DHCP re-init leak

[ThomasNauwelaerts]

When the network connection is lost, the device must re-init the DHCP and request an ip to the DHCP server. When removing our ethernet cable from the device, than plugging it back in, we do 'pico_dhcp_initiate_negotiation' again. At that point +-400 bytes are leaking.

[ThomasNauwelaerts]

@jelledevleeschouwer ?

[jelledevleeschouwer]

@ThomasNauwelaerts, this should be investigated. Could you reproduce this issue on a host (i.e. linux) environment?

[ThomasNauwelaerts]

@jelledevleeschouwer , I'm not able to use a host environment.
Problem is, I'm not sure how to re-init DHCP, for example when the device is changed to another network, or the network changes. Now it just did the 'pico_dhcp_initiate_negotiation' (after 'pico_dhcp_client_abort'), but that's leaking memory.
It also seems to be impossible to detect if the current IP address isn't valid anymore?

[szemzoa]

The leak is coming from the function pico_dhcp_server_add_negotiation().
When the client renew the lease, the new request comes with a different xid.
The pico_dhcp_server_find_negotiation() will fail with the new xid as a key, and new struct pico_dhcp_server_negotiation will be inserted to the tree every time.

I don't know if this is totally ok or not, but looking up the entries with the hwaddr and deleting the old entry solves the memleak, something like this:

static void pico_dhcp_server_free_negotiation_hwaddr(struct pico_stack *S, struct pico_dhcp_hdr *hdr)
{
struct pico_dhcp_server_negotiation *dhcpn = NULL;
struct pico_tree_node *idx;

pico_tree_foreach(idx, &S->DHCPNegotiations) {
    dhcpn = idx->keyValue;
    if (memcmp(hdr->hwaddr, dhcpn->hwaddr.addr, PICO_SIZE_ETH) == 0) {
        pico_tree_delete(&S->DHCPNegotiations, dhcpn);
        PICO_FREE(dhcpn);
        break;
    }
}

}

[szemzoa]

oops, this one was for the DHCP server part, which is leaks too, I will try to find the client leak too...