Message ID | 20210520230339.267445-8-jmattson@google.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | KVM: nVMX: Fix vmcs02 PID use-after-free issue | expand |
On Thu, May 20, 2021, Jim Mattson wrote: > Don't allow posted interrupts to modify a stale posted interrupt > descriptor (including the initial value of 0). > > Empirical tests on real hardware reveal that a posted interrupt > descriptor referencing an unbacked address has PCI bus error semantics > (reads as all 1's; writes are ignored). However, kvm can't distinguish > unbacked addresses from device-backed (MMIO) addresses, so it should > really ask userspace for an MMIO completion. That's overly > complicated, so just punt with KVM_INTERNAL_ERROR. > > Don't return the error until the posted interrupt descriptor is > actually accessed. We don't want to break the existing kvm-unit-tests > that assume they can launch an L2 VM with a posted interrupt > descriptor that references MMIO space in L1. > > Fixes: 6beb7bd52e48 ("kvm: nVMX: Refactor nested_get_vmcs12_pages()") > Signed-off-by: Jim Mattson <jmattson@google.com> > --- > arch/x86/kvm/vmx/nested.c | 15 ++++++++++++++- > 1 file changed, 14 insertions(+), 1 deletion(-) > > diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c > index 706c31821362..defd42201bb4 100644 > --- a/arch/x86/kvm/vmx/nested.c > +++ b/arch/x86/kvm/vmx/nested.c > @@ -3175,6 +3175,15 @@ static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu) > offset_in_page(vmcs12->posted_intr_desc_addr)); > vmcs_write64(POSTED_INTR_DESC_ADDR, > pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr)); > + } else { > + /* > + * Defer the KVM_INTERNAL_ERROR exit until > + * someone tries to trigger posted interrupt > + * processing on this vCPU, to avoid breaking > + * existing kvm-unit-tests. Run the lines out to 80 chars. Also, can we change the comment to tie it to CPU behavior in someway? A few years down the road, "existing kvm-unit-tests" may not have any relevant meaning, and it's not like kvm-unit-tests is bug free either. E.g. something like /* * Defer the KVM_INTERNAL_ERROR exit until posted * interrupt processing actually occurs on this vCPU. * Until that happens, the descriptor is not accessed, * and userspace can technically rely on that behavior. */ > + */ > + vmx->nested.pi_desc = NULL; > + pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR); > } > } > if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12)) > @@ -3689,10 +3698,14 @@ static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu) > void *vapic_page; > u16 status; > > - if (!vmx->nested.pi_desc || !vmx->nested.pi_pending) > + if (!vmx->nested.pi_pending) > return 0; > > + if (!vmx->nested.pi_desc) > + goto mmio_needed; > + > vmx->nested.pi_pending = false; > + > if (!pi_test_and_clear_on(vmx->nested.pi_desc)) > return 0; > > -- > 2.31.1.818.g46aad6cb9e-goog >
On Mon, May 24, 2021 at 4:22 PM Sean Christopherson <seanjc@google.com> wrote: > > On Thu, May 20, 2021, Jim Mattson wrote: > > Don't allow posted interrupts to modify a stale posted interrupt > > descriptor (including the initial value of 0). > > > > Empirical tests on real hardware reveal that a posted interrupt > > descriptor referencing an unbacked address has PCI bus error semantics > > (reads as all 1's; writes are ignored). However, kvm can't distinguish > > unbacked addresses from device-backed (MMIO) addresses, so it should > > really ask userspace for an MMIO completion. That's overly > > complicated, so just punt with KVM_INTERNAL_ERROR. > > > > Don't return the error until the posted interrupt descriptor is > > actually accessed. We don't want to break the existing kvm-unit-tests > > that assume they can launch an L2 VM with a posted interrupt > > descriptor that references MMIO space in L1. > > > > Fixes: 6beb7bd52e48 ("kvm: nVMX: Refactor nested_get_vmcs12_pages()") > > Signed-off-by: Jim Mattson <jmattson@google.com> > > --- > > arch/x86/kvm/vmx/nested.c | 15 ++++++++++++++- > > 1 file changed, 14 insertions(+), 1 deletion(-) > > > > diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c > > index 706c31821362..defd42201bb4 100644 > > --- a/arch/x86/kvm/vmx/nested.c > > +++ b/arch/x86/kvm/vmx/nested.c > > @@ -3175,6 +3175,15 @@ static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu) > > offset_in_page(vmcs12->posted_intr_desc_addr)); > > vmcs_write64(POSTED_INTR_DESC_ADDR, > > pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr)); > > + } else { > > + /* > > + * Defer the KVM_INTERNAL_ERROR exit until > > + * someone tries to trigger posted interrupt > > + * processing on this vCPU, to avoid breaking > > + * existing kvm-unit-tests. > > Run the lines out to 80 chars. Also, can we change the comment to tie it to > CPU behavior in someway? A few years down the road, "existing kvm-unit-tests" > may not have any relevant meaning, and it's not like kvm-unit-tests is bug free > either. E.g. something like > > /* > * Defer the KVM_INTERNAL_ERROR exit until posted > * interrupt processing actually occurs on this vCPU. > * Until that happens, the descriptor is not accessed, > * and userspace can technically rely on that behavior. > */ Okay...except for the fact that kvm will rather gratuitously process posted interrupts in situations where hardware won't. That makes it difficult to tie this to hardware behavior.
