2
0
mirror of https://github.com/xcat2/confluent.git synced 2026-11-19 08:52:20 +00:00
Files
confluent/confluent_osdeploy/el9-diskless/profiles/default/scripts/imageboot.sh
T
Markus Hilger 5473f46cf3 Do not autoactivate LVM on local disks on diskless boot
An old stateful OS on LVM was activated at diskless boot and kept the
disk busy, so repartitioning it for scratch failed. Set an empty
auto_activation_volume_list unless the image sets one. Ubuntu also
needs it in the initramfs, whose udev rules activate LVM earlier.
2026-11-09 21:35:40 +01:00

274 lines
13 KiB
Bash

. /lib/dracut-lib.sh
confluent_whost=$confluent_mgr
if [[ "$confluent_whost" == *:* ]] && [[ "$confluent_whost" != "["* ]]; then
confluent_whost="[$confluent_mgr]"
fi
mkdir -p /mnt/remoteimg /mnt/remote /mnt/overlay /sysroot
imagemethod=$(getarg confluent_imagemethod)
if grep installtodisk /proc/cmdline > /dev/null; then
# installimage copies from the mounted image, which an untethered or
# uncompressed boot would extract to RAM and unmount first
imagemethod=tethered
fi
if [ "untethered" = "$imagemethod" -o "uncompressed" = "$imagemethod" ]; then
mount -t tmpfs untethered /mnt/remoteimg
# Fall back to the other managers when the deployment server fails
while ! for url in https://$confluent_whost/confluent-public/os/$confluent_profile/rootimg.sfs $confluent_urls; do curl -f --connect-timeout 10 --speed-limit 1 --speed-time 30 $url -o /mnt/remoteimg/rootimg.sfs && break; done; do
echo "Failed downloading image, retrying..."
sleep $((RANDOM%3)).$((RANDOM))
done
else
confluent_urls="$confluent_urls https://$confluent_whost/confluent-public/os/$confluent_profile/rootimg.sfs"
/opt/confluent/bin/urlmount $confluent_urls /mnt/remoteimg
fi
/opt/confluent/bin/confluent_imginfo /mnt/remoteimg/rootimg.sfs > /tmp/rootimg.info
loopdev=$(losetup -f)
export mountsrc=$loopdev
losetup -r $loopdev /mnt/remoteimg/rootimg.sfs
if grep '^Format: confluent_crypted' /tmp/rootimg.info > /dev/null; then
while ! curl -sf -H "CONFLUENT_NODENAME: $nodename" -H "CONFLUENT_APIKEY: $(cat /etc/confluent/confluent.apikey)" https://$confluent_whost/confluent-api/self/profileprivate/pending/rootimg.key > /tmp/rootimg.key; do
echo "Unable to retrieve private key from $confluent_mgr (verify that confluent can access /var/lib/confluent/private/os/$confluent_profile/pending/rootimg.key)"
sleep 1
done
cipher=$(head -n 1 /tmp/rootimg.key)
key=$(tail -n 1 /tmp/rootimg.key)
len=$(wc -c /mnt/remoteimg/rootimg.sfs | awk '{print $1}')
len=$(((len-4096)/512))
dmsetup create cryptimg --table "0 $len crypt $cipher $key 0 $loopdev 8"
/opt/confluent/bin/confluent_imginfo /dev/mapper/cryptimg > /tmp/rootimg.info
mountsrc=/dev/mapper/cryptimg
fi
rootimgformat=$(awk -F': ' '/^Format:/ {print $2; exit}' /tmp/rootimg.info)
if [ "$rootimgformat" = squashfs ]; then
mount -o ro $mountsrc /mnt/remote
elif [ "$rootimgformat" = confluent_multisquash ]; then
tail -n +3 /tmp/rootimg.info | awk '{gsub("/", "_"); print "echo 0 " $4 " linear '$mountsrc' " $3 " | dmsetup create mproot" $7}' > /tmp/setupmount.sh
. /tmp/setupmount.sh
cat /tmp/setupmount.sh |awk '{printf "mount /dev/mapper/"$NF" "; sub("mproot", ""); gsub("_", "/"); print "/mnt/remote"$NF}' > /tmp/mountparts.sh
