getsockname returns the actual length of the address via sin_len. That is why it is called that way (sending it the address of a local variable).

My contribution to WAGing is that getsockname's 3rd parameter was not always socklen_t * . It is/was int * on some (older?) platforms.

Are you porting to an odd platform?
What is the prototype for getsockname on your platform?
Is Autoconf screwing up its detection of getsockname's prototype?

That and line 632 was " if ((tp = gettosbyname("interactive", "tcp")) && "
in my copy of the source but I don't know if you've otherwise edited it.

Are you sure that line number is correct or have you edited the file in question? The only reason I ask is that the source I grabbed off the MIT site has line 632 and 633 as

\nif ((tp = gettosbyname("interactive", "tcp")) &&\n        (setsockopt(f, IPPROTO_IP, IP_TOS, &tp->t_tos, sizeof(int)) < 0))\n

If that code was somehow getting ifdeffed in (I don't think it should on Linux) then the two messages you list could make sense.

The warning could result from the compiler not having a prototype for gettosbyname and defaulting to returning int, which then triggers the warning of trying to convert the int result to a pointer (tp)

The error could then be explained by the derefencing of the tp pointer in the setsockopt call before the struct type had been fully specified (the struct tosent *tp; line would in effect be a forward declaration that never got completed before it was used because the compiler doesn't know what "struct tosent" is).