Add DTLS support to NSS, contributed by Eric Rescorla.

net/third_party/nss/ssl/dtls1con.c

+/*
+ * DTLS Protocol
+ *
+ * ***** BEGIN LICENSE BLOCK *****
+ * Version: MPL 1.1/GPL 2.0/LGPL 2.1
+ *
+ * The contents of this file are subject to the Mozilla Public License Version
+ * 1.1 (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ * http://www.mozilla.org/MPL/
+ *
+ * Software distributed under the License is distributed on an "AS IS" basis,
+ * WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License
+ * for the specific language governing rights and limitations under the
+ * License.
+ *
+ * The Original Code is the Netscape security libraries.
+ *
+ * The Initial Developer of the Original Code is
+ * Netscape Communications Corporation.
+ * Portions created by the Initial Developer are Copyright (C) 1994-2000
+ * the Initial Developer. All Rights Reserved.
+ *
+ * Contributor(s):
+ *   Eric Rescorla <ekr@rtfm.com>
+ *
+ * Alternatively, the contents of this file may be used under the terms of
+ * either the GNU General Public License Version 2 or later (the "GPL"), or
+ * the GNU Lesser General Public License Version 2.1 or later (the "LGPL"),
+ * in which case the provisions of the GPL or the LGPL are applicable instead
+ * of those above. If you wish to allow use of your version of this file only
+ * under the terms of either the GPL or the LGPL, and not to allow others to
+ * use your version of this file under the terms of the MPL, indicate your
+ * decision by deleting the provisions above and replace them with the notice
+ * and other provisions required by the GPL or the LGPL. If you do not delete
+ * the provisions above, a recipient may use your version of this file under
+ * the terms of any one of the MPL, the GPL or the LGPL.
+ *
+ * ***** END LICENSE BLOCK ***** */
+/* $Id: $ */
+
+#include "ssl.h"
+#include "sslimpl.h"
+#include "sslproto.h"
+
+#ifndef PR_ARRAY_SIZE
+#define PR_ARRAY_SIZE(a) (sizeof(a)/sizeof((a)[0]))
+#endif
+
+static SECStatus dtls_TransmitMessageFlight(sslSocket *ss);
+static void dtls_RetransmitTimerExpiredCb(sslSocket *ss);
+static SECStatus dtls_SendSavedWriteData(sslSocket *ss);
+
+/* -28 adjusts for the IP/UDP header */
+static const PRUint16 COMMON_MTU_VALUES[] = {
+    1500 - 28,  /* Ethernet MTU */
+    1280 - 28,  /* IPv6 minimum MTU */
+    576 - 28,   /* Common assumption */
+    256 - 28    /* We're in serious trouble now */
+};
+
+#define DTLS_COOKIE_BYTES 32
+
+/* List copied from ssl3con.c:cipherSuites */
+static const ssl3CipherSuite nonDTLSSuites[] = {
+#ifdef NSS_ENABLE_ECC
+    TLS_ECDHE_ECDSA_WITH_RC4_128_SHA,
+    TLS_ECDHE_RSA_WITH_RC4_128_SHA,
+#endif  /* NSS_ENABLE_ECC */
+    TLS_DHE_DSS_WITH_RC4_128_SHA,
+#ifdef NSS_ENABLE_ECC
+    TLS_ECDH_RSA_WITH_RC4_128_SHA,
+    TLS_ECDH_ECDSA_WITH_RC4_128_SHA,
+#endif  /* NSS_ENABLE_ECC */
+    SSL_RSA_WITH_RC4_128_MD5,
+    SSL_RSA_WITH_RC4_128_SHA,
+    TLS_RSA_EXPORT1024_WITH_RC4_56_SHA,
+    SSL_RSA_EXPORT_WITH_RC4_40_MD5,
+    0 /* End of list marker */
+};
+
+/* Map back and forth between TLS and DTLS versions in wire format.
+ * Mapping table is:
+ *
+ * TLS             DTLS
+ * 1.1 (0302)      1.0 (feff)
+ */
+SSL3ProtocolVersion
+dtls_TLSVersionToDTLSVersion(SSL3ProtocolVersion tlsv)
+{
+    /* Anything other than TLS 1.1 is an error, so return
+     * the invalid version ffff. */
+    if (tlsv != SSL_LIBRARY_VERSION_TLS_1_1)
+        return 0xffff;
+
+    return SSL_LIBRARY_VERSION_DTLS_1_0_WIRE;
+}
+
+/* Map known DTLS versions to known TLS versions.
+ * - Invalid versions (< 1.0) return a version of 0
+ * - Versions > known return a version one higher than we know of
+ * to accomodate a theoretically newer version */
+SSL3ProtocolVersion
+dtls_DTLSVersionToTLSVersion(SSL3ProtocolVersion dtlsv)
+{
+    if (MSB(dtlsv) == 0xff) {
+        return 0;
+    }
+
+    if (dtlsv == SSL_LIBRARY_VERSION_DTLS_1_0_WIRE)
+        return SSL_LIBRARY_VERSION_TLS_1_1;
+
+    /* Return a fictional higher version than we know of */
+    return SSL_LIBRARY_VERSION_TLS_1_1 + 1;
+}
+
+/* On this socket, Disable non-DTLS cipher suites in the argument's list */
+SECStatus
+ssl3_DisableNonDTLSSuites(sslSocket * ss)
+{
+    const ssl3CipherSuite * suite;
+
+    for (suite = nonDTLSSuites; *suite; ++suite) {
+        SECStatus rv = ssl3_CipherPrefSet(ss, *suite, PR_FALSE);
+
+        PORT_Assert(rv == SECSuccess); /* else is coding error */
+    }
+    return SECSuccess;
+}
+
+/* Allocate a DTLSQueuedMessage.
