On Oct 01, 2001 14:01 +0400, Vladimir V. Saveliev wrote:
> "Jonas Jensen" wrote:
> > It seems to me that I have a problem with my IDE somewhere below reiserfs
> > that needs to be worked out. However, it still seems to be a bug in reiserfs
> > that corrupts my filesystem when it gets confused, instead of just giving up
> > so it would work the next time I remounted the partition.
> >
>
> IMHO, when hardware starts to fail - it is time to think about changing it.
> Reiserfs has not way to know when it should give up. It sends correct data
> to disk, broken hardware writes it wrong. Who did corrupt the data then?
However, since hardware failures DO happen (disks are by far the most
likely to fail) then it is also the responsibility of the filesystem
to detect bad values and not further corrupt the filesystem. This is
what the config value "REISERFS_EXTRA_CHECKS" does, but Hans complains
it slows down the fs too much for distros to ship with it enabled (while
the distros want the customer data to be as safe as possible).
Maybe some of these checks need to be moved outside the CONFIG and be
done all of the time. While it may slow down reiserfs a bit (as small
as possible if only important values are checked only once per access),
it is worth it for the filesystem to turn read-only (or optionally panic)
if corruption is detected, before it damages more of the filesystem.
These checks are doubly important in a tree-based fs, because even a small
error can cause a large amount of corruption, while for a fixed-layout fs
these problems can easily be detected before harm is done.
Note that a small bit of performance loss in the "benchmark war" is far
preferrable to a total loss of user confidence in reiserfs. See the
numerous complaints about the "data is corrupt on a crash" thread, to see
why people are rightfully upset about such problems, regardless of whether
it is in the FAQ or not. A filesystem full of valid metadata is useless
to a user if the user loses their data a minute after they stopped editing
the file. At that point nobody cares how fast the fs is, or whether it
took 2 seconds or 20 minutes to do recovery.
Cheers, Andreas
--
Andreas Dilger \ "If a man ate a pound of pasta and a pound of antipasto,
\ would they cancel out, leaving him still hungry?"
http://www-mddsp.enel.ucalgary.ca/People/adilger/ -- Dogbert
