From 32aba97e7a7d9329bd0b4124da7c3beff2bb9b02 Mon Sep 17 00:00:00 2001
From: Willy Tarreau <w@1wt.eu>
Date: Mon, 8 Dec 2014 19:59:07 +0100
Subject: EXP: support splicing all the time

This seems to work fine, though it is still possible to disable it using
-dS and/or without setting option splice-response.
---
 src/proto_http.c | 87 ++++++++++++++++++++++++++++++++++++++++++++++++++++----
 src/session.c    |  2 +-
 2 files changed, 82 insertions(+), 7 deletions(-)

diff --git a/src/proto_http.c b/src/proto_http.c
index da47ae9..cb5a8e4 100644
--- a/src/proto_http.c
+++ b/src/proto_http.c
@@ -10,6 +10,7 @@
  *
  */
 
+#define _GNU_SOURCE
 #include <ctype.h>
 #include <errno.h>
 #include <fcntl.h>
@@ -22,6 +23,7 @@
 #include <sys/socket.h>
 #include <sys/stat.h>
 #include <sys/types.h>
+#include <sys/uio.h>
 
 #include <netinet/tcp.h>
 
@@ -55,6 +57,7 @@
 #include <proto/log.h>
 #include <proto/hdr_idx.h>
 #include <proto/pattern.h>
+#include <proto/pipe.h>
 #include <proto/proto_tcp.h>
 #include <proto/proto_http.h>
 #include <proto/proxy.h>
@@ -3186,6 +3189,7 @@ struct hterm_ctx {
 
 static struct chunk hterm_chunk;
 static struct buffer *hterm_buf;
+static struct pipe *hterm_master_pipe;
 
 static int httpterm_send_http_headers(struct stream_interface *si)
 {
@@ -3330,13 +3334,68 @@ static void httpterm_io_handler(struct stream_interface *si)
 			tmp_chunk.str = hterm_chunk.str + offset;
 			tmp_chunk.len = max;
 
-			/* Note that we cannot use hterm_chunk.str + offset below as we'd corrupt the buffer */
-			hterm_buf->p = hterm_buf->data;
-			hterm_buf->i = max;
-			hterm_buf->o = 0;
+			swp = NULL;
+			if (buffer_empty(si->ib->buf)) {
+				if (!hctx->req_chunked &&
+				    (si->ib->flags & CF_KERN_SPLICING)) {
 
-			swp = bi_swpbuf(si->ib, hterm_buf);
-			if (!swp || swp == hterm_buf) {
+					if (unlikely(si->ib->pipe == NULL)) {
+						if (pipes_used >= global.maxpipes || !(si->ib->pipe = get_pipe())) {
+							si->ib->flags &= ~CF_KERN_SPLICING;
+							goto abort_splice;
+						}
+					}
+					if (si->ib->pipe->data < hctx->req_size) {
+						int ret;
+
+						if (global.tune.pipesize && max > global.tune.pipesize)
+							max = global.tune.pipesize;
+
+						ret = tee(hterm_master_pipe->cons, si->ib->pipe->prod, max, SPLICE_F_NONBLOCK);
+						max = ret;
+						if (ret > 0) {
+							si->ib->flags |= CF_READ_PARTIAL;
+							si->ib->pipe->data += ret;
+							if (si->ib->to_forward) {
+								unsigned long fwd = ret;
+								if (si->ib->to_forward != CHN_INFINITE_FORWARD) {
+									if (fwd > si->ib->to_forward)
+										fwd = si->ib->to_forward;
+									si->ib->to_forward -= fwd;
+								}
+							}
+						}
+						else if (ret == 0 || errno == EAGAIN) {
+							si->flags |= SI_FL_WAIT_ROOM;
+							goto fail;
+						}
+					}
+					swp = hterm_buf; /* pretend the operation was performed */
+				}
+				else {
+				abort_splice:
+					/* Attempt a zero-copy without splice */
+
+					/* Note that we cannot use hterm_chunk.str + offset
+					 * below as we'd corrupt the buffer.
+					 */
+					hterm_buf->p = hterm_buf->data;
+					hterm_buf->i = max;
+					hterm_buf->o = 0;
+
+					swp = bi_swpbuf(si->ib, hterm_buf);
+					if (swp == hterm_buf)
+						swp = NULL;
+				}
+			}
+			else if (!hctx->req_chunked &&
+				 (si->ib->flags & CF_KERN_SPLICING)) {
+				/* let the buffer flush */
+				si->flags |= SI_FL_WAIT_ROOM;
+				goto fail;
+			}
+
+			if (!swp) {
 				/* zero copy failed */
 				if (bi_putchk(si->ib, &tmp_chunk) == -1)
 					goto fail;
@@ -3433,6 +3492,22 @@ int http_handle_httpterm(struct session *s, struct channel *req)
 		hterm_chunk.len = hterm_chunk.size;
 	}
 
+	if (unlikely(!hterm_master_pipe && (global.tune.options & GTUNE_USE_SPLICE))) {
+		hterm_master_pipe = get_pipe();
+		if (hterm_master_pipe) {
+			int total = 0;
+			int ret;
+			struct iovec v = { .iov_base = hterm_chunk.str,
+					   .iov_len = hterm_chunk.size, };
+
+			do {
+				ret = vmsplice(hterm_master_pipe->prod, &v, 1, SPLICE_F_NONBLOCK);
+				if (ret > 0)
+					total += ret;
+			} while (ret > 0 && (global.tune.pipesize && total < global.tune.pipesize));
+		}
+	}
+
 	appctx = si_appctx(si);
 	appctx->st0 = appctx->st1 = appctx->st2 = 0;
 
diff --git a/src/session.c b/src/session.c
index ad65963..49f5cbf 100644
--- a/src/session.c
+++ b/src/session.c
@@ -2483,7 +2483,7 @@ struct task *process_session(struct task *t)
 	    s->rep->to_forward &&
 	    (global.tune.options & GTUNE_USE_SPLICE) &&
 	    (objt_conn(s->si[0].end) && __objt_conn(s->si[0].end)->xprt && __objt_conn(s->si[0].end)->xprt->snd_pipe) &&
-	    (objt_conn(s->si[1].end) && __objt_conn(s->si[1].end)->xprt && __objt_conn(s->si[1].end)->xprt->rcv_pipe) &&
+	    //(objt_conn(s->si[1].end) && __objt_conn(s->si[1].end)->xprt && __objt_conn(s->si[1].end)->xprt->rcv_pipe) &&
 	    (pipes_used < global.maxpipes) &&
 	    (((s->fe->options2|s->be->options2) & PR_O2_SPLIC_RTR) ||
 	     (((s->fe->options2|s->be->options2) & PR_O2_SPLIC_AUT) &&
-- 
1.7.12.1