patches for PR 41813 - Real EIO being returned to userland as EFAULT

[Chuck Silvers <chuq%chuq.com@localhost>]

hi folks,

I've put a diff with fixes for this problem on all the platforms for which
it exists in ftp://ftp.netbsd.org/pub/NetBSD/misc/chs/onfault/diff.onfault.*.

I've tested it on these platforms:

. alpha
. hp300
. hppa
. mips (pmax)
. powerpc/ibm4xx (explora)
. sh3 (hpcsh)
. sparc
. sparc64
. sun3
. vax

but there are a bunch of others (mostly m68k) that could use a test.
a few platforms needed no changes (x86, arm, powerpc/oea).

I tested using the diff.onfault-debug.* patch in the same directory,
following these steps on a kernel with both patches applied:

cat /etc/passwd
 <change chuq_uf to 1 using ddb or the mychroot.c hack if the platform
  doesn't have a way to enter ddb from the console.>
cat /etc/daily
 <see that cat shows ECHILD as the error.>
reboot
 <see that reboot also fails with ECHILD.>
 <change chuq_uf back to 0.>
cat /etc/daily
 <see that cat works>
reboot
 <see that reboot works>


there are also two more test programs in that directory
that exercise various error cases for copyinstr() and copyoutstr().

I'll commit all this in a few days if no one finds anything wrong.

-Chuck