Re: [PATCH] arp/ndp: don't send messages on uninitialized tap interface

[Stefano Brivio <sbrivio@redhat.com>]

On Wed, 26 Nov 2025 19:53:16 -0500
Jon Maloy <jmaloy@redhat.com> wrote:

> When running pasta without --config-net, the tap interface is opened
> and assigned a valid file descriptor, but intentionally not brought
> up in the namespace. This is the expected behavior when the user wants
> to configure the namespace manually.
> 
> However, in PASTA mode the code is attempting to send ARP announcements
> and NDP messages (initial requests and unsolicited NAs) based solely on
> whether c->fd_tap is valid, without checking if the interface actually
> is up and ready to transmit. This results in send failures, and when
> debug is activated (pasta -d) we see error printouts for these early
> messages.
> 
> We now add new function tap_is_ready() which checks both conditions:
> - Whether fd_tap is valid (all modes)
> - Whether the tap interface is up (pasta mode only). In this mode, we
>   use the existing c->pasta_conf_ns flag, which indicates whether
>   pasta_ns_conf() configured and brought up the interface. This test
>   is simple, and good enough for now.
> 
> We update all functions that send unsolicited ARP/NDP messages to
> check with the new function before making any send attempt.
> 
> This eliminates spurious send errors when starting pasta without
> --config-net, while preserving correct behavior when the interface
> is properly initialized.
> 
> Signed-off-by: Jon Maloy <jmaloy@redhat.com>
> ---
>  arp.c |  3 +++
>  ndp.c |  5 +++--
>  tap.c | 27 +++++++++++++++++++++++++++
>  tap.h |  1 +
>  4 files changed, 34 insertions(+), 2 deletions(-)
> 
> diff --git a/arp.c b/arp.c
> index 33b03cf..bb042e9 100644
> --- a/arp.c
> +++ b/arp.c
> @@ -168,6 +168,9 @@ void arp_announce(const struct ctx *c, struct in_addr *ip,
>  		struct arpmsg am;
>  	} __attribute__((__packed__)) msg;
>  
> +	if (!tap_is_ready(c))
> +		return;
> +
>  	/* Ethernet header */
>  	msg.eh.h_proto = htons(ETH_P_ARP);
>  	memcpy(msg.eh.h_dest, MAC_BROADCAST, sizeof(msg.eh.h_dest));
> diff --git a/ndp.c b/ndp.c
> index a33239d..eb9e313 100644
> --- a/ndp.c
> +++ b/ndp.c
> @@ -227,7 +227,8 @@ static void ndp_na(const struct ctx *c, const struct in6_addr *dst,
>   */
>  void ndp_unsolicited_na(const struct ctx *c, const struct in6_addr *addr)
>  {
> -	ndp_na(c, &in6addr_ll_all_nodes, addr);
> +	if (tap_is_ready(c))
> +		ndp_na(c, &in6addr_ll_all_nodes, addr);
>  }
>  
>  /**
> @@ -411,7 +412,7 @@ void ndp_timer(const struct ctx *c, const struct timespec *now)
>  	time_t max_rtr_adv_interval = DEFAULT_MAX_RTR_ADV_INTERVAL;
>  	time_t min_rtr_adv_interval, interval;
>  
> -	if (c->fd_tap < 0 || c->no_ra || now->tv_sec < next_ra)
> +	if (!tap_is_ready(c) || c->no_ra || now->tv_sec < next_ra)
>  		return;

Thanks for catching this one as well! I actually spotted the warning a
while ago but never had time to investigate.

Applied.

-- 
Stefano