How do I properly determine if my client is still connected to the server with C sockets? -

i have client connected server via c berkeley sockets. (try to) periodically check see if connection still valid calling recv() on socket msg_dontwait , msg_peek. however, i'm finding call returns eagain both when connection down in cases (e.g., after connection's interface shut down) , when there no data read. hence, there no way me distinguish whether connection still valid or not using method.

here's function:

/* return 1 if still connected, 0 otherwise */  int still_connected() {     int nbytes_recvd = 0;     char buf[2];     int err;      nbytes_recvd = recv(fd,buf,sizeof(buf),msg_peek|msg_dontwait);     err = errno;      if ( nbytes_recvd == 0 )     {         return 0;     }     else if ( nbytes_recvd < 0 )     {        if ( (err == enotsock) ||             (err == enotconn) ||             (err == einval) ||             (err == econnrefused) ||             (err == ebadf) ||             (err == eopnotsupp) )        {           return -1;        }     }      /* nbytes_recvd > 0 or eagain or ewouldblock */     /* eagain if connection down because shut down    interface */      return 1; } 

how can accurately check whether connection still valid?

from point of view of 1 of 2 endpoints of tcp connection, there no difference between peer had disappeared , peer still present isn't sending data. can't distinguish these situations recv() or other method.

the way detect if peer down, has reset, has disappeared, or has become unreachable, send data , see happens. there 2 options:

• enable tcp keepalives (so_keepalive). cause tcp send keepalive packet once in while , expect ack peer. if peer fails ack, can learn connection has been broken.
• send actual application data (or application-level keepalive message) on schedule of choice. similarily, if peer fails respond data (or socket error after sending it), can learn connection has been broken.