Re: [RFC PATCH 02/10] accel: Use qemu_security_policy_taint(), mark KVM and Xen as safe
It's been a while...
Daniel P. Berrangé writes:
> On Thu, Sep 09, 2021 at 01:20:16AM +0200, Philippe Mathieu-Daudé wrote:
>> Add the AccelClass::secure_policy_supported field to classify
>> safe (within security boundary) vs unsafe accelerators.
>>
>> Signed-off-by: Philippe Mathieu-Daudé
>> ---
>> include/qemu/accel.h | 5 +
>> accel/kvm/kvm-all.c | 1 +
>> accel/xen/xen-all.c | 1 +
>> softmmu/vl.c | 3 +++
>> 4 files changed, 10 insertions(+)
>>
>> diff --git a/include/qemu/accel.h b/include/qemu/accel.h
>> index 4f4c283f6fc..895e30be0de 100644
>> --- a/include/qemu/accel.h
>> +++ b/include/qemu/accel.h
>> @@ -44,6 +44,11 @@ typedef struct AccelClass {
>> hwaddr start_addr, hwaddr size);
>> #endif
>> bool *allowed;
>> +/*
>> + * Whether the accelerator is withing QEMU security policy boundary.
>> + * See: https://www.qemu.org/contribute/security-process/
>> + */
>> +bool secure_policy_supported;
>
> The security handling policy is a high level concept that is
> open to variation over time and also by downstream distro
> vendors.
Moreover, the concept of "tainting" isn't limited to "because
security".
> At a code level we should be dealing in a more fundamental
> concept. At an accelerator level we should really jsut
> declare whether or not the accelerator impl is considered
> to be secure against malicious guest code.
>
> eg
>
> /* Whether this accelerator is secure against execution
> * of malciious guest machine code */
> bool secure;
What I'd like to see is a separation of "assertions", like "not meant to
serve as security boundary", and policy. Yes, this is vague. Take it
as food for thought.
>> /*
>> * Array of global properties that would be applied when specific
>> * accelerator is chosen. It works like MachineClass.compat_props
>> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
>> index 0125c17edb8..eb6b9e44df2 100644
>> --- a/accel/kvm/kvm-all.c
>> +++ b/accel/kvm/kvm-all.c
>> @@ -3623,6 +3623,7 @@ static void kvm_accel_class_init(ObjectClass *oc, void
>> *data)
>> ac->init_machine = kvm_init;
>> ac->has_memory = kvm_accel_has_memory;
>> ac->allowed = &kvm_allowed;
>> +ac->secure_policy_supported = true;
>>
>> object_class_property_add(oc, "kernel-irqchip", "on|off|split",
>> NULL, kvm_set_kernel_irqchip,
>> diff --git a/accel/xen/xen-all.c b/accel/xen/xen-all.c
>> index 69aa7d018b2..57867af5faf 100644
>> --- a/accel/xen/xen-all.c
>> +++ b/accel/xen/xen-all.c
>> @@ -198,6 +198,7 @@ static void xen_accel_class_init(ObjectClass *oc, void
>> *data)
>> ac->setup_post = xen_setup_post;
>> ac->allowed = &xen_allowed;
>> ac->compat_props = g_ptr_array_new();
>> +ac->secure_policy_supported = true;
>>
>> compat_props_add(ac->compat_props, compat, G_N_ELEMENTS(compat));
>>
>> diff --git a/softmmu/vl.c b/softmmu/vl.c
>> index 92c05ac97ee..e4f94e159c3 100644
>> --- a/softmmu/vl.c
>> +++ b/softmmu/vl.c
>> @@ -2388,6 +2388,9 @@ static int do_configure_accelerator(void *opaque,
>> QemuOpts *opts, Error **errp)
>> return 0;
>> }
>>
>> +qemu_security_policy_taint(!ac->secure_policy_supported,
>> + "%s accelerator", acc);
>
> We need this information to be introspectable, becuase stuff printed
> to stderr is essentially opaque to libvirt and mgmt apps above.
>
> We don't have a convenient "query-accel" command but I think this
> could possibly fit into 'query-target'. ie the TargetInfo struct
> gain a field:
>
>
> ##
> # @TargetInfo:
> #
> # Information describing the QEMU target.
> #
> # @arch: the target architecture
> # @secure: Whether the currently active accelerator for this target
> # is secure against execution of malicous guest code
> #
> # Since: 1.2
> ##
> { 'struct': 'TargetInfo',
> 'data': { 'arch': 'SysEmuTarget',
> 'secure': 'bool'} }
My preferred means of introspection is QAPI schema introspection. For
QMP, that's query-qmp-schema. For CLI, it doesn't exist, yet.
