slice entries + stripe/mirror

[Allan Fields]

Last night I ran into a few issues.

Slice entries such as both /dev/ad12bs1 and /dev/ad12cs1 when I've
created a FreeBSD slice/partition on provider /dev/ad12 and previously
had /dev/ad12s1[bcd] proper seem error prone.  As a work around I've
simply updated /etc/fstab to use /dev/ad{8,10,12,14}b as swap.  E.g.
gstripe is using consumers /dev/ad{8,10}d which previously were showing
up as /dev/ad{8,10}s1d.  The change occurred when I did
geom_{stripe,mirror}_load="YES" in loader.conf.

I've seen this behaviour for a while now since early 5.x and am still
seeing it in 6.2-RELEASE.  It drives the rc scripts bonkers because they
can't find the correct device entries anymore, so files like /etc/fstab
and /etc/rc.conf have to be updated.  Anytime device entries like
/dev/ad0bs1 start to appear, as a sysadmin/user you might question why.
 ("Can't FreeBSD get it the right way around?", etc.)

The other issue is related to mirrors and setting geom_mirror_load="YES"
in /boot/loader.conf.  I had a dormant or erroneous mirror entry gm0 on
provider ad0 which is my boot disk. (Did I even put that there?)  While
this wasn't an issue when I was configurting mirrors manually, I started
the system up to find a new gm0 mirror that I had obviously forgotten
about now existed, and worse, it auto-activated it in addition to the
other mirror devices removing the dev entries for ad0s1[a-g] but
thankfully keeping the disklabel and creating mirror/gm0s1[a-g].


Just a thought..

The question with on-disk meta-data/auto that can possibly do funky
things to disks if they are plugged in on the wrong channel, etc. is a
concern.  Are there cases where it is possible for an admin to
inadvertently screw their geom config and/or destroy volumes?  Maybe the
user should be (default, optional) prompted first before a stripe or
raid starts to rebuild at boot time?

Prompt user first: I do know is possible with eli password prompt at
boot for instance.  As long as it "just works", there isn't any issue,
but what border-line cases exist, I'd like to know.

Thanks..
  Allan Fields <afields at ncf.ca>