It was found that a malicious HVM guest administrator can cause DoS, specifically prevent use of physical CPU for significant, perhaps indefinite period. When a benign exception occurs while delivering another benign exception, it is architecturally specified that these would be delivered sequentially. There are, however, cases where this results in an infinite loop inside the CPU, which (in the virtualized case) can be broken only by intercepting delivery of the respective exception.
When a guest sets up a hardware breakpoint covering a data structure involved in delivering #DB (Debug Exception), upon completion of the delivery of the first exception another #DB will need to be delivered. The effects slightly differ depending on further guest characteristics:
Guests running in 32-bit mode would be expected to sooner or later encounter another fault due to the stack pointer decreasing during each iteration of the loop. The most likely case would be #PF (Page Fault) due to running into unmapped virtual space. However, an infinite loop cannot be excluded (e.g. when the guest is running with paging disabled).
Guests running in long mode, but not using the IST (Interrupt Stack Table) feature for the IDT entry corresponding to #DB would behave similarly to guests running in 32-bit mode, just that the larger virtual address space allows for a much longer loop. The loop can't, however, be infinite, as eventually the stack pointer would move into non-canonical address space, causing #SS (Stack Fault) instead.
Guests running in long mode and using the IST for the IDT entry corresponding to #DB would enter an infinite loop, as the stack pointer wouldn't change between #DB instances.
If a host watchdog (Xen or dom0) is in use, this can lead to a watchdog timeout and consequently a reboot of the host. If another, innocent, guest, is configured with a watchdog, this issue can lead to a reboot of such a guest.
A privileged user inside guest could use this flaw to crash the host kernel resulting in DoS.
For KVM virtualisation, it only affects the AMD processor support, as for Intel it already intercepts the #DB exception.
Upstream KVM patch: ------------------- -> http://permalink.gmane.org/gmane.linux.kernel/2082332
References: ----------- -> http://www.openwall.com/lists/oss-security/2015/11/10/1
ISSUE DESCRIPTION =================
Guests are currently permitted to modify all of the (writable) bits in the PCI command register of devices passed through to them. This in particular allows them to disable memory and I/O decoding on the device unless the device is an SR-IOV virtual function, in which case subsequent accesses to the respective MMIO or I/O port ranges would - - on PCI Express devices - lead to Unsupported Request responses. The treatmeant of such errors is platform specific.
IMPACT ======
In the event that the platform surfaces aforementioned UR responses as Non-Maskable Interrupts, and either the OS is configured to treat NMIs as fatal or (e.g. via ACPI's APEI) the platform tells the OS to treat these errors as fatal, the host would crash, leading to a Denial of Service.
VULNERABLE SYSTEMS ==================
Xen versions 3.3 and onwards are vulnerable due to supporting PCI pass-through. Upstream Linux versions 3.1 and onwards are vulnerable due to supporting PCI backend functionality. Other Linux versions as well as other OS versions may be vulnerable too.
Any domain which is given access to a non-SR-IOV virtual function PCI Express device can take advantage of this vulnerability.
MITIGATION ==========
This issue can be avoided by not assigning PCI Express devices other than SR-IOV virtual functions to untrusted guests.
RESOLUTION ==========
Applying the attached patch resolves this issue for upstream Linux.
xsa120.patch Linux 3.19
$ sha256sum xsa120.patch 5167215293d4a8a05f090fca5b20eb5878213a0158a0e7a12c245553db81a855 xsa120.patch
An issue was discovered in Xen through 4.14.x. Out of bounds event channels are available to 32-bit x86 domains. The so called 2-level event channel model imposes different limits on the number of usable event channels for 32-bit x86 domains vs 64-bit or Arm (either bitness) ones. 32-bit x86 domains can use only 1023 channels, due to limited space in their shared (between guest and Xen) information structure, whereas all other domains can use up to 4095 in this model. The recording of the respective limit during domain initialization, however, has occurred at a time where domains are still deemed to be 64-bit ones, prior to actually honoring respective domain properties. At the point domains get recognized as 32-bit ones, the limit didn't get updated accordingly. Due to this misbehavior in Xen, 32-bit domains (including Domain 0) servicing other domains may observe event channel allocations to succeed when they should really fail. Subsequent use of such event channels would then possibly lead to corruption of other parts of the shared info structure. An unprivileged guest may cause another domain, in particular Domain 0, to misbehave. This may lead to a Denial of Service (DoS) for the entire system. All Xen versions from 4.4 onwards are vulnerable. Xen versions 4.3 and earlier are not vulnerable. Only x86 32-bit domains servicing other domains are vulnerable. Arm systems, as well as x86 64-bit domains, are not vulnerable.
