Panic on Unix-domain sockets with length 108

[sunfishcode]

On Linux, Unix-domain sockets with a path of length 108 don't get a terminator within sun_path:

https://man7.org/linux/man-pages/man7/unix.7.html#BUGS

Rustix currently panics on such paths:

#1004 (comment)

[kevinmehall]

The best way to fix this probably involves allowing SocketAddrUnix to hold the 108 byte path. But, since SocketAddrUnix wraps sockaddr_un, it does not have space for 108 bytes plus a nul terminator, and that's a problem for SocketAddrUnix::path returning CStr.

It seems like Path or OsStr would be a more appropriate return type for that function anyway -- may want that breaking change for 1.0.

[sunfishcode]

The best way to fix this probably involves allowing SocketAddrUnix to hold the 108 byte path. But, since SocketAddrUnix wraps sockaddr_un, it does not have space for 108 bytes plus a nul terminator, and that's a problem for SocketAddrUnix::path returning CStr.

It seems like Path or OsStr would be a more appropriate return type for that function anyway -- may want that breaking change for 1.0.

Thanks for pointing out that this should be fixed for 1.0. I've now filed #1371 as a fix. A complication with OsStr is that this code gets used in no-std contexts. I think we can instead have path return a Cow<CStr> so that it can return a CString when it needs to allocate to add a NUL, and add new path_bytes function which returns the raw bytes with the NUL excluded.