.. _asa: Security Advisory ################# Addressed in ACRN v2.7 ************************ We recommend that all developers upgrade to this v2.7 release (or later), which addresses the following security issue discovered in previous releases: ----- - Heap-use-after-free happens in ``MEVENT mevent_handle`` The file descriptor of ``mevent`` could be closed in another thread while being monitored by ``epoll_wait``. This causes a heap-use-after-free error in the ``mevent_handle()`` function. **Affected Release:** v2.6 and earlier Addressed in ACRN v2.6 ************************ We recommend that all developers upgrade to this v2.6 release (or later), which addresses the following security issue discovered in previous releases: ----- - Memory leakage vulnerability in ``devicemodel/hw/pci/xhci.c`` De-initializing of emulated USB devices results in a memory leakage issue as some resources allocated for transfer are not properly released. **Affected Release:** v2.5 and earlier. Addressed in ACRN v2.5 ************************ We recommend that all developers upgrade to this v2.5 release (or later), which addresses the following security issues that were discovered in previous releases: ----- - NULL Pointer Dereference in ``devicemodel/hw/pci/virtio/virtio_net.c`` ``virtio_net_ping_rxq()`` function tries to set ``vq->used->flags`` without validating pointer ``vq->used``, which may be NULL and cause a NULL pointer dereference. **Affected Release:** v2.4 and earlier. - NULL Pointer Dereference in ``hw/pci/virtio/virtio.c`` ``vq_endchains`` function tries to read ``vq->used->idx`` without validating pointer ``vq->used``, which may be NULL and cause a NULL pointer dereference. **Affected Release:** v2.4 and earlier. - NULL Pointer Dereference in ``devicemodel/hw/pci/xhci.c`` The ``trb`` pointer in ``pci_xhci_complete_commands`` function may be from user space and may be NULL. Accessing it without validating may cause a NULL pointer dereference. **Affected Release:** v2.4 and earlier. - Buffer overflow in ``hypervisor/arch/x86/vtd.c`` Malicious input ``index`` for function ``dmar_free_irte`` may trigger buffer overflow on array ``irte_alloc_bitmap[]``. **Affected Release:** v2.4 and earlier. - Page Fault in ``devicemodel/core/mem.c`` ``unregister_mem_int()`` function frees any entry when it is valid, which is not expected. (only entries to be removed from RB tree can be freed). This will cause a page fault when next RB tree iteration happens. **Affected Release:** v2.4 and earlier - Heap-use-after-free happens in VIRTIO timer_handler With virtio polling mode enabled, a timer is running in the virtio backend service. The timer will also be triggered if its frontend driver didn't do the device reset on shutdown. A freed virtio device could be accessed in the polling timer handler. **Affected Release:** v2.4 and earlier Addressed in ACRN v2.3 ************************ We recommend that all developers upgrade to this v2.3 release (or later), which addresses the following security issue that was discovered in previous releases: ----- - NULL Pointer Dereference in ``devicemodel\hw\pci\virtio\virtio_mei.c`` ``vmei_proc_tx()`` function tries to find the ``iov_base`` by calling function ``paddr_guest2host()``, which may return NULL (the ``vd`` struct control by the User VM OS). There is a use of ``iov_base`` afterward that can cause a NULL pointer dereference (CVE-2020-28346). **Affected Release:** v2.2 and earlier. Addressed in ACRN v2.1 ************************ We recommend that all developers upgrade to this v2.1 release (or later), which addresses the following security issue that was discovered in previous releases: ----- - Missing access control restrictions in the Hypervisor component A malicious entity with root access in the Service VM userspace could abuse the PCIe assign/de-assign Hypercalls via crafted ioctls and payloads. This attack can result in a corrupt state and Denial of Service (DoS) for previously assigned PCIe devices to the Service VM at runtime. **Affected Release:** v2.0 and v1.6.1. Addressed in ACRN v1.6.1 ************************ We recommend that all developers upgrade to this v1.6.1 release (or later), which addresses the following security issue that was discovered in previous releases: ----- - Service VM kernel Crashes