Re: [PATCH RFC v2 4/6] x86: Disable PTI on compatibility mode

[Andy Lutomirski]

>> On Feb 15, 2018, at 4:08 PM, Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>> 
>> On Thu, Feb 15, 2018 at 3:29 PM, Andy Lutomirski <luto@xxxxxxxxxx> wrote:
>> 
>> It's possible we could get away with adding the prctl but making the
>> default be that only the bitness that matches the program being run is
>> allowed.  After all, it's possible that CRIU is literally the only
>> program that switches bitness using the GDT.  (DOSEMU2 definitely does
>> cross-bitness stuff, but it uses the LDT as far as I know.)  And I've
>> never been entirely sure that CRIU fully counts toward the Linux
>> "don't break ABI" guarantee.
> 
> Ugh.
> 
> There are just _so_ many reasons to dislike that.
> 
> It's not that I don't think we could try to encourage it, but this
> whole "security depends on it being in sync" seems really like a
> fundamentally bad design.

If we're going to do Nadav's thing, I think we have no choice.  We could say that Nadav's idea of turning off PTI for 32-bit is just too messy, though.

> 
>> Linus, how would you feel about, by default, preventing 64-bit
>> programs from long-jumping to __USER32_CS and vice versa?
> 
> How? It's a standard GDT entry. Are you going to start switching the
> GDT around every context switch?

That's the idea.  We already switch out three GDT entries for TLS.  Switching two more isn't going to kill us.

> 
> I *thought* that user space can just do a far jump on its own. But
> it's so long since I had to care that I may have forgotten all the
> requirements for going between "compatibility mode" and real long
> mode.
> 
> I just feel this all is a nightmare. I can see how you would want to
> think that compatibility mode doesn't need PTI, but at the same time
> it feels like a really risky move to do this.
> 
> I can see one thread being in compatibiilty mode, and another being in
> long mode, and sharing the address space. But even with just one
> thread, I'm not seeing how you keep user mode from going from
> compatibility mode to L mode with just a far jump.
> 
> But maybe you have some clever scheme in mind that guarantees that
> there are no issues, or maybe I've just forgotten all the details of
> long mode vs compat mode.

The clever scheme is that we have a new (maybe default) compat-and-i-mean-it mode that removes the DPL=3 L code segment from the GDT and prevents opportunistic SYSRET.