Check if socket is connected or not

Question

I have an application which needs to send some data to a server at some time. The easy way would be to close the connection and then open it again when I want to send something. But I want to keep the connection open so when I want to send data, I first check the connection using this function:

bool is_connected(int sock)
{
    unsigned char buf;
    int err = recv(sock,&buf,1,MSG_PEEK);
    return err == -1 ? false : true;
}

The bad part is that this doesn't work. It hangs when there is no data to receive. What can I do? How can I check if the connection is still open?

Solution 1

Don't check first and then send. It's wasted effort and won't work anyway -- the status can change between when you check and when you send. Just do what you want to do and handle the error if it fails.

To check status, use:

int error_code;
int error_code_size = sizeof(error_code);
getsockopt(socket_fd, SOL_SOCKET, SO_ERROR, &error_code, &error_code_size);

Solution 2

You need to enable non-blocking behavior, by setting O_NONBLOCK using fcntl. One easy but non-standard way to do a non-blocking read would be to use:

recv(sock, &buf, 1, MSG_PEEK | MSG_DONTWAIT);

Afterwards, you must check errno if it fails. It can fail with EAGAIN or it can fail with EBADF or ENOTCONN etc.

---

Obviously, the simplest and cleanest way to deal with this would be to avoid "forgetting" if the socket is connected or not. You'll notice if the socket becomes disconnected once a recv returns 0 or a send returns EPIPE.

Solution 3

Default use of TCP doesn't allow very timely detection of dead sockets (outside of normal closure) so I'll suggest that an "is_connected" function like this is largely useless for all practical purposes. Consider implementing an application-layer keep-alive and track if it's alive based on timely responses (or lack thereof).

edit: after posting i see BoBTFish's link, which is effectively the same thing.

Comments

[BoBTFish]

stackoverflow.com/questions/4142012/…

[user207421]

This won't detect all connection failures. Only a send() can do that.

[cnicutar]

@EJP That's true in fact. I was just improving the solution of the op. What you are suggesting is the only bullet-proof solution, but it entails sending.

[Rafael Baptista]

True... but sometimes it is useful to check if a socket is still connected. Setting it to non-block, and then reading with peek_msg is a good way to elicit an error code to check it the socket is still connected.

[user207421]

@RafaelBaptista Why? Using it will check. Testing first cannot work reliably, for the reason David gives above.

[che]

@EJP: I want to check client connection is alive so I don't waste resources in case it is not.

[David Schwartz]

@che Right, so check by using it. Then if it's not alive, you can stop wasting resources. The check itself is a waste of resources.

[Seth]

If you are monitoring connections, say for a clustered app, it can be nice to periodically test connections to ensure that any failover can be performed even when connections are not in use. That would be a decent reason to test a connection.

[David Schwartz]

@seth Which you do by using it.

[avernus]

Also, it is worth mentioning here that since we are using MSG_DONTWAIT flag, we do not need a separate fcntl() syscall to mark socket non-blocking via O_NONBLOCK: MSG_DONTWAIT makes socket non-blocking here.