nuttx/net/tcp/tcp_send.c

717 lines
19 KiB
C
Raw Normal View History

/****************************************************************************
* net/tcp/tcp_send.c
*
* Copyright (C) 2007-2010, 2012, 2015, 2018-2019 Gregory Nutt. All rights
* reserved.
* Author: Gregory Nutt <gnutt@nuttx.org>
*
* Adapted for NuttX from logic in uIP which also has a BSD-like license:
*
* Original author Adam Dunkels <adam@dunkels.com>
* Copyright () 2001-2003, Adam Dunkels.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. The name of the author may not be used to endorse or promote
* products derived from this software without specific prior
* written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
* OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
* DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
* NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
****************************************************************************/
/****************************************************************************
* Included Files
****************************************************************************/
#include <nuttx/config.h>
#if defined(CONFIG_NET) && defined(CONFIG_NET_TCP)
#include <assert.h>
#include <stdint.h>
#include <string.h>
#include <debug.h>
#include <nuttx/net/netconfig.h>
#include <nuttx/net/netdev.h>
#include <nuttx/net/netstats.h>
#include <nuttx/net/ip.h>
#include <nuttx/net/tcp.h>
#include "netdev/netdev.h"
#include "devif/devif.h"
#include "inet/inet.h"
#include "tcp/tcp.h"
#include "utils/utils.h"
/****************************************************************************
* Private Functions
****************************************************************************/
/****************************************************************************
* Name: tcp_header
*
* Description:
* Get the length of the IP header
*
* Input Parameters:
* dev - The device driver structure to use in the send operation
*
* Returned Value:
* The length of the IP header (IPv4_HDRLEN or IPv6_HDRLEN)
*
****************************************************************************/
static inline FAR struct tcp_hdr_s *tcp_header(FAR struct net_driver_s *dev)
{
#ifdef CONFIG_NET_IPv6
#ifdef CONFIG_NET_IPv4
if (IFF_IS_IPv6(dev->d_flags))
#endif
{
return TCPIPv6BUF;
}
#endif /* CONFIG_NET_IPv6 */
#ifdef CONFIG_NET_IPv4
#ifdef CONFIG_NET_IPv6
else
#endif
{
return TCPIPv4BUF;
}
#endif /* CONFIG_NET_IPv4 */
}
/****************************************************************************
2014-06-25 02:12:49 +02:00
* Name: tcp_sendcommon
*
* Description:
* We're done with the input processing. We are now ready to send a reply
* Our job is to fill in all the fields of the TCP and IP headers before
* calculating the checksum and finally send the packet.
*
* Input Parameters:
* dev - The device driver structure to use in the send operation
* conn - The TCP connection structure holding connection information
*
* Returned Value:
* None
*
* Assumptions:
* Called with the network locked.
*
****************************************************************************/
2014-06-28 00:48:12 +02:00
static void tcp_sendcommon(FAR struct net_driver_s *dev,
FAR struct tcp_conn_s *conn,
FAR struct tcp_hdr_s *tcp)
{
/* Set TCP sequence numbers and port numbers */
memcpy(tcp->ackno, conn->rcvseq, 4);
memcpy(tcp->seqno, conn->sndseq, 4);
tcp->srcport = conn->lport;
tcp->destport = conn->rport;
/* Set the TCP window */
if (conn->tcpstateflags & TCP_STOPPED)
{
/* If the connection has issued TCP_STOPPED, we advertise a zero
* window so that the remote host will stop sending data.
