test(xcat-core): cover Genesis boot files written with the image arch

When the forward nodeset osimage writes the image attributes and then
fails, the node keeps the arch of the image. ipxe_secureboot_shim.sh
replays the saved shell, standby or runcmd state before it writes the
saved arch back, so nodeset writes the Genesis kernel and initrd of the
image arch. The chdef that follows corrects the database, not those
boot files.

The nodeset stub now records the Genesis kernel and initrd that a
shell, standby or runcmd state selects from the current node arch, and
can fail nodeset osimage after it updates the attributes. Two cases
start an x86_64 node at shell and at runcmd=bmcsetup with a ppc64le
image, and assert the x86_64 kernel and initrd. Both fail against the
current helper.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>

xCAT-test/bats/ipxe_secureboot_shim_restore.bats

@@ -56,6 +56,8 @@ setup()
     export STUB_LSDEF_FAIL_ATTR=
     export STUB_NODESET_RESTORE_RC=0
     export STUB_CHDEF_RC=0
+    # Fail the forward nodeset osimage after it wrote the image attributes into the node.
+    export STUB_NODESET_FORWARD_FAIL=
     # Which chdef netboot= call to fail, counted from 1. The helper sets the method and later
     # puts it back, so one return code for both cannot say which call failed.
     export STUB_CHDEF_NETBOOT_FAIL_NTH=0
@@ -73,6 +75,7 @@ setup()
     export STUB_OS="$OS"
     export STUB_STORE="$BATS_TEST_TMPDIR/store"
     export STUB_LIB="$BIN/../stublib.sh"
+    export STUB_BOOTCFG="$BATS_TEST_TMPDIR/bootcfg"
 
     cat >"$STUB_LIB" <<'STUB'
 # The node starts with the STUB_* values. A test that defines its own osimage rows before the
@@ -134,6 +137,7 @@ STUB
     # nodeset osimage[=<image>] does what setdestiny does to the node: it writes provmethod, os,
     # arch and profile from the osimage row, sets currchain to boot, and writes the UEFI script.
     # Any other destiny sets currstate, and the actions after its first comma become currchain.
+    # shell, standby and runcmd also write a Genesis boot configuration for the current arch.
     cat >"$BIN/nodeset" <<'STUB'
 #!/bin/bash
 echo "nodeset $*" >>"$STUB_CALLS"
@@ -156,6 +160,7 @@ case "$2" in
         put profile "$(cat "$row/profile")"
         put currchain boot
         put currstate "install $(get os)-$(get arch)-$(get profile)"
+        [ -z "$STUB_NODESET_FORWARD_FAIL" ] || { echo "Error: cannot write the boot files" >&2; exit 1; }
         {
             echo '#!gpxe'
             echo 'imgfetch -n kernel http://${next-server}/tftpboot/xcat/osimage/vmlinuz'
@@ -170,6 +175,13 @@ case "$2" in
         [ "$STUB_NODESET_RESTORE_RC" = 0 ] || exit "$STUB_NODESET_RESTORE_RC"
         put currstate "${2%%,*}"
         case "$2" in *,*) put currchain "${2#*,}" ;; esac
+        # shell, standby and runcmd boot Genesis, and nodeset selects its kernel and initrd from
+        # the arch of the node at the time of the call.
+        case "${2%%,*}" in
+            shell | standby | runcmd=*)
+                printf 'kernel genesis.kernel.%s\ninitrd genesis.fs.%s\n' "$(get arch)" "$(get arch)" >"$STUB_BOOTCFG"
+                ;;
+        esac
         exit 0
         ;;
 esac
@@ -489,3 +501,39 @@ STUB
     [[ "$output" != *IPXE_SHIM_OK* ]]
     [[ "$output" == *IPXE_SHIM_FAIL* ]]
 }
+
+# nodeset osimage writes the image attributes before it writes the boot files, so a forward call
+# that fails can leave the node with the arch of the image. A replay of shell, standby or runcmd
+# then writes the Genesis kernel and initrd of that arch, and a later chdef of the database does
+# not rewrite those files.
+@test "a Genesis state goes back with the kernel and initrd of the original arch" {
+    define_image "$OS-ppc64le-install-compute" "$OS" ppc64le compute
+    STUB_BEFORE=shell
+    STUB_PROVMETHOD="$OS-ppc64le-install-compute"
+    STUB_NODESET_FORWARD_FAIL=1
+    run "$SCRIPT" "$NODE"
+    [ "$status" -ne 0 ]
+    [[ "$output" != *IPXE_SHIM_OK* ]]
+    grep -q "^nodeset $NODE shell$" "$CALLS"
+    grep -qx 'kernel genesis.kernel.x86_64' "$STUB_BOOTCFG"
+    grep -qx 'initrd genesis.fs.x86_64' "$STUB_BOOTCFG"
+    [ "$(node_attr arch)" = x86_64 ]
+    [ "$(node_attr provmethod)" = "$OS-ppc64le-install-compute" ]
+    [ "$(node_attr currstate)" = shell ]
+}
+
+# The same for a runcmd state, which also restores the pending chain after the replay.
+@test "a runcmd state goes back with the kernel and initrd of the original arch" {
+    define_image "$OS-ppc64le-install-compute" "$OS" ppc64le compute
+    STUB_BEFORE='runcmd=bmcsetup'
+    STUB_CURRCHAIN=standby
+    STUB_PROVMETHOD="$OS-ppc64le-install-compute"
+    STUB_NODESET_FORWARD_FAIL=1
+    run "$SCRIPT" "$NODE"
+    [ "$status" -ne 0 ]
+    [[ "$output" != *IPXE_SHIM_OK* ]]
+    grep -qx 'kernel genesis.kernel.x86_64' "$STUB_BOOTCFG"
+    grep -qx 'initrd genesis.fs.x86_64' "$STUB_BOOTCFG"
+    [ "$(node_attr arch)" = x86_64 ]
+    [ "$(node_attr currchain)" = standby ]
+}