From: Jon Maloy <jmaloy@redhat.com>
To: passt-dev@passt.top, sbrivio@redhat.com, lvivier@redhat.com,
dgibson@redhat.com, jmaloy@redhat.com
Subject: [PATCH v2] tcp: unify payload and flags l2 frames array
Date: Fri, 1 Nov 2024 20:51:32 -0400 [thread overview]
Message-ID: <20241102005132.1408340-1-jmaloy@redhat.com> (raw)
In order to reduce static memory and code footprint, we merge
the array for l2 flag frames into the one for payload frames.
This change also ensures that no flag message will be sent out
over the l2 media bypassing already queued payload messages.
Performance measurements with iperf3, where we force all
traffic via the tap queue, show no significant difference:
Dual traffic both directions sinmultaneously, with patch:
========================================================
host->ns:
--------
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-100.00 sec 36.3 GBytes 3.12 Gbits/sec 4759 sender
[ 5] 0.00-100.04 sec 36.3 GBytes 3.11 Gbits/sec receiver
ns->host:
---------
[ ID] Interval Transfer Bitrate
[ 5] 0.00-100.00 sec 321 GBytes 27.6 Gbits/sec receiver
Dual traffic both directions sinmultaneously, without patch:
============================================================
host->ns:
--------
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-100.00 sec 35.0 GBytes 3.01 Gbits/sec 6001 sender
[ 5] 0.00-100.04 sec 34.8 GBytes 2.99 Gbits/sec receiver
ns->host
--------
[ ID] Interval Transfer Bitrate
[ 5] 0.00-100.00 sec 345 GBytes 29.6 Gbits/sec receiver
Single connection, with patch:
==============================
host->ns:
---------
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-100.00 sec 138 GBytes 11.8 Gbits/sec 922 sender
[ 5] 0.00-100.04 sec 138 GBytes 11.8 Gbits/sec receiver
ns->host:
-----------
[ ID] Interval Transfer Bitrate
[ 5] 0.00-100.00 sec 430 GBytes 36.9 Gbits/sec receiver
Single connection, without patch:
=================================
host->ns:
------------
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-100.00 sec 139 GBytes 11.9 Gbits/sec 900 sender
[ 5] 0.00-100.04 sec 139 GBytes 11.9 Gbits/sec receiver
ns->host:
---------
[ ID] Interval Transfer Bitrate
[ 5] 0.00-100.00 sec 440 GBytes 37.8 Gbits/sec receiver
Signed-off-by: Jon Maloy <jmaloy@redhat.com>
----
v2: - Adapted to and rebased on latest release.
- Removed redundant tcp_flags_push() function.
- Added measurement results.
Signed-off-by: Jon Maloy <jmaloy@redhat.com>
---
tcp.c | 1 -
tcp_buf.c | 66 ++++++++++++--------------------------------------
tcp_buf.h | 1 -
tcp_internal.h | 15 ------------
4 files changed, 16 insertions(+), 67 deletions(-)
diff --git a/tcp.c b/tcp.c
index 10ad06a..b17d5fe 100644
--- a/tcp.c
+++ b/tcp.c
@@ -936,7 +936,6 @@ bool tcp_flow_defer(const struct tcp_tap_conn *conn)
/* cppcheck-suppress [constParameterPointer, unmatchedSuppression] */
void tcp_defer_handler(struct ctx *c)
{
- tcp_flags_flush(c);
tcp_payload_flush(c);
}
diff --git a/tcp_buf.c b/tcp_buf.c
index 274e313..58df4cc 100644
--- a/tcp_buf.c
+++ b/tcp_buf.c
@@ -33,6 +33,7 @@
#include "tcp_internal.h"
#include "tcp_buf.h"
+#define PAYLOAD_FLAGS htons(0x5010) /* doff = 5, ack = 1 */
#define TCP_FRAMES_MEM 128
#define TCP_FRAMES \
(c->mode == MODE_PASTA ? 1 : TCP_FRAMES_MEM)
@@ -59,22 +60,10 @@ static_assert(MSS6 <= sizeof(tcp_payload[0].data), "MSS6 is greater than 65516")
static struct tcp_tap_conn *tcp_frame_conns[TCP_FRAMES_MEM];
static unsigned int tcp_payload_used;
