id	summary	reporter	owner	description	type	status	priority	component	version	resolution	keywords	cc
6	ntrip server connection reset with ntripcaster	jshih@…	stuerze	"Bug report from Jun Shih (Jun.Shih@novatel.com)

=========== first message ===================
I have seen the following ntrip server reset condition in my log:
[02/Nov/2010:21:26:43] [22:Source Thread] add_chunk: hex [], read=21
[02/Nov/2010:21:26:43] [22:Source Thread] add_chunk: http chunk left=0
[02/Nov/2010:21:26:43] [22:Source Thread] DEBUG: Source received 0 bytes in try
2, total 0, errno: 0
[02/Nov/2010:21:26:43] [22:Source Thread] Didn't receive data from source after
800 milliseconds, assuming it died...

A bit debugging found that this caused by the ""http chunk left=0"" condition,  the recv will return 0 because recv length parameter  is zero  (con->http_chunk->left - read_bytes =0) in source.c:663

                len = recv(con->sock, con->food.source->chunk[con->food.source->cid].data + read_bytes, con->http_chunk->left - read_bytes, 0);

when len=0, subsequently maxread=0 as well.  These cause the while loop to terminate on line 737 or 744.  I believe this code is the same as 2.0.14.

=========== second message ===================
But in my case connection is not shutdown (I can see from wireshark).  Note that recv returns 0 if the length parameter is zero (which is not undocumented in the manual) and it does not return -1 in this case.   I have attached a sample program to demonstrate that, to test “ telnet localhost 5000” and type some chars and a return from another terminal.

[jshih@ntriptest tcp_server]$ make tcp_recv
cc     tcp_recv.c   -o tcp_recv
[jshih@ntriptest tcp_server]$ ./tcp_recv 

               TCPServer Waiting for client on port 5000
               got a connection from (127.0.0.1 , 43891)bytes_recieved 0 errno 0
               bytes_recieved 0 errno 0
               bytes_recieved 0 errno 0

I know my server is continuously sending but with inserted delay test.  Further  I changed the code to demonstrate that, there are data in the pipe in my case:

                if (con->http_chunk->left < SOURCE_READSIZE) {
                    len = recv(con->sock, con->food.source->chunk[con->food.source->cid].data + read_bytes, con->http_chunk->left - read_bytes, 0);
                    xa_debug (4, ""Recv LEN1: recv left=%d read=%d len:%d errno:%d"", con->http_chunk->left, read_bytes,len, errno);

                   if (len == 0)
                  {
                    len = recv(con->sock, con->food.source->chunk[con->food.source->cid].data + read_bytes, con->http_chunk->left - read_bytes+1, 0);     
                    xa_debug (4, ""Recv LEN2: recv left=%d read=%d len:%d errno:%d"", con->http_chunk->left, read_bytes,len, errno);
            
                   len = 0;
                 }
                maxread = (int)(0.5 * con->http_chunk->left);

Below is the log capture:

 [03/Nov/2010:15:53:08] [6:Source Thread] add_chunk: hex [], read=21
[03/Nov/2010:15:53:08] [6:Source Thread] add_chunk: http chunk left=0
[03/Nov/2010:15:53:08] [6:Source Thread] Recv LEN1: recv left=0 read=0 len:0 errno:0
[03/Nov/2010:15:53:08] [6:Source Thread] Recv LEN2: recv left=0 read=0 len:1 errno:0

As you can see if I immediately give a length parameter=1, I successfully  recv’ed 1 byte.  So there are data.

Also I made a change to source.c (see attached).   I run for hours without connection drop (previously was dropping every a few minutes), please review the changes to see if this is a right solution.

=== changes in source.c ====
736c736
<               } else if (len == 0) {
---
>               } else if (len == 0 && con->http_chunk->left) {
744c744
<       } while ((read_bytes < maxread) && (tries < source_read_tries));
---
>       } while ((con->http_chunk->left==0 || read_bytes < maxread) && (tries < source_read_tries));

======================

"	defect	closed	normal	Professional Caster	ntripcaster 2.0.9	fixed	"recv() in source.c ; http chunk left=0""-condition"	Jun.Shih@…