KVM: arm64: vgic-its: Avoid potential UAF in LPI translation cache
authorOliver Upton <oliver.upton@linux.dev>
Thu, 4 Jan 2024 18:32:32 +0000 (18:32 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 25 Jan 2024 23:35:48 +0000 (15:35 -0800)
commit ad362fe07fecf0aba839ff2cc59a3617bd42c33f upstream.

There is a potential UAF scenario in the case of an LPI translation
cache hit racing with an operation that invalidates the cache, such
as a DISCARD ITS command. The root of the problem is that
vgic_its_check_cache() does not elevate the refcount on the vgic_irq
before dropping the lock that serializes refcount changes.

Have vgic_its_check_cache() raise the refcount on the returned vgic_irq
and add the corresponding decrement after queueing the interrupt.

Cc: stable@vger.kernel.org
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20240104183233.3560639-1-oliver.upton@linux.dev
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/arm64/kvm/vgic/vgic-its.c

index 5fe2365..c420723 100644 (file)
@@ -584,7 +584,11 @@ static struct vgic_irq *vgic_its_check_cache(struct kvm *kvm, phys_addr_t db,
        unsigned long flags;
 
        raw_spin_lock_irqsave(&dist->lpi_list_lock, flags);
+
        irq = __vgic_its_check_cache(dist, db, devid, eventid);
+       if (irq)
+               vgic_get_irq_kref(irq);
+
        raw_spin_unlock_irqrestore(&dist->lpi_list_lock, flags);
 
        return irq;
@@ -763,6 +767,7 @@ int vgic_its_inject_cached_translation(struct kvm *kvm, struct kvm_msi *msi)
        raw_spin_lock_irqsave(&irq->irq_lock, flags);
        irq->pending_latch = true;
        vgic_queue_irq_unlock(kvm, irq, flags);
+       vgic_put_irq(kvm, irq);
 
        return 0;
 }