Switch to pull-and-image strategy: flash-one.sh, disable netboot TFTP

scripts/flash-one.sh

@@ -0,0 +1,96 @@
+#!/bin/bash
+# Flash one planck node's NVMe stick with a fresh Raspberry Pi OS.
+# Usage: flash-one.sh <nodename> <device>   e.g. flash-one.sh planck016 sda
+# Run as root (via pkexec). Wipes the target device completely.
+set -euo pipefail
+
+NODE=$1
+DEV=$2
+IMG=/tmp/opencode/raspios-lite.img
+PUBKEY=/home/adamcarr/.ssh/id_ed25519.pub
+PASSFILE=/home/adamcarr/code/personal/pi-swarm/configs/node-password.txt
+EEPROM_UPD=/tmp/opencode/netboot/pieeprom-revert.upd
+EEPROM_SIG=/tmp/opencode/netboot/pieeprom-revert.sig
+
+fail() { echo "ERROR: $*" >&2; exit 1; }
+
+# --- safety checks --------------------------------------------------------
+[[ "$NODE" =~ ^planck0(0[1-9]|1[0-9]|20)$ ]] || fail "bad node name: $NODE"
+[[ -b /dev/$DEV ]] || fail "/dev/$DEV is not a block device"
+[[ "$DEV" == nvme* ]] && fail "refusing NVMe device name (looks like internal disk)"
+grep -q "^/dev/$DEV" /proc/mounts && fail "device has mounted partitions, unplug/replug first"
+
+SIZE_BYTES=$(cat /sys/block/$DEV/size 2>/dev/null) || fail "cannot read /sys/block/$DEV/size"
+SIZE_GB=$(( SIZE_BYTES * 512 / 1024 / 1024 / 1024 ))
+(( SIZE_GB >= 200 && SIZE_GB <= 300 )) || fail "device is ${SIZE_GB}GB, expected ~240GB - wrong stick or wrong port?"
+
+MODEL=$(lsblk -dn -o MODEL /dev/$DEV | head -1)
+SERIAL=$(lsblk -dn -o SERIAL /dev/$DEV | head -1)
+echo "Target: /dev/$DEV  ${SIZE_GB}GB  model=$MODEL serial=$SERIAL"
+echo "Node:   $NODE"
+echo "WIPING THIS DEVICE IN 3 SECONDS (ctrl-c to abort)"
+sleep 3
+
+# --- write image -----------------------------------------------------------
+wipefs -a /dev/$DEV
+dd if=$IMG of=/dev/$DEV bs=4M conv=fsync status=progress
+partprobe /dev/$DEV
+sleep 1
+
+# --- grow root partition to fill the disk ---------------------------------
+echo ",+" | sfdisk --no-reread -N 2 /dev/$DEV
+partprobe /dev/$DEV
+sleep 1
+e2fsck -fp /dev/${DEV}2 >/dev/null
+resize2fs /dev/${DEV}2 >/dev/null
+
+# --- configure -------------------------------------------------------------
+B=/mnt/planck-boot
+R=/mnt/planck-root
+mkdir -p $B $R
+mount /dev/${DEV}1 $B
+mount /dev/${DEV}2 $R
+
+# enable ssh on first boot
+touch $B/ssh
+
+# planck020 only: its EEPROM is stuck in network-boot mode, these files
+# tell the bootloader to flip it back to normal on first power-up
+if [ "$NODE" = "planck020" ]; then
+    cp $EEPROM_UPD $B/pieeprom.upd
+    cp $EEPROM_SIG $B/pieeprom.sig
+fi
+
+# hostname
+echo "$NODE" > $R/etc/hostname
+sed -i "s/^127.0.1.1.*/127.0.1.1	$NODE/" $R/etc/hosts
+
+# timezone (match old cluster)
+rm -f $R/etc/localtime
+ln -s /usr/share/zoneinfo/America/Los_Angeles $R/etc/localtime
+echo "America/Los_Angeles" > $R/etc/timezone
+
+# user adamcarr (uid 1000), password from passfile, groups like the old nodes
+HASH=$(openssl passwd -6 "$(cat $PASSFILE)")
+echo "adamcarr:x:1000:1000:Adam Carr:/home/adamcarr:/bin/bash" >> $R/etc/passwd
+echo "adamcarr:$HASH:20000:0:99999:7:::" >> $R/etc/shadow
+echo "adamcarr:x:1000:" >> $R/etc/group
+awk -F: -v OFS=: '
+    BEGIN{split("sudo adm dialout cdrom audio video plugdev games users input render netdev",t," "); for(i in t) want[t[i]]=1}
+    want[$1] { if ($4=="") $4="adamcarr"; else $4=$4",adamcarr" }
+    {print}' $R/etc/group > $R/etc/group.new && mv $R/etc/group.new $R/etc/group
+mkdir -p $R/home/adamcarr/.ssh
+cp $PUBKEY $R/home/adamcarr/.ssh/authorized_keys
+chown -R 1000:1000 $R/home/adamcarr
+chmod 700 $R/home/adamcarr/.ssh
+chmod 600 $R/home/adamcarr/.ssh/authorized_keys
+echo "adamcarr ALL=(ALL) NOPASSWD: ALL" > $R/etc/sudoers.d/010-adamcarr
+chmod 440 $R/etc/sudoers.d/010-adamcarr
+
+sync
+umount $B $R
+sync
+eject /dev/$DEV 2>/dev/null || true
+
+echo ""
+echo "DONE: $NODE flashed. Safe to unplug now."