Before upgrading your system, it is strongly recommended that you make a full backup, or at least back up any data or configuration information you can't afford to lose. The upgrade tools and process are quite reliable, but a hardware failure in the middle of an upgrade could result in a severely damaged system.

The main things you'll want to back up are the contents of /etc, /var/lib/dpkg, /var/lib/aptitude/pkgstates and the output of dpkg --get-selections "*" (the quotes are important).

The upgrade process itself does not modify anything in the /home directory. However, some applications (e.g. parts of the Mozilla suite, and the GNOME and KDE desktop environments) are known to overwrite existing user settings with new defaults when a new version of the application is first started by a user. As a precaution, you may want to make a backup of the hidden files and directories (“dotfiles”) in users' home directories. This backup may help to restore or recreate the old settings. You may also want to inform users about this.

Any package installation operation must be run with superuser privileges, so either log in as root or use su or sudo to gain the necessary access rights.

The upgrade has a few preconditions; you should check them before actually executing the upgrade.

It's wise to inform all users in advance of any upgrades you're planning, although users accessing your system via an ssh connection should notice little during the upgrade, and should be able to continue working.

If you wish to take extra precautions, back up or unmount the /home partition before upgrading.

You will probably have to do a kernel upgrade when upgrading to lenny, so a reboot will normally be necessary. Typically, this will be done after the upgrade is finished.

Because of the many changes in the kernel between etch and lenny regarding drivers, hardware discovery and the naming and ordering of device files, there is a real risk that you may experience problems rebooting your system after the upgrade. A lot of known potential issues are documented in this and the next chapters of these Release Notes.

For that reason it makes sense to ensure that you will be able to recover if your system should fail to reboot or, for remotely managed systems, fail to bring up networking.

If you are upgrading remotely via an ssh link it is highly recommended that you take the necessary precautions to be able to access the server through a remote serial terminal. There is a chance that, after upgrading the kernel and rebooting, some devices will be renamed (as described in Section 4.6.2, “Device enumeration reordering” ) and you will have to fix the system configuration through a local console. Also, if the system is rebooted accidentally in the middle of an upgrade there is a chance you will need to recover using a local console.

The most obvious thing to try first is to reboot with your old kernel. However, for various reasons documented elsewhere in this document, this is not guaranteed to work.

If that fails, you will need an alternative way to boot your system so you can access and repair it. One option is to use a special rescue image or a Linux live CD. After booting from that, you should be able to mount your root file system and chroot into it to investigate and fix the problem.

Another option we'd like to recommend is to use the rescue mode of the lenny Debian Installer. The advantage of using the installer is that you can choose between its many installation methods for one that best suits your situation. For more information, please consult the section “Recovering a Broken System” in chapter 8 of the Installation Guide and the Debian Installer FAQ.

The distribution upgrade should be done either locally from a textmode virtual console (or a directly connected serial terminal), or remotely via an ssh link.

In order to gain extra safety margin when upgrading remotely, we suggest that you run upgrade processes in the virtual console provided by the screen program, which enables safe reconnection and ensures the upgrade process is not interrupted even if the remote connection process fails.

	Important
	You should not upgrade using telnet, rlogin, rsh, or from an X session managed by xdm, gdm or kdm etc on the machine you are upgrading. That is because each of those services may well be terminated during the upgrade, which can result in an inaccessible system that is only half-upgraded.

Users using the LILO bootloader should note that the default settings for initramfs-tools now generate an initramfs that is too large for LILO to load. Such users should either switch to grub, or edit the file /etc/initramfs-tools/initramfs.conf, changing the line

MODULES=most

to read

MODULES=dep

Note, however, that doing this will cause initramfs-tools to install onto the initramfs only those modules that are required for the particular hardware that it is being run on. If you want to generate boot media that will work on more hardware than just the machine you're generating it on, you should leave the line as

MODULES=most

and make sure you do not use LILO.

In some cases, the use of apt-get for installing packages instead of aptitude might make aptitude consider a package as “unused” and schedule it for removal. In general, you should make sure the system is fully up-to-date and “clean” before proceeding with the upgrade.

Because of this you should review if there are any pending actions in the package manager aptitude. If a package is scheduled for removal or update in the package manager, it might negatively impact the upgrade procedure. Note that correcting this is only possible if your sources.list still points to etch and not to stable or lenny; see Section A.2, “Checking your sources list”.

To perform this review, launch aptitude in “visual mode” and press g (“Go”). If it shows any actions, you should review them and either fix them or implement the suggested actions. If no actions are suggested you will be presented with a message saying “No packages are scheduled to be installed, removed, or upgraded”.

