Bug 780643

Same here, as well with new installations of openSUSE 12.2.

happens at boot with ntpd:
ntp[1783]: /etc/rc.status: line 57: /dev/stderr: No such device or address

and sometimes while running (restarting npcd):
npcd[22338]: Shutting down npcd /etc/rc.status: line 57: /dev/stderr: No such device or address

seems that in the start environment /dev/stderr is not available (ntpd in chroot), maybe devtmpfs is not fully populated then.

Conflicting line 57 in /etc/rc.status of package aaa_base-12.2-2.11.1 is

start|stop|reload|restart|try-restart|force-reload|status) echo "redirecting to systemctl" >/dev/stderr ;;

by substituting ">/dev/stderr" with ">&2" everything is fine. See patch:


--- orig/etc/rc.status	2012-11-05 19:15:09.407475863 +0100
+++ fixed/etc/rc.status	2012-11-05 19:15:01.855257145 +0100
@@ -54,7 +54,7 @@ if test -z "$SYSTEMD_NO_WRAP" && /bin/mo
 	    ;;
 	esac
 	case "$1" in
-	    start|stop|reload|restart|try-restart|force-reload|status) echo "redirecting to systemctl" >/dev/stderr ;;
+	    start|stop|reload|restart|try-restart|force-reload|status) echo "redirecting to systemctl" >&2 ;;
 	    *) unset _rc_base ;;
 	esac
 	if test -n "$_rc_base" -a -x /bin/systemctl ; then

and with mysql:

mysql[10912]: Shutting down service MySQL ..done
mysql[10985]: /etc/rc.status: line 57: /dev/stderr: No such device or address
mysql[10985]: Starting service MySQL ..done

/dev/stderr should be a symlink to /dev/fd/2, which in turn is a symlink to /proc/self/fd/2. Example:

$ namei -x /dev/stderr 
f: /dev/stderr
 D /
 D dev
 l stderr -> fd/2
   l fd -> /proc/self/fd
     D /
     D proc
     l self -> 23605
       d 23605
     d fd
   l 2 -> /dev/pts/0
     D /
     D dev
     D pts
     c 0

Please check how does /dev/stderr look like in your case.

I modified the /etc/rc.status script by inserting your

/usr/bin/namei -x /dev/stderr > tmpfile

before the actual command of 

echo "redirecting to systemctl" >/dev/stderr


The output is:
"""
f: /dev/stderr
 D /
 D dev
 l stderr -> fd/2
   l fd -> /proc/self/fd
     D /
     D proc
     l self -> 32502
       d 32502
     d fd
   l 2 -> socket:[1004453]
socket:[1004453] - No such file or directory
"""

Hope this helps.

So /proc/self/fd/2 is a socket even in your shell? What shell is it? A tty login, ssh login, something else?

The above output was generated while using a /bin/bash shell via ssh login on a virtual machine. To rule out an login shell/ssh error I collected more info while rebooting the machine including $SHELL and $TERM. The $1 of rc.status was in both cases "status":


Shutdown:
---- scope: status -- SHELL: /bin/bash -- TERM: dumb ----
f: /dev/stderr
 D /
 D dev
 l stderr -> fd/2
   l fd -> /proc/self/fd
     D /
     D proc
     l self -> 2594
       d 2594
     d fd
   l 2 -> socket:[12470]
socket:[12470] - No such file or directory


Boot:
---- scope: status -- SHELL: /bin/bash -- TERM: dumb ----
f: /dev/stderr
 D /
 D dev
 l stderr -> fd/2
   l fd -> /proc/self/fd
     D /
     D proc
     l self -> 1363
       d 1363
     d fd
   l 2 -> socket:[8960]
socket:[8960] - No such file or directory

This is a kernel bug as there is no such links below /proc/<pid>/fd for sockets pairs.  And as systemd uses socketpairs instead of pseudo ttys this cause the trouble.  A workaround would be that systemd would use pseudo ttys instead of sockets pairs for interactive jobs.  This could be done similar as startpar from SysVinit it does.

*** This bug has been marked as a duplicate of bug 728774 ***