Hi ;)

Trying to make my own network game. For now made simply sending. Can easily send message to clients and from client to server. But now thinking how to send for example how changed one of client information. I only imagine that can be done by send method with some string parser on receive. Before sending I make format string (from to command parameters), and on receiving I will parse it. Or maybe possible to send so struct?

In my network model server at the same time is client (he can play to, just re-translate clients info to other clients). Every client (and server) holds info about whole players. And on connection/disconnection every client list will update. If any client moves, it will send info to server, and server sends to other players, and player will be updated/repainted. So I need info from what ip, to what ip (or all ips), and where moved player. This is my first network, so my approach could be bad...

Sending on sockets is similar to writing to files. Reading from sockets is similar to reading from files. Thus, you typically need to prefix each message with a message type, and perhaps a message length, so that you can easily tell when a full message is received, and start parsing it.

Yes, you can send structs, just like you can write structs to a file.

The main problem is that send() may not send all the bytes you request to send, and recv() may not receive all the bytes you request to receive. Thus, you typically need an outgoing buffer before send, and an incoming buffer after recv.

Like what hplus0603 said- you need to put some header on your messages (msg-type-ordinal) matching the info grouping type being transmitted, so the other side can decide what to do with that message, and the handler for that message type would require either having a fixed data size or immediately expect a data block size number following that msg-type-ordinal (for dynamicly sized messages). Then your incoming message handler will be able to interpret the incoming message and figure out if the entire message has arrived (in your network buffer) so that it can be properly processed -- otherwise it will likely have to wait til the entire message has arrived.

This kind of size specifying method is sometimes called 'chunking' where the actaul data is a contiguous block of a fixed or dynamic size and that size is included in the message being processed (some kind of header)

(similar example is graphics files which can have multiple 'chunks' of different data and have offsets included in the file headers and sub headers to be able to handle variable data block sizes)

For now I sending "commands": from ip; to ip; type (connect,disconnect,move); any parameter. On receiving I split whole string in 4 parts (that would be same as before sending). And then doing something depending on what type was. For now its working mostly time. But sometimes on receiving I get more then one message...But thinking that it's not "command" problem (making/parsing) :)

sometimes on receiving I get more then one message

That is exactly what we've been telling you: Sending data over TCP is like writing to a file that you can't seek on. Receiving messages is like reading from a file that you can't seek on.
Not only will you get multiple messages in one receive, you may get half a message in one receive, and the other half in the next receive.
You need to use some way of splitting messages. Either send a length prefix, or send a text-mode command terminator, like a newline character.
Then receive all you can into a buffer, and look for complete messages at the beginning of this buffer. When you find complete messages, execute them, and remove them from the buffer, retaining whatever else is left in the buffer.