Implement write pipeline support in libssl

Use the new pipeline cipher capability to encrypt multiple records being
written out all in one go. Two new SSL/SSL_CTX parameters can be used to
control how this works: max_pipelines and split_send_fragment.

max_pipelines defines the maximum number of pipelines that can ever be used
in one go for a single connection. It must always be less than or equal to
SSL_MAX_PIPELINES (currently defined to be 32). By default only one
pipeline will be used (i.e. normal non-parallel operation).

split_send_fragment defines how data is split up into pipelines. The number
of pipelines used will be determined by the amount of data provided to the
SSL_write call divided by split_send_fragment. For example if
split_send_fragment is set to 2000 and max_pipelines is 4 then:
SSL_write called with 0-2000 bytes == 1 pipeline used
SSL_write called with 2001-4000 bytes == 2 pipelines used
SSL_write called with 4001-6000 bytes == 3 pipelines used
SSL_write_called with 6001+ bytes == 4 pipelines used

split_send_fragment must always be less than or equal to max_send_fragment.
By default it is set to be equal to max_send_fragment. This will mean that
the same number of records will always be created as would have been
created in the non-parallel case, although the data will be apportioned
differently. In the parallel case data will be spread equally between the
pipelines.

Reviewed-by: Tim Hudson <tjh@openssl.org>

6 files changed, 401 insertions, 235 deletions

diff --git a/ssl/record/rec_layer_d1.c b/ssl/record/rec_layer_d1.c
index a1b0e9e7db..cd033e008f 100644
--- a/ssl/record/rec_layer_d1.c
+++ b/ssl/record/rec_layer_d1.c
@@ -1039,7 +1039,7 @@ int do_dtls1_write(SSL *s, int type, const unsigned char *buf,
     SSL3_BUFFER *wb;
     SSL_SESSION *sess;
 
-    wb = &s->rlayer.wbuf;
+    wb = &s->rlayer.wbuf[0];
 
     /*
      * first check if there is a SSL3_BUFFER still being written out.  This
@@ -1128,7 +1128,7 @@ int do_dtls1_write(SSL *s, int type, const unsigned char *buf,
 
     /* first we compress */
     if (s->compress != NULL) {
-        if (!ssl3_do_compress(s)) {
+        if (!ssl3_do_compress(s, wr)) {
             SSLerr(SSL_F_DO_DTLS1_WRITE, SSL_R_COMPRESSION_FAILURE);
             goto err;
         }
@@ -1145,7 +1145,7 @@ int do_dtls1_write(SSL *s, int type, const unsigned char *buf,
      */
 
     if (mac_size != 0) {
-        if (s->method->ssl3_enc->mac(s,
+        if (s->method->ssl3_enc->mac(s, wr,
                 &(p[SSL3_RECORD_get_length(wr) + eivlen]), 1) < 0)
             goto err;
         SSL3_RECORD_add_length(wr, mac_size);
@@ -1158,7 +1158,7 @@ int do_dtls1_write(SSL *s, int type, const unsigned char *buf,
     if (eivlen)
         SSL3_RECORD_add_length(wr, eivlen);
 
-    if (s->method->ssl3_enc->enc(s, 1) < 1)
+    if (s->method->ssl3_enc->enc(s, wr, 1, 1) < 1)
         goto err;
 
     /* record length after mac and block padding */
diff --git a/ssl/record/rec_layer_s3.c b/ssl/record/rec_layer_s3.c
index 6a4f92f9ba..3a232e5807 100644
--- a/ssl/record/rec_layer_s3.c
+++ b/ssl/record/rec_layer_s3.c
@@ -141,6 +141,8 @@ void RECORD_LAYER_init(RECORD_LAYER *rl, SSL *s)
 
 void RECORD_LAYER_clear(RECORD_LAYER *rl)
 {
+    unsigned int pipes;
+
     rl->rstate = SSL_ST_READ_HEADER;
 
     /* Do I need to clear read_ahead? As far as I can tell read_ahead did not
@@ -161,7 +163,9 @@ void RECORD_LAYER_clear(RECORD_LAYER *rl)
     rl->wpend_buf = NULL;
 
     SSL3_BUFFER_clear(&rl->rbuf);
-    SSL3_BUFFER_clear(&rl->wbuf);
+    for(pipes = 0; pipes < rl->numwpipes; pipes++)
+        SSL3_BUFFER_clear(&rl->wbuf[pipes]);
+    rl->numwpipes = 0;
     SSL3_RECORD_clear(&rl->rrec);
     SSL3_RECORD_clear(&rl->wrec);
 