+ *
+ * Called from dtls_QueueMessage()
+ */
+static DTLSQueuedMessage *
+dtls_AllocQueuedMessage(PRUint16 epoch, SSL3ContentType type,
+                        const unsigned char *data, PRUint32 len)
+{
+    DTLSQueuedMessage *msg = NULL;
+
+    msg = PORT_ZAlloc(sizeof(DTLSQueuedMessage));
+    if (!msg)
+        return NULL;
+
+    msg->data = PORT_Alloc(len);
+    if (!msg->data) {
+        PORT_Free(msg);
+        return NULL;
+    }
+    PORT_Memcpy(msg->data, data, len);
+
+    msg->len = len;
+    msg->epoch = epoch;
+    msg->type = type;
+
+    return msg;
+}
+
+/*
+ * Free a handshake message
+ *
+ * Called from dtls_FreeHandshakeMessages()
+ */
+static void
+dtls_FreeHandshakeMessage(DTLSQueuedMessage *msg)
+{
+    if (!msg)
+        return;
+
+    PORT_ZFree(msg->data, msg->len);
+    PORT_Free(msg);
+}
+
+/*
+ * Free a list of handshake messages
+ *
+ * Called from:
+ *              dtls_HandleHandshake()
+ *              ssl3_DestroySSL3Info()
+ */
+void
+dtls_FreeHandshakeMessages(PRCList *list)
+{
+    PRCList *cur_p;
+
+    while (!PR_CLIST_IS_EMPTY(list)) {
+        cur_p = PR_LIST_TAIL(list);
+        PR_REMOVE_LINK(cur_p);
+        dtls_FreeHandshakeMessage((DTLSQueuedMessage *)cur_p);
+    }
+}
+
+/* Called only from ssl3_HandleRecord, for each (deciphered) DTLS record.
+ * origBuf is the decrypted ssl record content and is expected to contain
+ * complete handshake records
+ * Caller must hold the handshake and RecvBuf locks.
+ *
+ * Note that this code uses msg_len for two purposes:
+ *
+ * (1) To pass the length to ssl3_HandleHandshakeMessage()
+ * (2) To carry the length of a message currently being reassembled
+ *
+ * However, unlike ssl3_HandleHandshake(), it is not used to carry
+ * the state of reassembly (i.e., whether one is in progress). That
+ * is carried in recvdHighWater and recvdFragments.
+ */
+#define OFFSET_BYTE(o) (o/8)
+#define OFFSET_MASK(o) (1 << (o%8))
+
+SECStatus
+dtls_HandleHandshake(sslSocket *ss, sslBuffer *origBuf)
+{
+    /* TODO(ekr@rtfm.com): OK for now.
+     * This doesn't work properly with asynchronous certificate validation.
+     * because that returns a WOULDBLOCK error. The current DTLS
+     * applications do not need asynchronous validation, but in the
+     * future we will need to add this.
+     */
+    sslBuffer buf = *origBuf;
+    SECStatus rv = SECSuccess;
+
+    PORT_Assert(ss->opt.noLocks || ssl_HaveRecvBufLock(ss));
+    PORT_Assert(ss->opt.noLocks || ssl_HaveSSL3HandshakeLock(ss));
+
+    while (buf.len > 0) {
+        PRUint8 type;
+        PRUint32 message_length;
+        PRUint16 message_seq;
+        PRUint32 fragment_offset;
+        PRUint32 fragment_length;
+        PRUint32 offset;
+
+        if (buf.len < 12) {
+            PORT_SetError(SSL_ERROR_RX_MALFORMED_HANDSHAKE);
+            rv = SECFailure;
+            break;
+        }
+
+        /* Parse the header */
+        type = buf.buf[0];
+        message_length = (buf.buf[1] << 16) | (buf.buf[2] << 8) | buf.buf[3];
+        message_seq = (buf.buf[4] << 8) | buf.buf[5];
+        fragment_offset = (buf.buf[6] << 16) | (buf.buf[7] << 8) | buf.buf[8];
+        fragment_length = (buf.buf[9] << 16) | (buf.buf[10] << 8) | buf.buf[11];
+        
+#define MAX_HANDSHAKE_MSG_LEN 0x1ffff   /* 128k - 1 */
+        if (message_length > MAX_HANDSHAKE_MSG_LEN) {
+            (void)ssl3_DecodeError(ss);
+            PORT_SetError(SSL_ERROR_RX_RECORD_TOO_LONG);
+            return SECFailure;
+        }
+#undef MAX_HANDSHAKE_MSG_LEN
+
+        buf.buf += 12;
+        buf.len -= 12;
+
+        /* This fragment must be complete */
+        if (buf.len < fragment_length) {
+            PORT_SetError(SSL_ERROR_RX_MALFORMED_HANDSHAKE);
+            rv = SECFailure;
+            break;
+        }
+
+        /* Sanity check the packet contents */
+        if ((fragment_length + fragment_offset) > message_length) {
+            PORT_SetError(SSL_ERROR_RX_MALFORMED_HANDSHAKE);
+            rv = SECFailure;
+            break;
+        }
+
+        /* There are three ways we could not be ready for this packet.
+         *
+         * 1. It's a partial next message.
+         * 2. It's a partial or complete message beyond the next
+         * 3. It's a message we've already seen
+         *
+         * If it's the complete next message we accept it right away.
