unsigned long geek = random();

A few months ago while working on a Linux system using XFS, I came across an interesting “feature”. When moving a directory such that its owner changed, the moved directory’s mtime was changed to the current date and time.

For example:

[0] mj@majestic:~/tmp$ mount |grep home
/dev/mapper/data-home on /home type xfs (rw)
[0] mj@majestic:~/tmp$ mkdir test
[0] mj@majestic:~/tmp$ ls -ld test
drwxr-sr-x 2 mj mj 6 Jun 18 15:28 test
[0] mj@majestic:~/tmp$ touch -t 200801011530 test
[0] mj@majestic:~/tmp$ ls -ld test
drwxr-sr-x 2 mj mj 6 Jan 1 15:30 test
[0] mj@majestic:~/tmp$ stat test
File: `test’
Size: 6 Blocks: 0 IO Block: 4096 directory
Device: fd00h/64768d Inode: 951267331 Links: 2
Access: (2755/drwxr-sr-x) Uid: ( 1000/ mj) Gid: ( 1000/ mj)
Access: 2008-01-01 15:30:00.000000000 +0000
Modify: 2008-01-01 15:30:00.000000000 +0000
Change: 2008-06-18 15:29:08.173750666 +0100
[0] mj@majestic:~/tmp$ mv test test1
[0] mj@majestic:~/tmp$ ls -ld test1
drwxr-sr-x 2 mj mj 6 Jan 1 15:30 test1
[0] mj@majestic:~/tmp$ mv test1 ..
[0] mj@majestic:~/tmp$ ls -ld ../test1
drwxr-sr-x 2 mj mj 6 Jun 18 15:30 ../test1
File: `../test1′
Size: 6 Blocks: 0 IO Block: 4096 directory
Device: fd00h/64768d Inode: 951267331 Links: 2
Access: (2755/drwxr-sr-x) Uid: ( 1000/ mj) Gid: ( 1000/ mj)
Access: 2008-01-01 15:30:00.000000000 +0000
Modify: 2008-06-18 15:30:02.814078187 +0100
Change: 2008-06-18 15:30:02.814078187 +0100


I’d never seen this happen before, so I tried to reproduce the behaviour on systems using ext3, UFS and HFS+ filesystems. None of them updated the moved directory’s mtime when the parent directory changed, which is what one would expect. XFS does have an active mailing list, so I reported my findings and was quickly provided with a patch that fixed the problem. Kudos to the XFS developers for providing a fix so quickly.

Rather odd that no-one had spotted this behaviour in the past though…

pkgsrc doesn’t officially support re-bootstrapping once it’s been installed (ie updating the core packages by doing a fresh bootstrap). There is a way to force one by removing a few core packages first though:
pkg_delete -ff bootstrap-mk-files bmake tnftp pax pkg_install
cd /usr/pkgsrc/bootstrap
./bootstrap


As long as the fresh bootstrap is done using the same parameters as used originally (prefix, database directory, etc), everything should work as before.

I tested this a short while ago on an OS X 10.5 system using a pkgsrc -current tree and it worked perfectly.

In the first part of this short series, I detailed the reasoning behind my need for a new mail server. In this second part, I’m going to detail my mail architecture as well as the software choices I made and why.

All mail for my various domains is delivered to the primary MX, which is a hosted server sitting in the US running Debian GNU/Linux (unfortunately my hosting provider doesn’t support NetBSD). It runs postfix and makes use of a variety of checks within postfix itself as well as postgrey for greylisting. I use a fairly conservative list of RBLs and, in conjunction with greylisting, they stop most spam from being accepted. Why postfix? Well, I stopped using sendmail over ten years ago, and although I’ve had good results with Exim in the past, these days I’m just most comfortable with postfix and it suits my needs perfectly.

The RBLs I use:

• zen.spamhaus.org
• cbl.abuseat.org
• list.dsbl.org

Once mail has been received by my primary MX, it is delivered to local mailboxes, one per user. None of the users read their mail from the US server, however. All the mail is downloaded to the local mail server via SSL-secured POP3 and accessed here, either locally or via IMAP. The local mail server is a Sun Ultra 2 running NetBSD/sparc64 3.1_STABLE.

