|
Bizarre Solbourne Corruption
When we first got our new Solbourne, we were seeing bizarre file system
corruption in /usr/tmp (e.g. gigabyte files, block devices,
partially-allocated i-nodes) and infrequent but odd panics ("rwip type",
"Memory address alignment", "already doing softints on this cpu"). Other
than that, the system has been stable, solid and quite fast. Finally,
one night after one of these odd panics, fsck discovered a smashed
primary superblock in /usr/tmp, but even "fsck -b 32" wouldn't fix the
smashed superblock. Barry and I happened to be the only ones in the
office, so we resorted to patching the superblock by hand; in the
process, we had to look at the disk geometry and happened to look at the
disk partition table for the drive that contains /usr/tmp. Barry noticed
that /usr/tmp was a 130-cylinder partition, but the next partition
started only 103 cylinders after the start of /usr/tmp; we had
overlapping file systems, which we quickly fixed, and the machine has
been fine ever since.
It turns out that our system administrator is more severely dyslexic than
I had realised, and had read "130" and "103" as the same number.
Normally some one else would have checked his arithmetic, but in the
hurry to get the new machine going, somehow no one did check. I had
known that he had severe trouble spelling; we once got a postcard from
him containing three one-syllable words and our receptionist shouted `We
got a postcard from Spike, and there are *no* spelling mistakes!'.
|