Re: [9fans] making better use of ext2fs?

Re: [9fans] making better use of ext2fs?

[Jonathan Sergent]

>apart from not sync'ing on shutdown

Even there, it's not been to bad to me--rolling blackouts, ^P.  I did
end up with some strange corruption once when this happened, but wrap
made it easy to figure out which system files got busted, and it
hasn't happened since then.

OTOH I suffered through many instances of having to suffer through
e2fsck and losing data when Linux ran on the same machine.  (And I
couldn't use the packaging database to check for what was busted,
because it turned into a subdirectory after the fsck was done.)

Am I confused or is the problem with data corruption on unclean
shutdown with kfs the same as with ext2, namely that there aren't
synchronous metadata updates?

Re: [9fans] making better use of ext2fs?

[forsyth]

>>Am I confused or is the problem with data corruption on unclean
>>shutdown with kfs the same as with ext2, namely that there aren't
>>synchronous metadata updates?

disk/kfs updates most metadata synchronously (look
for localfs and Bimm), which
might be why some people complain about its performance.
the file server kernel is (just) slightly more relaxed about it.

that reduces the scrambling, but doesn't ensure that
the data blocks from the last file you wrote have been
written out, hence the need to sync.

Re: [9fans] making better use of ext2fs?

[forsyth]

>>One thing I've noticed is that people report kfs sometimes gets flaky
>>when used in a stand-alone configuration.

apart from not sync'ing on shutdown, i suspect many of the problems people had with
kfs were mainly caused by bugs elsewhere in the kernel (eg, the ones that
caused trouble when pages were paged out).
perhaps kfs deserves a bad name for other reasons, but i don't think this
is one of them.  it's worth noting that if disk/kfs is buggy, it's quite
serious, because the main file server uses similar code.
i've been running various versions of kfs on my thinkpads for
more than five years.

[9fans] making better use of ext2fs?

[Aharon Robbins]

Greetings 9fans,

I've been following 9fans for quite a while over the years, and pretty
steadily since the 3rd release.

One thing I've noticed is that people report kfs sometimes gets flaky
when used in a stand-alone configuration.

So, in a thinking-out-loud sort of mode, I'm wondering if maybe it
might be worth investigating use of ext2fs instead of kfs for that
purpose.  This has the non-trivial advantage that, for those 9people
who also use Linux, recovery tools are available just by rebooting.
It also has the potential to make install/bootstrapping easier:

	fdisk	# set up a separate ext2 partition
	newfs /dev/xxx	# put a filesystem on it
	mount /dev/xxx /mnt
	cd /mnt ; tar -xvpzf p9tarball.tgz
	# edit various config files using whatever editor
	# you have under Linux that you like
	umount /mnt
	dd if=p9bootblock of=/dev/xxx bs=512 count=1

	# edit lilo config
	/sbin/lilo
	shutdown -r now

This is, obviously, just an outline of the steps, and there's obviously
a ``non-trivial'' amount of work involved.  OTOH, if making Plan 9 easier
to install and use is an acknowledged---if secondary---goal, then perhaps
this is worth considering?

The reason I suggest it is that the ext2fs server already exists and works
(for some definition of "works" :-), whereas a bsdfs server would probably
have to be written from scratch --- I'm not trying to start a Linux
vs. *BSD (flame-)war here.

In other words, this is a way to provide an additional bootstrapping
mechanism; I'm not suggesting that this replace the current install.

I almost didn't post this, because I hate it when I'm on the receiving
end of "here's a great idea, why don't you do the work?".  OTOH, I
wanted to offer the thought.  And if someone can run with it, great.
If not, maybe one day I'll be able to, and in the meantime we're no
worse off than we are now.

Thanks,

Arnold Robbins
arnold@skeeve.com