The 'virt-what' program prints facts that reflect the hypervisor that
the guest is running under.
The new complementary 'virt-what-cvm' program prints facts that reflect
the confidential virtualization technology the guest is running under,
if any.
It is kept as a separate tool, rather than incorporating the facts into
'virt-what' output because it is considering a different aspect of the
virtualization. Furthermore there are specific security concerns around
the usage of facts reported by 'virt-what-cvm'.
The tool has been tested in a number of environments
* Azure confidential guest with AMD SEV-SNP (GA)
* Azure confidential guest with Intel TDX (technology preview)
* Fedora 37 QEMU/KVM guest with AMD SEV (GA)
* Fedora 37 QEMU/KVM guest with AMD SEV-ES (GA)
* Fedora 38 QEMU/KVM guest with AMD SEV-SNP + SVSM (devel snapshot)
Signed-off-by: Daniel P. Berrangé <berrange@xxxxxxxxxx>
---
.gitignore | 3 +
Makefile.am | 12 +-
configure.ac | 3 +
virt-what-cvm.c | 404 ++++++++++++++++++++++++++++++++++++++++++++++
virt-what-cvm.pod | 195 ++++++++++++++++++++++
5 files changed, 613 insertions(+), 4 deletions(-)
create mode 100644 virt-what-cvm.c
create mode 100644 virt-what-cvm.pod
diff --git a/.gitignore b/.gitignore
index 4833fd6..ba897a1 100644
--- a/.gitignore
+++ b/.gitignore
@@ -26,5 +26,8 @@ Makefile.in
/test-driver
/virt-what
/virt-what-cpuid-helper
+/virt-what-cvm
+/virt-what-cvm.1
+/virt-what-cvm.txt
/virt-what.1
/virt-what.txt
diff --git a/Makefile.am b/Makefile.am
index 5435132..2050bef 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -24,20 +24,24 @@ EXTRA_DIST = .gitignore virt-what.in virt-what.pod
SUBDIRS = . tests
sbin_SCRIPTS = virt-what
+sbin_PROGRAMS = virt-what-cvm
libexec_PROGRAMS = virt-what-cpuid-helper
if HOST_CPU_IA64
libexec_PROGRAMS += virt-what-ia64-xen-rdtsc-test
endif
+virt_what_cvm_LDADD = $(TPM2_TSS_LIBS)
+virt_what_cvm_CFLAGS = $(TPM2_TSS_CFLAGS)
+
if HAVE_POD2MAN
-CLEANFILES += virt-what.1 virt-what.txt
-man_MANS = virt-what.1
+CLEANFILES += virt-what.1 virt-what-cvm.1 virt-what.txt virt-what-cvm.txt
+man_MANS = virt-what.1 virt-what-cvm.1
-virt-what.1: virt-what.pod
+%.1: %.pod
pod2man -c "Virtualization Support" --release "$(PACKAGE)-$(VERSION)" \
$? > $@
-virt-what.txt: virt-what.pod
+%.txt: %.pod
pod2text $? > $@
endif
diff --git a/configure.ac b/configure.ac
index a28a716..77b7665 100644
--- a/configure.ac
+++ b/configure.ac
@@ -32,6 +32,9 @@ dnl Architecture we are compiling for.
AC_CANONICAL_HOST
AM_CONDITIONAL([HOST_CPU_IA64], [ test "x$host_cpu" = "xia64" ])
+PKG_HAVE_DEFINE_WITH_MODULES(TPM2_TSS, tss2-esys, [tpm2-tss package])
+
+
dnl List of tests.
tests="\
alibaba-cloud-arm \
diff --git a/virt-what-cvm.c b/virt-what-cvm.c
new file mode 100644
index 0000000..407efb4
--- /dev/null
+++ b/virt-what-cvm.c
@@ -0,0 +1,404 @@
+/* virt-what-cvm-helper: Are we running inside confidential VM
+ * Copyright (C) 2023 Red Hat Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include "config.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <string.h>
+#include <stdbool.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <getopt.h>
+#ifdef HAVE_TPM2_TSS
+#include <tss2/tss2_esys.h>
+#include <assert.h>
+#endif
+
+static bool dodebug = false;
+
+#define debug(...) do { if (dodebug) fprintf(stderr, __VA_ARGS__); } while(0)
+
+/*
+ * AMD64 Architecture Programmer’s Manual Volume 3:
+ * General-Purpose and System Instructions.
