Re: [Linuxptp-users] PTP4L daemon seems to have a problem with to a link local address for IPv6

[Richard C. <ric...@gm...>]

On Fri, Nov 23, 2012 at 03:33:03PM +0100, Mario Molitor wrote:
> Hallo Richard,
> I have make a observation that the PTP4L daemon seems to have a problem with to a link local address for IPv6 .
> For IPV6 tests I have changed PTP_PRIMARY_MCAST_IP6ADDR  to a link local address.
> (#define PTP_PRIMARY_MCAST_IP6ADDR "FF02:0:0:0:0:0:0:181")

Why do you think this should be a link local address?

(I chose the global scope in order to maximise the chance of reaching
a master when running slave, and vice versa. I confess that I don't
really know what scope is correct to use here.)

> After this change I have saw that the communication  and explicit the udp6_send didn't work.
> 
> I have review the the code of  udp6_send and I have saw that the
> sin6_sopce_id of struct sockaddr_in6 addr is not correct initialize.

Are you running E2E mode?

It looks like we will have to set sin6_scope_id for the Pdelay
messages in P2P mode in any case. I will do some testing on this to
find out.

> 
> I have corrected this and the communication problem disappears.
> 
> --- a/udp6.c
> +++ b/udp6.c
> @@ -135,6 +135,8 @@ enum { MC_PRIMARY, MC_PDELAY };
>  
>  static struct in6_addr mc6_addr[2];
>  
> +static unsigned ip6_scope_id;
> +
>  static int udp6_open(struct transport *t, char *name, struct fdarray *fda,
>  		    enum timestamp_type ts_type)
>  {
> @@ -156,7 +158,8 @@ static int udp6_open(struct transport *t, char *name, struct fdarray *fda,
>  
>  	if (sk_timestamping_init(efd, name, ts_type, TRANS_UDP_IPV6))
>  		goto no_timestamping;
> -
> +	
> +	ip6_scope_id = if_nametoindex(name);

This won't work if you have more than one port. The particular index
must be remembered in a private field derived from the struct
transport.

>  	fda->fd[FD_EVENT] = efd;
>  	fda->fd[FD_GENERAL] = gfd;
>  	return 0;
> @@ -181,11 +184,14 @@ static int udp6_send(struct transport *t, struct fdarray *fda, int event, int pe
>  	ssize_t cnt;
>  	int fd = event ? fda->fd[FD_EVENT] : fda->fd[FD_GENERAL];
>  	struct sockaddr_in6 addr;
> +	memset(&addr, 0, sizeof(struct sockaddr_in6)); /* Init */

This makes sense in any case. Better to use variable name, and the
comment isn't helpful.

	memset(&addr, 0, sizeof(addr));

> +	
   ^
That is a stray tab.

Thanks,
Richard