On 12/12/24 20:13, Sean Christopherson wrote:
On Thu, Dec 12, 2024, Paolo Bonzini wrote:
On 12/12/24 09:07, Zhao Liu wrote:
On Thu, Dec 12, 2024 at 11:26:28AM +0800, Binbin Wu wrote:
Date: Thu, 12 Dec 2024 11:26:28 +0800
From: Binbin Wu <binbin.wu@xxxxxxxxxxxxxxx>
Subject: [PATCH] i386/kvm: Set return value after handling
KVM_EXIT_HYPERCALL
X-Mailer: git-send-email 2.46.0
Userspace should set the ret field of hypercall after handling
KVM_EXIT_HYPERCALL. Otherwise, a stale value could be returned to KVM.
Fixes: 47e76d03b15 ("i386/kvm: Add KVM_EXIT_HYPERCALL handling for KVM_HC_MAP_GPA_RANGE")
Reported-by: Farrah Chen <farrah.chen@xxxxxxxxx>
Signed-off-by: Binbin Wu <binbin.wu@xxxxxxxxxxxxxxx>
Tested-by: Farrah Chen <farrah.chen@xxxxxxxxx>
---
To test the TDX code in kvm-coco-queue, please apply the patch to the QEMU,
otherwise, TDX guest boot could fail.
A matching QEMU tree including this patch is here:
https://github.com/intel-staging/qemu-tdx/releases/tag/tdx-qemu-upstream-v6.1-fix_kvm_hypercall_return_value
Previously, the issue was not triggered because no one would modify the ret
value. But with the refactor patch for __kvm_emulate_hypercall() in KVM,
https://lore.kernel.org/kvm/20241128004344.4072099-7-seanjc@xxxxxxxxxx/, the
value could be modified.
Could you explain the specific reasons here in detail? It would be
helpful with debugging or reproducing the issue.
---
target/i386/kvm/kvm.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/target/i386/kvm/kvm.c b/target/i386/kvm/kvm.c
index 8e17942c3b..4bcccb48d1 100644
--- a/target/i386/kvm/kvm.c
+++ b/target/i386/kvm/kvm.c
@@ -6005,10 +6005,14 @@ static int kvm_handle_hc_map_gpa_range(struct kvm_run *run)
static int kvm_handle_hypercall(struct kvm_run *run)
{
+ int ret = -EINVAL;
+
if (run->hypercall.nr == KVM_HC_MAP_GPA_RANGE)
- return kvm_handle_hc_map_gpa_range(run);
+ ret = kvm_handle_hc_map_gpa_range(run);
+
+ run->hypercall.ret = ret;
ret may be negative but hypercall.ret is u64. Do we need to set it to
-ret?
If ret is less than zero, will stop the VM anyway as
RUN_STATE_INTERNAL_ERROR.
If this has to be fixed in QEMU, I think there's no need to set anything
if ret != 0; also because kvm_convert_memory() returns -1 on error and
that's not how the error would be passed to the guest.
However, I think the right fix should simply be this in KVM:
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 83fe0a78146f..e2118ba93ef6 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -10066,6 +10066,7 @@ unsigned long __kvm_emulate_hypercall(struct kvm_vcpu *vcpu, unsigned long nr,
}
vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
+ vcpu->run->ret = 0;
vcpu->run->hypercall.ret
vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
vcpu->run->hypercall.args[0] = gpa;
vcpu->run->hypercall.args[1] = npages;
While there is arguably a change in behavior of the kernel both with
the patches in kvm-coco-queue and with the above one, _in practice_
the above change is one that userspace will not notice.
I agree that KVM should initialize "ret", but I don't think '0' is the right
value. KVM shouldn't assume userspace will successfully handle the hypercall.
What happens if KVM sets vcpu->run->hypercall.ret to a non-zero value, e.g. -KVM_ENOSYS?
Unfortunately QEMU is never writing vcpu->run->hypercall.ret, so the
guest sees -KVM_ENOSYS; this is basically the same bug that Binbin is
fixing, just with a different value passed to the guest.
In other words, the above one-liner is pulling the "don't break
userspace" card.
Paolo
