[TUHS] v5 and v6 kernel is mode 777

Noel Chiappa jnc at mercury.lcs.mit.edu
Mon Feb 23 06:19:45 AEST 2015


    > From: Dave Horsfall <dave at horsfall.org>

    >> Once UNIX has booted, it has no use for 'unix' (or whatever file it
    >> booted from)

    > Didn't "ps" try and read its symbol table?

Sorry, meant 'UNIX the monolithic kernel'; yes, ps and siblings (e.g. iostat)
need to get the running system's symbol table.


    > I had fun days when I booted, say, "/unix.new", and "ps" wouldn't
    > sodding work...

Know that feeling! I added the following to one of the kernel data files:

    char *endsys &end;

and then in programs which grab the system's symbol table, I have an nlist()
entry:

  "_endsys",

with the follwing code:

  /* Check that the namelist applies to the current system.
   */

  checknms(symfile)
  char	*symfile;

  {	char	*chkloc, *chkval;

	if (nl[0].type == 0)
		cerror("No namelist\n");

	chkloc = nl[ENDSYS].value;
	chkval = rdloc(chkloc);

	if (chkval != nl[END].value) {
		cerror("Symbol table in %s doesn't match running system\n",
		       symfile);
		}
  }

on the theory that pretty much any change at all is going to result in a
change in the system's size (and thus the address of 'end').

Although in a split I/D system, this may not be true (you could change the
code, and have the data+BSS remain the same size); I should probably check
the location of 'etext' as well...

Anyway, a rebuilt system may result in the address of 'endsys' changing, and
thus the rdloc() won't return the contents of the running system's 'endsys',
but the chances of an essentially-random fetch being the same as the value of
'end' in /unix are pretty slim, I would say...

	Noel



More information about the TUHS mailing list