You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 
gitea/vendor/github.com/42wim/sshsig/pem/encode.go

346 lines
8.8 KiB

// Modified by 42wim
//
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// Package pem implements the PEM data encoding, which originated in Privacy
// Enhanced Mail. The most common use of PEM encoding today is in TLS keys and
// certificates. See RFC 1421.
package pem
import (
"bytes"
"encoding/base64"
"errors"
"io"
"sort"
"strings"
)
// A Block represents a PEM encoded structure.
//
// The encoded form is:
// -----BEGIN Type-----
// Headers
// base64-encoded Bytes
// -----END Type-----
// where Headers is a possibly empty sequence of Key: Value lines.
type Block struct {
Type string // The type, taken from the preamble (i.e. "RSA PRIVATE KEY").
Headers map[string]string // Optional headers.
Bytes []byte // The decoded bytes of the contents. Typically a DER encoded ASN.1 structure.
}
// getLine results the first \r\n or \n delineated line from the given byte
// array. The line does not include trailing whitespace or the trailing new
// line bytes. The remainder of the byte array (also not including the new line
// bytes) is also returned and this will always be smaller than the original
// argument.
func getLine(data []byte) (line, rest []byte) {
i := bytes.IndexByte(data, '\n')
var j int
if i < 0 {
i = len(data)
j = i
} else {
j = i + 1
if i > 0 && data[i-1] == '\r' {
i--
}
}
return bytes.TrimRight(data[0:i], " \t"), data[j:]
}
// removeSpacesAndTabs returns a copy of its input with all spaces and tabs
// removed, if there were any. Otherwise, the input is returned unchanged.
//
// The base64 decoder already skips newline characters, so we don't need to
// filter them out here.
func removeSpacesAndTabs(data []byte) []byte {
if !bytes.ContainsAny(data, " \t") {
// Fast path; most base64 data within PEM contains newlines, but
// no spaces nor tabs. Skip the extra alloc and work.
return data
}
result := make([]byte, len(data))
n := 0
for _, b := range data {
if b == ' ' || b == '\t' {
continue
}
result[n] = b
n++
}
return result[0:n]
}
var (
pemStart = []byte("\n-----BEGIN ")
pemEnd = []byte("\n-----END ")
pemEndOfLine = []byte("-----")
)
// Decode will find the next PEM formatted block (certificate, private key
// etc) in the input. It returns that block and the remainder of the input. If
// no PEM data is found, p is nil and the whole of the input is returned in
// rest.
func Decode(data []byte) (p *Block, rest []byte) {
// pemStart begins with a newline. However, at the very beginning of
// the byte array, we'll accept the start string without it.
rest = data
if bytes.HasPrefix(data, pemStart[1:]) {
rest = rest[len(pemStart)-1 : len(data)]
} else if i := bytes.Index(data, pemStart); i >= 0 {
rest = rest[i+len(pemStart) : len(data)]
} else {
return nil, data
}
typeLine, rest := getLine(rest)
if !bytes.HasSuffix(typeLine, pemEndOfLine) {
return decodeError(data, rest)
}
typeLine = typeLine[0 : len(typeLine)-len(pemEndOfLine)]
p = &Block{
Headers: make(map[string]string),
Type: string(typeLine),
}
for {
// This loop terminates because getLine's second result is
// always smaller than its argument.
if len(rest) == 0 {
return nil, data
}
line, next := getLine(rest)
i := bytes.IndexByte(line, ':')
if i == -1 {
break
}
// TODO(agl): need to cope with values that spread across lines.
key, val := line[:i], line[i+1:]
key = bytes.TrimSpace(key)
val = bytes.TrimSpace(val)
p.Headers[string(key)] = string(val)
rest = next
}
var endIndex, endTrailerIndex int
// If there were no headers, the END line might occur
// immediately, without a leading newline.
if len(p.Headers) == 0 && bytes.HasPrefix(rest, pemEnd[1:]) {
endIndex = 0
endTrailerIndex = len(pemEnd) - 1
} else {
endIndex = bytes.Index(rest, pemEnd)
endTrailerIndex = endIndex + len(pemEnd)
}
if endIndex < 0 {
return decodeError(data, rest)
}
// After the "-----" of the ending line, there should be the same type
// and then a final five dashes.
endTrailer := rest[endTrailerIndex:]
endTrailerLen := len(typeLine) + len(pemEndOfLine)
if len(endTrailer) < endTrailerLen {
return decodeError(data, rest)
}
restOfEndLine := endTrailer[endTrailerLen:]
endTrailer = endTrailer[:endTrailerLen]
if !bytes.HasPrefix(endTrailer, typeLine) ||
!bytes.HasSuffix(endTrailer, pemEndOfLine) {
return decodeError(data, rest)
}
// The line must end with only whitespace.
if s, _ := getLine(restOfEndLine); len(s) != 0 {
return decodeError(data, rest)
}
base64Data := removeSpacesAndTabs(rest[:endIndex])
p.Bytes = make([]byte, base64.StdEncoding.DecodedLen(len(base64Data)))
n, err := base64.StdEncoding.Decode(p.Bytes, base64Data)
if err != nil {
return decodeError(data, rest)
}
p.Bytes = p.Bytes[:n]
// the -1 is because we might have only matched pemEnd without the
// leading newline if the PEM block was empty.
