[vdr] [watchdog-crash] SVDRP should not block

[Udo Richter <udo_richter@gmx.de>]

Hi list.

I've successfully tracked down a bug that caused several VDR restarts on 
my machine lately. The problem is caused by vdradmin's SVDRP connection 
and VDR dying of watchdog expire. I've logged such an incident with 
tcpdump to see what happens, and the expire happened because vdradmin 
requested epg data and then blocked the TCP/IP connection for 2min45s 
without accepting any data.

At the time of the crash, two recordings were running and two instances 
of noad were scanning, so there was noticeable load on the machine. 
vdradmin was idle for 2 hours, thats probably the cause for the long 
delay in processing. (after the 2m45s, vdradmin processed all remaining 
buffered data normally, and even tried to request timers before the 
connection terminated.)

VDR should not block when sending SVDRP data, or at least should timeout 
soon. In any case, the SVDRP connection should be terminated before the 
watchdog timer expires. Keeping recordings running is more important 
than answering epg requests.

(I know there is a vdradmin version that directly reads epg.data, but 
that doesnt solve the underlying problem.)

I've attached a testcase perl script that blocks VDR via SVDRP and 
forces a watchdog expire. Obviously, these things can happen in reality, 
and not responding to the remote is not the worst possible problem.

Cheers,

Udo

#!/usr/bin/perl
use Socket;

socket(SOCK, PF_INET, SOCK_STREAM, getprotobyname('tcp')) or die "socket";
connect(SOCK, sockaddr_in(2001, inet_aton("localhost"))) or die "connect";
select(SOCK); $| = 1;
print SOCK "LSTE\r\n";
sleep(240);
close(SOCK) or die "close";