linux-kernel - [PATCH] KVM: ioapic: break infinite recursion on lazy EOI

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>] [day] [month] [year] [list]

Message-Id: <20201024081724.2799401-1-pbonzini@redhat.com>
Date:   Sat, 24 Oct 2020 04:17:24 -0400
From:   Paolo Bonzini <pbonzini@...hat.com>
To:     linux-kernel@...r.kernel.org, kvm@...r.kernel.org
Cc:     Vitaly Kuznetsov <vkuznets@...hat.com>,
        Alex Williamson <alex.williamson@...hat.com>
Subject: [PATCH] KVM: ioapic: break infinite recursion on lazy EOI

From: Vitaly Kuznetsov <vkuznets@...hat.com>

During shutdown the IOAPIC trigger mode is reset to edge triggered
while the vfio-pci INTx is still registered with a resampler.
This allows us to get into an infinite loop:

ioapic_set_irq
  -> ioapic_lazy_update_eoi
  -> kvm_ioapic_update_eoi_one
  -> kvm_notify_acked_irq
  -> kvm_notify_acked_gsi
  -> (via irq_acked fn ptr) irqfd_resampler_ack
  -> kvm_set_irq
  -> (via set fn ptr) kvm_set_ioapic_irq
  -> kvm_ioapic_set_irq
  -> ioapic_set_irq

Commit 7faab2faca0c acknowledges that this recursion loop exists
and tries to avoid it at the call to ioapic_lazy_update_eoi, but at
this point the scenario is already set, we have an edge interrupt with
resampler on the same gsi.

Fortunately, the only user of irq ack notifiers (in addition to resamplefd)
is i8254 timer interrupt reinjection.  These are edge-triggered, so in
principle they would need the call to kvm_ioapic_update_eoi_one from
ioapic_lazy_update_eoi, but they already disable AVIC(*), so they don't
need the lazy EOI behavior.  Therefore, remove the call to
kvm_ioapic_update_eoi_one from ioapic_lazy_update_eoi.

This fixes CVE-2020-27152.  Note that this issue cannot happen with
SR-IOV assigned devices because virtual functions do not have INTx,
only MSI.

Suggested-by: Paolo Bonzini <pbonzini@...hat.com>
Tested-by: Alex Williamson <alex.williamson@...hat.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@...hat.com>
Signed-off-by: Paolo Bonzini <pbonzini@...hat.com>
---
 arch/x86/kvm/ioapic.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/arch/x86/kvm/ioapic.c b/arch/x86/kvm/ioapic.c
index d057376bd3d3..698969e18fe3 100644
--- a/arch/x86/kvm/ioapic.c
+++ b/arch/x86/kvm/ioapic.c
@@ -197,12 +197,9 @@ static void ioapic_lazy_update_eoi(struct kvm_ioapic *ioapic, int irq)

 		/*
 		 * If no longer has pending EOI in LAPICs, update
-		 * EOI for this vetor.
+		 * EOI for this vector.
 		 */
 		rtc_irq_eoi(ioapic, vcpu, entry->fields.vector);
-		kvm_ioapic_update_eoi_one(vcpu, ioapic,
-					  entry->fields.trig_mode,
-					  irq);
 		break;
 	}
 }
-- 
2.26.2