Here are the symptoms:
1. Cron takes a LONG time to launch jobs. We're talking 5-10 minutes EACH
time a cron job is spawned. The time increases geometrically as the
user's location in the password file increases. That is, a root (first
entry) cron job starts immediately. User ID 6000 takes 10 minutes.
Needless to say, this makes cron useless for 80-90% or the users, because
if they start up a job say every 15 minutes, the system bogs down until
it crashes.
2. Sendmail verify takes a LONG time whenever it gets a bad username. We're
talking 30 minutes here. HP has actually patched this with a special
sendmail for us, but it seems relevant.
3. Finger fails with "fork failed: no more processes" whenever someone fingers
the machine from a remote site. Local fingers work, UNLESS you are root.
This is my favorite problem, because I think it demonstrates the root of
the problem, which is that the routine getpwent in HP-UX 8 and 9 has a
BAD bug that only comes into full effect when you are on a trusted system
(i.e. /.secure/etc/password exists AND is readable by the process). That's
why I think finger bombs as root.
Anyway, HP had promised to fix all of these problems at HP-UX 9.0, but they
let us down on 1 and 3. In fact, they let us down on 2 as well, in that we
are still using the patched sendmail from HP-UX 8.
It should be obvious that I know only enough Unix to make me dangerous.
I would like to be able to say that I wish no one else has these problems,
but I'll say instead that I hope someone else has figured out how to fix them.
Standard disclaimer applies here: These views are mine alone.
--
* Noel Hunter, Academic Systems Administrator, Wake Forest University *
* no...@wfu.edu TEL:(919) 759-5812 MAIL:PO Box 7408 Winston-Salem, NC 27109 *
We have this problem too. Around 5000 accounts. Mater of fact,
sendmail takes forever to vrfy, period. Valid username or not.
>3. Finger fails with "fork failed: no more processes" whenever someone fingers
> the machine from a remote site. Local fingers work, UNLESS you are root.
> This is my favorite problem, because I think it demonstrates the root of
> the problem, which is that the routine getpwent in HP-UX 8 and 9 has a
> BAD bug that only comes into full effect when you are on a trusted system
> (i.e. /.secure/etc/password exists AND is readable by the process). That's
> why I think finger bombs as root.
This would go with my working theory as well.
-rob
---
Rob Francis Unix Systems Specialist, University Computing Services
Indiana University, fran...@indiana.edu (812)855-0327
Root cron jobs take 15 seconds on my 8.x 8x7 before the child is spawned.
That's bad. Since I start maybe 4 cron jobs every 5 minutes, that means
20% of the cpu is taken up starting cron jobs - which are themselves
incredibly trivial. I finally combined all 4 jobs into one shell script
and spawn one job that does 4 things.
>
>2. Sendmail verify takes a LONG time whenever it gets a bad username. We're
> talking 30 minutes here. HP has actually patched this with a special
> sendmail for us, but it seems relevant.
That's fairly serious. sendmail v8 takes less than a second to find a bad
username (1000 passwd entries)
>
>3. Finger fails with "fork failed: no more processes" whenever someone fingers
> the machine from a remote site. Local fingers work, UNLESS you are root.
> This is my favorite problem, because I think it demonstrates the root of
> the problem, which is that the routine getpwent in HP-UX 8 and 9 has a
> BAD bug that only comes into full effect when you are on a trusted system
> (i.e. /.secure/etc/password exists AND is readable by the process). That's
> why I think finger bombs as root.
>
finger as root on the 8.0 8x7 with /.secure/etc/passwd does not show this
problem.
HP now tells me that they have a beta versoin of a patch to address the
getpwent performance problems. More to follow... If anyone has tried the
beta patch, could you provide an evaluation?
>: In article <2c9fds$9...@quad.wfunet.wfu.edu>,
>: Noel Hunter <no...@ac.wfu.edu> wrote:
>: >After two years of wrnagling with HP over problems which appear to be the
>: >result of our "huge" password file (I never thought 6,000 was that big, but
>HP now tells me that they have a beta versoin of a patch to address the
>getpwent performance problems. More to follow... If anyone has tried the
>beta patch, could you provide an evaluation?
Would anyone please advise where I can get this "beta" version patch ?
Thanks a lot.
Regards,
--
S.T. Wong | BITNET: A86...@CUCSC.BITNET
Computer Services Centre | Internet: st-...@cuhk.hk
The Chinese University of Hong Kong | Tel. No: (852) 609 8825
Shatin, N.T., Hong Kong | FAX No: (852) 603 5001