Greetings,
Environment: Solaris 8, Exim 4.10, Exiscan 4.10-16
I've failed to find this in the list archives.
Most of the time, exiwhat works just fine (AFAICT). However, occasionally
it causes one (or possibly more) of the exim processes to panic. When it
happens to be the listening daemon, this is somewhat embarrassing, though
not until several hours later :-(
Some examples from the paniclog (wrapped for readability) :-
2002-10-17 22:21:18 Cannot open process log file
"/var/spool/exim/spool/exim-process.info":
Permission denied: euid=49992 egid=49992
2002-10-27 11:01:27 Cannot open process log file
"/var/spool/exim/spool/exim-process.info":
Permission denied: euid=49992 egid=49992
2002-10-27 12:04:10 Cannot open process log file
"/var/spool/exim/spool/exim-process.info":
Permission denied: euid=49992 egid=49992
2002-10-27 12:04:10 184NJU-0000eZ-00 Cannot open process log file
"/var/spool/exim/spool/exim-process.info":
Permission denied: euid=49992 egid=49992
2002-10-27 12:04:14 185ZRv-0006mN-00 Cannot open process log file
"/var/spool/exim/spool/exim-process.info":
Permission denied: euid=49992 egid=49992
User 49992 is exim; group 49992 is exim; exim_user is set to exim.
There are (virtually) no local deliveries on the m/c in question, so no
uids other than exim and root should be involved. The spool directory
permissions are
drwxr-xr-x 6 exim exim 512 Oct 29 09:26 spool
and the created file is (normally)
-rw-r----- 1 exim exim 4807 Oct 29 09:26 exim-process.info
My _guess_ is that there is some sort of timing/race problem between
different exim processes trying to create and/or open the file. I've
stared at open_log() in log.c, and it seems theoretically possible that
one process with euid=root could create and own the file, and another
process with euid=exim could fail as above before root has time to
chown/chmod the file. However, I don't know enough to know if the
conditions for this could actually happen in practice.
It goes without saying, of course, that the problem never occurs when I
try to truss() things. Which reinforces my suspicion that it is a timing
thing.
Anyone?
TIA,
Richard Hall