+         * This is the common case for short messages
+         */
+        if ((message_seq == ss->ssl3.hs.recvMessageSeq)
+            && (fragment_offset == 0)
+            && (fragment_length == message_length)) {
+            /* Complete next message. Process immediately */
+            ss->ssl3.hs.msg_type = (SSL3HandshakeType)type;
+            ss->ssl3.hs.msg_len = message_length;
+
+            /* At this point we are advancing our state machine, so
+             * we can free our last flight of messages */
+            dtls_FreeHandshakeMessages(ss->ssl3.hs.lastMessageFlight);
+            ss->ssl3.hs.recvdHighWater = -1;
+            dtls_CancelTimer(ss);
+
+            /* Reset the timer to the initial value if the retry counter
+             * is 0, per Sec. 4.2.4.1 */
+            if (ss->ssl3.hs.rtRetries == 0) {
+                ss->ssl3.hs.rtTimeoutMs = INITIAL_DTLS_TIMEOUT_MS;
+            }
+
+            rv = ssl3_HandleHandshakeMessage(ss, buf.buf, ss->ssl3.hs.msg_len);
+            if (rv == SECFailure) {
+                /* Do not attempt to process rest of messages in this record */
+                break;
+            }
+        } else {
+            if (message_seq < ss->ssl3.hs.recvMessageSeq) {
+                /* Case 3: we do an immediate retransmit if we're
+                 * in a waiting state*/
+                if (ss->ssl3.hs.rtTimerCb == NULL) {
+                    /* Ignore */
+                } else if (ss->ssl3.hs.rtTimerCb ==
+                         dtls_RetransmitTimerExpiredCb) {
+                    SSL_TRC(30, ("%d: SSL3[%d]: Retransmit detected",
+                                 SSL_GETPID(), ss->fd));
+                    /* Check to see if we retransmitted recently. If so,
+                     * suppress the triggered retransmit. This avoids
+                     * retransmit wars after packet loss.
+                     * This is not in RFC 5346 but should be
+                     */
+                    if ((PR_IntervalNow() - ss->ssl3.hs.rtTimerStarted) >
+                        (ss->ssl3.hs.rtTimeoutMs / 4)) {
+                            SSL_TRC(30,
+                            ("%d: SSL3[%d]: Shortcutting retransmit timer",
+                            SSL_GETPID(), ss->fd));
+
+                            /* Cancel the timer and call the CB,
+                             * which re-arms the timer */
+                            dtls_CancelTimer(ss);
+                            dtls_RetransmitTimerExpiredCb(ss);
+                            rv = SECSuccess;
+                            break;
+                        } else {
+                            SSL_TRC(30,
+                            ("%d: SSL3[%d]: We just retransmitted. Ignoring.",
+                            SSL_GETPID(), ss->fd));
+                            rv = SECSuccess;
+                            break;
+                        }
+                } else if (ss->ssl3.hs.rtTimerCb == dtls_FinishedTimerCb) {
+                    /* Retransmit the messages and re-arm the timer
+                     * Note that we are not backing off the timer here.
+                     * The spec isn't clear and my reasoning is that this
+                     * may be a re-ordered packet rather than slowness,
+                     * so let's be aggressive. */
+                    dtls_CancelTimer(ss);
+                    rv = dtls_TransmitMessageFlight(ss);
+                    if (rv == SECSuccess) {
+                        rv = dtls_StartTimer(ss, dtls_FinishedTimerCb);
+                    }
+                    if (rv != SECSuccess)
+                        return rv;
+                    break;
+                }
+            } else if (message_seq > ss->ssl3.hs.recvMessageSeq) {
+                /* Case 2
+                 *
+                 * Ignore this message. This means we don't handle out of
+                 * order complete messages that well, but we're still
+                 * compliant and this probably does not happen often
+                 *
+                 * TODO(ekr@rtfm.com): OK for now. Maybe do something smarter
+                 * at some point?
+                 */
+            } else {
+                /* Case 1
+                 *
+                 * Buffer the fragment for reassembly
+                 */
+                /* Make room for the message */
+                if (ss->ssl3.hs.recvdHighWater == -1) {
+                    PRUint32 map_length = OFFSET_BYTE(message_length) + 1;
+
+                    rv = sslBuffer_Grow(&ss->ssl3.hs.msg_body, message_length);
+                    if (rv != SECSuccess)
+                        break;
+                    /* Make room for the fragment map */
+                    rv = sslBuffer_Grow(&ss->ssl3.hs.recvdFragments,
+                                        map_length);
+                    if (rv != SECSuccess)
+                        break;
+
+                    /* Reset the reassembly map */
+                    ss->ssl3.hs.recvdHighWater = 0;
+                    PORT_Memset(ss->ssl3.hs.recvdFragments.buf, 0,
+                                ss->ssl3.hs.recvdFragments.space);
+                    ss->ssl3.hs.msg_type = (SSL3HandshakeType)type;
+                    ss->ssl3.hs.msg_len = message_length;
+                }
+
+                /* If we have a message length mismatch, abandon the reassembly
+                 * in progress and hope that the next retransmit will give us
+                 * something sane
+                 */
+                if (message_length != ss->ssl3.hs.msg_len) {
+                    ss->ssl3.hs.recvdHighWater = -1;
+                    PORT_SetError(SSL_ERROR_RX_MALFORMED_HANDSHAKE);