-static struct tap_hdr tcp_flags_tap_hdr[TCP_FRAMES_MEM];
-/* IPv4 headers for TCP segment without payload */
-static struct iphdr tcp4_flags_ip[TCP_FRAMES_MEM];
-/* TCP segments without payload for IPv4 frames */
-static struct tcp_flags_t tcp_flags[TCP_FRAMES_MEM];
-
-static unsigned int tcp_flags_used;
-
-/* IPv6 headers for TCP segment without payload */
-static struct ipv6hdr tcp6_flags_ip[TCP_FRAMES_MEM];
-
/* recvmsg()/sendmsg() data for tap */
static struct iovec iov_sock [TCP_FRAMES_MEM + 1];
static struct iovec tcp_l2_iov[TCP_FRAMES_MEM][TCP_NUM_IOVS];
-static struct iovec tcp_l2_flags_iov[TCP_FRAMES_MEM][TCP_NUM_IOVS];
/**
* tcp_update_l2_buf() - Update Ethernet header buffers with addresses
@@ -107,13 +96,6 @@ void tcp_sock_iov_init(const struct ctx *c)
tcp_payload[i].th.ack = 1;
}
- for (i = 0; i < ARRAY_SIZE(tcp_flags); i++) {
- tcp6_flags_ip[i] = ip6;
- tcp4_flags_ip[i] = iph;
- tcp_flags[i].th.doff = sizeof(struct tcphdr) / 4;
- tcp_flags[i].th.ack = 1;
- }
-
for (i = 0; i < TCP_FRAMES_MEM; i++) {
struct iovec *iov = tcp_l2_iov[i];
@@ -121,25 +103,6 @@ void tcp_sock_iov_init(const struct ctx *c)
iov[TCP_IOV_ETH].iov_len = sizeof(struct ethhdr);
iov[TCP_IOV_PAYLOAD].iov_base = &tcp_payload[i];
}
-
- for (i = 0; i < TCP_FRAMES_MEM; i++) {
- struct iovec *iov = tcp_l2_flags_iov[i];
-
- iov[TCP_IOV_TAP] = tap_hdr_iov(c, &tcp_flags_tap_hdr[i]);
- iov[TCP_IOV_ETH].iov_len = sizeof(struct ethhdr);
- iov[TCP_IOV_PAYLOAD].iov_base = &tcp_flags[i];
- }
-}
-
-/**
- * tcp_flags_flush() - Send out buffers for segments with no data (flags)
- * @c: Execution context
- */
-void tcp_flags_flush(const struct ctx *c)
-{
- tap_send_frames(c, &tcp_l2_flags_iov[0][0], TCP_NUM_IOVS,
- tcp_flags_used);
- tcp_flags_used = 0;
}
/**
@@ -171,7 +134,7 @@ static void tcp_revert_seq(const struct ctx *c, struct tcp_tap_conn **conns,
}
/**
- * tcp_payload_flush() - Send out buffers for segments with data
+ * tcp_payload_flush() - Send out buffers for segments with data or flags
* @c: Execution context
*/
void tcp_payload_flush(const struct ctx *c)
@@ -197,37 +160,35 @@ void tcp_payload_flush(const struct ctx *c)
*/
int tcp_buf_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags)
{
- struct tcp_flags_t *payload;
+ struct tcp_payload_t *payload;
struct iovec *iov;
size_t optlen;
size_t l4len;
uint32_t seq;
int ret;
- iov = tcp_l2_flags_iov[tcp_flags_used];
+ iov = tcp_l2_iov[tcp_payload_used];
if (CONN_V4(conn)) {
- iov[TCP_IOV_IP] = IOV_OF_LVALUE(tcp4_flags_ip[tcp_flags_used]);
+ iov[TCP_IOV_IP] = IOV_OF_LVALUE(tcp4_payload_ip[tcp_payload_used]);
iov[TCP_IOV_ETH].iov_base = &tcp4_eth_src;
} else {
- iov[TCP_IOV_IP] = IOV_OF_LVALUE(tcp6_flags_ip[tcp_flags_used]);
+ iov[TCP_IOV_IP] = IOV_OF_LVALUE(tcp6_payload_ip[tcp_payload_used]);
iov[TCP_IOV_ETH].iov_base = &tcp6_eth_src;
}
payload = iov[TCP_IOV_PAYLOAD].iov_base;
seq = conn->seq_to_tap;
ret = tcp_prepare_flags(c, conn, flags, &payload->th,
- &payload->opts, &optlen);
+ (struct tcp_syn_opts *)&payload->data, &optlen);
if (ret <= 0)
return ret;
- tcp_flags_used++;
+ tcp_payload_used++;
l4len = tcp_l2_buf_fill_headers(conn, iov, optlen, NULL, seq, false);
iov[TCP_IOV_PAYLOAD].iov_len = l4len;
-
if (flags & DUP_ACK) {
- struct iovec *dup_iov;
+ struct iovec *dup_iov = tcp_l2_iov[tcp_payload_used++];
- dup_iov = tcp_l2_flags_iov[tcp_flags_used++];
memcpy(dup_iov[TCP_IOV_TAP].iov_base, iov[TCP_IOV_TAP].iov_base,
iov[TCP_IOV_TAP].iov_len);