*/
tcp->wnd[0] = 0;
tcp->wnd[1] = 0;
}
else
{
/* Update the TCP received window based on I/O buffer availability */
uint32_t rcvseq = tcp_getsequence(conn->rcvseq);
uint32_t recvwndo = tcp_get_recvwindow(dev, conn);
/* Update the Receiver Window */
conn->rcv_adv = rcvseq + recvwndo;
#ifdef CONFIG_NET_TCP_WINDOW_SCALE
recvwndo >>= conn->rcv_scale;
#endif
/* Set the TCP Window */
tcp->wnd[0] = recvwndo >> 8;
tcp->wnd[1] = recvwndo & 0xff;
}
tcp->urgp[0] = 0;
tcp->urgp[1] = 0;
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
/* Update device buffer length before setup the IP header */
iob_update_pktlen(dev->d_iob, dev->d_len, false);
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
/* Calculate chk & build L3 header */
#ifdef CONFIG_NET_IPv6
#ifdef CONFIG_NET_IPv4
if (IFF_IS_IPv6(dev->d_flags))
#endif
{
ninfo("do IPv6 IP header build!\n");
ipv6_build_header(IPv6BUF, dev->d_len - IPv6_HDRLEN,
IP_PROTO_TCP,
netdev_ipv6_srcaddr(dev, conn->u.ipv6.raddr),
conn->u.ipv6.raddr,
conn->sconn.ttl, conn->sconn.s_tclass);
/* Calculate TCP checksum. */
tcp->tcpchksum = 0;
tcp->tcpchksum = ~tcp_ipv6_chksum(dev);
#ifdef CONFIG_NET_STATISTICS
g_netstats.ipv6.sent++;
#endif
}
#endif /* CONFIG_NET_IPv6 */
#ifdef CONFIG_NET_IPv4
#ifdef CONFIG_NET_IPv6
else
#endif
{
ninfo("do IPv4 IP header build!\n");
ipv4_build_header(IPv4BUF, dev->d_len, IP_PROTO_TCP,
&dev->d_ipaddr, &conn->u.ipv4.raddr,
conn->sconn.ttl, conn->sconn.s_tos, NULL);
/* Calculate TCP checksum. */
tcp->tcpchksum = 0;
tcp->tcpchksum = ~tcp_ipv4_chksum(dev);
#ifdef CONFIG_NET_STATISTICS
g_netstats.ipv4.sent++;
#endif
}
#endif /* CONFIG_NET_IPv4 */
ninfo("Outgoing TCP packet length: %d bytes\n", dev->d_len);
#ifdef CONFIG_NET_STATISTICS
g_netstats.tcp.sent++;
#endif
#if !defined(CONFIG_NET_TCP_WRITE_BUFFERS)
if ((tcp->flags & (TCP_SYN | TCP_FIN)) != 0)
{
/* Remember sndseq that will be used in case of a possible
* SYN or FIN retransmission
*/
conn->rexmit_seq = tcp_getsequence(conn->sndseq);
/* Advance sndseq by +1 because SYN and FIN occupy
* one sequence number (RFC 793)
*/
net_incr32(conn->sndseq, 1);
}
#else
/* REVISIT for the buffered mode */
#endif
}
/****************************************************************************
* Public Functions
****************************************************************************/
/****************************************************************************
2014-06-25 02:12:49 +02:00
* Name: tcp_send
*
* Description:
* Setup to send a TCP packet
*
* Input Parameters:
* dev - The device driver structure to use in the send operation
* conn - The TCP connection structure holding connection information
* flags - flags to apply to the TCP header
* len - length of the message (includes the length of the IP and TCP
* headers)
*
* Returned Value:
* None
*
* Assumptions:
* Called with the network locked.
*
****************************************************************************/
2014-06-28 00:48:12 +02:00
void tcp_send(FAR struct net_driver_s *dev, FAR struct tcp_conn_s *conn,
2014-06-25 02:12:49 +02:00
uint16_t flags, uint16_t len)
{
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
FAR struct tcp_hdr_s *tcp;
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
if (dev->d_iob == NULL)
{
return;
}
tcp = tcp_header(dev);
tcp->flags = flags;
dev->d_len = len;
#ifdef CONFIG_NET_TCP_SELECTIVE_ACK
if ((conn->flags & TCP_SACK) && (flags == TCP_ACK) && conn->nofosegs > 0)
{
int optlen = conn->nofosegs * sizeof(struct tcp_sack_s);
int i;
tcp->optdata[0] = TCP_OPT_NOOP;
tcp->optdata[1] = TCP_OPT_NOOP;
tcp->optdata[2] = TCP_OPT_SACK;
tcp->optdata[3] = TCP_OPT_SACK_PERM_LEN + optlen;
optlen += 4;
for (i = 0; i < conn->nofosegs; i++)
{
ninfo("TCP SACK [%d]"
"[%" PRIu32 " : %" PRIu32 " : %" PRIu32 "]\n", i,
conn->ofosegs[i].left, conn->ofosegs[i].right,
TCP_SEQ_SUB(conn->ofosegs[i].right, conn->ofosegs[i].left));
tcp_setsequence(&tcp->optdata[4 + i * 2 * sizeof(uint32_t)],
conn->ofosegs[i].left);
tcp_setsequence(&tcp->optdata[4 + (i * 2 + 1) * sizeof(uint32_t)],
conn->ofosegs[i].right);
}
dev->d_len += optlen;
tcp->tcpoffset = ((TCP_HDRLEN + optlen) / 4) << 4;
}
else
#endif /* CONFIG_NET_TCP_SELECTIVE_ACK */
{
tcp->tcpoffset = (TCP_HDRLEN / 4) << 4;
}
tcp_sendcommon(dev, conn, tcp);
#if defined(CONFIG_NET_STATISTICS) && \
defined(CONFIG_NET_TCP_DEBUG_DROP_SEND)
#pragma message \
"CONFIG_NET_TCP_DEBUG_DROP_SEND is selected, this is debug " \
"feature to drop the tcp send packet on the floor, " \
"please confirm the configuration again if you do not want " \
"debug the TCP stack."
