> I think that a bad "seek", i.e., a read or write beyond the end of the
> physical media, caused Linux to be trashed. Therefore, subsequent writes
> were to __OTHER__ devices, not even "known" to Linux.
This is indeed a problem, and not limited to SCSI devices. I'm using an
IDE disk under DMA, and until a few kernels ago, could try a large
recursive copy from one MSDOS partition to another, and sometimes cause a
read beyond the end of some device or another. The resulting mayhem kills
Linux and often required a power cycle to get the drive talking again.
Was probably a MSDOG bug
Chris.