On Mon, May 24, 2021, Jim Mattson wrote: > On Mon, May 24, 2021 at 4:22 PM Sean Christopherson <seanjc@google.com> wrote: > > > > On Thu, May 20, 2021, Jim Mattson wrote: > > > Don't allow posted interrupts to modify a stale posted interrupt > > > descriptor (including the initial value of 0). > > > > > > Empirical tests on real hardware reveal that a posted interrupt > > > descriptor referencing an unbacked address has PCI bus error semantics > > > (reads as all 1's; writes are ignored). However, kvm can't distinguish > > > unbacked addresses from device-backed (MMIO) addresses, so it should > > > really ask userspace for an MMIO completion. That's overly > > > complicated, so just punt with KVM_INTERNAL_ERROR. > > > > > > Don't return the error until the posted interrupt descriptor is > > > actually accessed. We don't want to break the existing kvm-unit-tests > > > that assume they can launch an L2 VM with a posted interrupt > > > descriptor that references MMIO space in L1. > > > > > > Fixes: 6beb7bd52e48 ("kvm: nVMX: Refactor nested_get_vmcs12_pages()") > > > Signed-off-by: Jim Mattson <jmattson@google.com> > > > --- > > > arch/x86/kvm/vmx/nested.c | 15 ++++++++++++++- > > > 1 file changed, 14 insertions(+), 1 deletion(-) > > > > > > diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c > > > index 706c31821362..defd42201bb4 100644 > > > --- a/arch/x86/kvm/vmx/nested.c > > > +++ b/arch/x86/kvm/vmx/nested.c > > > @@ -3175,6 +3175,15 @@ static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu) > > > offset_in_page(vmcs12->posted_intr_desc_addr)); > > > vmcs_write64(POSTED_INTR_DESC_ADDR, > > > pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr)); > > > + } else { > > > + /* > > > + * Defer the KVM_INTERNAL_ERROR exit until > > > + * someone tries to trigger posted interrupt > > > + * processing on this vCPU, to avoid breaking > > > + * existing kvm-unit-tests. > > > > Run the lines out to 80 chars. Also, can we change the comment to tie it to > > CPU behavior in someway? A few years down the road, "existing kvm-unit-tests" > > may not have any relevant meaning, and it's not like kvm-unit-tests is bug free > > either. E.g. something like > > > > /* > > * Defer the KVM_INTERNAL_ERROR exit until posted > > * interrupt processing actually occurs on this vCPU. > > * Until that happens, the descriptor is not accessed, > > * and userspace can technically rely on that behavior. > > */ > Okay...except for the fact that kvm will rather gratuitously process > posted interrupts in situations where hardware won't. That makes it > difficult to tie this to hardware behavior. Hrm, true, but we can at say that KVM won't bail if there's zero chance of posted interrupts being processed. I hope?
