delay in spawning new getty after logging out on console

[Matt Zimmerman]

Here's what is logged to syslog during this time:

Jul 30 00:25:36 localhost hal.hotplug[23486]: waiting for vc info at /class/vc/vcs2
Jul 30 00:25:36 localhost hal.hotplug[23486]: Dont know how to wait for vc at
/class/vc/vcs2; sleeping 1000 ms
Jul 30 00:25:36 localhost hal.hotplug[23492]: waiting for vc info at /class/vc/vcsa2
Jul 30 00:25:36 localhost hal.hotplug[23492]: Dont know how to wait for vc at
/class/vc/vcsa2; sleeping 1000 ms
Jul 30 00:25:39 localhost udev[23495]: removing device node '/dev/vcs2'
Jul 30 00:25:39 localhost udev[23497]: removing device node '/dev/vcsa2'
Jul 30 00:25:40 localhost udev[23503]: creating device node '/dev/vcs2'
Jul 30 00:25:40 localhost udev[23504]: creating device node '/dev/vcsa2'
Jul 30 00:25:46 localhost hal.hotplug[23527]: waiting for vc info at /class/vc/vcs2
Jul 30 00:25:46 localhost hal.hotplug[23527]: Dont know how to wait for vc at
/class/vc/vcs2; sleeping 1000 ms
Jul 30 00:25:46 localhost hal.hotplug[23533]: waiting for vc info at /class/vc/vcsa2
Jul 30 00:25:46 localhost hal.hotplug[23533]: Dont know how to wait for vc at
/class/vc/vcsa2; sleeping 1000 ms
Jul 30 00:25:49 localhost udev[23537]: removing device node '/dev/vcsa2'
Jul 30 00:25:49 localhost udev[23536]: removing device node '/dev/vcs2'
Jul 30 00:25:50 localhost udev[23544]: creating device node '/dev/vcsa2'
Jul 30 00:25:50 localhost udev[23545]: creating device node '/dev/vcs2'

[Jeff Waugh (jdub)]

See the thread, but this is the mail you want, with PATCH LOVING!

http://freedesktop.org/pipermail/hal/2004-July/000623.html

[Matt Zimmerman (mdz)]

It turns out that the cause is actually this:

Aug 13 18:00:31 localhost getty[23272]: /dev/tty2: already in use

Which causes getty to sleep for a full 10 seconds. Reassigning to lamont as
util-linux maintainer

[Matt Zimmerman (mdz)]

This is actually starting to look like a race condition in udev. Here's my
hypothesis for the sequence of events:

- getty starts up
- kernel vc device is created
- hotplug add event is emitted, which invokes udev
- udev sleeps because the sysfs devices haven't been created yet
- sysfs devices are created
- getty errors out, closing the console device
- kernel vc device is destroyed
- sysfs devices disappear
- udev wakes up, and continues to wait for the sysfs entries

[LaMont Jones (lamont)]

*** Bug 8230 has been marked as a duplicate of this bug. ***

[LaMont Jones (lamont)]

This from debian bug #226443 has the following. This looks like a low-risk fix
to a rather annoying issue, and is in debian 2.12-8. (2.12-7ubuntu2 differs
from 2.12-7 only in the init script prettification. Hence we'll need to patch
whichever way we go, and simply uploading 2.12-7ubuntu3 seems simplest for the
preview freeze period.)

thoughts?
lamont

Resent-From: Samuel Thibault <email address hidden>

Today, I took my printk() and debuggued that thing right from the
kernel. The problem relies in userspace, as expected: some programs
hold an fd on /dev/tty3. Actually, there are two of them (and actually
only one: getty itself !):

- init will open /dev/tty0 and dup() it as standard output and standard
  error for getty. *If* the current vt is 3, /dev/tty0 will actually get
  to /dev/tty3 (that can be seen in the kernel in tty->name), so that
  getty actually already gets an fd on the tty he wants to open. *But*
  close(0) is done *after* VT_OPENQRY is done ! Hence someone (getty,
  actually) indeed already holds the tty and VT_OPENQRY doesn't give it.
- getty itself opens /dev/tty0 ! Just to be able to call VT_OPENQRY,
  actually. If the current vt is 3, it gets /dev/tty3, and of course,
  VT_OPENQRY doesn't give it !

So the patch I'm submitting just moves the close(0) *before* calling
VT_OPENQRY, and opens /dev/tty1 instead of /dev/tty0 to avoid using
/dev/tty3. This doesn't hurt since no check is done for /dev/tty1.
I now have no "getty fails to initialize /dev/tty[2-6]" anymore...

Please apply and ask for a urgent sarge build asap.

Regards,
Samuel Thibault

--- login-utils/agetty.c.buggy 2004-09-20 15:48:01.000000000 +0200
+++ login-utils/agetty.c 2004-09-20 15:48:38.000000000 +0200
@@ -634,6 +634,9 @@
        if ((st.st_mode & S_IFMT) != S_IFCHR)
            error(_("/dev/%s: not a character device"), tty);

+ (void) close(0);
+ errno = 0; /* ignore close(2) errors */
+
        /*
         * Try to avoid opening a vt that is already open, as this will
         * mean that the keyboard will be unusable.
@@ -663,7 +666,7 @@
                for ( i = 0 ; i < MAX_NR_CONSOLES ; i++ )
                    fds[i] = -1;

- if ((fd = open("/dev/tty0", O_WRONLY, 0) ) < 0
+ if ((fd = open("/dev/tty1", O_WRONLY, 0) ) < 0
                    && errno != ENOENT)
                    error(_("/dev/tty0: cannot open: %m"));

@@ -704,9 +707,6 @@

        /* Open the tty as standard input. */

- (void) close(0);
- errno = 0; /* ignore close(2) errors */
-
        debug(_("open(2)\n"));
        if (open(tty, O_RDWR|O_NONBLOCK, 0) != 0)
            error(_("/dev/%s: cannot open as standard input: %m"), tty);

[Jeff Waugh (jdub)]

upload approved

[LaMont Jones (lamont)]

Fixed in 2.12-7ubuntu5