@@ -176,7 +180,7 @@ void RECORD_LAYER_release(RECORD_LAYER *rl)
 {
     if (SSL3_BUFFER_is_initialised(&rl->rbuf))
         ssl3_release_read_buffer(rl->s);
-    if (SSL3_BUFFER_is_initialised(&rl->wbuf))
+    if (rl->numwpipes > 0)
         ssl3_release_write_buffer(rl->s);
     SSL3_RECORD_release(&rl->rrec);
 }
@@ -188,7 +192,8 @@ int RECORD_LAYER_read_pending(RECORD_LAYER *rl)
 
 int RECORD_LAYER_write_pending(RECORD_LAYER *rl)
 {
-    return SSL3_BUFFER_get_left(&rl->wbuf) != 0;
+    return (rl->numwpipes > 0)
+            && SSL3_BUFFER_get_left(&rl->wbuf[rl->numwpipes-1]) != 0;
 }
 
 int RECORD_LAYER_set_data(RECORD_LAYER *rl, const unsigned char *buf, int len)
@@ -433,10 +438,10 @@ int ssl3_write_bytes(SSL *s, int type, const void *buf_, int len)
     int tot;
     unsigned int n, nw;
 #if !defined(OPENSSL_NO_MULTIBLOCK) && EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK
-    unsigned int max_send_fragment;
+    unsigned int max_send_fragment, split_send_fragment, maxpipes;
     unsigned int u_len = (unsigned int)len;
 #endif
-    SSL3_BUFFER *wb = &s->rlayer.wbuf;
+    SSL3_BUFFER *wb = &s->rlayer.wbuf[0];
     int i;
 
     if (len < 0) {
@@ -622,13 +627,70 @@ int ssl3_write_bytes(SSL *s, int type, const void *buf_, int len)
     }
 
     n = (len - tot);
+
+    split_send_fragment = s->split_send_fragment;
+    /*
+     * If max_pipelines is 0 then this means "undefined" and we default to
+     * 1 pipeline. Similaraly if the cipher does not support pipelined
+     * processing then we also only use 1 pipeline, or if we're not using
+     * explicit IVs
+     */
+    maxpipes = s->max_pipelines;
+    if (maxpipes > SSL_MAX_PIPELINES) {
+        /*
+         * We should have prevented this when we set max_pipelines so we
+         * shouldn't get here
+        */
+        SSLerr(SSL_F_SSL3_WRITE_BYTES, ERR_R_INTERNAL_ERROR);
+        return -1;
+    }
+    if (maxpipes == 0
+            || s->enc_write_ctx == NULL
+            || !(EVP_CIPHER_flags(EVP_CIPHER_CTX_cipher(s->enc_write_ctx))
+                 & EVP_CIPH_FLAG_PIPELINE)
+            || !SSL_USE_EXPLICIT_IV(s))
+        maxpipes = 1;
+    if (s->max_send_fragment == 0 || split_send_fragment > s->max_send_fragment
+            || split_send_fragment == 0) {
+        /*
+         * We should have prevented this when we set the split and max send
+         * fragments so we shouldn't get here
+        */
+        SSLerr(SSL_F_SSL3_WRITE_BYTES, ERR_R_INTERNAL_ERROR);
+        return -1;
+    }
+
     for (;;) {
-        if (n > s->max_send_fragment)
-            nw = s->max_send_fragment;
+        unsigned int pipelens[SSL_MAX_PIPELINES], tmppipelen, remain;
+        unsigned int numpipes, j;
+
+        if (n == 0)
+            numpipes = 1;
         else
-            nw = n;
+            numpipes = ((n - 1) / split_send_fragment) + 1;
+        if (numpipes > maxpipes)
+            numpipes = maxpipes;
+
+        if (n / numpipes >= s->max_send_fragment) {
+            /*
+             * We have enough data to completely fill all available
+             * pipelines
+             */
+            for (j = 0; j < numpipes; j++) {
+                pipelens[j] = s->max_send_fragment;
+            }
+        } else {
+            /* We can partially fill all available pipelines */
+            tmppipelen = n / numpipes;
+            remain = n % numpipes;
+            for (j = 0; j < numpipes; j++) {
+                pipelens[j] = tmppipelen;
+                if (j < remain)
+                    pipelens[j]++;
+            }
+        }
 
-        i = do_ssl3_write(s, type, &(buf[tot]), nw, 0);
+        i = do_ssl3_write(s, type, &(buf[tot]), pipelens, numpipes, 0);
         if (i <= 0) {
             /* XXX should we ssl3_release_write_buffer if i<0? */
             s->rlayer.wnum = tot;
@@ -657,23 +719,28 @@ int ssl3_write_bytes(SSL *s, int type, const void *buf_, int len)
 }
 
 int do_ssl3_write(SSL *s, int type, const unsigned char *buf,
-                  unsigned int len, int create_empty_fragment)
+                  unsigned int *pipelens, unsigned int numpipes,
+                  int create_empty_fragment)
 {
-    unsigned char *p, *plen;
+    unsigned char *outbuf[SSL_MAX_PIPELINES], *plen[SSL_MAX_PIPELINES];
+    SSL3_RECORD wr[SSL_MAX_PIPELINES];
     int i, mac_size, clear = 0;
     int prefix_len = 0;
     int eivlen;
     size_t align = 0;
-    SSL3_RECORD *wr;
-    SSL3_BUFFER *wb = &s->rlayer.wbuf;
+    SSL3_BUFFER *wb;
     SSL_SESSION *sess;
+    unsigned int totlen = 0;
+    unsigned int j;
 
+    for (j = 0; j < numpipes; j++)
+        totlen += pipelens[j];
     /*
      * first check if there is a SSL3_BUFFER still being written out.  This
      * will happen with non blocking IO
      */
-    if (SSL3_BUFFER_get_left(wb) != 0)
-        return (ssl3_write_pending(s, type, buf, len));
+    if (RECORD_LAYER_write_pending(&s->rlayer))
+        return (ssl3_write_pending(s, type, buf, totlen));
 
