Re: [PATCH 24/24] x86/syscall/64: Move the checking for sysret to C code

[Lai Jiangshan]

On 2021/9/10 15:20, Nikolay Borisov wrote:
> On 31.08.21 г. 20:50, Lai Jiangshan wrote:
> > From: Lai Jiangshan <laijs@xxxxxxxxxxxxxxxxx>
> >
> > Like do_fast_syscall_32() which checks whether it can return to userspace
> > via fast instructions before the function returns, do_syscall_64()
> > also checks whether it can use sysret to return to userspace before
> > do_syscall_64() returns via C code.  And a bunch of ASM code can be removed.
> >
> > No functional change intended.
> >
> > Signed-off-by: Lai Jiangshan <laijs@xxxxxxxxxxxxxxxxx>
>
> <snip>
>
> > +/*
> > + * Check if it can use SYSRET.
> > + *
> > + * Try to use SYSRET instead of IRET if we're returning to
> > + * a completely clean 64-bit userspace context.
> > + *
> > + * Returns 0 to return using IRET or 1 to return using SYSRET.
> > + */
> > +static __always_inline int can_sysret(struct pt_regs *regs)
>
> nit: Since this is a predicate function why not simply return bool ?

I don't have any preference.

The choice came from my limitation of the needed knowledge.

I followed the design of do_fast_syscall_32() which returns a 4-byte word
to indicate if it can fast return to userspace,  and I know how to test the
result in ASM for a 4-byte word.  If it was a bool, I don't know how to
test the result in ASM.

> > +{
> > +	/* In the Xen PV case we must use iret anyway. */
> > +	if (static_cpu_has(X86_FEATURE_XENPV))
> > +		return 0;
> > +
> > +	/* SYSRET requires RCX == RIP && R11 == RFLAGS */
> > +	if (regs->ip != regs->cx || regs->flags != regs->r11)
> > +		return 0;
> > +
> > +	/* CS and SS must match SYSRET */
> > +	if (regs->cs != __USER_CS || regs->ss != __USER_DS)
> > +		return 0;
> > +
> > +	/*
> > +	 * On Intel CPUs, SYSRET with non-canonical RCX/RIP will #GP
> > +	 * in kernel space.  This essentially lets the user take over
> > +	 * the kernel, since userspace controls RSP.
> > +	 */
> > +	if (regs->cx != canonical_address(regs->cx))
> > +		return 0;
> > +
> > +	/*
> > +	 * SYSCALL clears RF when it saves RFLAGS in R11 and SYSRET cannot
> > +	 * restore RF properly. If the slowpath sets it for whatever reason, we
> > +	 * need to restore it correctly.
> > +	 *
> > +	 * SYSRET can restore TF, but unlike IRET, restoring TF results in a
> > +	 * trap from userspace immediately after SYSRET.  This would cause an
> > +	 * infinite loop whenever #DB happens with register state that satisfies
> > +	 * the opportunistic SYSRET conditions.  For example, single-stepping
> > +	 * this user code:
> > +	 *
> > +	 *           movq	$stuck_here, %rcx
> > +	 *           pushfq
> > +	 *           popq %r11
> > +	 *   stuck_here:
> > +	 *
> > +	 * would never get past 'stuck_here'.
> > +	 */
> > +	if (regs->r11 & (X86_EFLAGS_RF | X86_EFLAGS_TF))
> > +		return 0;
> > +
> > +	return 1;
> > +}
> > +
> > +/* Returns 0 to return using IRET or 1 to return using SYSRET. */
> > +__visible noinstr int do_syscall_64(struct pt_regs *regs, int nr)
>
> nit: Ditto about bool
>
> >   {
> >   	add_random_kstack_offset();
> >   	nr = syscall_enter_from_user_mode(regs, nr);
> > @@ -84,6 +154,7 @@ __visible noinstr void do_syscall_64(struct pt_regs *regs, int nr)
> >   
> >   	instrumentation_end();
> >   	syscall_exit_to_user_mode(regs);
> > +	return can_sysret(regs);
> >   }
> >   #endif
> >   
>
> <snip>