linux-kernel - Re: [PATCH v2 1/4] mm: Introduce vm_uffd

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <htdtwazk66dcrfkpwh5fgtk5ruzzo2zy54i5w6kbg45lqtyect@tt22z7hcdfnx>
Date: Wed, 2 Jul 2025 22:00:51 -0400
From: "Liam R. Howlett" <Liam.Howlett@...cle.com>
To: Peter Xu <peterx@...hat.com>
Cc: Nikita Kalyazin <kalyazin@...zon.com>,
        Lorenzo Stoakes <lorenzo.stoakes@...cle.com>,
        Suren Baghdasaryan <surenb@...gle.com>, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org, Vlastimil Babka <vbabka@...e.cz>,
        Muchun Song <muchun.song@...ux.dev>, Mike Rapoport <rppt@...nel.org>,
        Hugh Dickins <hughd@...gle.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        James Houghton <jthoughton@...gle.com>, Michal Hocko <mhocko@...e.com>,
        David Hildenbrand <david@...hat.com>,
        Andrea Arcangeli <aarcange@...hat.com>,
        Oscar Salvador <osalvador@...e.de>,
        Axel Rasmussen <axelrasmussen@...gle.com>,
        Ujwal Kundur <ujwal.kundur@...il.com>
Subject: Re: [PATCH v2 1/4] mm: Introduce vm_uffd_ops API

* Peter Xu <peterx@...hat.com> [250702 17:36]:
> On Wed, Jul 02, 2025 at 05:24:02PM -0400, Liam R. Howlett wrote:
> > That's because the entry point is from a function pointer, so [3] won't
> > help at all.
> > 
> > It is recreating the situation that existed for the vma through the
> > vm_ops in mmap, but for uffd.  And at a lower level (page tables).  I do not
> > want to relive that experience.
> > 
> > We are not doing this.  It is for the benefit of everyone that we are
> > not doing this.
> 
> Is the vma issue about "allowing vma->vm_flags to be modified anywhere"
> issue?  Or is there a pointer to the issue being discussed if not?

The issue is passing pointers of structs that are protected by locks or
ref counters into modules to do what they please.

vma->vm_flags was an example of where we learned how wrong this can go.

There is also the concern of the state of the folio on return from the
callback.  The error handling gets messy quick.

Now, imagine we have something that gets a folio, but then we find a
solution for contention of a lock or ref count (whatever is next), but
it doesn't work because the mm code has been bleeding into random
modules and we have no clue what that module is supposed to be doing, or
we can't make the necessary change because this module will break
userspace, or cause a performance decrease, or any other random thing
that we cannot work around without rewriting (probably suboptimally)
something we don't maintain.

Again, these are examples of how this can go bad but not an exhaustive
list by any means.

So the issue is with allowing modules to play with the folio and page
tables on their own.

If this is outside the mm, we probably won't even be Cc'ed on modules
that use it.

And do we want to be Cc'ed on modules that want to use it?

We will most likely be Cc'ed or emailed directly on the resulting memory
leak/security issue that results in what should be mm code.  It'll be a
Saturday because it always is.. :)

Even the example use code had a potential ref leak that you found [1].

> > 
> > We need to find another way.
> 
> Could you suggest something?  The minimum goal is to allow guest-memfd
> support minor faults.

Mike brought up another idea, that seems worth looking into.

Thanks,
Liam

[1] https://lore.kernel.org/all/7455220c-e35b-4509-b7c3-a78fde5b12d5@amazon.com/