     /* If we have an alert to send, lets send it */
     if (s->s3->alert_dispatch) {
@@ -683,14 +750,13 @@ int do_ssl3_write(SSL *s, int type, const unsigned char *buf,
         /* if it went, fall through and send more stuff */
     }
 
-    if (!SSL3_BUFFER_is_initialised(wb))
-        if (!ssl3_setup_write_buffer(s))
+    if (s->rlayer.numwpipes < numpipes)
+        if (!ssl3_setup_write_buffer(s, numpipes))
             return -1;
 
-    if (len == 0 && !create_empty_fragment)
+    if (totlen == 0 && !create_empty_fragment)
         return 0;
 
-    wr = &s->rlayer.wrec;
     sess = s->session;
 
     if ((sess == NULL) ||
@@ -720,7 +786,9 @@ int do_ssl3_write(SSL *s, int type, const unsigned char *buf,
              * 'prefix_len' bytes are sent out later together with the actual
              * payload)
              */
-            prefix_len = do_ssl3_write(s, type, buf, 0, 1);
+            unsigned int tmppipelen = 0;
+
+            prefix_len = do_ssl3_write(s, type, buf, &tmppipelen, 1, 1);
             if (prefix_len <= 0)
                 goto err;
 
@@ -737,6 +805,7 @@ int do_ssl3_write(SSL *s, int type, const unsigned char *buf,
     }
 
     if (create_empty_fragment) {
+        wb = &s->rlayer.wbuf[0];
 #if defined(SSL3_ALIGN_PAYLOAD) && SSL3_ALIGN_PAYLOAD!=0
         /*
          * extra fragment would be couple of cipher blocks, which would be
@@ -746,38 +815,24 @@ int do_ssl3_write(SSL *s, int type, const unsigned char *buf,
         align = (size_t)SSL3_BUFFER_get_buf(wb) + 2 * SSL3_RT_HEADER_LENGTH;
         align = (0-align) & (SSL3_ALIGN_PAYLOAD - 1);
 #endif
-        p = SSL3_BUFFER_get_buf(wb) + align;
+        outbuf[0] = SSL3_BUFFER_get_buf(wb) + align;
         SSL3_BUFFER_set_offset(wb, align);
     } else if (prefix_len) {
-        p = SSL3_BUFFER_get_buf(wb) + SSL3_BUFFER_get_offset(wb) + prefix_len;
+        wb = &s->rlayer.wbuf[0];
+        outbuf[0] = SSL3_BUFFER_get_buf(wb) + SSL3_BUFFER_get_offset(wb)
+                    + prefix_len;
     } else {
+        for (j=0; j < numpipes; j++) {
+            wb = &s->rlayer.wbuf[j];
 #if defined(SSL3_ALIGN_PAYLOAD) && SSL3_ALIGN_PAYLOAD!=0
-        align = (size_t)SSL3_BUFFER_get_buf(wb) + SSL3_RT_HEADER_LENGTH;
-        align = (0-align) & (SSL3_ALIGN_PAYLOAD - 1);
+            align = (size_t)SSL3_BUFFER_get_buf(wb) + SSL3_RT_HEADER_LENGTH;
+            align = (-align) & (SSL3_ALIGN_PAYLOAD - 1);
 #endif
-        p = SSL3_BUFFER_get_buf(wb) + align;
-        SSL3_BUFFER_set_offset(wb, align);
+            outbuf[j] = SSL3_BUFFER_get_buf(wb) + align;
+            SSL3_BUFFER_set_offset(wb, align);
+        }
     }
 
-    /* write the header */
-
-    *(p++) = type & 0xff;
-    SSL3_RECORD_set_type(wr, type);
-
-    *(p++) = (s->version >> 8);
-    /*
-     * Some servers hang if iniatial client hello is larger than 256 bytes
-     * and record version number > TLS 1.0
-     */
-    if (SSL_get_state(s) == TLS_ST_CW_CLNT_HELLO
-        && !s->renegotiate && TLS1_get_version(s) > TLS1_VERSION)
-        *(p++) = 0x1;
-    else
-        *(p++) = s->version & 0xff;
-
-    /* field where we are to write out packet length */
-    plen = p;
-    p += 2;
     /* Explicit IV length, block ciphers appropriate version flag */
     if (s->enc_write_ctx && SSL_USE_EXPLICIT_IV(s)) {
         int mode = EVP_CIPHER_CTX_mode(s->enc_write_ctx);
@@ -796,94 +851,132 @@ int do_ssl3_write(SSL *s, int type, const unsigned char *buf,
     } else
         eivlen = 0;
 
-    /* lets setup the record stuff. */
-    SSL3_RECORD_set_data(wr, p + eivlen);
-    SSL3_RECORD_set_length(wr, (int)len);
-    SSL3_RECORD_set_input(wr, (unsigned char *)buf);
 
+    totlen = 0;
+    /* Clear our SSL3_RECORD structures */
+    memset(wr, 0, sizeof wr);
+    for (j=0; j < numpipes; j++) {
+        /* write the header */
+        *(outbuf[j]++) = type & 0xff;
+        SSL3_RECORD_set_type(&wr[j], type);
 
