path: root/manage.sh

#!/usr/bin/env bash
# Top-level entry point for management.
# Run without args for the interactive menu.
set -euo pipefail

# Resolve paths relative to this script, not the caller's CWD
cd -- "$(dirname -- "$(readlink -f -- "${BASH_SOURCE[0]}")")"

die() { echo "❌ $*" >&2; exit 1; }

usage() {
  cat <<'EOF'
usage: manage.sh                 # interactive menu
       manage.sh add <HOST> [flags]
       manage.sh update <HOST> [switch|boot]
       manage.sh update-all
       manage.sh base            # rebuild base image, restore + template on proxmox

add flags:
  --cpu N     cores            (default 1)
  --mem MB    memory in MB     (default 1024)
  --disk SIZE e.g. 10G         (default 10G)
  --vlan TAG  network tag      (default 30)
  --tf       show plan, ask for approval, then tofu/terraform apply
  --ssh      wait for ssh to come up
  --deploy   nix/deploy.sh <HOST> jefe
  --all      shorthand for --tf --ssh --deploy
  --yes      non-interactive: auto-approve the plan, skip prompts
EOF
}

hosts() { ls nix/per-host | grep -v '^base$'; }

confirm() {
  local r=""
  read -rp "$1 [y/N]: " r || r=""
  [[ $r =~ ^[Yy]$ ]]
}

# Map a numeric selection (or a literal name) to a host name
resolve_host() {
  local input="$1" list=()
  readarray -t list < <(hosts)
  if [[ $input =~ ^[0-9]+$ ]]; then
    (( input >= 1 && input <= ${#list[@]} )) || die "invalid selection: $input (pick 1..${#list[@]})"
    printf '%s' "${list[input-1]}"
  else
    printf '%s' "$input"
  fi
}

pick_host() {
  local host="${1:-}"
  if [[ -z $host ]]; then
    local list=() i
    readarray -t list < <(hosts)
    {
      echo "Hosts:"
      for i in "${!list[@]}"; do printf ' %2d) %s\n' "$((i+1))" "${list[i]}"; done
    } >&2
    read -rp "Host name/number: " input || die "aborted"
    [[ -n $input ]] || die "aborted"
  fi
  host=$(resolve_host "${host:-$input}")
  [[ -d "nix/per-host/$host" ]] || die "no such host: '$host' (expected nix/per-host/$host)"
  printf '%s' "$host"
}

# ---- actions ----

cmd_update_host() {
  local host target
  host=$(pick_host "${1:-}")
  target="${2:-}"
  if [[ -z $target ]]; then
    read -rp "Deploy action [switch/boot, default switch]: " target || target=""
    target="${target:-switch}"
  fi
  [[ $target == switch || $target == boot ]] || die "action must be 'switch' or 'boot'"
  nix/deploy.sh "$host" jefe "$target"
}

cmd_update_all() {
  echo "→ updating flake inputs ..."
  (cd nix && nix flake update)
  local host
  for host in $(hosts); do
    if ! ssh -o ConnectTimeout=5 "$host" true 2>/dev/null; then
      echo "→ [$host] offline, skipping"
      continue
    fi
    # `boot` avoids the auto-switch restriction; the reboot activates it.
    if nix/deploy.sh "$host" jefe boot; then
      ssh "$host" sudo reboot || true
    else
      echo "→ [$host] deploy failed, not rebooting"
    fi
  done
}

cmd_update_base() {
  echo "→ building base image ..."
  (cd nix && nix flake update && nixos-rebuild build-image --image-variant proxmox --flake .#base)
  local img
  img=$(ls -t nix/result/*.vma.zst 2>/dev/null | head -n1) || die "no image found in nix/result/"
  echo "→ image: $(basename "$img")"
  echo "→ copying to lan.hesh:/var/lib/vz/dump/ ..."
  scp "$img" lan.hesh:/var/lib/vz/dump/

  # 104 is the clone source in TF (full clones, so no linked-clone breakage).
  confirm "→ Destroy template 104 and restore from this image?" || {
    echo "→ stopped after copy; restore manually"
    return 0
  }
  ssh lan.hesh 'qm destroy 104 --purge'
  ssh lan.hesh "qmrestore /var/lib/vz/dump/$(basename "$img") 104 --storage local-lvm"
  ssh lan.hesh 'qm template 104'
  echo "✅ template 104 updated"
}