If you have configured APT to install certain packages from a distribution other than stable (e.g. from testing), you may have to change your APT pinning configuration (stored in /etc/apt/preferences) to allow the upgrade of packages to the versions in the new stable release. Further information on APT pinning can be found in apt_preferences(5).

Regardless of the method used for upgrading, it is recommended that you check the status of all packages first, and verify that all packages are in an upgradable state. The following command will show any packages which have a status of Half-Installed or Failed-Config, and those with any error status.

# dpkg --audit

You could also inspect the state of all packages on your system using dselect, aptitude, or with commands such as

# dpkg -l | pager

or

# dpkg --get-selections "*" > ~/curr-pkgs.txt

It is desirable to remove any holds before upgrading. If any package that is essential for the upgrade is on hold, the upgrade will fail.

Note that aptitude uses a different method for registering packages that are on hold than apt-get and dselect. You can identify packages on hold for aptitude with

# aptitude search "~ahold" | grep "^.h"

If you want to check which packages you had on hold for apt-get, you should use

# dpkg --get-selections | grep hold

If you changed and recompiled a package locally, and didn't rename it or put an epoch in the version, you must put it on hold to prevent it from being upgraded.

The “hold” package state for aptitude can be changed using:

# aptitude hold package_name

Replace hold with unhold to unset the “hold” state.

If there is anything you need to fix, it is best to make sure your sources.list still refers to etch as explained in Section A.2, “Checking your sources list”.

If you have listed the proposed-updates section in your /etc/apt/sources.list file, you should remove it from that file before attempting to upgrade your system. This is a precaution to reduce the likelihood of conflicts.

If you have any non-Debian packages on your system, you should be aware that these may be removed during the upgrade because of conflicting dependencies. If these packages were installed by adding an extra package archive in your /etc/apt/sources.list, you should check if that archive also offers packages compiled for lenny and change the source line accordingly at the same time as your source lines for Debian packages.

Some users may have unofficial backported “newer” versions of packages that are in Debian installed on their etch system. Such packages are most likely to cause problems during an upgrade as they may result in file conflicts^[3]. Section 4.5.8, “Possible issues during upgrade” has some information on how to deal with file conflicts if they should occur.

The default configuration is set up for installation from main Debian Internet servers, but you may wish to modify /etc/apt/sources.list to use other mirrors, preferably a mirror that is network-wise closest to you.

Debian HTTP or FTP mirror addresses can be found at http://www.debian.org/distrib/ftplist (look at the “list of Debian mirrors” section). HTTP mirrors are generally speedier than FTP mirrors.

For example, suppose your closest Debian mirror is http://mirrors.kernel.org. When inspecting that mirror with a web browser or FTP program, you will notice that the main directories are organized like this:

http://mirrors.kernel.org/debian/dists/lenny/main/binary-i386/...
http://mirrors.kernel.org/debian/dists/lenny/contrib/binary-i386/...

To use this mirror with apt, you add this line to your sources.list file:

deb http://mirrors.kernel.org/debian lenny main contrib

Note that the `dists' is added implicitly, and the arguments after the release name are used to expand the path into multiple directories.

After adding your new sources, disable the previously existing “deb” lines in sources.list by placing a hash sign (#) in front of them.

Instead of using HTTP or FTP package mirrors, you may wish to modify /etc/apt/sources.list to use a mirror on a local disk (possibly mounted over NFS).

For example, your package mirror may be under /var/ftp/debian/, and have main directories like this:

/var/ftp/debian/dists/lenny/main/binary-i386/...
/var/ftp/debian/dists/lenny/contrib/binary-i386/...

To use this with apt, add this line to your sources.list file:

deb file:/var/ftp/debian lenny main contrib

Note that the `dists' is added implicitly, and the arguments after the release name are used to expand the path into multiple directories.

After adding your new sources, disable the previously existing “deb” lines in sources.list by placing a hash sign (#) in front of them.

If you want to use CDs only, comment out the existing “deb” lines in /etc/apt/sources.list by placing a hash sign (#) in front of them.

Make sure there is a line in /etc/fstab that enables mounting your CD-ROM drive at the /cdrom mount point (the exact /cdrom mount point is required for apt-cdrom). For example, if /dev/hdc is your CD-ROM drive, /etc/fstab should contain a line like:

/dev/hdc /cdrom auto defaults,noauto,ro 0 0

Note that there must be no spaces between the words defaults,noauto,ro in the fourth field.

