KVM: SEV: Do COPY_ENC_CONTEXT_FROM with both VMs locked
authorPaolo Bonzini <pbonzini@redhat.com>
Tue, 23 Nov 2021 00:50:33 +0000 (19:50 -0500)
committerPaolo Bonzini <pbonzini@redhat.com>
Tue, 30 Nov 2021 08:54:13 +0000 (03:54 -0500)
Now that we have a facility to lock two VMs with deadlock
protection, use it for the creation of mirror VMs as well.  One of
COPY_ENC_CONTEXT_FROM(dst, src) and COPY_ENC_CONTEXT_FROM(src, dst)
would always fail, so the combination is nonsensical and it is okay to
return -EBUSY if it is attempted.

This sidesteps the question of what happens if a VM is
MOVE_ENC_CONTEXT_FROM'd at the same time as it is
COPY_ENC_CONTEXT_FROM'd: the locking prevents that from
happening.

Cc: Peter Gonda <pgonda@google.com>
Cc: Sean Christopherson <seanjc@google.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211123005036.2954379-10-pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/svm/sev.c

index 74b6459..025d973 100644 (file)
@@ -1955,77 +1955,59 @@ int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd)
 {
        struct file *source_kvm_file;
        struct kvm *source_kvm;
-       struct kvm_sev_info source_sev, *mirror_sev;
+       struct kvm_sev_info *source_sev, *mirror_sev;
        int ret;
 
        source_kvm_file = fget(source_fd);
        if (!file_is_kvm(source_kvm_file)) {
                ret = -EBADF;
-               goto e_source_put;
+               goto e_source_fput;
        }
 
        source_kvm = source_kvm_file->private_data;
-       mutex_lock(&source_kvm->lock);
-
-       if (!sev_guest(source_kvm)) {
-               ret = -EINVAL;
-               goto e_source_unlock;
-       }
+       ret = sev_lock_two_vms(kvm, source_kvm);
+       if (ret)
+               goto e_source_fput;
 
-       /* Mirrors of mirrors should work, but let's not get silly */
-       if (is_mirroring_enc_context(source_kvm) || source_kvm == kvm) {
+       /*
+        * Mirrors of mirrors should work, but let's not get silly.  Also
+        * disallow out-of-band SEV/SEV-ES init if the target is already an
+        * SEV guest, or if vCPUs have been created.  KVM relies on vCPUs being
+        * created after SEV/SEV-ES initialization, e.g. to init intercepts.
+        */
+       if (sev_guest(kvm) || !sev_guest(source_kvm) ||
+           is_mirroring_enc_context(source_kvm) || kvm->created_vcpus) {
                ret = -EINVAL;
-               goto e_source_unlock;
+               goto e_unlock;
        }
 
-       memcpy(&source_sev, &to_kvm_svm(source_kvm)->sev_info,
-              sizeof(source_sev));
-
        /*
         * The mirror kvm holds an enc_context_owner ref so its asid can't
         * disappear until we're done with it
         */
+       source_sev = &to_kvm_svm(source_kvm)->sev_info;
        kvm_get_kvm(source_kvm);
 
-       fput(source_kvm_file);
-       mutex_unlock(&source_kvm->lock);
-       mutex_lock(&kvm->lock);
-
-       /*
-        * Disallow out-of-band SEV/SEV-ES init if the target is already an
-        * SEV guest, or if vCPUs have been created.  KVM relies on vCPUs being
-        * created after SEV/SEV-ES initialization, e.g. to init intercepts.
-        */
-       if (sev_guest(kvm) || kvm->created_vcpus) {
-               ret = -EINVAL;
-               goto e_mirror_unlock;
-       }
-
        /* Set enc_context_owner and copy its encryption context over */
        mirror_sev = &to_kvm_svm(kvm)->sev_info;
        mirror_sev->enc_context_owner = source_kvm;
        mirror_sev->active = true;
-       mirror_sev->asid = source_sev.asid;
-       mirror_sev->fd = source_sev.fd;
-       mirror_sev->es_active = source_sev.es_active;
-       mirror_sev->handle = source_sev.handle;
+       mirror_sev->asid = source_sev->asid;
+       mirror_sev->fd = source_sev->fd;
+       mirror_sev->es_active = source_sev->es_active;
+       mirror_sev->handle = source_sev->handle;
        INIT_LIST_HEAD(&mirror_sev->regions_list);
+       ret = 0;
+
        /*
         * Do not copy ap_jump_table. Since the mirror does not share the same
         * KVM contexts as the original, and they may have different
         * memory-views.
         */
 
-       mutex_unlock(&kvm->lock);
-       return 0;
-
-e_mirror_unlock:
-       mutex_unlock(&kvm->lock);
-       kvm_put_kvm(source_kvm);
-       return ret;
-e_source_unlock:
-       mutex_unlock(&source_kvm->lock);
-e_source_put:
+e_unlock:
+       sev_unlock_two_vms(kvm, source_kvm);
+e_source_fput:
        if (source_kvm_file)
                fput(source_kvm_file);
        return ret;