diff --git a/src/uu/csplit/src/csplit.rs b/src/uu/csplit/src/csplit.rs
index b5004c362..cd55ef7df 100644
--- a/src/uu/csplit/src/csplit.rs
+++ b/src/uu/csplit/src/csplit.rs
@@ -99,15 +99,17 @@ where
     let patterns: Vec<patterns::Pattern> = patterns::get_patterns(&patterns[..])?;
     let ret = do_csplit(&mut split_writer, patterns, &mut input_iter);
 
-    // consume the rest
-    input_iter.rewind_buffer();
-    if let Some((_, line)) = input_iter.next() {
-        split_writer.new_writer()?;
-        split_writer.writeln(&line?)?;
-        for (_, line) in input_iter {
+    // consume the rest, unless there was an error
+    if ret.is_ok() {
+        input_iter.rewind_buffer();
+        if let Some((_, line)) = input_iter.next() {
+            split_writer.new_writer()?;
             split_writer.writeln(&line?)?;
+            for (_, line) in input_iter {
+                split_writer.writeln(&line?)?;
+            }
+            split_writer.finish_split();
         }
-        split_writer.finish_split();
     }
     // delete files on error by default
     if ret.is_err() && !options.keep_files {
diff --git a/tests/by-util/test_csplit.rs b/tests/by-util/test_csplit.rs
index 5bd384de3..03b8c92fc 100644
--- a/tests/by-util/test_csplit.rs
+++ b/tests/by-util/test_csplit.rs
@@ -575,7 +575,7 @@ fn test_skip_to_match_context_underflow() {
     let (at, mut ucmd) = at_and_ucmd!();
     ucmd.args(&["numbers50.txt", "%5%-10"])
         .fails()
-        .stdout_is("141\n")
+        .stdout_is("")
         .stderr_is("csplit: '%5%-10': line number out of range\n");
 
     let count = glob(&at.plus_as_string("xx*"))
@@ -586,14 +586,13 @@ fn test_skip_to_match_context_underflow() {
     let (at, mut ucmd) = at_and_ucmd!();
     ucmd.args(&["numbers50.txt", "%5%-10", "-k"])
         .fails()
-        .stdout_is("141\n")
+        .stdout_is("")
         .stderr_is("csplit: '%5%-10': line number out of range\n");
 
     let count = glob(&at.plus_as_string("xx*"))
         .expect("counting splits")
         .count();
-    assert_eq!(count, 1);
-    assert_eq!(at.read("xx00"), generate(1, 51));
+    assert_eq!(count, 0);
 }
 
 #[test]
@@ -1225,13 +1224,12 @@ fn test_corner_case4() {
     assert_eq!(at.read("xx02"), generate(26, 51));
 }
 
-// NOTE: differs from gnu's output: the empty split is not written
 #[test]
 fn test_up_to_match_context_underflow() {
     let (at, mut ucmd) = at_and_ucmd!();
     ucmd.args(&["numbers50.txt", "/5/-10"])
         .fails()
-        .stdout_is("0\n141\n")
+        .stdout_is("0\n")
         .stderr_is("csplit: '/5/-10': line number out of range\n");
 
     let count = glob(&at.plus_as_string("xx*"))
@@ -1242,26 +1240,24 @@ fn test_up_to_match_context_underflow() {
     let (at, mut ucmd) = at_and_ucmd!();
     ucmd.args(&["numbers50.txt", "/5/-10", "-k"])
         .fails()
-        .stdout_is("0\n141\n")
+        .stdout_is("0\n")
         .stderr_is("csplit: '/5/-10': line number out of range\n");
 
     let count = glob(&at.plus_as_string("xx*"))
         .expect("counting splits")
         .count();
-    assert_eq!(count, 2);
+    assert_eq!(count, 1);
     assert_eq!(at.read("xx00"), "");
-    assert_eq!(at.read("xx01"), generate(1, 51));
 }
 
 // the offset is out of range because of the first pattern
-// NOTE: output different than gnu's: the empty split is written but the rest of the input file is not
 #[test]
 fn test_line_num_range_with_up_to_match1() {
     let (at, mut ucmd) = at_and_ucmd!();
     ucmd.args(&["numbers50.txt", "10", "/12/-5"])
         .fails()
         .stderr_is("csplit: '/12/-5': line number out of range\n")
-        .stdout_is("18\n0\n123\n");
+        .stdout_is("18\n0\n");
 
     let count = glob(&at.plus_as_string("xx*"))
         .expect("there should be splits created")
@@ -1272,26 +1268,24 @@ fn test_line_num_range_with_up_to_match1() {
     ucmd.args(&["numbers50.txt", "10", "/12/-5", "-k"])
         .fails()
         .stderr_is("csplit: '/12/-5': line number out of range\n")
-        .stdout_is("18\n0\n123\n");
+        .stdout_is("18\n0\n");
 
     let count = glob(&at.plus_as_string("xx*"))
         .expect("there should be splits created")
         .count();
-    assert_eq!(count, 3);
+    assert_eq!(count, 2);
     assert_eq!(at.read("xx00"), generate(1, 10));
     assert_eq!(at.read("xx01"), "");
-    assert_eq!(at.read("xx02"), generate(10, 51));
 }
 
 // the offset is out of range because more lines are needed than physically available
-// NOTE: output different than gnu's: the empty split is not written but the rest of the input file is
 #[test]
 fn test_line_num_range_with_up_to_match2() {
     let (at, mut ucmd) = at_and_ucmd!();
     ucmd.args(&["numbers50.txt", "10", "/12/-15"])
         .fails()
         .stderr_is("csplit: '/12/-15': line number out of range\n")
-        .stdout_is("18\n0\n123\n");
+        .stdout_is("18\n0\n");
 
     let count = glob(&at.plus_as_string("xx*"))
         .expect("there should be splits created")
@@ -1302,15 +1296,14 @@ fn test_line_num_range_with_up_to_match2() {
     ucmd.args(&["numbers50.txt", "10", "/12/-15", "-k"])
         .fails()
         .stderr_is("csplit: '/12/-15': line number out of range\n")
-        .stdout_is("18\n0\n123\n");
+        .stdout_is("18\n0\n");
 
     let count = glob(&at.plus_as_string("xx*"))
         .expect("there should be splits created")
         .count();
-    assert_eq!(count, 3);
+    assert_eq!(count, 2);
     assert_eq!(at.read("xx00"), generate(1, 10));
     assert_eq!(at.read("xx01"), "");
-    assert_eq!(at.read("xx02"), generate(10, 51));
 }
 
 // NOTE: output different than gnu's: the pattern /10/ is matched but should not