liveinst fails after resize of image

[mdda]

I have created a new Fedora spin 'EeeDora' specifically for the EeePC.  The
Live Image boots fine (very fine).  Live install works in manual mode, but
with a kickstart file (I'd like to make this completely automatic) Anaconda
goes through the installation (writing packages=100%), but gives an
exception in the last stages.  This leaves the machine in an unusable state.

What appears to be happening is that the root fs is not being mounted on
/mnt/sysimage/ after the resize.  Looking at /dev/sda1, I can see everything
got installed up to that stage.  

anaconda None exception report
Traceback (most recent call first):
  File "/usr/lib/anaconda/fsset.py", line 1202, in write
    f = open (prefix + "/etc/fstab", "w")
  File "/usr/lib/anaconda/livecd.py", line 313, in doPostInstall
    anaconda.id.fsset.write(anaconda.rootPath)
  File "/usr/lib/anaconda/backend.py", line 195, in doPostInstall
    anaconda.backend.doPostInstall(anaconda)
  File "/usr/lib/anaconda/dispatch.py", line 203, in moveStep
    rc = stepFunc(self.anaconda)
  File "/usr/lib/anaconda/dispatch.py", line 126, in gotoNext
    self.moveStep()
  File "/usr/lib/anaconda/gui.py", line 1048, in nextClicked
    self.anaconda.dispatch.gotoNext()
  File "/usr/lib/anaconda/iw/progress_gui.py", line 67, in renderCallback
    self.intf.icw.nextClicked()
  File "/usr/lib/anaconda/gui.py", line 1075, in handleRenderCallback
    self.currentWindow.renderCallback()
IOError: [Errno 2] No such file or directory: '/mnt/sysimage/etc/fstab'

http://www.nabble.com/file/p14207396/error1 error1 


Just me hacking around... : If I force ext2 partitions to be mounted (by
jamming isMountable() ) then anaconda gets to complete - but the machine is
still unbootable because /dev/root doesn't get created/found properly.  

Can someone shed some light on this?  My knowledge of Python approaches
zero, so it's tricky to sort this out (particularly since I'm doing this on
the small keyboard of the Eee, booted on the live image...)

Thanks
Martin
:-)
-- 
View this message in context:
http://www.nabble.com/liveinst-fails-after-resize-of-image-tf4960453.html...
Sent from the Fedora Livecd List mailing list archive at Nabble.com.

[Douglas McClendon]

mdda wrote:
>  I have created a new Fedora spin 'EeeDora' specifically for
> the EeePC.  The
>  Live Image boots fine (very fine).  Live install works in manual mode, but
>  with a kickstart file (I'd like to make this completely automatic) Anaconda
>  goes through the installation (writing packages=100%), but gives an
>  exception in the last stages.  This leaves the machine in an unusable state. 
Running anaconda with a kickstart from a livecd, is something that I 
haven't tried to use at all, but have been meaning to.  If this doesn't 
get fixed by someone else soon, at some point I'll attempt to recreate 
it, and debug it.

-dmc

>  What appears to be happening is that the root fs is not being mounted
> on
>  /mnt/sysimage/ after the resize.  Looking at /dev/sda1, I can see everything
>  got installed up to that stage.  
>  
>  anaconda None exception report
>  Traceback (most recent call first):
>    File "/usr/lib/anaconda/fsset.py", line 1202, in write
>      f = open (prefix + "/etc/fstab", "w")
>    File "/usr/lib/anaconda/livecd.py", line 313, in doPostInstall
>      anaconda.id.fsset.write(anaconda.rootPath)
>    File "/usr/lib/anaconda/backend.py", line 195, in doPostInstall
>      anaconda.backend.doPostInstall(anaconda)
>    File "/usr/lib/anaconda/dispatch.py", line 203, in moveStep
>      rc = stepFunc(self.anaconda)
>    File "/usr/lib/anaconda/dispatch.py", line 126, in gotoNext
>      self.moveStep()
>    File "/usr/lib/anaconda/gui.py", line 1048, in nextClicked
>      self.anaconda.dispatch.gotoNext()
>    File "/usr/lib/anaconda/iw/progress_gui.py", line 67, in renderCallback
>      self.intf.icw.nextClicked()
>    File "/usr/lib/anaconda/gui.py", line 1075, in handleRenderCallback
>      self.currentWindow.renderCallback()
>  IOError: [Errno 2] No such file or directory: '/mnt/sysimage/etc/fstab'
>  
>  http://www.nabble.com/file/p14207396/error1 error1 
>  
>  
>  Just me hacking around... : If I force ext2 partitions to be mounted (by
>  jamming isMountable() ) then anaconda gets to complete - but the machine is
>  still unbootable because /dev/root doesn't get created/found properly.  
>  
>  Can someone shed some light on this?  My knowledge of Python approaches
>  zero, so it's tricky to sort this out (particularly since I'm doing this on
>  the small keyboard of the Eee, booted on the live image...)
>  
>  Thanks
>  Martin
>  :-)