+ * Chapter: E4.1 - Maximum Extended Function Number and Vendor String
+ * https://www.amd.com/system/files/TechDocs/24594.pdf
+ */
+#define CPUID_GET_HIGHEST_FUNCTION 0x80000000
+
+/*
+ * AMD64 Architecture Programmer’s Manual Volume 3:
+ * General-Purpose and System Instructions.
+ * Chapter: E4.17 - Encrypted Memory Capabilities
+ * https://www.amd.com/system/files/TechDocs/24594.pdf
+ */
+#define CPUID_AMD_GET_ENCRYPTED_MEMORY_CAPABILITIES 0x8000001f
+
+/*
+ * AMD64 Architecture Programmer’s Manual Volume 3:
+ * General-Purpose and System Instructions.
+ * Chapter: 15.34.10 - SEV_STATUS MSR
+ * https://www.amd.com/system/files/TechDocs/24593.pdf
+ */
+#define MSR_AMD64_SEV 0xc0010131
+
+/*
+ * Intel® TDX Module v1.5 Base Architecture Specification
+ * Chapter: 11.2
+ * https://www.intel.com/content/www/us/en/content-details/733575/intel-tdx-module-v1-5-base-architecture-specification.html
+ */
+
+#define CPUID_INTEL_TDX_ENUMERATION 0x21
+
+
+#define CPUID_SIG_AMD "AuthenticAMD"
+#define CPUID_SIG_INTEL "GenuineIntel"
+#define CPUID_SIG_INTEL_TDX "IntelTDX "
+
+/*
+ * This TPM NV data format is not explicitly documented anywhere,
+ * but the header definition is present in code at:
+ *
+ * https://github.com/kinvolk/azure-cvm-tooling/blob/main/az-snp-vtpm/src/hcl.rs
+ */
+#define TPM_AZURE_HCLA_REPORT_INDEX 0x01400001
+
+struct TPMAzureHCLAHeader {
+ uint32_t signature;
+ uint32_t version;
+ uint32_t report_len;
+ uint32_t report_type;
+ uint32_t unknown[4];
+};
+
+/* The bytes for "HCLA" */
+#define TPM_AZURE_HCLA_SIGNATURE 0x414C4348
+#define TPM_AZURE_HCLA_VERSION 0x1
+#define TPM_AZURE_HCLA_REPORT_TYPE_SNP 0x2
+
+#if defined(__x86_64__)
+
+#ifdef HAVE_TPM2_TSS
+static char *
+tpm_nvread(uint32_t nvindex, size_t *retlen)
+{
+ TSS2_RC rc;
+ ESYS_CONTEXT *ctx = NULL;
+ ESYS_TR primary = ESYS_TR_NONE;
+ ESYS_TR session = ESYS_TR_NONE;
+ ESYS_TR nvobj = ESYS_TR_NONE;
+ TPM2B_NV_PUBLIC *pubData = NULL;
+ TPMT_SYM_DEF sym = {
+ .algorithm = TPM2_ALG_AES,
+ .keyBits = { .aes = 128 },
+ .mode = { .aes = TPM2_ALG_CFB }
+ };
+ char *ret;
+ size_t retwant;
+
+ rc = Esys_Initialize(&ctx, NULL, NULL);
+ if (rc != TSS2_RC_SUCCESS)
+ return NULL;
+
+ rc = Esys_Startup(ctx, TPM2_SU_CLEAR);
+ debug("tpm startup %d\n", rc);
+ if (rc != TSS2_RC_SUCCESS)
+ goto error;
+
+ rc = Esys_StartAuthSession(ctx, ESYS_TR_NONE, ESYS_TR_NONE,
+ ESYS_TR_NONE, ESYS_TR_NONE, ESYS_TR_NONE,
+ NULL, 0,
+ &sym, TPM2_ALG_SHA256, &session);
+ debug("tpm auth session %d\n", rc);
+ if (rc != TSS2_RC_SUCCESS)
+ goto error;
+
+ rc = Esys_TR_FromTPMPublic(ctx, nvindex, ESYS_TR_NONE,
