Re: [PATCH] arm64: Fix PTRACE_SYSEMU semantics

[Catalin Marinas]

On Wed, May 20, 2020 at 11:03:30AM +0100, Sudeep Holla wrote:
> On Tue, May 19, 2020 at 01:07:27PM +0100, Catalin Marinas wrote:
> > On Mon, May 18, 2020 at 12:41:20PM +0100, Will Deacon wrote:
> > > On Fri, May 15, 2020 at 06:22:53PM -0400, Keno Fischer wrote:
> > > > Quoth the man page:
> > > > ```
> > > >        If the tracee was restarted by PTRACE_SYSCALL or PTRACE_SYSEMU, the
> > > >        tracee enters syscall-enter-stop just prior to entering any system
> > > >        call (which will not be executed if the restart was using
> > > >        PTRACE_SYSEMU, regardless of any change made to registers at this
> > > >        point or how the tracee is restarted after this stop).
> > > > ```
> > > >
> > > > The parenthetical comment is currently true on x86 and powerpc,
> > > > but not currently true on arm64. arm64 re-checks the _TIF_SYSCALL_EMU
> > > > flag after the syscall entry ptrace stop. However, at this point,
> > > > it reflects which method was used to re-start the syscall
> > > > at the entry stop, rather than the method that was used to reach it.
> > > > Fix that by recording the original flag before performing the ptrace
> > > > stop, bringing the behavior in line with documentation and x86/powerpc.
> > > >
> > > > Signed-off-by: Keno Fischer <keno@xxxxxxxxxxxxxxxxxx>
> > > > ---
> > > >  arch/arm64/kernel/ptrace.c | 8 +++++---
> > > >  1 file changed, 5 insertions(+), 3 deletions(-)
> > > >
> > > > diff --git a/arch/arm64/kernel/ptrace.c b/arch/arm64/kernel/ptrace.c
> > > > index b3d3005d9515..b67b4d14aa17 100644
> > > > --- a/arch/arm64/kernel/ptrace.c
> > > > +++ b/arch/arm64/kernel/ptrace.c
> > > > @@ -1829,10 +1829,12 @@ static void tracehook_report_syscall(struct pt_regs *regs,
> > > >
> > > >  int syscall_trace_enter(struct pt_regs *regs)
> > > >  {
> > > > -	if (test_thread_flag(TIF_SYSCALL_TRACE) ||
> > > > -		test_thread_flag(TIF_SYSCALL_EMU)) {
> > > > +	u32 flags = READ_ONCE(current_thread_info()->flags) &
> > > > +		(_TIF_SYSCALL_EMU | _TIF_SYSCALL_TRACE);
> > > > +
> > > > +	if (flags) {
> > >
> > > nit: but I'd rather the '&' operation was in the conditional so that the
> > > 'flags' variable holds all of the flags.
> > >
> > > With that:
> > >
> > > Acked-by: Will Deacon <will@xxxxxxxxxx>
> > >
> > > Also needs:
> > >
> > > Cc: <stable@xxxxxxxxxxxxxxx>
> > > Fixes: f086f67485c5 ("arm64: ptrace: add support for syscall emulation")
> > >
> > > Catalin -- can you pick this up for 5.7 please, with my 'nit' addressed?
> >
> > I'll queue it with the above addressed. I think flags also needs to be
> > unsigned long rather than u32.
> >
> > However, before sending the pull request, I'd like Sudeep to confirm
> > that it doesn't break his original use-case for this feature.
> 
> I just tested it with my simple programs I had before. I have also asked
> teams working on gvisor to test. They have tested it and see no
> regression. I will ask them to reply here.
> 
> Tested-by: Sudeep Holla <sudeep.holla@xxxxxxx>

Thanks Sudeep.

-- 
Catalin