# ---- add a new host (scaffold + optional full flow) ----

cmd_add() {
  local HOST="" CPU=1 MEM=1024 DISK="10G" VLAN=30
  local CPU_GIVEN=0 MEM_GIVEN=0 DISK_GIVEN=0 VLAN_GIVEN=0
  local RUN_TF=0 RUN_SSH=0 RUN_DEPLOY=0 AUTO_YES=0

  # Interactive menu: offer the full flow up front
  if (( $# == 0 )) && [[ -t 0 ]]; then
    local r=""
    read -rp "→ Run full flow (tofu apply, ssh wait, deploy)? [Y/n]: " r || r=""
    [[ $r =~ ^[Nn]$ ]] || { RUN_TF=1; RUN_SSH=1; RUN_DEPLOY=1; }
  fi

  while [[ $# -gt 0 ]]; do
    case "$1" in
      -h|--help) usage; return 0 ;;
      --tf) RUN_TF=1 ;;
      --ssh) RUN_SSH=1 ;;
      --deploy) RUN_DEPLOY=1 ;;
      --all) RUN_TF=1; RUN_SSH=1; RUN_DEPLOY=1 ;;
      -y|--yes) AUTO_YES=1 ;;
      --cpu)  [[ $# -ge 2 ]] || die "--cpu needs a value"; CPU=$2; CPU_GIVEN=1; shift ;;
      --mem)  [[ $# -ge 2 ]] || die "--mem needs a value"; MEM=$2; MEM_GIVEN=1; shift ;;
      --disk) [[ $# -ge 2 ]] || die "--disk needs a value"; DISK=$2; DISK_GIVEN=1; shift ;;
      --vlan) [[ $# -ge 2 ]] || die "--vlan needs a value"; VLAN=$2; VLAN_GIVEN=1; shift ;;
      -*) usage >&2; die "unknown flag: $1" ;;
      *) [[ -z $HOST ]] || die "unexpected argument: $1"; HOST=$1 ;;
    esac
    shift
  done

  if [[ -z $HOST ]]; then
    read -rp "Host name: " HOST
  fi
  [[ $HOST =~ ^[a-z0-9]([a-z0-9-]*[a-z0-9])?$ ]] || die "invalid host name: '$HOST'"
  [[ -d nix/per-host/$HOST && $HOST != base ]] && echo "→ note: nix/per-host/$HOST already exists"

  [[ $CPU =~ ^[0-9]+$ ]] || die "--cpu must be an integer"
  [[ $MEM =~ ^[0-9]+$ ]] || die "--mem must be an integer (MB)"
  [[ $VLAN =~ ^[0-9]+$ && $VLAN -le 4094 ]] || die "--vlan must be 0..4094"
  [[ $DISK =~ ^[0-9]+[KMGT]B?$ ]] || die "--disk must look like 10G / 100G / 512M"

  # Prompt (interactive only) for anything not given; defaults come from monitoring-2.
  if [[ -t 0 ]] && ! (( AUTO_YES )); then
    (( CPU_GIVEN ))  || { read -rp "CPU cores [$CPU]: " r  || r=""; CPU=${r:-$CPU}; }
    (( MEM_GIVEN ))  || { read -rp "Memory MB [$MEM]: " r  || r=""; MEM=${r:-$MEM}; }
    (( DISK_GIVEN )) || { read -rp "Disk size [$DISK]: " r || r=""; DISK=${r:-$DISK}; }
    (( VLAN_GIVEN )) || { read -rp "VLAN tag [$VLAN]: " r  || r=""; VLAN=${r:-$VLAN}; }
  fi

  local SNAKE="${HOST//-/_}"
  local SOPS_PREFIX
  SOPS_PREFIX=$(sed -E 's/-[0-9]+$//' <<<"$HOST")  # torrents-2 -> torrents (for path_regex)

  # ---- host keys + age key ----
  if [[ -f "nix/host-keys/$HOST/ssh_host_ed25519_key.pub" ]]; then
    echo "→ [$HOST] host keys exist, skipping generation"
  else
    nix/generate-keys.sh "$HOST" >/dev/null
    echo "→ [$HOST] generated host keys"
  fi

  if ! command -v ssh-to-age >/dev/null; then
    die "ssh-to-age not found (direnv should provide it via nix/.envrc)"
  fi
  local AGE_KEY
  AGE_KEY=$(ssh-to-age < "nix/host-keys/$HOST/ssh_host_ed25519_key.pub")
  [[ $AGE_KEY == age1* ]] || die "failed to derive age key for $HOST"

