OpenSuSE 11.1 install experiences from a Linux expert

I’m a Linux fan and user since August 1992 and when I tried to install opensuse 11.1 x86_64 on my hp/compaq nx6325 (network install from inside our company network)
I encountered the following difficulties:

• setup tries to DNS-lookup download.opensuse.org,
  which is bound to fail behind a serious firewall
  (i.e. on a 10.x.x.x network)

  workaround: go to expert->start shell and do
  an “echo 1.2.3.4 download.opensuse.org >> /etc/hosts”

  I was confused to see network install to be able to
  with a http proxy, but unconditionally attempt a DNS
  lookup

• The non-support of LILO is a pretty dumb move.
  GRUB works pretty much only in the MBR of the boot disk,
  while lilo works reliably and great from pretty much
  everywhere, in particular in partition boot blocks
  of partitions at the end of large disk

• the boot manager configuration dialog seems pretty
  broken, it killed the HP recovery feature in my MBR
  although I DID find and disable the “overwrite MBR”
  and I did check the boot device in lilo.conf to be
  /dev/sda3 several times before the bootmanager install
  was performed.

• Aborting the installation procedure halfway wedged the
  installer pretty thoroughly, nagging me with
  hundreds of failure popups which showed in the “details”
  the reason “User aborted”. Abort simply doesn’t work

• the installed kernel fails to boot, it constantly dies
  with “PANIC: early exception 0e rip 10:fffff…”
  Booting the Net installation CD and the installed
  opensuse through the “Boot installed system from
  harddisk” is the only way to run the system

• When YOU is started from an interactive shell
  (sudo bash, export DISPLAY=…) then exiting YOU
  with cancel will immediately bring up a textual YOU
  in that shell…

• The HTTP-proxy that I had entered for installation
  (and that did still work on second stage installation
  after reboot is not written into /etc/sysconfig/proxy

• right after first bootup, after i got YOU up and
  finally running&connecting, I started downloading
  available patches. Halfway through download and
  installation, I suddenly received an network error
  popup from YOU. For some reason unknown to me the
  braindead NetworkManager had kicked in an killed the
  default route from the original (ifup-based) network
  configuration from installation (which was done with
  a static IP).

Is there any projection when a bootable x86_64 kernel will become available for opensuse 11.1? (one that doesn’t die with PANIC: early exception?)

The underlying Kernel bug seems to have been fixed mid of October 2008?

-martin

PS: the decision to de-support Lilo is a pretty bad news.
GRUB is nowhere near mature.

In 1993 I used “shoelace” (in the root partition bootblock) from an extended partition of a secondary SCSI-disk off a Adaptec 1542 SCSI-controller (primary disk was IDE).

It took several years before LILO grew up to do this (while the lilo manpage&readme continued to claim that it was impossible…). Maybe when GRUB grows up it will also one day be able to do things like this, but today GRUB will hardly work anywhere else than in the MBR of your first hard disk (which is not where I want to have it).

Adjusting lilo.conf fixed the PANICing kernel.
It seems that these days the kernels and initrds have grown so large that you need to enable additional tricks.

I replaced the global lilo parameters with this:

timeout = 150
large-memory
lba32
prompt
default = Linux
vga = ask
compact
boot = /dev/sda3

other necessary fixes to the lilo.conf created
by the SuSE installer were removing the offending
“root” lines from the Windows “other” boot entries,
removing “message=/boot/message” and fixing the
timeout value (above).

btw. the VGA values shown when “ask” is configured
are actually HEX values, while the vga value configured
in Lilo is a decimal value. So I’m using vga=835
now, which is 343hex (1400x1050x24).

It seems that in order to get that installation
booting under VMware (Player) as well, I will need to add some kernel modules to initrd (like mptspi scsi driver for the VMWare LSI logic scsi controller)…

Why are distros so extremly alien to moving an installation around between different hardwares these days? The initrd on the installation CD is not such royal PITA when it comes to drivers/modules.

-Martin

Because if the initrd were to contain every module conceivably needed it would be bloated and take longer to load and probe the hardware. When you are installing you don’t mind waiting so much. Waiting gets a bit old when you just want to get to work on a normal startup. Things have changed. Maybe that’s the price of progress—bloat.

whey I try to mount an NTFS filesystem without giving “-t ntfs” to mount, the mount command will wedge completely and eternally (trying to get something called “fuse” running). Only a kill -9 and umount -f will recover it.
Including “-t ntfs” when mounting works smooth. What is wrong with that fuse thing that will wedge the machine?

-Martin

on those constant interruptions to install…

We corrected these by simply choosing another mirror. Were having consistent problems ourselves with timeouts on http://downloads.suse.org.

I think that many your experiences is bugs, which noone reported so it is not fixed. I suggest you to report it to bugzilla and it gone in next release.
lilo is deprecated due to grub can fit almost all cases. If you have any case where grub doesn’t work, please report it also. (grub doesn’t have any problem with booting from partition and let live disc MBR (yast2-bootloader rewrite your MBR with generic boot-code))

Any reason you are not using ntfs-3g?

No, a local DNS server should be queried, which can make DNS queries through the firewall. Commonly this is the Router in SOHO networks.

I was confused to see network install to be able to
with a http proxy, but unconditionally attempt a DNS
lookup

This is not the true cause of your confusion.

• The non-support of LILO is a pretty dumb move.
  GRUB works pretty much only in the MBR of the boot disk,
  while lilo works reliably and great from pretty much
  everywhere, in particular in partition boot blocks
  of partitions at the end of large disk

GRUB is more commonly used, and has the advantage of not having to update a map file on kernel changes. It works installed into physical partitions, not the MBR, and supports large disks to, though a small boot partition on fastest part of disk is preferred to minimise boot times (same as with lilo).

The only reason lilo is still supported, is because of the architectures that require it. GRUB has been preferred since SuSE 8.0 or so, it works for most ppl, where lilo was found much harder.

• the boot manager configuration dialog seems pretty
  broken, it killed the HP recovery feature in my MBR
  although I DID find and disable the “overwrite MBR”
  and I did check the boot device in lilo.conf to be
  /dev/sda3 several times before the bootmanager install
  was performed.

A backup of MBR is made, you can use the “active” partition method and install GRUB into a partition.

• Aborting the installation procedure halfway wedged the
  installer pretty thoroughly, nagging me with
  hundreds of failure popups which showed in the “details”
  the reason “User aborted”. Abort simply doesn’t work

Known bug, see Fate and Bugzilla if you need details.

• the installed kernel fails to boot, it constantly dies
  with “PANIC: early exception 0e rip 10:fffff…”
  Booting the Net installation CD and the installed
  opensuse through the “Boot installed system from
  harddisk” is the only way to run the system

That’s booting the installed kernel, it would appear your boot loader installation didn’t go as well as you thought.

• When YOU is started from an interactive shell
  (sudo bash, export DISPLAY=…) then exiting YOU
  with cancel will immediately bring up a textual YOU
  in that shell…

Please report to Bugzilla.

• The HTTP-proxy that I had entered for installation
  (and that did still work on second stage installation
  after reboot is not written into /etc/sysconfig/proxy

The proxy setting for software manager purposes is written to ~root/.curlrc, to set up proxies for end user, there’s a YaST module for the purpose, or you can edit config file directly.

Merely setting the proxy variables, doesn’t mean the browsers will use them. Proxy set up has been regarded as a configuration task, so you need to activate the system variables.

• right after first bootup, after i got YOU up and
  finally running&connecting, I started downloading
  available patches. Halfway through download and
  installation, I suddenly received an network error
  popup from YOU. For some reason unknown to me the
  braindead NetworkManager had kicked in an killed the
  default route from the original (ifup-based) network
  configuration from installation (which was done with
  a static IP).

I use traditional ifup/down and avoid Network Manager and have not seen this problem. But I use KDE not GNOME.

Is there any projection when a bootable x86_64 kernel will become available for opensuse 11.1? (one that doesn’t die with PANIC: early exception?)