+ ESYS_TR_NONE, ESYS_TR_NONE, &nvobj);
+ debug("tpm from public %d\n", rc);
+ if (rc != TSS2_RC_SUCCESS)
+ goto error;
+
+ rc = Esys_NV_ReadPublic(ctx, nvobj, ESYS_TR_NONE,
+ ESYS_TR_NONE, ESYS_TR_NONE,
+ &pubData, NULL);
+ debug("tpm read public %d\n", rc);
+ if (rc != TPM2_RC_SUCCESS)
+ goto error;
+
+ retwant = pubData->nvPublic.dataSize;
+ free(pubData);
+ *retlen = 0;
+ ret = malloc(retwant);
+ assert(ret);
+ while (*retlen < retwant) {
+ size_t want = retwant - *retlen;
+ TPM2B_MAX_NV_BUFFER *data = NULL;
+ if (want > 1024)
+ want = 1024;
+ rc = Esys_NV_Read(ctx, ESYS_TR_RH_OWNER, nvobj, session, ESYS_TR_NONE, ESYS_TR_NONE,
+ want, *retlen, &data);
+ debug("tpm nv read %d\n", rc);
+ if (rc != TPM2_RC_SUCCESS) {
+ free(ret);
+ goto error;
+ }
+
+ memcpy(ret + *retlen, data->buffer, data->size);
+ *retlen += data->size;
+ free(data);
+ }
+
+ return ret;
+
+ error:
+ if (nvobj != ESYS_TR_NONE)
+ Esys_FlushContext(ctx, nvobj);
+ if (session != ESYS_TR_NONE)
+ Esys_FlushContext(ctx, session);
+ if (primary != ESYS_TR_NONE)
+ Esys_FlushContext(ctx, primary);
+ Esys_Finalize(&ctx);
+ *retlen = 0;
+ return NULL;
+}
+#else /* ! HAVE_TPM2_TSS */
+static char *
+tpm_nvread(uint32_t nvindex, size_t *retlen)
+{
+ return NULL;
+}
+#endif /* ! HAVE_TPM2_TSS */
+
+/* Copied from the Linux kernel definition in
+ * arch/x86/include/asm/processor.h
+ */
+static inline void
+cpuid (uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx)
+{
+ debug("CPUID func %x %x\n", *eax, *ecx);
+ asm volatile ("cpuid"
+ : "=a" (*eax), "=b" (*ebx), "=c" (*ecx), "=d" (*edx)
+ : "0" (*eax), "2" (*ecx)
+ : "memory");
+ debug("CPUID result %x %x %x %x\n", *eax, *ebx, *ecx, *edx);
+}
+
+
+static uint32_t
+cpuid_leaf (uint32_t eax, char *sig)
+{
+ uint32_t *sig32 = (uint32_t *) sig;
+
+ cpuid (&eax, &sig32[0], &sig32[2], &sig32[1]);
+ sig[12] = 0; /* \0-terminate the string to make string comparison possible */
+ debug("CPUID sig %s\n", sig);
+ return eax;
+}
+
+#define MSR_DEVICE "/dev/cpu/0/msr"
+
+static uint64_t
+msr (off_t index)
+{
+ uint64_t ret;
+ int fd = open (MSR_DEVICE, O_RDONLY);
+ if (fd < 0) {
+ debug ("Cannot open MSR device %s", MSR_DEVICE);
+ return 0;
+ }
+
+ if (pread (fd, &ret, sizeof(ret), index) != sizeof(ret))
+ ret = 0;
+
+ close (fd);
+
+ debug ("MSR %llx result %llx\n", (unsigned long long)index,
+ (unsigned long long)ret);
+ return ret;
+}
+
+bool
+cpu_sig_amd_azure (void)
+{
+ size_t datalen = 0;
+ char *data = tpm_nvread(TPM_AZURE_HCLA_REPORT_INDEX, &datalen);
+ struct TPMAzureHCLAHeader *header = (struct TPMAzureHCLAHeader *)data;
+ bool ret;
+
+ if (!data)
