linux-kernel - Re: [PATCH] KVM: x86/mmu: Reject direct bits in gpa passed to KVM_PRE_FAULT

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: Windows password security audit tool. GUI, reports in PDF.

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

Message-ID: <CABgObfafukOFkeR09krA5GVb3itfpHNHjBM0aRZj5t4EKJv7Dw@mail.gmail.com>
Date: Thu, 12 Jun 2025 07:34:12 +0200
From: Paolo Bonzini <pbonzini@...hat.com>
To: Xiaoyao Li <xiaoyao.li@...el.com>
Cc: linux-kernel@...r.kernel.org, kvm@...r.kernel.org, seanjc@...gle.com, 
	yan.y.zhao@...el.com
Subject: Re: [PATCH] KVM: x86/mmu: Reject direct bits in gpa passed to KVM_PRE_FAULT_MEMORY

On Thu, Jun 12, 2025 at 7:29 AM Xiaoyao Li <xiaoyao.li@...el.com> wrote:
>
> On 6/12/2025 12:49 PM, Paolo Bonzini wrote:
> > Only let userspace pass the same addresses that were used in KVM_SET_USER_MEMORY_REGION
> > (or KVM_SET_USER_MEMORY_REGION2); gpas in the the upper half of the address space
> > are an implementation detail of TDX and KVM.
> >
> > Extracted from a patch by Sean Christopherson <seanjc@...gle.com>.
> >
> > Signed-off-by: Paolo Bonzini <pbonzini@...hat.com>
> > ---
> >   arch/x86/kvm/mmu/mmu.c | 3 +++
> >   1 file changed, 3 insertions(+)
> >
> > diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> > index a4040578b537..4e06e2e89a8f 100644
> > --- a/arch/x86/kvm/mmu/mmu.c
> > +++ b/arch/x86/kvm/mmu/mmu.c
> > @@ -4903,6 +4903,9 @@ long kvm_arch_vcpu_pre_fault_memory(struct kvm_vcpu *vcpu,
> >       if (!vcpu->kvm->arch.pre_fault_allowed)
> >               return -EOPNOTSUPP;
> >
> > +     if (kvm_is_gfn_alias(vcpu->kvm, gpa_to_gfn(range->gpa)))
> > +             return -EINVAL;
>
> Do we need to worry about the case (range->gpa + range->size) becomes alias?

No, because the function only processes a single page and everything
in the non-aliased part of the address space *can* be prefaulted.

KVM's generic kvm_vcpu_pre_fault_memory() call will see the EINVAL on
a later invocation and will stop processing the part of the request
that is has the shared/direct bit set.

Paolo

>
> >       /*
> >        * reload is efficient when called repeatedly, so we can do it on
> >        * every iteration.
>