amd64/128524: No geom documentation for loading gjournal
joe at zircon.seattle.wa.us
Sat Nov 1 14:00:12 PDT 2008
>Synopsis: No geom documentation for loading gjournal
>Arrival-Date: Sat Nov 01 21:00:11 UTC 2008
>Originator: Joe Kelsey
>Release: FreeBSD 7.1-PRERELEASE amd64
Joseph M. Kelsey
System: FreeBSD zircon.zircon.seattle.wa.us 7.1-PRERELEASE FreeBSD 7.1-PRERELEASE #26: Mon Oct 27 20:08:55 PDT 2008 root at zircon.zircon.seattle.wa.us:/usr/obj/usr/src/sys/ZIRCON amd64
I am trying to use gjournal on two different disk drives and
have been frustrated over the lack of documentation.
To start with, there is no explanation over why certain commands
are necessary. For instance, what does the "gjournal load"
command do? The man page says that the geom command documents
it. However, none of the actual "load" commands are necessary
in the standard 7.x kernel build, since everything is loaded by
However, for some unknowable reason, you actually have to issue
the gjournal load command in order to actually cause the various
.journal devices to magically appear. This is especially true
at boot time. This is in spite of the fact that
geom_journal_load is not documented anywhere, relying on someone
just guessing that it will cause the right thing to happen.
Why do I need to do this? How does this whole thing interact
with labels? Why do I sometimes get label devices with, for
example, labels1a.journal and sometimes label? What sequence of
events causes the label to move from /dev/label/ufs to /dev/ufs
to other palces? How to I actually control it?
Another problem occurs when you want to experiment with
journalling. As soon as you get one method of journalling
started, you can never change it because the journal stuff will
never let go. For instance, I wanted to reformat my disk to try
different ways of setting up partitions on the drive to
effectively support journalling. What are the recommended sizes
to use for the data versus journal provider? Should I do a
newfs on both? I have setteled on 10% for journal and the rest
for data with no actual file system on the journal provider.
However it took a long time to repartition the disk drive since
the stupid journal would not let go of the drive. I had to
boot into single user without loading the journal stuff in order
to actually have disklabel -e work correctly. There needs to be
a reliable way to cause the journalling to let go of the device
in order to actually accomplish something.
Run gjournal without doing gjournal load. Look at the various
things that do not happen correctly.
The latest handbook on my disk is different from the handbook on
freebsd.org. Why is that?
More information about the freebsd-amd64