lirent
9b17bdfa33
Force a real, platform-anchored OEM BIOS identity into SMBIOS type0 (patch 0024 + spoof_bios_version/spoof_bios_date getters) so the guest does not read an empty or firmware-default BIOS vendor via WMI/dmidecode. Vendor coheres with the board; version uses the vendor real format. Inert without a seed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
3.3 KiB
Anti-detect patch series
Quilt patches injected (by make prepare) into pve-qemu/debian/patches/series,
applied in order during the .deb build. Each wires a stock QEMU literal to a
spoof_*() getter from the in-tree module (hw/misc/spoof.{c,h}, dropped in by
the Makefile). The getter returns the seed-derived value, or the stock default
when no spoof-seed is set — so the whole series is inert until a VM opts in.
Authored against the pinned pve-qemu/qemu tree (QEMU 11.0) so the context lines
match exactly. Naming: 0002 = infra, 0010+ = one aspect each.
| patch | covers |
|---|---|
| 0002-x86-machine-spoof-properties | register spoof-seed / spoof-hv / spoof-waet / spoof-vmgenid / spoof-pvpanic machine props |
| 0010-acpi-table-header | ACPI OEM id / table id / creator + FADT hypervisor-vendor |
| 0011-acpi-policy-waet-vmgenid | drop the WAET table; skip the vmgenid SSDT when policy = hide |
| 0012-acpi-device-hids | vmgenid _HID (mask), pvpanic _HID (hide), fw_cfg _HID |
| 0013-cpuid-kvm-sig-freq | CPUID KVM signature + leaf 0x16 frequency |
| 0014-fwcfg-signatures | fw_cfg selector + DMA signatures |
| 0015-smbios-vm-bit | clear the SMBIOS type0 "VM" characteristic bit |
| 0016-edid-monitor | EDID vendor / name / model / serial / manufacture-date |
| 0017-storage-identity | IDE model / serial / fw / cdrom + NVMe model / serial / fw |
| 0018-pci-subsystem-id | realize-time OEM-brand of emulated PCI subsystem ids (SVID/SSID) |
| 0019-machine-desc | machine desc string → platform board model |
| 0021-smbios-identity | SMBIOS type3 chassis / type4 socket + cpu-mfr / type11 OEM / type17 memory |
| 0022-storage-extra | IDE WWN + rotation rate; NVMe EUI-64 + NGUID |
| 0023-cpu-microcode | CPU microcode revision (IA32_UCODE_REV), vendor-positioned |
| 0024-smbios-bios-type0 | SMBIOS type0 BIOS vendor / version / date (platform-anchored OEM firmware) |
Why 0018 is subsystem-only (the PCI-id problem)
The PCI vendor:device id IS the guest driver-binding contract: NetKVM binds
1af4:1000, the ivshmem driver binds 1af4:1110, the GPU driver binds its id.
Spoofing those breaks the device (no driver matches). So 0018 only rewrites the
subsystem id (not used for binding) — and even that skips 0x1af4, because
legacy-virtio encodes the device type in the subsystem id. Class-bound devices
(NVMe / xHCI / AHCI bind by class code, not id) could take a real vendor:device,
but that needs a per-role allowlist. virtio / ivshmem / GPU ids are instead
de-fingerprinted by device choice (e1000e / SATA) and vfio x-pci-*, not by
id spoofing.
Out of scope (handled by config or upstream, not these patches)
- PCI vendor:device id for class-bound devices (qemu-nvme
1b36:0010→ a real NVMe vendor;nec-usb-xhci): safe in principle, would need a per-role allowlist in the0018realize hook. - HDA codec id (
QEMU_HDA_ID_VENDOR 0x1af4): a compile-time constant baked into astatic constcodec descriptor; spoofing needs a runtime override of the GET_PARAMETER verb response. - vmport/vmmouse (VMware backdoor port 0x5658): disable via
-machine vmport=off(a machine option). - hv-mode / hypervisor bit: via the CPU model flags (
cpu: host,hidden=1+ hv enlightenments). The module'sspoof_hv_modepolicy is the intent; enforcement is configuration.