1 May
2003
1 May
'03
3:49 p.m.
Carl Lowenstein:
Isn't this really true of Unix systems of any age, when doing fsck
on a mounted root file system?
Some middle-elderly BSD systems--4.1 and possibly 4.0--
managed the buffer pool in such a way that the super-block
of a mounted file system was kept in the original buffer,
with device and block number correctly stored in the struct
buf header. Hence if fsck wrote to the block device rather
than the raw one, the super-block came out right even when
checking a mounted file system; in particular there was no
need to reboot.
This convenience was abolished in either 4.2 or 4.3 (I am
travelling right now and cannot check manuals and
sources). I never quite understood why, though I never
looked at the source code in the later systems. The
scheme found in most current systems, in which the
root starts out read-only, is a better idea anyway.
Norman Wilson
Toronto ON (normally)