To verify it works, insert a CD and try running

# mount /cdrom    # this will mount the CD to the mount point
# ls -alF /cdrom  # this should show the CD's root directory
# umount /cdrom   # this will unmount the CD

Next, run:

# apt-cdrom add

for each Debian Binary CD-ROM you have, to add the data about each CD to APT's database.

It is strongly recommended that you use the /usr/bin/script program to record a transcript of the upgrade session. Then if a problem occurs, you will have a log of what happened, and if needed, can provide exact information in a bug report. To start the recording, type:

# script -t 2>~/upgrade-lenny.time -a ~/upgrade-lenny.script

or similar. Do not put the typescript file in a temporary directory such as /tmp or /var/tmp (files in those directories may be deleted during the upgrade or during any restart).

The typescript will also allow you to review information that has scrolled off-screen. Just switch to VT2 (using Alt+F2) and, after logging in, use less -R ~root/upgrade-lenny.script to view the file.

After you have completed the upgrade, you can stop script by typing exit at the prompt.

If you have used the -t switch for script you can use the scriptreplay program to replay the whole session:

# scriptreplay ~/upgrade-lenny.time ~/upgrade-lenny.script

First the list of available packages for the new release needs to be fetched. This is done by executing:

# aptitude update

Running this the first time new sources are updated will print out some warnings related to the availability of the sources. These warnings are harmless and will not appear if you rerun the command again.

You have to make sure before upgrading your system that you have sufficient hard disk space when you start the full system upgrade described in Section 4.5.7, “Upgrading the rest of the system”. First, any package needed for installation that is fetched from the network is stored in /var/cache/apt/archives (and the partial/ subdirectory, during download), so you must make sure you have enough space on the file system partition that holds /var/ to temporarily download the packages that will be installed in your system. After the download, you will probably need more space in other file system partitions in order to both install upgraded packages (which might contain bigger binaries or more data) and new packages that will be pulled in for the upgrade. If your system does not have sufficient space you might end up with an incomplete upgrade that might be difficult to recover from.

Both aptitude and apt will show you detailed information of the disk space needed for the installation. Before executing the upgrade, you can see this estimate by running:

# aptitude -y -s -f --with-recommends dist-upgrade
[ ... ]
XXX upgraded, XXX newly installed, XXX to remove and XXX not upgraded.
Need to get xx.xMB/yyyMB of archives. After unpacking AAAMB will be used.
Would download/install/remove packages.

	Note
	Running this command at the beginning of the upgrade process may give an error, for the reasons described in the next sections. In that case you will need to wait until you've done the minimal system upgrade as in Section 4.5.6, “Minimal system upgrade” and upgraded your kernel before running this command to estimate the disk space.

If you do not have enough space for the upgrade, make sure you free up space beforehand. You can:

Note that in order to safely remove packages, it is advisable to switch your sources.list back to etch as described in Section A.2, “Checking your sources list”.

Several bug reports have shown that the versions of the aptitude and apt packages in etch are often unable to handle the upgrade to lenny. In lenny, apt is better at dealing with complex chains of packages requiring immediate configuration and aptitude is smarter at searching for solutions to satisfy the dependencies. These two features are heavily involved during the dist-upgrade to lenny, so it is necessary to upgrade these two packages before upgrading anything else.

The following command will upgrade both aptitude and apt:

# aptitude install aptitude apt dpkg

This step will also automatically upgrade libc6 and locales. At this point, some running services will be restarted, including xdm, gdm and kdm. As a consequence, local X11 sessions might be disconnected.

Upgrading with apt

Please note that using apt-get is not recommended for the upgrade from etch to lenny. If you do not have aptitude installed you are recommended to install it first. If you want to perform the upgrade with apt or if the upgrade with aptitude failed and you want to try the upgrade with apt' dependency chain resolution algorithm, you should run: # apt-get install apt Note that you will have to adapt other aptitude commands to use apt-get instead.

aptitude maintains a list of packages that were installed automatically (for instance, as dependencies of another package). In lenny, apt now has this feature as well.

The first time the lenny version of aptitude is run, it will read in its list of automatically installed packages and convert it for use with the lenny version of apt. If you have aptitude installed, you should at least issue one aptitude command to do the conversion. One way to do this is by searching for a non-existent package:

# aptitude search "?false"

Because of certain necessary package conflicts between etch and lenny, running aptitude dist-upgrade directly will often remove large numbers of packages that you will want to keep. We therefore recommend a two-part upgrade process, first a minimal upgrade to overcome these conflicts, then a full dist-upgrade.

First, run:

# aptitude safe-upgrade

This has the effect of upgrading those packages which can be upgraded without requiring any other packages to be removed or installed.

