On Thu, Feb 01, 2024 at 05:12:11PM -0500, Jon Maloy wrote:

The kernel may support recvmsg(MSG_PEEK), starting reading data from a given offset set by the SO_PEEK_OFF socket option. This makes it possible to avoid repeated reading of already read initial bytes of a received message, hence saving read cycles when forwarding TCP messages in the host->name space direction.

In this commit, we add functionality to leverage this feature when available, while we fall back to the previous behavior when not.

Measurements with iperf3 shows that throughput increases with 15-20 percent in the host->namespace direction when this feature is used.

Signed-off-by: Jon Maloy --- tcp.c | 35 +++++++++++++++++++++++++++++++++-- 1 file changed, 33 insertions(+), 2 deletions(-)

diff --git a/tcp.c b/tcp.c index 905d26f..58674eb 100644 --- a/tcp.c +++ b/tcp.c @@ -505,6 +505,7 @@ static struct tcp_buf_seq_update tcp6_l2_buf_seq_update[TCP_FRAMES_MEM]; static unsigned int tcp6_l2_buf_used;

/* recvmsg()/sendmsg() data for tap */ +static bool peek_offset_cap = false; static char tcp_buf_discard [MAX_WINDOW]; static struct iovec iov_sock [TCP_FRAMES_MEM + 1];

@@ -582,6 +583,14 @@ static_assert(ARRAY_SIZE(tc_hash) >= FLOW_MAX, int init_sock_pool4 [TCP_SOCK_POOL_SIZE]; int init_sock_pool6 [TCP_SOCK_POOL_SIZE];

A function comnment would be nice.

+static void set_peek_offset(int s, int offset) +{ + if (!peek_offset_cap) + return; + if (setsockopt(s, SOL_SOCKET, SO_PEEK_OFF, &offset, sizeof(offset))) + perror("Failed to set SO_PEEK_OFF\n"); +} + /** * tcp_conn_epoll_events() - epoll events mask for given connection state * @events: Current connection events @@ -1951,7 +1960,7 @@ static void tcp_conn_from_tap(struct ctx *c, if (bind(s, (struct sockaddr *)&addr6_ll, sizeof(addr6_ll))) goto cancel; } - + set_peek_offset(s, 0); conn = &flow->tcp; conn->f.type = FLOW_TCP; conn->sock = s; @@ -2174,6 +2183,12 @@ static int tcp_data_from_sock(struct ctx *c, struct tcp_tap_conn *conn) if (iov_rem) iov_sock[fill_bufs].iov_len = iov_rem;

+ /* Don't use discard buffer if SO_PEEK_OFF is supported. */ + if (peek_offset_cap) { + mh_sock.msg_iov = &iov_sock[1]; + mh_sock.msg_iovlen -= 1; + } + /* Receive into buffers, don't dequeue until acknowledged by guest. */ do len = recvmsg(s, &mh_sock, MSG_PEEK); @@ -2195,7 +2210,9 @@ static int tcp_data_from_sock(struct ctx *c, struct tcp_tap_conn *conn) return 0; }

- sendlen = len - already_sent; + sendlen = len; + if (!peek_offset_cap) + sendlen -= already_sent; if (sendlen <= 0) { conn_flag(c, conn, STALLED); return 0; @@ -2367,6 +2384,7 @@ static int tcp_data_from_tap(struct ctx *c, struct tcp_tap_conn *conn, max_ack_seq, conn->seq_to_tap); conn->seq_ack_from_tap = max_ack_seq; conn->seq_to_tap = max_ack_seq; + set_peek_offset(conn->sock, 0);

I don't think a zero offset is entirely correct here. Instead I think it should be (max_ack_seq - conn->seq_ack_from_tap). We're retransmitting from the segment immediately after max_ack_seq, so we want that offset, minus the offset of the the "TRUNC" pointer in the stream, which is conn->seq_ack_from_tap. Now, usually that will be 0, because tcp_sock_consume() just above will have advanced the TRUNC pointer up to max_ack_seq, so tcp_update_seqack_from_tap() will have set conn->seq_ack_from_tap equal to max_ack_seq. However, tcp_sock_consume() can fail, in which case neither the TRUNC pointer nor conn->seq_ack_from_tap will be advanced.

tcp_data_from_sock(c, conn); }

@@ -2718,6 +2736,7 @@ void tcp_listen_handler(struct ctx *c, union epoll_ref ref, tcp_splice_conn_from_sock(c, ref.tcp_listen, &flow->tcp_splice, s, (struct sockaddr *)&sa)) return; + set_peek_offset(s, 0);

Since this is specific to "tap" connections, I think this would be better moved into tcp_tap_conn_from_sock().

tcp_tap_conn_from_sock(c, ref.tcp_listen, &flow->tcp, s, (struct sockaddr *)&sa, now); @@ -3042,6 +3061,7 @@ static void tcp_sock_refill_init(const struct ctx *c) int tcp_init(struct ctx *c) { unsigned b; + int s;

for (b = 0; b < TCP_HASH_TABLE_SIZE; b++) tc_hash[b] = FLOW_SIDX_NONE; @@ -3065,6 +3085,17 @@ int tcp_init(struct ctx *c) NS_CALL(tcp_ns_socks_init, c); }

+ /* Probe for SO_PEEK_OFF support */ + s = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP); + if (s < 0) { + perror("Temporary tcp socket creation failed\n"); + } else { + if (!setsockopt(s, SOL_SOCKET, SO_PEEK_OFF, &(int){0}, sizeof(int))) { + peek_offset_cap = true; + } + close(s); + } + printf("SO_PEEK_OFF%ssupported\n", peek_offset_cap ? " " : " not "); return 0; }

I think you're also missing one call to set_peek_offset(). You've got a call for the fast re-transmit case (duplicate ack), but not for the "slow retransmit" case (ack timeout), which is in tcp_timer_handler(). *thinks* the peek pointer basically needs to always be kept in sync with conn->seq_to_tap. When seq_to_tap is advanced because we've peeked more data, the kernel should automatically update the peek pointer to match. But any other place we adjust seq_to_tap probably needs a setsockopt(). That looks to be the "slow" retransmit case in tcp_timer_handler(), but also the "ack sequence gap" bit at the top of tcp_data_from_sock(). I'm not really sure what that case is about, but I think we need to adjust SO_PEEK_OFF. I wonder if it would be safe to have a helper function that adjusts both seq_to_tap and SO_PEEK_OFF. -- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson