Hello everyone, today I tried to set up network access to a service using stunnel 4.35. The underlying service, by default, exits after every request+reply. When running stunnel in daemon mode, this seems to upset things such that subsequent connections hang. I cannot quite pinpoint whether it's a config problem, design (of my setup) problem, or bug, so any help's welcome ...

More info:

The underlying service accesses a Nagios' livestatus interface

http://mathias-kettner.de/checkmk_livestatus.html#H1:Remote%20access%20to%20...

using the "unixcat" tool provided. Unless you specifically request connection keepalives, unixcat will close/exit after answering one request, *or* timing out waiting for one. That "exit procedure", AFAICT, isn't different from what the imapd or pppd in the stunnel(8) manpage's EXAMPLES section do ...

I need to secure access to the service, so I'm throwing client cert verification and libwrap at the connections. The current daemon mode configuration (with several commented-out attempts) reads, slightly anonymized: client=no verify=3 pid=/var/run/stunnel/livestatus.pid #setuid=stunnel #socket = a:SO_REUSEADDR=0 #foreground= yes

[livestatus]

accept = [...PORT...]

cert = [...SERVER...].pem key = [...SERVER...].key CAfile = [...CA...].pem CApath = [...TempStagingDirectory...]

exec = /usr/local/bin/unixcat execargs = unixcat /var/lib/nagios3/rw/live pty = yes

retry = no session = 3

#TIMEOUTclose = 0 #TIMEOUTbusy = 5 #TIMEOUTidle = 5

After starting the stunnel daemon, I can perform *one* normal connection:

# openssl s_client -CAfile [...CA...].pem -showcerts \

...

-connect [...SERVER...]:[...PORT...] -cert [...CLIENT...].cert \ -key [...CLIENT...].key -state

CONNECTED(00000003) SSL_connect:before/connect initialization SSL_connect:SSLv2/v3 write client hello A SSL_connect:SSLv3 read server hello A

[...]

SSL handshake has read 1743 bytes and written 1933 bytes

New, TLSv1/SSLv3, Cipher is AES256-SHA Server public key is 1024 bit Compression: NONE Expansion: NONE SSL-Session: Protocol : TLSv1 Cipher : AES256-SHA Session-ID: 1B4221E2917E3BD49AA2102C2CCE60A624DB256D6B4D61F9D0D77EA843DC31BC Session-ID-ctx: Master-Key: 57E6DF8E7E0A741B718E356467450DD00969465295A9C6FE03416ED16E74AF93058EC168EF6DF1B9F3F637F505FF6EF4 Key-Arg : None Start Time: 1298390481 Timeout : 300 (sec) Verify return code: 0 (ok)

---

At that point, I can type in one request, then unixcat terminates, stunnel closes the connection and s_client (?) reports:

read:errno=104

Typical log output on the server side:

LOG5[30042:139819109037792]: Reading configuration from file /etc/stunnel/livestatus.conf LOG5[30042:139819109037792]: Peer certificate location [...TempStagingDirectory...] LOG5[30042:139819109037792]: Configuration successful LOG5[30042:139819109037792]: No limit detected for the number of clients LOG5[30043:139819109037792]: stunnel 4.35 on x86_64-unknown-linux-gnu with OpenSSL 0.9.8g 19 Oct 2007 LOG5[30043:139819109037792]: Threading:PTHREAD SSL:ENGINE Sockets:POLL,IPv6

[...]

LOG5[30043:1075198288]: Service livestatus accepted connection from [...CLIENT...]:39287 LOG5[30043:1075198288]: Certificate accepted: depth=1, [...CA...] LOG5[30043:1075198288]: Certificate accepted: depth=0, [...CLIENT...] LOG3[30043:1075198288]: readsocket: Input/output error (5) LOG5[30043:1075198288]: Connection reset: 2293 bytes sent to SSL, 33 bytes sent to socket

Rerunning the exact same command stops dead after the

CONNECTED(00000003)

and manually "telnet"ing to the port from a different host doesn't behave in a normal way (closing the connection after I type the first line of non-SSL stuff) anymore, either. Running the unixcat command from a local command line, however, still works normally.

I tried running stunnel off inetd, which seems to work, so I'd guess it's the server-side SSL session cacheing which interferes. However, I don't see any related stunnel config keywords in the manpage, and stunnel offering a cached session to a connection *from a different client host* sound odd to me, too ... ?

Kind regards, J. Bern