Need to use tcpip_api_call() instead of tcpip_callback(), since the
former waits for the tcpip task to complete and thus prevents potential
data races with subsequent TCP/IP tasks.
This fixes a bug where we ping our own IP and the request itself
bounces back to the raw receive function and is incorrectly treated as
reply. (this bug was discovered when fixing ICMPv6 pings with incorrect
checksums, while the ping request was dropped in icmpv6.c due to wrong
checksum, but was also fed to raw layers where it was treated as
"correct" response, so the PINGv6 to ourselves still worked)
0 is a valid socket description, so should be accepted after we create a
socket. In IDF, though, it's just a theoretical issue, as customers have
to configure:
* FD_SETSIZE = CONFIG_LWIP_MAX_SOCKETS (from makefiles)
* CONFIG_VFS_SUPPORT_IO = n (from menuconfig)
(this configuration is currently broken, but if we adjust the IDF here
and there we could reproduce the issue)
