func skipCRLF(r *bufio.Reader) error {
// There maybe servers using single '\n' for line ending
if _, err := r.ReadSlice('\n'); err != nil {
errl.Println("Error reading CRLF:", err)
return err
}
return nil
}
// Learned from net.textproto. One difference is that this one keeps the
// ending '\n' in the returned line. Buf if there's only CRLF in the line,
// return nil for the line.
func readContinuedLineSlice(r *bufio.Reader) ([]byte, error) {
// feedly.com request headers contains things like:
// "$Authorization.feedly: $FeedlyAuth\r\n", so we must test for only
// continuation spaces.
isspace := func(b byte) bool {
return b == ' ' || b == '\t'
}
// Read the first line.
line, err := r.ReadSlice('\n')
if err != nil {
return nil, err
}
// There are servers that use \n for line ending, so trim first before check ending.
// For example, the 404 page for http://plan9.bell-labs.com/magic/man2html/1/2l
trimmed := TrimSpace(line)
if len(trimmed) == 0 {
if len(line) > 2 {
return nil, fmt.Errorf("malformed end of headers, len: %d, %#v", len(line), string(line))
}
return nil, nil
}
if isspace(line[0]) {
return nil, fmt.Errorf("malformed header, start with space: %#v", string(line))
}
// Optimistically assume that we have started to buffer the next line
// and it starts with an ASCII letter (the next header key), so we can
// avoid copying that buffered data around in memory and skipping over
// non-existent whitespace.
if r.Buffered() > 0 {
peek, err := r.Peek(1)
if err == nil && !isspace(peek[0]) {
return line, nil
}
}
var buf []byte
buf = append(buf, trimmed...)
// Read continuation lines.
for skipSpace(r) > 0 {
line, err := r.ReadSlice('\n')
if err != nil {
break
}
buf = append(buf, ' ')
buf = append(buf, TrimTrailingSpace(line)...)
}
buf = append(buf, '\r', '\n')
return buf, nil
}
// Only add headers that are of interest for a proxy into request/response's header map.
func (h *Header) parseHeader(reader *bufio.Reader, raw *bytes.Buffer, url *URL) (err error) {
h.ContLen = -1
dummyLastLine := []byte{}
// Read request header and body
var s, name, val, lastLine []byte
for {
if s, err = reader.ReadSlice('\n'); err != nil {
return
}
// There are servers that use \n for line ending, so trim first before check ending.
// For example, the 404 page for http://plan9.bell-labs.com/magic/man2html/1/2l
trimmed := TrimSpace(s)
if len(trimmed) == 0 { // end of headers
return
}
if (s[0] == ' ' || s[0] == '\t') && lastLine != nil { // multi-line header
// I've never seen multi-line header used in headers that's of interest.
// Disable multi-line support to avoid copy for now.
errl.Printf("Multi-line support disabled: %v %s", url, s)
return errNotSupported
// combine previous line with current line
// trimmed = bytes.Join([][]byte{lastLine, []byte{' '}, trimmed}, nil)
}
if name, val, err = splitHeader(trimmed); err != nil {
return
}
// Wait Go to solve/provide the string<->[]byte optimization
kn := string(name)
if parseFunc, ok := headerParser[kn]; ok {
// lastLine = append([]byte(nil), trimmed...) // copy to avoid next read invalidating the trimmed line
lastLine = dummyLastLine
val = TrimSpace(val)
if len(val) == 0 {
continue
}
parseFunc(h, ASCIIToLower(val), raw)
} else {
// mark this header as not of interest to proxy
lastLine = nil
}
if hopByHopHeader[kn] {
continue
}
raw.Write(s)
// debug.Printf("len %d %s", len(s), s)
}
return
}
// Use this function until we find Trailer headers actually in use.
func skipTrailer(r *bufio.Reader) error {
// It's possible to get trailer headers, but the body will always end with
// a line with just CRLF.
for {
s, err := r.ReadSlice('\n')
if err != nil {
errl.Println("skip trailer:", err)
return err
}
if len(s) == 2 && s[0] == '\r' && s[1] == '\n' {
return nil
}
errl.Printf("skip trailer: %#v", string(s))
if len(s) == 1 || len(s) == 2 {
return fmt.Errorf("malformed chunk body end: %#v", string(s))
}
}
}