+                    rv = SECFailure;
+                    break;
+                }
+
+                /* Now copy this fragment into the buffer */
+                PORT_Assert((fragment_offset + fragment_length) <=
+                            ss->ssl3.hs.msg_body.space);
+                PORT_Memcpy(ss->ssl3.hs.msg_body.buf + fragment_offset,
+                            buf.buf, fragment_length);
+
+                /* This logic is a bit tricky. We have two values for
+                 * reassembly state:
+                 *
+                 * - recvdHighWater contains the highest contiguous number of
+                 *   bytes received
+                 * - recvdFragments contains a bitmask of packets received
+                 *   above recvdHighWater
+                 *
+                 * This avoids having to fill in the bitmask in the common
+                 * case of adjacent fragments received in sequence
+                 */
+                if (fragment_offset <= ss->ssl3.hs.recvdHighWater) {
+                    /* Either this is the adjacent fragment or an overlapping
+                     * fragment */
+                    ss->ssl3.hs.recvdHighWater = fragment_offset +
+                                                 fragment_length;
+                } else {
+                    for (offset = fragment_offset;
+                         offset < fragment_offset + fragment_length;
+                         offset++) {
+                        ss->ssl3.hs.recvdFragments.buf[OFFSET_BYTE(offset)] |=
+                            OFFSET_MASK(offset);
+                    }
+                }
+
+                /* Now figure out the new high water mark if appropriate */
+                for (offset = ss->ssl3.hs.recvdHighWater;
+                     offset < ss->ssl3.hs.msg_len; offset++) {
+                    /* Note that this loop is not efficient, since it counts
+                     * bit by bit. If we have a lot of out-of-order packets,
+                     * we should optimize this */
+                    if (ss->ssl3.hs.recvdFragments.buf[OFFSET_BYTE(offset)] &
+                        OFFSET_MASK(offset)) {
+                        ss->ssl3.hs.recvdHighWater++;
+                    } else {
+                        break;
+                    }
+                }
+
+                /* If we have all the bytes, then we are good to go */
+                if (ss->ssl3.hs.recvdHighWater == ss->ssl3.hs.msg_len) {
+                    ss->ssl3.hs.recvdHighWater = -1;
+
+                    rv = ssl3_HandleHandshakeMessage(ss,
+                                                     ss->ssl3.hs.msg_body.buf,
+                                                     ss->ssl3.hs.msg_len);
+                    if (rv == SECFailure)
+                        break; /* Skip rest of record */
+
+                    /* At this point we are advancing our state machine, so
+                     * we can free our last flight of messages */
+                    dtls_FreeHandshakeMessages(ss->ssl3.hs.lastMessageFlight);
+                    dtls_CancelTimer(ss);
+
+                    /* If there have been no retries this time, reset the
+                     * timer value to the default per Section 4.2.4.1 */
+                    if (ss->ssl3.hs.rtRetries == 0) {
+                        ss->ssl3.hs.rtTimeoutMs = INITIAL_DTLS_TIMEOUT_MS;
+                    }
+                }
+            }
+        }
+
+        buf.buf += fragment_length;
+        buf.len -= fragment_length;
+    }
+
+    origBuf->len = 0;   /* So ssl3_GatherAppDataRecord will keep looping. */
+
+    /* TODO(ekr@rtfm.com): OK for now. In future handle rv == SECWouldBlock
+     * safely in order to deal with asynchronous certificate verification */
+    return rv;
+}
+
+/* Enqueue a message (either handshake or CCS)
+ *
+ * Called from:
+ *              dtls_StageHandshakeMessage()
+ *              ssl3_SendChangeCipherSpecs()
+ */
+SECStatus dtls_QueueMessage(sslSocket *ss, SSL3ContentType type,
+    const SSL3Opaque *pIn, PRInt32 nIn)
+{
+    SECStatus rv = SECSuccess;
+    DTLSQueuedMessage *msg = NULL;
+
+    PORT_Assert(ss->opt.noLocks || ssl_HaveSSL3HandshakeLock(ss));
+    PORT_Assert(ss->opt.noLocks || ssl_HaveXmitBufLock(ss));
+
+    msg = dtls_AllocQueuedMessage(ss->ssl3.cwSpec->epoch, type, pIn, nIn);
+
+    if (!msg) {
+        PORT_SetError(SEC_ERROR_NO_MEMORY);
+        rv = SECFailure;
+    } else {
+        PR_APPEND_LINK(&msg->link, ss->ssl3.hs.lastMessageFlight);
+    }
+
+    return rv;
+}
+
+/* Add DTLS handshake message to the pending queue
+ * Empty the sendBuf buffer.
+ * This function returns SECSuccess or SECFailure, never SECWouldBlock.
+ * Always set sendBuf.len to 0, even when returning SECFailure.
+ *
+ * Called from:
+ *              ssl3_AppendHandshakeHeader()
+ *              dtls_FlushHandshake()
+ */
+SECStatus
+dtls_StageHandshakeMessage(sslSocket *ss)
+{
+    SECStatus rv = SECSuccess;
+
+    PORT_Assert(ss->opt.noLocks || ssl_HaveSSL3HandshakeLock(ss));
+    PORT_Assert(ss->opt.noLocks || ssl_HaveXmitBufLock(ss));
+
+    /* This function is sometimes called when no data is actually to
+     * be staged, so just return SECSuccess. */
+    if (!ss->sec.ci.sendBuf.buf || !ss->sec.ci.sendBuf.len)
+        return rv;
+
+    rv = dtls_QueueMessage(ss, content_handshake,
+                           ss->sec.ci.sendBuf.buf, ss->sec.ci.sendBuf.len);
+
+    /* Whether we succeeded or failed, toss the old handshake data. */
+    ss->sec.ci.sendBuf.len = 0;
+    return rv;
+}
+
+/* Enqueue the handshake message in sendBuf (if any) and then
+ * transmit the resulting flight of handshake messages.
