Some problems

Some problems

[Dipl.-Ing. Michael Niederle]

Hi!

I continue testing nilfs as a root file system. Until now I have no script to
cleanly unmount the root file system. (In the layered approach (aufs) I used
before this was not necessary.) So I just called sync and then shut down the
system.

The next boot took some time, because the superblock was "broken". Here an
extract from the kernel log:

[    1.407323] NILFS warning: broken superblock. using spare superblock.
[    1.408569] NILFS warning: broken superblock. using spare superblock.
[  461.587031] segctord starting. Construction interval = 5 seconds, CP
frequency < 30 seconds [  461.591873] NILFS warning: mounting unchecked fs
[  461.629865] NILFS: recovery complete.

Is nilfs doing a sequential scan over the whole partition to find its spare
superblock?

Then I wanted to remove some snapshots:

> lscp -s
                 CNO        DATE     TIME  MODE  FLG   NBLKINC       ICNT
                 685  2009-05-13 09:38:20   ss    -         35     568722
                1330  2009-05-13 10:59:41   ss    i       7629     579584
                4466  2009-05-14 19:15:28   ss    -         23     589270
                8250  2009-05-16 03:40:42   ss    i       6202     597141
                9910  2009-05-17 19:14:54   ss    -         49     626715
               10380  2009-05-17 20:21:53   ss    -         18     647228
               10620  2009-05-23 21:45:41   ss    i       6054     646978

>rmcp 685
rmcp: 685: operation not permitted

What's the cause of this message? (rmcp also failed for all other snapshots;
the failed rmcp did not produce any kernel message)

Greetings, Michael

Re: Some problems

[Ryusuke Konishi]

Hi!,
On Sat, 23 May 2009 22:11:20 +0200, "Dipl.-Ing. Michael Niederle" wrote:
> Hi!
> 
> I continue testing nilfs as a root file system. Until now I have no script to
> cleanly unmount the root file system. (In the layered approach (aufs) I used
> before this was not necessary.) So I just called sync and then shut down the
> system.

I made a Debian lenny on my pendrive.  It looks remount my root nilfs
partition readonly on shutdown, and no trouble so far.

> The next boot took some time, because the superblock was "broken". Here an
> extract from the kernel log:
> 
> [    1.407323] NILFS warning: broken superblock. using spare superblock.
> [    1.408569] NILFS warning: broken superblock. using spare superblock.
> [  461.587031] segctord starting. Construction interval = 5 seconds, CP
> frequency < 30 seconds [  461.591873] NILFS warning: mounting unchecked fs
> [  461.629865] NILFS: recovery complete.
> 
> Is nilfs doing a sequential scan over the whole partition to find its spare
> superblock?

No, the location of spare superblock is determined by size of the divice.
It's in tail of the partition.

I think the spare superblock pointed to a certain old segment and it
took some time to scan forward from the old segment to the latest
segment.

It may take about a minute, but something is wrong if it took several
minutes.

> Then I wanted to remove some snapshots:
> 
> > lscp -s
>                  CNO        DATE     TIME  MODE  FLG   NBLKINC       ICNT
>                  685  2009-05-13 09:38:20   ss    -         35     568722
>                 1330  2009-05-13 10:59:41   ss    i       7629     579584
>                 4466  2009-05-14 19:15:28   ss    -         23     589270
>                 8250  2009-05-16 03:40:42   ss    i       6202     597141
>                 9910  2009-05-17 19:14:54   ss    -         49     626715
>                10380  2009-05-17 20:21:53   ss    -         18     647228
>                10620  2009-05-23 21:45:41   ss    i       6054     646978
> 
> >rmcp 685
> rmcp: 685: operation not permitted
> 
> What's the cause of this message? (rmcp also failed for all other snapshots;
> the failed rmcp did not produce any kernel message)

rmcp is allowed only for plain checkpoints.
So, you have to change it to the plain checkpoint like

 # chcp cp 685 1330 ...
 # rmcp 685 ...

Actually, you don't have to do the rmcp because the garbage collection
will delete plain checkpoints.

I remember someone asked a simliar question before.

I think we should make the error message more user-friendly.  And we
should allow force option of rmcp to remove snapshots; the current
force option does not have this meaning.

Cheers,
Ryusuke Konishi

Re: Some problems

[Ryusuke Konishi]

On Sun, 24 May 2009 17:51:01 +0900 (JST), Ryusuke Konishi wrote:
> > >rmcp 685
> > rmcp: 685: operation not permitted
> > 
> > What's the cause of this message? (rmcp also failed for all other snapshots;
> > the failed rmcp did not produce any kernel message)
> 
> rmcp is allowed only for plain checkpoints.
> So, you have to change it to the plain checkpoint like
> 
>  # chcp cp 685 1330 ...
>  # rmcp 685 ...
> 
> Actually, you don't have to do the rmcp because the garbage collection
> will delete plain checkpoints.
> 
> I remember someone asked a simliar question before.
> 
> I think we should make the error message more user-friendly.  And we
> should allow force option of rmcp to remove snapshots; the current
> force option does not have this meaning.

I found the current error code against the removal for snapshots is
badly assigned, and rmcp cannot distinguish it from real permission
errors.

So, first I'm thinking to change the error code to EBUSY or something.

Regards,
Ryusuke Konishi