Re: configure should avoid exec-shield breakage

[Dave Love]

address@hidden (Kim F. Storm) writes:

> David PONCE <address@hidden> writes:
>
>>
>> WDYT of the following (untested) change to also check that exec-shield
>> is enabled.

For what it's worth, I'd have done something like:

case `cat /proc/sys/kernel/exec-shield 2>/dev/null` in
  ""|0) ;;
  *) type setarch 2>/dev/null && SETARCH='setarch i386' ;;
esac

> Does it matter ?

It makes it somewhat less likely to fail on unknown systems, e.g. if
exec-shield is off, and setarch i386 doesn't work.  It's probably not
a big win, though.

> I would guess that setarch is a noop if it is disabled.

No, it calls personality(2), which is why setarch(8) is wrong.  That
may or may not be harmless if `i386' is an allowed arg, which it isn't
on, say, ppc (should that get the same kernel patch).  Actually, it's
not obvious that it's harmless to use `personality(PER_LINUX32)' for
an amd64 64-bit build, so I guess the setarch could break that.
Perhaps someone can check Fedora on amd64.

> And what if exec-shield is enabled _after_ you run configure ?

It's the same as if you clobber anything else configure has checked...