kern/121481: data rot on disk

[Bernard Steiner]

>Number:         121481
>Category:       kern
>Synopsis:       data rot on disk
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          change-request
>Submitter-Id:   current-users
>Arrival-Date:   Fri Mar 07 22:20:00 UTC 2008
>Closed-Date:
>Last-Modified:
>Originator:     Bernard Steiner
>Release:        6.3
>Organization:
>Environment:
FreeBSD grimma 6.3-STABLE FreeBSD 6.3-STABLE #16: Wed Mar  5 19:14:26 CET 2008     root at grimma.anydomain.de:/usr/obj/usr/src/sys/GRIMMA  amd64

>Description:
Setup:
I keep around ten thousand digital images from my camera on disk.
The disk in question is a geom mirror (using load balancing) of two
same size partitions (both h) on two slices (both 1) on two identical
ST3500630AS/3.AAK SATA drives connected via standard SATA cables to
the same VIA 6421 SATA150 controller.
The gmirror contains a ufs2 file system and _is_usually_mounted_read-only_.
The only time I mount the file system read-write is when I actually
copy new images to it. 
May I also point out that I
(a) changed to the amd64 platform,
(b) installed the file system mentioned above and
(c) started using GNOME which for some reason or other has crashed (or rendered unusable, as in destroyed console access) my system on a daily basis,
at about the same time.
 
After I copied said images to the file system, I md5 summed them all.
It appears that now, a few weeks later, I get md5 sum mis-matches.
 
Problem:
It is obvious there is some data rot going on somewhere.
It is yet unclear when and how this data rot should occur.
The obvious place might be a hardware problem; alas, the
kernel does not report such a one.
Note that graid3 only provides EIO on parity mis-match, and gmirror does not even provide for that.
I have now implemented a ufs2 on top of a gmirror using two
geli devices both with hmac/md5. 
 
I request to enhance geli with "none" encryption.

I also request for a RAID-6 (or double parity) functionality with possible parity auto-repair geom class to be developed.

>How-To-Repeat:

>Fix:


>Release-Note:
>Audit-Trail:
>Unformatted: