CASMTRIAGE-6885 - fix etc/resolv.conf resolution for broken symlinks.

CHANGELOG.md

@@ -7,6 +7,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.13.1] - 2024-04-12
+### Changed
+- CASMTRIAGE-6885 - fix etc/resolv.conf resolution when it is a broken symlink.
+
 ## [2.13.0] - 2024-03-01
 ### Added
 - CASMCMS-8821 - add support for remote build jobs.

scripts/buildenv-sidecar.sh

@@ -93,7 +93,7 @@ setup_user_shell() {
     chmod 600 /etc/cray/ims
 
     # Change signal location if user if jailed and not running on remote
-    if [ "$SSH_JAIL" = "True" -a "$REMOTE_BUILD_NODE" = ""]
+    if [ "$SSH_JAIL" = "True" -a -z "$REMOTE_BUILD_NODE"]
     then
         SIGNAL_FILE_COMPLETE=$IMAGE_ROOT_PARENT/image-root/tmp/complete
         SIGNAL_FILE_FAILED=$IMAGE_ROOT_PARENT/image-root/tmp/failed
@@ -158,10 +158,10 @@ setup_resolv() {
     local IMAGE_ROOT_DIR=$1
     local IMAGE_ROOT_RESOLV=$IMAGE_ROOT_DIR/etc/resolv.conf
     local TMP_RESOLV=$IMAGE_ROOT_DIR/tmp/resolv.conf
-    if [ -f "$IMAGE_ROOT_RESOLV" ]; then
+    if [ -f "$IMAGE_ROOT_RESOLV" -o -L "$IMAGE_ROOT_RESOLV" ]; then
         mv "$IMAGE_ROOT_RESOLV" "$TMP_RESOLV"
     fi
-    cp /etc/resolv.conf "$IMAGE_ROOT_RESOLV"
+    cp --remove-destination /etc/resolv.conf "$IMAGE_ROOT_RESOLV"
 }
 
 restore_resolv() {
@@ -170,7 +170,7 @@ restore_resolv() {
     local IMAGE_ROOT_RESOLV=$IMAGE_ROOT_DIR/etc/resolv.conf
     local TMP_RESOLV=$IMAGE_ROOT_DIR/tmp/resolv.conf
     rm "$IMAGE_ROOT_RESOLV"
-    if [ -f "$TMP_RESOLV" ]; then
+    if [ -f "$TMP_RESOLV" -o -L "$TMP_RESOLV" ]; then
         mv "$TMP_RESOLV" "$IMAGE_ROOT_RESOLV"
     fi
 }
@@ -200,13 +200,19 @@ case "$IMS_ACTION" in
         fi
         ;;
     customize)
-        # copy ca root key into the image before user shell
-        copy_ca_root_key "$IMAGE_ROOT_PARENT/image-root"
-        setup_resolv "$IMAGE_ROOT_PARENT/image-root"
+        # copy ca root key into the image before user shell if running locally
+        if [[ -z "${REMOTE_BUILD_NODE}" ]]; then
+            copy_ca_root_key "$IMAGE_ROOT_PARENT/image-root"
+            setup_resolv "$IMAGE_ROOT_PARENT/image-root"
+        fi
 
         echo "Running user shell for customize action"
         setup_user_shell
-        restore_resolv "$IMAGE_ROOT_PARENT/image-root"
+
+        # only restore the resolv file if running locally
+        if [[ -z "${REMOTE_BUILD_NODE}" ]]; then
+            restore_resolv "$IMAGE_ROOT_PARENT/image-root"
+        fi
 
         if [ -f "$SIGNAL_FILE_FAILED" ]; then
             echo "Customization of image root marked as failed. A new IMS image will not be created."

scripts/remote_customize_entrypoint.sh

@@ -41,18 +41,20 @@ setup_resolv() {
     # Copy the container's /etc/resolv.conf to the image root for customization
     local IMAGE_ROOT_RESOLV=$IMAGE_ROOT_DIR/etc/resolv.conf
     local TMP_RESOLV=$IMAGE_ROOT_DIR/tmp/resolv.conf
-    if [ -f "$IMAGE_ROOT_RESOLV" ]; then
+    if [ -f "$IMAGE_ROOT_RESOLV" -o -L "$IMAGE_ROOT_RESOLV" ]; then
         mv "$IMAGE_ROOT_RESOLV" "$TMP_RESOLV"
+    else
+        echo "Did not find an existing /etc/resolv.conf file"
     fi
-    cp /etc/resolv.conf "$IMAGE_ROOT_RESOLV"
+    cp --remove-destination /etc/resolv.conf "$IMAGE_ROOT_RESOLV"
 }
 
 restore_resolv() {
     # Restore the image root's old /etc/resolv.conf after customization
     local IMAGE_ROOT_RESOLV=$IMAGE_ROOT_DIR/etc/resolv.conf
     local TMP_RESOLV=$IMAGE_ROOT_DIR/tmp/resolv.conf
     rm "$IMAGE_ROOT_RESOLV"
-    if [ -f "$TMP_RESOLV" ]; then
+    if [ -f "$TMP_RESOLV" -o -L "$TMP_RESOLV" ]; then
         mv "$TMP_RESOLV" "$IMAGE_ROOT_RESOLV"
     fi
 }
@@ -109,6 +111,10 @@ fi
 mkdir -p $IMAGE_ROOT_PARENT
 unsquashfs -f -d $IMAGE_ROOT_DIR /data/image.sqsh
 
+# clear any signal flags that were left in the image
+rm $USER_SIGNAL_FILE_FAILED
+rm $USER_SIGNAL_FILE_COMPLETE
+
 # Copy cray certs into image
 copy_ca_root_key
 
@@ -196,10 +202,11 @@ fi
 restore_resolv
 
 # if successful, package up the results into a squashfs file to transfer back to worker node
-if [ ! -f "$USER_SIGNAL_FILE_FAILED" ]; then
-  # Remove the successful signal file and put failed flag in place in case squash fails
+if [ -f "$USER_SIGNAL_FILE_COMPLETE" ]; then
+  # Remove the signal flags before squash and put failed flag in place in case squash fails
   # NOTE: the sshd entrypoint script will be waiting for $SIGNAL_FILE_REMOTE_EXITING
-  rm $SIGNAL_FILE_COMPLETE
+  rm $USER_SIGNAL_FILE_FAILED
+  rm $USER_SIGNAL_FILE_COMPLETE
   touch $SIGNAL_FILE_FAILED
 
   # Make the squashfs formatted archive to transfer results back to job
@@ -215,6 +222,10 @@ if [ ! -f "$USER_SIGNAL_FILE_FAILED" ]; then
     rm $SIGNAL_FILE_FAILED
     touch $SIGNAL_FILE_COMPLETE
   fi
+else
+  # failed so set the correct signal files
+  rm $USER_SIGNAL_FILE_FAILED
+  touch $SIGNAL_FILE_FAILED
 fi
 
 # Signal that this is finished