+ *
+ * Called from:
+ *              ssl3_FlushHandshake()
+ */
+SECStatus
+dtls_FlushHandshakeMessages(sslSocket *ss, PRInt32 flags)
+{
+    SECStatus rv = SECSuccess;
+
+    PORT_Assert(ss->opt.noLocks || ssl_HaveSSL3HandshakeLock(ss));
+    PORT_Assert(ss->opt.noLocks || ssl_HaveXmitBufLock(ss));
+
+    rv = dtls_StageHandshakeMessage(ss);
+    if (rv != SECSuccess)
+        return rv;
+
+    if (!(flags & ssl_SEND_FLAG_FORCE_INTO_BUFFER)) {
+        rv = dtls_TransmitMessageFlight(ss);
+        if (rv != SECSuccess)
+            return rv;
+        
+        if (!(flags & ssl_SEND_FLAG_NO_RETRANSMIT)) {
+            ss->ssl3.hs.rtRetries = 0;
+            rv = dtls_StartTimer(ss, dtls_RetransmitTimerExpiredCb);
+        }
+    }
+
+    return rv;
+}
+
+/* The callback for when the retransmit timer expires
+ *
+ * Called from:
+ *              dtls_CheckTimer()
+ *              dtls_HandleHandshake()
+ */
+static void
+dtls_RetransmitTimerExpiredCb(sslSocket *ss)
+{
+    SECStatus rv = SECFailure;
+
+    ss->ssl3.hs.rtRetries++;
+
+    if (!(ss->ssl3.hs.rtRetries % 3)) {
+        /* If one of the messages was potentially greater than > MTU,
+         * then downgrade. Do this every time we have retransmitted a
+         * message twice, per RFC 6347 Sec. 4.1.1 */
+        dtls_SetMTU(ss, ss->ssl3.hs.maxMessageSent - 1);
+    }
+        
+    rv = dtls_TransmitMessageFlight(ss);
+    if (rv == SECSuccess) {
+
+        /* Re-arm the timer */
+        rv = dtls_RestartTimer(ss, PR_TRUE, dtls_RetransmitTimerExpiredCb);
+    }
+
+    if (rv == SECFailure) {
+        /* TODO(ekr@rtfm.com): OK for now. In future maybe signal the stack
+         * that we couldn't transmit. For now, let the read handle any real
+         * network errors */
+    }
+}
+
+/* Transmit a flight of handshake messages, stuffing them
+ * into as few records as seems reasonable
+ *
+ * Called from:
+ *             dtls_FlushHandshake()
+ *             dtls_RetransmitTimerExpiredCb()
+ */
+static SECStatus
+dtls_TransmitMessageFlight(sslSocket *ss)
+{
+    SECStatus rv = SECSuccess;
+    PRCList *msg_p;
+    PRUint16 room_left = ss->ssl3.mtu;
+    PRInt32 sent;
+
+    ssl_GetXmitBufLock(ss);
+    ssl_GetSpecReadLock(ss);
+
+    /* DTLS does not buffer its handshake messages in
+     * ss->pendingBuf, but rather in the lastMessageFlight
+     * structure. This is just a sanity check that
+     * some programming error hasn't inadvertantly
+     * stuffed something in ss->pendingBuf
+     */
+    PORT_Assert(!ss->pendingBuf.len);
+    for (msg_p = PR_LIST_HEAD(ss->ssl3.hs.lastMessageFlight);
+         msg_p != ss->ssl3.hs.lastMessageFlight;
+         msg_p = PR_NEXT_LINK(msg_p)) {
+        DTLSQueuedMessage *msg = (DTLSQueuedMessage *)msg_p;
+
+        /* The logic here is:
+         *
+         * 1. If this is a message that will not fit into the remaining
+         *    space, then flush.
+         * 2. If the message will now fit into the remaining space,
+         *    encrypt, buffer, and loop.
+         * 3. If the message will not fit, then fragment.
+         *
+         * At the end of the function, flush.
+         */
+        if ((msg->len + SSL3_BUFFER_FUDGE) > room_left) {
+            /* The message will not fit into the remaining space, so flush */
+            rv = dtls_SendSavedWriteData(ss);
+            if (rv != SECSuccess)
+                break;
+
+            room_left = ss->ssl3.mtu;
+        }
+
+        if ((msg->len + SSL3_BUFFER_FUDGE) <= room_left) {
+            /* The message will fit, so encrypt and then continue with the
+             * next packet */
+            sent = ssl3_SendRecord(ss, msg->epoch, msg->type,
+                                   msg->data, msg->len,
+                                   ssl_SEND_FLAG_FORCE_INTO_BUFFER |
+                                   ssl_SEND_FLAG_USE_EPOCH);
+            if (sent != msg->len) {
+                rv = SECFailure;
+                if (sent != -1) {
+                    PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
+                }
+                break;
+            }
+
+            room_left = ss->ssl3.mtu - ss->pendingBuf.len;
+        } else {
+            /* The message will not fit, so fragment.
+             *
+             * TODO(ekr@rtfm.com): OK for now. Arrange to coalesce the
+             * last fragment of this message with the next message
+             * if possible. That would be more efficient.
+             */
+            PRUint32 fragment_offset = 0;
+            unsigned char fragment[DTLS_MAX_MTU]; /* >= than largest
+                                                   * plausible MTU */
+
+            /* Assert that we have already flushed */
+            PORT_Assert(room_left == ss->ssl3.mtu);
+
+            /* Case 3: We now need to fragment this message
+             * DTLS only supports fragmenting handshaking messages */
+            PORT_Assert(msg->type == content_handshake);
+
+            /* The headers consume 12 bytes so the smalles possible
+             *  message (i.e., an empty one) is 12 bytes
+             */
+            PORT_Assert(msg->len >= 12);
+
+            while ((fragment_offset + 12) < msg->len) {
+                PRUint32 fragment_len;
+                const unsigned char *content = msg->data + 12;
+                PRUint32 content_len = msg->len - 12;
+
+                /* The reason we use 8 here is that that's the length of
+                 * the new DTLS data that we add to the header */
+                fragment_len = PR_MIN(room_left - (SSL3_BUFFER_FUDGE + 8),
+                                      content_len - fragment_offset);
+                PORT_Assert(fragment_len < DTLS_MAX_MTU - 12);
+                /* Make totally sure that we are within the buffer.
