# We built on top of tiny-cloud . /lib/tiny-cloud/common IMDS_ENDPOINT="169.254.169.254" . /lib/tiny-cloud/cloud/"$CLOUD"/imds _imds() { wget --quiet --timeout 1 --output-document - \ --header "$(_imds_header)" \ "http://$IMDS_ENDPOINT/$IMDS_URI/$1$IMDS_QUERY" } # Todo: This should go into a yaml file query_imds() { MAC=$(_imds meta-data/mac) AVAILABILITY_ZONE=$(_imds meta-data/placement/availability-zone) REGION=$(echo ${AVAILABILITY_ZONE} | sed "s/[a-z]$//") INSTANCE_ID=$(_imds meta-data/instance-id) cat <> /var/lib/cloud/meta-data AVAILABILITY_ZONE=$AVAILABILITY_ZONE REGION=$REGION INSTANCE_ID=$INSTANCE_ID IP_ADDRESS=$(_imds meta-data/local-ipv4) PUBLIC_IP_ADDRESS=$(_imds meta-data/public-ipv4 || true) DEFAULT_GW_INTERFACE=$(ip -o route get 8.8.8.8 | awk '{print $5}') MAC=$MAC VPC_CIDR_RANGE=$(_imds meta-data/network/interfaces/macs/${MAC}/vpc-ipv4-cidr-block) SUBNET=$(_imds meta-data/network/interfaces/macs/${MAC}/subnet-ipv4-cidr-block) _META_HOSTNAME=$(_imds meta-data/hostname) DOMAIN_NAME=\${_META_HOSTNAME#*.} AWS_ACCOUNT_ID=$(_imds meta-data/network/interfaces/macs/${MAC}/owner-id) INSTANCE_LIFE_CYCLE=$(_imds meta-data/instance-life-cycle) INSTANCE_TYPE=$(_imds meta-data/instance-type) EOF } # Todo: This should go into a yaml file get_tags() { # via metadata AWS restricts tags to NOT have " " or "/" ;-( # Replace all /:.- with _ for valid variable names for key in $(_imds meta-data/tags/instance); do value="$(_imds meta-data/tags/instance/$key)" key=$(echo ${key//[\/:.-]/_} | tr '[:lower:]' '[:upper:]') echo "$key=\"$value\"" >> /var/lib/cloud/meta-data done #while read _key value; do # key=$(echo ${_key//[\/:.-]/_} | tr '[:lower:]' '[:upper:]') # echo "$key=\"$value\"" >> /var/lib/cloud/meta-data #done < <(aws ec2 describe-tags --filters "Name=resource-id,Values=${INSTANCE_ID}" --query 'Tags[*].[Key,Value]' --region $REGION --output text) } # extract user-data args and cloud meta-data into /var/lib/cloud/meta-data get_meta_data() { if [ ! -f /var/lib/cloud/meta-data ]; then echo '#!/bin/bash' > /var/lib/cloud/meta-data query_imds get_tags fi if [ ! -f /etc/cloudbender.conf ]; then bash /var/lib/cloud/user-data extract_parameters fi } import_meta_data() { . /etc/cloudbender.conf . /var/lib/cloud/meta-data export AWS_DEFAULT_REGION=$REGION export AWS_DEFAULT_OUTPUT=text # some basic logic if [ "$DEBUG" == "None" -o "$DEBUG" == "False" ]; then unset DEBUG LAUNCH_HOOK="CloudBenderLaunchHook" fi # Workaround for current CFN ASG_ hack _key=$(echo $AWS_CLOUDFORMATION_LOGICAL_ID | tr '[:lower:]' '[:upper:]') [ -n "$(eval echo \$${_key}_CUSTOMHOSTNAME)" ] && CUSTOMHOSTNAME="$(eval echo \$${_key}_CUSTOMHOSTNAME)" [ -n "$(eval echo \$${_key}_VOLUMES)" ] && VOLUMES="$(eval echo \$${_key}_VOLUMES)" return 0 } # setup_instance, various OS tweaks impossible to do via AMI baking setup_instance() { # create machine-id to emulate systemd [ -f /etc/machine-id ] || uuidgen > /etc/machine-id # add and mount bpf file system add_once /etc/fstab "bpffs /sys/fs/bpf bpf rw,nosuid,nodev,noexec,relatime,mode=700 0 0" mount -a # Ensure certain mounts are shared to run containers later, eg. cilium, falco mount --make-shared /sys/fs/cgroup mount --make-shared /sys/fs/bpf mount --make-shared /sys add_once /etc/hosts "${IP_ADDRESS} ${_META_HOSTNAME} ${HOSTNAME}" # Set system wide default region for boto3 echo "export AWS_DEFAULT_REGION=$REGION" > /etc/profile.d/aws.sh case "$CLOUD" in aws) setup_sns_alarms ;; *) ewarn "Unsupported Cloud: $CLOUD" # return 1 ;; esac } ################ # IAM SSH KEYS # ################ configure_sshd() { # Change Listen port local port=${SSHPORT:-"22"} [ -w /etc/ssh/sshd_config ] && sed -i -e 's/^[\s#]*Port\s.*/Port '$port'/' /etc/ssh/sshd_config case "$CLOUD" in aws) # on AWS call IAM for allowed groups and actual keys local group=${SSHKEYIAMGROUP:-""} local role=${SSHKEYIAMROLE:-"arn:aws:iam::000000000000:role/Undefined"} [ $role == "arn:aws:iam::000000000000:role/Undefined" ] && role="" if [ -n "$group" ]; then # Configure SSHD sed -i -e 's,^[\s#]*AuthorizedKeysCommand\s.*,AuthorizedKeysCommand /usr/sbin/get_iam_sshkeys.py --user %u --group '$group' --iamRole "'$role'",' /etc/ssh/sshd_config sed -i -e 's,^[\s#]*AuthorizedKeysCommandUser\s.*,AuthorizedKeysCommandUser nobody,' /etc/ssh/sshd_config ebegin "added $group to SSH admin keys" fi ;; *) ewarn "Unsupported Cloud: $CLOUD" # return 1 ;; esac } # Persist host keys persistent_sshd_hostkeys() { # Top level is artifact to be able to limit the SSM IAM permissions local ssm_path=$1 local key_types="dsa ecdsa ed25519 rsa" # If host keys exist on SSM try to download RET=0 for key in $key_types; do (aws ssm get-parameters --names "${ssm_path}/host_${key}.tgz" --with-decryption --query 'Parameters[0].Value' | base64 -d | tar xzf - --directory=/ 1>/dev/null 2>&1) \ && log -t user-data info "Restored ssh_host_${key}_key from SSM" || RET=1 done # Update keys if any key couldn't be restored from SSM if [ $RET -eq 1 ]; then for key in $key_types; do if [ -r /etc/ssh/ssh_host_${key}_key -a -r /etc/ssh/ssh_host_${key}_key.pub ]; then (aws ssm put-parameter --name "${ssm_path}/host_${key}.tgz" --type SecureString --value \ "$(tar czf - /etc/ssh/ssh_host_${key}_key /etc/ssh/ssh_host_${key}_key.pub | base64)" --overwrite) \ && log -t user-data info "Uploaded ssh_host_${key}_key to SSM" fi done fi } # either plain custom hostname or # - `unique:` eg. `uniq:kube-worker-{:02}` -> kube-worker-01 # - `myownip: ` eg. `myip: nodegroup-` -> nodegroup-1.2.3.4 set_hostname() { local custom_hostname=$(echo $1 | awk -F. '{ print $1 }') if [ -n "$custom_hostname" ]; then if [[ "$custom_hostname" == unique:* ]]; then new_hostname=$(uniq_hostname.py $AWS_EC2LAUNCHTEMPLATE_ID $INSTANCE_ID ${custom_hostname##unique:}) elif [[ "$custom_hostname" == myownip:* ]]; then local _ip=$(echo $IP_ADDRESS | sed -e 's/\./-/g') new_hostname=$(echo "${custom_hostname##myownip:}$_ip") else new_hostname=$custom_hostname fi FQDN="${new_hostname}.