520e1220bb
TLS 1.3 deployment is currently blocked by buggy middleboxes throughout the ecosystem. As an experiment to better understand these bugs and the problems they are causing, implement TLS 1.3 variants with alternate encodings. These are still the same protocol, only encoded slightly differently. We will use what we learn from these experiments to guide the TLS 1.3 deployment strategy and proposals to the IETF, if any. These experiments only target the basic 1-RTT TLS 1.3 handshake. Based on what we learn from this experiment, we may try future variations to explore 0-RTT and HelloRetryRequest. When enabled, the server supports all TLS 1.3 variants while the client is configured to use a particular variant. Change-Id: I532411d1abc41314dc76acce0246879b754b4c61 Reviewed-on: https://boringssl-review.googlesource.com/17327 Reviewed-by: David Benjamin <davidben@google.com> Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
478 lines
14 KiB
Go
478 lines
14 KiB
Go
// Copyright 2014 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.
|
|
|
|
// DTLS implementation.
|
|
//
|
|
// NOTE: This is a not even a remotely production-quality DTLS
|
|
// implementation. It is the bare minimum necessary to be able to
|
|
// achieve coverage on BoringSSL's implementation. Of note is that
|
|
// this implementation assumes the underlying net.PacketConn is not
|
|
// only reliable but also ordered. BoringSSL will be expected to deal
|
|
// with simulated loss, but there is no point in forcing the test
|
|
// driver to.
|
|
|
|
package runner
|
|
|
|
import (
|
|
"bytes"
|
|
"errors"
|
|
"fmt"
|
|
"io"
|
|
"math/rand"
|
|
"net"
|
|
)
|
|
|
|
func wireToVersion(vers uint16, isDTLS bool) (uint16, bool) {
|
|
if isDTLS {
|
|
switch vers {
|
|
case VersionDTLS12:
|
|
return VersionTLS12, true
|
|
case VersionDTLS10:
|
|
return VersionTLS10, true
|
|
}
|
|
} else {
|
|
switch vers {
|
|
case VersionSSL30, VersionTLS10, VersionTLS11, VersionTLS12:
|
|
return vers, true
|
|
case tls13DraftVersion, tls13ExperimentVersion:
|
|
return VersionTLS13, true
|
|
}
|
|
}
|
|
|
|
return 0, false
|
|
}
|
|
|
|
func (c *Conn) dtlsDoReadRecord(want recordType) (recordType, *block, error) {
|
|
recordHeaderLen := dtlsRecordHeaderLen
|
|
|
|
if c.rawInput == nil {
|
|
c.rawInput = c.in.newBlock()
|
|
}
|
|
b := c.rawInput
|
|
|
|
// Read a new packet only if the current one is empty.
|
|
var newPacket bool
|
|
if len(b.data) == 0 {
|
|
// Pick some absurdly large buffer size.
|
|
b.resize(maxCiphertext + recordHeaderLen)
|
|
n, err := c.conn.Read(c.rawInput.data)
|
|
if err != nil {
|
|
return 0, nil, err
|
|
}
|
|
if c.config.Bugs.MaxPacketLength != 0 && n > c.config.Bugs.MaxPacketLength {
|
|
return 0, nil, fmt.Errorf("dtls: exceeded maximum packet length")
|
|
}
|
|
c.rawInput.resize(n)
|
|
newPacket = true
|
|
}
|
|
|
|
// Read out one record.
|
|
//
|
|
// A real DTLS implementation should be tolerant of errors,
|
|
// but this is test code. We should not be tolerant of our
|
|
// peer sending garbage.
|
|
if len(b.data) < recordHeaderLen {
|
|
return 0, nil, errors.New("dtls: failed to read record header")
|
|
}
|
|
typ := recordType(b.data[0])
|
|
vers := uint16(b.data[1])<<8 | uint16(b.data[2])
|
|
// Alerts sent near version negotiation do not have a well-defined
|
|
// record-layer version prior to TLS 1.3. (In TLS 1.3, the record-layer
|
|
// version is irrelevant.)
|
|
if typ != recordTypeAlert {
|
|
if c.haveVers {
|
|
if vers != c.wireVersion {
|
|
c.sendAlert(alertProtocolVersion)
|
|
return 0, nil, c.in.setErrorLocked(fmt.Errorf("dtls: received record with version %x when expecting version %x", vers, c.wireVersion))
|
|
}
|
|
} else {
|
|
// Pre-version-negotiation alerts may be sent with any version.