On Mon, May 24, 2021 at 4:45 PM Sean Christopherson <seanjc@google.com> wrote: > > On Mon, May 24, 2021, Jim Mattson wrote: > > On Mon, May 24, 2021 at 4:22 PM Sean Christopherson <seanjc@google.com> wrote: > > > > > > On Thu, May 20, 2021, Jim Mattson wrote: > > > > Don't allow posted interrupts to modify a stale posted interrupt > > > > descriptor (including the initial value of 0). > > > > > > > > Empirical tests on real hardware reveal that a posted interrupt > > > > descriptor referencing an unbacked address has PCI bus error semantics > > > > (reads as all 1's; writes are ignored). However, kvm can't distinguish > > > > unbacked addresses from device-backed (MMIO) addresses, so it should > > > > really ask userspace for an MMIO completion. That's overly > > > > complicated, so just punt with KVM_INTERNAL_ERROR. > > > > > > > > Don't return the error until the posted interrupt descriptor is > > > > actually accessed. We don't want to break the existing kvm-unit-tests > > > > that assume they can launch an L2 VM with a posted interrupt > > > > descriptor that references MMIO space in L1. > > > > > > > > Fixes: 6beb7bd52e48 ("kvm: nVMX: Refactor nested_get_vmcs12_pages()") > > > > Signed-off-by: Jim Mattson <jmattson@google.com> > > > > --- > > > > arch/x86/kvm/vmx/nested.c | 15 ++++++++++++++- > > > > 1 file changed, 14 insertions(+), 1 deletion(-) > > > > > > > > diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c > > > > index 706c31821362..defd42201bb4 100644 > > > > --- a/arch/x86/kvm/vmx/nested.c > > > > +++ b/arch/x86/kvm/vmx/nested.c > > > > @@ -3175,6 +3175,15 @@ static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu) > > > > offset_in_page(vmcs12->posted_intr_desc_addr)); > > > > vmcs_write64(POSTED_INTR_DESC_ADDR, > > > > pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr)); > > > > + } else { > > > > + /* > > > > + * Defer the KVM_INTERNAL_ERROR exit until > > > > + * someone tries to trigger posted interrupt > > > > + * processing on this vCPU, to avoid breaking > > > > + * existing kvm-unit-tests. > > > > > > Run the lines out to 80 chars. Also, can we change the comment to tie it to > > > CPU behavior in someway? A few years down the road, "existing kvm-unit-tests" > > > may not have any relevant meaning, and it's not like kvm-unit-tests is bug free > > > either. E.g. something like > > > > > > /* > > > * Defer the KVM_INTERNAL_ERROR exit until posted > > > * interrupt processing actually occurs on this vCPU. > > > * Until that happens, the descriptor is not accessed, > > > * and userspace can technically rely on that behavior. > > > */ > > Okay...except for the fact that kvm will rather gratuitously process > > posted interrupts in situations where hardware won't. That makes it > > difficult to tie this to hardware behavior. > > Hrm, true, but we can at say that KVM won't bail if there's zero chance of posted > interrupts being processed. I hope? Zero chance in KVM, or zero chance on hardware? For instance, set TPR high enough to block the posted interrupt vector from being delivered, and there is zero chance of posted interrupts being processed by hardware. However, if another L1 vCPU sends that vector by IPI, there is a 100% chance that KVM will bail, because it ignores TPR for processing posted interrupts.
On Mon, May 24, 2021, Jim Mattson wrote: > On Mon, May 24, 2021 at 4:45 PM Sean Christopherson <seanjc@google.com> wrote: > > > > On Mon, May 24, 2021, Jim Mattson wrote: > > > On Mon, May 24, 2021 at 4:22 PM Sean Christopherson <seanjc@google.com> wrote: > > > > > > > > On Thu, May 20, 2021, Jim Mattson wrote: > > > > > Don't allow posted interrupts to modify a stale posted interrupt > > > > > descriptor (including the initial value of 0). > > > > > > > > > > Empirical tests on real hardware reveal that a posted interrupt > > > > > descriptor referencing an unbacked address has PCI bus error semantics > > > > > (reads as all 1's; writes are ignored). However, kvm can't distinguish > > > > > unbacked addresses from device-backed (MMIO) addresses, so it should > > > > > really ask userspace for an MMIO completion. That's overly > > > > > complicated, so just punt with KVM_INTERNAL_ERROR. > > > > > > > > > > Don't return the error until the posted interrupt descriptor is > > > > > actually accessed. We don't want to break the existing kvm-unit-tests > > > > > that assume they can launch an L2 VM with a posted interrupt > > > > > descriptor that references MMIO space in L1. > > > > > > > > > > Fixes: 6beb7bd52e48 ("kvm: nVMX: Refactor nested_get_vmcs12_pages()") > > > > > Signed-off-by: Jim Mattson <jmattson@google.com> > > > > > --- > > > > > arch/x86/kvm/vmx/nested.c | 15 ++++++++++++++- > > > > > 1 file changed, 14 insertions(+), 1 deletion(-) > > > > > > > > > > diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c > > > > > index 706c31821362..defd42201bb4 100644 > > > > > --- a/arch/x86/kvm/vmx/nested.c > > > > > +++ b/arch/x86/kvm/vmx/nested.c > > > > > @@ -3175,6 +3175,15 @@ static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu) > > > > > offset_in_page(vmcs12->posted_intr_desc_addr)); > > > > > vmcs_write64(POSTED_INTR_DESC_ADDR, > > > > > pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr)); > > > > > + } else { > > > > > + /* > > > > > + * Defer the KVM_INTERNAL_ERROR exit until > > > > > + * someone tries to trigger posted interrupt > > > > > + * processing on this vCPU, to avoid breaking > > > > > + * existing kvm-unit-tests. > > > > > > > > Run the lines out to 80 chars. Also, can we change the comment to tie it to > > > > CPU behavior in someway? A few years down the road, "existing kvm-unit-tests" > > > > may not have any relevant meaning, and it's not like kvm-unit-tests is bug free > > > > either. E.g. something like > > > > > > > > /* > > > > * Defer the KVM_INTERNAL_ERROR exit until posted > > > > * interrupt processing actually occurs on this vCPU. > > > > * Until that happens, the descriptor is not accessed, > > > > * and userspace can technically rely on that behavior. > > > > */ > > > Okay...except for the fact that kvm will rather gratuitously process > > > posted interrupts in situations where hardware won't. That makes it > > > difficult to tie this to hardware behavior. > > > > Hrm, true, but we can at say that KVM won't bail if there's zero chance of posted > > interrupts being processed. I hope? > Zero chance in KVM, or zero chance on hardware? I was hoping hardware... > For instance, set TPR high enough to block the posted interrupt vector > from being delivered, and there is zero chance of posted interrupts > being processed by hardware. However, if another L1 vCPU sends that > vector by IPI, there is a 100% chance that KVM will bail, because it > ignores TPR for processing posted interrupts. Can we instead word it along the lines of: Defer the KVM_INTERNAL_EXIT until KVM actually attempts to consume the posted interrupt descriptor on behalf of the vCPU. Note, KVM may process posted interrupts when it architecturally should not. Bugs aside, userspace can at least rely on KVM to not process posted interrupts if there is no (posted?) interrupt activity whatsoever.
