strip(1): will it goof anything?

[Keith Packard]

In article <2886 at sun.uucp> guy at sun.uucp (Guy Harris) writes:
>> One note.  An unmonitored strip, perhaps stuck inside a find, will
>> silently remove setuid bits throughout the find path...be careful.
>
>Or run it as "root".  I can't speak for System V (thanks to COFF, "strip"
>consists of no less than 8 ".c" files, and I'm not about to see how it does
>the stripping), but it may not remove the setuid bits there.
>
>	Guy Harris

It depends on the configuration of the kernel; version 7 and derivatives
have a kernel define (INSECURE) that, if not defined, causes all
set-uid, set-gid bits to be cleared on *any* modification to the file.
Many commercial unix systems (Ultrix for example) do define this flag so
running the find on them will cause the set-uid bits to be cleared, 
even when running as root.  For example, in my 2.9 kernel we have
the little bit of code (from rdwri.c):

writei(ip)
register struct inode *ip;
{
	.
	.
	.

#ifndef	INSECURE
	/*
	 * clear set-uid/gid on any write
	 */
	ip->i_mode &= ~(ISUID|ISGID);
#endif

Not having the 4.2 truncate call I cannot tell if that will as
well clear the bits but I suspect they were overly cautious...

Of course, most hand built kernels do not define INSECURE as it
isn't a problem if you are cautious about certain set-uid programs.

keith packard
...!tektronix!reed!motel6!keith