Blocking socket returns EAGAIN

Question

One of my projects on Linux uses blocking sockets. Things happen very serially so non-blocking would just make things more complicated. Anyway, I am finding that often a recv() call is returning -1 with errno set to EAGAIN.

The man page only really mentions this happening for non-blocking sockets, which makes sense. With non-blocking, the socket may or may not be available so you might need to try again.

What would cause it to happen for a blocking socket? Can I do anything to avoid it?

At the moment, my code to deal with it looks something like this (I have it throw an exception on error, but beyond that it is a very simple wrapper around recv()):

int ret;
do {
    ret = ::recv(socket, buf, len, flags | MSG_NOSIGNAL);
} while(ret == -1 && errno == EAGAIN);


if(ret == -1) {
    throw socket_error(strerror(errno));
}
return ret;

Is this even correct? The EAGAIN condition gets hit pretty often.

EDIT: some things which I've noticed which may be relevant.

1. I do set a read timeout on the socket using setsockopts(), but it is set to 30 seconds. the EAGAIN's happen way more often than once every 30 secs. CORRECTION my debugging was flawed, EAGAIN's don't happen as often as I thought they did. Perhaps it is the timeout triggering.

2. For connecting, I want to be able to have connect timeout, so I temporarily set the socket to non-blocking. That code looks like this:

   int      error = 0;
   fd_set   rset;
   fd_set   wset;
   int      n;
   const SOCKET sock = m_Socket;
   
   // set the socket as nonblocking IO
   const int flags = fcntl (sock, F_GETFL, 0);
   fcntl(sock, F_SETFL, flags | O_NONBLOCK);
   
   errno = 0;
   
   // we connect, but it will return soon
   n = ::connect(sock, addr, size_addr);
   
   if(n < 0) { 
       if (errno != EINPROGRESS) {
           return -1;
       }
   } else if (n == 0) {
       goto done;
   }
   
   FD_ZERO(&rset);
   FD_ZERO(&wset);
   FD_SET(sock, &rset);
   FD_SET(sock, &wset);
   
   struct timeval tval;
   tval.tv_sec = timeout;
   tval.tv_usec = 0;
   
   // We "select()" until connect() returns its result or timeout
   n = select(sock + 1, &rset, &wset, 0, timeout ? &tval : 0);
   if(n == 0) {    
       errno = ETIMEDOUT;
       return -1;
   }
   
   if (FD_ISSET(sock, &rset) || FD_ISSET(sock, &wset)) {
       socklen_t len = sizeof(error);
       if (getsockopt(SOL_SOCKET, SO_ERROR, &error, &len) < 0) {
           return -1;
       }
   } else {
       return -1;
   }
   
   done:
   // We change the socket options back to blocking IO
   if (fcntl(sock, F_SETFL, flags) == -1) {
       return -1;
   }
   return 0;
   

The idea is that I set it to non-blocking, attempt a connect and select on the socket so I can enforce a timeout. Both the set and restore fcntl() calls return successfully, so the socket should end up in blocking mode again when this function completes.

Solution 1

It's possible that you have a nonzero receive timeout set on the socket (via setsockopt(sock, SOL_SOCKET, SO_RCVTIMEO,...)) as that would also cause recv to return EAGAIN

Solution 2

Is it possible that you're using MSG_DONTWAIT is being specified as part of your flags? The man page says EAGAIN will occur if no data is available and this flag is specified.

If you really want to force a block until the recv() is somewhat successful, you may wish to use the MSG_WAITALL flag.

Comments

[Jason]

I just grepped my source tree, MSG_DONTWAIT is not used.

[Jason]

yes, but it is set to 30000 milliseconds, I get the EAGAIN's way more often than that. Pretty much constatnly.

[Jason]

CORRECTION my debugging was flawed, EAGAIN's don't happen as often as I thought they did. Perhaps it is the timeout triggering.

[yixiang]

For anyone who want a reference, check here and search "SO_RCVTIMEO".