/* Debug feature to drop the tcp received packet on the floor */
if ((flags & TCP_PSH) != 0)
{
if ((g_netstats.tcp.sent %
CONFIG_NET_TCP_DEBUG_DROP_SEND_PROBABILITY) == 0)
{
uint32_t seq = tcp_getsequence(tcp->seqno);
ninfo("TCP DROP SNDPKT: "
"[%d][%" PRIu32 " : %" PRIu32 " : %d]\n",
g_netstats.tcp.sent, seq, TCP_SEQ_ADD(seq, dev->d_sndlen),
dev->d_sndlen);
dev->d_len = 0;
}
}
#endif
}
/****************************************************************************
2014-06-25 02:12:49 +02:00
* Name: tcp_reset
*
* Description:
* Send a TCP reset (no-data) message
*
* Input Parameters:
* dev - The device driver structure to use in the send operation
*
* Returned Value:
* None
*
* Assumptions:
* Called with the network locked.
*
****************************************************************************/
void tcp_reset(FAR struct net_driver_s *dev, FAR struct tcp_conn_s *conn)
{
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
FAR struct tcp_hdr_s *tcp;
uint32_t ackno;
uint16_t tmp16;
uint16_t acklen = 0;
uint8_t seqbyte;
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
if (dev->d_iob == NULL)
{
return;
}
#ifdef CONFIG_NET_STATISTICS
g_netstats.tcp.rst++;
#endif
/* TCP setup */
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
tcp = tcp_header(dev);
if ((tcp->flags & TCP_SYN) != 0 || (tcp->flags & TCP_FIN) != 0)
{
acklen++;
}
#ifdef CONFIG_NET_IPv6
#ifdef CONFIG_NET_IPv4
2019-08-28 00:20:39 +02:00
if (IFF_IS_IPv6(dev->d_flags))
#endif
{
FAR struct ipv6_hdr_s *ip = IPv6BUF;
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
acklen += (ip->len[0] << 8 | ip->len[1]);
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
/* Set the packet length to the size of the IPv6 + TCP headers */
dev->d_len = IPv6TCP_HDRLEN;
}
#endif /* CONFIG_NET_IPv6 */
#ifdef CONFIG_NET_IPv4
#ifdef CONFIG_NET_IPv6
else
#endif
{
FAR struct ipv4_hdr_s *ip = IPv4BUF;
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
acklen += (ip->len[0] << 8) + ip->len[1] - (ip->vhl & 0x0f) * 4;
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
/* Set the packet length to the size of the IPv4 + TCP headers */
dev->d_len = IPv4TCP_HDRLEN;
}
#endif /* CONFIG_NET_IPv4 */
acklen -= (tcp->tcpoffset >> 4) << 2;
tcp->flags = TCP_RST | TCP_ACK;
tcp->tcpoffset = 5 << 4;
/* Flip the seqno and ackno fields in the TCP header. */
seqbyte = tcp->seqno[3];
tcp->seqno[3] = tcp->ackno[3];
tcp->ackno[3] = seqbyte;
seqbyte = tcp->seqno[2];
tcp->seqno[2] = tcp->ackno[2];
tcp->ackno[2] = seqbyte;
seqbyte = tcp->seqno[1];
tcp->seqno[1] = tcp->ackno[1];
tcp->ackno[1] = seqbyte;
seqbyte = tcp->seqno[0];
tcp->seqno[0] = tcp->ackno[0];
tcp->ackno[0] = seqbyte;
/* We also have to increase the sequence number we are
* acknowledging. If the least significant byte overflowed, we need
* to propagate the carry to the other bytes as well.
