From b74558259c5149e5edd79348b70eb34177cbeea0 Mon Sep 17 00:00:00 2001 From: Wanpeng Li Date: Wed, 22 Nov 2017 14:04:00 -0800 Subject: [PATCH] KVM: VMX: Fix vmx->nested freeing when no SMI handler MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Reported by syzkaller: ------------[ cut here ]------------ WARNING: CPU: 5 PID: 2939 at arch/x86/kvm/vmx.c:3844 free_loaded_vmcs+0x77/0x80 [kvm_intel] CPU: 5 PID: 2939 Comm: repro Not tainted 4.14.0+ #26 RIP: 0010:free_loaded_vmcs+0x77/0x80 [kvm_intel] Call Trace: vmx_free_vcpu+0xda/0x130 [kvm_intel] kvm_arch_destroy_vm+0x192/0x290 [kvm] kvm_put_kvm+0x262/0x560 [kvm] kvm_vm_release+0x2c/0x30 [kvm] __fput+0x190/0x370 task_work_run+0xa1/0xd0 do_exit+0x4d2/0x13e0 do_group_exit+0x89/0x140 get_signal+0x318/0xb80 do_signal+0x8c/0xb40 exit_to_usermode_loop+0xe4/0x140 syscall_return_slowpath+0x206/0x230 entry_SYSCALL_64_fastpath+0x98/0x9a The syzkaller testcase will execute VMXON/VMLAUCH instructions, so the vmx->nested stuff is populated, it will also issue KVM_SMI ioctl. However, the testcase is just a simple c program and not be lauched by something like seabios which implements smi_handler. Commit 05cade71cf (KVM: nSVM: fix SMI injection in guest mode) gets out of guest mode and set nested.vmxon to false for the duration of SMM according to SDM 34.14.1 "leave VMX operation" upon entering SMM. We can't alloc/free the vmx->nested stuff each time when entering/exiting SMM since it will induce more overhead. So the function vmx_pre_enter_smm() marks nested.vmxon false even if vmx->nested stuff is still populated. What it expected is em_rsm() can mark nested.vmxon to be true again. However, the smi_handler/rsm will not execute since there is no something like seabios in this scenario. The function free_nested() fails to free the vmx->nested stuff since the vmx->nested.vmxon is false which results in the above warning. This patch fixes it by also considering the no SMI handler case, luckily vmx->nested.smm.vmxon is marked according to the value of vmx->nested.vmxon in vmx_pre_enter_smm(), we can take advantage of it and free vmx->nested stuff when L1 goes down. Reported-by: Dmitry Vyukov Cc: Paolo Bonzini Cc: Radim Krčmář Cc: Dmitry Vyukov Reviewed-by: Liran Alon Fixes: 05cade71cf (KVM: nSVM: fix SMI injection in guest mode) Signed-off-by: Wanpeng Li Signed-off-by: Paolo Bonzini --- arch/x86/kvm/vmx.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c index 04f19b099617..4704aaf6d19e 100644 --- a/arch/x86/kvm/vmx.c +++ b/arch/x86/kvm/vmx.c @@ -7415,10 +7415,11 @@ static inline void nested_release_vmcs12(struct vcpu_vmx *vmx) */ static void free_nested(struct vcpu_vmx *vmx) { - if (!vmx->nested.vmxon) + if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon) return; vmx->nested.vmxon = false; + vmx->nested.smm.vmxon = false; free_vpid(vmx->nested.vpid02); vmx->nested.posted_intr_nv = -1; vmx->nested.current_vmptr = -1ull;