Vincent Meyer <[EMAIL PROTECTED]> writes:
> Also, funny thing - typing the "sync" command from the console still resulted
> in the not cleanly unmounted message, and some minor errors with fdck the
> next time the system was booted.
which is just normal; sync() only flush dirty buffers, so that any
pending writes get physically commited on disk (data, metadata tables,
...). [whell, there's eide writeback problem, some costly scsi
controller that have cache with their own battery, ...]
but the fs remains opened by the kernel,
when the kernel open an fs (but those who can't store that info like
vfat), it marks it as dirty, so that if you press the reset button,
fsck try to repair it or the journal code try to replay any pending
write operation when the reset occured.
so sync let the fs in its dirty state. if you want it to be in "clean"
state, you've to umount it or remount it ro.
nothing anormal here.
> When I first learned UNIX stuff - like 1985, there was a bug in sync, that
> required that it be executed at least twice in order to be sure the buffers
> synced. Can I assume sometime in the last 15 years this got fixed? was
> AT&T unix / Berkley 4.3 hybrid on a Stride 440
this is irrelevant.
