Press S or M does not work for more than one partition on error

[Mathieu Alorent]

Binary package hint: mountall

On 10.04 Lucid Server/Desktop installations, if a filesystem that is specified in /etc/fstab cannot be mounted for some reason (USB disk not connected, partition deleted, samba share not available etc.) the boot process hangs displaying "Press S to Skip or M for Manual recovery".

It works when there is only one partition on error, but with more than one, during the second message, pressing S/M does nothing and boot process hang there.

As seen in bug #571444, a workaround is to press alt-sysrq-i and edit your fstab to remove missing drives.

This bug looks like bug #571444, but on #571444 there is another issue which is about "no plymouth theme, so no message displayed". That's why I've opened a new bug specifically for "Press S or M does not work for more than one partition on error" problem.

I've found when trying to debug/trace the problem that the problem come from plymouth_answer function at lines 3037/3038.

the test :
        if ((plymouth_mnt != NULL) && (plymouth_mnt->error != plymouth_error))
                return;
is true as "plymouth_mnt->error" variable as been reset to "ERROR_NONE"

I'll post a patch.

[Mathieu Alorent (kumy)]

Here is a patch :)

I've also build a new package, you could find it in my PPA.

https://launchpad.net/~kumy/+archive/ppa?field.series_filter=lucid

( sudo add-apt-repository ppa:kumy/ppa )

[Brian Murray (brian-murray)]

I tried recreating this on Natty by creating two entries in my /etc/fstab for 2 devices that do not exist. After pressing S to skip the first entry, I was then presented the option to press S again for the second partition.

[Colin Watson (cjwatson)]

Thanks for your patch.

The third change to src/mountall.c appears unnecessary. mnt->error was already set to ERROR_NONE immediately above (we know that the 'if (status)' block can't have been run, since that unconditionally returns).

[Colin Watson (cjwatson)]

Oh, never mind, you're moving that code. I'll reread with more coffee!

[Colin Watson (cjwatson)]

In the first change to src/mountall.c (in run_mount), mnt->error can only be ERROR_NONE or ERROR_BORED due to the check at the top of this function. The effect of that change is therefore that mnt->error may be ERROR_BORED in the period between spawn being called and run_mount_finished being called from the event loop (this isn't a problem in the other two branches of that if statement because they both pass wait == TRUE to spawn). I think that this is wrong; it means that, if the mount takes long enough that plymouth_update is called in the meantime, then the message "The disk drive for %s is not ready yet or not present" will be displayed even though mount is running on that device.

So, let's analyse the second and third changes. Those move 'mnt->error = ERROR_NONE' down to below the 'if (status)' block. The only case whose behaviour that can change is that where mount exits non-zero on a remote filesystem. It appears that currently mountall is intentionally not printing errors about those; see bug 504224. I can see that it might make some sense to zero out any previous ERROR_BORED error, for consistency, but that's the only effect I can see this having.

I'm certainly not opposed to this kind of fix, and I'm sorry that we've taken so long to respond to your patch; the bug as described is definitely an important one to fix. However, I want to understand patches completely before I apply them, to make sure that I'm not breaking something else in the process. Could you explain the exact situation where this patch changes behaviour, with reference to the code?

[Scott Moser (smoser)]

I'm un-subscribing ubuntu-sponsors, as Collin asked for input. When you sort that out, please move this from 'incomplete' and subscribe sponsors team again.

[Daniel Nyström (speakman)]

I'm running Natty and has the latest updates as of 1 June 2011. I'm experiencing as described in the bug report; having four NFS entries failing at boot, there is no reaction when pressing either S nor M when asked for.

Taking me away from this required the Alternate CD on an USB-disk and booting for rescue.

[Steve Langasek (vorlon)]

Several people have tried to reproduce this issue without success in recent versions of mountall. I've tried myself in saucy, and can't reproduce it. The request for a reproducer was sent two years ago. With no more information, I'm closing this bug as invalid (and presumed fixed in recent releases).

[Rolf Leggewie (r0lf)]

lucid has seen the end of its life and is no longer receiving any updates. Marking the lucid task for this ticket as "Won't Fix".