If it did, then it would tell us that (QAPIfied) -accel takes an
argument @accel of a certain enumeration type. We could then tack
suitable feature flags to the enumeration type's values.
If we make the feature flags "special", i.e. known to QAPI, we can then
tie them to policy, like special feature flag 'deprecated' is tied to
policy configured with -compat deprecated-{input,output}=...
Alternatively, leave policy to the management application.
QAPI schema feature flags plus policy are is not a *complete* solution,
just like feature flag 'deprecated' and -compat are not a complete
solution for handling use of deprecated interfaces: we can and do
deprecate usage that isn't tied to a syntactic element in the QAPI
schema. Example: commit a9b305ba291 deprecated use of socket chardev
option wait together with server=true.
It is, however, a solution for
Re: [RFC PATCH 02/10] accel: Use qemu_security_policy_taint(), mark KVM and Xen as safe
On Thu, Sep 09, 2021 at 01:20:16AM +0200, Philippe Mathieu-Daudé wrote:
> Add the AccelClass::secure_policy_supported field to classify
> safe (within security boundary) vs unsafe accelerators.
>
> Signed-off-by: Philippe Mathieu-Daudé
> ---
> include/qemu/accel.h | 5 +
> accel/kvm/kvm-all.c | 1 +
> accel/xen/xen-all.c | 1 +
> softmmu/vl.c | 3 +++
> 4 files changed, 10 insertions(+)
>
> diff --git a/include/qemu/accel.h b/include/qemu/accel.h
> index 4f4c283f6fc..895e30be0de 100644
> --- a/include/qemu/accel.h
> +++ b/include/qemu/accel.h
> @@ -44,6 +44,11 @@ typedef struct AccelClass {
> hwaddr start_addr, hwaddr size);
> #endif
> bool *allowed;
> +/*
> + * Whether the accelerator is withing QEMU security policy boundary.
within
> + * See: https://www.qemu.org/contribute/security-process/
> + */
> +bool secure_policy_supported;
> /*
> * Array of global properties that would be applied when specific
> * accelerator is chosen. It works like MachineClass.compat_props
--
Eric Blake, Principal Software Engineer
Red Hat, Inc. +1-919-301-3266
Virtualization: qemu.org | libvirt.org
Re: [RFC PATCH 02/10] accel: Use qemu_security_policy_taint(), mark KVM and Xen as safe
On Thu, Sep 09, 2021 at 01:20:16AM +0200, Philippe Mathieu-Daudé wrote:
> Add the AccelClass::secure_policy_supported field to classify
> safe (within security boundary) vs unsafe accelerators.
>
> Signed-off-by: Philippe Mathieu-Daudé
> ---
> include/qemu/accel.h | 5 +
> accel/kvm/kvm-all.c | 1 +
> accel/xen/xen-all.c | 1 +
> softmmu/vl.c | 3 +++
> 4 files changed, 10 insertions(+)
>
> diff --git a/include/qemu/accel.h b/include/qemu/accel.h
> index 4f4c283f6fc..895e30be0de 100644
> --- a/include/qemu/accel.h
> +++ b/include/qemu/accel.h
> @@ -44,6 +44,11 @@ typedef struct AccelClass {
> hwaddr start_addr, hwaddr size);
> #endif
> bool *allowed;
> +/*
> + * Whether the accelerator is withing QEMU security policy boundary.
> + * See: https://www.qemu.org/contribute/security-process/
> + */
> +bool secure_policy_supported;
The security handling policy is a high level concept that is
open to variation over time and also by downstream distro
vendors.
At a code level we should be dealing in a more fundamental
concept. At an accelerator level we should really jsut
declare whether or not the accelerator impl is considered
to be secure against malicious guest code.
