assumption that
> only libcs use vsyscall is accurate), and it provides full hardening
> for new binaries.
>
> The chosen value of ARCH_VSYSCALL_CONTROL should avoid conflicts
> with other x86-64 arch_prctl requests.  The fact that with
> vsyscall=emulate, reading the vsyscall region is still possible
> even after a zero ARCH_VSYSCALL_CONTROL is considered limitation
> in the current implementation and may change in a future kernel
> version.
>
> Future arch_prctls requests commonly used at process startup can imply
> ARCH_VSYSCALL_CONTROL with a zero argument, so that a separate system
> call for disabling vsyscall is avoided.
>
> Signed-off-by: Florian Weimer <fweimer@xxxxxxxxxx>
> Acked-by: Andrei Vagin <avagin@xxxxxxxxx>
> ---
> v3: Remove warning log message.  Split out test.
> v2: ARCH_VSYSCALL_CONTROL instead of ARCH_VSYSCALL_LOCKOUT.  New tests
>      for the toggle behavior.  Implement hiding [vsyscall] in
>      /proc/PID/maps and test it.  Various other test fixes cleanups
>      (e.g., fixed missing second argument to gettimeofday).
>
> arch/x86/entry/vsyscall/vsyscall_64.c | 7 ++++++-
>   arch/x86/include/asm/mmu.h            | 6 ++++++
>   arch/x86/include/uapi/asm/prctl.h     | 2 ++
>   arch/x86/kernel/process_64.c          | 7 +++++++
>   4 files changed, 21 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/entry/vsyscall/vsyscall_64.c b/arch/x86/entry/vsyscall/vsyscall_64.c
> index fd2ee9408e91..6fc524b9f232 100644
> --- a/arch/x86/entry/vsyscall/vsyscall_64.c
> +++ b/arch/x86/entry/vsyscall/vsyscall_64.c
> @@ -174,6 +174,9 @@ bool emulate_vsyscall(unsigned long error_code,
>   
>   	tsk = current;
>   
> +	if (tsk->mm->context.vsyscall_disabled)
> +		goto sigsegv;
> +

Is there a reason you didn't just change the check earlier in the 
function to:

if (vsyscall_mode == NONE || current->mm->context.vsyscall_disabled)

Also, I still think the prctl should not be available if 
vsyscall=emulate.  Either we should fully implement it or we should not 
implement.  We could even do:

pr_warn_once("userspace vsyscall hardening request ignored because you 
have vsyscall=emulate.  Unless you absolutely need vsyscall=emulate, 
update your system to use vsyscall=xonly.\n");

and thus encourage good behavior.

--Andy