Hello,
we've been using perdition as a pop3/pop3s/imap/imaps proxy for about
four years now, first with Debian Sarge package and now under Etch.
And throughout this time I've seen pop3s (and from the looks of it
the same happens with imaps) processes stuck in connect, like this:
---
16836 ? S 5:31 0 120 32179 2204 0.0 perdition.pop3s
28070 ? S 0:00 0 120 32311 1564 0.0 \_ perdition.pop3s: connect
7782 ? S 0:00 0 120 32311 1564 0.0 \_ perdition.pop3s: connect
24468 ? S 0:00 0 120 32311 1568 0.0 \_ perdition.pop3s: connect
14180 ? S 0:00 0 120 32311 1568 0.0 \_ perdition.pop3s: connect
13503 ? S 0:00 0 120 32311 1564 0.0 \_ perdition.pop3s: connect
---
They never die off, keep the connection open, there is no traffic and the
other end might be long gone. Last trace in the logs is always like this:
---
Feb 5 22:05:16 pp11 perdition[7782]: Connect: hi.mi.ts.u->203.216.5.113
---
It must be something related to the SSL'ness of these service, since I'm
not seeing this happening ever for imap/pop3. Alas a lot of people do use
TLS with those, so it's not a generic SSL issue. Maybe the master process
could kick a child handling connections in the head after "timeout"
seconds in connect state?
If more information is needed I can try to provide it, but note that with a
rate of roughly 35 pops per second I'm a bit weary to turn on
debugging. ^_-
This may or may not be related to another SSL related issue, which will
be for the sake of making searches in the archive more likely to find good
keywords in a separate mail.
Regards,
Christian
--
Christian Balzer Network/Systems Engineer NOC
chibi(a)gol.com Global OnLine Japan/Fusion Network Services
http://www.gol.com/
Hi,
I've tried perdition_1.19~rc4-3 but I still get a segfault. I'm
running cyrus 2.4.8, and I can avoid the segfault under two scenarios:
1) With "sasl_mech_list: LOGIN PLAIN" and "allowplaintext: no" in my
cyrus config
2) With "sasl_mech_list: PLAIN" and "allowplaintext: yes" in my cyrus config
In my environment, we have a client which can't do TLS, so a special
sieve daemon which allows plaintext listens on local host for it.
Originally #2 also offered LOGIN and I that's how I ran into the
segfault.
It seems to be something in the sieve banner parsing. I've made a
python script to act as a sieve server (see attached):
./server.py
Then I run perdition:
perdition.managesieve -d --no_daemon --listen_port=2006 -C
--outgoing_server=127.0.0.1:2005 -f '' --pid_file /tmp/sieve.pid
Then I telnet to 2006 and paste :
AUTHENTICATE "PLAIN" "<real auth string>"
I either get an OK or a segfault in perdition. server.py will cycle
through the specified banners, so you can continue to telnet to try
each banner test case out.
Let me know if you need any further information. Thanks.
--
Steven Kurylo