On Mon, May 24, 2021 at 5:11 PM Sean Christopherson <seanjc@google.com> wrote: > > Can we instead word it along the lines of: > > Defer the KVM_INTERNAL_EXIT until KVM actually attempts to consume the posted > interrupt descriptor on behalf of the vCPU. Note, KVM may process posted > interrupts when it architecturally should not. Bugs aside, userspace can at > least rely on KVM to not process posted interrupts if there is no (posted?) > interrupt activity whatsoever. How about: Defer the KVM_INTERNAL_EXIT until KVM tries to access the contents of the VMCS12 posted interrupt descriptor. (Note that KVM may do this when it should not, per the architectural specification.)
On Mon, May 24, 2021, Jim Mattson wrote: > On Mon, May 24, 2021 at 5:11 PM Sean Christopherson <seanjc@google.com> wrote: > > > > Can we instead word it along the lines of: > > > > Defer the KVM_INTERNAL_EXIT until KVM actually attempts to consume the posted > > interrupt descriptor on behalf of the vCPU. Note, KVM may process posted > > interrupts when it architecturally should not. Bugs aside, userspace can at > > least rely on KVM to not process posted interrupts if there is no (posted?) > > interrupt activity whatsoever. > > How about: > > Defer the KVM_INTERNAL_EXIT until KVM tries to access the contents of > the VMCS12 posted interrupt descriptor. (Note that KVM may do this > when it should not, per the architectural specification.) Works for me!
diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c index 706c31821362..defd42201bb4 100644 --- a/arch/x86/kvm/vmx/nested.c +++ b/arch/x86/kvm/vmx/nested.c @@ -3175,6 +3175,15 @@ static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu) offset_in_page(vmcs12->posted_intr_desc_addr)); vmcs_write64(POSTED_INTR_DESC_ADDR, pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr)); + } else { + /* + * Defer the KVM_INTERNAL_ERROR exit until + * someone tries to trigger posted interrupt + * processing on this vCPU, to avoid breaking + * existing kvm-unit-tests. + */ + vmx->nested.pi_desc = NULL; + pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR); } } if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12)) @@ -3689,10 +3698,14 @@ static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu) void *vapic_page; u16 status; - if (!vmx->nested.pi_desc || !vmx->nested.pi_pending) + if (!vmx->nested.pi_pending) return 0; + if (!vmx->nested.pi_desc) + goto mmio_needed; + vmx->nested.pi_pending = false; + if (!pi_test_and_clear_on(vmx->nested.pi_desc)) return 0;
Don't allow posted interrupts to modify a stale posted interrupt descriptor (including the initial value of 0). Empirical tests on real hardware reveal that a posted interrupt descriptor referencing an unbacked address has PCI bus error semantics (reads as all 1's; writes are ignored). However, kvm can't distinguish unbacked addresses from device-backed (MMIO) addresses, so it should really ask userspace for an MMIO completion. That's overly complicated, so just punt with KVM_INTERNAL_ERROR. Don't return the error until the posted interrupt descriptor is actually accessed. We don't want to break the existing kvm-unit-tests that assume they can launch an L2 VM with a posted interrupt descriptor that references MMIO space in L1. Fixes: 6beb7bd52e48 ("kvm: nVMX: Refactor nested_get_vmcs12_pages()") Signed-off-by: Jim Mattson <jmattson@google.com> --- arch/x86/kvm/vmx/nested.c | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-)