  # ---- .sops.yaml: key anchor ----
  if grep -q "&server_${SNAKE} " nix/.sops.yaml; then
    echo "→ [$HOST] .sops.yaml anchor exists, skipping"
  else
    awk -v line="  - &server_${SNAKE} ${AGE_KEY}" '
      !done && /^creation_rules:/ { print line; done=1 }
      { print }' nix/.sops.yaml > nix/.sops.yaml.tmp && mv nix/.sops.yaml.tmp nix/.sops.yaml
    echo "→ [$HOST] added age key to .sops.yaml"
  fi

  # ---- .sops.yaml: creation rule (extend an existing prefix rule if present) ----
  if grep -qE "^ +- \*server_${SNAKE}\$" nix/.sops.yaml; then
    echo "→ [$HOST] .sops.yaml creation rule exists, skipping"
  elif grep -q "path_regex: .secrets/${SOPS_PREFIX}\." nix/.sops.yaml; then
    awk -v prefix="$SOPS_PREFIX" -v newline="          - *server_${SNAKE}" '
      /^  - path_regex/ { inrule=0 }
      $0 ~ ("  - path_regex: .secrets/" prefix) { inrule=1 }
      inrule && /^          - / { last=NR }
      /^$/ { inrule=0 }
      { lines[NR]=$0 }
      END {
        for (i=1; i<=NR; i++) {
          print lines[i]
          if (i == last) print newline
        }
      }' nix/.sops.yaml > nix/.sops.yaml.tmp && mv nix/.sops.yaml.tmp nix/.sops.yaml
    echo "→ [$HOST] extended existing creation rule for 'secrets/${SOPS_PREFIX}.*'"
  else
    cat >> nix/.sops.yaml <<EOF

  - path_regex: 'secrets/${SOPS_PREFIX}.*\.yaml$'
    key_groups:
      - age:
          - *admin_me
          - *admin_me_2
          - *server_${SNAKE}
EOF
    echo "→ [$HOST] added creation rule for 'secrets/${SOPS_PREFIX}.*'"
  fi

  # ---- per-host config ----
  if [[ -f "nix/per-host/$HOST/configuration.nix" ]]; then
    echo "→ [$HOST] per-host config exists, skipping"
  else
    mkdir -p "nix/per-host/$HOST"
    cat > "nix/per-host/$HOST/configuration.nix" <<EOF
{ config, pkgs, modulesPath, lib, system, ... }:

{
  config = {
    sops.age.sshKeyPaths = [ "/etc/ssh/ssh_host_ed25519_key" ];
    sops.defaultSopsFile = ../../secrets/${HOST}.yaml;

    networking.hostName = "${HOST}";
  };
}
EOF
    echo "→ [$HOST] created nix/per-host/$HOST/configuration.nix"
  fi

  # ---- flake.nix: nixosConfiguration ----
  if grep -q "\"${HOST}\" = nixpkgs.lib.nixosSystem" nix/flake.nix; then
    echo "→ [$HOST] flake.nix entry exists, skipping"
  else
    local FLAKE_BLOCK
    FLAKE_BLOCK=$(cat <<EOF
      "${HOST}" = nixpkgs.lib.nixosSystem {
        system = "x86_64-linux";
        modules = [
          ./per-host/base/configuration.nix
          ./per-host/${HOST}/configuration.nix
          sops-nix.nixosModules.sops
        ];
      };
EOF
)
    awk -v block="$FLAKE_BLOCK" '
      { lines[NR]=$0 }
      END {
        # insert before the close of nixosConfigurations ("    };" + "  };"),
        # falling back to the last "    };"
        ins=0
        for (i=NR; i>=1; i--) {
          if (lines[i] == "    };" && (i == NR || lines[i+1] == "  };")) { ins=i; break }
        }
        for (i=1; i<=NR; i++) {
          if (i == ins) { print block; print "" }
          print lines[i]
        }
      }' nix/flake.nix > nix/flake.nix.tmp && mv nix/flake.nix.tmp nix/flake.nix
    echo "→ [$HOST] added nixosConfigurations entry to flake.nix"
  fi

  # ---- secrets ----
  if [[ -f "nix/secrets/$HOST.yaml" ]]; then
    echo "→ [$HOST] nix/secrets/$HOST.yaml exists, skipping"
  else
    command -v sops >/dev/null || die "sops not found (direnv should provide it via nix/.envrc)"
    printf 'foo: bar\n' > "nix/secrets/$HOST.yaml"
    (cd nix && sops -e -i "secrets/$HOST.yaml") || die "sops encrypt failed"
    echo "→ [$HOST] created nix/secrets/$HOST.yaml (foo: bar)"
  fi

