[PATCH v17 07/10] mm: introduce memfd_secret system call to create "secret" memory areas
rppt at kernel.org
Thu Feb 11 06:29:46 EST 2021
On Thu, Feb 11, 2021 at 11:02:07AM +0100, David Hildenbrand wrote:
> Another thought regarding "doesn't have _any_ backing storage"
> What are the right semantics when it comes to memory accounting/commit?
> As secretmem does not have
> a) any backing storage
> b) cannot go to swap
> The MAP_NORESERVE vs. !MAP_NORESERVE handling gets a little unclear. Why
> "reserve swap space" if the allocations cannot ever go to swap? Sure, we
> want to "reserve physical memory", but in contrast to other users that can
> go to swap.
> Of course, this is only relevant for MAP_PRIVATE secretmem mappings. Other
> MAP_SHARED assumes there is no need for reserving swap space as it can just
> go to the backing storage. (yeah, tmpfs/shmem is weird in that regard as
> well, but again, it's a bit different)
In that sense seceremem is as weird as tmpfs and it only allows MAP_SHARED.
More information about the linux-riscv