The next step will vary depending on the set of packages that you have installed. These release notes give general advice about which method should be used, but if in doubt, it is recommended that you examine the package removals proposed by each method before proceeding.

Some common packages that are expected to be removed include base-config, hotplug, xlibs, netkit-inetd, python2.3, xfree86-common, and xserver-common. For more information about packages obsoleted in lenny, see Section 4.10, “Obsolete packages”.

You are now ready to continue with the main part of the upgrade. Execute:

# aptitude dist-upgrade

This will perform a complete upgrade of the system, i.e. install the newest available versions of all packages, and resolve all possible dependency changes between packages in different releases. If necessary, it will install some new packages (usually new library versions, or renamed packages), and remove any conflicting obsoleted packages.

When upgrading from a set of CD-ROMs (or DVDs), you will be asked to insert specific CDs at several points during the upgrade. You might have to insert the same CD multiple times; this is due to inter-related packages that have been spread out over the CDs.

New versions of currently installed packages that cannot be upgraded without changing the install status of another package will be left at their current version (displayed as “held back”). This can be resolved by either using aptitude to choose these packages for installation or by trying aptitude -f install package.

If an operation using aptitude, apt-get, or dpkg fails with the error

E: Dynamic MMap ran out of room

the default cache space is insufficient. You can solve this by either removing or commenting lines you don't need in /etc/apt/sources.list or increasing the cache size. The cache size can be increased by setting APT::Cache-Limit in /etc/apt/apt.conf. The following command will set it to a value that should be sufficient for the upgrade:

# echo 'APT::Cache-Limit "12500000";' >> /etc/apt/apt.conf

This assumes that you do not yet have this variable set in that file.

Sometimes it's necessary to enable the APT::Force-LoopBreak option in APT to be able to temporarily remove an essential package due to a Conflicts/Pre-Depends loop. aptitude will alert you of this and abort the upgrade. You can work around this by specifying the option -o APT::Force-LoopBreak=1 on the aptitude command line.

It is possible that a system's dependency structure can be so corrupt as to require manual intervention. Usually this means using aptitude or

# dpkg --remove package_name

to eliminate some of the offending packages, or

# aptitude -f install
# dpkg --configure --pending

In extreme cases you might have to force re-installation with a command like

# dpkg --install /path/to/package_name.deb

File conflicts should not occur if you upgrade from a “pure” etch system, but can occur if you have unofficial backports installed. A file conflict will result in an error like:

Unpacking <package-foo> (from <package-foo-file>) ...
dpkg: error processing <package-foo> (--install):
 trying to overwrite `<some-file-name>',
 which is also in package <package-bar>
dpkg-deb: subprocess paste killed by signal (Broken pipe)
 Errors were encountered while processing:
 <package-foo>

You can try to solve a file conflict by forcibly removing the package mentioned on the last line of the error message:

# dpkg -r --force-depends package_name

After fixing things up, you should be able to resume the upgrade by repeating the previously described aptitude commands.

During the upgrade, you will be asked questions regarding the configuration or re-configuration of several packages. When you are asked if any file in the /etc/init.d or /etc/terminfo directories, or the /etc/manpath.config file should be replaced by the package maintainer's version, it's usually necessary to answer `yes' to ensure system consistency. You can always revert to the old versions, since they will be saved with a .dpkg-old extension.

If you're not sure what to do, write down the name of the package or file and sort things out at a later time. You can search in the typescript file to review the information that was on the screen during the upgrade.

When you dist-upgrade from etch to lenny, it is strongly recommended that you install a new linux-image-2.6-* metapackage. This package may be installed automatically by the dist-upgrade process. You can verify this by running:

# dpkg -l "linux-image*" | grep ^ii

If you do not see any output, then you will need to install a new linux-image package by hand. To see a list of available linux-image-2.6 metapackages, run:

# apt-cache search linux-image-2.6- | grep -v transition

If you are unsure about which package to select, run uname -r and look for a package with a similar name. For example, if you see '2.6.18-6-686', it is recommended that you install linux-image-2.6-686. (Note that the k7 flavor no longer exists; if you are currently using the k7 kernel flavor, you should install the 686 flavor instead.) You may also use apt-cache to see a long description of each package in order to help choose the best one available. For example:

# apt-cache show linux-image-2.6-686

You should then use aptitude install to install it. Once this new kernel is installed you should reboot at the next available opportunity to get the benefits provided by the new kernel version.

For the more adventurous there is an easy way to compile your own custom kernel on Debian GNU/Linux. Install the kernel-package tool and read the documentation in /usr/share/doc/kernel-package.