-    /*
-     * we now 'read' from wr->input, wr->length bytes into wr->data
-     */
+        *(outbuf[j]++) = (s->version >> 8);
+        /*
+         * Some servers hang if iniatial client hello is larger than 256 bytes
+         * and record version number > TLS 1.0
+         */
+        if (SSL_get_state(s) == TLS_ST_CW_CLNT_HELLO
+            && !s->renegotiate && TLS1_get_version(s) > TLS1_VERSION)
+            *(outbuf[j]++) = 0x1;
+        else
+            *(outbuf[j]++) = s->version & 0xff;
 
-    /* first we compress */
-    if (s->compress != NULL) {
-        if (!ssl3_do_compress(s)) {
-            SSLerr(SSL_F_DO_SSL3_WRITE, SSL_R_COMPRESSION_FAILURE);
-            goto err;
-        }
-    } else {
-        memcpy(wr->data, wr->input, wr->length);
-        SSL3_RECORD_reset_input(wr);
-    }
+        /* field where we are to write out packet length */
+        plen[j] = outbuf[j];
+        outbuf[j] += 2;
 
-    /*
-     * we should still have the output to wr->data and the input from
-     * wr->input.  Length should be wr->length. wr->data still points in the
-     * wb->buf
-     */
+        /* lets setup the record stuff. */
+        SSL3_RECORD_set_data(&wr[j], outbuf[j] + eivlen);
+        SSL3_RECORD_set_length(&wr[j], (int)pipelens[j]);
+        SSL3_RECORD_set_input(&wr[j], (unsigned char *)&buf[totlen]);
+        totlen += pipelens[j];
 
-    if (!SSL_USE_ETM(s) && mac_size != 0) {
-        if (s->method->ssl3_enc->mac(s, &(p[wr->length + eivlen]), 1) < 0)
-            goto err;
-        SSL3_RECORD_add_length(wr, mac_size);
-    }
+        /*
+         * we now 'read' from wr->input, wr->length bytes into wr->data
+         */
 
-    SSL3_RECORD_set_data(wr, p);
-    SSL3_RECORD_reset_input(wr);
+        /* first we compress */
+        if (s->compress != NULL) {
+            if (!ssl3_do_compress(s, &wr[j])) {
+                SSLerr(SSL_F_DO_SSL3_WRITE, SSL_R_COMPRESSION_FAILURE);
+                goto err;
+            }
+        } else {
+            memcpy(wr[j].data, wr[j].input, wr[j].length);
+            SSL3_RECORD_reset_input(&wr[j]);
+        }
 
-    if (eivlen) {
         /*
-         * if (RAND_pseudo_bytes(p, eivlen) <= 0) goto err;
+         * we should still have the output to wr->data and the input from
+         * wr->input.  Length should be wr->length. wr->data still points in the
+         * wb->buf
          */
-        SSL3_RECORD_add_length(wr, eivlen);
-    }
 
-    if (s->method->ssl3_enc->enc(s, 1) < 1)
-        goto err;
+        if (!SSL_USE_ETM(s) && mac_size != 0) {
+            if (s->method->ssl3_enc->mac(s, &wr[j],
+                    &(outbuf[j][wr[j].length + eivlen]), 1) < 0)
+                goto err;
+            SSL3_RECORD_add_length(&wr[j], mac_size);
+        }
 
-    if (SSL_USE_ETM(s) && mac_size != 0) {
-        if (s->method->ssl3_enc->mac(s, p + wr->length, 1) < 0)
-            goto err;
-        SSL3_RECORD_add_length(wr, mac_size);
+
+        SSL3_RECORD_set_data(&wr[j], outbuf[j]);
+        SSL3_RECORD_reset_input(&wr[j]);
+
+        if (eivlen) {
+            /*
+             * if (RAND_pseudo_bytes(p, eivlen) <= 0) goto err;
+             */
+            SSL3_RECORD_add_length(&wr[j], eivlen);
+        }
     }
 
-    /* record length after mac and block padding */
-    s2n(SSL3_RECORD_get_length(wr), plen);
+    if (s->method->ssl3_enc->enc(s, wr, numpipes, 1) < 1)
+        goto err;
 
-    if (s->msg_callback)
-        s->msg_callback(1, 0, SSL3_RT_HEADER, plen - 5, 5, s,
-                        s->msg_callback_arg);
+    for (j=0; j < numpipes; j++) {
+        if (SSL_USE_ETM(s) && mac_size != 0) {
+            if (s->method->ssl3_enc->mac(s, &wr[j],
+                                         outbuf[j] + wr[j].length, 1) < 0)
+                goto err;
+            SSL3_RECORD_add_length(&wr[j], mac_size);
+        }
 
-    /*
-     * we should now have wr->data pointing to the encrypted data, which is
-     * wr->length long
-     */
-    SSL3_RECORD_set_type(wr, type);  /* not needed but helps for debugging */
-    SSL3_RECORD_add_length(wr, SSL3_RT_HEADER_LENGTH);
+        /* record length after mac and block padding */
+        s2n(SSL3_RECORD_get_length(&wr[j]), plen[j]);
+
+        if (s->msg_callback)
+            s->msg_callback(1, 0, SSL3_RT_HEADER, plen[j] - 5, 5, s,
+                            s->msg_callback_arg);
 
