Why we do partitions the way we do,

how to get at the data when stuff blows up
and call the guest a guest, not host.

diff --git a/input/howto/install-kvm.creole b/input/howto/install-kvm.creole
index e648ec7..e4c760d 100644 (file)
--- a/input/howto/install-kvm.creole
+++ b/input/howto/install-kvm.creole
@@ -1,5 +1,9 @@
 == Setup a new kvm domain fast ==
 
+=== guest setup ===
+
+or: how to install Debian.
+
 Define a shell function
 
 {{{
@@ -7,7 +11,7 @@ Define a shell function
 setup_env() {
  echo -n "New hostname: " &&
  export LC_ALL=C &&
- read host &&
+ read guest &&
  target=/mnt/target &&
  mirror=`cat /etc/apt/sources.list /etc/apt/sources.list.d/debian.list 2>/dev/null | awk '/^deb.*debian/ {print $2; exit}'` &&
  vg=`vgdisplay -c | awk -F: '{print $1;exit}' | sed 's/ *//g'` &&
@@ -24,25 +28,35 @@ setup_env
 
 Then install base:
 
+The way we lay out the filesystems by default is that we have one 4g /
+filesystem, a swap, and a tiny boot filesystem.  On the host we make
+a new LVM logical volume for each of the three.  Only the LV that will
+take the guest's boot will actually be partitioned - into a single boot
+partition.  That's so we can install grub into the MBR and have the system
+start just like a real system.  Root and swap are put directly onto the
+logical volume, without partitioning it at all.  This makes getting to the
+data from the host easier - no need to run kpartx - and it makes growing
+trivial.
+
 {{{
 #######
 # install base
 
  apt-get install debootstrap kpartx &&
- lvcreate -L 128m -n "$host"-boot /dev/"$vg" &&
- lvcreate -L 4g -n "$host"-root /dev/"$vg" &&
- lvcreate -L 4g -n "$host"-swap /dev/"$vg" &&
+ lvcreate -L 128m -n "$guest"-boot /dev/"$vg" &&
+ lvcreate -L 4g -n "$guest"-root /dev/"$vg" &&
+ lvcreate -L 4g -n "$guest"-swap /dev/"$vg" &&
  : &&
- ( echo ',,L,*' | sfdisk /dev/"$vg"/"$host"-boot ) &&
- kpartx -v -a /dev/"$vg"/"$host"-boot &&
- mkfs.ext3 /dev/mapper/"$vg"-"$host"--boot1 &&
- mkfs.ext3 /dev/"$vg"/"$host"-root &&
- mkswap /dev/"$vg"/"$host"-swap &&
+ ( echo ',,L,*' | sfdisk /dev/"$vg"/"$guest"-boot ) &&
+ kpartx -v -a /dev/"$vg"/"$guest"-boot &&
+ mkfs.ext3 /dev/mapper/"$vg"-"$guest"--boot1 &&
+ mkfs.ext3 /dev/"$vg"/"$guest"-root &&
+ mkswap /dev/"$vg"/"$guest"-swap &&
  : &&
  mkdir -p "$target" &&
- mount /dev/"$vg"/"$host"-root "$target" &&
+ mount /dev/"$vg"/"$guest"-root "$target" &&
  mkdir -p "$target/boot" &&
- mount /dev/mapper/"$vg"-"$host"--boot1 "$target/boot" &&
+ mount /dev/mapper/"$vg"-"$guest"--boot1 "$target/boot" &&
 
  cd "$target" &&
  debootstrap --variant=minbase lenny . "$mirror"
@@ -53,7 +67,7 @@ And finalize the setup:
 {{{
 #######
 # finish setup
- echo "$host" > etc/hostname &&
+ echo "$guest" > etc/hostname &&
  cat > etc/hosts << EOF &&
 127.0.0.1       localhost
 
@@ -79,7 +93,7 @@ EOF
  chroot . apt-get install grub &&
  cp -av usr/lib/grub/x86_64-pc boot/grub &&
  grub << EOF &&
-device (hd0) /dev/mapper/$vg-$host--boot
+device (hd0) /dev/mapper/$vg-$guest--boot
 root (hd0,0)
 setup (hd0)
 quit
@@ -94,9 +108,9 @@ And a fstab and a boot loader config
 
 {{{
  # doesn't work: chroot . update-grub
- rootuuid=`vol_id /dev/"$vg"/"$host"-root "$target" | awk -F= '$1=="ID_FS_UUID" {print $2}'` &&
- swapuuid=`vol_id /dev/"$vg"/"$host"-swap "$target" | awk -F= '$1=="ID_FS_UUID" {print $2}'` &&
- bootuuid=`vol_id /dev/mapper/"$vg"-"$host"--boot1  | awk -F= '$1=="ID_FS_UUID" {print $2}'` &&
+ rootuuid=`vol_id /dev/"$vg"/"$guest"-root "$target" | awk -F= '$1=="ID_FS_UUID" {print $2}'` &&
+ swapuuid=`vol_id /dev/"$vg"/"$guest"-swap "$target" | awk -F= '$1=="ID_FS_UUID" {print $2}'` &&
+ bootuuid=`vol_id /dev/mapper/"$vg"-"$guest"--boot1  | awk -F= '$1=="ID_FS_UUID" {print $2}'` &&
  cat > boot/grub/menu.lst << EOF &&
 default 0
 timeout 5
@@ -141,18 +155,46 @@ And unmount:
   cd / &&
  umount "$target"/boot &&
  umount "$target" &&
- kpartx -v -d /dev/"$vg"/"$host"-boot &&
+ kpartx -v -d /dev/"$vg"/"$guest"-boot &&
  rmdir "$target"
 }}}
 
-Setup a new kvm domain by creating a new file in /etc/dm-virt/`hostname/$host.xml.
+=== virsh setup ===
+
+Setup a new kvm domain by creating a new file in /etc/dm-virt/`hostname/$guest.xml.
 
 * Properly configure hostname
 * Pick a new uuid ({{{uuidgen}}})
 * Setup block devices properly
 * pick a new and unique mac address (on d.o every kvm host has their own mac address space and the last block is changed for the guests, as in {{{..:..:..:..:<host byte>:<guest byte>}}}. )
 * virsh commands:
+** {{{virsh help}}}
 ** {{{virsh define foo.xml}}}
 ** {{{virsh start foo}}}
 ** {{{virsh destroy foo}}}
 ** {{{virsh vncdisplay foo}}} (and ssh -L 5900:localhost:<5900+x> $host  and  vnc localhost)
+
+
+=== when stuff goes wrong ===
+
+To get to the guest data from the host:
+
+{{{
+  setup_env
+ kpartx -v -a /dev/"$vg"/"$guest"-boot &&
+ mkdir -p "$target" &&
+ mount /dev/"$vg"/"$guest"-root "$target" &&
+ mkdir -p "$target/boot" &&
+ mount /dev/mapper/"$vg"-"$guest"--boot1 "$target/boot"
+}}}
+
+and once you're done:
+{{{
+ cd / &&
+ umount "$target"/boot &&
+ umount "$target" &&
+ kpartx -v -d /dev/"$vg"/"$guest"-boot &&
+ rmdir "$target"
+}}}
+
+Make sure that the filesystem isn't being mounted twice - i.e. never start the guest while the filesystems are mounted, and never mount them while the guest is running.