How to get a remote unix Process ID?

If I understand the question correctly - you are running a network-aware application that somebody connects to, and you want to figure out who it is that has connected to it - then no, there is no reliable way to do that.

There have been several attempts to do something like this, most notably the Authentication Server as described in RFC 931, commonly known as 'auth' or 'identd'.  However, all of those mechanisms ultimately rely on information that has to be provided by a process running on the remote host, and that's where the problem comes in: can you really trust it?  Can you really assume that when some process on some machine tells you that it is running as user ID 505 and username 'mxkhec', this really is true?  In the end, the answer is 'no'.

In case you are not really worried about security, there is one simple and a couple of more complicated ways to try to figure out who has connected to you.  All of them require that you know the IP addresses and port numbers for the connection - both local and remote.  This can be done easily in most programming languages: the getsockname(2) and getpeername(2) system calls in C, the similar getsockname() and getpeername() functions in Perl, or the peername/sockname methods of the Perl IO::Socket module, and so on.  If you want to limit yourself to shell scripting, this can be done using e.g. the 'getpeername' and 'getsockname' utilities from the netpipes package - <URL:http://web.purplefrog.com/~thoth/netpipes/ftp/>.  All of these functions or utilities should be applied to the file descriptor which holds the socket - if you have opened a telnet connection, you would usually have the choice of file descriptors 0, 1, and 2, though there are cases when those file descriptors would actually be connected to virtual terminals (/dev/tty* pseudo-devices) and not real sockets.  In some cases, such as when using the Secure Shell (SSH), the server would set an environment variable with this information, but sometimes it won't.

As an example, imagine that I have opened a TCP connection from a machine named 'straylight' to another machine named 'db', and something on 'db' wants to figure out who am I on 'straylight'.  That "something" (most likely a program) would invoke the getsockname() and getpeername() system calls, and find out that the connection was made from 10.0.8.129 port 4356 (this is straylight) to 10.0.8.9 port 22 (this is db).

Once you have established the local and remote addresses and ports, the easiest way to actually determine the user ID / username would be to use the auth protocol: open a TCP connection to port 113 of the remote machine and send a line of text containing first the remote port number, then your local port number, separated by a comma (and for compatibility with all kinds of identd's out there, it never hurts to put spaces around the comma), like this:

[roam@db:pts/3 ~/tmp]$ telnet 10.0.8.129 113
Trying 10.0.8.129...
Connected to 10.0.8.129.
Escape character is '^]'.
4356 , 22
4356 , 22 : USERID : FreeBSD : roam
Connection closed by foreign host.
[roam@db:pts/3 ~/tmp]$

Here, the first line - "4356 , 22" - is what I typed (or what the program on 'db' would send).  Thus, I made a request to the identd server on straylight to tell me what is the program that has port 4356 on that side, and port 22 on mine.  The auth server replied with the next line - '4356 , 22 : USERID : FreeBSD : roam' - telling me that the program is running as user 'roam', and (most ident servers won't do that for security reasons) that the machine is running the FreeBSD OS.  For more information about the auth/ident protocol, take a look at RFC 931, e.g. at <URL:http://www.faqs.org/rfcs/rfc931.html>.

Another way for the program on db to determine who has connected to it from straylight would be to connect back to straylight and run some program there.  There are two programs often used for the purpose: the 'fuser' utility on System V OS's, and the 'lsof' program on pretty much all Unix-like OS's.  If you have 'fuser' on your machine, you can connect to straylight and execute the 'fuser -n tcp 22,10.0.8.129,4356' command - see the fuser(1) or fuser(8) manual page for more info.  If you have installed the 'lsof' utility, again, you have to connect to straylight and run the 'lsof -n -i tcp@10.0.8.129:4356' command; again, for more information, consult the lsof(1) manual page.

Note that in all three scenarios: an auth query, fuser, or lsof - you are dependent on whether the remote server will really behave itself and give an 'honest' answer, and there is absolutely no way to verify that answer.  A prime example of this was the reason why the auth service, once widely used for authentication, is now pretty much disabled everywhere: it could be depended upon once, when most machines were Unix terminal servers and nobody but the actual system administrator could start processes that could bind to port 113, but now that everybody has either a Windows or their own personal Unix machine, it is very, very probable that the response will not come from a "real" auth service, but from some program specifically instructed to give that particular response so to fool any authentication attempts.  The same goes for the fuser and lsof methods - you can never be sure that the fuser or lsof commands that you are executing are the "real" commands, and not some backdoored/trojaned versions tailored specifically for this reason.

Hope that helped a bit, and sorry for the long-winded explanation :)

G'luck,
Peter