Re: [PATCH v2 0/6] x86/alternatives: text_poke() fixes

[Nadav Amit]

at 12:02 PM, Nadav Amit <namit@xxxxxxxxxx> wrote:

> at 11:56 AM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> 
>> On Sun, Sep 02, 2018 at 10:32:18AM -0700, Nadav Amit wrote:
>>> This patch-set addresses some issues that were raised in a recent
>>> correspondence and might affect the security and the correctness of code
>>> patching. (Note that patching performance is not addressed by this
>>> patch-set).
>>> 
>>> The main issue that the patches deal with is the fact that the fixmap
>>> PTEs that are used for patching are available for access from other
>>> cores and might be exploited. They are not even flushed from the TLB in
>>> remote cores, so the risk is even higher. Address this issue by
>>> introducing a temporary mm that is only used during patching.
>>> Unfortunately, due to init ordering, fixmap is still used during
>>> boot-time patching. Future patches can eliminate the need for it.
>> 
>> Remind me; why are we doing it like this instead of fixing fixmap?
>> Because while this fixes the text_poke crud, it does leave fixmap
>> broken.
> 
> Do you have other fixmap mappings in mind that are modified after boot?

Oh.. I misunderstood you. You mean: why not to make the fixmap mappings that
are used for text_poke() as private ones.

Well, the main reason is that it can require synchronizations of the
different page-tables whenever a module is loaded/unloaded. The fixmap
region shares a PGD and PUD with the modules area in x86-64.

In contrast, the proposed solution uses a different PGD, so no
synchronization between page-tables is needed when modules are loaded.
Remember that module memory is allocated even when BPF programs are
installed, which can be rather common scenario.