Re: [PATCH] mm: larger stack guard gap, between vmas

From: Ximin Luo
Date: Tue Jul 04 2017 - 10:29:03 EST

Michal Hocko:
> On Tue 04-07-17 13:21:02, Ben Hutchings wrote:
>> On Tue, 2017-07-04 at 14:00 +0200, Michal Hocko wrote:
>>> On Tue 04-07-17 12:36:11, Ben Hutchings wrote:
>>>> On Tue, 2017-07-04 at 12:42 +0200, Michal Hocko wrote:
>>>>> On Tue 04-07-17 11:47:28, Willy Tarreau wrote:
>>>>>> On Tue, Jul 04, 2017 at 11:35:38AM +0200, Michal Hocko wrote:
>>>> [...]
>>>>>> But wouldn't this completely disable the check in case such a guard page
>>>>>> is installed, and possibly continue to allow the collision when the stack
>>>>>> allocation is large enough to skip this guard page ?
>>>>> Yes and but a PROT_NONE would fault and as the changelog says, we _hope_
>>>>> that userspace does the right thing.
>>>> It may well not be large enough, because of the same wrong assumptions
>>>> that resulted in the kernel's guard page not being large enough. We
>>>> should count it as part of the guard gap but not a substitute.
>>> yes, you are right of course. But isn't this a bug on their side
>>> considering they are managing their _own_ stack gap?
>> Yes it's their bug, but you know the rule - don't break user-space.
> Absolutely, that is why I belive we should consider the prev VMA but
> doing anything more just risks for new regressions. Or why do you think
> that not-checking them would cause a regression?
>>> Our stack gap
>>> management is a best effort thing and two such approaches competing will
>>> always lead to weird cornercases. That was my assumption when saying
>>> that I am not sure this is really _worth_ it. We should definitely try
>>> to workaround clashes but that's about it. If others think that we
>>> should do everything to prevent even those issues I will not oppose
>>> of course. It just adds more cycles to something that is a weird case
>>> already.
>> I don't want odd behaviour to weaken the stack guard.
>>> [...]
>>>> This *doesn't* fix the LibreOffice regression on i386.
>>> Are there any details about this regression?
>> Here:
>> I haven't reproduced it in Writer, but if I use Base to create a new
>> HSQLDB database it reliably crashes (HSQLDB is implemented in Java).
> I haven't read through previous 169 comments but I do not see any stack
> trace. Ideally with info proc mapping that would tell us the memory
> layout.

I've written up an explanation of what happens in the Rust case here:

Hopefully I got the details about Linux correct - I only had them explained to me last night - please reply on that page if not.


GPG: ed25519/56034877E1F87C35
GPG: rsa4096/1318EFAC5FBBDBCE