-    if (create_empty_fragment) {
         /*
-         * we are in a recursive call; just return the length, don't write
-         * out anything here
+         * we should now have wr->data pointing to the encrypted data, which is
+         * wr->length long
          */
-        return SSL3_RECORD_get_length(wr);
+        SSL3_RECORD_set_type(&wr[j], type);  /* not needed but helps for debugging */
+        SSL3_RECORD_add_length(&wr[j], SSL3_RT_HEADER_LENGTH);
+
+        if (create_empty_fragment) {
+            /*
+             * we are in a recursive call; just return the length, don't write
+             * out anything here
+             */
+            if (j > 0) {
+                /* We should never be pipelining an empty fragment!! */
+                SSLerr(SSL_F_DO_SSL3_WRITE, ERR_R_INTERNAL_ERROR);
+                goto err;
+            }
+            return SSL3_RECORD_get_length(wr);
+        }
+
+        /* now let's set up wb */
+        SSL3_BUFFER_set_left(&s->rlayer.wbuf[j],
+            prefix_len + SSL3_RECORD_get_length(&wr[j]));
     }
 
-    /* now let's set up wb */
-    SSL3_BUFFER_set_left(wb, prefix_len + SSL3_RECORD_get_length(wr));
+
 
     /*
      * memorize arguments so that ssl3_write_pending can detect bad write
      * retries later
      */
-    s->rlayer.wpend_tot = len;
+    s->rlayer.wpend_tot = totlen;
     s->rlayer.wpend_buf = buf;
     s->rlayer.wpend_type = type;
-    s->rlayer.wpend_ret = len;
+    s->rlayer.wpend_ret = totlen;
 
     /* we now just need to write the buffer */
-    return ssl3_write_pending(s, type, buf, len);
+    return ssl3_write_pending(s, type, buf, totlen);
  err:
     return -1;
 }
@@ -893,7 +986,8 @@ int ssl3_write_pending(SSL *s, int type, const unsigned char *buf,
                        unsigned int len)
 {
     int i;
-    SSL3_BUFFER *wb = &s->rlayer.wbuf;
+    SSL3_BUFFER *wb = s->rlayer.wbuf;
+    unsigned int currbuf = 0;
 
 /* XXXX */
     if ((s->rlayer.wpend_tot > (int)len)
@@ -905,19 +999,28 @@ int ssl3_write_pending(SSL *s, int type, const unsigned char *buf,
     }
 
     for (;;) {
+        /* Loop until we find a buffer we haven't written out yet */
+        if (SSL3_BUFFER_get_left(&wb[currbuf]) == 0
+                && currbuf < s->rlayer.numwpipes - 1) {
+            currbuf++;
+            continue;
+        }
         clear_sys_error();
         if (s->wbio != NULL) {
             s->rwstate = SSL_WRITING;
             i = BIO_write(s->wbio,
-                (char *)&(SSL3_BUFFER_get_buf(wb)[SSL3_BUFFER_get_offset(wb)]),
-                (unsigned int)SSL3_BUFFER_get_left(wb));
+                (char *)&(SSL3_BUFFER_get_buf(&wb[currbuf])[
+                                SSL3_BUFFER_get_offset(&wb[currbuf])]),
+                (unsigned int)SSL3_BUFFER_get_left(&wb[currbuf]));
         } else {
             SSLerr(SSL_F_SSL3_WRITE_PENDING, SSL_R_BIO_NOT_SET);
             i = -1;
         }
-        if (i == SSL3_BUFFER_get_left(wb)) {
-            SSL3_BUFFER_set_left(wb, 0);
-            SSL3_BUFFER_add_offset(wb, i);
+        if (i == SSL3_BUFFER_get_left(&wb[currbuf])) {
+            SSL3_BUFFER_set_left(&wb[currbuf], 0);
+            SSL3_BUFFER_add_offset(&wb[currbuf], i);
+            if (currbuf + 1 < s->rlayer.numwpipes)
+                continue;
             s->rwstate = SSL_NOTHING;
             return (s->rlayer.wpend_ret);
         } else if (i <= 0) {
@@ -926,12 +1029,12 @@ int ssl3_write_pending(SSL *s, int type, const unsigned char *buf,
                  * For DTLS, just drop it. That's kind of the whole point in
                  * using a datagram service
                  */
-                SSL3_BUFFER_set_left(wb, 0);
+                SSL3_BUFFER_set_left(&wb[currbuf], 0);
             }
             return (i);
         }
-        SSL3_BUFFER_add_offset(wb, i);
-        SSL3_BUFFER_add_left(wb, -i);
+        SSL3_BUFFER_add_offset(&wb[currbuf], i);
+        SSL3_BUFFER_add_left(&wb[currbuf], -i);
     }
 }
 