${DOMAIN_NAME}" echo ${new_hostname} > /etc/hostname hostname $new_hostname export HOSTNAME=$new_hostname # add new hostname to hosts add_once /etc/hosts "${IP_ADDRESS} ${FQDN} ${new_hostname}" log -t user-data info "Hostname updated to ${new_hostname}." # hup syslog to update loghost macro /etc/init.d/syslog-ng reload # update Route53 entry for VPC internal FQDN route53.py --fqdn $FQDN --record $IP_ADDRESS # update our Name Tag to FQDN or PrivateDNSName to allow easy indentification in the AWS UI aws ec2 create-tags --resources $INSTANCE_ID --tags Key=Name,Value=$FQDN else aws ec2 create-tags --resources $INSTANCE_ID --tags Key=Name,Value=${HOSTNAME}.${REGION}.compute.internal fi } # various early volume functions attach_ebs() { local volId="$1" local device="$2" local tries=30 while true; do _json="$(aws ec2 describe-volumes --volume-ids $volId --region $REGION --output json)" rc=$?; [ $rc -ne 0 ] && return $rc vol_status=$(echo "$_json" | jq -r .Volumes[].State) attachedId=$(echo "$_json" | jq -r .Volumes[].Attachments[].InstanceId) [ "$attachedId" = "$INSTANCE_ID" ] && break if [ "$vol_status" = "available" ]; then aws ec2 attach-volume --volume-id "$volId" --instance-id "$INSTANCE_ID" --region "$REGION" --device "$device" > /dev/null rc=$?; [ $rc -ne 0 ] && return $rc break fi # if attached but not to us -> detach if [ "$vol_status" = "in-use" ]; then aws ec2 detach-volume --volume-id "$volId" --region "$REGION" --force rc=$?; [ $rc -ne 0 ] && return $rc fi ((tries=tries-1)) [ $tries -eq 0 ] && return 1 sleep 5 done } _parse_volume() { # Todo: proper checks once all is yaml # For now just replace ':' echo $1 | sed -e 's/:/ /g' } # mount optional remote volumes mount_volumes() { local volumes="$1" for vol in $volumes; do # Todo: check volume type and call matching func read volType volId volDevice volPath < <(_parse_volume $vol) [ "$volType" != "ebs" ] && { echo "Unknown volume type $volType"; break; } attach_ebs $volId $volDevice rc=$? [ $rc -ne 0 ] && { ewarn "error trying to attach $volId"; break; } # wait for the block device to become available while true; do mdev -s test -b $volDevice && break sleep 1 done # check volume for existing filesystem type=$(file -Lbs $volDevice) if [[ "$type" =~ "XFS filesystem" ]]; then xfs_repair $volDevice >/dev/null 2>&1 else mkfs.xfs -qf $volDevice >/dev/null fi # mount mkdir -p $volPath mount -t xfs -o noatime $volDevice $volPath ebegin "mounting $volDevice at $volPath" done } unmount_volumes() { local volumes="$1" for vol in $volumes; do read volType volId volDevice volPath < <(_parse_volume $vol) umount $volPath && aws ec2 detach-volume --volume-id "$volId" --instance-id $INSTANCE_ID --region $REGION > /dev/null done } # msg used for sns event, last one wins msg() { MSG="$@"; log -t user-data info "$@"; } # Generic retry command wrapper, incl. timeout of 30s # $1 = number of tries; 0 = forever # $2 = number of seconds to sleep between tries # $@ actual command retry() { local tries=$1 local waitfor=$2 shift 2 while true; do # Only use timeout of $1 is an executable, call directly if function type -tf $1 >/dev/null && { timeout 30 $@ && return; } || { $@ && return; } ((tries=tries-1)) [ $tries -eq 0 ] && return 1 sleep $waitfor done } add_swap() { [ -f /.swapfile ] || { dd if=/dev/zero of=/.swapfile bs=1M count=$1 && chmod 600 /.swapfile && mkswap /.swapfile && swapon /.swapfile; } grep -q "/.swapfile" /etc/fstab || echo "/.swapfile none swap sw 0 0" >> /etc/fstab sysctl -w vm.swappiness=10 } # Get SSM secure string base64 decoded # $0 SSM_PATH, value to stdout get_secret() { aws ssm get-parameter --name ${1,,} --with-decryption --query 'Parameter.Value' | base64 -d } # Store values as base64 on SSM # $0 SSM_PATH VALUE put_secret() { aws ssm put-parameter --name ${1,,} --type SecureString --value "$(echo "$2" | base64 -w0)" --overwrite } # Gets existing passphrase or creates new passphrase and stores it init_passphrase() { local _URL=$1 local _PPFILE=$2 # If secret already exists noop [ -f $_PPFILE ] && return 0 get_secret $_URL > $_PPFILE && chmod 600 $_PPFILE || \ { xxd -l16 -p /dev/random > $_PPFILE; chmod 600 $_PPFILE; put_secret $_URL "$(cat $_PPFILE)"; } } asg_heartbeat() { [ -n "$LAUNCH_HOOK" ] && aws autoscaling record-lifecycle-action-heartbeat --instance-id $INSTANCE_ID --lifecycle-hook-name $LAUNCH_HOOK --auto-scaling-group-name $AWS_AUTOSCALING_GROUPNAME || true } setup_sns_alarms() { # store SNS message json template cat < /var/lib/cloud/sns_alarm.json { "Source": "CloudBender", "AWSAccountId": "$AWS_ACCOUNT_ID", "Region": "$REGION", "Artifact": "$ARTIFACT", "Asg": "$AWS_AUTOSCALING_GROUPNAME", "Instance": "$INSTANCE_ID", "ip": "$IP_ADDRESS" } EOF cat <<'EOF' > /var/lib/cloud/sns_alarm.sh #!/bin/bash SUBJECT=$1 MSG=$2 LEVEL=${3:-Info} ATTACHMENT=${4:-""} EMOJI=${5:-""} EOF if [ -n "$ALARMSNSARN" ]; then cat <> /var/lib/cloud/sns_alarm.sh jq -M --arg subject "\$SUBJECT" --arg level "\$LEVEL" --arg msg "\$MSG" --arg attachment "\$ATTACHMENT" --arg emoji "\$EMOJI" --arg hostname "\$HOSTNAME" '.Subject = \$subject | .Level = \$level | .Message = \$msg | .Attachment = \$attachment | .Emoji = \$emoji | .Hostname = \$hostname' < /var/lib/cloud/sns_alarm.json | sed -e 's/\\\\\\\\/\\\\/g' > /tmp/sns.json aws sns publish --region ${REGION} --target-arn $ALARMSNSARN --message file:///tmp/sns.json EOF fi chmod +x /var/lib/cloud/sns_alarm.sh } exit_trap() { set +e trap - ERR EXIT local ERR_CODE=$1 local ERR_LINE="$2" local ERR_FUNC="$3" local ERR_CMD="$4" if [ $ERR_CODE -ne 0 ]; then CFN_STATUS="FAILURE" RESULT="ABANDON" else CFN_STATUS="SUCCESS" RESULT="CONTINUE" fi # Add SNS events on demand if [ -n "$ALARMSNSARN" ]; then if [ $ERR_CODE -ne 0 ]; then LEVEL="Error" SUBJECT="Error during cloud-init." if [ $ERR_LINE -ne 1 ]; then MSG="$ERR_CMD failed in $ERR_FUNC at $ERR_LINE. Return: $ERR_CODE" ATTACHMENT="$(pr -tn $0 | tail -n+$((ERR_LINE - 3)) | head -n7)" else MSG="$ERR_CMD" fi if [ -n "$DEBUG" ]; then SUBJECT="$SUBJECT Instance kept running for debug." else SUBJECT="$SUBJECT Instance terminated by ASG lifecycle hook." fi else LEVEL="Info" SUBJECT="ZDT Alpine Instance launched." fi if [ -z "${DISABLECLOUDBENDERSNSSCALINGEVENTS}" ] || [ "$LEVEL" != "Info" ]; then /var/lib/cloud/sns_alarm.sh "$SUBJECT" "$MSG" "$LEVEL" "$ATTACHMENT" fi # Disable scaling events during shutdown [ -n "${DISABLECLOUDBENDERSNSSCALINGEVENTS}" ] && echo "DISABLE_SCALING_EVENTS=1" >> /etc/cloudbender.conf