Apparent race in the startup code

jdow jdow at earthlink.net
Thu Oct 13 12:29:28 UTC 2005


It appears that there is a race in the startup code for FedoraCore 4.

On my configuration for ppp0, my PPPoE port, it seems network startup
somehow gets around to "ifup ppp0" time before several nice prerequisite
items are completed. I discovered this by running logger messages to
diagnose what part of my "ip-up.local" script was not running. The
results proved erratic. It appears syslog is not always started as
ppp starts if I have it setup to start on boot. It also appears that
at least some of the require files are unreachable at that time,
although I fail to see why this should be true for the likes of a
custom iptables script in /etc/sysconfig/network-scripts. Nonetheless
when I deferred starting ppp until just before ntpd is started the
startup and iptables configuration took place perfectly. (I built
a special /etc/init.d/ppp0 file setup to link in as a service at
S57ppp0.)

Has anybody else noticed some obscure sequencing problems in this area?
I am not sure if the 2.6.12 series kernels had this problem.

The 2.6.13-1526 kernel surely does, for me. The system logging seemed
to kick in about half way through the sequence of operations I had in
the ip-up.local file. (It kills any fetchmails that may be running,
fires off the iptables script to accommodate the new dynamic address,
it runs the dyndns script, then starts or restarts ntpd, and finishes
with restarting the fetcmails used to get email from the ISP. The logger
seems to kick in about half way through the dyndns activities, most of
the time. And the great python of putridity only knows why something
like "if [ -f /etc/sysconfig/network-scripts/iptables ] ; then" should
fail when the indicated file is most assuredly present and executable
with suitable permissions. Subsequent direct runs of the script work.
And subsequent "ipdown ppp0;ipup ppp0" sequences work just fine.

One hypothesis that comes to mind is that there is a period of time
during which newly mounted ext3fs partitions are mounted read-only to
allow concurrent booting and fsck-ing and I am getting caught as the
root partition is remounted read-write with it in transition during the
instants I am searching for the iptables script. (I have several and
automatically select the most recent. This gives me a roll back position
I rather like.)

{^_^}   This REALLY has me scratching my head over it.




More information about the fedora-list mailing list