diff --git a/ssl/record/record.h b/ssl/record/record.h
index a3b50dcfcf..6105dedd44 100644
--- a/ssl/record/record.h
+++ b/ssl/record/record.h
@@ -252,10 +252,12 @@ typedef struct record_layer_st {
     int read_ahead;
     /* where we are when reading */
     int rstate;
+
+    unsigned int numwpipes;
     /* read IO goes into here */
     SSL3_BUFFER rbuf;
     /* write IO goes into here */
-    SSL3_BUFFER wbuf;
+    SSL3_BUFFER wbuf[SSL_MAX_PIPELINES];
     /* each decoded record goes in here */
     SSL3_RECORD rrec;
     /* goes out from here */
@@ -326,16 +328,19 @@ unsigned int RECORD_LAYER_get_rrec_length(RECORD_LAYER *rl);
 __owur int ssl3_pending(const SSL *s);
 __owur int ssl3_write_bytes(SSL *s, int type, const void *buf, int len);
 __owur int do_ssl3_write(SSL *s, int type, const unsigned char *buf,
-                         unsigned int len, int create_empty_fragment);
+                         unsigned int *pipelens, unsigned int numpipes,
+                         int create_empty_fragment);
 __owur int ssl3_read_bytes(SSL *s, int type, int *recvd_type,
                            unsigned char *buf, int len, int peek);
 __owur int ssl3_setup_buffers(SSL *s);
-__owur int ssl3_enc(SSL *s, int send_data);
-__owur int n_ssl3_mac(SSL *ssl, unsigned char *md, int send_data);
+__owur int ssl3_enc(SSL *s, SSL3_RECORD *inrecs, unsigned int numpipes,
+                    int send);
+__owur int n_ssl3_mac(SSL *ssl, SSL3_RECORD *rec, unsigned char *md, int send);
 __owur int ssl3_write_pending(SSL *s, int type, const unsigned char *buf,
                        unsigned int len);
-__owur int tls1_enc(SSL *s, int snd);
-__owur int tls1_mac(SSL *ssl, unsigned char *md, int snd);
+__owur int tls1_enc(SSL *s, SSL3_RECORD *recs, unsigned int numpipes,
+                    int send);
+__owur int tls1_mac(SSL *ssl, SSL3_RECORD *rec, unsigned char *md, int send);
 int DTLS_RECORD_LAYER_new(RECORD_LAYER *rl);
 void DTLS_RECORD_LAYER_free(RECORD_LAYER *rl);
 void DTLS_RECORD_LAYER_clear(RECORD_LAYER *rl);
diff --git a/ssl/record/record_locl.h b/ssl/record/record_locl.h
index 333d3c947d..f44cda16e5 100644
--- a/ssl/record/record_locl.h
+++ b/ssl/record/record_locl.h
@@ -119,7 +119,7 @@
 /* Functions/macros provided by the RECORD_LAYER component */
 
 #define RECORD_LAYER_get_rbuf(rl)               (&(rl)->rbuf)
-#define RECORD_LAYER_get_wbuf(rl)               (&(rl)->wbuf)
+#define RECORD_LAYER_get_wbuf(rl)               ((rl)->wbuf)
 #define RECORD_LAYER_get_rrec(rl)               (&(rl)->rrec)
 #define RECORD_LAYER_get_wrec(rl)               (&(rl)->wrec)
 #define RECORD_LAYER_set_packet(rl, p)          ((rl)->packet = (p))
@@ -165,7 +165,7 @@ void SSL3_BUFFER_clear(SSL3_BUFFER *b);
 void SSL3_BUFFER_set_data(SSL3_BUFFER *b, const unsigned char *d, int n);
 void SSL3_BUFFER_release(SSL3_BUFFER *b);
 __owur int ssl3_setup_read_buffer(SSL *s);
-__owur int ssl3_setup_write_buffer(SSL *s);
+__owur int ssl3_setup_write_buffer(SSL *s, unsigned int numwpipes);
 int ssl3_release_read_buffer(SSL *s);
 int ssl3_release_write_buffer(SSL *s);
 
@@ -194,7 +194,7 @@ void SSL3_RECORD_release(SSL3_RECORD *r);
 int SSL3_RECORD_setup(SSL3_RECORD *r);
 void SSL3_RECORD_set_seq_num(SSL3_RECORD *r, const unsigned char *seq_num);
 int ssl3_get_record(SSL *s);
-__owur int ssl3_do_compress(SSL *ssl);
+__owur int ssl3_do_compress(SSL *ssl, SSL3_RECORD *wr);
 __owur int ssl3_do_uncompress(SSL *ssl);
 void ssl3_cbc_copy_mac(unsigned char *out,
                        const SSL3_RECORD *rec, unsigned md_size);
diff --git a/ssl/record/ssl3_buffer.c b/ssl/record/ssl3_buffer.c
index 3c03499f59..576533c31e 100644
--- a/ssl/record/ssl3_buffer.c
+++ b/ssl/record/ssl3_buffer.c
@@ -176,13 +176,15 @@ int ssl3_setup_read_buffer(SSL *s)
     return 0;
 }
 
-int ssl3_setup_write_buffer(SSL *s)
+int ssl3_setup_write_buffer(SSL *s, unsigned int numwpipes)
 {
     unsigned char *p;
     size_t len, align = 0, headerlen;
     SSL3_BUFFER *wb;
+    unsigned int currpipe;
+
+    s->rlayer.numwpipes = numwpipes;
 
-    wb = RECORD_LAYER_get_wbuf(&s->rlayer);
 
     if (SSL_IS_DTLS(s))
         headerlen = DTLS1_RT_HEADER_LENGTH + 1;
@@ -193,20 +195,25 @@ int ssl3_setup_write_buffer(SSL *s)
     align = (-SSL3_RT_HEADER_LENGTH) & (SSL3_ALIGN_PAYLOAD - 1);
 #endif
 