+ return false;
+
+ if (datalen < sizeof(struct TPMAzureHCLAHeader)) {
+ debug ("TPM data len is too small to be an Azure HCLA report");
+ return false;
+ }
+
+ debug ("Azure TPM HCLA report header sig %x ver %x type %x\n",
+ header->signature, header->version, header->report_type);
+
+ ret = (header->signature == TPM_AZURE_HCLA_SIGNATURE &&
+ header->version == TPM_AZURE_HCLA_VERSION &&
+ header->report_type == TPM_AZURE_HCLA_REPORT_TYPE_SNP);
+ debug ("Azure TPM HCLA report present ? %d\n", ret);
+
+ free(data);
+ return ret;
+}
+
+static void
+cpu_sig_amd (void)
+{
+ uint32_t eax, ebx, ecx, edx;
+ uint64_t msrval;
+
+ eax = CPUID_GET_HIGHEST_FUNCTION;
+ ebx = ecx = edx = 0;
+
+ cpuid (&eax, &ebx, &ecx, &edx);
+
+ if (eax < CPUID_AMD_GET_ENCRYPTED_MEMORY_CAPABILITIES)
+ return;
+
+ eax = CPUID_AMD_GET_ENCRYPTED_MEMORY_CAPABILITIES;
+ ebx = ecx = edx = 0;
+
+ cpuid (&eax, &ebx, &ecx, &edx);
+
+ /* bit 1 == CPU supports SEV feature
+ *
+ * Note, Azure blocks this CPUID leaf from its SEV-SNP
+ * guests, so we must fallback to probing the TPM which
+ * exposes a SEV-SNP attestation report as evidence.
+ */
+ if (!(eax & (1 << 1))) {
+ debug ("No sev in CPUID, try azure TPM NV\n");
+
+ if (cpu_sig_amd_azure()) {
+ puts ("amd-sev-snp");
+ puts ("azure-hcl");
+ } else {
+ debug("No azure TPM NV\n");
+ }
+ return;
+ }
+
+ msrval = msr (MSR_AMD64_SEV);
+
+ /* Test reverse order, since the SEV-SNP bit implies
+ * the SEV-ES bit, which implies the SEV bit */
+ if (msrval & (1 << 2)) {
+ puts ("amd-sev-snp");
+ } else if (msrval & (1 << 1)) {
+ puts ("amd-sev-es");
+ } else if (msrval & (1 << 0)) {
+ puts ("amd-sev");
+ }
+}
+
+static void
+cpu_sig_intel (void)
+{
+ uint32_t eax, ebx, ecx, edx;
+ char sig[13];
+
+ eax = CPUID_GET_HIGHEST_FUNCTION;
+ ebx = ecx = edx = 0;
+
+ cpuid (&eax, &ebx, &ecx, &edx);
+ debug ("CPUID max function: %x %x %x %x\n", eax, ebx, ecx,edx);
+
+ if (eax < CPUID_INTEL_TDX_ENUMERATION)
+ return;
+
+ memset (sig, 0, sizeof sig);
+ cpuid_leaf (CPUID_INTEL_TDX_ENUMERATION, sig);
+
+ if (memcmp (sig, CPUID_SIG_INTEL_TDX, sizeof(sig)) == 0)
+ puts ("intel-tdx");
+}
+
+static void
+cpu_sig (void)
+{
+ char sig[13];
+
+ memset (sig, 0, sizeof sig);
+ cpuid_leaf (0, sig);
+
+ if (memcmp (sig, CPUID_SIG_AMD, sizeof(sig)) == 0)
+ cpu_sig_amd ();
+ else if (memcmp (sig, CPUID_SIG_INTEL, sizeof(sig)) == 0)
+ cpu_sig_intel ();
+}
+
+#else /* !x86_64 */
+
+static void
+cpu_sig (void)
+{
+ /* nothing for other architectures */
+}
+
+#endif
+
+int
+main(int argc, char **argv)
+{
+ int c;
+
+ while (true) {
+ int option_index = 0;