*/
ackno = tcp_addsequence(tcp->ackno, acklen);
tcp_setsequence(tcp->ackno, ackno);
/* Swap port numbers. */
tmp16 = tcp->srcport;
tcp->srcport = tcp->destport;
tcp->destport = tmp16;
/* Initialize the rest of the tcp header to sane values.
*/
tcp->wnd[0] = 0;
tcp->wnd[1] = 0;
tcp->urgp[0] = 0;
tcp->urgp[0] = 0;
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
/* Update device buffer length before setup the IP header */
iob_update_pktlen(dev->d_iob, dev->d_len, false);
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
/* Calculate chk & build L3 header */
#ifdef CONFIG_NET_IPv6
#ifdef CONFIG_NET_IPv4
if (IFF_IS_IPv6(dev->d_flags))
#endif
{
FAR struct ipv6_hdr_s *ipv6 = IPv6BUF;
ipv6_build_header(ipv6, dev->d_len - IPv6_HDRLEN,
IP_PROTO_TCP,
netdev_ipv6_srcaddr(dev, ipv6->srcipaddr),
ipv6->srcipaddr,
conn ? conn->sconn.ttl : IP_TTL_DEFAULT,
conn ? conn->sconn.s_tos : 0);
tcp->tcpchksum = 0;
tcp->tcpchksum = ~tcp_ipv6_chksum(dev);
}
#endif /* CONFIG_NET_IPv6 */
#ifdef CONFIG_NET_IPv4
#ifdef CONFIG_NET_IPv6
else
#endif
{
FAR struct ipv4_hdr_s *ipv4 = IPv4BUF;
ipv4_build_header(IPv4BUF, dev->d_len, IP_PROTO_TCP,
&dev->d_ipaddr, (FAR in_addr_t *)ipv4->srcipaddr,
conn ? conn->sconn.ttl : IP_TTL_DEFAULT,
conn ? conn->sconn.s_tos : 0, NULL);
tcp->tcpchksum = 0;
tcp->tcpchksum = ~tcp_ipv4_chksum(dev);
}
#endif /* CONFIG_NET_IPv4 */
}
/****************************************************************************
* Name: tcp_rx_mss
*
* Description:
* Return the MSS to advertize to the peer.
*
* Input Parameters:
* dev - The device driver structure
*
* Returned Value:
* The MSS value.
*
****************************************************************************/
uint16_t tcp_rx_mss(FAR struct net_driver_s *dev)
{
uint16_t tcp_mss;
#ifdef CONFIG_NET_IPv6
#ifdef CONFIG_NET_IPv4
if (IFF_IS_IPv6(dev->d_flags))
#endif
{
tcp_mss = TCP_IPv6_MSS(dev);
}
#endif /* CONFIG_NET_IPv6 */
#ifdef CONFIG_NET_IPv4
#ifdef CONFIG_NET_IPv6
else
#endif
{
tcp_mss = TCP_IPv4_MSS(dev);
}
#endif /* CONFIG_NET_IPv4 */
return tcp_mss;
}
/****************************************************************************
* Name: tcp_synack
*
* Description:
* Send the SYN, ACK, or SYNACK response.
*
* - SYN and SYNACK are sent only from the TCP state machine.
* - ACK may be sent alone only if delayed ACKs are enabled and the ACK
* delay timeout occurs.
*
* Input Parameters:
* dev - The device driver structure to use in the send operation
* conn - The TCP connection structure holding connection information
* ack - The ACK response to send
*
* Returned Value:
* None
*
* Assumptions:
* Called with the network locked.