. /tmp/mountparts.sh
fi
#mount -t tmpfs overlay /mnt/overlay
if [ ! "uncompressed" = "$imagemethod" ]; then
modprobe zram
memtot=$(grep ^MemTotal: /proc/meminfo|awk '{print $2}')
memtot=$((memtot/2))$(grep ^MemTotal: /proc/meminfo | awk '{print $3'})
echo $memtot > /sys/block/zram0/disksize
mkfs.xfs /dev/zram0 > /dev/null
fi
TETHERED=0
if [ "untethered" = "$imagemethod" -o "uncompressed" = "$imagemethod" ]; then
if [ "untethered" = "$imagemethod" ]; then
mount -o discard /dev/zram0 /sysroot
else
mount -t tmpfs disklessroot /sysroot
fi
if ! extract_untethered_rootimg; then
# a partly extracted root would boot into a broken system
echo "Unable to boot: the root filesystem could not be extracted into memory, the image may be too large for an untethered boot on this node"
while :; do sleep 86400; done
fi
else
TETHERED=1
mount -o discard /dev/zram0 /mnt/overlay
if [ ! -f /tmp/mountparts.sh ]; then
mkdir -p /mnt/overlay/upper /mnt/overlay/work
mount -t overlay -o upperdir=/mnt/overlay/upper,workdir=/mnt/overlay/work,lowerdir=/mnt/remote disklessroot /sysroot
else
for srcmount in $(cat /tmp/mountparts.sh | awk '{print $3}'); do
mkdir -p /mnt/overlay${srcmount}/upper /mnt/overlay${srcmount}/work
mount -t overlay -o upperdir=/mnt/overlay${srcmount}/upper,workdir=/mnt/overlay${srcmount}/work,lowerdir=${srcmount} disklesspart /sysroot${srcmount#/mnt/remote}
done
fi
fi
mkdir -p /sysroot/etc/ssh
mkdir -p /sysroot/etc/confluent
mkdir -p /sysroot/root/.ssh
cp /root/.ssh/* /sysroot/root/.ssh
chmod 700 /sysroot/root/.ssh
cp /etc/confluent/* /sysroot/etc/confluent/
cp /etc/ssh/*key* /sysroot/etc/ssh/
for pubkey in /etc/ssh/ssh_host*key.pub; do
certfile=${pubkey/.pub/-cert.pub}
privfile=${pubkey%.pub}
if [ -s $certfile ]; then
echo HostCertificate $certfile >> /sysroot/etc/ssh/sshd_config
fi
echo HostKey $privfile >> /sysroot/etc/ssh/sshd_config
done
mkdir -p /sysroot/dev /sysroot/sys /sysroot/proc /sysroot/run
if [ ! -z "$autocons" ]; then
autocons=${autocons%,*}
mkdir -p /run/systemd/generator/getty.target.wants
ln -s /usr/lib/systemd/system/serial-getty@.service /run/systemd/generator/getty.target.wants/serial-getty@${autocons}.service
fi
while [ ! -e /sysroot/sbin/init ]; do
echo "Failed to access root filesystem or it is missing /sbin/init"
echo "System should be accessible through ssh at port 2222 with the appropriate key"
while [ ! -e /sysroot/sbin/init ]; do
sleep 1
done
done
rootpassword=$(grep ^rootpassword: /etc/confluent/confluent.deploycfg)
rootpassword=${rootpassword#rootpassword: }
if [ "$rootpassword" = "null" ]; then
rootpassword=""
fi
if [ ! -z "$rootpassword" ]; then
sed -i "s@root:[^:]*:@root:$rootpassword:@" /sysroot/etc/shadow
fi
for i in /ssh/*.ca; do
echo '@cert-authority *' $(cat $i) >> /sysroot/etc/ssh/ssh_known_hosts
done
echo HostbasedAuthentication yes >> /sysroot/etc/ssh/sshd_config
echo HostbasedUsesNameFromPacketOnly yes >> /sysroot/etc/ssh/sshd_config
echo IgnoreRhosts no >> /sysroot/etc/ssh/sshd_config
sshconf=/sysroot/etc/ssh/ssh_config
if [ -d /sysroot/etc/ssh/ssh_config.d/ ]; then
sshconf=/sysroot/etc/ssh/ssh_config.d/01-confluent.conf
fi
echo 'Host *' >> $sshconf
echo ' HostbasedAuthentication yes' >> $sshconf
