[1/4] MIPS/seccomp: Fix indirect syscall args

Since commit 669c4092225f ("MIPS: Give __secure_computing() access to
syscall arguments."), upon syscall entry when seccomp is enabled,
syscall_trace_enter() passes a carefully prepared struct seccomp_data
containing syscall arguments to __secure_computing(). Unfortunately it
directly uses mips_get_syscall_arg() and fails to take into account the
indirect O32 system calls (i.e. syscall(2)) which put the system call
number in a0 and have the arguments shifted up by one entry.

We can't just revert that commit as samples/bpf/tracex5 would break
again, so use syscall_get_arguments() which already takes indirect
syscalls into account instead of directly using mips_get_syscall_arg(),
similar to what populate_seccomp_data() does.

This also removes the redundant error checking of the
mips_get_syscall_arg() return value (get_user() already zeroes the
result if an argument from the stack can't be loaded).

Reported-by: James Cowgill <James.Cowgill@imgtec.com>
Fixes: 669c4092225f ("MIPS: Give __secure_computing() access to syscall arguments.")
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: David Daney <david.daney@cavium.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Will Drewry <wad@chromium.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: netdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: linux-mips@linux-mips.org
---
It would have been much simpler for MIPS arch code to just pass a NULL
seccomp_data to secure_computing() so populate_seccomp_data() would take
care of fetching arguments, as it did for MIPS prior to commit
669c4092225f ("MIPS: Give __secure_computing() access to syscall
arguments."), but as that commit mentions it breaks samples/bpf/tracex5,
which relies on sd being non-NULL at entry to __seccomp_filter().

Arguably the samples/bpf/tracex5 test is flawed, at least for every arch
except x86 (and now MIPS).
---
 arch/mips/kernel/ptrace.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

On Fri, Aug 11, 2017 at 1:56 PM, James Hogan <james.hogan@imgtec.com> wrote:
> Since commit 669c4092225f ("MIPS: Give __secure_computing() access to
> syscall arguments."), upon syscall entry when seccomp is enabled,
> syscall_trace_enter() passes a carefully prepared struct seccomp_data
> containing syscall arguments to __secure_computing(). Unfortunately it
> directly uses mips_get_syscall_arg() and fails to take into account the
> indirect O32 system calls (i.e. syscall(2)) which put the system call
> number in a0 and have the arguments shifted up by one entry.
>
> We can't just revert that commit as samples/bpf/tracex5 would break
> again, so use syscall_get_arguments() which already takes indirect
> syscalls into account instead of directly using mips_get_syscall_arg(),
> similar to what populate_seccomp_data() does.
>
> This also removes the redundant error checking of the
> mips_get_syscall_arg() return value (get_user() already zeroes the
> result if an argument from the stack can't be loaded).
>
> Reported-by: James Cowgill <James.Cowgill@imgtec.com>
> Fixes: 669c4092225f ("MIPS: Give __secure_computing() access to syscall arguments.")
> Signed-off-by: James Hogan <james.hogan@imgtec.com>
> Cc: Ralf Baechle <ralf@linux-mips.org>
> Cc: David Daney <david.daney@cavium.com>
> Cc: Kees Cook <keescook@chromium.org>
> Cc: Andy Lutomirski <luto@amacapital.net>
> Cc: Will Drewry <wad@chromium.org>
> Cc: Oleg Nesterov <oleg@redhat.com>
> Cc: Alexei Starovoitov <ast@kernel.org>
> Cc: Daniel Borkmann <daniel@iogearbox.net>
> Cc: netdev@vger.kernel.org
> Cc: linux-kernel@vger.kernel.org
> Cc: linux-mips@linux-mips.org
> ---
> It would have been much simpler for MIPS arch code to just pass a NULL
> seccomp_data to secure_computing() so populate_seccomp_data() would take
> care of fetching arguments, as it did for MIPS prior to commit
> 669c4092225f ("MIPS: Give __secure_computing() access to syscall
> arguments."), but as that commit mentions it breaks samples/bpf/tracex5,
> which relies on sd being non-NULL at entry to __seccomp_filter().
>
> Arguably the samples/bpf/tracex5 test is flawed, at least for every arch
> except x86 (and now MIPS).

Weird. Yeah, that sample is broken. Allowing NULL sd is totally fine.
The point is that seccomp will use syscall_get_arguments() when it's
NULL (which is effectively what this is doing...)

The reason sd can be _non_-NULL is when an architecture has access to
the args in some way that might be faster than calling
syscall_get_arguments().

Regardless, I'm fine with this change. It should either be this or
reverting 669c4092225f, but it looks like kprobes of
__seccomp_filter() is desired on MIPS...

Reviewed-by: Kees Cook <keescook@chromium.org>

-Kees

> ---
>  arch/mips/kernel/ptrace.c | 10 ++++------
>  1 file changed, 4 insertions(+), 6 deletions(-)
>
> diff --git a/arch/mips/kernel/ptrace.c b/arch/mips/kernel/ptrace.c
> index 6dd13641a418..1395654cfc8d 100644
> --- a/arch/mips/kernel/ptrace.c
> +++ b/arch/mips/kernel/ptrace.c
> @@ -872,15 +872,13 @@ asmlinkage long syscall_trace_enter(struct pt_regs *regs, long syscall)
>         if (unlikely(test_thread_flag(TIF_SECCOMP))) {
>                 int ret, i;
>                 struct seccomp_data sd;
> +               unsigned long args[6];
>
>                 sd.nr = syscall;
>                 sd.arch = syscall_get_arch();
> -               for (i = 0; i < 6; i++) {
> -                       unsigned long v, r;
> -
> -                       r = mips_get_syscall_arg(&v, current, regs, i);
> -                       sd.args[i] = r ? 0 : v;
> -               }
> +               syscall_get_arguments(current, regs, 0, 6, args);
> +               for (i = 0; i < 6; i++)
> +                       sd.args[i] = args[i];
>                 sd.instruction_pointer = KSTK_EIP(current);
>
>                 ret = __secure_computing(&sd);
> --
> 2.13.2
>