Re: TLB flushes on fixmap changes

[Andy Lutomirski]

> On Aug 26, 2018, at 9:47 AM, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> 
>> On Sun, Aug 26, 2018 at 7:20 AM, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
>> 
>> 
>>>> On Aug 25, 2018, at 9:43 PM, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
>>>> 
>>>>> On Sat, Aug 25, 2018 at 9:21 PM, Andy Lutomirski <luto@xxxxxxxxxx> wrote:
>>>>> On Sat, Aug 25, 2018 at 7:23 PM, Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
>>>>> On Fri, 24 Aug 2018 21:23:26 -0700
>>>>> Andy Lutomirski <luto@xxxxxxxxxx> wrote:
>>>>>> Couldn't text_poke() use kmap_atomic()?  Or, even better, just change CR3?
>>>>> 
>>>>> No, since kmap_atomic() is only for x86_32 and highmem support kernel.
>>>>> In x86-64, it seems that returns just a page address. That is not
>>>>> good for text_poke, since it needs to make a writable alias for RO
>>>>> code page. Hmm, maybe, can we mimic copy_oldmem_page(), it uses ioremap_cache?
>>>>> 
>>>> 
>>>> I just re-read text_poke().  It's, um, horrible.  Not only is the
>>>> implementation overcomplicated and probably buggy, but it's SLOOOOOW.
>>>> It's totally the wrong API -- poking one instruction at a time
>>>> basically can't be efficient on x86.  The API should either poke lots
>>>> of instructions at once or should be text_poke_begin(); ...;
>>>> text_poke_end();.
>>>> 
>>>> Anyway, the attached patch seems to boot.  Linus, Kees, etc: is this
>>>> too scary of an approach?  With the patch applied, text_poke() is a
>>>> fantastic exploit target.  On the other hand, even without the patch
>>>> applied, text_poke() is every bit as juicy.
>>> 
>>> I tried to convince Ingo to use this method for doing "write rarely"
>>> and he soundly rejected it. :) I've always liked this because AFAICT,
>>> it's local to the CPU. I had proposed it in
>>> https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git/commit/?h=kspp/write-rarely&id=9ab0cb2618ebbc51f830ceaa06b7d2182fe1a52d
>> 
>> Ingo, can you clarify why you hate it?  I personally would rather use CR3, but CR0 seems like a fine first step, at least for text_poke.
> 
> Sorry, it looks like it was tglx, not Ingo:
> 
> https://lkml.kernel.org/r/alpine.DEB.2.20.1704071048360.1716@nanos
> 
> This thread is long, and one thing that I think went unanswered was
> "why do we want this to be fast?" the answer is: for doing page table
> updates. Page tables are becoming a bigger target for attacks now, and
> it's be nice if they could stay read-only unless they're getting
> updated (with something like this).
> 
> 

It kind of sounds like tglx would prefer the CR3 approach. And indeed my patch has a serious problem wrt the NMI code.