Migration from the KQ Implementation

Fajar A. Nugraha list at fajar.net
Wed May 4 23:14:53 EDT 2011

On Thu, May 5, 2011 at 8:08 AM, Gordan Bobic <gordan.bobic at gmail.com> wrote:
>> So that's really question...beyond those things what
>> would a ZFS fsck do that wouldn't require a custom recovery job?
> The same things that ext3 fsck does when the journal doesn't ensure a clean
> fs.

We're really getting off-topic here.

Here's my summary/opinion:

Sun/Oracle has made it perfectly clear that they had no intention of
creating fsck-like tool for zfs. There are people who feels the need
for it. if you're one of them, your best bet is to try contacting
Oracle sales people and see how then can help you.

Brian has made it clear that there is no plan to create a
Linux-specific fsck for zfs. IMHO it's a good decision, as it would
take very significant effort to do so (something that even Oracle is
not willing to invest in for Solaris at this point)

Latest version of zfs (in Solaris Express) has been solid enough that
it can recover from most situations that would normally require fsck
for other filesystems (including . If you STILL have unrecoverable
pool when using that version, asking in zfs-discuss at opensolaris.org or
get Oracle support might be the best way to go. When you said
"enlisting the services of one of the original ZFS developers and some
of his custom tools" I suspect the engineer was Victor Latushkin and
the problem is related to this thread:
Such fixes is already incorporated in latest version of zfs.

While most filesystems included in Linux kernel have fcsk, some do not
(or don't have a "fixing" version). Examples include:
- squashfs (a read-only compressed filesystem)
- btrfs (fsck.btrfs is currently not capable of fixing errors)

There are cases where even an fsck will not help (like a raidz1 on ZFS
or raid5 on Linux md with two broken disks), where IMHO the data won't
be recoverable by design.

So to sum it up:
- If you STILL have problems with an unrecoverable pool, trying latest
version of Solaris Express might solve it. If not, asking on
zfs-discuss at opensolaris.org will be a good place
- this list should better be focused on discussing zfs implementation
on Linux, and not about features that aren't even exist in upstream


More information about the zfs-discuss mailing list