On Tue, 8 Mar 2022 15:12:45 +0100 Ilya Maximets wrote:
> >> diff --git a/include/uapi/linux/openvswitch.h b/include/uapi/linux/openvswitch.h
> >> index 9d1710f20505..ab6755621e02 100644
> >> --- a/include/uapi/linux/openvswitch.h
> >> +++ b/include/uapi/linux/openvswitch.h
> >> @@ -351,11 +351,16 @@ enum ovs_key_attr {
> >> OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4, /* struct ovs_key_ct_tuple_ipv4 */
> >> OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6, /* struct ovs_key_ct_tuple_ipv6 */
> >> OVS_KEY_ATTR_NSH, /* Nested set of ovs_nsh_key_* */
> >> - OVS_KEY_ATTR_IPV6_EXTHDRS, /* struct ovs_key_ipv6_exthdr */
> >>
> >> #ifdef __KERNEL__
> >> OVS_KEY_ATTR_TUNNEL_INFO, /* struct ip_tunnel_info */
> >> #endif
> >> + /* User space decided to squat on types 30 and 31 */
> >> + OVS_KEY_ATTR_IPV6_EXTHDRS = 32, /* struct ovs_key_ipv6_exthdr */
> >> + /* WARNING: <scary warning to avoid the problem coming back> */
>
> Yes, that is something that I had in mind too. The only thing that makes
> me uncomfortable is OVS_KEY_ATTR_TUNNEL_INFO = 30 here. Even though it
> doesn't make a lot of difference, I'd better keep the kernel-only attributes
> at the end of the enumeration. Is there a better way to handle kernel-only
> attribute?
My thought was to leave the kernel/userspace only types "behind" to
avoid perpetuating the same constructs.
Johannes's point about userspace to userspace messages makes the whole
thing a little less of an aberration.
Is there a reason for the types to be hidden under __KERNEL__?
Or someone did that in a misguided attempt to save space in attr arrays
when parsing?
> Also, the OVS_KEY_ATTR_ND_EXTENSIONS (31) attribute used to store IPv6 Neighbor
> Discovery extensions is currently implemented only for userspace, but nothing
> actually prevents us having the kernel implementation. So, we need a way to
> make it usable by the kernel in the future.
The "= 32" leaves the earlier attr types as reserved so nothing
prevents us from defining them later. But..
> > It might be nicer to actually document here in what's at least supposed
> > to be the canonical documentation of the API what those types were used
> > for.
>
> I agree with that.
Should we add the user space types to the kernel header and remove the
ifdef __KERNEL__ around TUNNEL_INFO, then?
On 3/8/22 17:17, Jakub Kicinski wrote:
> On Tue, 8 Mar 2022 15:12:45 +0100 Ilya Maximets wrote:
>>>> diff --git a/include/uapi/linux/openvswitch.h b/include/uapi/linux/openvswitch.h
>>>> index 9d1710f20505..ab6755621e02 100644
>>>> --- a/include/uapi/linux/openvswitch.h
>>>> +++ b/include/uapi/linux/openvswitch.h
>>>> @@ -351,11 +351,16 @@ enum ovs_key_attr {
>>>> OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4, /* struct ovs_key_ct_tuple_ipv4 */
>>>> OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6, /* struct ovs_key_ct_tuple_ipv6 */
>>>> OVS_KEY_ATTR_NSH, /* Nested set of ovs_nsh_key_* */
>>>> - OVS_KEY_ATTR_IPV6_EXTHDRS, /* struct ovs_key_ipv6_exthdr */
>>>>
>>>> #ifdef __KERNEL__
>>>> OVS_KEY_ATTR_TUNNEL_INFO, /* struct ip_tunnel_info */
>>>> #endif
>>>> + /* User space decided to squat on types 30 and 31 */
>>>> + OVS_KEY_ATTR_IPV6_EXTHDRS = 32, /* struct ovs_key_ipv6_exthdr */
>>>> + /* WARNING: <scary warning to avoid the problem coming back> */
>>
>> Yes, that is something that I had in mind too. The only thing that makes
>> me uncomfortable is OVS_KEY_ATTR_TUNNEL_INFO = 30 here. Even though it
>> doesn't make a lot of difference, I'd better keep the kernel-only attributes
>> at the end of the enumeration. Is there a better way to handle kernel-only
>> attribute?
