[lwip-members] Bug: memory leak in etharp.c (Martin Glunz)

[leon . woestenberg]

Hello all,

there exist some nasty code in the etharp.c that leaks memory.

I think the bug report by Martin Glunz catches most of it, quote:

"When etharp_output() returns an ARP request instead of
the original packet, the pbuf containing the request
will not be freed."

I have verified this and yes, this seems to be a leak.

IT AFFECTS ALL CURRENT ETHERNET NETWORK INTERFACE DRIVERS
AS THE BUG FIX MUST OCCUR THERE (in driver_output()).

"My fix is the insertion of a new function
etharp_arp_free() that should be called after the
packet is sent."

This function does work, but I suggest to rename it to something
like etharp_output_sent().

Another alternative patch is to do this in the Ethernet drivers:
q = etharp_output(p);
if (q)
{
  driver_transmit(q);
  if (q != p) pbuf_free(q); /* if an ARP request was returned, free it */
}

Martin submitted another bug report, which I have not verified this
yet.

In addition to that, I have fixed a NULL pointer issue in etharp.c that
will
only bring problems in pre-emptive scheduling situations.

I will check in my fixes to etharp.c which I am confident about and will
not touch the rest.

All in all, etharp.c needs a code audit. I'm currently doing this, but I
welcome the eyes of more people.

Regards,

Leon Woestenberg