On Mon, 2014-04-28 at 17:41 -0700, Justin Maggard wrote:
> If you are using a bond interface, the bond interface and the slave
> interfaces will share the same MAC address. This can eventually
> cause connman to crash, because several things depend on a unique
> MAC address per interface.
>
> We can just skip slave interfaces altogether, since connman can't
> do anything with them anyway.
> ---
> src/rtnl.c | 12 +++++++++---
> 1 file changed, 9 insertions(+), 3 deletions(-)
>
> diff --git a/src/rtnl.c b/src/rtnl.c
> index be1bce9..19cbe20 100644
> --- a/src/rtnl.c
> +++ b/src/rtnl.c
> @@ -431,6 +431,13 @@ static void process_newlink(unsigned short type, int
> index, unsigned flags,
> address.ether_addr_octet[4],
> address.ether_addr_octet[5]);
>
> + if ((flags & IFF_SLAVE) ||
This part of the test looks ok.
> + memcmp(&address, &compare, ETH_ALEN) == 0) {
This part of the test breaks all rtnl messaging for interfaces that
don't have ethernet addressing. Cellular connections are the first ones
that comes to mind.
> + connman_info("%s {newlink} skipping index %d address %s flags
> %x",
> + ifname, index, str, flags);
> + return;
> + }
> +
So slave interfaces should be filtered out if nothing else than ethernet
bonding creates them. The info output must spell out that it's filtered
out due to being a slave interface so we get better debugging.
> switch (type) {
> case ARPHRD_ETHER:
> case ARPHRD_LOOPBACK:
> @@ -442,9 +449,8 @@ static void process_newlink(unsigned short type, int
> index, unsigned flags,
> break;
> }
>
> - if (memcmp(&address, &compare, ETH_ALEN) != 0)
> - connman_info("%s {newlink} index %d address %s mtu %u",
> - ifname, index, str, mtu);
> + connman_info("%s {newlink} index %d address %s mtu %u",
> + ifname, index, str, mtu);
Yes, that 'if (memcmp...' looks a little artificial. Ok to remove it.
Cheers,
Patrik
_______________________________________________
connman mailing list
[email protected]
https://lists.connman.net/mailman/listinfo/connman