2.6.27.7-9-default works for me and is x86_64.

PS: the decision to de-support Lilo is a pretty bad news.
GRUB is nowhere near mature.

I think lilo is still required for EFI, so I don’t think lilo is unsupported, it’s just not the default for good reasons.

GRUB will hardly work anywhere else than in the MBR of your first hard disk (which is not where I want to have it).

??? !!! ???

I have multiple GRUB installs from different distro’s, they aren’t all in the MBR

It works fine for me.

Most of the installation problems over the network would be overcome if you set the dhcp server to supply the details. If these aren’t passed by the dhcp server, you need to set up the network manually, which means supplying the correct dns and routing information for the network.
No network is ever going to wrk if it can’t find a valid dns server, so this isn’t a suse problem, but a problem with you not supplying the correct details for your non-standard setup. Try searching the installion guide for a fix (I know, I do the same - try first, read the guide later if it doesn’t work!)
I miss the old redhat/suse style network install that allowed a local network install via ftp or http - all you needed was to supply the ip address and login details of the local installation server.
I have always used the traditional ifup method for network control, but I have read about problems using network manager, so don’t use it.

Most distros now use grub as a boot manager.

How do you mean, miss it? This functionality is still on the NET install CD. I’ve already done a few installs using NFS from anther host on the LAN, but HTTP, SMB, FTP and even TFTP are supported. It’s a good way to install if you have only a CD drive, but not enough memory for the live CD, and you have a donor host on the LAN. The NET install CD is not just for Internet installs, there is more functionality there.

Nope. In a sensibly managed company internal network with 10.x.x.x addressing there is absolutely no direct connection to the internet, and internal DNS is a seperate universe. You will not see anything besides internal addresses. So for scenarios where an HTTP proxy is used, a DNS lookup of arbitrary “internet addresses” is a BAD IDEA. Only the client-side HTTP Proxy that is directly connected to the internet needs to be able to resolve the address from the URL.

There are a number of broken Java applets out there that fail for the same flawed assumption. That’s why my hosts files are growing with nonsense-IP-addresses, just to make software not break on such flawed assumptions.

I know that GRUB is being used quite often these days.
Don’t get me wrong, I like GRUB in several ways, including the pretty winter graphics animation, but I encountered a setup where GRUB simply didn’t work, so after 2 full days I gave up. I had a linux installation in an 8 Gig partition at the end of a 80 Gig drive, grub installed on a floppy, but GRUB didn’t see the partition. The LBA addressing in GRUB is either defective or missing.

Debian maintains both grub menu.lst and lilo.conf, so it took me exactly 2 minutes to set the boot device to /dev/fd0 in /etc/lilo.conf and call lilo -v -v – and it just worked.

The disadvantage of Lilo is that it “hardwires” all sector references of kernel images and initrds that it can load, whereas grub can read filesystems and once you got it running in your environment, additions or changes are as simple as editing a text file.

But the clear advantage of Lilo is that the only connection to any fancy boot-manager is the lilo boot-block, and practically every bootmanager under the sun can handle partition bootblocks adequately, including something as simple as a floppy disk.

This may become important when some reckless OS installer has overwritten your existing/prefered boot loader with his own during installation, usually without asking and without giving you a choice.

I HATE IT when OS installers do that, no matter where they come from (Redmond or other)!

That’s an explanation but no excuse.

The (static) network configuration that I entered during install was persited into the network configuration – why wasn’t that done with the HTTP proxy configuration?
At least for the proxy configuration that YOU needs when you first login after a fresh install.

I also use KDE. I have no idea why the Network Manager kicked in halfway during YOU all on his own and all of a sudden. After getting a popup with a “no route to host” error i checked with “netstat -r” and then tried ifdown/ifup–which told me that Network Manager was now active. The installation did create a working ifup-configuration, and I prefer ifup over Network manager. I have no clue why it kicked in.