+ static struct option long_options[] = {
+ {"debug", no_argument, 0, 'd' },
+ {"version", no_argument, 0, 'v' },
+ {"help", no_argument, 0, 'h'},
+ {0, 0, 0, 0 }
+ };
+
+ c = getopt_long(argc, argv, "dvh",
+ long_options, &option_index);
+ if (c == -1)
+ break;
+
+ switch (c) {
+ case 'd':
+ dodebug = true;
+ break;
+ case 'v':
+ fprintf(stdout, "%s\n", PACKAGE_VERSION);
+ exit(EXIT_SUCCESS);
+ break;
+ case 'h':
+ default: /* '?' */
+ fprintf(c == 'h' ? stdout : stderr,
+ "Usage: %s [--debug|-d] [--help|-h] [--version|-v]\n",
+ argv[0]);
+ exit(c == 'h' ? EXIT_SUCCESS : EXIT_FAILURE);
+ }
+ }
+
+ if (!dodebug)
+ setenv("TSS2_LOG", "all+none", 1);
+
+ cpu_sig ();
+
+ exit(EXIT_SUCCESS);
+}
diff --git a/virt-what-cvm.pod b/virt-what-cvm.pod
new file mode 100644
index 0000000..12cfc6a
--- /dev/null
+++ b/virt-what-cvm.pod
@@ -0,0 +1,195 @@
+=encoding utf8
+
+=head1 NAME
+
+virt-what-cvm - detect if we are running in a confidential virtual machine
+
+=head1 SUMMARY
+
+virt-what-cvm [options]
+
+=head1 DESCRIPTION
+
+C<virt-what-cvm> is a tool which can be used to detect if the program
+is running in a confidential virtual machine.
+
+The program prints out a list of "facts" about the confidential virtual
+machine, derived from heuristics. One fact is printed per line.
+
+If nothing is printed and the script exits with code 0 (no error),
+then it can mean I<either> that the program is running on bare-metal
+I<or> the program is running inside a non-confidential virtual machine,
+I<or> inside a type of confidential virtual machine which we don't know
+about or cannot detect.
+
+=head1 FACTS
+
+=over 4
+
+=item B<amd-sev>
+
+This is a confidential guest running with AMD SEV technology
+
+Status: tested on Fedora 37 QEMU+KVM
+
+=item B<amd-sev-es>
+
+This is a confidential guest running with AMD SEV-ES technology
+
+Status: tested on Fedora 37 QEMU+KVM
+
+=item B<amd-sev-snp>
+
+This is a confidential guest running with AMD SEV-SNP technology
+
+Status: tested on Microsoft Azure SEV-SNP CVM
+
+Status: tested on Fedora 38 QEMU+KVM SEV-SNP (devel snapshot)
+
+=item B<intel-tdx>
+
+This is a confidential guest running with Intel TDX technology
+
+Status: tested on Microsoft Azure TDX CVM (preview)
+
+=item B<azure-hcl>
+
+This is a confidential guest running unenlightened under the
+Azure HCL (Host Compatibility Layer). This will be paired with
+B<amd-sev-snp>.
+
+Status: tested on Microsoft Azure SEV-SNP CVM
+
+=back
+
+=head1 EXIT STATUS
+
+Programs that use or wrap C<virt-what-cvm> should check that the exit
+status is 0 before they attempt to parse the output of the command.