[Jeremy Katz]

On Fri, 2007-12-07 at 05:17 -0800, mdda wrote:
>  I have created a new Fedora spin 'EeeDora' specifically for
> the EeePC.  The
>  Live Image boots fine (very fine).  Live install works in manual mode, but
>  with a kickstart file (I'd like to make this completely automatic) Anaconda
>  goes through the installation (writing packages=100%), but gives an
>  exception in the last stages.  This leaves the machine in an unusable state. 
What does the kickstart config you're using for the install look like?

Jeremy

[mdda]

>  What does the kickstart config you're using for the install look
> like? 
My idea is to fix up everything, but let the user make changes - and I've
found a way to make it work.

First : The (unchanged) Kickstart file : 
http://www.nabble.com/file/p14226119/eee-live-install.ks eee-live-install.ks 

Some pertinent points :
* I'm forcing the install partition to be ext2 (the rumour is that this will
wear down the Eee's internal flash least)
* I found that the sizing of the (faked) partition within the ISO (at
4096Mb) was too large for the Eee's 3.8Gb flash drive - and that the
minsizing thing wasn't behaving somehow (my ISO-building machine is fully
yum updated).  So I fixed this hurdle by forcing the the size to (like)
2048Mb in livecd-creator.
* To be on the safe side, I commented out the call to
self._resizeRootfs(anaconda, wait) in _doFilesystemMangling (this may have
been overkill, but it occurred to me that 'resize2fs' wouldn't be happy
resizing a mounted ext2 partition, so that might be causing problems)
* Most significantly (?), since the writing of $SYSIMAGE/etc/fstab seemed to
be failing because $SYSIMAGE/etc wasn't there, I brute-forced the following
inside doPostInstall in /usr/lib/anaconda/livecd.py :

    def doPostInstall(self, anaconda):
        self._doFilesystemMangling(anaconda)

        # maybe heavy handed, but it'll do
        anaconda.id.bootloader.args.append("rhgb quiet")
        anaconda.id.desktop.setDefaultRunLevel(5)

### added vvv
        r = anaconda.id.fsset.getEntryByMountPoint("/")
        r.mount(anaconda.rootPath)                
### added ^^^

        # now write out the "real" fstab and mtab
        anaconda.id.fsset.write(anaconda.rootPath)

This finally did the trick, and now everything writes and boots without
complaining.  

Hopefully, somewhere in this hackery is a clue to what's actually messing
up.  I'm pretty sure there's a more subtle way of doing it - but anaconda is
a big machine, and I feel like I'm looking at it through a key-hole.

- Martin

-- 
View this message in context:
http://www.nabble.com/liveinst-fails-after-resize-of-image-tf4960453.html...
Sent from the Fedora Livecd List mailing list archive at Nabble.com.

[Jeremy Katz]

On Fri, 2007-12-07 at 23:28 -0800, mdda wrote:
>  > What does the kickstart config you're using for the install
> look like?
>  My idea is to fix up everything, but let the user make changes - and I've
>  found a way to make it work.
>  
>  First : The (unchanged) Kickstart file : 
>  http://www.nabble.com/file/p14226119/eee-live-install.ks eee-live-install.ks 
>  
>  Some pertinent points :
>  * I'm forcing the install partition to be ext2 (the rumour is that this will
>  wear down the Eee's internal flash least) 
Is the partition on your live image ext2 also?  If not, then I can
definitely see that leading to some of the weirdness you're describing.

