Start passthrough VM and its dependents by device identity, not address

Adding a PCIe card can renumber the buses behind it, so a hostpci entry
that names a PCI address can pass through the wrong device. These start
the VM only after finding each device by vendor/device/subsystem ID or
by its MAC-pinned interface, repoint its resource mapping if it moved,
and refuse to start on anything ambiguous. Its dependents then wait for
a port on it to answer before starting, so they find their shares up.

start-dependents.sh

@@ -0,0 +1,117 @@
+#!/bin/bash
+# Start the guests that depend on the passthrough VM, once it is serving.
+#
+# Usage:
+#   ./start-dependents.sh [--check]
+#
+# Run at boot by dependents-start.service, after passthrough-vm-start.service
+# has started the VM they need (a NAS, typically). The guests to start, in
+# order, are listed in conf/dependents.conf; give each of them onboot: 0 so that
+# only this starts them. VM and container IDs both work.
+#
+# Before starting anything it waits for WAIT_FOR (host:port) to accept a TCP
+# connection, for up to WAIT_SECS seconds, because a VM that has started is not
+# yet one that is serving. If it never answers the guests are started anyway,
+# with a warning, which is no worse than not waiting at all.
+#
+# A guest that fails to start does not stop the others; the script exits
+# non-zero at the end so the unit shows as failed. Guests already running are
+# left alone. Nothing is started while prox-maintenance-mode.sh has autostart
+# disabled.
+#
+# Options:
+#   --check   Show what would be started and test WAIT_FOR once; start nothing
+#
+# The config is looked for relative to this script, so this works both in a
+# checkout and deployed at /opt/ergosteur. Override with DEPENDENTS_CONFIG.
+
+set -uo pipefail
+
+# Print the header comment block: everything from line 2 up to the first blank line.
+usage() {
+    sed -n '2,/^$/p' "$0" | sed -e 's/^#//' -e 's/^ //'
+}
+
+SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+CONFIG=${DEPENDENTS_CONFIG:-$SCRIPT_DIR/conf/dependents.conf}
+MAINT_STATE=/var/tmp/proxmox-autostart-backup.txt  # written by prox-maintenance-mode.sh --enable
+CHECK_ONLY=false
+
+while [[ $# -gt 0 ]]; do
+    case $1 in
+        -h|--help) usage; exit 0 ;;
+        --check)   CHECK_ONLY=true; shift ;;
+        *) echo "Error: unknown option '$1'." >&2; usage >&2; exit 1 ;;
+    esac
+done
+
+[[ -r $CONFIG ]] || { echo "Error: cannot read $CONFIG" >&2; exit 1; }
+GUESTS=()
+WAIT_FOR=""
+WAIT_SECS=300
+# shellcheck source=conf/dependents.conf.example
+source "$CONFIG"
+
+if [[ -e $MAINT_STATE ]]; then
+    echo "maintenance mode is on ($MAINT_STATE exists); not starting anything"
+    exit 0
+fi
+if [[ ${#GUESTS[@]} -eq 0 ]]; then
+    echo "no GUESTS listed in $CONFIG; nothing to do"
+    exit 0
+fi
+
+# True once host:port accepts a TCP connection.
+answers() {
+    timeout 3 bash -c "exec 3<>/dev/tcp/${1%:*}/${1##*:}" 2>/dev/null
+}
+
+if [[ -n $WAIT_FOR ]]; then
+    if $CHECK_ONLY; then
+        if answers "$WAIT_FOR"; then
+            echo "$WAIT_FOR answers"
+        else
+            echo "$WAIT_FOR is not answering right now"
+        fi
+    else
+        echo "waiting up to ${WAIT_SECS}s for $WAIT_FOR"
+        deadline=$((SECONDS + WAIT_SECS))
+        until answers "$WAIT_FOR"; do
+            if (( SECONDS >= deadline )); then
+                echo "WARNING: $WAIT_FOR still not answering after ${WAIT_SECS}s; starting the guests anyway"
+                break
+            fi
+            sleep 5
+        done
+        answers "$WAIT_FOR" && echo "$WAIT_FOR answers"
+    fi
+fi
+
+failed=0
+for id in "${GUESTS[@]}"; do
+    if [[ -e /etc/pve/qemu-server/$id.conf ]]; then
+        tool=qm conf=/etc/pve/qemu-server/$id.conf
+    elif [[ -e /etc/pve/lxc/$id.conf ]]; then
+        tool=pct conf=/etc/pve/lxc/$id.conf
+    else
+        echo "ERROR: no VM or container $id on this node" >&2
+        failed=1
+        continue
+    fi
+    if grep -q '^onboot: 1' "$conf"; then
+        echo "WARNING: $id has onboot: 1, so Proxmox starts it before the VM it needs; set onboot: 0"
+    fi
+    if [[ $($tool status "$id" 2>/dev/null) == "status: running" ]]; then
+        echo "$id is already running"
+        continue
+    fi
+    if $CHECK_ONLY; then
+        echo "would start $id ($tool)"
+    elif $tool start "$id"; then
+        echo "started $id ($tool)"
+    else
+        echo "ERROR: $tool start $id failed" >&2
+        failed=1
+    fi
+done
+exit "$failed"