tail: don't panic on '-c +N' with a huge N

src/uu/tail/src/tail.rs

@@ -480,7 +480,10 @@ fn bounded_tail(file: &mut File, settings: &Settings) -> UResult<()> {
         FilterMode::Bytes(Signum::Positive(count)) if count > &1 => {
             // GNU `tail` seems to index bytes and lines starting at 1, not
             // at 0. It seems to treat `+0` and `+1` as the same thing.
-            file.seek(SeekFrom::Start(*count - 1)).unwrap();
+            // Clamp to the file length so a start past EOF (or above `i64::MAX`) yields empty
+            // output instead of panicking on an `EINVAL` seek (#13887).
+            let len = file.metadata()?.len();
+            file.seek(SeekFrom::Start((*count - 1).min(len)))?;
         }
         _ => {}
     }

tests/by-util/test_tail.rs

@@ -857,6 +857,22 @@ fn test_bytes_single() {
         .stdout_is_fixture("foobar_bytes_single.expected");
 }
 
+#[test]
+fn test_positive_bytes_overflowing_offset_does_not_panic() {
+    // Regression test for #13887: `tail -c +N` on a regular file larger than the block size seeks
+    // to byte N-1. A very large N (greater than `i64::MAX`) makes the seek fail with `EINVAL`;
+    // `tail` used to `.unwrap()` and abort. That start is past the end of the file, so the output
+    // must simply be empty.
+    let (at, mut ucmd) = at_and_ucmd!();
+    // Larger than sane_blksize (~4 KiB) so the seek code path is taken.
+    at.write("big", &"x".repeat(8192));
+    ucmd.arg("-c")
+        .arg("+18446744073709551615") // u64::MAX
+        .arg("big")
+        .succeeds()
+        .no_output();
+}
+
 #[test]
 fn test_bytes_stdin() {
     new_ucmd!()