  # ---- Terraform VM (only if it doesn't exist) ----
  local TF_CREATED=0
  if grep -q "proxmox_vm_qemu\" \"${SNAKE}\"" terraform/main.tf; then
    echo "→ [$HOST] TF resource proxmox_vm_qemu.${SNAKE} exists, skipping"
  else
    local ORDER
    ORDER=$(grep -E '^\s*order\s*=' terraform/main.tf \
              | grep -oE '[0-9]+' | sort -n | tail -n1)
    ORDER=$(( ${ORDER:-0} + 1 ))
    cat >> terraform/main.tf <<EOF

resource "proxmox_vm_qemu" "${SNAKE}" {
  target_node        = "hesh"
  name               = "${HOST}"
  memory             = ${MEM}
  start_at_node_boot = true
  boot               = "order=virtio0;net0"
  clone_id           = 104
  full_clone         = true
  scsihw             = "virtio-scsi-single"

  startup_shutdown {
    order = ${ORDER}
  }

  cpu {
    cores = ${CPU}
  }

  disk {
    type    = "disk"
    slot    = "virtio0"
    size    = "${DISK}"
    storage = "local-lvm"
    format  = "raw"
  }

  network {
    id     = 0
    model  = "virtio"
    bridge = "vmbr1"
    tag    = ${VLAN}
    mtu    = 1420
  }
}
EOF
    TF_CREATED=1
    echo "→ [$HOST] added proxmox_vm_qemu.${SNAKE} to terraform/main.tf (startup order ${ORDER})"
  fi

  # ---- ~/.ssh/config ----
  local SSH_CONFIG="${HOME}/.ssh/config"
  if [[ -f $SSH_CONFIG ]] && grep -qE "^Host ${HOST}\$" "$SSH_CONFIG"; then
    echo "→ [$HOST] ~/.ssh/config block exists, skipping"
  else
    mkdir -p "${HOME}/.ssh"
    touch "$SSH_CONFIG"
    cat >> "$SSH_CONFIG" <<EOF

Host ${HOST}
    User jefe
    ProxyJump  lan.labj
    # required for first ssh
    # remove after ./deploy.sh has run
    HostName nixos
    StrictHostKeyChecking no
    UserKnownHostsFile /dev/null
EOF
    echo "→ [$HOST] added ~/.ssh/config block"
  fi

  # ---- README Hosts section ----
  if grep -q "^### ${HOST}\$" nix/README.md; then
    echo "→ [$HOST] README section exists, skipping"
  else
    local PURPOSE=""
    if [[ -t 0 ]] && ! (( AUTO_YES )); then
      read -rp "Host purpose for README (blank = skip): " PURPOSE || PURPOSE=""
    fi
    cat >> nix/README.md <<EOF

### ${HOST}
EOF
    if [[ -n ${PURPOSE// /} ]]; then
      printf '\n%s\n' "$PURPOSE" >> nix/README.md
    fi
    echo "→ [$HOST] added README Hosts section"
  fi

  # ---- Phase: tofu plan/approve/apply (idempotent; plan is shown before apply) ----
  local TF_APPLIED=0 TF_CLEAN=0 TF_BLOCKED=0
  if (( RUN_TF )); then
    local TOFU=()
    if command -v tofu >/dev/null; then TOFU=(tofu)
    elif command -v terraform >/dev/null; then TOFU=(terraform)
    else
      die "tofu/terraform not on PATH (direnv provides it via nix/.envrc)"
    fi
    echo "→ [$HOST] running ${TOFU[0]} plan ..."
    rm -f terraform/tfplan   # clear any stale plan
    local PLAN_OUT
    PLAN_OUT=$( cd terraform &&
      # shellcheck disable=SC1091
      source con.env && "${TOFU[@]}" plan -out=tfplan ) || die "tofu plan failed"
    printf '%s\n' "$PLAN_OUT"

    if grep -q "No changes." <<<"$PLAN_OUT"; then
      echo "→ [$HOST] no infrastructure changes; apply not needed"
      TF_CLEAN=1
    else
      local approved=0
      if (( AUTO_YES )); then
        approved=1
      elif [[ -t 0 ]]; then
        read -rp "→ Apply the above plan? [y/N]: " r || r=""
        if [[ $r =~ ^[Yy]$ ]]; then approved=1; fi
      else
        echo "→ [$HOST] no tty available for approval; use --yes to auto-approve"
      fi
      if (( approved )); then
        echo "→ [$HOST] applying ..."
        ( cd terraform &&
          # shellcheck disable=SC1091
          source con.env && "${TOFU[@]}" apply tfplan ) || die "tofu apply failed"
        echo "→ [$HOST] apply done"
        TF_APPLIED=1
      else
        echo "→ [$HOST] apply declined"
        TF_BLOCKED=1
      fi
    fi
    rm -f terraform/tfplan
  fi

