Re: [LineControl-development] linesrv 2.1.15 leaks filedescriptors

[S. F. <lin...@sr...>]

Hi,

> just today I discovered why I cannot restart linesrv while my
> connection is  up:
> # lsof -i -P -n | grep 16007
> linesrv  18856  0  3u  IPv4 46809507  TCP 192.168.0.30:16007
> (LISTEN) linesrv  18856  0  4u  IPv4 46809508  UDP
> 192.168.0.30:16007 linesrv  18856  0  7u  IPv4 46809598  TCP
> SRV:16007->ME:36731 (ESTABLISHED) dhclient 18956  0  3u  IPv4
> 46809507  TCP 192.168.0.30:16007 (LISTEN) dhclient 18956  0  4u 
> IPv4 46809508  UDP 192.168.0.30:16007 dhclient 18956  0  7u  IPv4
> 46809598  TCP SRV:16007->ME:36731 (ESTABLISHED)
> 
> SRV == My server's ip
> ME  == my workstation ip
> 
> This looks a lot like linesrv is fork()ing for running up/down
> scripts without  closing unneeded filedescriptors after the fork
> (the child should do so  though).

As far as I remember... you're right. I assumed that the forked
processes get terminated before the parent process gets terminated. In
your case this doesn't apply.

> the config for the device is:
> line           myline1
> interface      eth1
> con_type       netdev
> script_up      /sbin/ifup eth1
> script_dn      /sbin/ifdown eth1
> line_baud_up   2000000
> line_baud_down 2000000
> allow_manually yes
> script_esc     /sbin/ifdown eth0
> 
> 
> Would be nice if this could be fixed in case it's not fixed already.
> 
> linesrv-2.1/server/NEWS doesn't mention such a change between 2.1.15
> and  2.1.20 so I guess it's still there.

Hmm... I just wrote a rather long paragraph about why there's a
time to wait before a tcp socket can be bound to a previously used
tcp port. Then I probably got the problem:

You run linesrv, start the connection and the script_up doesn't
terminate? Your linesrv dies (why??? shouldn't!)/gets killed and the
script_up keeps running whith the tcp-16007 socket still open, right?
Then you can't restart linesrv... ok, that's no surprise anymore ;).

The neccessary filedescriptors can be accessed via the global
variables "server" and "cltlist" (see global_vars.h). Do we need
to close all (what happens with the connections to the clients
if we issue a close() on their file descriptors?) or just the
server socket?

Closing needs to be done between fork() and exec*(). So this is
probably in execute.c.
I'll be glad if someone could do this for me as I've quite a
few examinations in the next weeks and I regard this as a low
priority bug.

Due to bad design (design??) of linesrv one has to make sure
not to call close() for a file descriptor which is not open yet,
e.g. when exec_dont_care*() gets called before the socket
is opened. Maybe we're lucky and exec_dont_care*() gets only
called with an open socket... has to be checked.

Greetings & thanks for the report

Stefan Fuchs
--
http://srf.ch/