On Fri, Jan 10, 2025 at 05:19:54PM +0100, David Hildenbrand wrote:
On 10.01.25 17:14, Matthew Wilcox wrote:
On Fri, Jan 10, 2025 at 04:48:03PM +0100, David Hildenbrand wrote:
If I would have to guess, I would assume that we have a refcount issue such
that we succeed in splitting a folio while concurrently mapping it.
That would seem hard to accomplish, because both hold the folio lock,
so it wouldn't be just a refcount bug but also a locking bug. Not sure
what this is though.
Yeah, but we also have
https://lkml.kernel.org/r/6774bf44.050a0220.25abdd.098a.GAE@xxxxxxxxxx
That one is a UAF on the vma, so it's either a different issue, or the
problem is with the VMA refcount/lookup/..., not the folio refcount.
cc'ing the relevant maintainers.