An issue was discovered in Xen through 4.14.x. There is mishandling of the constraint that once-valid event channels may not turn invalid. Logic in the handling of event channel operations in Xen assumes that an event channel, once valid, will not become invalid over the life time of a guest. However, operations like the resetting of all event channels may involve decreasing one of the bounds checked when determining validity. This may lead to bug checks triggering, crashing the host. An unprivileged guest may be able to crash Xen, leading to a Denial of Service (DoS) for the entire system. All Xen versions from 4.4 onwards are vulnerable. Xen versions 4.3 and earlier are not vulnerable. Only systems with untrusted guests permitted to create more than the default number of event channels are vulnerable. This number depends on the architecture and type of guest. For 32-bit x86 PV guests, this is 1023; for 64-bit x86 PV guests, and for all ARM guests, this number is 4095. Systems where untrusted guests are limited to fewer than this number are not vulnerable. Note that xl and libxl limit maxeventchannels to 1023 by default, so systems using exclusively xl, libvirt+libxl, or their own toolstack based on libxl, and not explicitly setting maxeventchannels, are not vulnerable.
The dommuupdate function in arch/x86/mm.c in Xen 3.2.x through 4.4.x does not properly manage page references, which allows remote domains to cause a denial of service by leveraging control over an HVM guest and a crafted MMUMACHPHYSUPDATE.
The acceleration support for the "REP MOVS" instruction in Xen 4.4.x, 3.2.x, and earlier lacks properly bounds checking for memory mapped I/O (MMIO) emulated in the hypervisor, which allows local HVM guests to cause a denial of service (host crash) via unspecified vectors.
The compatibility mode hypercall argument translation in Xen 3.3.x through 4.4.x, when running on a 64-bit hypervisor, allows local 32-bit HVM guests to cause a denial of service (host crash) via vectors involving altering the high halves of registers while in 64-bit mode.
arch/x86/x86emulate/x86emulate.c in Xen 3.2.1 through 4.4.x does not properly check privileges, which allows local HVM guest users to gain privileges or cause a denial of service (crash) via a crafted (1) CALL, (2) JMP, (3) RETF, (4) LCALL, (5) LJMP, or (6) LRET far branch instruction.
The x86emulate function in arch/x86/x86emulate/x86emulate.c in Xen 3.3.x through 4.4.x does not check the supervisor mode permissions for instructions that generate software interrupts, which allows local HVM guest users to cause a denial of service (guest crash) via unspecified vectors.
Race condition in HVMOPtrackdirtyvram in Xen 4.0.0 through 4.4.x does not ensure possession of the guarding lock for dirty video RAM tracking, which allows certain local guest domains to cause a denial of service via unspecified vectors.
The x86emulate function in arch/x86/x86emulate/x86emulate.c in Xen 4.4.x and earlier does not properly check supervisor mode permissions, which allows local HVM users to cause a denial of service (guest crash) or gain guest kernel mode privileges via vectors involving an (1) HLT, (2) LGDT, (3) LIDT, or (4) LMSW instruction.
The hvmmsrreadintercept function in arch/x86/hvm/hvm.c in Xen 4.1 through 4.4.x uses an improper MSR range for x2APIC emulation, which allows local HVM guests to cause a denial of service (host crash) or read data from the hypervisor or other guests via unspecified vectors.
The evtchnfifosetpending function in Xen 4.4.x allows local guest users to cause a denial of service (host crash) via vectors involving an uninitialized FIFO-based event channel control block when (1) binding or (2) moving an event to a different VCPU.
Certain MMU virtualization operations in Xen 4.2.x through 4.4.x before the xsa97-hap patch, when using Hardware Assisted Paging (HAP), are not preemptible, which allows local HVM guest to cause a denial of service (vcpu consumption) by invoking these operations, which process every page assigned to a guest, a different vulnerability than CVE-2014-5149.
Certain MMU virtualization operations in Xen 4.2.x through 4.4.x, when using shadow pagetables, are not preemptible, which allows local HVM guest to cause a denial of service (vcpu consumption) by invoking these operations, which process every page assigned to a guest, a different vulnerability than CVE-2014-5146.
Xen 4.4.x, when running a 64-bit kernel on an ARM system, does not properly handle traps from the guest domain that use a different address width, which allows local guest users to cause a denial of service (host crash) via a crafted 32-bit process.