|
|
if expect := c.config.Bugs.ExpectInitialRecordVersion; expect != 0 && vers != expect {
|
|
c.sendAlert(alertProtocolVersion)
|
|
return 0, nil, c.in.setErrorLocked(fmt.Errorf("dtls: received record with version %x when expecting version %x", vers, expect))
|
|
}
|
|
}
|
|
}
|
|
epoch := b.data[3:5]
|
|
seq := b.data[5:11]
|
|
// For test purposes, require the sequence number be monotonically
|
|
// increasing, so c.in includes the minimum next sequence number. Gaps
|
|
// may occur if packets failed to be sent out. A real implementation
|
|
// would maintain a replay window and such.
|
|
if !bytes.Equal(epoch, c.in.seq[:2]) {
|
|
c.sendAlert(alertIllegalParameter)
|
|
return 0, nil, c.in.setErrorLocked(fmt.Errorf("dtls: bad epoch"))
|
|
}
|
|
if bytes.Compare(seq, c.in.seq[2:]) < 0 {
|
|
c.sendAlert(alertIllegalParameter)
|
|
return 0, nil, c.in.setErrorLocked(fmt.Errorf("dtls: bad sequence number"))
|
|
}
|
|
copy(c.in.seq[2:], seq)
|
|
n := int(b.data[11])<<8 | int(b.data[12])
|
|
if n > maxCiphertext || len(b.data) < recordHeaderLen+n {
|
|
c.sendAlert(alertRecordOverflow)
|
|
return 0, nil, c.in.setErrorLocked(fmt.Errorf("dtls: oversized record received with length %d", n))
|
|
}
|
|
|
|
// Process message.
|
|
b, c.rawInput = c.in.splitBlock(b, recordHeaderLen+n)
|
|
ok, off, _, alertValue := c.in.decrypt(b)
|
|
if !ok {
|
|
// A real DTLS implementation would silently ignore bad records,
|
|
// but we want to notice errors from the implementation under
|
|
// test.
|
|
return 0, nil, c.in.setErrorLocked(c.sendAlert(alertValue))
|
|
}
|
|
b.off = off
|
|
|
|
// TODO(nharper): Once DTLS 1.3 is defined, handle the extra
|
|
// parameter from decrypt.
|
|
|
|
// Require that ChangeCipherSpec always share a packet with either the
|
|
// previous or next handshake message.
|
|
if newPacket && typ == recordTypeChangeCipherSpec && c.rawInput == nil {
|
|
return 0, nil, c.in.setErrorLocked(fmt.Errorf("dtls: ChangeCipherSpec not packed together with Finished"))
|
|
}
|
|
|
|
return typ, b, nil
|
|
}
|
|
|
|
func (c *Conn) makeFragment(header, data []byte, fragOffset, fragLen int) []byte {
|
|
fragment := make([]byte, 0, 12+fragLen)
|
|
fragment = append(fragment, header...)
|
|
fragment = append(fragment, byte(c.sendHandshakeSeq>>8), byte(c.sendHandshakeSeq))
|
|
fragment = append(fragment, byte(fragOffset>>16), byte(fragOffset>>8), byte(fragOffset))
|
|
fragment = append(fragment, byte(fragLen>>16), byte(fragLen>>8), byte(fragLen))
|
|
fragment = append(fragment, data[fragOffset:fragOffset+fragLen]...)
|
|
return fragment
|
|
}
|
|
|
|
func (c *Conn) dtlsWriteRecord(typ recordType, data []byte) (n int, err error) {
|
|
if typ != recordTypeHandshake {
|
|
// Only handshake messages are fragmented.
|
|
n, err = c.dtlsWriteRawRecord(typ, data)
|
|
if err != nil {
|
|
return
|
|
}
|
|
|
|
if typ == recordTypeChangeCipherSpec {
|
|
err = c.out.changeCipherSpec(c.config)
|
|
if err != nil {
|
|
return n, c.sendAlertLocked(alertLevelError, err.(alert))
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
if c.out.cipher == nil && c.config.Bugs.StrayChangeCipherSpec {
|
|
_, err = c.dtlsWriteRawRecord(recordTypeChangeCipherSpec, []byte{1})
|
|
if err != nil {
|
|
return
|
|
}
|
|
}
|
|
|
|
maxLen := c.config.Bugs.MaxHandshakeRecordLength
|
|
if maxLen <= 0 {
|
|
maxLen = 1024
|
|
}
|
|
|
|
// Handshake messages have to be modified to include fragment
|
|
// offset and length and with the header replicated. Save the
|
|
// TLS header here.