-Martin

It should not matter that the client gets a DNS server which is purely internal. When a proxy is used, there will be no lookup at the client, only at the proxy server. Of course an app that assumes access to the Internet is broken, but the OpenSUSE installer and updater, and for that matter, Firefox and all other web browsers, are not broken, and do not require resolution of external addresses, when a proxy is specified.

Ken
I missed a few distros and it isn’t readily apparent from the installation and I haven’t searched for the updated details.
Can you please start a new thread and tell us how or where it is?
In the old (v9.x) versions, the network install was a 1.4 meg floppy which asked for the installation server details.

See new thread as requested.

I’m sorry, but the opensuse 11.1 NET installer does it the other way round. It first asks for “download.opensuse.org” and tries to resolve it via DNS before it asks for an HTTP proxy. And you can only work around this bug by going through the expert to shell exit and appending a random IP for download.opensuse.org to /etc/hosts, e.g. “echo 1.2.3.4 download.opensuse.org > /etc/hosts”

-Martin

Looking up download.opensuse.org is a reasonable thing to do, to verify the DNS server. The network is running split horizon DNS in a way that is not as beneficial or mandatory as you assert.

There is known brokenness in the C library that can make DNS resolution fail with V6 addresses. That hits me and I was able to do netinstall without the work round you recommend.

Giving the IP adress as dotted quad suffices.

If you assertion “In a sensibly managed company internal network with 10.x.x.x addressing … internal DNS is a seperate universe. You will not see anything besides internal addresses.” was true, do you really think you’d be running into broken software?

Running split-horizon, DNS is perfectly possible wihout breaking lookup, and use of NAT is common to do things like port forwarding.

But the clear advantage of Lilo is that the only connection to any fancy boot-manager is the lilo boot-block, and practically every bootmanager under the sun can handle partition bootblocks adequately, including something as simple as a floppy disk.

??? !!! !!!

GRUB runs in various stages, you can chain load it from lilo, GRUB or “another Boot manager”. So what you claim just seems to be based on a false assumption.

HUH?

Trying to resolve download.opensuse.org via DNS when an HTTP proxy is used is obviously dumb. The bug in the installer is that it asks for the download hostname and filepath BEFORE asking whether an HTTP proxy is needed.

Giving the IP address probably only works when that IP address is the REAL IP Address of download.opensuse.org, because what you enter there will be put into the URL that gets sent to the HTTP proxy.

However, as I described, in a seperate DNS universe the real ip address of download.opensuse.org is NOT available.
the only reason why that /etc/hosts entry is necessary is to work around the obviously inappropriate and completely unnecessary DNS lookup by the opensuse 11.1 NET installer.

And unless you happen to know a valid real IP address of a hosts that carries the opensuse 11.1 installation files, you REALLLLLY want to supply a textual hostname and not an IP address, so that the HTTP proxy can actually connect to that repository.

-Martin

A year ago I had a PC (HP Pavillion d530 I think) with an 80 GByte SATA hard disk. I reduced the existing Windows (NTFS) partition and added an 8GB Linux installation at the end (Debian). However, when booting, GRUB (0.97) didn’t find any data on that partition, it simply was unable to boot a linux kernel from there. Lilo worked immediately. The LBA support in GRUB is either defective or missing.

On our company firewall we’re running a BIND with split horizon since 1995, so that we would get the hostnames instead of just ip-addresses into the proxy&gateway logfiles. (I modified BIND to do this. When I talked to Paul Vixie on the IETF Meeting in Stockholm he said that he wasn’t interested in adding this into the official BIND distribution, and a few others had created similar patches).

Seems like split horizon DNS is less uncommon today.

However there is a certain danger with concurrent visibility of both universes, because the answers from both universes contradict each other. Having a DNS Server that knows how to do split horizon and simple clients querying it is usually no problem, but I certainly would not tell the address of that split-horizon DNS server to any other caching DNS server; because they’re likely going to be confused.

-Martin

I think I just found the explanantion for the GRUB failure here:

OpenSUSE.us : ext3 and Grub - e2fsprogs incompatibility

Yup, I had installed my debian system on ext3
(on a partition that I had manually formatted).

-Martin