I've been having problems with my filesystems ever since I installed the first
1.3 kernel. With newer versions, I had a short period with no errors, but it
just started again. Having some time on my hands, I decided to find out what
happened.
A number of inodes, starting at offset 0x4000 ranging to offset 0x43ff, were
overwritten by a piece of a log-file, which normally is written on the same
filesystem (so it wasn't an overrun from the partition in front of the disk).
The system was quite loaded at that time (I was moving, unpacking, pruning,
cvs'ing, and more of that stuff that loads a fs), but apart from that I
haven't got a clue why this block was written wrongly.
As none of the files involved were important, I cleared the inodes and removed
the directory entries with debugfs.
At about the same time, I suddenly found that a backup I created on another
filesystem had all of its blocks cross-linked to a large number of files (51,
to be precise). As this is my most important filesystem, I dismounted it an
ran an fsck immediately - so I have no detailed data here (luckily it was
cross-linked only to object files, so damage was low).
I have all filesystems currently mounted with check=strict and
errors=remount-ro, so I hope that when it happens again I can provide detailed
information (what is exactly needed, anyway?). In the meantime, this is the
first time that I am worrying about data integrity under Linux since
0.99pl14...
My system: 486DX66, 24Mb Ram, NCR53c810 (5 MHz async). The SCSI drives are
fairly new (6 months and 1.5 year) and both haven't added bad blocks to their
list apart from the manufacturing defects - so I really think this is a
software problem. My current Linux is 1.3.68, haven't got the time to upgrade
yet (sorry for that :-)).
Cees
-- Cees de Groot <C.deGroot@inter.NL.net> OpenLink Software, Inc.