+                 * Note that the only way that fragment len could get
+                 * adjusted here is if
+                 *
+                 * (a) we are in release mode so the PORT_Assert is compiled out
+                 * (b) either the MTU table is inconsistent with DTLS_MAX_MTU
+                 * or ss->ssl3.mtu has become corrupt.
+                 */
+                fragment_len = PR_MIN(fragment_len, DTLS_MAX_MTU - 12);
+
+                /* Construct an appropriate-sized fragment */
+                /* Type, length, sequence */
+                PORT_Memcpy(fragment, msg->data, 6);
+
+                /* Offset */
+                fragment[6] = (fragment_offset >> 16) & 0xff;
+                fragment[7] = (fragment_offset >> 8) & 0xff;
+                fragment[8] = (fragment_offset) & 0xff;
+
+                /* Fragment length */
+                fragment[9] = (fragment_len >> 16) & 0xff;
+                fragment[10] = (fragment_len >> 8) & 0xff;
+                fragment[11] = (fragment_len) & 0xff;
+
+                PORT_Memcpy(fragment + 12, content + fragment_offset,
+                            fragment_len);
+
+                /*
+                 *  Send the record. We do this in two stages
+                 * 1. Encrypt
+                 */
+                sent = ssl3_SendRecord(ss, msg->epoch, msg->type,
+                                       fragment, fragment_len + 12,
+                                       ssl_SEND_FLAG_FORCE_INTO_BUFFER |
+                                       ssl_SEND_FLAG_USE_EPOCH);
+                if (sent != (fragment_len + 12)) {
+                    rv = SECFailure;
+                    if (sent != -1) {
+                        PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
+                    }
+                    break;
+                }
+                
+                /* 2. Flush */
+                rv = dtls_SendSavedWriteData(ss);
+                if (rv != SECSuccess)
+                    break;
+
+                fragment_offset += fragment_len;
+            }
+        }
+    }
+
+    /* Finally, we need to flush */
+    if (rv == SECSuccess)
+        rv = dtls_SendSavedWriteData(ss);
+
+    /* Give up the locks */
+    ssl_ReleaseSpecReadLock(ss);
+    ssl_ReleaseXmitBufLock(ss);
+
+    return rv;
+}
+
+/* Flush the data in the pendingBuf and update the max message sent
+ * so we can adjust the MTU estimate if we need to.
+ * Wrapper for ssl_SendSavedWriteData.
+ *
+ * Called from dtls_TransmitMessageFlight()
+ */
+static
+SECStatus dtls_SendSavedWriteData(sslSocket *ss)
+{
+    PRInt32 sent;
+
+    sent = ssl_SendSavedWriteData(ss);
+    if (sent < 0)
+        return SECFailure;
+
+    /* We should always have complete writes b/c datagram sockets
+     * don't really block */
+    if (ss->pendingBuf.len > 0) {
+        ssl_MapLowLevelError(SSL_ERROR_SOCKET_WRITE_FAILURE);
+        return SECFailure;
+    }
+
+    /* Update the largest message sent so we can adjust the MTU
+     * estimate if necessary */
+    if (sent > ss->ssl3.hs.maxMessageSent)
+        ss->ssl3.hs.maxMessageSent = sent;
+
+    return SECSuccess;
+}
+
+/* Compress, MAC, encrypt a DTLS record. Allows specification of
+ * the epoch using epoch value. If use_epoch is PR_TRUE then
+ * we use the provided epoch. If use_epoch is PR_FALSE then
+ * whatever the current value is in effect is used.
+ *
+ * Called from ssl3_SendRecord()
+ */
+SECStatus
+dtls_CompressMACEncryptRecord(sslSocket *        ss,
+                              DTLSEpoch          epoch,
+                              PRBool             use_epoch,
+                              SSL3ContentType    type,
+                              const SSL3Opaque * pIn,
+                              PRUint32           contentLen,
+                              sslBuffer        * wrBuf)
+{
+    SECStatus rv = SECFailure;
+    ssl3CipherSpec *          cwSpec;
+
+    ssl_GetSpecReadLock(ss);    /********************************/
+
+    /* The reason for this switch-hitting code is that we might have
+     * a flight of records spanning an epoch boundary, e.g.,
+     *
+     * ClientKeyExchange (epoch = 0)
+     * ChangeCipherSpec (epoch = 0)
+     * Finished (epoch = 1)
+     *
+     * Thus, each record needs a different cipher spec. The information
+     * about which epoch to use is carried with the record.
