Strange panic with today's kernel

[Harold Gutch]

On Thu, Jan 08, 2004 at 02:02:42PM +0100, Simon 'corecode' Schubert wrote:
> On 08.01.2004, at 09:19, Matthew Dillon wrote:
> >    the issue is... it's simple:  init crashed because you ripped the
> >    binary out from under it and replaced it with a new binary.  The
> >    old already-running init tried to page something in and BANG!
> 
> shouldn't something in the kernel prevent such things to happen?
> Something like "text file busy" when writing to the file or such...

I believe this only happens if you write to a binary that's
currently running, but NOT if you remove it (after which you then
can create a new file with just that name).


> besides, I thought install creates a tmp file and renames that to the 
> destination. Then init should still exist on disk, until last reference 
> is closed?

Yeah, I also thought that a binary would reference the file on
until the last reference was closed, so in this case it would
still reference a file on the disk for which no directory entry
exists.  This would happen as long as init is running (= until
the next reboot, when the new init binary is loaded).


bye,
  Harold