Add socklen_t

[dmik]

This is missing since long. We have __socklen_t which is an unsigned 32-bit integer (__uint32_t) and in place of socklen_t our headers use u_int in structs (eg and cmsghdr) and int in function calls. Ie for structs it’s completely safe to replace u_int with socklen_t while for functions some thinking is needed because of the difference in signedness.

As POSIX suggests, socklen_t should not be larger than 2^31-1 which is a 32 bit signed integer. And this is what TCP/IP for OS/2 (which LIBC internally uses) implies too. For msghdr it’s a terrible mess there though, according to the TCP/IP 4.21 Programming Reference at least. Ie msg_namelen is int (which matches socklen_t) but msg_controllen is u_int (should be socklen_t too in POSIX). The OS/2 Toolkit headers define both as u_int though. Even more messy.

[dmik]

I actually think that we should just redefine everything to socklen_t (signed int) to be POSIX compatible on the source level. I don’t think that these length fields are expected to carry values more than 2^31-1 in the TCP/IP stack code anyway so we should end up binary level compatible too.

[SilvanScherrer]

just to give your brain a little boost :) this is done here already https://github.com/bitwiseworks/libcx/blob/master/src/net/libcx/net.h

[dmik]

Note that I also added sa_family_t and sockaddr_storage definitions which are part of the modern POSIX too.

Note however that I did not change the socket function prototypes to use socklen_t instead of int as it's too much work for now (and also requires a lot of changes in the underlying backend code). I will create a new ticket for it.