The 2006 Linux Filesystems Workshop (Part II)

Posted Jul 6, 2006 22:01 UTC (Thu) by adobriyan (guest, #30858) [Link]

From: Theodore Ts'o

You've obviously never kept several dozen reiserfs filesystem images
(for use with Xen or User-Mode Linux) on a reiserfs filesystem, and
then had a hardware failure bad enough that the fsck had to try to
rebuild the b-tree, I take it?

From: Hans Reiser

That is fixed in V4.  Until people start to use V4 they should compress
their V3 backup images that they store on V3, or store them on separate
partitions.  I regret that fixing it without a disk format change was
not possible.

Looks like fsck.reiserfs can liberate reiserfs filesystem image out of loopback prison and glue it to the main one.

Posted Jul 20, 2006 13:40 UTC (Thu) by Duncan (guest, #6647) [Link]

You may not see this at this late date, but others will. Said as someone
running a 100% reiserfs system, semi-patiently waiting for reiser4, so I'm
certainly not anti-Hans-Reiser in any way, but as a realist, Hans' style
simply doesn't work well at such conferences.

He's extremely bright, possibly more so than many/most/all regular kernel
or fs hackers. However, like many extremely bright people, he's also
rather less socially skilled than most, particularly among "peers", as
he's used to being so much brighter than anyone else around that no one
can stand up to his force of logic, which means he's used to always
being "right" even when he's not, and to always getting his way virtually
without question, simply because no one else has the ability or resources
to question him on his level. As a result, he simply doesn't have the
social negotiation skills that most of us lessor mortals end up developing
because there's always someone around who can show us up.

When he gets on LKML, flames nearly always ensue, with little good coming
directly out of it. Many kernel hackers now simply refuse to be involved
at all, which is a shame as it has hampered what nearly all involved
realize /could/ be a major advance in Linux filesystems. OTOH, no one
blames them because it ultimately ends up being a matter of personal
sanity and a defense mechanism staying away from him. Go read some of the
exchanges and the insults he has thrown around and tell me you can
honestly disagree.

What has actually happened, the progress reiser4 /has/ made towards
integration, has very often been the result of Hans' employees quietly
working things out with the rest of the Linux community pretty
much /despite/ Hans' fireworks and presence, rather than /because/ of it.

There's also some serious history involved. After reiserfs (reiser3) was
included in the kernel, Hans and Namesys basically took off and abandoned
it. Their argument, not entirely unreasonable, was that it's now stable
and in bugfix mode, and you don't add new features to a stable
version, /particularly/ when the software in question is a filesystem, and
people's data is at stake if those new features add bugs that risk that
data!

OTOH, the kernel is a living/breathing/changing code collection, with
various components not always entirely stable at the same time. As the
kernel changed and the rest of its major filesystems got extended
attribute support, and data=ordered and data=journaled support for
reiserfs matured and was tested enough to be added to the mainline kernel,
Hans Reiser and Namesys was nowhere around to add it, and refused to do so
pointing to the stable thing, regardless of the fact that it was needed to
keep up with the rest of the kernel.

The result was Chris Mason and other kernel hackers had to take up the
slack, and end up maintaining code that had effectively been "dumped" on
them, and that had been accepted before it met the normal kernel coding
conventions, making it far harder to maintain. The result of that is that
the kernel hackers are being FAR harder on reiser4 than the were on
reiserfs, knowing from experience that soon after it gets into the kernel,
Hans Reiser and Namesys may well be off developing reiser5 or whatever the
next big thing is, leaving the kernel maintainers to cope the best they
can. No /wonder/ they are demanding the new code fit the coding
conventions and style of the rest of the kernel, this time! To a man
used to being right, not only because he very often is, but because he has
few peers, few that even come /close/ to being able to challenge him, this
is seems a rather nasty rebuke.

Back to the conference, however. It was Chris Mason that finally
shepherded the data=ordered and data=journaled updates of reiserfs into
the mainline kernel, Chris Mason that handled much of the extended
attribute reiserfs work, and Chris Mason that has been pretty much the
point man on reiserfs, and likely will be the point man on reiser4 as
well, in terms of working with the other kernel filesystems and the rest
of the kernel. Unlike Hans Reiser, he has demonstrated his ability to
work with others, including the rest of the core kernel and file systems
teams.

Thus, even if Hans Reiser would have had a valuable viewpoint and valuable
knowledge to contribute, and that he certainly WOULD have, from a
practical viewpoint, were he to attend the conference, far less would have
likely actually gotten hashed out. Far better that Chris Mason represent
reiserfs/reiser4, and something actually get done, than Hans, and only
fighting and flaming and bitter recriminations result.

Hans is a very gifted man, very good at what he does, developing file
systems, and at least acceptably good running a company (I can't rightly
judge /how/ good, but the company is still in business, and still making
payroll, so it's not /bad/). However, he's the wrong guy to have at a
kernel file systems conference if you want to get anything done. That's
just the way he is. Let's appreciate him and help him contribute where
he's best, and I'm certainly very happy to use his filesystems, but
please, keep him away from those conferences! =8^O

Duncan