Prevent inconsistent CPU state after sequence of dlclose/dlopen

[Mathieu Desnoyers]

On 2025-01-10 12:24, Adhemerval Zanella Netto wrote:
> 
> 
> On 10/01/25 14:15, Mathieu Desnoyers wrote:
>> On 2025-01-10 12:10, Florian Weimer wrote:
>>> * Mathieu Desnoyers:
>>>
>>>> On 2025-01-10 11:54, Peter Zijlstra wrote:
>>>>> On Fri, Jan 10, 2025 at 10:55:36AM -0500, Mathieu Desnoyers wrote:
>>>>>> Hi,
>>>>>>
>>>>>> I was discussing with Mark Rutland recently, and he pointed out that a
>>>>>> sequence of dlclose/dlopen mapping new code at the same addresses in
>>>>>> multithreaded environments is an issue on ARM, and possibly on Intel/AMD
>>>>>> with the newer TLB broadcast maintenance.
>>>>> What is the exact race? Should not munmap() invalidate the TLBs
>>>>> before
>>>>> it allows overlapping mmap() to complete?
>>>>
>>>> The race Mark mentioned (on ARM) is AFAIU the following scenario:
>>>>
>>>> CPU 0                     CPU 1
>>>>
>>>> - dlopen()
>>>>     - mmap PROT_EXEC @addr
>>>>                             - fetch insn @addr, CPU state expects unchanged insn.
>>>>                             - execute unrelated code
>>>> - dlclose(addr)
>>>>     - munmap @addr
>>>> - dlopen()
>>>>     - mmap PROT_EXEC @addr
>>>>                             - fetch new insn @addr. Incoherent CPU state.
>>>
>>> Unmapping an object while code is executing in it is undefined.
>>
>> That's not the scenario though. In this scenario, CPU 1 executes
>> _unrelated code_ while we unmap @addr.
> 
> But in this scenario you still a concurrent dlclose while you have a running
> thread executing code from that module, right? Or am I still missing something
> here?

No.

> 
> Or, are you saying that even after dlopen returns (assuming the scenario where
> it maps the code in a previous used mapping), the CPU is in an inconsistent
> state unless MEMBARRIER_CMD_PRIVATE_EXPEDITED_SYNC_CORE is issued?

Yes, this is it.

And the issue is not specific to dlopen/dlclose. We can have a similar
issue if we have a sequence of:

dlopen  @addr
dlclose
mmap PROT_EXEC|PROT_WRITE  @addr
   - JIT writes some code and jumps to it.

So it appears to be something we may want to fix at the kernel level.

Thanks,

Mathieu


-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com