I repost here an updated version of the bug resport I posted on the eagle-usg.org forum under the title "blocage pppoa".
Platform : Linux Fedora Core 2, last kernel release (2.6.8-1.521), eagle-usb-1.9.9.1 or eagle-usb-2.0.0rc3. Processor Athlon XP 2400+, 512 MO RAM, Modem Sagem Fast 800, Ethernet card Realtek (this may be related). The Ethernet card is connected to a network comprising one or two other machines (a desktop and a laptop), the machine with the modem acts as a server, with Samba (used only on the local network), ssh, httpd... ADSL 512Kbit/s, Tiscali. The machine usualy acts as a router, with NAT and firewall (iptabes). The firewall rejects most connections from the internet, except ssh. It has been configured through "system-config-security-level". Users can also connect to the server through XDMCP from an other machine (used as an X terminal). Problem : pppoa blocks, using 100% CPU, mostly in system mode. The machine becomes almost unusable : a single interaction under X11 takes several minutes, including to switch to console mode (Ctrl-Alt-F1). A shell is a bit more responsive in console mode, so that a few investigations can be done (including the use of SysReq magic key, if enabled). I did not notice anything anormal regarding the memory use. I noticed the problem for the first time after several modifications of the environment : kernel update, eagle-usb update, and modification of the configuration of the firewall. As the problem happen in some specific conditions, I cannot say which after which modification it happened. Conditions : There is no clear relation between the activity of the machine and the problem, but it always happen when another machine communicate with the server through Ethernet. This is most often when another user connects to the server through XDMCP, but it also happened while synchronizing files between the laptop and the server (using Unison). There is no clear relation between the activity on the ADSL network and the problem : it often arise when there is no particular activity on ADSL... But there are very frequent connexion attempts on Samba ports (on average, once every 2 to 5 seconds). The modification of the firewall done before the problem was related the the processing of unauthorized connections (from DROP to REJECT, if I remember well). In fact, I suspect that the problem is related to a conjunction of events on ADSL and on the Ethernet interface. To reproduce the problem, I usually connects to the server from another machine through XDMCP, and start the screen saver. It usualy takes a few dozen minutes, and pppoa can block even if there is no activity on the server apart from the screensaver (ie, there is no user at all on both machines). Tests done : - upgraded from eagle-usb-1.9.9.1 to eagle-usb-2.0.0rc3, without changes. - suppressed NAT and routing (so -that other machines on the network can non longer access the Internet), without changes. - use of SysReq while blocked : unable to get a stack trace from pppoa : I can take note of the actual message if it can be usefull. - suppressed the firewall : I did it yesterday. The machine runned 4 hours without blocking, but more time would be needed to draw a definitive conclusion from this. So, the open questions seems to be : - is the probleme related to a modification from 1.9.8 to 1.9.9? - is the problem related to the kernel udpate? - is the probleme related to the firewall (iptables). In this case, is it related to some specific rule? NB : for personnal replies, native french speaking people can write in french... -- Gilles BERGER SABBATEL LSR-IMAG, BP 72 - 38402 SAINT MARTIN D'HERES CEDEX - FRANCE Tel: (33) 04 76 82 72 68 Fax : (33) 04 76 82 72 87 http://www-lsr.imag.fr/Les.Personnes/Gilles.Berger-Sabbatel
smime.p7s
Description: S/MIME cryptographic signature
