Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20190225124330.613028745@infradead.org> <20190225125231.936952143@infradead.org>
 <CALCETrW-p4iF-RMecyNt1JDwKh74c_tw67FCW7Yr_Fd3adtGUw@mail.gmail.com>
 <20190225161003.GL32477@hirez.programming.kicks-ass.net> <3F83AE74-3F9E-42BB-8F9C-3033F72B3EF5@amacapital.net>
 <20190225163745.GF32494@hirez.programming.kicks-ass.net>
In-Reply-To: <20190225163745.GF32494@hirez.programming.kicks-ass.net>
From:   Andy Lutomirski <luto@kernel.org>
Date:   Mon, 25 Feb 2019 08:49:06 -0800
Message-ID: <CALCETrU7nW8JVbu79VppNhn2RSxvSiqanXgOkwboUqbJYnQyTQ@mail.gmail.com>
Subject: Re: [PATCH 2/6] x86/ia32: Fix ia32_restore_sigcontext AC leak
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     Linus Torvalds <torvalds@linux-foundation.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        "H. Peter Anvin" <hpa@zytor.com>,
        Julien Thierry <julien.thierry@arm.com>,
        Will Deacon <will.deacon@arm.com>,
        Ingo Molnar <mingo@kernel.org>,
        Catalin Marinas <catalin.marinas@arm.com>,
        James Morse <james.morse@arm.com>, valentin.schneider@arm.com,
        Brian Gerst <brgerst@gmail.com>,
        Josh Poimboeuf <jpoimboe@redhat.com>,
        Andrew Lutomirski <luto@kernel.org>,
        Borislav Petkov <bp@alien8.de>,
        Denys Vlasenko <dvlasenk@redhat.com>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Feb 25, 2019 at 8:37 AM Peter Zijlstra <peterz@infradead.org> 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 =3D GET_SEG(seg) | 3;            \
> > > +    regs->seg =3D GET_SEG(seg);            \
> > > } while (0)
> >
> > And this.
> >
> > Unfortunately, whether we want the | 3 varies by segment. For FS and
> > GS, we definitely don=E2=80=99t want it, since 0 is a common and import=
ant
> > 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 =3D GET_SEG(seg);    \
> > > +    unsigned int pre =3D (seg);        \
> > >    unsigned int cur =3D get_user_seg(seg);    \
> > > -    pre |=3D 3;                \
>
> And here ?
>
> > >    if (pre !=3D 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=3D=
execve&id=3D0618fe90d8224979f70b15d33dcae75a403592e5

>
> 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 =3D GET_SEG(seg);        \
> +       unsigned int pre =3D (seg) | 3;           \
>         unsigned int cur =3D get_user_seg(seg);   \
> -       pre |=3D 3;                               \
>         if (pre !=3D 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 =3D 0;
> +       u16 gs, fs, es, ds;
>         void __user *buf;
>         u32 tmp;
>
> @@ -79,16 +79,10 @@ static int ia32_restore_sigcontext(struct pt_regs *re=
gs,
>         current->restart_block.fn =3D 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 =3D GET_SEG(gs);
> +               fs =3D GET_SEG(fs);
> +               ds =3D GET_SEG(ds);
> +               es =3D 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 *r=
egs,
>                 buf =3D 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 |=3D fpu__restore_sig(buf, 1);
>
>         force_iret();

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