I know that TCP provides stream-like data transmission, but the main question is - what situations can occur while sending data over TCP?
1. The message can be split to N chunks to fit in MTU size.
2. Two messages can be read in 1 recv call.
Can there be the next situation?
MTU for example 1500 bytes.
Client calls send with 1498 bytes data.
Client calls send with 100 bytes data.
Server calls recv and receives 1500 bytes data.
Server calls recv and receives 98 bytes data.
So it end up with situation when 2 bytes from second client send will be received in first server recv.
My protocol defined as foolows:
4 bytes - data length
data content.
I wonder can I came up with situation when 4 bytes (data length) will be split into 2 chunks?
Yes, a stream of bytes may be split on any byte boundary. You certainly can have your 4 byte data length header split in any of 8 different ways:
4
1-3
2-2
3-1
1-1-2
1-2-1
2-1-1
1-1-1-1
Some of these are more likely to occur than others, but you must account for them. Code that could handle this might look something like the following:
unsigned char buf[4];
size_t len = 0;
while (len < sizeof(buf)) {
ssize_t n = recv(s, buf+len, sizeof(buf)-len, 0);
if (n < 0) {
// error handling here
}
len += n;
}
length = buf[0] | (buf[1] << 8) | (buf[2] << 16) | (buf[3] << 24);
I always write my applications in a manner that expects the data to become fragmented somehow. It's not hard to do once you come up with a good design.
What's the best way to monitor a socket for new data and then process that data?
