Establish build machine layout

1 files changed, 124 insertions, 1 deletions

diff --git a/buildos b/buildos
index 7aa9a0e..ec2be0b 100755
--- a/buildos
+++ b/buildos
@@ -14,7 +14,8 @@
 #
 owd="$(pwd)"
 trap "{ cd '$owd'; exit 1; }" ERR
-set -o errtrace # Trap in functions.
+set -o errtrace   # Trap in functions.
+shopt -s nullglob # Expand patterns than don't match to empty.
 
 # Note: diagnostics goes to stdout.
 #
@@ -320,6 +321,128 @@ if [ -z "$tc" ]; then
   info "no buildos.toolchain_url specified, not bootstrapping"
 fi
 
+# Cleanup /build/machines/ of any stray snapshots or deleted machines.
+#
+diag=()
+fail=
+for v in /build/machines/*; do
+  if [ ! -d "$v" ]; then
+    diag+=("$v: error: invalid volume")
+    fail="true"
+    continue
+  fi
+
+  cd "$v"
+
+  for m in *; do
+    if [ ! -d "$m" ]; then
+      diag+=("$v/$m: error: invalid machine")
+      fail="true"
+      continue
+    fi
+
+    cd "$m"
+
+    # If there is no current machine symlink, then delete the whole thing.
+    #
+    d=
+    if [ ! -L "$m" ]; then
+      d="true"
+    fi
+
+    # Examine each machine subvolume.
+    #
+    for s in "$m"-*; do
+      if [ ! -d "$s" ]; then
+	diag+=("$v/$m/$s: error: invalid machine subvolume")
+	fail="true"
+	continue
+      fi
+
+      # Unless we are deleting the whole thing, keep initial and bootstrapped
+      # (for known toolchains) subvolumes.
+      #
+      if [ -z "$d" ]; then
+	# <name>-<N>
+	#
+	if [[ "$s" =~ ^"$m"-[0-9]+$ ]]; then
+	  continue
+	fi
+
+	# <name>-<toolchain>
+	#
+	f=
+	for tn in "${!toolchains[@]}"; do
+	  if [[ "$s" =~ ^"$m"-"$tn"$ ]]; then
+	    f="true"
+	    break
+	  fi
+	done
+
+	if [ -n "$f" ]; then
+	  continue
+	fi
+      fi
+
+      # This is either a stray working submodule or a bootsrapped subvolume
+      # for a toolchain that was deleted (or we are deleting everything).
+      #
+      if ! btrfs property set -ts "$s" ro false; then
+	diag+=("$v/$m/$s: error: unable to change subvolume property")
+	fail="true"
+	continue
+      fi
+
+      if ! btrfs subvolume delete "$s"; then
+	diag+=("$v/$m/$s: error: unable to delete subvolume")
+	fail="true"
+	continue
+      fi
+
+      diag+=("$v/$m/$s: info: deleted subvolume")
+    done
+
+    cd "$v"
+
+    # Delete the machine directory (which we expect to be now empty).
+    #
+    if [ -n "$d" ]; then
+      if ! rmdir "$m"; then
+	diag+=("$v/$m: error: unable to delete machine directory")
+	fail="true"
+	continue
+      fi
+
+      diag+=("$v/$m: info: deleted machine directory")
+    fi
+  done
+  cd "$owd"
+done
+
+function print_diag ()
+{
+  local p
+  for p in "${diag[@]}"; do
+    echo "  $p"
+  done
+}
+
+if [ "${#diag[@]}" -gt 0 ]; then
+  if [ -z "$fail" ]; then
+    s="cleaned up entries in /build/machines/"
+  else
+    s="invalid entries in /build/machines/, halting"
+  fi
+
+  print_diag | email "$s"
+  info "$s" && print_diag 2>&1
+
+  if [ -n "$fail" ]; then
+    info "correct and restart the monitor (systemctl restart buildos)"
+    exit 1
+  fi
+fi
+
 # Monitoring loop.
 #
 while true; do