Re: [PATCH 2/6] x86/ia32: Fix ia32_restore_sigcontext AC leak

From: Andy Lutomirski
Date: Mon Feb 25 2019 - 11:49:24 EST


On Mon, Feb 25, 2019 at 8:37 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Mon, Feb 25, 2019 at 08:29:12AM -0800, Andy Lutomirski wrote:
>
> > > diff --git a/arch/x86/ia32/ia32_signal.c b/arch/x86/ia32/ia32_signal.c
> > > index 321fe5f5d0e9..e04eeeddcc35 100644
> > > --- a/arch/x86/ia32/ia32_signal.c
> > > +++ b/arch/x86/ia32/ia32_signal.c
> > > @@ -53,17 +53,16 @@
> > > #define GET_SEG(seg) ({ \
> > > unsigned short tmp; \
> > > get_user_ex(tmp, &sc->seg); \
> > > - tmp; \
> > > + tmp | 3; \
> > > })
> > >
> >
> > Drop this part.
> >
> > > #define COPY_SEG_CPL3(seg) do { \
> > > - regs->seg = GET_SEG(seg) | 3; \
> > > + regs->seg = GET_SEG(seg); \
> > > } while (0)
> >
> > And this.
> >
> > Unfortunately, whether we want the | 3 varies by segment. For FS and
> > GS, we definitely donât want it, since 0 is a common and important
> > value, and 3 is a deeply screwed up value. (3 is legal to *write* to
> > GS, and it sticks, but IRET silently changes it to 0, because the
> > original 386 designers (I assume) confused themselves.
>
> > >
> > > #define RELOAD_SEG(seg) { \
> > > - unsigned int pre = GET_SEG(seg); \
> > > + unsigned int pre = (seg); \
> > > unsigned int cur = get_user_seg(seg); \
> > > - pre |= 3; \
>
> And here ?
>
> > > if (pre != cur) \
> > > set_user_seg(seg, pre); \
> > > }
>
> Thing is; afaict the current code _always_ does |3 on any GET_SET()
> result.
>
> If you want to change that; I'm fine with that, but lets not do that in
> this same patch.

Ugh, you're right. I bet I can come up with some awful case involving
ptrace and sigreturn where this causes problems.

I have a patch series I need to dust off that deletes this entire
file. It's this and its parents:

https://git.kernel.org/pub/scm/linux/kernel/git/luto/linux.git/commit/?h=execve&id=0618fe90d8224979f70b15d33dcae75a403592e5

>
> So then?
>
> diff --git a/arch/x86/ia32/ia32_signal.c b/arch/x86/ia32/ia32_signal.c
> index 321fe5f5d0e9..4d5fcd47ab75 100644
> --- a/arch/x86/ia32/ia32_signal.c
> +++ b/arch/x86/ia32/ia32_signal.c
> @@ -61,9 +61,8 @@
> } while (0)
>
> #define RELOAD_SEG(seg) { \
> - unsigned int pre = GET_SEG(seg); \
> + unsigned int pre = (seg) | 3; \
> unsigned int cur = get_user_seg(seg); \
> - pre |= 3; \
> if (pre != cur) \
> set_user_seg(seg, pre); \
> }
> @@ -72,6 +71,7 @@ static int ia32_restore_sigcontext(struct pt_regs *regs,
> struct sigcontext_32 __user *sc)
> {
> unsigned int tmpflags, err = 0;
> + u16 gs, fs, es, ds;
> void __user *buf;
> u32 tmp;
>
> @@ -79,16 +79,10 @@ static int ia32_restore_sigcontext(struct pt_regs *regs,
> current->restart_block.fn = do_no_restart_syscall;
>
> get_user_try {
> - /*
> - * Reload fs and gs if they have changed in the signal
> - * handler. This does not handle long fs/gs base changes in
> - * the handler, but does not clobber them at least in the
> - * normal case.
> - */
> - RELOAD_SEG(gs);
> - RELOAD_SEG(fs);
> - RELOAD_SEG(ds);
> - RELOAD_SEG(es);
> + gs = GET_SEG(gs);
> + fs = GET_SEG(fs);
> + ds = GET_SEG(ds);
> + es = GET_SEG(es);
>
> COPY(di); COPY(si); COPY(bp); COPY(sp); COPY(bx);
> COPY(dx); COPY(cx); COPY(ip); COPY(ax);
> @@ -106,6 +100,17 @@ static int ia32_restore_sigcontext(struct pt_regs *regs,
> buf = compat_ptr(tmp);
> } get_user_catch(err);
>
> + /*
> + * Reload fs and gs if they have changed in the signal
> + * handler. This does not handle long fs/gs base changes in
> + * the handler, but does not clobber them at least in the
> + * normal case.
> + */
> + RELOAD_SEG(gs);
> + RELOAD_SEG(fs);
> + RELOAD_SEG(ds);
> + RELOAD_SEG(es);
> +
> err |= fpu__restore_sig(buf, 1);
>
> force_iret();

Looks good to me. The order of the restores shouldn't matter at all.