|
|
//
|
|
// TODO(davidben): This assumes that data contains exactly one
|
|
// handshake message. This is incompatible with
|
|
// FragmentAcrossChangeCipherSpec. (Which is unfortunate
|
|
// because OpenSSL's DTLS implementation will probably accept
|
|
// such fragmentation and could do with a fix + tests.)
|
|
header := data[:4]
|
|
data = data[4:]
|
|
|
|
isFinished := header[0] == typeFinished
|
|
|
|
if c.config.Bugs.SendEmptyFragments {
|
|
fragment := c.makeFragment(header, data, 0, 0)
|
|
c.pendingFragments = append(c.pendingFragments, fragment)
|
|
}
|
|
|
|
firstRun := true
|
|
fragOffset := 0
|
|
for firstRun || fragOffset < len(data) {
|
|
firstRun = false
|
|
fragLen := len(data) - fragOffset
|
|
if fragLen > maxLen {
|
|
fragLen = maxLen
|
|
}
|
|
|
|
fragment := c.makeFragment(header, data, fragOffset, fragLen)
|
|
if c.config.Bugs.FragmentMessageTypeMismatch && fragOffset > 0 {
|
|
fragment[0]++
|
|
}
|
|
if c.config.Bugs.FragmentMessageLengthMismatch && fragOffset > 0 {
|
|
fragment[3]++
|
|
}
|
|
|
|
// Buffer the fragment for later. They will be sent (and
|
|
// reordered) on flush.
|
|
c.pendingFragments = append(c.pendingFragments, fragment)
|
|
if c.config.Bugs.ReorderHandshakeFragments {
|
|
// Don't duplicate Finished to avoid the peer
|
|
// interpreting it as a retransmit request.
|
|
if !isFinished {
|
|
c.pendingFragments = append(c.pendingFragments, fragment)
|
|
}
|
|
|
|
if fragLen > (maxLen+1)/2 {
|
|
// Overlap each fragment by half.
|
|
fragLen = (maxLen + 1) / 2
|
|
}
|
|
}
|
|
fragOffset += fragLen
|
|
n += fragLen
|
|
}
|
|
if !isFinished && c.config.Bugs.MixCompleteMessageWithFragments {
|
|
fragment := c.makeFragment(header, data, 0, len(data))
|
|
c.pendingFragments = append(c.pendingFragments, fragment)
|
|
}
|
|
|
|
// Increment the handshake sequence number for the next
|
|
// handshake message.
|
|
c.sendHandshakeSeq++
|
|
return
|
|
}
|
|
|
|
func (c *Conn) dtlsFlushHandshake() error {
|
|
// This is a test-only DTLS implementation, so there is no need to
|
|
// retain |c.pendingFragments| for a future retransmit.
|
|
var fragments [][]byte
|
|
fragments, c.pendingFragments = c.pendingFragments, fragments
|
|
|
|
if c.config.Bugs.ReorderHandshakeFragments {
|
|
perm := rand.New(rand.NewSource(0)).Perm(len(fragments))
|
|
tmp := make([][]byte, len(fragments))
|
|
for i := range tmp {
|
|
tmp[i] = fragments[perm[i]]
|
|
}
|
|
fragments = tmp
|
|
} else if c.config.Bugs.ReverseHandshakeFragments {
|
|
tmp := make([][]byte, len(fragments))
|
|
for i := range tmp {
|
|
tmp[i] = fragments[len(fragments)-i-1]
|
|
}
|
|
fragments = tmp
|
|
}
|
|
|
|
maxRecordLen := c.config.Bugs.PackHandshakeFragments
|
|
maxPacketLen := c.config.Bugs.PackHandshakeRecords
|
|
|
|
// Pack handshake fragments into records.
