In a lot of places, we use this kind of comparison to detect if a
device has a lower link:
dev->ifindex != dev_get_iflink(dev)
This seems to be a leftover of the pre-netns days, when the ifindex
was unique over the whole system. Nowadays, with network namespaces,
it's very easy to create a device with the same ifindex as its lower
link:
ip netns add main
ip netns add peer
ip -net main link add dummy0 type dummy
ip -net main link add link dummy0 macvlan0 netns peer type macvlan
ip -net main link show type dummy
9: dummy0: <BROADCAST,NOARP> mtu 1500 qdisc noop ...
ip -net peer link show type macvlan
9: macvlan0@if9: <BROADCAST,MULTICAST> mtu 1500 qdisc noop ...
To detect if a device has a lower link, we can simply check the
existence of the dev->netdev_ops->ndo_get_iflink operation, instead of
checking its return value. In particular, I attempted to fix one of
these checks in commit feadc4b6cf42 ("rtnetlink: always put IFLA_LINK
for links with a link-netnsid"), but this patch isn't correct, since
tunnel devices can export IFLA_LINK_NETNSID without IFLA_LINK. That
patch needs to be reverted.
This series will fix all those bogus comparisons, and export missing
IFLA_LINK_NETNSID attributes in bridge and ipv6 dumps.
ipvlan and geneve are also missing the get_link_net operation, so
userspace can't know when those device are cross-netns. There are a
couple of other device types that have an ndo_get_iflink op but no
get_link_net (virt_wifi, ipoib), and should probably also have a
get_link_net.
Sabrina Dubroca (12):
ipvlan: add get_link_net
geneve: add get_link_net
Revert "rtnetlink: always put IFLA_LINK for links with a link-netnsid"
rtnetlink: always put IFLA_LINK for links with ndo_get_iflink
bridge: always put IFLA_LINK for ports with ndo_get_iflink
bridge: advertise IFLA_LINK_NETNSID when dumping bridge ports
ipv6: always put IFLA_LINK for devices with ndo_get_iflink
ipv6: advertise IFLA_LINK_NETNSID when dumping ipv6 addresses
net: link_watch: fix operstate when the link has the same index as the
device
net: link_watch: fix detection of urgent events
batman-adv: fix iflink detection in batadv_is_on_batman_iface
batman-adv: fix detection of lower link in batadv_get_real_netdevice
drivers/net/can/vxcan.c | 2 +-
drivers/net/geneve.c | 8 ++++++++
drivers/net/ipvlan/ipvlan_main.c | 9 +++++++++
drivers/net/veth.c | 2 +-
include/net/rtnetlink.h | 4 ++++
net/batman-adv/hard-interface.c | 4 ++--
net/bridge/br_netlink.c | 4 +++-
net/core/link_watch.c | 4 ++--
net/core/rtnetlink.c | 25 ++++++++++++-------------
net/ipv6/addrconf.c | 11 ++++++++++-
10 files changed, 52 insertions(+), 21 deletions(-)
--
2.28.0
Before Adding radios to my setup I connected to computers with three NICs each. I added all three interfaces to the mesh interface bat0 on each. I then run iperf across it and all the traffic seems to go on one interface. I run iperf3 with -p 4 so there are multiple streams. Changing it to bonding does not seem to change the behavior.
batctl o - shows all three interfaces
batctl n - shows three interfaces -This I thought seemed odd as its one neighbor across three links
batctl tg - shows all clients Via one address
If anyone can point me at what to look at next or what might be wrong would help.
I am using BATMAN_V version 2019.4 in kernel 5.4.68.
If this is not the best place to ask questions regarding mesh networks utilizing the batman-adv kernel module, I apologies and please point me to where I need to be.
I'm looking to set up distributed mesh network using the batman-adv Linux kernel module. However, I don't want to have to statically assign IP addresses to all my nodes therefore my first thought was to use DHCP. The problem arises in my scenario that any node could come and go in the mesh network as they move in and out range of the network. Therefore manually allocating a single or even a few DHCP servers isn't realistic as that DHCP server may drop out of the network at anytime. Is there a dynamic way to reassign the DHCP server based on the nodes still within the network when the previous DHCP server drops from the network?
Thanks,
-tdev