fi [ -n "$LAUNCH_HOOK" ] && aws autoscaling complete-lifecycle-action --lifecycle-action-result $RESULT --instance-id $INSTANCE_ID --lifecycle-hook-name $LAUNCH_HOOK --auto-scaling-group-name ${AWS_AUTOSCALING_GROUPNAME} || true if [ -n "${AWS_CLOUDFORMATION_LOGICAL_ID}" ]; then aws cloudformation signal-resource --stack-name ${AWS_CLOUDFORMATION_STACK_NAME} --logical-resource-id ${AWS_CLOUDFORMATION_LOGICAL_ID} --unique-id ${INSTANCE_ID} --status ${CFN_STATUS} fi # timestamp being done end_uptime=$(awk '{print $1}' < /proc/uptime) log -t user-data info "Exiting user-data. Duration: $(echo "$end_uptime-$start_uptime" | bc) seconds" # Shutdown / poweroff if we ran into error and not DEBUG [ $ERR_CODE -ne 0 -a -z "$DEBUG" ] && poweroff exit 0 } ### S3 based locking MUTEX=mutex MUTEX_OWNER=$HOSTNAME MUTEX_TIMEOUT=600 release_lock() { local S3LOCK=$1 rm -f $MUTEX aws s3 rm $S3LOCK } # Lock not timed out and we own it: 0 # Lock not timed out and someone else owns it: 1 # Lock timed out: 2 verify_lock() { local S3LOCK=$1 aws s3 cp $S3LOCK $MUTEX _host=$(grep "MUTEX_OWNER=" $MUTEX | sed -e 's/MUTEX_OWNER=//') _time=$(grep "MUTEX_TIME=" $MUTEX | sed -e 's/MUTEX_TIME=//') # Check for timestamp and timeout let timepassed=$(date -u +%s)-$_time [ $timepassed -gt $MUTEX_TIMEOUT ] && return 2 [ "$_host" == "$MUTEX_OWNER" ] && return 0 return 1 } aquire_lock() { local S3LOCK=$1 echo "MUTEX_OWNER=${MUTEX_OWNER}" > $MUTEX echo "MUTEX_TIME=$(date -u +%s)" >> $MUTEX aws s3 cp $MUTEX $S3LOCK # verify we actually got the lock sleep 2 verify_lock $S3LOCK } check_lock() { local S3LOCK=$1 aws s3 ls $S3LOCK && rc=$? || rc=$? # No LOCK ? -> get it ! if [ $rc -ne 0 ]; then aquire_lock $S3LOCK else verify_lock $S3LOCK && rc=$? || rc=$? # Lock timeout -> we get it if [ $rc -eq 2 ]; then aquire_lock $S3LOCK # we already own it elif [ $rc -eq 0 ]; then return 0 # someone else has a valid lock else return 1 fi fi } # All things networking enable_ip_forwarding() { local interface=$1 modprobe nf_conntrack cat < /etc/sysctl.d/40-ip-forward.conf net.ipv4.ip_forward=1 net.ipv4.ip_local_port_range = 1024 65535 net.ipv4.conf.$interface.send_redirects=0 net.ipv4.conf.all.rp_filter = 1 net.ipv4.conf.all.accept_redirects = 0 net.ipv6.conf.default.forwarding = 1 net.ipv6.conf.all.forwarding = 1 net.netfilter.nf_conntrack_max = 524288 EOF sysctl -p /etc/sysctl.d/40-ip-forward.conf } enable_iptables() { rc-update add iptables /etc/init.d/iptables save } register_service_dns() { if [ -n "$DNSZONE" -a -n "$SERVICENAME" ]; then _IP=${PUBLIC_IP_ADDRESS:-$IP_ADDRESS} [ -n "SERVICEPRIVATE" ] && _IP=$IP_ADDRESS route53.py --fqdn "${SERVICENAME}.${DNSZONE}" --record $_IP cat <> /etc/local.d/route53.stop echo "Deleting Route53 record for ${SERVICENAME}.${DNSZONE}" >> /tmp/shutdown.log route53.py --delete --fqdn "${SERVICENAME}.${DNSZONE}" --record ${PUBLIC_IP_ADDRESS:-$IP_ADDRESS} EOF chmod +x /etc/local.d/route53.stop # Short cut our public IP to private one to allow talking to our own service name add_once /etc/hosts "${IP_ADDRESS} ${SERVICENAME}.