echo ' EnableSSHKeysign yes' >> $sshconf
echo ' HostbasedKeyTypes *ed25519*' >> $sshconf
curl -sf -H "CONFLUENT_NODENAME: $nodename" -H "CONFLUENT_APIKEY: $(cat /etc/confluent/confluent.apikey)" https://$confluent_whost/confluent-api/self/nodelist > /sysroot/etc/ssh/shosts.equiv
cp /sysroot/etc/ssh/shosts.equiv /sysroot/root/.shosts
if grep ^ssh_keys: /etc/group > /dev/null; then
chmod 640 /sysroot/etc/ssh/*_key
chroot /sysroot chgrp ssh_keys /etc/ssh/*_key
fi
cp /tls/*.pem /sysroot/etc/pki/ca-trust/source/anchors/
chroot /sysroot/ update-ca-trust
curl -sf https://$confluent_whost/confluent-public/os/$confluent_profile/scripts/onboot.service > /sysroot/etc/systemd/system/onboot.service
mkdir -p /sysroot/opt/confluent/bin
curl -sf https://$confluent_whost/confluent-public/os/$confluent_profile/scripts/onboot.sh > /sysroot/opt/confluent/bin/onboot.sh
chmod +x /sysroot/opt/confluent/bin/onboot.sh
cp /opt/confluent/bin/apiclient /sysroot/opt/confluent/bin
set_selinux_context system_u:object_r:systemd_unit_file_t:s0 /sysroot/etc/systemd/system/onboot.service
set_selinux_context system_u:object_r:bin_t:s0 /sysroot/opt/confluent/bin/onboot.sh
set_selinux_context system_u:object_r:bin_t:s0 /sysroot/opt/confluent/bin/apiclient
ln -s /etc/systemd/system/onboot.service /sysroot/etc/systemd/system/multi-user.target.wants/onboot.service
cp /etc/confluent/functions /sysroot/etc/confluent/functions
# give every node its own stable machine-id instead of the one baked into the
# image, which every node booting it would otherwise share (the initramfs has
# openssl but no sha512sum)
machineid=$( (echo $nodename; cat /sys/devices/virtual/dmi/id/product_uuid 2> /dev/null) | openssl dgst -sha512 -r | head -c 32)
if [ ${#machineid} -eq 32 ]; then
echo $machineid > /sysroot/etc/machine-id
chmod 444 /sysroot/etc/machine-id
fi
# Do not autoactivate LVM on local disks unless the image sets auto_activation_volume_list
if [ -e /sysroot/etc/lvm/lvm.conf ] && ! grep -qs '^[[:space:]]*auto_activation_volume_list' /sysroot/etc/lvm/lvm.conf /sysroot/etc/lvm/lvmlocal.conf; then
printf 'activation {\n\tauto_activation_volume_list = []\n}\n' >> /sysroot/etc/lvm/lvmlocal.conf
fi
rm -f /sysroot/etc/dracut.conf.d/diskless.conf # remove diskless dracut from runtime, to make kdump happier
if [ -x /sysroot/usr/sbin/matchpathcon ] && grep -Eq '^SELINUX=(enforcing|permissive)' /sysroot/etc/selinux/config 2> /dev/null; then
# Files written to /sysroot before policy load are unlabeled.
selinuxpolicy=$(sed -n 's/^SELINUXTYPE=//p' /sysroot/etc/selinux/config)
fcontexts=/etc/selinux/$selinuxpolicy/contexts/files/file_contexts
if [ $TETHERED -eq 1 ]; then
# Label only the upper layer, using globstar since the initramfs has no find.
shopt -s globstar dotglob nullglob
relabel=()
if [ -f /tmp/mountparts.sh ]; then
overlayparts=$(awk '{print $3}' /tmp/mountparts.sh)
else
overlayparts=/mnt/remote
fi
for srcmount in $overlayparts; do
upperdir=/mnt/overlay/upper
if [ -f /tmp/mountparts.sh ]; then
upperdir=/mnt/overlay${srcmount}/upper
fi
prefix=${srcmount#/mnt/remote}
prefix=${prefix%/}
relabel+=("${prefix:-/}")
for relpath in "$upperdir"/**; do
relabel+=("$prefix${relpath#$upperdir}")
