[PATCH] KVM: X86: Fix residual mmio emulation request to userspace

[Wanpeng Li]

Reported by syzkaller:

The kvm-intel.unrestricted_guest=0

   WARNING: CPU: 5 PID: 1014 at /home/kernel/data/kvm/arch/x86/kvm//x86.c:7227 kvm_arch_vcpu_ioctl_run+0x38b/0x1be0 [kvm]
   CPU: 5 PID: 1014 Comm: warn_test Tainted: G        W  OE   4.13.0-rc3+ #8
   RIP: 0010:kvm_arch_vcpu_ioctl_run+0x38b/0x1be0 [kvm]
   Call Trace:
    ? put_pid+0x3a/0x50
    ? rcu_read_lock_sched_held+0x79/0x80
    ? kmem_cache_free+0x2f2/0x350
    kvm_vcpu_ioctl+0x340/0x700 [kvm]
    ? kvm_vcpu_ioctl+0x340/0x700 [kvm]
    ? __fget+0xfc/0x210
    do_vfs_ioctl+0xa4/0x6a0
    ? __fget+0x11d/0x210
    SyS_ioctl+0x79/0x90
    entry_SYSCALL_64_fastpath+0x23/0xc2
    ? __this_cpu_preempt_check+0x13/0x20

The syszkaller folks reported a residual mmio emulation request to userspace 
due to vm86 fails to emulate inject real mode interrupt(fails to read CS) and 
incurs a triple fault. The vCPU returns to userspace with vcpu->mmio_needed == true 
and KVM_EXIT_SHUTDOWN exit reason. However, the syszkaller testcase constructs 
several threads to launch the same vCPU, the thread which lauch this vCPU after 
the thread whichs get the vcpu->mmio_needed == true and KVM_EXIT_SHUTDOWN will 
trigger the warning.

   #define _GNU_SOURCE
   #include <pthread.h>
   #include <stdio.h>
   #include <stdlib.h>
   #include <string.h>
   #include <sys/wait.h>
   #include <sys/types.h>
   #include <sys/stat.h>
   #include <sys/mman.h>
   #include <fcntl.h>
   #include <unistd.h>
   #include <linux/kvm.h>
   #include <stdio.h>
   
   int kvmcpu;
   struct kvm_run *run;
   
   void* thr(void* arg)
   {
     int res;
     res = ioctl(kvmcpu, KVM_RUN, 0);
     printf("ret1=%d exit_reason=%d suberror=%d\n",
         res, run->exit_reason, run->internal.suberror);
     return 0;
   }
   
   void test()
   {
     int i, kvm, kvmvm;
     pthread_t th[4];
   
     kvm = open("/dev/kvm", O_RDWR);
     kvmvm = ioctl(kvm, KVM_CREATE_VM, 0);
     kvmcpu = ioctl(kvmvm, KVM_CREATE_VCPU, 0);
     run = (struct kvm_run*)mmap(0, 4096, PROT_READ|PROT_WRITE, MAP_SHARED, kvmcpu, 0);
     srand(getpid());
     for (i = 0; i < 4; i++) {
       pthread_create(&th[i], 0, thr, 0);
       usleep(rand() % 10000);
     }
     for (i = 0; i < 4; i++)
       pthread_join(th[i], 0);
   }
   
   int main()
   {
     for (;;) {
       int pid = fork();
       if (pid < 0)
         exit(1);
       if (pid == 0) {
         test();
         exit(0);
       }
       int status;
       while (waitpid(pid, &status, __WALL) != pid) {}
     }
     return 0;
   }

This patch fixes it by resetting the vcpu->mmio_needed once we receive 
the triple fault to avoid the residue.

Reported-by: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
Cc: Radim KrÄmÃÅ <rkrcmar@xxxxxxxxxx>
Cc: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
Signed-off-by: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
---
 arch/x86/kvm/vmx.c | 1 +
 arch/x86/kvm/x86.c | 1 +
 2 files changed, 2 insertions(+)

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 8e4a2dc..77ab10b 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -5864,6 +5864,7 @@ static int handle_external_interrupt(struct kvm_vcpu *vcpu)
 static int handle_triple_fault(struct kvm_vcpu *vcpu)
 {
 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
+	vcpu->mmio_needed = 0;
 	return 0;
 }
 
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 72d82ab..1e143f7 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -6776,6 +6776,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
 		}
 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
+			vcpu->mmio_needed = 0;
 			r = 0;
 			goto out;
 		}
-- 
2.7.4