-    if (wb->buf == NULL) {
-        len = s->max_send_fragment
-            + SSL3_RT_SEND_MAX_ENCRYPTED_OVERHEAD + headerlen + align;
+    len = s->max_send_fragment
+        + SSL3_RT_SEND_MAX_ENCRYPTED_OVERHEAD + headerlen + align;
 #ifndef OPENSSL_NO_COMP
-        if (ssl_allow_compression(s))
-            len += SSL3_RT_MAX_COMPRESSED_OVERHEAD;
+    if (ssl_allow_compression(s))
+        len += SSL3_RT_MAX_COMPRESSED_OVERHEAD;
 #endif
-        if (!(s->options & SSL_OP_DONT_INSERT_EMPTY_FRAGMENTS))
-            len += headerlen + align + SSL3_RT_SEND_MAX_ENCRYPTED_OVERHEAD;
+    if (!(s->options & SSL_OP_DONT_INSERT_EMPTY_FRAGMENTS))
+        len += headerlen + align + SSL3_RT_SEND_MAX_ENCRYPTED_OVERHEAD;
 
-        if ((p = OPENSSL_malloc(len)) == NULL)
-            goto err;
-        wb->buf = p;
-        wb->len = len;
+    wb = RECORD_LAYER_get_wbuf(&s->rlayer);
+    for (currpipe = 0; currpipe < numwpipes; currpipe++) {
+        if (wb[currpipe].buf == NULL) {
+            if ((p = OPENSSL_malloc(len)) == NULL) {
+                s->rlayer.numwpipes = currpipe;
+                goto err;
+            }
+            wb[currpipe].buf = p;
+            wb[currpipe].len = len;
+        }
     }
 
     return 1;
@@ -220,7 +227,7 @@ int ssl3_setup_buffers(SSL *s)
 {
     if (!ssl3_setup_read_buffer(s))
         return 0;
-    if (!ssl3_setup_write_buffer(s))
+    if (!ssl3_setup_write_buffer(s, 1))
         return 0;
     return 1;
 }
@@ -228,11 +235,17 @@ int ssl3_setup_buffers(SSL *s)
 int ssl3_release_write_buffer(SSL *s)
 {
     SSL3_BUFFER *wb;
+    unsigned int pipes;
 
-    wb = RECORD_LAYER_get_wbuf(&s->rlayer);
+    pipes = s->rlayer.numwpipes;
+    while (pipes > 0) {
+        wb = &RECORD_LAYER_get_wbuf(&s->rlayer)[pipes - 1];
 
-    OPENSSL_free(wb->buf);
-    wb->buf = NULL;
+        OPENSSL_free(wb->buf);
+        wb->buf = NULL;
+        pipes--;
+    }
+    s->rlayer.numwpipes = 0;
     return 1;
 }
 
diff --git a/ssl/record/ssl3_record.c b/ssl/record/ssl3_record.c
index 7f89cc0a1c..ad19621b11 100644
--- a/ssl/record/ssl3_record.c
+++ b/ssl/record/ssl3_record.c
@@ -373,7 +373,7 @@ int ssl3_get_record(SSL *s)
         }
         rr->length -= mac_size;
         mac = rr->data + rr->length;
-        i = s->method->ssl3_enc->mac(s, md, 0 /* not send */ );
+        i = s->method->ssl3_enc->mac(s, rr, md, 0 /* not send */ );
         if (i < 0 || CRYPTO_memcmp(md, mac, (size_t)mac_size) != 0) {
             al = SSL_AD_BAD_RECORD_MAC;
             SSLerr(SSL_F_SSL3_GET_RECORD,
@@ -382,7 +382,7 @@ int ssl3_get_record(SSL *s)
         }
     }
 
-    enc_err = s->method->ssl3_enc->enc(s, 0);
+    enc_err = s->method->ssl3_enc->enc(s, rr, 1, 0);
     /*-
      * enc_err is:
      *    0: (in non-constant time) if the record is publically invalid.
@@ -449,7 +449,7 @@ int ssl3_get_record(SSL *s)
             mac = &rr->data[rr->length];
         }
 
-        i = s->method->ssl3_enc->mac(s, md, 0 /* not send */ );
+        i = s->method->ssl3_enc->mac(s, rr, md, 0 /* not send */ );
         if (i < 0 || mac == NULL
             || CRYPTO_memcmp(md, mac, (size_t)mac_size) != 0)
             enc_err = -1;
@@ -542,13 +542,11 @@ int ssl3_do_uncompress(SSL *ssl)
     return (1);
 }
 
-int ssl3_do_compress(SSL *ssl)
+int ssl3_do_compress(SSL *ssl, SSL3_RECORD *wr)
 {
 #ifndef OPENSSL_NO_COMP
     int i;
-    SSL3_RECORD *wr;
 
