Re: [XEN PATCH] xen/common: validate shared memory guest address overlap with guest RAM

[Julien Grall <julien@xxxxxxx>]

Hi Joan,

Thank you for the patch.

On 14/04/2026 09:59, Joan Bae wrote:
> Currently, process_shm() does not check whether the guest physical
> address of a shared memory region overlaps with the domain's allocated RAM 
> banks.
> Neither process_shm() nor p2m_set_entry() checks for existing
> mappings, so the RAM mapping is silently overwritten if a user
> specifies a guest physical address that falls within the guest RAM
> range. Since construct_domain() loads the kernel after process_shm(),
> the kernel can end up in shared memory pages. This can cause:
> - Another domain corrupting the kernel via shared memory write
> - Silent guest crash with no error message from Xen
This seems to be solving one specific issue (RAM clashing with shared 
memory) but I believe this could also happen with other kind of mappings 
because, as you said, p2m_set_entry() doesn't check any overlap.
So I would rather prefer if we solve the problem once and for all. This 
would mean modifying p2m_set_entry() (or one of its top caller). 
Although, we would need to be careful to not break memory hypercalls 
which may rely on overwriting existing mappings.
Cheers,

--
Julien Grall