net/http: limit chunked data overhead

The chunked transfer encoding adds some overhead to
the content transferred. When writing one byte per
chunk, for example, there are five bytes of overhead
per byte of data transferred: "1\r\nX\r\n" to send "X".

Chunks may include "chunk extensions",
which we skip over and do not use.
For example: "1;chunk extension here\r\nX\r\n".

A malicious sender can use chunk extensions to add
about 4k of overhead per byte of data.
(The maximum chunk header line size we will accept.)

Track the amount of overhead read in chunked data,
and produce an error if it seems excessive.

Fixes #64433
Fixes CVE-2023-39326

Change-Id: I40f8d70eb6f9575fb43f506eb19132ccedafcf39
Reviewed-on: https://team-review.git.corp.google.com/c/golang/go-private/+/2076135
Reviewed-by: Tatiana Bradley <[email protected]>
Reviewed-by: Roland Shoemaker <[email protected]>
Reviewed-on: https://go-review.googlesource.com/c/go/+/547335
Reviewed-by: Dmitri Shuralyov <[email protected]>
Reviewed-by: Dmitri Shuralyov <[email protected]>
LUCI-TryBot-Result: Go LUCI <[email protected]>

src/net/http/internal/chunked.go

@@ -39,7 +39,8 @@ type chunkedReader struct {
 	n        uint64 // unread bytes in chunk
 	err      error
 	buf      [2]byte
-	checkEnd bool // whether need to check for \r\n chunk footer
+	checkEnd bool  // whether need to check for \r\n chunk footer
+	excess   int64 // "excessive" chunk overhead, for malicious sender detection
 }
 
 func (cr *chunkedReader) beginChunk() {
@@ -49,10 +50,36 @@ func (cr *chunkedReader) beginChunk() {
 	if cr.err != nil {
 		return
 	}
+	cr.excess += int64(len(line)) + 2 // header, plus \r\n after the chunk data
+	line = trimTrailingWhitespace(line)
+	line, cr.err = removeChunkExtension(line)
+	if cr.err != nil {
+		return
+	}
 	cr.n, cr.err = parseHexUint(line)
 	if cr.err != nil {
 		return
 	}
+	// A sender who sends one byte per chunk will send 5 bytes of overhead
+	// for every byte of data. ("1\r\nX\r\n" to send "X".)
+	// We want to allow this, since streaming a byte at a time can be legitimate.
+	//
+	// A sender can use chunk extensions to add arbitrary amounts of additional
+	// data per byte read. ("1;very long extension\r\nX\r\n" to send "X".)
+	// We don't want to disallow extensions (although we discard them),
+	// but we also don't want to allow a sender to reduce the signal/noise ratio
+	// arbitrarily.
+	//
+	// We track the amount of excess overhead read,
+	// and produce an error if it grows too large.
+	//
+	// Currently, we say that we're willing to accept 16 bytes of overhead per chunk,
+	// plus twice the amount of real data in the chunk.
+	cr.excess -= 16 + (2 * int64(cr.n))
+	cr.excess = max(cr.excess, 0)
+	if cr.excess > 16*1024 {
+		cr.err = errors.New("chunked encoding contains too much non-data")
+	}
 	if cr.n == 0 {
 		cr.err = io.EOF
 	}
@@ -140,11 +167,6 @@ func readChunkLine(b *bufio.Reader) ([]byte, error) {
 	if len(p) >= maxLineLength {
 		return nil, ErrLineTooLong
 	}
-	p = trimTrailingWhitespace(p)
-	p, err = removeChunkExtension(p)
-	if err != nil {
-		return nil, err
-	}
 	return p, nil
 }
 

src/net/http/internal/chunked_test.go

@@ -239,3 +239,62 @@ func TestChunkEndReadError(t *testing.T) {
 		t.Errorf("expected %v, got %v", readErr, err)
 	}
 }
+
+func TestChunkReaderTooMuchOverhead(t *testing.T) {
+	// If the sender is sending 100x as many chunk header bytes as chunk data,
+	// we should reject the stream at some point.
+	chunk := []byte("1;")
+	for i := 0; i < 100; i++ {
+		chunk = append(chunk, 'a') // chunk extension
+	}
+	chunk = append(chunk, "\r\nX\r\n"...)
+	const bodylen = 1 << 20
+	r := NewChunkedReader(&funcReader{f: func(i int) ([]byte, error) {
+		if i < bodylen {
+			return chunk, nil
+		}
+		return []byte("0\r\n"), nil
+	}})
+	_, err := io.ReadAll(r)
+	if err == nil {
+		t.Fatalf("successfully read body with excessive overhead; want error")
+	}
+}
+
+func TestChunkReaderByteAtATime(t *testing.T) {
+	// Sending one byte per chunk should not trip the excess-overhead detection.
+	const bodylen = 1 << 20
+	r := NewChunkedReader(&funcReader{f: func(i int) ([]byte, error) {
+		if i < bodylen {
+			return []byte("1\r\nX\r\n"), nil
+		}
+		return []byte("0\r\n"), nil
+	}})
+	got, err := io.ReadAll(r)
+	if err != nil {
+		t.Errorf("unexpected error: %v", err)
+	}
+	if len(got) != bodylen {
+		t.Errorf("read %v bytes, want %v", len(got), bodylen)
+	}
+}
+
+type funcReader struct {
+	f   func(iteration int) ([]byte, error)
+	i   int
+	b   []byte
+	err error
+}
+
+func (r *funcReader) Read(p []byte) (n int, err error) {
+	if len(r.b) == 0 && r.err == nil {
+		r.b, r.err = r.f(r.i)
+		r.i++
+	}
+	n = copy(p, r.b)
+	r.b = r.b[n:]
+	if len(r.b) > 0 {
+		return n, nil
+	}
+	return n, r.err
+}