-    wr = RECORD_LAYER_get_wrec(&ssl->rlayer);
     i = COMP_compress_block(ssl->compress, wr->data,
                             SSL3_RT_MAX_COMPRESSED_LENGTH,
                             wr->input, (int)wr->length);
@@ -572,7 +570,7 @@ int ssl3_do_compress(SSL *ssl)
  *   -1: if the record's padding is invalid or, if sending, an internal error
  *       occurred.
  */
-int ssl3_enc(SSL *s, int send)
+int ssl3_enc(SSL *s, SSL3_RECORD *inrecs, unsigned int numpipes, int send)
 {
     SSL3_RECORD *rec;
     EVP_CIPHER_CTX *ds;
@@ -580,16 +578,15 @@ int ssl3_enc(SSL *s, int send)
     int bs, i, mac_size = 0;
     const EVP_CIPHER *enc;
 
+    rec = inrecs;
     if (send) {
         ds = s->enc_write_ctx;
-        rec = RECORD_LAYER_get_wrec(&s->rlayer);
         if (s->enc_write_ctx == NULL)
             enc = NULL;
         else
             enc = EVP_CIPHER_CTX_cipher(s->enc_write_ctx);
     } else {
         ds = s->enc_read_ctx;
-        rec = RECORD_LAYER_get_rrec(&s->rlayer);
         if (s->enc_read_ctx == NULL)
             enc = NULL;
         else
@@ -646,13 +643,14 @@ int ssl3_enc(SSL *s, int send)
  *   -1: if the record's padding/AEAD-authenticator is invalid or, if sending,
  *       an internal error occurred.
  */
-int tls1_enc(SSL *s, int send)
+int tls1_enc(SSL *s, SSL3_RECORD *recs, unsigned int numpipes, int send)
 {
-    SSL3_RECORD *rec;
     EVP_CIPHER_CTX *ds;
-    unsigned long l;
+    size_t reclen[SSL_MAX_PIPELINES];
+    unsigned char buf[SSL_MAX_PIPELINES][EVP_AEAD_TLS1_AAD_LEN];
     int bs, i, j, k, pad = 0, ret, mac_size = 0;
     const EVP_CIPHER *enc;
+    unsigned int ctr;
 
     if (send) {
         if (EVP_MD_CTX_md(s->write_hash)) {
@@ -660,7 +658,6 @@ int tls1_enc(SSL *s, int send)
             OPENSSL_assert(n >= 0);
         }
         ds = s->enc_write_ctx;
-        rec = RECORD_LAYER_get_wrec(&s->rlayer);
         if (s->enc_write_ctx == NULL)
             enc = NULL;
         else {
@@ -673,14 +670,19 @@ int tls1_enc(SSL *s, int send)
             else
                 ivlen = 0;
             if (ivlen > 1) {
-                if (rec->data != rec->input)
-                    /*
-                     * we can't write into the input stream: Can this ever
-                     * happen?? (steve)
-                     */
-                    fprintf(stderr, "tls1_enc(): rec->data != rec->input\n");
-                else if (RAND_bytes(rec->input, ivlen) <= 0)
-                    return -1;
+                for (ctr = 0; ctr < numpipes; ctr++) {
+                    if (recs[ctr].data != recs[ctr].input) {
+                        /*
+                         * we can't write into the input stream: Can this ever
+                         * happen?? (steve)
+                         */
+                        SSLerr(SSL_F_TLS1_ENC, ERR_R_INTERNAL_ERROR);
+                        return -1;
+                    } else if (RAND_bytes(recs[ctr].input, ivlen) <= 0) {
+                        SSLerr(SSL_F_TLS1_ENC, ERR_R_INTERNAL_ERROR);
+                        return -1;
+                    }
+                }
             }
         }
     } else {
@@ -689,7 +691,6 @@ int tls1_enc(SSL *s, int send)
             OPENSSL_assert(n >= 0);
         }
         ds = s->enc_read_ctx;
-        rec = RECORD_LAYER_get_rrec(&s->rlayer);
         if (s->enc_read_ctx == NULL)
             enc = NULL;
         else
@@ -697,97 +698,146 @@ int tls1_enc(SSL *s, int send)
     }
 
     if ((s->session == NULL) || (ds == NULL) || (enc == NULL)) {
-        memmove(rec->data, rec->input, rec->length);
-        rec->input = rec->data;
+        for (ctr = 0; ctr < numpipes; ctr++) {
+            memmove(recs[ctr].data, recs[ctr].input, recs[ctr].length);
+            recs[ctr].input = recs[ctr].data;
+        }
         ret = 1;
     } else {
-        l = rec->length;
-        bs = EVP_CIPHER_CTX_block_size(ds);
+        bs = EVP_CIPHER_block_size(EVP_CIPHER_CTX_cipher(ds));
+
+        if (numpipes > 1) {
+            if(!(EVP_CIPHER_flags(EVP_CIPHER_CTX_cipher(ds))
+                                  & EVP_CIPH_FLAG_PIPELINE)) {
+                /*
+                 * We shouldn't have been called with pipeline data if the
+                 * cipher doesn't support pipelining
+                 */
+                SSLerr(SSL_F_TLS1_ENC, SSL_R_PIPELINE_FAILURE);
+                return -1;
+            }
+        }
+        for (ctr = 0; ctr < numpipes; ctr++) {
+            reclen[ctr] = recs[ctr].length;
+
+            if (EVP_CIPHER_flags(EVP_CIPHER_CTX_cipher(ds))
+                    & EVP_CIPH_FLAG_AEAD_CIPHER) {
+                unsigned char *seq;
+
+