axgbe: Fix setting promisc mode

[zlei]

Summary

Ethernet drivers should respect IFF_PROMISC rather than IFF_PPROMISC.
The latter is for user-requested promisc mode, it implies the former
but not vice versa. Some in-kernel components such as if_bridge(4) and
bpf(4) will set promisc mode for interfaces on-demand.

While here, update the debugging message to be not confusing.

This was spotted while reviewing markj@ 's work D46524.

Tested by: ???
MFC after: 1 week

Test Plan

Either of the following should suffice.

1. Do traffic sniffing on axgbe interface. The interface will enter promisc mode and should see packets not for us.
2. Create if_bridge(4) and bring axgbe interface a member. Make traffic over if_bridge.

[zlei]

I do not have axgbe hardware so can not test it by myself. I'd appreciate if someone could test this fix !

[zlei]

sys/dev/axgbe/if_axgbe_pci.c

2358

Well, the allmulti mode (IFF_ALLMULTI) should also be handled correctly, but that is out of scope of this change.

[markj]

This looks ok, but yes it needs to be tested. I don't have this hardware.

[franco_opnsense.org]

Let's get this in for the sake of correctness although my testing was inconclusive and will circle back eventually.

[stephan.dewt_yahoo.co.uk]

Just to comment on the unconditional promisc mode initialization, this is shown in more detail here: https://github.com/freebsd/freebsd-src/blob/main/sys/dev/axgbe/xgbe-dev.c#L2032-L2036.

At the time tests showed that multicast traffic was filtered even when specifically enabled in hardware, with the above change as the only feasible workaround.

[zlei]

In D46794#1076531, @stephan.dewt_yahoo.co.uk wrote:

Just to comment on the unconditional promisc mode initialization, this is shown in more detail here: https://github.com/freebsd/freebsd-src/blob/main/sys/dev/axgbe/xgbe-dev.c#L2032-L2036.

Emm, that is unfortunate.

At the time tests showed that multicast traffic was filtered even when specifically enabled in hardware, with the above change as the only feasible workaround.

Is that specific to FreeBSD only ? I do not see this workaround in Linux driver https://github.com/torvalds/linux/blob/master/drivers/net/ethernet/amd/xgbe/xgbe-dev.c#L2855 .

[stephan.dewt_yahoo.co.uk]

In D46794#1076717, @zlei wrote:

Is that specific to FreeBSD only ? I do not see this workaround in Linux driver https://github.com/torvalds/linux/blob/master/drivers/net/ethernet/amd/xgbe/xgbe-dev.c#L2855 .

I don't know for sure, but chances are this issue has never been seen on Linux given the low adoption.

FWIW, I just re-tested this and indeed CARP packets are filtered out without the "RA" bit set. In fact, even with promiscuous mode on (https://github.com/freebsd/freebsd-src/blob/main/sys/dev/axgbe/xgbe-dev.c#L974), unicast packets not destined for this MAC are filtered out as well.

[zlei]

In D46794#1077535, @stephan.dewt_yahoo.co.uk wrote:

In D46794#1076717, @zlei wrote:

Is that specific to FreeBSD only ? I do not see this workaround in Linux driver https://github.com/torvalds/linux/blob/master/drivers/net/ethernet/amd/xgbe/xgbe-dev.c#L2855 .

I don't know for sure, but chances are this issue has never been seen on Linux given the low adoption.

It will be good if the Linux's version can be tested. So that we can confirm whether it is firmware/hardware/design issue or (probably) not .

FWIW, I just re-tested this and indeed CARP packets are filtered out without the "RA" bit set. In fact, even with promiscuous mode on (https://github.com/freebsd/freebsd-src/blob/main/sys/dev/axgbe/xgbe-dev.c#L974), unicast packets not destined for this MAC are filtered out as well.

The line 974 is actually for multicast promiscuous mode (IFF_ALLMULTI). The current code do not handle it correctly. See my previous comment under axgbe_if_promisc_set().
May you please test the unicast promiscuous mode only right now ( without the "RA" bit set )?

A quick setup would be directly connected two ports / hosts via cross twist cable or DAC cable, as nowadays even cheap L2 switch is smart and will filter out unwanted packets.