linuxrc: do not rely on chroot evaluating PATH

When trying to install a new kernel on catbus.sparc.dev.gentoo.org, the
newly built initrd consistently claimed that it could not find init, and
dropped to the rescue shell.

However, init was there just fine in /newroot (as before).

I dug out the command that is actually run inside linuxrc,

        elif ! chroot "${CHROOT}" test -x /${init#/} 1>/dev/null 2>&1

and tried to run that manually, which led to a rather strange error message

rescueshell / # chroot /newroot test -x /lib/systemd/systemd
chroot: can't execute 'test': File name too long

Some more research led me to the busybox manpage (where here chroot comes
from):
https://busybox.net/downloads/BusyBox.html#chroot

chroot
    chroot NEWROOT [PROG [ARGS]]
    Run PROG with root directory set to NEWROOT

Note, the third argument is *not* a command (as with usual chroot, see the
manpage from coreutils chroot) but a program!

Bug: https://bugs.gentoo.org/842027
Signed-off-by: Andreas K. Hüttel <dilfridge@gentoo.org>

1 files changed, 1 insertions, 1 deletions

diff --git a/defaults/linuxrc b/defaults/linuxrc
index ff08ba2..15fbf7c 100644
--- a/defaults/linuxrc
+++ b/defaults/linuxrc
@@ -1376,7 +1376,7 @@ do
 	if  ! mountpoint "${CHROOT}" 1>/dev/null 2>&1
 	then
 		bad_msg "${CHROOT} is not a mountpoint; Was root device (${REAL_ROOT}) not mounted?"
-	elif ! chroot "${CHROOT}" test -x /${init#/} 1>/dev/null 2>&1
+	elif ! chroot "${CHROOT}" /usr/bin/test -x /${init#/} 1>/dev/null 2>&1
 	then
 		mounted_root_device=$(mountpoint -n /newroot 2>/dev/null | awk '{ print $1 }')
 		bad_msg "init (${init}) not found in mounted root device (${mounted_root_device})!"