Re: [PATCH] bridge: extend BR_ISOLATE to full split-horizon

[Alexandra Winter <wintera@linux.ibm.com>]

On 09.12.21 17:08, Nikolay Aleksandrov wrote:
> On 09/12/2021 17:42, Jakub Kicinski wrote:
>> On Thu,  9 Dec 2021 13:14:32 +0100 David Lamparter wrote:
>>> Split-horizon essentially just means being able to create multiple
>>> groups of isolated ports that are isolated within the group, but not
>>> with respect to each other.
>>>
>>> The intent is very different, while isolation is a policy feature,
>>> split-horizon is intended to provide functional "multiple member ports
>>> are treated as one for loop avoidance."  But it boils down to the same
>>> thing in the end.
>>>
>>> Signed-off-by: David Lamparter <equinox@diac24.net>
>>> Cc: Nikolay Aleksandrov <nikolay@nvidia.com>
>>> Cc: Alexandra Winter <wintera@linux.ibm.com>
>>
>> Does not apply to net-next, you'll need to repost even if the code is
>> good. Please put [PATCH net-next] in the subject.
>>
> 
> Hi,
> For some reason this patch didn't make it to my inbox.. Anyway I was
> able to see it now online, a few comments (sorry can't do them inline due
> to missing mbox patch):
> - please drop the sysfs part, we're not extending sysfs anymore
> - split the bridge change from the driver
> - drop the /* BR_ISOLATED - previously BIT(16) */ comment
> - [IFLA_BRPORT_HORIZON_GROUP] = NLA_POLICY_MIN(NLA_S32, 0), why not just { .type = NLA_U32 } ?
> - just forbid having both set (tb[IFLA_BRPORT_ISOLATED] && tb[IFLA_BRPORT_HORIZON_GROUP])
>   user-space should use just one of the two, if isolated is set then it overwrites any older
>   IFLA_BRPORT_HORIZON_GROUP settings, that should simplify things considerably
Yes, please keep it compatible with userspace setting IFLA_BRPORT_ISOLATED only.
> 
> Why the limitation (UAPI limited to positive signed int. (recommended ifindex namespace)) ?
> You have the full unsigned space available, user-space can use it as it sees fit.
> You can just remove the comment about recommended ifindex.
> 
> Also please extend the port isolation self-test with a test for a different horizon group.
> 
> Thanks,
>  Nik
> 
> 
> 
>