>  * I found that the sizing of the (faked) partition within the ISO
> (at
>  4096Mb) was too large for the Eee's 3.8Gb flash drive - and that the
>  minsizing thing wasn't behaving somehow (my ISO-building machine is fully
>  yum updated).  So I fixed this hurdle by forcing the the size to (like)
>  2048Mb in livecd-creator. 
davidz said something along these lines, but I did an install on
Thursday or so into a virtual disk that was < 4 gigs without problems.
And even with swap.  So there's something weird going on with the Eee in
particular, but I'm not sure what.  :-/

>  * To be on the safe side, I commented out the call to
>  self._resizeRootfs(anaconda, wait) in _doFilesystemMangling (this may have
>  been overkill, but it occurred to me that 'resize2fs' wouldn't be happy
>  resizing a mounted ext2 partition, so that might be causing problems) 
The resize is done without the filesystem mounted, so this shouldn't be
needed

>  * Most significantly (?), since the writing of $SYSIMAGE/etc/fstab
> seemed to
>  be failing because $SYSIMAGE/etc wasn't there, I brute-forced the following
>  inside doPostInstall in /usr/lib/anaconda/livecd.py :
>  
>      def doPostInstall(self, anaconda):
>          self._doFilesystemMangling(anaconda) 
_doFilesystemMangling ends up remounting the filesystems, so they should
be mounted at that point

Jeremy

[mdda]

Sorry it took so long for me to get to the bottom of this...

> > * I'm forcing the install partition to be ext2 (the rumour is
> that this
> > will
> > wear down the Eee's internal flash least)
>  Is the partition on your live image ext2 also?  If not, then I can
>  definitely see that leading to some of the weirdness you're describing. 
Right - this is the main problem (and I'm amazed that just remounting an
ext3 raw image as ext2 seemed to work at all).  

I had to rework livecd-creator to be able to create both ext3 and ext2
partitions.  Ideally, it should be a command-line option (since ext2 is
pretty important for flash drive installs).  Mayflower was also hard-wired
to expect an ext3 partition - so I've got an autodetector version.

It would be even cleaner if .../LiveOS/ext3fs.img were renamed
/LiveOS/ext3.img (otherwise the squashfs img should become
.../LiveOS/squashfsfs.img) - and I'm sure there was already a patch proposed
for this.

As it stands, I have to have special versions of livecd-creator and
mayflower embedded with my project, which is a pity.  But at least the
anaconda installer works unchanged, although I have to update
/usr/sbin/liveinst to accept command line parameters to pass to anaconda so
that I can pass a kickstart file.

Result : Now I've got a clean live image with one-click install onto an ext2
rootfs on the Eee's internal flash drive.  (But it has been a bit of a
struggle).

I'll check out the 4096Mb resizing thing separately. 

Martin
:-)

PS: One alternative (to changing livecd-creator and mayflower) would be to
detect within liveinst whether the partition fstype of the source and the
target was the same - if it is, then do the current 8Mb-at-a-time copy,
otherwise do a more traditional file copy thing.

-- 
View this message in context:
http://www.nabble.com/liveinst-fails-after-resize-of-image-tp14207396p143...
Sent from the Fedora Livecd List mailing list archive at Nabble.com.

[Jeremy Katz]

On Wed, 2007-12-12 at 13:49 -0800, mdda wrote:
>  Sorry it took so long for me to get to the bottom of this...
>  
>  >> * I'm forcing the install partition to be ext2 (the rumour is that this
>  >> will
>  >> wear down the Eee's internal flash least)
>  > Is the partition on your live image ext2 also?  If not, then I can
>  > definitely see that leading to some of the weirdness you're describing.
>  
>  Right - this is the main problem (and I'm amazed that just remounting an
>  ext3 raw image as ext2 seemed to work at all).   
This has always worked and has always been supposed to.

>  I had to rework livecd-creator to be able to create both ext3 and
> ext2
>  partitions.  Ideally, it should be a command-line option (since ext2 is
>  pretty important for flash drive installs).  Mayflower was also hard-wired
>  to expect an ext3 partition - so I've got an autodetector version. 
Patches accepted :)

>  PS: One alternative (to changing livecd-creator and mayflower) would
> be to
>  detect within liveinst whether the partition fstype of the source and the
>  target was the same - if it is, then do the current 8Mb-at-a-time copy,
>  otherwise do a more traditional file copy thing. 
A traditional file copy was ... painfully slow when I tried it last.  So
it's probably something we should avoid if we want.  And it should be
practical to allow an ext2 filesystem to be detected and work just as
well.  Plus, that will help when we eventually want to use ext4 :)

Jeremy