Linux Ethernet Bridging - Jun 2022 - [Bridge] [PATCH V3 net-next 1/4] net: bridge: add fdb flag to extent locked port feature

On Thu, Jun 02, 2022 at 01:30:06PM +0300, Nikolay Aleksandrov
wrote:
> On 02/06/2022 13:17, Hans Schultz wrote:
> > On tor, jun 02, 2022 at 12:33, Nikolay Aleksandrov <razor at
blackwall.org> wrote:
> >> On 02/06/2022 12:17, Hans Schultz wrote:
> >>> On tis, maj 31, 2022 at 17:23, Ido Schimmel <idosch at
nvidia.com> wrote:
> >>>> On Tue, May 31, 2022 at 11:34:21AM +0200, Hans Schultz
wrote:
> > 
> >>> Another issue is that
> >>> bridge fdb add MAC dev DEV master static
> >>> seems to add the entry with the SELF flag set, which I
don't think is
> >>> what we would want it to do or?
> >>
> >> I don't see such thing (hacked iproute2 to print the flags
before cmd):
> >> $ bridge fdb add 00:11:22:33:44:55 dev vnet110 master static
> >> flags 0x4
> >>
> >> 0x4 = NTF_MASTER only
> >>
> > 
> > I also get 0x4 from iproute2, but I still get SELF entries when I look
> > with:
> > bridge fdb show dev DEV
> > 
> 
> after the above add:
> $ bridge fdb show dev vnet110 | grep 00:11
> 00:11:22:33:44:55 master virbr0 static
I think Hans is testing with mv88e6xxx which dumps entries directly from
HW via ndo_fdb_dump(). See dsa_slave_port_fdb_do_dump() which sets
NTF_SELF.

Hans, are you seeing the entry twice? Once with 'master' and once with
'self'?
> 
> >>> Also the replace command is not really supported properly as
it is. I
> >>> have made a fix for that which looks something like this:
> >>>
> >>> diff --git a/net/bridge/br_fdb.c b/net/bridge/br_fdb.c
> >>> index 6cbb27e3b976..f43aa204f375 100644
> >>> --- a/net/bridge/br_fdb.c
> >>> +++ b/net/bridge/br_fdb.c
> >>> @@ -917,6 +917,9 @@ static int fdb_add_entry(struct net_bridge
*br, struct net_bridge_port *source,
> >>>                 if (flags & NLM_F_EXCL)
> >>>                         return -EEXIST;
> >>>  
> >>> +               if (flags & NLM_F_REPLACE)
> >>> +                       modified = true;
> >>> +
> >>>                 if (READ_ONCE(fdb->dst) != source) {
> >>>                         WRITE_ONCE(fdb->dst, source);
> >>>                         modified = true;
> >>>
> >>> The argument for always sending notifications to the driver in
the case
> >>> of replace is that a replace command will refresh the entries
timeout if
> >>> the entry is the same. Any thoughts on this?
> >>
> >> I don't think so. It always updates its "used"
timer, not its "updated" timer which is the one
> >> for expire. A replace that doesn't actually change anything on
the entry shouldn't generate
> >> a notification.
> > 
> > Okay, so then there is missing checks on flags as the issue arose from
> > replacing locked entries with dynamic entries. I will do another fix
> > based on flags as modified needs to be true for the driver to get
notified.
>

On tor, jun 02, 2022 at 13:39, Ido Schimmel <idosch at nvidia.com>
wrote:
> On Thu, Jun 02, 2022 at 01:30:06PM +0300, Nikolay Aleksandrov wrote:
>> On 02/06/2022 13:17, Hans Schultz wrote:
>> > On tor, jun 02, 2022 at 12:33, Nikolay Aleksandrov <razor at
blackwall.org> wrote:
>> >> On 02/06/2022 12:17, Hans Schultz wrote:
>> >>> On tis, maj 31, 2022 at 17:23, Ido Schimmel <idosch at
nvidia.com> wrote:
>> >>>> On Tue, May 31, 2022 at 11:34:21AM +0200, Hans Schultz
wrote:
>> > 
>> >>> Another issue is that
>> >>> bridge fdb add MAC dev DEV master static
>> >>> seems to add the entry with the SELF flag set, which I
don't think is
>> >>> what we would want it to do or?
>> >>
>> >> I don't see such thing (hacked iproute2 to print the flags
before cmd):
>> >> $ bridge fdb add 00:11:22:33:44:55 dev vnet110 master static
>> >> flags 0x4
>> >>
>> >> 0x4 = NTF_MASTER only
>> >>
>> > 
>> > I also get 0x4 from iproute2, but I still get SELF entries when I
look
>> > with:
>> > bridge fdb show dev DEV
>> > 
>> 
>> after the above add:
>> $ bridge fdb show dev vnet110 | grep 00:11
>> 00:11:22:33:44:55 master virbr0 static
>
> I think Hans is testing with mv88e6xxx which dumps entries directly from
> HW via ndo_fdb_dump(). See dsa_slave_port_fdb_do_dump() which sets
> NTF_SELF.
>
> Hans, are you seeing the entry twice? Once with 'master' and once
with
> 'self'?
>
Well yes, but I get some additional entries with 'self' for different
vlans. So from clean adding a random fdb entry I get 4 entries on the
port, 2 with 'master' and two with 'self'.
It looks like this:

# bridge fdb add  00:22:33:44:55:66 dev eth6 master static
# bridge fdb show dev eth6 | grep 55
00:22:33:44:55:66 vlan 1 master br0 offload static
00:22:33:44:55:66 master br0 offload static
00:22:33:44:55:66 vlan 1 self static
00:22:33:44:55:66 vlan 4095 self static

If I do a replace of a locked entry I only get one with the 'self' flag.

>
> I think Hans is testing with mv88e6xxx which dumps entries directly from
> HW via ndo_fdb_dump(). See dsa_slave_port_fdb_do_dump() which sets
> NTF_SELF.
>
> Hans, are you seeing the entry twice? Once with 'master' and once
with
> 'self'?
>
When replacing a locked entry it looks like this:

# bridge fdb show dev eth6 | grep 4c
00:4c:4c:4c:4c:4c vlan 1 master br0 extern_learn offload locked

# bridge fdb replace 00:4c:4c:4c:4c:4c dev eth6 vlan 1 master static ; bridge
fdb show dev eth6 | grep 4c
00:4c:4c:4c:4c:4c vlan 1 self static

The problem is then that the function
br_fdb_find_rcu(br,eth_hdr(skb)->h_source, vid);
, where the h_source and vid is the entry above, does not find the entry.
My hypothesis was then that this is because of the 'self' flag that I
see.

I am thinking that the function dsa_slave_port_fdb_do_dump() is only for
debug, and thus does not really set any flags in the bridge modules FDB,
but then I don't understand why the above find function does not find
the entry?