kernel/linux.git/virt, branch linux-4.20.y

kvm: fix kvm_ioctl_create_device() reference counting (CVE-2019-6974)

2019-02-12T19:02:37+00:00

commit cfa39381173d5f969daf43582c95ad679189cbc9 upstream. kvm_ioctl_create_device() does the following: 1. creates a device that holds a reference to the VM object (with a borrowed reference, the VM's refcount has not been bumped yet) 2. initializes the device 3. transfers the reference to the device to the caller's file descriptor table 4. calls kvm_get_kvm() to turn the borrowed reference to the VM into a real reference The ownership transfer in step 3 must not happen before the reference to the VM becomes a proper, non-borrowed reference, which only happens in step 4. After step 3, an attacker can close the file descriptor and drop the borrowed reference, which can cause the refcount of the kvm object to drop to zero. This means that we need to grab a reference for the device before anon_inode_getfd(), otherwise the VM can disappear from under us. Fixes: 852b6d57dc7f ("kvm: add device control API") Cc: stable@kernel.org Signed-off-by: Jann Horn Signed-off-by: Paolo Bonzini Signed-off-by: Greg Kroah-Hartman

kvm: Change offset in kvm_write_guest_offset_cached to unsigned

2019-02-12T19:02:27+00:00

[ Upstream commit 7a86dab8cf2f0fdf508f3555dddfc236623bff60 ] Since the offset is added directly to the hva from the gfn_to_hva_cache, a negative offset could result in an out of bounds write. The existing BUG_ON only checks for addresses beyond the end of the gfn_to_hva_cache, not for addresses before the start of the gfn_to_hva_cache. Note that all current call sites have non-negative offsets. Fixes: 4ec6e8636256 ("kvm: Introduce kvm_write_guest_offset_cached()") Reported-by: Cfir Cohen Signed-off-by: Jim Mattson Reviewed-by: Cfir Cohen Reviewed-by: Peter Shier Reviewed-by: Krish Sadhukhan Reviewed-by: Sean Christopherson Signed-off-by: Radim Krčmář Signed-off-by: Sasha Levin

arm64: KVM: Skip MMIO insn after emulation

2019-02-12T19:02:22+00:00

[ Upstream commit 0d640732dbebed0f10f18526de21652931f0b2f2 ] When we emulate an MMIO instruction, we advance the CPU state within decode_hsr(), before emulating the instruction effects. Having this logic in decode_hsr() is opaque, and advancing the state before emulation is problematic. It gets in the way of applying consistent single-step logic, and it prevents us from being able to fail an MMIO instruction with a synchronous exception. Clean this up by only advancing the CPU state *after* the effects of the instruction are emulated. Cc: Peter Maydell Reviewed-by: Alex Bennée Reviewed-by: Christoffer Dall Signed-off-by: Mark Rutland Signed-off-by: Marc Zyngier Signed-off-by: Sasha Levin

KVM: arm/arm64: vgic: Fix off-by-one bug in vgic_get_irq()

2019-01-09T16:46:04+00:00

commit c23b2e6fc4ca346018618266bcabd335c0a8a49e upstream. When using the nospec API, it should be taken into account that: "...if the CPU speculates past the bounds check then * array_index_nospec() will clamp the index within the range of [0, * size)." The above is part of the header for macro array_index_nospec() in linux/nospec.h Now, in this particular case, if intid evaluates to exactly VGIC_MAX_SPI or to exaclty VGIC_MAX_PRIVATE, the array_index_nospec() macro ends up returning VGIC_MAX_SPI - 1 or VGIC_MAX_PRIVATE - 1 respectively, instead of VGIC_MAX_SPI or VGIC_MAX_PRIVATE, which, based on the original logic: /* SGIs and PPIs */ if (intid <= VGIC_MAX_PRIVATE) return &vcpu->arch.vgic_cpu.private_irqs[intid]; /* SPIs */ if (intid <= VGIC_MAX_SPI) return &kvm->arch.vgic.spis[intid - VGIC_NR_PRIVATE_IRQS]; are valid values for intid. Fix this by calling array_index_nospec() macro with VGIC_MAX_PRIVATE + 1 and VGIC_MAX_SPI + 1 as arguments for its parameter size. Fixes: 41b87599c743 ("KVM: arm/arm64: vgic: fix possible spectre-v1 in vgic_get_irq()") Cc: stable@vger.kernel.org Signed-off-by: Gustavo A. R. Silva [dropped the SPI part which was fixed separately] Signed-off-by: Marc Zyngier Signed-off-by: Greg Kroah-Hartman

