Fixed hub upgrade PG migration failing with Permission denied on old binaries

Ticket: ENT-12809
Changelog: Title

Author: nickanderson

packaging/common/cfengine-hub/postinstall.sh

@@ -631,7 +631,9 @@ migrate_db_using_dump_file() {
     safe_mv "$PREFIX/share" postgresql "$PREFIX/share.new"
     # restore old backup
     safe_cp "$BACKUP_DIR" data "$PREFIX/state/pg"
-    safe_cp "$BACKUP_DIR" bin "$PREFIX"
+    safe_cp_nolink "$BACKUP_DIR" bin "$PREFIX"
+    # Ensure restored old binaries are executable by cfpostgres.
+    chmod 755 "$PREFIX/bin"/* 2>/dev/null || true
     on_files cp "$BACKUP_DIR/lib" "$PREFIX/lib"
     safe_cp "$BACKUP_DIR/lib" postgresql "$PREFIX/lib"
     safe_cp "$BACKUP_DIR/share" postgresql "$PREFIX/share"
@@ -745,6 +747,22 @@ do_migration() {
     else
       DEBUG=0
     fi
+
+    # Preflight: verify cfpostgres can execute the old binaries before any
+    # destructive operation. If not, attempt a one-shot chmod repair.
+    if ! su cfpostgres -c "LD_LIBRARY_PATH=$BACKUP_DIR/lib/ $BACKUP_DIR/bin/pg_ctl --version" >/dev/null 2>&1; then
+      cf_console echo "cfpostgres cannot execute $BACKUP_DIR/bin/pg_ctl — attempting mode repair"
+      chmod 755 "$BACKUP_DIR/bin"/* 2>/dev/null || true
+      chmod 711 "$BACKUP_DIR" "$BACKUP_DIR/bin" 2>/dev/null || true
+      if ! su cfpostgres -c "LD_LIBRARY_PATH=$BACKUP_DIR/lib/ $BACKUP_DIR/bin/pg_ctl --version" >/dev/null 2>&1; then
+        cf_console echo "ERROR: cfpostgres still cannot execute $BACKUP_DIR/bin/pg_ctl after chmod repair."
+        ls -la "$BACKUP_DIR/bin/" >&2
+        cf_console echo "Please fix permissions on $BACKUP_DIR/bin/ and retry the upgrade."
+        exit 1
+      fi
+      cf_console echo "Mode repair succeeded, continuing migration."
+    fi
+
     cd /tmp
     cf_console echo "Migrating database using pg_upgrade utility..."
     cf_console echo

packaging/common/cfengine-hub/preinstall.sh

@@ -252,7 +252,13 @@ if migrating_postgres; then
     # the postgresql.conf file was modified, we should try to use it after migration
     cp -a "$BACKUP_DIR/data/postgresql.conf" "$BACKUP_DIR/data/postgresql.conf.modified"
   fi
-  safe_cp "$PREFIX" bin "$BACKUP_DIR"
+  # Copy, don't hardlink: policy managing perms under $PREFIX/state/pg (e.g.
+  # CFE-951 in masterfiles) would otherwise change the live binaries too.
+  safe_cp_nolink "$PREFIX" bin "$BACKUP_DIR"
+  # Normalize executable permissions so cfpostgres can run old PG binaries
+  # during migration, whatever mode damage the source already had.
+  chmod 755 "$BACKUP_DIR/bin"/* 2>/dev/null || true
+  chmod 711 "$BACKUP_DIR/bin" 2>/dev/null || true
   on_files cp "$PREFIX/lib" "$BACKUP_DIR/lib"
   safe_cp "$PREFIX/lib" postgresql "$BACKUP_DIR/lib"
   safe_cp "$PREFIX/share" postgresql "$BACKUP_DIR/share"

packaging/common/script-templates/script-common.sh

@@ -136,6 +136,21 @@ safe_cp() {
     return 1
 }
 
+safe_cp_nolink() {
+    # Like `safe_cp`, but never creates hardlinks, so the copy does not share
+    # mode and owner with the source. Args are same as in safe_cp.
+    test "$#" -eq 3 || return 2
+    from="$1"
+    name="$2"
+    to="$3"
+    if cp -a "$from/$name" "$to"; then
+        return 0
+    fi
+    echo "Copy failed, so removing partially-copied data and aborting"
+    rm -rf "$to/$name"
+    return 1
+}
+
 safe_mv() {
     # "safe" alternative to `mv`. Executes `safe_cp` and deletes source dir if
     # that succeeds.