[Bug 839166] Re: mountall hangs while trying to send dbus? messages for /proc

[Steve Langasek]

On Mon, May 21, 2012 at 08:05:16PM -0000, DaveHansen wrote:
> We can certainly do a lot better than let a package install while silently
> letting "large chunks of its brain" be missing.  As you said,
> /etc/init/mountall*.conf are much more than mere config files: they're
> large chunks of mountall's brain.

I would certainly welcome it if we had better tooling options around this,
but that's really out of scope for a bug report against mountall, which is
already using the existing interfaces as designed.  That would be better
discussed on the ubuntu-devel mailing list.

> I also can't imagine a once-a-minute warning message saying "mountall
> waiting for $FOO to complete" could complicate the code that much, or
> even an override key that would give the user some hope of reaching a
> shell.

Well, without a reproducible scenario where mountall actually stalls in this
way, we'd be working in the dark here.  And missing upstart jobs is not at
all a common failure scenario.

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to mountall in Ubuntu.
https://bugs.launchpad.net/bugs/839166

Title:
  mountall hangs while trying to send dbus? messages for /proc

Status in “mountall” package in Ubuntu:
  Invalid

Bug description:
  Symptoms: system hangs forever at boot
  Versions: originally discovered with 2.25ubuntu1, but reproduced with 2.31 as well
  Description:	Ubuntu 11.04
  Release:	11.04
  # apt-cache policy mountall
  mountall:
    Installed: 2.25ubuntu1
    Candidate: 2.25ubuntu1
    Version table:
   *** 2.25ubuntu1 0
          500 http://ubuntu.osuosl.org/ubuntu/ natty/main amd64 Packages
          100 /var/lib/dpkg/status

  Alt-sysrq-k shows that it was it was waiting for mountall.  Adding some debugging shows that it's waiting on this line of code in mounted() for /proc:
          emit_event ("mounted", mnt);
  Which in turn is waiting on:
          pending_call = NIH_SHOULD (upstart_emit_event (upstart,
                                                         name, env, mnt ? TRUE : FALSE,
                                                         NULL, emit_event_error, NULL,
                                                       NIH_DBUS_TIMEOUT_NEVER));

  Changing that NIH_DBUS_TIMEOUT_NEVER to something like "90 * 1000"
  will let the system at least boot (after 90 seconds of course),
  although into somewhat of a crippled state because mountall dies,
  although the rest of the system seems to start OK.

  The first problem is that the system simply hangs.  None of the
  mountall key commands like 'S' to skip mounts are yet available.
  There is also no message about what it is doing.  Basically, it's
  really hard to debug.  Some messages from mountall would probably be
  really handy here.  Perhaps it could pass a timeout value in and at
  least spit something to the console, like:

  int wait = mnt ? TRUE : FALSE,
  while (wait) {
          pending_call = NIH_SHOULD (upstart_emit_event (upstart,..., MY_TIMEOUT));
          if (pending_call)
              break;
          print_helpful_message_here();
  }

  The other part is that I have no idea how this could have ever worked.  upstart_emit_event() looks to be talking to dbus, but dbus does not get started until local-filesystems are mounted (from /etc/init/dbus.conf):
  start on local-filesystems

  The 'local-filesystems' event appears to me in trigger_events() to only be triggered *after* virtual filesystems:
          /* Enforce local only after virtual filesystems triggered */
  So, since /proc is a virtual filesystem, it makes sense that it would hang trying to notify dbus since dbus is not yet started.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/mountall/+bug/839166/+subscriptions