+     */
+    if (use_epoch) {
+        if (ss->ssl3.cwSpec->epoch == epoch)
+            cwSpec = ss->ssl3.cwSpec;
+        else if (ss->ssl3.pwSpec->epoch == epoch)
+            cwSpec = ss->ssl3.pwSpec;
+        else
+            cwSpec = NULL;
+    } else {
+        cwSpec = ss->ssl3.cwSpec;
+    }
+
+    if (cwSpec) {
+        rv = ssl3_CompressMACEncryptRecord(cwSpec, ss->sec.isServer, PR_TRUE,
+                                           type, pIn, contentLen, wrBuf);
+    } else {
+        PR_NOT_REACHED("Couldn't find a cipher spec matching epoch");
+        PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
+    }
+    ssl_ReleaseSpecReadLock(ss); /************************************/
+
+    return rv;
+}
+
+/* Start a timer
+ *
+ * Called from:
+ *             dtls_HandleHandshake()
+ *             dtls_FlushHAndshake()
+ *             dtls_RestartTimer()
+ */
+SECStatus
+dtls_StartTimer(sslSocket *ss, DTLSTimerCb cb)
+{
+    PORT_Assert(ss->ssl3.hs.rtTimerCb == NULL);
+
+    ss->ssl3.hs.rtTimerStarted = PR_IntervalNow();
+    ss->ssl3.hs.rtTimerCb = cb;
+
+    return SECSuccess;
+}
+
+/* Restart a timer with optional backoff
+ *
+ * Called from dtls_RetransmitTimerExpiredCb()
+ */
+SECStatus
+dtls_RestartTimer(sslSocket *ss, PRBool backoff, DTLSTimerCb cb)
+{
+    if (backoff) {
+        ss->ssl3.hs.rtTimeoutMs *= 2;
+        if (ss->ssl3.hs.rtTimeoutMs > MAX_DTLS_TIMEOUT_MS)
+            ss->ssl3.hs.rtTimeoutMs = MAX_DTLS_TIMEOUT_MS;
+    }
+
+    return dtls_StartTimer(ss, cb);
+}
+
+/* Cancel a pending timer
+ *
+ * Called from:
+ *              dtls_HandleHandshake()
+ *              dtls_CheckTimer()
+ */
+void
+dtls_CancelTimer(sslSocket *ss)
+{
+    PORT_Assert(ss->opt.noLocks || ssl_HaveRecvBufLock(ss));
+
+    ss->ssl3.hs.rtTimerCb = NULL;
+}
+
+/* Check the pending timer and fire the callback if it expired
+ *
+ * Called from ssl3_GatherCompleteHandshake()
+ */
+void
+dtls_CheckTimer(sslSocket *ss)
+{
+    if (!ss->ssl3.hs.rtTimerCb)
+        return;
+
+    if ((PR_IntervalNow() - ss->ssl3.hs.rtTimerStarted) >
+        PR_MillisecondsToInterval(ss->ssl3.hs.rtTimeoutMs)) {
+        /* Timer has expired */
+        DTLSTimerCb cb = ss->ssl3.hs.rtTimerCb;
+        
+        /* Cancel the timer so that we can call the CB safely */
+        dtls_CancelTimer(ss);
+
+        /* Now call the CB */
+        cb(ss);
+    }
+}
+
+/* The callback to fire when the holddown timer for the Finished
+ * message expires and we can delete it
+ *
+ * Called from dtls_CheckTimer()
+ */
+void
+dtls_FinishedTimerCb(sslSocket *ss)
+{
+    ssl3_DestroyCipherSpec(ss->ssl3.pwSpec, PR_FALSE);
+}
+
+/* Cancel the Finished hold-down timer and destroy the
+ * pending cipher spec. Note that this means that
+ * successive rehandshakes will fail if the Finished is
+ * lost.
+ *
+ * TODO(ekr@rtfm.com): OK for now. Figure out how to
+ * handle the combination of Finished lost and rehandshake
+ */
+void
+dtls_RehandshakeCleanup(sslSocket *ss)
+{
+    dtls_CancelTimer(ss);
+    ssl3_DestroyCipherSpec(ss->ssl3.pwSpec, PR_FALSE);
+}
+
+/* Set the MTU to the next step less than or equal to the
+ * advertised value. Also used to downgrade the MTU by
+ * doing dtls_SetMTU(ss, biggest packet set).
+ *
+ * Passing 0 means set this to the largest MTU known
+ * (effectively resetting the PMTU backoff value).
+ *
+ * Called by:
+ *            ssl3_InitState()
+ *            dtls_RetransmitTimerExpiredCb()
+ */
+void
+dtls_SetMTU(sslSocket *ss, PRUint16 advertised)
+{
+    int i;
+
+    if (advertised == 0) {
+        ss->ssl3.mtu = COMMON_MTU_VALUES[0];
+        SSL_TRC(30, ("Resetting MTU to %d", ss->ssl3.mtu));
+        return;
+    }
+        
+    for (i = 0; i < PR_ARRAY_SIZE(COMMON_MTU_VALUES); i++) {
+        if (COMMON_MTU_VALUES[i] <= advertised) {
+            ss->ssl3.mtu = COMMON_MTU_VALUES[i];
+            SSL_TRC(30, ("Resetting MTU to %d", ss->ssl3.mtu));
+            return;
+        }
+    }
+
+    /* Fallback */
+    ss->ssl3.mtu = COMMON_MTU_VALUES[PR_ARRAY_SIZE(COMMON_MTU_VALUES)-1];
+    SSL_TRC(30, ("Resetting MTU to %d", ss->ssl3.mtu));
+}
+
+/* Called from ssl3_HandleHandshakeMessage() when it has deciphered a
+ * DTLS hello_verify_request
+ * Caller must hold Handshake and RecvBuf locks.