|
|
var records [][]byte
|
|
for _, fragment := range fragments {
|
|
if n := c.config.Bugs.SplitFragments; n > 0 {
|
|
if len(fragment) > n {
|
|
records = append(records, fragment[:n])
|
|
records = append(records, fragment[n:])
|
|
} else {
|
|
records = append(records, fragment)
|
|
}
|
|
} else if i := len(records) - 1; len(records) > 0 && len(records[i])+len(fragment) <= maxRecordLen {
|
|
records[i] = append(records[i], fragment...)
|
|
} else {
|
|
// The fragment will be appended to, so copy it.
|
|
records = append(records, append([]byte{}, fragment...))
|
|
}
|
|
}
|
|
|
|
// Format them into packets.
|
|
var packets [][]byte
|
|
for _, record := range records {
|
|
b, err := c.dtlsSealRecord(recordTypeHandshake, record)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
if i := len(packets) - 1; len(packets) > 0 && len(packets[i])+len(b.data) <= maxPacketLen {
|
|
packets[i] = append(packets[i], b.data...)
|
|
} else {
|
|
// The sealed record will be appended to and reused by
|
|
// |c.out|, so copy it.
|
|
packets = append(packets, append([]byte{}, b.data...))
|
|
}
|
|
c.out.freeBlock(b)
|
|
}
|
|
|
|
// Send all the packets.
|
|
for _, packet := range packets {
|
|
if _, err := c.conn.Write(packet); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// dtlsSealRecord seals a record into a block from |c.out|'s pool.
|
|
func (c *Conn) dtlsSealRecord(typ recordType, data []byte) (b *block, err error) {
|
|
recordHeaderLen := dtlsRecordHeaderLen
|
|
maxLen := c.config.Bugs.MaxHandshakeRecordLength
|
|
if maxLen <= 0 {
|
|
maxLen = 1024
|
|
}
|
|
|
|
b = c.out.newBlock()
|
|
|
|
explicitIVLen := 0
|
|
explicitIVIsSeq := false
|
|
|
|
if cbc, ok := c.out.cipher.(cbcMode); ok {
|
|
// Block cipher modes have an explicit IV.
|
|
explicitIVLen = cbc.BlockSize()
|
|
} else if aead, ok := c.out.cipher.(*tlsAead); ok {
|
|
if aead.explicitNonce {
|
|
explicitIVLen = 8
|
|
// The AES-GCM construction in TLS has an explicit nonce so that
|
|
// the nonce can be random. However, the nonce is only 8 bytes
|
|
// which is too small for a secure, random nonce. Therefore we
|
|
// use the sequence number as the nonce.
|
|
explicitIVIsSeq = true
|
|
}
|
|
} else if _, ok := c.out.cipher.(nullCipher); !ok && c.out.cipher != nil {
|
|
panic("Unknown cipher")
|
|
}
|
|
b.resize(recordHeaderLen + explicitIVLen + len(data))
|
|
// TODO(nharper): DTLS 1.3 will likely need to set this to
|
|
// recordTypeApplicationData if c.out.cipher != nil.
|
|
b.data[0] = byte(typ)
|
|
vers := c.wireVersion
|
|
if vers == 0 {
|
|
// Some TLS servers fail if the record version is greater than
|
|
// TLS 1.0 for the initial ClientHello.
|
|
if c.isDTLS {
|
|
vers = VersionDTLS10
|
|
} else {
|
|
vers = VersionTLS10
|
|
}
|
|
}
|
|
b.data[1] = byte(vers >> 8)
|
|
b.data[2] = byte(vers)
|
|
// DTLS records include an explicit sequence number.