If possible, it is to your advantage to upgrade the kernel package separately from the main dist-upgrade to reduce the chances of a temporarily non-bootable system. Note that this should only be done after the minimal upgrade process described in Section 4.5.6, “Minimal system upgrade”.

lenny features a more robust mechanism for hardware discovery than previous releases. However, this may cause changes in the order devices are discovered on your system, affecting the order in which device names are assigned. For example, if you have two network adapters that are associated with two different drivers, the devices eth0 and eth1 refer to may be swapped. Please note that the new mechanism means that if you e.g. exchange ethernet adapters in a running lenny system, the new adapter will also get a new interface name.

For network devices, you can avoid this reordering by using udev rules, more specifically, through the definitions at /etc/udev/rules.d/70-persistent-net.rules^[4]. Alternatively you can use the ifrename utility to bind physical devices to specific names at boot time. See ifrename(8) and iftab(5) for more information. The two alternatives (udev and ifrename) should not be used at the same time.

For storage devices, you can avoid this reordering by using initramfs-tools and configuring it to load storage device driver modules in the same order they are currently loaded. To do this, identify the order the storage modules on your system were loaded by looking at the output of lsmod. lsmod lists modules in the reverse order that they were loaded in, i.e., the first module in the list was the last one loaded. Note that this will only work for devices which the kernel enumerates in a stable order (like PCI devices).

However, removing and reloading modules after initial boot will affect this order. Also, your kernel may have some drivers linked statically, and these names will not appear in the output of lsmod. You may be able to decipher these driver names and load order from looking at /var/log/kern.log, or the output of dmesg.

Add these module names to /etc/initramfs-tools/modules in the order they should be loaded at boot time. Some module names may have changed between etch and lenny. For example, sym53c8xx_2 has become sym53c8xx.

You will then need to regenerate your initramfs image(s) by executing update-initramfs -u -k all.

Once you are running a lenny kernel and udev, you may reconfigure your system to access disks by an alias that is not dependent upon driver load order. These aliases reside in the /dev/disk/ hierarchy.

If an initrd created with initramfs-tools is used to boot the system, in some cases the creation of device files by udev can happen too late for the boot scripts to act on.

The usual symptoms are that the boot will fail because the root file system cannot be mounted and you are dropped into a debug shell. But if you check afterwards, all devices that are needed are present in /dev. This has been observed in cases where the root file system is on a USB disk or on RAID, especially if LILO is used.

A workaround for this issue is to use the boot parameter rootdelay=9. The value for the timeout (in seconds) may need to be adjusted.

If you are using lilo as your bootloader (it is the default bootloader for some installations of etch) it is strongly recommended that you rerun lilo after the upgrade:

# /sbin/lilo

Notice this is needed even if you did not upgrade your system's kernel, as lilo's second stage will change due to the package upgrade.

Also, review the contents of your /etc/kernel-img.conf and make sure that you have do_bootloader = Yes in it. That way the bootloader will always be rerun after a kernel upgrade.

If you encounter any issues when running lilo, review the symbolic links in / to vmlinuz and initrd and the contents of your /etc/lilo.conf for discrepancies.

If you forgot to rerun lilo before the reboot or the system is accidentally rebooted before you could do this manually, your system might fail to boot. Instead of the lilo prompt, you will only see LI when booting the system^[5]. See Section 4.1.3, “Prepare for recovery” for information on how to recover from this.

One can avoid this problem entirely by using an identifier for the root filesystem that does not change from one boot to the next. There are two possible methods for doing this - labeling the filesystem, or using the filesystem's universally unique identifier (UUID). These methods are supported in Debian since the 'etch' release.

The two approaches have advantages and disadvantages. The labeling approach is more readable, but there may be problems if another filesystem on your machine has the same label. The UUID approach is uglier, but having two clashing UUIDs is highly unlikely.

For the examples below we assume the root filesystem is on /dev/hda6. We also assume your system has a working udev installation and ext2 or ext3 filesystems.

To implement the labeling approach:

To implement the UUID approach:

Some packages from etch have been split into several packages in lenny, often to improve system maintainability. To ease the upgrade path in such cases, lenny often provides “dummy” packages: empty packages that have the same name as the old package in etch with dependencies that cause the new packages to be installed. These “dummy” packages are considered obsolete packages after the upgrade and can be safely removed.

Most (but not all) dummy packages' descriptions indicate their purpose. Package descriptions for dummy packages are not uniform, however, so you might also find deborphan with the --guess options useful to detect them in your system. Note that some dummy packages are not intended to be removed after an upgrade but are, instead, used to keep track of the current available version of a program over time.