eg
/* Whether this accelerator is secure against execution
* of malciious guest machine code */
bool secure;
> /*
> * Array of global properties that would be applied when specific
> * accelerator is chosen. It works like MachineClass.compat_props
> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
> index 0125c17edb8..eb6b9e44df2 100644
> --- a/accel/kvm/kvm-all.c
> +++ b/accel/kvm/kvm-all.c
> @@ -3623,6 +3623,7 @@ static void kvm_accel_class_init(ObjectClass *oc, void
> *data)
> ac->init_machine = kvm_init;
> ac->has_memory = kvm_accel_has_memory;
> ac->allowed = &kvm_allowed;
> +ac->secure_policy_supported = true;
>
> object_class_property_add(oc, "kernel-irqchip", "on|off|split",
> NULL, kvm_set_kernel_irqchip,
> diff --git a/accel/xen/xen-all.c b/accel/xen/xen-all.c
> index 69aa7d018b2..57867af5faf 100644
> --- a/accel/xen/xen-all.c
> +++ b/accel/xen/xen-all.c
> @@ -198,6 +198,7 @@ static void xen_accel_class_init(ObjectClass *oc, void
> *data)
> ac->setup_post = xen_setup_post;
> ac->allowed = &xen_allowed;
> ac->compat_props = g_ptr_array_new();
> +ac->secure_policy_supported = true;
>
> compat_props_add(ac->compat_props, compat, G_N_ELEMENTS(compat));
>
> diff --git a/softmmu/vl.c b/softmmu/vl.c
> index 92c05ac97ee..e4f94e159c3 100644
> --- a/softmmu/vl.c
> +++ b/softmmu/vl.c
> @@ -2388,6 +2388,9 @@ static int do_configure_accelerator(void *opaque,
> QemuOpts *opts, Error **errp)
> return 0;
> }
>
> +qemu_security_policy_taint(!ac->secure_policy_supported,
> + "%s accelerator", acc);
We need this information to be introspectable, becuase stuff printed
to stderr is essentially opaque to libvirt and mgmt apps above.
We don't have a convenient "query-accel" command but I think this
could possibly fit into 'query-target'. ie the TargetInfo struct
gain a field:
##
# @TargetInfo:
#
# Information describing the QEMU target.
#
# @arch: the target architecture
# @secure: Whether the currently active accelerator for this target
# is secure against execution of malicous guest code
#
# Since: 1.2
##
{ 'struct': 'TargetInfo',
'data': { 'arch': 'SysEmuTarget',
'secure': 'bool'} }
Regards,
Daniel
--
|: https://berrange.com -o-https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org -o-https://fstop138.berrange.com :|
|: https://entangle-photo.org-o-https://www.instagram.com/dberrange :|
[RFC PATCH 02/10] accel: Use qemu_security_policy_taint(), mark KVM and Xen as safe
Add the AccelClass::secure_policy_supported field to classify
safe (within security boundary) vs unsafe accelerators.
Signed-off-by: Philippe Mathieu-Daudé
---
include/qemu/accel.h | 5 +
accel/kvm/kvm-all.c | 1 +
accel/xen/xen-all.c | 1 +
softmmu/vl.c | 3 +++
4 files changed, 10 insertions(+)
diff --git a/include/qemu/accel.h b/include/qemu/accel.h
index 4f4c283f6fc..895e30be0de 100644
--- a/include/qemu/accel.h
+++ b/include/qemu/accel.h
@@ -44,6 +44,11 @@ typedef struct AccelClass {
hwaddr start_addr, hwaddr size);
#endif
bool *allowed;
+/*
+ * Whether the accelerator is withing QEMU security policy boundary.
+ * See: https://www.qemu.org/contribute/security-process/
+ */
+bool secure_policy_supported;
/*
* Array of global properties that would be applied when specific
* accelerator is chosen. It works like MachineClass.compat_props
diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
index 0125c17edb8..eb6b9e44df2 100644
--- a/accel/kvm/kvm-all.c
+++ b/accel/kvm/kvm-all.c
@@ -3623,6 +3623,7 @@ static void kvm_accel_class_init(ObjectClass *oc, void
*data)
ac->init_machine = kvm_init;
ac->has_memory = kvm_accel_has_memory;
ac->allowed = &kvm_allowed;
+ac->secure_policy_supported = true;
object_class_property_add(oc, "kernel-irqchip", "on|off|split",
NULL, kvm_set_kernel_irqchip,
diff --git a/accel/xen/xen-all.c b/accel/xen/xen-all.c
index 69aa7d018b2..57867af5faf 100644
--- a/accel/xen/xen-all.c
+++ b/accel/xen/xen-all.c
@@ -198,6 +198,7 @@ static void xen_accel_class_init(ObjectClass *oc, void
*data)
ac->setup_post = xen_setup_post;
ac->allowed = &xen_allowed;
ac->compat_props = g_ptr_array_new();
+ac->secure_policy_supported = true;
compat_props_add(ac->compat_props, compat, G_N_ELEMENTS(compat));
diff --git a/softmmu/vl.c b/softmmu/vl.c
index 92c05ac97ee..e4f94e159c3 100644
--- a/softmmu/vl.c
+++ b/softmmu/vl.c
@@ -2388,6 +2388,9 @@ static int do_configure_accelerator(void *opaque,
QemuOpts *opts, Error **errp)
return 0;
}
+qemu_security_policy_taint(!ac->secure_policy_supported,
+ "%s accelerator", acc);
+
return 1;
}
--
2.31.1