+ */
+SECStatus
+dtls_HandleHelloVerifyRequest(sslSocket *ss, SSL3Opaque *b, PRUint32 length)
+{
+    int                 errCode = SSL_ERROR_RX_MALFORMED_HELLO_VERIFY_REQUEST;
+    SECStatus           rv;
+    PRInt32             temp;
+    SECItem             cookie = {siBuffer, NULL, 0};
+    SSL3AlertDescription desc   = illegal_parameter;
+
+    SSL_TRC(3, ("%d: SSL3[%d]: handle hello_verify_request handshake",
+        SSL_GETPID(), ss->fd));
+    PORT_Assert(ss->opt.noLocks || ssl_HaveRecvBufLock(ss));
+    PORT_Assert(ss->opt.noLocks || ssl_HaveSSL3HandshakeLock(ss));
+
+    if (ss->ssl3.hs.ws != wait_server_hello) {
+        errCode = SSL_ERROR_RX_UNEXPECTED_HELLO_VERIFY_REQUEST;
+        desc    = unexpected_message;
+        goto alert_loser;
+    }
+
+    /* The version */
+    temp = ssl3_ConsumeHandshakeNumber(ss, 2, &b, &length);
+    if (temp < 0) {
+        goto loser;     /* alert has been sent */
+    }
+
+    if (temp != SSL_LIBRARY_VERSION_DTLS_1_0_WIRE) {
+        /* Note: this will need adjustment for DTLS 1.2 per Section 4.2.1 */
+        goto alert_loser;
+    }
+
+    /* The cookie */
+    rv = ssl3_ConsumeHandshakeVariable(ss, &cookie, 1, &b, &length);
+    if (rv != SECSuccess) {
+        goto loser;     /* alert has been sent */
+    }
+    if (cookie.len > DTLS_COOKIE_BYTES) {
+        desc = decode_error;
+        goto alert_loser;       /* malformed. */
+    }
+
+    PORT_Memcpy(ss->ssl3.hs.cookie, cookie.data, cookie.len);
+    ss->ssl3.hs.cookieLen = cookie.len;
+
+
+    ssl_GetXmitBufLock(ss);             /*******************************/
+
+    /* Now re-send the client hello */
+    rv = ssl3_SendClientHello(ss, PR_TRUE);
+
+    ssl_ReleaseXmitBufLock(ss);         /*******************************/
+
+    if (rv == SECSuccess)
+        return rv;
+
+alert_loser:
+    (void)SSL3_SendAlert(ss, alert_fatal, desc);
+
+loser:
+    errCode = ssl_MapLowLevelError(errCode);
+    return SECFailure;
+}
+
+/* Initialize the DTLS anti-replay window
+ *
+ * Called from:
+ *              ssl3_SetupPendingCipherSpec()
+ *              ssl3_InitCipherSpec()
+ */
+void
+dtls_InitRecvdRecords(DTLSRecvdRecords *records)
+{
+    PORT_Memset(records->data, 0, sizeof(records->data));
+    records->left = 0;
+    records->right = DTLS_RECVD_RECORDS_WINDOW - 1;
+}
+
+/*
+ * Has this DTLS record been received? Return values are:
+ * -1 -- out of range to the left
+ *  0 -- not received yet
+ *  1 -- replay
+ *
+ *  Called from: dtls_HandleRecord()
+ */
+int
+dtls_RecordGetRecvd(DTLSRecvdRecords *records, PRUint64 seq)
+{
+    PRUint64 offset;
+
+    /* Out of range to the left */
+    if (seq < records->left) {
+        return -1;
+    }
+
+    /* Out of range to the right; since we advance the window on
+     * receipt, that means that this packet has not been received
+     * yet */
+    if (seq > records->right)
+        return 0;
+
+    offset = seq % DTLS_RECVD_RECORDS_WINDOW;
+
+    return !!(records->data[offset / 8] & (1 << (offset % 8)));
+}
+
+/* Update the DTLS anti-replay window
+ *
+ * Called from ssl3_HandleRecord()
+ */
+void
+dtls_RecordSetRecvd(DTLSRecvdRecords *records, PRUint64 seq)
+{
+    PRUint64 offset;
+
+    if (seq < records->left)
+        return;
+
+    if (seq > records->right) {
+        PRUint64 new_left;
+        PRUint64 new_right;
+        PRUint64 right;
+
+        /* Slide to the right; this is the tricky part
+         *
+         * 1. new_top is set to have room for seq, on the
+         *    next byte boundary by setting the right 8
+         *    bits of seq
+         * 2. new_left is set to compensate.
+         * 3. Zero all bits between top and new_top. Since
+         *    this is a ring, this zeroes everything as-yet
+         *    unseen. Because we always operate on byte
+         *    boundaries, we can zero one byte at a time
+         */
+        new_right = seq | 0x07;
+        new_left = (new_right - DTLS_RECVD_RECORDS_WINDOW) + 1;
+
+        for (right = records->right + 8; right <= new_right; right += 8) {
+            offset = right % DTLS_RECVD_RECORDS_WINDOW;
+            records->data[offset / 8] = 0;
+        }
+
+        records->right = new_right;
+        records->left = new_left;
+    }
+
+    offset = seq % DTLS_RECVD_RECORDS_WINDOW;
+
+    records->data[offset / 8] |= (1 << (offset % 8));
+}
+
+SECStatus
+DTLS_GetTimeout(PRFileDesc *socket, PRIntervalTime *timeout)
+{
+    sslSocket * ss = NULL;
+    PRIntervalTime elapsed;
+    PRIntervalTime desired;
+
+    ss = ssl_FindSocket(socket);
+
+    if (!ss)
+        return SECFailure;
+
+    if (!IS_DTLS(ss))
+        return SECFailure;
+
+    if (!ss->ssl3.hs.rtTimerCb)
+        return SECFailure;
+
+    elapsed = PR_IntervalNow() - ss->ssl3.hs.rtTimerStarted;
+    desired = PR_MillisecondsToInterval(ss->ssl3.hs.rtTimeoutMs);
+    if (elapsed > desired) {
+        /* Timer expired */
+        *timeout = PR_INTERVAL_NO_WAIT;
+    } else {
+        *timeout = desired - elapsed;
+    }
+
+    return SECSuccess;
+}