can't reboot when root fs mounted r/o

Bump. This is urgent - people can't boot to single user mode, remount r/o, fsck, and then reboot.

I've done a little more debugging, the results of which I'll post later, but for the moment just wanted to note that this is exacerbated by the fact that magic sysrq keys don't work in single user mode for some reason.

Here's an image of the freeze I transcribed above (a different run), in case I made any important typos.

The following is a hackish fix which disables the log_action_end_msg, but fixes the problem. I suspect the problem is in /lib/lsb/init-functions, but I'm not having much luck determining where so far.

And here's a screenshot of what rebooting from a r/o fs looks like with my hackish patch applied (image attached in case there are any typing errors on my part. I disabled some debug "echo here"s in the patch).

...
* Disabling laptop mode...
/etc/rc6.d/K99laptop-mode: 85: cannot create: Directory nonexistent.
rm: missing operand
Try `rm --help' for more information.
* Terminating all remaining processes... [OK]
...

(and reboot finishes properly without hanging).

I should also note that it might be the filesystem containing /var that is the problem - I have only / and /home on this machine, so I haven't debugged that further.

This also affects a coworker of mine, marking confirmed.

Closing the laptop-mode-tools task, I don't think there's anything to fix there.

Seems to me that whomever "owns" /etc/init.d/laptop-mode is the group responsible for this bug, as I have no problems with any other init scripts. If not, do you have any suggestions for what the sysvinit folks need to track down, Tormod? Thanks.

This is endemic on all hardy machines in our office, some upgraded, some freshly installed. Marking confirmed.

Sorry, I thought it was just complaining without doing more harm. But actually, if the /tmp filesystem is not writeable, mktemp fails and TEMPFILE is empty. Calling cat with the empty argument makes cat wait for input on stdin.

ski, can you try this patch? After verification, I'll make a debdiff and we should try to get an SRU for this.

The previous patch won't apply on the Hardy version, since that one has a dangling LF on line 58.

Ah ha, good sleuthing Tormod. I just tried the (second) patch - works like a charm. Thanks a bunch.

This bug was fixed in the package laptop-mode-tools - 1.42-1ubuntu1

---------------
laptop-mode-tools (1.42-1ubuntu1) intrepid; urgency=low

  * Merge from debian unstable (LP: #229579), remaining changes:
    - etc/init.d/laptop-mode: Check if laptop mode is disabled in
      /etc/default/laptop-mode (that's the file laptop_mode looks into, too,
      but it incorrectly evaluates the ENABLE_LAPTOP_MODE setting) as well
      as in /etc/default/acpi-support.
    - /usr/sbin/laptop_mode: Do not parse arguments, ignore "force" and
      "init" and let start/stop mean enable/disable.
    - /usr/sbin/laptop_mode: Do not read $ACTIVATE_WITH_POSSIBLE_DATA_LOSS
      from /var/run/laptop-mode-state to see if state has changed.
    - debian/rules: Do not ship acpi/apm scripts (we handle that ourselves)
    - debian/laptop-mode-tools.preinst: Remove any old acpi/apm scripts
    - debian/laptop-mode-tools.postinst: Create /etc/default/laptop-mode
      (with ENABLE_LAPTOP_MODE=false) if the file does not exist yet and we
      are on ppc.
    - debian/control: Change maintainer
  * Use #DEBHELPER# in .postinst and .preinst (LP: #226788)
  * Avoid using temporary files in init script to avoid hangs on
    shutdown if filesystem is read-only (LP: #206217)
  * Also accept "stop" as second argument since the init script
    calls laptop_mode with "init stop"

laptop-mode-tools (1.42-1) unstable; urgency=low

  * New upstream version 1.42:
    * Remove "echo -e" bashism in lm-profiler. Closes: #480606.
    * Allow different names for AC adapter in /sys/class/power_supply.
      Closes: #478937.
    * Add "hal-polling" module to control HAL polling of CD/DVD drives. (Taken
      from powertop tips.) Closes: #438782.
    * Add "bluetooth" module to disable bluetooth in battery mode. (Taken from
      powertop tips.)

laptop-mode-tools (1.41-1) unstable; urgency=high

  * New upstream version 1.41:
    * Support energy_* in /sys/class/power_supply interface. Closes: #477118.
    * Rename module "core" to "laptop-mode". Closes: #477310.
    * Add missing assignment to ACTIVATE_WITH_POSSIBLE_DATA_LOSS.
      Closes: #477172.

laptop-mode-tools (1.40-1) unstable; urgency=high

  * Upgrade standards version to 3.7.3.
  * Add hal to recommends due to upstream changes.
  * New upstream version 1.40:
    * Add support for getting lid button state from HAL instead of from
      /proc/acpi.
    * Support /sys/class/power_supply interface to determine battery levels.
      Closes: #475926
    * Add generic support for config file swapping. Closes: #475304
    * Add module for iwlwifi power management. Closes: #475407
    * Fully modularized configuration.

laptop-mode-tools (1.36-1) unstable; urgency=low

  * New upstream version 1.36.
  * Init script LSB header: depend on remote_fs instead of local_fs.
    Closes: #468657.
  * Remove ^M from /etc/init.d/laptop-mode. Closes: #459422.
  * Change package description to say "X11" instead of "X Windows".
    Closes: #464066.
  * Get rid of empty /usr/bin directory in package. (lintian warning)

 -- Tormod Volden <email address hidden> Mon, 12 May 2008 16:24:50 +0200

Can you apply this to hardy before 8.04.1? The patch is quite simple, and the problem quite tedious.

Sorry, it is too late for 8.04.1, but we can try to make an SRU for Hardy anyway. It will need a debdiff against the current Hardy version and the procedure from https://wiki.ubuntu.com/StableReleaseUpdates. Contributions are welcome.

 laptop-mode-tools (1.35-1ubuntu2) hardy-proposed; urgency=low
 .
   * Avoid using temporary files in init script to avoid hangs on
     shutdown if filesystem is read-only (LP: #206217)