tpm2: re-manufacture software TPM when state dir is incomplete

setup_swtpm() decided whether a software TPM had already been
manufactured by checking whether the state directory was empty. But
manufacture_swtpm() writes swtpm's config files before forking
swtpm_setup, so an interrupted manufacture leaves the directory
non-empty yet without a usable TPM. The next boot then mistook it for a
complete TPM and started swtpm against a broken state directory.

Keying off a swtpm state file like tpm2-00.permall is no better, as
swtpm_setup gives no guarantee any single one is written atomically or
last. Instead, have manufacture_swtpm() write a marker (.manufactured)
as its very last step, once swtpm_setup has exited successfully, and
gate on it: re-manufacture when it is missing in the initrd, and refuse
rather than start a broken TPM outside it.

Signed-off-by: Paul Meyer <katexochen0@gmail.com>

test/units/TEST-92-TPM2-SWTPM.sh

@@ -3,12 +3,16 @@
 set -eux
 set -o pipefail
 
-# Exercises the software TPM fallback (systemd-tpm2-swtpm.service) across a reboot. The VM boots in EFI mode
+# Exercises the software TPM fallback (systemd-tpm2-swtpm.service) across reboots. The VM boots in EFI mode
 # without a hardware/firmware TPM and with "systemd.tpm2_software_fallback=yes" (see the test's meson.build),
 # so systemd-tpm2-generator manufactures a software TPM on the ESP in the initrd and chainloads swtpm.
 #
 #   boot 0: the TPM is manufactured in the initrd; seal a secret to it and stash the blob.
-#   boot 1: the TPM state persisted on the ESP across the reboot, so the secret still unseals.
+#   boot 1: the TPM state persisted on the ESP across the reboot, so the secret still unseals. Then mimic a
+#           manufacture that was interrupted before it completed (drop everything but the config files, so the
+#           ".manufactured" marker is gone) and reboot.
+#   boot 2: setup_swtpm() must notice the missing marker and re-manufacture, rather than mistaking the
+#           leftover config files for a complete TPM and starting swtpm against a stateless directory.
 #
 # See systemd-tpm2-swtpm.service(8).
 
@@ -17,6 +21,8 @@ set -o pipefail
 
 CRED=/var/lib/systemd-tpm2-swtpm-test.cred
 PLAINTEXT="swtpm round-trip"
+# Marker (SWTPM_MANUFACTURED_MARKER) that manufacture_swtpm() indicates completion with.
+MARKER=.manufactured
 
 if [[ -n "${ASAN_OPTIONS:-}" ]]; then
     # swtpm_setup is not built with sanitizers, but does NSS lookups that pull in the ASan-instrumented
@@ -38,6 +44,15 @@ assert_swtpm_up() {
     assert_in '\+driver' "$(systemd-analyze has-tpm2 || :)"
 }
 
+# Locate swtpm's state directory on the ESP.
+swtpm_state_dir() {
+    local d
+    for d in /boot/loader/swtpm /efi/loader/swtpm; do
+        [[ -d "$d" ]] && { echo "$d"; return 0; }
+    done
+    return 1
+}
+
 case "$REBOOT_COUNT" in
     0)
         assert_swtpm_up
@@ -53,6 +68,31 @@ case "$REBOOT_COUNT" in
         # Persistence: the TPM state survived the reboot on the ESP, so the blob still unseals.
         echo -n "$PLAINTEXT" >/tmp/swtpm-plaintext
         systemd-creds decrypt --name= "$CRED" - | cmp /tmp/swtpm-plaintext -
+
+        # Mimic a manufacture interrupted after swtpm began writing TPM state but before the marker: stop
+        # swtpm, drop everything except the three config files, then leave a partial/corrupt state file
+        # behind. swtpm_setup --not-overwrite would refuse to recreate that, so recovery must clear it first.
+        statedir="$(swtpm_state_dir)"
+        systemctl stop systemd-tpm2-swtpm.service
+        find "$statedir" -mindepth 1 -maxdepth 1 -type f \
+             ! -name swtpm-localca.conf ! -name swtpm-localca.options ! -name swtpm_setup.conf -delete
+        echo "corrupt" >"$statedir/tpm2-00.permall"
+        test -e "$statedir/swtpm_setup.conf"
+        test ! -e "$statedir/$MARKER"
+        systemctl_final reboot
+        exec sleep infinity
+        ;;
+    2)
+        # setup_swtpm() must have re-manufactured instead of trusting the leftover config files: the marker is
+        # back and swtpm_setup re-ran swtpm_localca, recreating issuer-certificate.pem. The TPM must also work.
+        # Regression test for keying re-manufacture off an incomplete state directory.
+        assert_swtpm_up
+        statedir="$(swtpm_state_dir)"
+        test -e "$statedir/$MARKER"
+        test -e "$statedir/issuer-certificate.pem"
+        echo -n "$PLAINTEXT" >/tmp/swtpm-plaintext
+        systemd-creds encrypt --name= --with-key=tpm2 /tmp/swtpm-plaintext /tmp/swtpm-new.cred
+        systemd-creds decrypt --name= /tmp/swtpm-new.cred - | cmp /tmp/swtpm-plaintext -
         touch /testok
         ;;
     *)