xen/xsa164.patch
Charles Arnold 31905d81fa - Update to Xen Version 4.6.1
xen-4.6.1-testing-src.tar.bz2
- Dropped patches now contained in tarball or unnecessary
  xen-4.6.0-testing-src.tar.bz2
  5604f239-x86-PV-properly-populate-descriptor-tables.patch
  561bbc8b-VT-d-don-t-suppress-invalidation-address-write-when-it-is-zero.patch
  561d2046-VT-d-use-proper-error-codes-in-iommu_enable_x2apic_IR.patch
  561d20a0-x86-hide-MWAITX-from-PV-domains.patch
  561e3283-x86-NUMA-fix-SRAT-table-processor-entry-parsing-and-consumption.patch
  5632118e-arm-Support-hypercall_create_continuation-for-multicall.patch
  56321222-arm-rate-limit-logging-from-unimplemented-PHYSDEVOP-and-HVMOP.patch
  56321249-arm-handle-races-between-relinquish_memory-and-free_domheap_pages.patch
  5632127b-x86-guard-against-undue-super-page-PTE-creation.patch
  5632129c-free-domain-s-vcpu-array.patch
  563212c9-x86-PoD-Eager-sweep-for-zeroed-pages.patch
  563212e4-xenoprof-free-domain-s-vcpu-array.patch
  563212ff-x86-rate-limit-logging-in-do_xen-oprof-pmu-_op.patch
  56323737-libxl-adjust-PoD-target-by-memory-fudge-too.patch
  56377442-x86-PoD-Make-p2m_pod_empty_cache-restartable.patch
  5641ceec-x86-HVM-always-intercept-AC-and-DB.patch
  56549f24-x86-vPMU-document-as-unsupported.patch
  5677f350-x86-make-debug-output-consistent-in-hvm_set_callback_via.patch
  xen-4.6.0-testing-src.tar.bz2
  xsa155-qemut-qdisk-double-access.patch
  xsa155-qemut-xenfb.patch
  xsa155-qemuu-qdisk-double-access.patch
  xsa155-qemuu-xenfb.patch
  xsa159.patch
  xsa160.patch
  xsa162-qemut.patch

OBS-URL: https://build.opensuse.org/package/show/Virtualization/xen?expand=0&rev=399
2016-02-12 16:58:27 +00:00

38 lines
1.4 KiB
Diff

References: bsc#958007 XSA-164
MSI-X: avoid array overrun upon MSI-X table writes
pt_msix_init() allocates msix->msix_entry[] to just cover
msix->total_entries entries. While pci_msix_readl() resorts to reading
physical memory for out of bounds reads, pci_msix_writel() so far
simply accessed/corrupted unrelated memory.
pt_iomem_map()'s call to cpu_register_physical_memory() registers a
page granular region, which is necessary as the Pending Bit Array may
share space with the MSI-X table (but nothing else is allowed to). This
also explains why pci_msix_readl() actually honors out of bounds reads,
but pci_msi_writel() doesn't need to.
This is XSA-164.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Index: xen-4.6.1-testing/tools/qemu-xen-traditional-dir-remote/hw/pt-msi.c
===================================================================
--- xen-4.6.1-testing.orig/tools/qemu-xen-traditional-dir-remote/hw/pt-msi.c
+++ xen-4.6.1-testing/tools/qemu-xen-traditional-dir-remote/hw/pt-msi.c
@@ -447,6 +447,13 @@ static void pci_msix_writel(void *opaque
return;
}
+ if ( addr - msix->mmio_base_addr >= msix->total_entries * 16 )
+ {
+ PT_LOG("Error: Out of bounds write to MSI-X table,"
+ " addr %016"PRIx64"\n", addr);
+ return;
+ }
+
entry_nr = (addr - msix->mmio_base_addr) / 16;
entry = &msix->msix_entry[entry_nr];
offset = ((addr - msix->mmio_base_addr) % 16) / 4;