send() return value

Hello, i got a question, and would be nice if someone can help me.

1) Is it safe to closesocket() right after send() in case send returns the correct ammount of bytes sent?

I thought it was, but now im thinking maybe im wrong.
After having problems with a program im writing, i did some research and it seems that the return value of send (bytes sent if no error) is the ammount of bytes "sent" to some TCP buffer, wich will later send the bytes to the peer.

2) So, closing the socket right after send() can make it impossible to the peer to recv if the socket has been closed before the bytes were actually sent through the network?

Thanks in advance.

****************************
Solved

The answer is (in case someone is interested)
Yes, that can happen.

Quote:

---

Originally Posted by stratoforce

Is it safe to closesocket() right after send() in case send returns the correct ammount of bytes sent?

---

The right sequence of closing a TCP connection without loosing data is:

Call shutdown() (how parameter = 1) when you finish sending the data.
Call closesocket() after recv() returns 0.



Regards
Doron Moraz

Thanx for the reply :)

I had a problem where the sending process closed the socket too early.
This happens once every 1000 packets, sometimes after 4k, its very random.

Im talking about custom packets, with a couple of fields, being the relevant ones for this post the "length" and some binary content.
So, i have been trying stuff and there are 2 solutions that worked.

1)I made the receiving process to do a shutdown() only after it received all the data, the entire packet ( i know the lenght because it is included at the beginning of each packet).
Then i do an aditional recv() on the sending process, and close the socket only after it returned zero.
Dunno if this is the correct way, but it worked.

2) Close the socket in the receiving process after it received all the data, then do an aditional recv() in the sending process, and close the socket only after it returned zero.
This approach workted too.

Maybe in this particular situation im having, shutdown() is not necesary. Any opinions/advices?
I feel there's something about shutdown im missing.

Btw, parameter 1 (SD_RECV) also sends a FIN?

Hi stratoforce,

Take a look at:
http://msdn.microsoft.com/en-us/libr...81(VS.85).aspx


Reagrds
Doron Moraz

Thanks :)

I had already read that, but reading it again + testing some stuff cleared things.