From 33958638b2ea26f7f0284edab66323e509b89473 Mon Sep 17 00:00:00 2001
From: "Dylan M. Taylor" <dylan@dylanmtaylor.com>
Date: Sat, 14 Nov 2026 02:06:54 +0000
Subject: [PATCH] Validate XFL archive end record candidates

---
 crates/xfl/src/lib.rs   | 47 +++++++++++++++++++++++++++++++++++------
 crates/xfl/src/tests.rs | 11 ++++++++++
 2 files changed, 51 insertions(+), 7 deletions(-)

diff --git a/crates/xfl/src/lib.rs b/crates/xfl/src/lib.rs
index ac6dcbc4c..4004d57ad 100644
--- a/crates/xfl/src/lib.rs
+++ b/crates/xfl/src/lib.rs
@@ -234,13 +234,46 @@ fn method_name(method: u16) -> &'static str {
 fn find_end(data: &[u8]) -> Result<(u64, u64, u64, usize), ZipError> {
     // The end record is in the last 22 + 65535 bytes (it may carry a comment).
     let lo = data.len().saturating_sub(22 + 65_535);
-    let eocd = (lo..data.len().saturating_sub(21))
-        .rev()
-        .find(|&i| {
-            u32_at(data, i) == Some(END)
-                && u16_at(data, i + 20).is_some_and(|comment| i.checked_add(22 + usize::from(comment)).is_some_and(|end| end <= data.len()))
-        })
-        .ok_or(ZipError::NotZip)?;
+    let mut eocd = None;
+    let mut fallback = None;
+    for candidate in (lo..data.len().saturating_sub(21)).rev() {
+        if u32_at(data, candidate) != Some(END)
+            || !u16_at(data, candidate + 20)
+                .is_some_and(|comment| candidate.checked_add(22 + usize::from(comment)).is_some_and(|end| end <= data.len()))
+        {
+            continue;
+        }
+        // Keep the earliest bounds-valid candidate's eventual parse error if no structurally
+        // consistent candidate is found. This preserves useful damaged-ZIP diagnostics.
+        fallback = Some(candidate);
+
+        let size = u32_at(data, candidate + 12).unwrap_or(u32::MAX);
+        let offset = u32_at(data, candidate + 16).unwrap_or(u32::MAX);
+        if size == u32::MAX || offset == u32::MAX {
+            // ZIP64's directory must end exactly where the ZIP64 end record starts.
+            let locator = candidate.checked_sub(20).filter(|&at| u32_at(data, at) == Some(LOCATOR64));
+            let valid = locator
+                .and_then(|at| u64_at(data, at + 8))
+                .and_then(|at| usize::try_from(at).ok())
+                .filter(|&at| u32_at(data, at) == Some(END64))
+                .is_some_and(|at| {
+                    let dir_size = u64_at(data, at + 40).and_then(|n| usize::try_from(n).ok());
+                    let dir_offset = u64_at(data, at + 48).and_then(|n| usize::try_from(n).ok());
+                    dir_size.zip(dir_offset).is_some_and(|(size, offset)| offset.checked_add(size) == Some(at))
+                });
+            if !valid {
+                continue;
+            }
+        } else {
+            let end = usize::try_from(offset).ok().and_then(|offset| usize::try_from(size).ok().and_then(|size| offset.checked_add(size)));
+            if end != Some(candidate) {
+                continue;
+            }
+        }
+        eocd = Some(candidate);
+        break;
+    }
+    let eocd = eocd.or(fallback).ok_or(ZipError::NotZip)?;
     let field16 = |o: usize| u16_at(data, eocd + o).ok_or(ZipError::Damaged("the end record is cut short"));
     let field32 = |o: usize| u32_at(data, eocd + o).ok_or(ZipError::Damaged("the end record is cut short"));
     let (disk, dir_disk, on_disk, count) = (field16(4)?, field16(6)?, field16(8)?, field16(10)?);
diff --git a/crates/xfl/src/tests.rs b/crates/xfl/src/tests.rs
index 7f981d772..9f9d5a13e 100644
--- a/crates/xfl/src/tests.rs
+++ b/crates/xfl/src/tests.rs
@@ -240,6 +240,17 @@ fn invalid_eocd_comment_candidate_does_not_mask_archive_end() {
     assert_eq!(fake + 22, z.len());
 }
 
+#[test]
+fn eocd_signature_inside_comment_is_not_an_empty_archive() {
+    let mut z = zip(&[("a.xml", b"x", false)]);
+    let end = z.len() - 22;
+    z[end + 20..end + 22].copy_from_slice(&22u16.to_le_bytes());
+    z.extend_from_slice(&END.to_le_bytes());
+    z.extend_from_slice(&[0; 18]);
+    let archive = Archive::open(&z).unwrap();
+    assert_eq!(archive.names().collect::<Vec<_>>(), ["a.xml"]);
+}
+
 #[test]
 fn central_directory_must_be_consumed_exactly() {
     let mut z = zip(&[("a.xml", b"x", false)]);