diff --git a/polars/polars-io/src/csv.rs b/polars/polars-io/src/csv.rs
index cdeeff6fde20..8ffbd5fb7a2c 100644
--- a/polars/polars-io/src/csv.rs
+++ b/polars/polars-io/src/csv.rs
@@ -41,7 +41,7 @@
 //! }
 //! ```
 //!
-use crate::csv_core::csv::{build_csv_reader, SequentialReader};
+use crate::csv_core::csv::{build_csv_reader, CoreReader};
 use crate::mmap::MmapBytesReader;
 use crate::utils::{resolve_homedir, to_arrow_compatible_df};
 use crate::{SerReader, SerWriter};
@@ -345,7 +345,7 @@ where
         self
     }
 
-    pub fn build_inner_reader(self) -> Result<SequentialReader<R>> {
+    pub fn build_inner_reader(self) -> Result<CoreReader<R>> {
         build_csv_reader(
             self.reader,
             self.stop_after_n_rows,
diff --git a/polars/polars-io/src/csv_core/csv.rs b/polars/polars-io/src/csv_core/csv.rs
index feadec6e4a79..a856151037c2 100644
--- a/polars/polars-io/src/csv_core/csv.rs
+++ b/polars/polars-io/src/csv_core/csv.rs
@@ -4,7 +4,6 @@ use crate::csv_core::{buffer::*, parser::*};
 use crate::mmap::MmapBytesReader;
 use crate::PhysicalIoExpr;
 use crate::ScanAggregation;
-use csv::ByteRecordsIntoIter;
 use polars_arrow::array::*;
 use polars_core::utils::accumulate_dataframes_vertical;
 use polars_core::{prelude::*, POOL};
@@ -18,13 +17,13 @@ use std::sync::atomic::Ordering;
 use std::sync::{atomic::AtomicUsize, Arc};
 
 /// CSV file reader
-pub struct SequentialReader<R: Read + MmapBytesReader> {
+pub struct CoreReader<R: Read + MmapBytesReader> {
     /// Explicit schema for the CSV file
     schema: SchemaRef,
     /// Optional projection for which columns to load (zero-based column indices)
     projection: Option<Vec<usize>>,
     /// File reader
-    record_iter: Option<ByteRecordsIntoIter<R>>,
+    reader: Option<R>,
     /// Current line number, used in error reporting
     line_number: usize,
     ignore_parser_errors: bool,
@@ -42,7 +41,7 @@ pub struct SequentialReader<R: Read + MmapBytesReader> {
     null_values: Option<Vec<String>>,
 }
 
-impl<R> fmt::Debug for SequentialReader<R>
+impl<R> fmt::Debug for CoreReader<R>
 where
     R: Read + MmapBytesReader,
 {
@@ -103,7 +102,7 @@ impl RunningSize {
     }
 }
 
-impl<R: Read + Sync + Send + MmapBytesReader> SequentialReader<R> {
+impl<R: Read + Sync + Send + MmapBytesReader> CoreReader<R> {
     /// Returns the schema of the reader, useful for getting the schema without reading
     /// record batches
     pub fn schema(&self) -> SchemaRef {
@@ -442,34 +441,30 @@ impl<R: Read + Sync + Send + MmapBytesReader> SequentialReader<R> {
     ) -> Result<DataFrame> {
         let n_threads = self.n_threads.unwrap_or_else(|| POOL.current_num_threads());
 
-        let mut df = match (&self.path, self.record_iter.is_some()) {
+        let mut df = match &self.path {
             // we have a path so we can mmap
-            (Some(p), _) => {
+            Some(p) => {
                 let file = std::fs::File::open(p)?;
                 let mmap = unsafe { memmap::Mmap::map(&file)? };
                 let bytes = mmap[..].as_ref();
                 self.parse_csv(n_threads, bytes, predicate.as_ref())?
             }
-            (None, true) => {
-                // get a hold of the reader + mmapreader
-                let mut r = std::mem::take(&mut self.record_iter)
-                    .unwrap()
-                    .into_reader()
-                    .into_inner();
+            None => {
+                let mut reader = self.reader.take().unwrap();
 
                 // we have a file so we can mmap
-                if let Some(file) = r.to_file() {
+                if let Some(file) = reader.to_file() {
                     let mmap = unsafe { memmap::Mmap::map(file)? };
                     let bytes = mmap[..].as_ref();
                     self.parse_csv(n_threads, bytes, predicate.as_ref())?
                 } else {
                     // we can get the bytes for free
-                    let bytes = if let Some(bytes) = r.to_bytes() {
+                    let bytes = if let Some(bytes) = reader.to_bytes() {
                         Cow::Borrowed(bytes)
                     } else {
                         // we have to read to an owned buffer to get the bytes.
                         let mut bytes = Vec::with_capacity(1024 * 128);
-                        r.read_to_end(&mut bytes)?;
+                        reader.read_to_end(&mut bytes)?;
                         if !bytes.is_empty()
                             && (bytes[bytes.len() - 1] != b'\n' || bytes[bytes.len() - 1] != b'\r')
                         {
@@ -478,10 +473,12 @@ impl<R: Read + Sync + Send + MmapBytesReader> SequentialReader<R> {
                         Cow::Owned(bytes)
                     };
 
-                    self.parse_csv(n_threads, &bytes, predicate.as_ref())?
+                    let out = self.parse_csv(n_threads, &bytes, predicate.as_ref())?;
+                    // restore reader for a potential second run
+                    self.reader = Some(reader);
+                    out
                 }
             }
-            _ => return Err(PolarsError::Other("file or reader must be set".into())),
         };
 
         if let Some(aggregate) = aggregate {
@@ -524,7 +521,7 @@ pub fn build_csv_reader<R: Read + Seek + Sync + Send + MmapBytesReader>(
     low_memory: bool,
     comment_char: Option<u8>,
     null_values: Option<NullValues>,
-) -> Result<SequentialReader<R>> {
+) -> Result<CoreReader<R>> {
     // check if schema should be inferred
     let delimiter = delimiter.unwrap_or(b',');
     let schema = match schema {
@@ -554,13 +551,10 @@ pub fn build_csv_reader<R: Read + Seek + Sync + Send + MmapBytesReader>(
         projection = Some(prj);
     }
 
-    let csv_reader = init_csv_reader(reader, has_header, delimiter, comment_char);
-    let record_iter = Some(csv_reader.into_byte_records());
-
-    Ok(SequentialReader {
+    Ok(CoreReader {
         schema,
         projection,
-        record_iter,
+        reader: Some(reader),
         line_number: if has_header { 1 } else { 0 },
         ignore_parser_errors,
         skip_rows,