KVM: arm/arm64: Fix VMID alloc race by reverting to lock-less

2019-01-09T16:46:04+00:00

commit fb544d1ca65a89f7a3895f7531221ceeed74ada7 upstream. We recently addressed a VMID generation race by introducing a read/write lock around accesses and updates to the vmid generation values. However, kvm_arch_vcpu_ioctl_run() also calls need_new_vmid_gen() but does so without taking the read lock. As far as I can tell, this can lead to the same kind of race: VM 0, VCPU 0 VM 0, VCPU 1 ------------ ------------ update_vttbr (vmid 254) update_vttbr (vmid 1) // roll over read_lock(kvm_vmid_lock); force_vm_exit() local_irq_disable need_new_vmid_gen == false //because vmid gen matches enter_guest (vmid 254) kvm_arch.vttbr = : read_unlock(kvm_vmid_lock); enter_guest (vmid 1) Which results in running two VCPUs in the same VM with different VMIDs and (even worse) other VCPUs from other VMs could now allocate clashing VMID 254 from the new generation as long as VCPU 0 is not exiting. Attempt to solve this by making sure vttbr is updated before another CPU can observe the updated VMID generation. Cc: stable@vger.kernel.org Fixes: f0cf47d939d0 "KVM: arm/arm64: Close VMID generation race" Reviewed-by: Julien Thierry Signed-off-by: Christoffer Dall Signed-off-by: Marc Zyngier Signed-off-by: Greg Kroah-Hartman

KVM: arm/arm64: vgic-v2: Set active_source to 0 when restoring state

2019-01-09T16:46:04+00:00

commit 60c3ab30d8c2ff3a52606df03f05af2aae07dc6b upstream. When restoring the active state from userspace, we don't know which CPU was the source for the active state, and this is not architecturally exposed in any of the register state. Set the active_source to 0 in this case. In the future, we can expand on this and exposse the information as additional information to userspace for GICv2 if anyone cares. Cc: stable@vger.kernel.org Signed-off-by: Christoffer Dall Signed-off-by: Marc Zyngier Signed-off-by: Greg Kroah-Hartman

KVM: arm/arm64: vgic: Cap SPIs to the VM-defined maximum

2019-01-09T16:46:04+00:00

commit bea2ef803ade3359026d5d357348842bca9edcf1 upstream. SPIs should be checked against the VMs specific configuration, and not the architectural maximum. Cc: stable@vger.kernel.org Signed-off-by: Marc Zyngier Signed-off-by: Greg Kroah-Hartman

KVM: arm/arm64: vgic: Do not cond_resched_lock() with IRQs disabled

2019-01-09T16:46:04+00:00

commit 2e2f6c3c0b08eed3fcf7de3c7684c940451bdeb1 upstream. To change the active state of an MMIO, halt is requested for all vcpus of the affected guest before modifying the IRQ state. This is done by calling cond_resched_lock() in vgic_mmio_change_active(). However interrupts are disabled at this point and we cannot reschedule a vcpu. We actually don't need any of this, as kvm_arm_halt_guest ensures that all the other vcpus are out of the guest. Let's just drop that useless code. Signed-off-by: Julien Thierry Suggested-by: Christoffer Dall Cc: stable@vger.kernel.org Signed-off-by: Marc Zyngier Signed-off-by: Greg Kroah-Hartman

arm/arm64: KVM: vgic: Force VM halt when changing the active state of GICv3 PPIs/SGIs

2019-01-09T16:46:03+00:00

commit 107352a24900fb458152b92a4e72fbdc83fd5510 upstream. We currently only halt the guest when a vCPU messes with the active state of an SPI. This is perfectly fine for GICv2, but isn't enough for GICv3, where all vCPUs can access the state of any other vCPU. Let's broaden the condition to include any GICv3 interrupt that has an active state (i.e. all but LPIs). Cc: stable@vger.kernel.org Reviewed-by: Christoffer Dall Signed-off-by: Marc Zyngier Signed-off-by: Greg Kroah-Hartman

KVM: fix unregistering coalesced mmio zone from wrong bus

2018-12-18T21:07:25+00:00

If you register a kvm_coalesced_mmio_zone with '.pio = 0' but then unregister it with '.pio = 1', KVM_UNREGISTER_COALESCED_MMIO will try to unregister it from KVM_PIO_BUS rather than KVM_MMIO_BUS, which is a no-op. But it frees the kvm_coalesced_mmio_dev anyway, causing a use-after-free. Fix it by only unregistering and freeing the zone if the correct value of 'pio' is provided. Reported-by: syzbot+f87f60bb6f13f39b54e3@syzkaller.appspotmail.com Fixes: 0804c849f1df ("kvm/x86 : add coalesced pio support") Signed-off-by: Eric Biggers Signed-off-by: Paolo Bonzini