Integration: #783 reads the cover image with a size cap (256 MiB), devices included

crates/engine/src/cmd/datamerge/mod.rs

@@ -768,17 +768,23 @@ fn read_path(path: &str) -> std::result::Result<Vec<u8>, String> {
     read_path_capped(path, MAX_DATA_FILE_BYTES)
 }
 
-fn read_path_capped(path: &str, cap: u64) -> std::result::Result<Vec<u8>, String> {
+/// The file at `path`, refused when it is larger than `cap` bytes. The read itself stops past
-    match std::fs::metadata(path) {
+/// `cap`, so a device or a file that grows while it is read can't exhaust memory.
-        Ok(m) if m.len() > cap => return Err(format!("{path}: the data file is larger than {} MiB", cap / (1024 * 1024))),
+pub(crate) fn read_path_capped(path: &str, cap: u64) -> std::result::Result<Vec<u8>, String> {
-        Err(e) if e.kind() == std::io::ErrorKind::NotFound => return Err(format!("{path}: the data file is missing")),
+    use std::io::Read;
-        _ => {}
+    let too_big = || format!("{path}: the file is larger than {} MiB", cap / (1024 * 1024));
+    let missing = |e: std::io::Error| {
+        if e.kind() == std::io::ErrorKind::NotFound { format!("{path}: the file is missing") } else { format!("{path}: {e}") }
+    };
+    if std::fs::metadata(path).map_err(missing)?.len() > cap {
+        return Err(too_big());
     }
-    match std::fs::read(path) {
+    let mut b = Vec::new();
-        Ok(b) => Ok(b),
+    std::fs::File::open(path).map_err(missing)?.take(cap.saturating_add(1)).read_to_end(&mut b).map_err(missing)?;
-        Err(e) if e.kind() == std::io::ErrorKind::NotFound => Err(format!("{path}: the data file is missing")),
+    if b.len() as u64 > cap {
-        Err(e) => Err(format!("{path}: {e}")),
+        return Err(too_big());
     }
+    Ok(b)
 }
 
 fn parent_dir(path: &str) -> Option<PathBuf> {

crates/engine/src/cmd/datamerge/tests.rs

@@ -1086,6 +1086,15 @@ fn oversized_data_file_is_refused_before_reading() {
     assert_eq!(read_path_capped(&path, 4096).unwrap().len(), 2048);
 }
 
+/// A device reports no length; the read still stops at the cap (an EPUB `coverImage` of
+/// `/dev/zero` must not read until memory runs out).
+#[cfg(unix)]
+#[test]
+fn endless_file_is_refused_at_the_cap() {
+    let msg = read_path_capped("/dev/zero", 1024).unwrap_err();
+    assert!(msg.contains("larger than"), "{msg}");
+}
+
 /// #198: with facing pages on, one record per page must not stack the copies at identical
 /// coordinates on one spread (leaving a later page empty).
 #[test]

crates/engine/src/cmd/export.rs

@@ -180,7 +180,7 @@ fn export_epub(s: &mut Session, p: &Value) -> Result<Value> {
     let mut warnings: Vec<String> = Vec::new();
     // A cover image file (PNG or JPEG), else the first page rendered when `cover` asks for it.
     let cover = match p.get("coverImage").and_then(Value::as_str).filter(|_| !fixed) {
-        Some(path) => Some(std::fs::read(path).map_err(|e| crate::EngineError::Other(format!("{path}: {e}")))?),
+        Some(path) => Some(super::datamerge::read_path_capped(path, super::datamerge::MAX_IMAGE_FILE_BYTES).map_err(crate::EngineError::Other)?),
         None => (want_cover && !fixed).then(|| rr.render_page(&st.doc, &s.cache, 0, 2.0, false, &ropts).map(|img| img.to_png())).flatten(),
     };
     if let Some(c) = &cover