${DNSZONE}" fi } # associate EIP # return 0 if we attache EIP # return 1 if we public did NOT change associate_eip() { local instance_id=$1 local eip=$(echo $2 | sed -e 's/\/32//' | grep -E -o "(25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)\.(25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)\.(25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)\.(25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)") || true local current_instance if [ -n "$eip" ]; then if [ "$eip" != "0.0.0.0" ]; then read eip_alloc_id eip_assoc_id current_instance < <(aws ec2 describe-addresses --public-ips $eip --query 'Addresses[*].[AllocationId,AssociationId,InstanceId]' || true) # If we already own and have the EIP attached -> done [ "$instance_id" == "$current_instance" ] && return if [ ! -z "$eip_alloc_id" ]; then if [[ "$eip_assoc_id" =~ ^eipassoc- ]]; then log -t user-data info "EIP $eip already associated via Association ID ${eip_assoc_id}. Disassociating." retry 3 10 aws ec2 disassociate-address --association-id $eip_assoc_id fi log -t user-data info "Associating Elastic IP $eip via Allocation ID $eip_alloc_id with Instance $instance_id" aws ec2 associate-address --no-allow-reassociation --instance-id $instance_id --allocation-id $eip_alloc_id return else log -t user-data warn "Elastic IP $eip address not found." fi else log -t user-data info "0.0.0.0 requested, keeping AWS assigned IP." fi else log -t user-data debug "Invalid or no ElasticIP defined. Skip" fi return 1 } # Accept incoming traffic for everything disable_source_dest_check() { aws ec2 modify-instance-attribute --instance-id ${INSTANCE_ID} --source-dest-check "{\"Value\": false}" } # Register ourself at route tables register_routes() { local rtb_id_list=$1 local route_cidr=${2:-"0.0.0.0/0"} for cidr in ${route_cidr//,/ }; do for rt in ${rtb_id_list//,/ }; do [[ "$rt" =~ ^rtb-[a-f0-9]*$ ]] || { log -t user-data warn "Invalid Route Table ID: $rt"; return 1; } aws ec2 create-route --route-table-id $rt --destination-cidr-block "${cidr}" --instance-id ${INSTANCE_ID} || \ aws ec2 replace-route --route-table-id $rt --destination-cidr-block "${cidr}" --instance-id ${INSTANCE_ID} done done } setup_nat() { local mode=$1 # Masquerade all outgoing traffic iptables -t nat -A POSTROUTING -o $DEFAULT_GW_INTERFACE -s ${VPC_CIDR_RANGE} -j MASQUERADE } setup_fluentbit() { local token="cloudbender" if [[ $FLUENTDURL == *@* ]]; then token=${FLUENTDURL%%@*} FLUENTD_URL=${FLUENTDURL##*@} fi LOG_FILES=$FLUENTDLOGFILES # Add a local file based syslog parser which does not require Priority cat < /etc/fluent-bit/metadata.conf # add some AWS metadata [FILTER] Name record_modifier Match * Record source.ip $IP_ADDRESS Record source.instance_id $INSTANCE_ID Record source.region $REGION Record source.account $AWS_ACCOUNT_ID Record source.conglomerate $CONGLOMERATE Record source.artifact $ARTIFACT EOF # install logrotate fragment cat < /etc/logrotate.d/fluentbit /var/log/fluentbit.log { rotate 3 missingok notifempty compress maxsize 10M daily postrotate rc-service fluent-bit restart endscript } EOF rc-update add fluent-bit default rc-service fluent-bit start }