dup_iov[TCP_IOV_ETH].iov_base = iov[TCP_IOV_ETH].iov_base;
@@ -237,8 +198,8 @@ int tcp_buf_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags)
dup_iov[TCP_IOV_PAYLOAD].iov_len = l4len;
}
- if (tcp_flags_used > TCP_FRAMES_MEM - 2)
- tcp_flags_flush(c);
+ if (tcp_payload_used > TCP_FRAMES_MEM - 2)
+ tcp_payload_flush(c);
return 0;
}
@@ -254,8 +215,10 @@ int tcp_buf_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags)
static void tcp_data_to_tap(const struct ctx *c, struct tcp_tap_conn *conn,
ssize_t dlen, int no_csum, uint32_t seq)
{
+ struct tcp_payload_t *payload;
const uint16_t *check = NULL;
struct iovec *iov;
+ uint16_t *flags;
size_t l4len;
conn->seq_to_tap = seq + dlen;
@@ -274,6 +237,9 @@ static void tcp_data_to_tap(const struct ctx *c, struct tcp_tap_conn *conn,
iov[TCP_IOV_IP] = IOV_OF_LVALUE(tcp6_payload_ip[tcp_payload_used]);
iov[TCP_IOV_ETH].iov_base = &tcp6_eth_src;
}
+ payload = iov[TCP_IOV_PAYLOAD].iov_base;
+ flags = &payload->th.window - 1;
+ *(flags) = PAYLOAD_FLAGS;
l4len = tcp_l2_buf_fill_headers(conn, iov, dlen, check, seq, false);
iov[TCP_IOV_PAYLOAD].iov_len = l4len;
if (++tcp_payload_used > TCP_FRAMES_MEM - 1)
diff --git a/tcp_buf.h b/tcp_buf.h
index 49c04d4..54f5e53 100644
--- a/tcp_buf.h
+++ b/tcp_buf.h
@@ -7,7 +7,6 @@
#define TCP_BUF_H
void tcp_sock_iov_init(const struct ctx *c);
-void tcp_flags_flush(const struct ctx *c);
void tcp_payload_flush(const struct ctx *c);
int tcp_buf_data_from_sock(const struct ctx *c, struct tcp_tap_conn *conn);
int tcp_buf_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags);
diff --git a/tcp_internal.h b/tcp_internal.h
index a5a47df..c846f60 100644
--- a/tcp_internal.h
+++ b/tcp_internal.h
@@ -134,21 +134,6 @@ struct tcp_syn_opts {
.ws = TCP_OPT_WS(ws_), \
})
-/**
- * struct tcp_flags_t - TCP header and data to send zero-length
- * segments (flags)
- * @th: TCP header
- * @opts TCP options
- */
-struct tcp_flags_t {
- struct tcphdr th;
- struct tcp_syn_opts opts;
-#ifdef __AVX2__
-} __attribute__ ((packed, aligned(32)));
-#else
-} __attribute__ ((packed, aligned(__alignof__(unsigned int))));
-#endif
-
extern char tcp_buf_discard [MAX_WINDOW];
void conn_flag_do(const struct ctx *c, struct tcp_tap_conn *conn,
--
@@ -134,21 +134,6 @@ struct tcp_syn_opts {
.ws = TCP_OPT_WS(ws_), \
})
-/**
- * struct tcp_flags_t - TCP header and data to send zero-length
- * segments (flags)
- * @th: TCP header
- * @opts TCP options
- */
-struct tcp_flags_t {
- struct tcphdr th;
- struct tcp_syn_opts opts;
-#ifdef __AVX2__
-} __attribute__ ((packed, aligned(32)));
-#else
-} __attribute__ ((packed, aligned(__alignof__(unsigned int))));
-#endif
-
extern char tcp_buf_discard [MAX_WINDOW];
void conn_flag_do(const struct ctx *c, struct tcp_tap_conn *conn,
--
2.45.2
next reply other threads:[~2024-11-02 0:51 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-11-02 0:51 Jon Maloy [this message]
2024-11-04 1:16 ` [PATCH v2] tcp: unify payload and flags l2 frames array David Gibson
2024-11-04 2:29 ` David Gibson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20241102005132.1408340-1-jmaloy@redhat.com \
--to=jmaloy@redhat.com \
--cc=dgibson@redhat.com \
--cc=lvivier@redhat.com \
--cc=passt-dev@passt.top \
--cc=sbrivio@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
Code repositories for project(s) associated with this public inbox
https://passt.top/passt
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for IMAP folder(s).