Wedge system == lost files

[Steve Kargl]

On Thu, Jan 25, 2018 at 05:38:11PM -0800, Kirk McKusick wrote:
> > To: freebsd-current at freebsd.org
> > Subject: Wedge system == lost files
> > 
> > So, my system just wedge itself.  No panic.  No keyboard.
> > No mouse.  No remote login.  Nothing.  Just wedged.
> > 
> > The system is FreeBSD 12.0-CURRENT r326432 Fri Dec 1 2017 amd64.
> > 
> > Imagine my surprise when I rebooted system and the file
> > I had been editing (and saving after every change) is
> > gone.  /usr/lost+found is empty.  The file pl19-21.tex is
> > gone along with a few files from latex.
> > 
> > -rw-r--r--  1 kargl  kargl  -      0 Jan 25 14:53 pl19-21.aux
> > -rw-r--r--  1 kargl  kargl  -      0 Jan 25 14:53 pl19-21.log
> > -rw-r--r--  1 kargl  kargl  -      0 Jan 25 14:53 pl19-21.out
> > -rw-r--r--  1 kargl  kargl  -      0 Jan 25 14:53 pl19-21.pdf
> > -rw-r--r--  1 kargl  kargl  -      0 Jan 25 14:52 pl19-21.tex
> > 
> > /dev/ada0p5 on /usr (ufs, local, journaled soft-updates)
> > 
> > Are there known issues with SUJ, again? 
> > 
> 
> I have not seen any recent issues with SUJ other than a complaint
> that it corrupted cylinder-group check hashes (which has been fixed).
> 
> A hanging system as you describe can occur if your system disk
> locks up, though the machine generally does remain pingable. But
> it would explain missing files if the system had not been able to
> flush them out (the writes simply accumulate in memory). But most
> editors do an fsync after each file is written which would have
> hung if you could not write to the disk.
> 
> So, regretably, I do not have an explanation. If you are suspicious
> about SUJ, you can disable it using:
> 
> 	tunefs -j disable /dev/ada0p5
> 
> That will leave soft updates enabled so will not cause a slowdown
> in your disk throughput.
> 

Thanks for the quick reply.

I suppose I'm surprise that pl19-21.tex is completely gone.
It was sitting on disk for awhile before I started editing it.  
Your explanation of the buffer cache not getting written to
disk probably explains why the other files are gone.  I'll
chalk it up as a burp with my system.   

-- 
Steve