>
> My thought was to leave the kernel/userspace only types "behind" to
> avoid perpetuating the same constructs.
>
> Johannes's point about userspace to userspace messages makes the whole
> thing a little less of an aberration.
>
> Is there a reason for the types to be hidden under __KERNEL__?
> Or someone did that in a misguided attempt to save space in attr arrays
> when parsing?
My impression is that OVS_KEY_ATTR_TUNNEL_INFO was hidden from the
user space just because it's not supposed to ever be used by the
user space application. Pravin or Jesse should know for sure.
>
>> Also, the OVS_KEY_ATTR_ND_EXTENSIONS (31) attribute used to store IPv6 Neighbor
>> Discovery extensions is currently implemented only for userspace, but nothing
>> actually prevents us having the kernel implementation. So, we need a way to
>> make it usable by the kernel in the future.
>
> The "= 32" leaves the earlier attr types as reserved so nothing
> prevents us from defining them later. But..
>
>>> It might be nicer to actually document here in what's at least supposed
>>> to be the canonical documentation of the API what those types were used
>>> for.
>>
>> I agree with that.
>
> Should we add the user space types to the kernel header and remove the
> ifdef __KERNEL__ around TUNNEL_INFO, then?
I don't think we need to actually define them, but we may list them
in the comment. I'm OK with either option though.
For the removal of ifdef __KERNEL__, that might be a good thing to do.
I'm just not sure what are the best practices here.
We'll need to make some code changes in user space to avoid warnings
about not all the enum members being used in 'switch'es. But that's
not a problem.
If you think that having a flat enum without 'ifdef's is a viable
option from a kernel's point of view, I'm all for it.
Maybe something like this (only checked that this compiles; 29 and
30 are correct numbers of these userspace attributes):
---
diff --git a/include/uapi/linux/openvswitch.h b/include/uapi/linux/openvswitch.h
index 9d1710f20505..86bc951be5bc 100644
--- a/include/uapi/linux/openvswitch.h
+++ b/include/uapi/linux/openvswitch.h
@@ -351,11 +351,19 @@ enum ovs_key_attr {
OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4, /* struct ovs_key_ct_tuple_ipv4 */
OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6, /* struct ovs_key_ct_tuple_ipv6 */
OVS_KEY_ATTR_NSH, /* Nested set of ovs_nsh_key_* */
- OVS_KEY_ATTR_IPV6_EXTHDRS, /* struct ovs_key_ipv6_exthdr */
-#ifdef __KERNEL__
- OVS_KEY_ATTR_TUNNEL_INFO, /* struct ip_tunnel_info */
-#endif
+ /* User space decided to squat on types 29 and 30. They are listed
+ * below, but should not be sent to the kernel:
+ *
+ * OVS_KEY_ATTR_PACKET_TYPE, be32 packet type
+ * OVS_KEY_ATTR_ND_EXTENSIONS, IPv6 Neighbor Discovery extensions
+ *
+ * WARNING: No new types should be added unless they are defined
+ * for both kernel and user space (no 'ifdef's). It's hard
+ * to keep compatibility otherwise. */
+ OVS_KEY_ATTR_TUNNEL_INFO = 31, /* struct ip_tunnel_info.
+ For in-kernel use only. */
+ OVS_KEY_ATTR_IPV6_EXTHDRS, /* struct ovs_key_ipv6_exthdr */
__OVS_KEY_ATTR_MAX
};
diff --git a/net/openvswitch/flow_netlink.c b/net/openvswitch/flow_netlink.c
index 8b4124820f7d..315064bada3e 100644
--- a/net/openvswitch/flow_netlink.c
+++ b/net/openvswitch/flow_netlink.c
@@ -346,7 +346,7 @@ size_t ovs_key_attr_size(void)
/* Whenever adding new OVS_KEY_ FIELDS, we should consider
* updating this function.
*/
- BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 30);
+ BUILD_BUG_ON(OVS_KEY_ATTR_MAX != 32);
return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
+ nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
---
Thoughts?
The same change can be ported to the user-space header, but with
types actually defined and not part of the comment. It may look
like this: https://pastebin.com/k8UWEZtR (without IPV6_EXTHDRS yet).
For the future, we'll try to find a way to define them in a separate
enum or will define them dynamically based on the policy dumped from
the currently running kernel. In any case no new userspace-only types
should be defined in that enum.
Best regards, Ilya Maximets.