update

tests/TC008-mount-errors.md

@@ -0,0 +1,77 @@
+# Test Case: Mount Error Conditions
+
+**Test ID**: TC008-mount-errors
+**Category**: Mount Operations
+**Priority**: High
+
+## Objective
+
+Verify graceful rejection of three distinct provider conditions: a
+checksum-valid bad magic at the correct field, a 1023-byte provider ending
+before the superblock, and a zero-length provider.
+
+## Fixture Qualification
+
+```sh
+work=$(mktemp -d /tmp/repo22-tc008.XXXXXX)
+tests/prepare_error_fixtures.sh "$work/fixtures"
+grep -E '^(bad-magic|truncated-before-super|empty-provider) ' \
+  "$work/fixtures/fixture-evidence.txt"
+python3 tests/erofs_fixture.py inspect "$work/fixtures/bad-magic.erofs"
+wc -c "$work/fixtures/bad-magic.erofs" \
+  "$work/fixtures/truncated-before-super.erofs" \
+  "$work/fixtures/empty-provider.erofs"
+(cd "$work/fixtures" && sha256 -c SHA256SUMS)
+```
+
+The bad-magic variant changes only absolute bytes 1024 through 1027 and
+preserves media size. It intentionally retains the original checksum: the
+production verifier's fixed-magic suffix calculation remains valid, while the
+canonical checksum over the now-invalid magic does not. Magic is rejected
+before checksum verification. The short and empty fixtures are intentional
+provider-size cases, not same-size corruption variants.
+
+## FreeBSD Procedure
+
+Create a fresh mount directory. Execute each case separately and record whether
+failure occurs at `mdconfig` or `nmount`; never assume a fixed md number.
+
+```sh
+mkdir "$work/mnt"
+for image in bad-magic.erofs truncated-before-super.erofs empty-provider.erofs
+do
+  set +e
+  unit=$(mdconfig -a -t vnode -f "$work/fixtures/$image" \
+    2>"$work/$image.md.err")
+  md_status=$?
+  set -e
+  printf '%s md_status=%s unit=%s\n' "$image" "$md_status" "$unit"
+  if test "$md_status" -eq 0; then
+    set +e
+    mount -t erofs -o ro "/dev/$unit" "$work/mnt" \
+      >"$work/$image.out" 2>"$work/$image.mount.err"
+    mount_status=$?
+    set -e
+    printf '%s mount_status=%s\n' "$image" "$mount_status"
+    test "$mount_status" -ne 0
+    set +e
+    truss -o "$work/$image.mount.truss" mount -t erofs -o ro \
+      "/dev/$unit" "$work/mnt" >/dev/null 2>&1
+    set -e
+    grep -E 'nmount.*ERR#' "$work/$image.mount.truss"
+    mdconfig -d -u "${unit#md}"
+  else
+    cat "$work/$image.md.err"
+  fi
+  ! mount -p | awk -v path="$work/mnt" '$2 == path { found=1 }
+    END { exit found ? 0 : 1 }'
+done
+```
+
+## Expected Results
+
+All three cases fail without panic or hang. Bad magic reaches `nmount` and
+returns `EINVAL`. A 1023-byte vnode provider reaches `nmount` and returns
+`ENXIO`; a zero-byte provider is rejected by `mdconfig` with `EINVAL`. Record
+the observed stage and errno rather than fabricating a mount result. No md unit
+or mount may remain after each case.