x86/pvm: Disable some unsupported syscalls and features

n the PVM guest, the LDT won't be loaded into hardware, rendering it ineffective. Consequently, the modify_ldt() syscall should be disabled. Additionally, the VSYSCALL address is not within the allowed address range, making full emulation of the vsyscall page unsupported in the PVM guest. It is recommended to use XONLY mode instead. Furthermore, SYSENTER (Intel) and SYSCALL32 (AMD) are not supported by the hypervisor, so they should not be used in VDSO. Suggested-by: Lai Jiangshan <[email protected]> Signed-off-by: Hou Wenlong <[email protected]>
loopholelabs · Feb 26, 2024 · ebec6f6 · ebec6f6
1 parent 34c183d
commit ebec6f6
Show file tree

Hide file tree

Showing 3 changed files with 11 additions and 0 deletions.
diff --git a/arch/x86/entry/vsyscall/vsyscall_64.c b/arch/x86/entry/vsyscall/vsyscall_64.c
@@ -378,6 +378,10 @@ void __init map_vsyscall(void)
 	extern char __vsyscall_page;
 	unsigned long physaddr_vsyscall = __pa_symbol(&__vsyscall_page);
 
+	/* Full emulation is not supported in PVM guest, use XONLY instead. */
+	if (vsyscall_mode == EMULATE && boot_cpu_has(X86_FEATURE_KVM_PVM_GUEST))
+		vsyscall_mode = XONLY;
+
 	/*
 	 * For full emulation, the page needs to exist for real.  In
 	 * execute-only mode, there is no PTE at all backing the vsyscall

diff --git a/arch/x86/kernel/ldt.c b/arch/x86/kernel/ldt.c
@@ -669,6 +669,9 @@ SYSCALL_DEFINE3(modify_ldt, int , func , void __user * , ptr ,
 {
 	int ret = -ENOSYS;
 
+	if (cpu_feature_enabled(X86_FEATURE_KVM_PVM_GUEST))
+		return (unsigned int)ret;
+
 	switch (func) {
 	case 0:
 		ret = read_ldt(ptr, bytecount);

diff --git a/arch/x86/kernel/pvm.c b/arch/x86/kernel/pvm.c
@@ -457,6 +457,10 @@ void __init pvm_early_setup(void)
 	setup_force_cpu_cap(X86_FEATURE_KVM_PVM_GUEST);
 	setup_force_cpu_cap(X86_FEATURE_PV_GUEST);
 
+	/* Don't use SYSENTER (Intel) and SYSCALL32 (AMD) in vdso. */
+	setup_clear_cpu_cap(X86_FEATURE_SYSENTER32);
+	setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
+
 	/* PVM takes care of %gs when switching to usermode for us */
 	pv_ops.cpu.load_gs_index = pvm_load_gs_index;
 	pv_ops.cpu.cpuid = pvm_cpuid;