Xen 4.4.x, when running on an ARM system and "handling an unknown system register access from 64-bit userspace," returns to an instruction of the trap handler for kernel space faults instead of an instruction that is associated with faults in 64-bit userspace, which allows local guest users to cause a denial of service (crash) and possibly gain privileges via a crafted process.
Xen 4.6.x, 4.5.x, 4.4.x, 4.3.x, and earlier do not initialize x86 FPU stack and XMM registers when XSAVE/XRSTOR are not used to manage guest extended register state, which allows local guest domains to obtain sensitive information from other domains via unspecified vectors.
The memoryexchange function in common/memory.c in Xen 3.2.x through 4.6.x does not properly release locks, which might allow guest OS administrators to cause a denial of service (deadlock or host crash) via unspecified vectors, related to XENMEMexchange error handling.
The libxl toolstack library in Xen 4.1.x through 4.6.x does not properly release mappings of files used as kernels and initial ramdisks when managing multiple domains in the same process, which allows attackers to cause a denial of service (memory and disk consumption) by starting domains.
The memoryexchange function in common/memory.c in Xen 3.2.x through 4.6.x does not properly hand back pages to a domain, which might allow guest OS administrators to cause a denial of service (host crash) via unspecified vectors related to domain teardown.
The (1) libxlsetmemorytarget function in tools/libxl/libxl.c and (2) libxlbuildpost function in tools/libxl/libxldom.c in Xen 3.4.x through 4.6.x do not properly calculate the balloon size when using the populate-on-demand (PoD) system, which allows local HVM guest users to cause a denial of service (guest crash) via unspecified vectors related to "heavy memory pressure."
Multiple memory leaks in Xen 4.0 through 4.6.x allow local guest administrators or domains with certain permission to cause a denial of service (memory consumption) via a large number of "teardowns" of domains with the vcpu pointer array allocated using the (1) XENDOMCTLmaxvcpus hypercall or the xenoprofile state vcpu pointer array allocated using the (2) XENOPROFgetbuffer or (3) XENOPROFsetpassive hypercall.
Xen 3.2.x through 4.6.x does not limit the number of printk console messages when logging certain pmu and profiling hypercalls, which allows local guests to cause a denial of service via a sequence of crafted (1) HYPERCALLxenoprofop hypercalls, which are not properly handled in the doxenoprofop function in common/xenoprof.c, or (2) HYPERVISORxenpmuop hypercalls, which are not properly handled in the doxenpmuop function in arch/x86/cpu/vpmu.c.
The hypercallcreatecontinuation function in arch/arm/domain.c in Xen 4.4.x through 4.6.x allows local guest users to cause a denial of service (host crash) via a preemptible hypercall to the multicall interface.
The modl2entry function in arch/x86/mm.c in Xen 3.4 through 4.6.x does not properly validate level 2 page table entries, which allows local PV guest administrators to gain privileges via a crafted superpage mapping.
Xen 4.4.x, 4.5.x, and 4.6.x does not limit the number of printk console messages when reporting unimplemented hypercalls, which allows local guests to cause a denial of service via a sequence of (1) HYPERVISORphysdevop hypercalls, which are not properly handled in the dophysdevop function in arch/arm/physdev.c, or (2) HYPERVISORhvmop hypercalls, which are not properly handled in the dohvmop function in arch/arm/hvm.c.
libxl in Xen 4.1.x through 4.6.x does not properly handle the readonly flag on disks when using the qemu-xen device model, which allows local guest users to write to a read-only disk image.
The xenmemaddtophysmapone function in arch/arm/mm.c in Xen 4.5.x, 4.4.x, and earlier does not limit the number of printk console messages when reporting a failure to retrieve a reference on a foreign page, which allows remote domains to cause a denial of service by leveraging permissions to map the memory of a foreign guest.
It was found that a guest can DoS a host by triggering an infinite loop in microcode. If a guest in 32-bit mode enabled alignment exceptions, puts the exception handler in ring 3, and then triggers an alignment exception with an unaligned stack, then the microcode will enter an infinite loop. Because there's no instruction boundary the core never receives another interrupt (including SMIs). The host kernel panics pretty quickly due to the effects.
A privileged user inside guest could use this flaw to crash the host kernel resulting in DoS.
Upstream KVM patch: ------------------- -> http://permalink.gmane.org/gmane.linux.kernel/2082329
References: ----------- -> http://www.openwall.com/lists/oss-security/2015/11/10/1