On Wednesday 20 July 2005 02:43, Robin H. Johnson wrote:
> This problem IS fixed in ~arch:
>
> line 190 of both vpopmail-5.4.10.ebuild and
> vpopmail-5.4.9-r2.ebuild: chmod 4711 ${D}${VPOP_HOME}/bin/vchkpw

Ahh okay, that explains things a bit.  I'm using x86, which means 
5.4.6.

> So if this is still a problem in arch, but works in ~arch, you
> SHOULD file a bug report.

Why not just wait for the newer releases to make it to arch?

> However the original reasoning for vchkpw NOT being setuid was
> that setuid is NOT always needed depending on which backend you
> are using.

I can confirm that - bincimap and qmail-pop3d run as root, so the 
setuid bit is not necessary.  I believe this is also the case for 
dovecot 1.0 beta releases, though there are no ebuilds for them so 
I haven't yet tested (<1.0 releases use libvpopmail directly 
instead of the checkpassword interface).  However it is necessary 
for any server running as a non-root user, i.e. qmail-smtpd.

Thus I believe this should have the same treatment as binaries like 
chsh - they won't work for non-root users without the setuid bit, 
but running as a non-root user is generally accepted.  If I want to 
be paranoid (which I am), I can use suidctl (which I do), and only 
uncomment the binary when I discover the need to.  There's not 
really any reverse of suidctl to my awareness.

Nor is there a use flag for qmail or similar on vpopmail, but the 
vpopmail ebuild requires qmail regardless of USE settings (postfix 
support is not present), so at least in the current state, since 
the package is built for qmail, it should assume qmail's non-root 
qmail-smtpd will need to access vchkpw.

I would encourage making vchkpw suid even if postfix is supported 
and used instead of qmail, because there are other softwares (i.e. 
IMAP & POP servers) which have a checkpassword interface which do 
may not run as the root user.

> And as I've mentioned before I'd like MORE reports of packages
> working well before they are moved to stable arch. Without those
> stable working reports I don't have any means to judge just how
> much testing has been done on a package, other than my own use of
> a package (and as such I do leave things longer than the 30 days,
> because I don't entirely trust them).

This sounds like a request for the QA team.  I tend to stay away 
from most ~arch packages simply because most of our systems are 
live production servers, but I'd be happy to test-drive new ebuilds 
of vpopmail if it would help get new versions into the stable tree 
faster.

Cheers,
-- 
Casey Allen Shobe | http://casey.shobe.info
[EMAIL PROTECTED] | cell 425-443-4653
AIM & Yahoo:  SomeLinuxGuy | ICQ:  1494523
SeattleServer.com, Inc. | http://www.seattleserver.com
-- 
[email protected] mailing list

Reply via email to