|
|
copy(b.data[3:11], c.out.outSeq[0:])
|
|
b.data[11] = byte(len(data) >> 8)
|
|
b.data[12] = byte(len(data))
|
|
if explicitIVLen > 0 {
|
|
explicitIV := b.data[recordHeaderLen : recordHeaderLen+explicitIVLen]
|
|
if explicitIVIsSeq {
|
|
copy(explicitIV, c.out.outSeq[:])
|
|
} else {
|
|
if _, err = io.ReadFull(c.config.rand(), explicitIV); err != nil {
|
|
return
|
|
}
|
|
}
|
|
}
|
|
copy(b.data[recordHeaderLen+explicitIVLen:], data)
|
|
c.out.encrypt(b, explicitIVLen, typ)
|
|
return
|
|
}
|
|
|
|
func (c *Conn) dtlsWriteRawRecord(typ recordType, data []byte) (n int, err error) {
|
|
b, err := c.dtlsSealRecord(typ, data)
|
|
if err != nil {
|
|
return
|
|
}
|
|
|
|
_, err = c.conn.Write(b.data)
|
|
if err != nil {
|
|
return
|
|
}
|
|
n = len(data)
|
|
|
|
c.out.freeBlock(b)
|
|
return
|
|
}
|
|
|
|
func (c *Conn) dtlsDoReadHandshake() ([]byte, error) {
|
|
// Assemble a full handshake message. For test purposes, this
|
|
// implementation assumes fragments arrive in order. It may
|
|
// need to be cleverer if we ever test BoringSSL's retransmit
|
|
// behavior.
|
|
for len(c.handMsg) < 4+c.handMsgLen {
|
|
// Get a new handshake record if the previous has been
|
|
// exhausted.
|
|
if c.hand.Len() == 0 {
|
|
if err := c.in.err; err != nil {
|
|
return nil, err
|
|
}
|
|
if err := c.readRecord(recordTypeHandshake); err != nil {
|
|
return nil, err
|
|
}
|
|
}
|
|
|
|
// Read the next fragment. It must fit entirely within
|
|
// the record.
|
|
if c.hand.Len() < 12 {
|
|
return nil, errors.New("dtls: bad handshake record")
|
|
}
|
|
header := c.hand.Next(12)
|
|
fragN := int(header[1])<<16 | int(header[2])<<8 | int(header[3])
|
|
fragSeq := uint16(header[4])<<8 | uint16(header[5])
|
|
fragOff := int(header[6])<<16 | int(header[7])<<8 | int(header[8])
|
|
fragLen := int(header[9])<<16 | int(header[10])<<8 | int(header[11])
|
|
|
|
if c.hand.Len() < fragLen {
|
|
return nil, errors.New("dtls: fragment length too long")
|
|
}
|
|
fragment := c.hand.Next(fragLen)
|
|
|
|
// Check it's a fragment for the right message.
|
|
if fragSeq != c.recvHandshakeSeq {
|
|
return nil, errors.New("dtls: bad handshake sequence number")
|
|
}
|
|
|
|
// Check that the length is consistent.
|
|
if c.handMsg == nil {
|
|
c.handMsgLen = fragN
|
|
if c.handMsgLen > maxHandshake {
|
|
return nil, c.in.setErrorLocked(c.sendAlert(alertInternalError))
|
|
}
|
|
// Start with the TLS handshake header,
|
|
// without the DTLS bits.
|
|
c.handMsg = append([]byte{}, header[:4]...)
|
|
} else if fragN != c.handMsgLen {
|
|
return nil, errors.New("dtls: bad handshake length")
|
|
}
|
|
|
|
// Add the fragment to the pending message.
|
|
if 4+fragOff != len(c.handMsg) {
|
|
return nil, errors.New("dtls: bad fragment offset")
|
|
}
|
|
if fragOff+fragLen > c.handMsgLen {
|
|
return nil, errors.New("dtls: bad fragment length")
|
|
}
|
|
c.handMsg = append(c.handMsg, fragment...)
|
|
}
|
|
c.recvHandshakeSeq++
|
|
ret := c.handMsg
|
|
c.handMsg, c.handMsgLen = nil, 0
|
|
return ret, nil
|
|
}
|
|
|
|
// DTLSServer returns a new DTLS server side connection
|
|
// using conn as the underlying transport.
|
|
// The configuration config must be non-nil and must have
|
|
// at least one certificate.
|
|
func DTLSServer(conn net.Conn, config *Config) *Conn {
|
|
c := &Conn{config: config, isDTLS: true, conn: conn}
|
|
c.init()
|
|
return c
|
|
}
|
|
|
|
// DTLSClient returns a new DTLS client side connection
|
|
// using conn as the underlying transport.
|
|
// The config cannot be nil: users must set either ServerHostname or
|
|
// InsecureSkipVerify in the config.
|
|
func DTLSClient(conn net.Conn, config *Config) *Conn {
|
|
c := &Conn{config: config, isClient: true, isDTLS: true, conn: conn}
|
|
c.init()
|
|
return c
|
|
}
|