Re: [PATCH v4 4/4] vduse: Add LSM hooks to check Virtio device type

[Maxime Coquelin <maxime.coquelin@redhat.com>]

Hello Paul,

On 11/8/23 03:31, Paul Moore wrote:
> On Oct 20, 2023 "Michael S. Tsirkin" <mst@redhat.com> wrote:
>>
>> This patch introduces LSM hooks for devices creation,
>> destruction and opening operations, checking the
>> application is allowed to perform these operations for
>> the Virtio device type.
>>
>> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
>> ---
>>   drivers/vdpa/vdpa_user/vduse_dev.c  | 12 +++++++
>>   include/linux/lsm_hook_defs.h       |  4 +++
>>   include/linux/security.h            | 15 ++++++++
>>   security/security.c                 | 42 ++++++++++++++++++++++
>>   security/selinux/hooks.c            | 55 +++++++++++++++++++++++++++++
>>   security/selinux/include/classmap.h |  2 ++
>>   6 files changed, 130 insertions(+)
> 
> My apologies for the late reply, I've been trying to work my way through
> the review backlog but it has been taking longer than expected; comments
> below ...

No worries, I have also been busy these days.

>> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
>> index 2aa0e219d721..65d9262a37f7 100644
>> --- a/security/selinux/hooks.c
>> +++ b/security/selinux/hooks.c
>> @@ -21,6 +21,7 @@
>>    *  Copyright (C) 2016 Mellanox Technologies
>>    */
>>   
>> +#include "av_permissions.h"
>>   #include <linux/init.h>
>>   #include <linux/kd.h>
>>   #include <linux/kernel.h>
>> @@ -92,6 +93,7 @@
>>   #include <linux/fsnotify.h>
>>   #include <linux/fanotify.h>
>>   #include <linux/io_uring.h>
>> +#include <uapi/linux/virtio_ids.h>
>>   
>>   #include "avc.h"
>>   #include "objsec.h"
>> @@ -6950,6 +6952,56 @@ static int selinux_uring_cmd(struct io_uring_cmd *ioucmd)
>>   }
>>   #endif /* CONFIG_IO_URING */
>>   
>> +static int vduse_check_device_type(u32 sid, u32 device_id)
>> +{
>> +	u32 requested;
>> +
>> +	if (device_id == VIRTIO_ID_NET)
>> +		requested = VDUSE__NET;
>> +	else if (device_id == VIRTIO_ID_BLOCK)
>> +		requested = VDUSE__BLOCK;
>> +	else
>> +		return -EINVAL;
>> +
>> +	return avc_has_perm(sid, sid, SECCLASS_VDUSE, requested, NULL);
>> +}
>> +
>> +static int selinux_vduse_dev_create(u32 device_id)
>> +{
>> +	u32 sid = current_sid();
>> +	int ret;
>> +
>> +	ret = avc_has_perm(sid, sid, SECCLASS_VDUSE, VDUSE__DEVCREATE, NULL);
>> +	if (ret)
>> +		return ret;
>> +
>> +	return vduse_check_device_type(sid, device_id);
>> +}
> 
> I see there has been some discussion about the need for a dedicated
> create hook as opposed to using the existing ioctl controls.  I think
> one important point that has been missing from the discussion is the
> idea of labeling the newly created device.  Unfortunately prior to a
> few minutes ago I hadn't ever looked at VDUSE so please correct me if
> I get some things wrong :)
> 
>  From what I can see userspace creates a new VDUSE device with
> ioctl(VDUSE_CREATE_DEV), which trigger the creation of a new
> /dev/vduse/XXX device which will be labeled according to the udev
> and SELinux configuration, likely with a generic udev label.  My
> question is if we want to be able to uniquely label each VDUSE
> device based on the process that initiates the device creation
> with the call to ioctl()?  If that is the case, we would need a
> create hook not only to control the creation of the device, but to
> record the triggering process' label in the new device; this label
> would then be used in subsequent VDUSE open and destroy operations.
> The normal device file I/O operations would still be subject to the
> standard SELinux file I/O permissions using the device file label
> assigned by systemd/udev when the device was created.

I don't think we need a unique label for VDUSE devices, but maybe
Michael thinks otherwise?

> 
>> +static int selinux_vduse_dev_destroy(u32 device_id)
>> +{
>> +	u32 sid = current_sid();
>> +	int ret;
>> +
>> +	ret = avc_has_perm(sid, sid, SECCLASS_VDUSE, VDUSE__DEVDESTROY, NULL);
>> +	if (ret)
>> +		return ret;
>> +
>> +	return vduse_check_device_type(sid, device_id);
>> +}
>> +
>> +static int selinux_vduse_dev_open(u32 device_id)
>> +{
>> +	u32 sid = current_sid();
>> +	int ret;
>> +
>> +	ret = avc_has_perm(sid, sid, SECCLASS_VDUSE, VDUSE__DEVOPEN, NULL);
>> +	if (ret)
>> +		return ret;
>> +
>> +	return vduse_check_device_type(sid, device_id);
>> +}
>> +
>>   /*
>>    * IMPORTANT NOTE: When adding new hooks, please be careful to keep this order:
>>    * 1. any hooks that don't belong to (2.) or (3.) below,
>> @@ -7243,6 +7295,9 @@ static struct security_hook_list selinux_hooks[] __ro_after_init = {
>>   #ifdef CONFIG_PERF_EVENTS
>>   	LSM_HOOK_INIT(perf_event_alloc, selinux_perf_event_alloc),
>>   #endif
>> +	LSM_HOOK_INIT(vduse_dev_create, selinux_vduse_dev_create),
>> +	LSM_HOOK_INIT(vduse_dev_destroy, selinux_vduse_dev_destroy),
>> +	LSM_HOOK_INIT(vduse_dev_open, selinux_vduse_dev_open),
>>   };
>>   
>>   static __init int selinux_init(void)
>> diff --git a/security/selinux/include/classmap.h b/security/selinux/include/classmap.h
>> index a3c380775d41..d3dc37fb03d4 100644
>> --- a/security/selinux/include/classmap.h
>> +++ b/security/selinux/include/classmap.h
>> @@ -256,6 +256,8 @@ const struct security_class_mapping secclass_map[] = {
>>   	  { "override_creds", "sqpoll", "cmd", NULL } },
>>   	{ "user_namespace",
>>   	  { "create", NULL } },
>> +	{ "vduse",
>> +	  { "devcreate", "devdestroy", "devopen", "net", "block", NULL} },
> 
> I think we can just call the permissions "create", "open", and "destroy"
> since the "dev" prefix is somewhat implied by this being a dedicated
> VDUSE object class.

Ack, I can remove the "dev" prefix in next revision.

> 
> I don't see where you are using the "net" and "block" permissions above,
> is this a leftover from a prior draft of this patch or are you planning
> to do something with these permissions?

It is actually used, but maybe not in a correct way.
If you look at each hook, there are two checks performed:
1. Check for the operation type: create/destroy/open
2. Check for the device type: block/net

It means that the application will have to combine one (or more)
operation type with one (or more) device type.

Does that make sense?

Thanks,
Maxime

> 
>>   	{ NULL }
>>     };
>>   
>> -- 
>> 2.41.0
> 
> --
> paul-moore.com
>