[nylug-talk] Re :Re: extfs journalling percentage (Chris Knadle)

[Chris Knadle]

On Tuesday 27 May 2008, Michael Bubb wrote:
> > From: Chris Knadle <Chris.Knadle at coredump.us>
...
> >> If I understand correctly the actual journal is not that big. Most of
> >> this 5% is reserved so that a superuser has access to  an otherwise
> >> full partition. Do you need 5% for this? (ie if you have a 250G
> >> parttion do you need 13G reserved).

...
> >   I'm not 100% positive, but I don't think the ext3 journal counts in
> > that 5% reserved for root-only usage.  Mainly I think that this is space
> > set aside for root for having a place to do emergency filesystem repair,
> > like using debugfs or similar utilities.

...
> Thank you Chris.
> Now I am intruiged as to how to use this admin space in an actual
> crash.  Guess I will try and google more on the purpose for this.

   I've only had to rescue an ext3 partition once.  It happened because of an 
overloaded circuit breaker due to not monitoring how much power we were 
running through a circuit, repeatedly bringing down half an entire server 
room, including the local mail server -- which corrupted the /var partition 
that was using ext3.  In this case any attempt at mounting the /var partition 
on the mail server would cause the kernel to hang hard without even a kernel 
panic as to why.  We assumed that the disk had been in the middle of a write 
operation (logs) when the power dropped and thus did something unexpected to 
the disk.  We were forced to use debugfs to glean what information from the 
unmounted /var partition that we could, onto another partition.  Hitting the 
trouble spot even with debugfs would hang the kernel hard, so we had to be 
careful to avoid that spot... once we "found out" where that was... by trial 
and error.
   We then re-made the ext3 filesystem on the /var partition, mounted it, and 
copied the data that we had acquired back.

   I believe doing the above required booting into a Red Hat 7.2 rescue 
envionment, which by default did not have device names made for the disk that 
had the problem, so we had to use mknod to make those before doing the work.  
So on every "trial and error" miss we had to re-make these devices as well as 
run an fsck on the ext3 partitions that had been mounted when the kernel hung 
during the rescue work.
   These days with udev the missing devices are likely no longer a problem.

   Now, I don't think we used the 5% "admin" space on any of the ext3 
parititions in the above case, however if all of the various filesystems were 
FULL, then that admin space would have been all we had to work with to rescue 
the /var partition.  So I can understand why it's needed even if I haven't 
actually had to use it.  ;-)

   -- Chris

-- 

Chris Knadle
Chris.Knadle at coredump.us