System call SELECT and TESTMASK

Question

I'm currently studying OS's and i came across a question that i cant quite figure it out.

I have this piece of pseudo-code:

for(;;) {
   mask = teskmask;
   select(MAXSOCKS,&mask,0,0);
   if(FD_ISSET(strmfd,&mask)) {
     clilen=sizeof(cliaddr);
     newfd=accept(strmfd,(struct sockaddr*)&cliaddr,&clilen);
     echo(newfd);
     close(newfd);
   }
   if (FD_ISSET(dgrmfd,&mask)) echo (dgrmfd);
}

Note: consider MAXSOCKS to be defined as whatever (doesn't matter here), strmfd to be a Stream socket, dgrmfd as a datagram socket, clilen is the size of client address, echo(newfd); is just a function to echo whats in the socket.

So my question is:

what is/whats for the testmask and mask and how is it used here?

I know that select blocks the process until some socket is available to read/write or exceptions.

Answer 1

The simplest way to get answer to your question is just to read select(2) manual page, which says:

int select(int nfds,
           fd_set *readfds, fd_set *writefds, fd_set *exceptfds,
           struct timeval *timeout);

or a quite similar declaration. So "testmask" is fd_set. Select() gets these masks on input and refills them on output, unless an error is signaled.

In all Unix systems I've seen fd_set is bit array up to FD_SETSIZE bits, and it could be hard to extend this to larger value, so a descriptor greater or equal to FD_SETSIZE can't be select()ed without target-specific hacking. So poll() is preferred for a large descriptor set. (And, platform-specific solutions as kqueue (*BSD, MacOSX), epoll (Linux), /dev/poll (Solaris) are better if relatively small subset of all descriptors is ready on each cycle; for Linux it's ~30% when poll() is more efficient than epoll-based scheme, measured for systems ~3 years ago.)

In Windows systems, fd_set is array of socket numbers (integers) with unlimited value, but limited total count (AFAIR, 64).

Using only standard macros to deal with fd_set gives code which is portable between Unix and Windows. But this code could be inefficient for both in different manner.

Answer 2

Actually the proper way to copy the select mask is to use the FD_COPY macro.

I have just posted an article about the difference between select, poll and epoll here: http://www.ulduzsoft.com/2014/01/select-poll-epoll-practical-difference-for-system-architects/ - I suggest you take a look on it as it covers many things you asked, and there's a chance it covers something you missed.