When Fuzzing HC_ASSIGN_PCIDEV and HC_DEASSIGN_PCIDEV NULL pointer dereference due to invalid address of PCI device to be assigned or de-assigned may result in kernel crash. The return value of 'pci_find_bus()' shall be validated before using in 'update_assigned_vf_state()'. **Affected Release:** v1.6. Addressed in ACRN v1.6 ********************** We recommend that all developers upgrade to this v1.6 release (or later), which addresses the following security issues that were discovered in previous releases: ----- - Hypervisor Crashes When Fuzzing HC_DESTROY_VM The input 'vdev->pdev' should be validated properly when handling HC_SET_PTDEV_INTR_INFO to ensure that the physical device is linked to 'vdev'; otherwise, the hypervisor crashes when fuzzing the hypercall HC_DESTROY_VM with crafted input. **Affected Release:** v1.5 and earlier. - Hypervisor Crashes When Fuzzing HC_VM_WRITE_PROTECT_PAGE The input GPA is not validated when handling this hypercall; an "Invalid GPA" that is not in the scope of the target VM's EPT address space results in the hypervisor crashing when handling this hypercall. **Affected Release:** v1.4 and earlier. - Hypervisor Crashes When Fuzzing HC_NOTIFY_REQUEST_FINISH The input is not validated properly when handing this hypercall; 'vcpu_id' should be less than 'vm->hw.created_vcpus' instead of 'MAX_VCPUS_PER_VM'. When the software fails to validate input properly, the hypervisor crashes when handling crafted inputs. **Affected Release:** v1.4 and earlier. Addressed in ACRN v1.4 ********************** We recommend that all developers upgrade to this v1.4 release (or later), which addresses the following security issues that were discovered in previous releases: ----- - Mitigation for Machine Check Error on Page Size Change Improper invalidation for page table updates by a virtual guest operating system for multiple Intel(R) Processors may allow an authenticated user to potentially enable denial of service of the host system via local access. A malicious guest kernel could trigger this issue, CVE-2018-12207. **Affected Release:** v1.3 and earlier. - AP Trampoline Is Accessible to the Service VM This vulnerability is triggered when validating the memory isolation between the VM and the hypervisor. The AP Trampoline code exists in the LOW_RAM region of the hypervisor but is potentially accessible to the Service VM. This could be used by an attacker to mount DoS attacks on the hypervisor if the Service VM is compromised. **Affected Release:** v1.3 and earlier. - Improper Usage Of the ``LIST_FOREACH()`` Macro Testing discovered that the MACRO ``LIST_FOREACH()`` was incorrectly used in some cases which could induce a "wild pointer" and cause the ACRN Device Model to crash. Attackers can potentially use this issue to cause denial of service (DoS) attacks. **Affected Release:** v1.3 and earlier. - Hypervisor Crashes When Fuzzing HC_SET_CALLBACK_VECTOR This vulnerability was reported by the Fuzzing tool for the debug version of ACRN. When the software fails to validate input properly, an attacker is able to craft the input in a form that is not expected by the rest of the application. This can lead to parts of the system receiving unintended inputs, which may result in an altered control flow, arbitrary control of a resource, or arbitrary code execution. **Affected Release:** v1.3 and earlier. - FILE Pointer Is Not Closed After Using This vulnerability was reported by the Fuzzing tool. Leaving the file unclosed will cause a leaking file descriptor and may cause unexpected errors in the Device Model program. **Affected Release:** v1.3 and earlier. - Descriptor of Directory Stream Is Referenced After Release This vulnerability was reported by the Fuzzing tool. A successful call to ``closedir(DIR *dirp)`` also closes the underlying file descriptor associated with ``dirp``. Access to the released descriptor may point to some arbitrary memory location or cause undefined behavior. **Affected Release:** v1.3 and earlier. - Mutex Is Potentially Kept in a Locked State Forever This vulnerability was reported by the Fuzzing tool. Here, pthread_mutex_lock/unlock pairing was not always done. Leaving a mutex in a locked state forever can cause program deadlock, depending on the usage scenario. **Affected Release:** v1.3 and earlier.