Re: [PATCH] Increase default MLOCK_LIMIT to 8 MiB

[Jason Gunthorpe]

On Wed, Nov 24, 2021 at 05:43:58PM +0100, David Hildenbrand wrote:
> On 24.11.21 16:34, Jason Gunthorpe wrote:
> > On Wed, Nov 24, 2021 at 03:14:00PM +0100, David Hildenbrand wrote:
> > 
> >> I'm not aware of any where you can fragment 50% of all pageblocks in the
> >> system as an unprivileged user essentially consuming almost no memory
> >> and essentially staying inside well-defined memlock limits. But sure if
> >> there are "many" people will be able to come up with at least one
> >> comparable thing. I'll be happy to learn.
> > 
> > If the concern is that THP's can be DOS'd then any avenue that renders
> > the system out of THPs is a DOS attack vector. Including all the
> > normal workloads that people run and already complain that THPs get
> > exhausted.
> > 
> > A hostile userspace can only quicken this process.
> 
> We can not only fragment THP but also easily smaller compound pages,
> with less impact though (well, as long as people want more than 0.1% per
> user ...).

My point is as long as userspace can drive this fragmentation, by any
means, we can never have DOS proof higher order pages, so lets not
worry so much about one of many ways to create fragmentation.

> >> My position that FOLL_LONGTERM for unprivileged users is a strong no-go
> >> stands as it is.
> > 
> > As this basically excludes long standing pre-existing things like
> > RDMA, XDP, io_uring, and more I don't think this can be the general
> > answer for mm, sorry.
> 
> Let's think about options to restrict FOLL_LONGTERM usage:

Which gives me the view that we should be talking about how to make
high order pages completely DOS proof, not about FOLL_LONGTERM.

To me that is exactly what ZONE_MOVABLE strives to achieve, and I
think anyone who cares about QOS around THP must include ZONE_MOVABLE
in their solution.

In all of this I am thinking back to the discussion about the 1GB THP
proposal which was resoundly shot down on the grounds that 2MB THP
*doesn't work* today due to the existing fragmentation problems.

> Another option would be not accounting FOLL_LONGTERM as RLIMIT_MEMLOCK,
> but instead as something that explicitly matches the differing
> semantics. 

Also a good idea, someone who cares about this should really put
pinned pages into the cgroup machinery (with correct accounting!)

> At the same time, eventually work on proper alternatives with mmu
> notifiers (and possibly without the any such limits) where possible
> and required.

mmu_notifiers is also bad, it just offends a different group of MM
concerns :)

Something like io_ring is registering a bulk amount of memory and then
doing some potentially long operations against it.

So to use a mmu_notifier scheme you'd have to block the mmu_notifier
invalidate_range_start until all the operations touching the memory
finish (and suspend new operations at the same time!).

Blocking the notifier like this locks up the migration/etc threads
completely, and is destructive to the OOM reclaim.

At least with a pinned page those threads don't even try to touch it
instead of getting stuck up.

> Don't get me wrong, I really should be working on other stuff, so I have
> limited brain capacity and time :) OTOH I'm willing to help at least
> discuss alternatives.

Haha, me too..

Jason