done
done
shopt -u globstar dotglob nullglob
# Not setfiles (recurses into the image) or restorecon (no-op before policy load).
chroot /sysroot /usr/sbin/matchpathcon -f $fcontexts "${relabel[@]}" | while IFS=$'\t' read -r relpath context; do
# deleted files leave whiteouts in the upper layer
if [ "$context" != "<<none>>" ] && { [ -e "/sysroot$relpath" ] || [ -L "/sysroot$relpath" ]; }; then
chcon -h "$context" "/sysroot$relpath"
fi
done
elif [ -x /sysroot/usr/bin/python3 ] && [ -x /sysroot/usr/sbin/setfiles ]; then
# The root is a local copy: label every inode without a label, which
# also covers sockets that unsquashfs creates without attributes.
# setfiles only recurses into directories that are new here.
chroot /sysroot /usr/bin/python3 -c '
import errno, os, sys
out = sys.stdout.buffer
rootdev = os.lstat(b"/").st_dev
def check(path):
try:
os.getxattr(path, b"security.selinux", follow_symlinks=False)
except OSError as e:
if e.errno == errno.ENODATA:
out.write(path + b"\0")
check(b"/")
for root, dirs, files in os.walk(b"/"):
for name in dirs + files:
check(os.path.join(root, name))
dirs[:] = [d for d in dirs if os.lstat(os.path.join(root, d)).st_dev == rootdev]
' | chroot /sysroot /usr/sbin/setfiles -0 -F -f - $fcontexts
relabelstatus=("${PIPESTATUS[@]}")
if [ "${relabelstatus[0]}" != 0 ] || [ "${relabelstatus[1]}" != 0 ]; then
echo "Warning: SELinux labeling of the root filesystem failed"
fi
else
echo "Warning: python3 or setfiles missing in the image, files written at boot stay unlabeled"
fi
fi
if grep installtodisk /proc/cmdline > /dev/null; then
. /etc/confluent/functions
run_remote installimage
exec reboot -f
fi
mv /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs
ln -s /sysroot/lib/modules/$(uname -r) /lib/modules/
if [ -d /lib/firmware ]; then
mv /lib/firmware /lib/firmware-ramfs
fi
ln -s /sysroot/lib/firmware /lib/firmware
kill $(grep -l ^/usr/lib/systemd/systemd-udevd /proc/*/cmdline|cut -d/ -f 3)
if grep debugssh /proc/cmdline >& /dev/null; then
debugssh=1
else
debugssh=0
fi
if [ $TETHERED -eq 1 ]; then
# In tethered mode, the double-caching is useful to get through tricky part of
# onboot with confignet. After that, it's excessive cache usage.
# Give the onboot script a hook to have us come in and enable directio to the
# squashfs and drop the cache of the rootimg so far
(
sleep 86400 &
ONBOOTPID=$!
mkdir -p /run/confluent
echo $ONBOOTPID > /run/confluent/onboot_sleep.pid
wait $ONBOOTPID
dd if=/mnt/remoteimg/rootimg.sfs iflag=nocache count=0 >& /dev/null
if [ $debugssh -eq 0 ]; then
rm -rf /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs /lib/firmware-ramfs /usr/lib64/libcrypto.so* /usr/lib64/systemd/ /kernel/ /usr/bin/ /usr/sbin/ /usr/libexec/
fi
) &
while [ ! -f /run/confluent/onboot_sleep.pid ]; do
sleep 0.1
done
elif [ $debugssh -eq 0 ]; then
rm -rf /lib/modules/$(uname -r) /lib/modules/$(uname -r)-ramfs /lib/firmware-ramfs /usr/lib64/libcrypto.so* /usr/lib64/systemd/ /kernel/ /usr/bin/ /usr/sbin/ /usr/libexec/
fi
if grep debugssh /proc/cmdline >& /dev/null; then
exec /opt/confluent/bin/start_root
else
exec /opt/confluent/bin/start_root -s # share mount namespace, keep kernel callbacks intact
fi