_, rest = getLine(rest[endIndex+len(pemEnd)-1:])
return
}
func decodeError(data, rest []byte) (*Block, []byte) {
// If we get here then we have rejected a likely looking, but
// ultimately invalid PEM block. We need to start over from a new
// position. We have consumed the preamble line and will have consumed
// any lines which could be header lines. However, a valid preamble
// line is not a valid header line, therefore we cannot have consumed
// the preamble line for the any subsequent block. Thus, we will always
// find any valid block, no matter what bytes precede it.
//
// For example, if the input is
//
// -----BEGIN MALFORMED BLOCK-----
// junk that may look like header lines
// or data lines, but no END line
//
// -----BEGIN ACTUAL BLOCK-----
// realdata
// -----END ACTUAL BLOCK-----
//
// we've failed to parse using the first BEGIN line
// and now will try again, using the second BEGIN line.
p, rest := Decode(rest)
if p == nil {
rest = data
}
return p, rest
}
// https://github.com/openssh/openssh-portable/blob/master/PROTOCOL.sshsig#L20-L21
// Should be on 76 chars, but strangely enough ssh breaks on 70
const pemLineLength = 70
type lineBreaker struct {
line [pemLineLength]byte
used int
out io.Writer
}
var nl = []byte{'\n'}
func (l *lineBreaker) Write(b []byte) (n int, err error) {
if l.used+len(b) < pemLineLength {
copy(l.line[l.used:], b)
l.used += len(b)
return len(b), nil
}
n, err = l.out.Write(l.line[0:l.used])
if err != nil {
return
}
excess := pemLineLength - l.used
l.used = 0
n, err = l.out.Write(b[0:excess])
if err != nil {
return
}
n, err = l.out.Write(nl)
if err != nil {
return
}
return l.Write(b[excess:])
}
func (l *lineBreaker) Close() (err error) {
if l.used > 0 {
_, err = l.out.Write(l.line[0:l.used])
if err != nil {
return
}
_, err = l.out.Write(nl)
}
return
}
func writeHeader(out io.Writer, k, v string) error {
_, err := out.Write([]byte(k + ": " + v + "\n"))
return err
}
// Encode writes the PEM encoding of b to out.
func Encode(out io.Writer, b *Block) error {
// Check for invalid block before writing any output.
for k := range b.Headers {
if strings.Contains(k, ":") {
return errors.New("pem: cannot encode a header key that contains a colon")
}
}
// All errors below are relayed from underlying io.Writer,
// so it is now safe to write data.
if _, err := out.Write(pemStart[1:]); err != nil {
return err
}
if _, err := out.Write([]byte(b.Type + "-----\n")); err != nil {
return err
}
if len(b.Headers) > 0 {
const procType = "Proc-Type"
h := make([]string, 0, len(b.Headers))
hasProcType := false
for k := range b.Headers {
if k == procType {
hasProcType = true
continue
}
h = append(h, k)
}
// The Proc-Type header must be written first.
// See RFC 1421, section 4.6.1.1
if hasProcType {
if err := writeHeader(out, procType, b.Headers[procType]); err != nil {
return err
}
}
// For consistency of output, write other headers sorted by key.
sort.Strings(h)
for _, k := range h {
if err := writeHeader(out, k, b.Headers[k]); err != nil {
return err
}
}
if _, err := out.Write(nl); err != nil {
return err
}
}
var breaker lineBreaker
breaker.out = out
b64 := base64.NewEncoder(base64.StdEncoding, &breaker)
if _, err := b64.Write(b.Bytes); err != nil {
return err
}
b64.Close()
breaker.Close()
if _, err := out.Write(pemEnd[1:]); err != nil {
return err
}
_, err := out.Write([]byte(b.Type + "-----\n"))
return err
}
// EncodeToMemory returns the PEM encoding of b.
//
// If b has invalid headers and cannot be encoded,
// EncodeToMemory returns nil. If it is important to
// report details about this error case, use Encode instead.
func EncodeToMemory(b *Block) []byte {
var buf bytes.Buffer
if err := Encode(&buf, b); err != nil {
return nil
}
return buf.Bytes()
}