  # ---- Phase: wait for ssh (VM first boots as 'nixos' via the ~/.ssh/config workaround) ----
  if (( RUN_SSH && TF_BLOCKED )); then
    echo "→ [$HOST] skipping ssh phase (apply was declined)"
  elif (( RUN_SSH )); then
    local SSH_TRIES=30 SSH_WAIT=5 ok=0
    echo "→ [$HOST] waiting for ssh (up to $((SSH_TRIES * SSH_WAIT))s) ..."
    local i
    for i in $(seq 1 "$SSH_TRIES"); do
      if ssh -o ConnectTimeout=5 "$HOST" true 2>/dev/null; then ok=1; break; fi
      sleep "$SSH_WAIT"
    done
    (( ok )) || die "ssh to $HOST failed after $SSH_TRIES attempts"
    echo "→ [$HOST] ssh ok"
  fi

  # ---- Phase: deploy (nixos-rebuild switch is idempotent) ----
  if (( RUN_DEPLOY && TF_BLOCKED )); then
    echo "→ [$HOST] skipping deploy phase (apply was declined)"
  elif (( RUN_DEPLOY )); then
    echo "→ [$HOST] deploying ..."
    nix/deploy.sh "$HOST" jefe
    echo "→ [$HOST] deploy done"
  fi

  # ---- Summary ----
  echo ""
  echo "================ ${HOST} ================"
  echo "  age key:            ${AGE_KEY}"
  echo "  sops prefix:        secrets/${SOPS_PREFIX}.*.yaml"
  echo "  flake target:       .#${HOST}"
  echo "  per-host config:    nix/per-host/${HOST}/configuration.nix"
  echo "  secrets:            nix/secrets/${HOST}.yaml (foo: bar)"
  if (( TF_CREATED )); then
    echo "  TF resource:        proxmox_vm_qemu.${SNAKE} (created: cpu=${CPU} mem=${MEM} disk=${DISK} vlan=${VLAN})"
  else
    echo "  TF resource:        proxmox_vm_qemu.${SNAKE} (already in TF)"
  fi
  echo ""
  echo "Next steps:"
  if (( TF_CREATED && ! RUN_TF )) || (( RUN_TF && ! TF_APPLIED && ! TF_CLEAN )); then
    cat <<EOF

  cd terraform
  source con.env
  tofu plan -out=tfplan
  tofu apply tfplan
EOF
  fi
  if (( ! RUN_SSH )); then
    cat <<EOF

  # test ssh (first boot uses the base image hostname 'nixos')
  ssh ${HOST} true && echo ok
EOF
  fi
  if (( ! RUN_DEPLOY )); then
    cat <<EOF

  # deploy
  nix/deploy.sh ${HOST} jefe
EOF
  fi
  cat <<EOF

  # then remove the StrictHostKeyChecking/UserKnownHostsFile workaround
  # from ~/.ssh/config for ${HOST}
EOF
}

# ---- menu ----

menu() {
  while true; do
    echo ""
    echo "=== home-lab manager ==="
    echo "  1) add a new host"
    echo "  2) update an existing host"
    echo "  3) update all hosts"
    echo "  4) update base image + template"
    echo "  q) quit"
    local choice=""
    read -rp "> " choice || exit 0
    case "$choice" in
      # subshells: a die/validation failure inside an action must not kill the menu
      1) ( cmd_add ) || echo "✗ add failed" ;;
      2) ( cmd_update_host ) || echo "✗ update failed" ;;
      3) ( cmd_update_all ) || echo "✗ update-all failed" ;;
      4) ( cmd_update_base ) || echo "✗ base update failed" ;;
      q|quit) exit 0 ;;
      "") ;;
      *) echo "? unknown choice: $choice" ;;
    esac
  done
}

case "${1:-}" in
  "" ) menu ;;
  -h|--help|help) usage ;;
  add) shift; cmd_add "$@" ;;
  update) shift; cmd_update_host "$@" ;;
  update-all) cmd_update_all ;;
  base) cmd_update_base ;;
  *) usage >&2; die "unknown command: $1" ;;
esac