Unexpected soft update inconsistency

[Uwe Doering]

Peter Schuller wrote:
> Hello,
> 
> What's the deal with soft updates and guaranteed consistency? Every time 
> journaling is brought up by someone, he/she is promptly told about how soft 
> updates does the job at least as well.  I never had a problem with this based 
> on what I have read about soft updates. However:
> 
> I *very* quickly ran into a case where I got an "unexpected soft update 
> inconsistency" after crashing the machine by doing something naughty with 
> Vinum while there was disk activity (note: the filesystem which exhibited the 
> problem was not on a vinum volume).
> 
> So my question is:
> 
> Do soft updates, or do they not, algorithmically guarantee filesystem 
> meta-data consistency in the event of a crash?

The design goal for Soft Updates, apart from the performance gain, was 
to keep the disk image in a recoverable state at all times and to limit 
data loss to the last couple of seconds.  This does not mean, however, 
that it guarantees that after a crash fsck(8) will never ask any questions.

Normally, the right thing to do is to answer with 'yes' when it offers 
to remove the file, since that file can be expected to be incomplete. 
This way you are kind of turning back the clock by a couple of seconds, 
from the file system's point of view, until it gets into the time range 
again where all files were still consistent (payload and meta data).

    Uwe
-- 
Uwe Doering         |  EscapeBox - Managed On-Demand UNIX Servers
gemini at geminix.org  |  http://www.escapebox.net