Socket arbitrarily connects - or doesnt

Question 1

Unless what you actually, really want to learn is low-level berkeley socket manipulation, I'd suggest you look at libevent or a similar library.

The structure of your main loop is a little unusual. You can clearly only handle one connection at a time, and you don't cope well with any connection attempts that happened while you were servicing a previous connection.

bind(mysocket, (struct sockaddr *)&serv, sizeof(struct sockaddr));

bind can fail, e.g. if another process has recently had the socket open and the OS hasn't finished cleaning up use of the port. You can change this behavior, but you should still check, from die.net's bind manpage

Return Value
  On success, zero is returned. On error, -1 is returned, and errno is set appropriately.

so

if(bind(mysocket, (struct sockaddr *)&serv, sizeof(struct sockaddr))) {
    perror("bind failed");
    exit(1);
}

listen() only needs to be called once, but also needs to be checked

if(listen(mysocket, 2)) {
    perror("listen failed");
    exit(1);
}

after this, if you are content to do the single-service approach, then you can do the following:

mysocket = socket(AF_INET, SOCK_STREAM, 0);
if(mysocket < 0) {
    perror("socket failed");
    exit(1);
}

if(bind(mysocket, (struct sockaddr *)&serv, sizeof(struct sockaddr))) {
    perror("bind failed");
    exit(1);
}

if(listen(mysocket, 2)) {
    perror("listen failed");
    exit(1);
}

for (;;) {
    consocket = accept(mysocket, (struct sockaddr *)&dest, &socksize);
    if(consocket < 0) // might return if the connection has already gone away.
        continue;
    if (!sendGreeting(consocket)) {
        // sendGreeting should return -1 if it was unable to send, 0 if successful
        while (!readLoop(consocket, recvBuf, MAXRCVLEN))
            ;
    }
    close(consocket);
}

readLoop would then be something like:

int readLoop(int socket, char* buffer, size_t bufSize) {
    int len = recv(socket, buffer, bufSize);
    if (len > 0)
        return processBuffer(socket, buffer, len);
    if (len < 0 && (errno == EINTR || errno == EAGAIN))
        return 0; // do-over
    return -1;
}

make sure that processBuffer also returns 0 or -1 accordingly.

As I mentioned above, there are still problems with this approach, but it's not my intent here to teach you everything you need to know about sockets in one pass :) If you want to further develop your socket knowledge, your next stop should be learning about select or poll with non-blocking sockets so that you can host multiple sockets and service them as they become active.

Question 2

Generally, you should use tcpdump/wireshark to see what packets are seen by you Rpi, and strace to see what your program does. My first guess about your connections sometimes not working would be loss of packets. By using wired LAN (Ethernet), you could rule this possibility out.

But the example server code that you're using is a rather bad example. Even if you only want to accept a single client connection at a time, your server should not use blocking waits for any remote message. You should read about using non-blocking I/O, select or poll, and look at examples using these. Also, please read about SO_REUSEADDR, you probably need that one in your server as well.

Question 3

This line code

char msg[] = " ";
do{
  scanf("%s",msg);

will fail miserably if the number of bytes scanned in is larger then 1 character, as msg provides exactly two bytes (from which one is always used as 0-terminator). Feeding more would write out of the bounds of msg and doing so will provoke undefined behaviuor.

To fix this providing at least a minimum of 255 characters to so:

char msg[256] = "";
do{
  scanf("%255s",msg);