+
+A non-zero exit status indicates some error, for example, an
+unrecognized command line argument. If the exit status is non-zero
+then the output "facts" (if any were printed) cannot be guaranteed and
+should be ignored.
+
+The exit status does I<not> have anything to do with whether the
+program is running on baremetal or under confidential virtualization,
+nor with whether C<virt-what-cvm> managed detection "correctly" (which
+is basically unknowable given the large variety of virtualization
+systems out there)
+
+=head1 RUNNING VIRT-WHAT-CVM FROM OTHER PROGRAMS
+
+C<virt-what-cvm> is designed so that you can easily run it from
+other programs or wrap it up in a library.
+
+Your program should check the exit status (see the section above).
+
+=head1 IMPORTANT NOTE
+
+This program detects whether it is likely to be running within a known
+confidential VM, but does I<NOT> prove that the environment is trustworthy.
+To attain trust in the environment requires an attestation report for the
+virtual machine, which is then verified by an already trusted 3rd party.
+
+The hardware features that this program relies on to establish facts
+about the confidential virtualization environment, are those features
+whose behaviour will be proved by verification of an attestation report.
+
+This program I<MAY> have false positives. ie it may report that it is a
+confidential VM when it is in fact a non-confidential VM faking it.
+
+This program I<SHOULD NOT> have false negatives. ie it should not fail to
+report existance of a confidential VM. Caveat that this only applies to
+environments which have been explicitly tested.
+
+If this program does print a fact, this can be used for enabling or
+disabling use of certain features, according to whether they are
+appropriate for a confidential environment. None the less, the VM
+I<MUST NOT> be trusted until an attestation report is verified.
+
+As a protection against false negatives from this tool, environments
+requiring high assurance should take one or more of these measures:
+
+ * The facts reported by this program I<SHOULD> should be measured
+ into one of the TPM PCRs
+ * The attestation report I<SHOULD> cover the facts reported by
+ this program
+ * The attestation report I<SHOULD> should cover the enablement
+ status of any features affected by decisions involving facts
+ reported by this tool
+
+=head1 SEE ALSO
+
+L<http://people.redhat.com/~rjones/virt-what/>,
+L<https://github.com/Azure/confidential-computing-cvm-guest-attestation>,
+L<https://virtee.io/>
+
+=head1 AUTHORS
+
+Daniel P. Berrangé <berrange @ redhat . com>
+
+=head1 COPYRIGHT
+
+(C) Copyright 2023 Red Hat Inc.,
+L<http://people.redhat.com/~rjones/virt-what/>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License
+along with this program; if not, write to the Free Software
+Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+=head1 REPORTING BUGS
+
+Bugs can be viewed on the Red Hat Bugzilla page:
+L<https://bugzilla.redhat.com/>.
+
+If you find a bug in virt-what-cvm, please follow these steps to report it:
+
+=over 4
+
+=item 1. Check for existing bug reports
+
+Go to L<https://bugzilla.redhat.com/> and search for similar bugs.
+Someone may already have reported the same bug, and they may even
+have fixed it.
+
+=item 2. Capture debug and error messages
+
+Run
+
+ virt-what-cvm -d > virt-what-cvm.log 2>&1
+
+and keep I<virt-what-cvm.log>. It may contain error messages which you
+should submit with your bug report.
+
+=item 3. Get version of virt-what-cvm.
+
+Run
+
+ virt-what-cvm --version
+
+=item 4. Submit a bug report.
+
+Go to L<https://bugzilla.redhat.com/> and enter a new bug.
+Please describe the problem in as much detail as possible.
+
+Remember to include the version numbers (step 3) and the debug
+messages file (step 2) and as much other detail as possible.
+
+=item 5. Assign the bug to rjones @ redhat.com
+
+Assign or reassign the bug to B<rjones @ redhat.com> (without the
+spaces). You can also send me an email with the bug number if you
+want a faster response.
+
+=back
--
2.40.1