Software I’m using on the local mail server:

• postfix
• dovecot for IMAP (over SSL) access. There are a number of IMAP/POP3 servers available, but I chose Dovecot because of its clean design, good security record and flexible support for mail storage, amongst other things.
• amavisd-new with spamassasin (with razor and Bayesian filtering enabled) and clamav for content filtering
• getmail to download mail from the US server
• mailgraph for simple reporting

All of the above are available in pkgsrc. As I have already done any RBL-based checks on the MX, I don’t do any of them locally.

Although FreeBSD is a platform supported by pkgsrc, only i386 supported is explicitly mentioned and the most recent bootstrap binaries are for FreeBSD/i386 5.3. Being the adventurous chap I am, I decided to bootstrap from source on my dual CPU Ultra 60 running FreeBSD/sparc64 6.2. In true pkgsrc style, it Just Worked. Since bootstrapping, I’ve build a number of fairly large packages without problem: zsh, ncurses, vim, postfix, perl.

Why pkgsrc and not ports on a FreeBSD system? Well, I have a finely tuned pkgsrc environment that builds packages with my set of defaults (MIT Kerberos support, for example), so I decided to stick with what works well for me.

I’ve uploaded the slides from last night’s *BSD talk to the CLUG wiki - grab ‘em here. Content licensed under a Creative Commons Attribution Share Alike license.

I’ve been network booting SPARC systems for a while now, ever since my last run in with a faulty floppy drive on a SPARCstation 2. NetBSD makes it easy - the standard installation includes a diskless client filesystem which can simply be extracted onto the boot server. It wasn’t quite so straightforward with FreeBSD, so here are a few pointers…

My boot server is running NetBSD 3.1 and I booted FreeBSD 6.2, so if you’re using different software you may need to make some adjustments. YMMV.

1. On the boot server, configure rarpd(8) as usual, adding the entry for your machine’s MAC address to /etc/ethers. For example (for a machine called test02):
   08:00:20:b2:2f:b6 test02
2. Extract the FreeBSD base fileset to the appropriate location on your boot server (for example, /export/install/fb62_sp64).
3. Extract the FreeBSD GENERIC kernel fileset to boot/ in your diskless filesystem.
4. Within your diskless root, symlink boot/GENERIC to boot/kernel (boot/GENERIC is a directory that contains the kernel and its modules).
5. Put boot/loaders from the diskless filesystem into your tftp root directory and symlink it to your machine’s IP address in hex. For example, the filename for 192.168.1.92 is C0A8015C.
6. Export your diskless root filesystem via NFS and add the necessary dhcpd.conf stanza. For example:
host test02.pimp.org.za {
hardware ethernet 08:00:20:b2:2f:b6;
fixed-address 192.168.1.92;
option host-name "test02";
option root-path "/export/install/fb62_sp64";
}
7. Boot your machine - “boot net” from the PROM should do it.

A few tips:

• It’s normally a good idea to update the machine’s OpenBoot PROM to the latest release. Old PROMs often have subtle bugs.
• Extracting FreeBSD filesets is simple:
cat 6.2-RELEASE/base/base.* > /tmp/base.tar.gz
tar -xzvpf /tmp/base.tar.gz -C /export/install/fb62_sp4


Update: Fixed incorrect command to extract sets (thanks John Messenger!)

Just a quick reminder about the topic of tonight’s CLUG talk - BSD Unix, given by, er, yours truly. The talk starts at 6:30pm, at the UCT Chemical Engineering Lecture Theatre.

Thanks to a lead from Mark, I now have even more Sun Ultra 60s - another five, for the princely total of R450 (about $60). They are of varying hardware specification, but I have 1×300Mhz CPU, 2×360Mhz CPUs and 2×450Mhz CPUs, somewhere in the region of 2.5GiB RAM and a few 4 and 9GiB disks. All the machines have dual width Elite3D framebuffers. All in all, a pretty good deal

My plan is to put together at least two dual CPU machines, one running FreeBSD and another probably running OpenSolaris. *sigh* If only NetBSD supported SMP on 64-bit SPARC systems.