*
****************************************************************************/
void tcp_synack(FAR struct net_driver_s *dev, FAR struct tcp_conn_s *conn,
uint8_t ack)
{
FAR struct tcp_hdr_s *tcp;
uint16_t tcp_mss;
int16_t optlen = 0;
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
if (dev->d_iob == NULL)
{
return;
}
/* Get the offset TCP header address for this packet */
tcp = tcp_header(dev);
/* Set the packet length for the TCP Maximum Segment Size */
dev->d_len = tcpip_hdrsize(conn);
/* Set the packet length for the TCP Maximum Segment Size */
#ifdef CONFIG_NET_TCPPROTO_OPTIONS
if (conn->user_mss != 0 && conn->user_mss < tcp_rx_mss(dev))
{
tcp_mss = conn->user_mss;
}
else
#endif
{
tcp_mss = tcp_rx_mss(dev);
}
/* Save the ACK bits */
tcp->flags = ack;
/* We send out the TCP Maximum Segment Size option with our ACK. */
tcp->optdata[optlen++] = TCP_OPT_MSS;
tcp->optdata[optlen++] = TCP_OPT_MSS_LEN;
tcp->optdata[optlen++] = tcp_mss >> 8;
tcp->optdata[optlen++] = tcp_mss & 0xff;
#ifdef CONFIG_NET_TCP_WINDOW_SCALE
if (tcp->flags == TCP_SYN ||
((tcp->flags == (TCP_ACK | TCP_SYN)) && (conn->flags & TCP_WSCALE)))
{
tcp->optdata[optlen++] = TCP_OPT_NOOP;
tcp->optdata[optlen++] = TCP_OPT_WS;
tcp->optdata[optlen++] = TCP_OPT_WS_LEN;
tcp->optdata[optlen++] = CONFIG_NET_TCP_WINDOW_SCALE_FACTOR;
}
#endif
#ifdef CONFIG_NET_TCP_SELECTIVE_ACK
if (tcp->flags == TCP_SYN ||
((tcp->flags == (TCP_ACK | TCP_SYN)) && (conn->flags & TCP_SACK)))
{
tcp->optdata[optlen++] = TCP_OPT_NOOP;
tcp->optdata[optlen++] = TCP_OPT_NOOP;
tcp->optdata[optlen++] = TCP_OPT_SACK_PERM;
tcp->optdata[optlen++] = TCP_OPT_SACK_PERM_LEN;
}
#endif
tcp->tcpoffset = ((TCP_HDRLEN + optlen) / 4) << 4;
dev->d_len += optlen;
/* Complete the common portions of the TCP message */
tcp_sendcommon(dev, conn, tcp);
}
/****************************************************************************
* Name: tcp_send_txnotify
*
* Description:
* Notify the appropriate device driver that we are have data ready to
* be send (TCP)
*
* Input Parameters:
* psock - Socket state structure
* conn - The TCP connection structure
*
* Returned Value:
* None
*
****************************************************************************/
void tcp_send_txnotify(FAR struct socket *psock,
FAR struct tcp_conn_s *conn)
{
#ifdef CONFIG_NET_IPv4
#ifdef CONFIG_NET_IPv6
/* If both IPv4 and IPv6 support are enabled, then we will need to select
* the device driver using the appropriate IP domain.
*/
if (psock->s_domain == PF_INET)
#endif
{
/* Notify the device driver that send data is available */
netdev_ipv4_txnotify(conn->u.ipv4.laddr, conn->u.ipv4.raddr);
}
#endif /* CONFIG_NET_IPv4 */
#ifdef CONFIG_NET_IPv6
#ifdef CONFIG_NET_IPv4
else /* if (psock->s_domain == PF_INET6) */
#endif /* CONFIG_NET_IPv4 */
{
/* Notify the device driver that send data is available */
DEBUGASSERT(psock->s_domain == PF_INET6);
netdev_ipv6_txnotify(conn->u.ipv6.laddr, conn->u.ipv6.raddr);
}
#endif /* CONFIG_NET_IPv6 */
}
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
/****************************************************************************
* Name: tcpip_hdrsize
*
* Description:
* Get the total size of L3 and L4 TCP header
*
* Input Parameters:
* conn The connection structure associated with the socket
*
* Returned Value:
* the total size of L3 and L4 TCP header
*
****************************************************************************/
uint16_t tcpip_hdrsize(FAR struct tcp_conn_s *conn)
{
uint16_t hdrsize = sizeof(struct tcp_hdr_s);
UNUSED(conn);
return net_ip_domain_select(conn->domain,
sizeof(struct ipv4_hdr_s) + hdrsize,
sizeof(struct ipv6_hdr_s) + hdrsize);
net/l2/l3/l4: add support of iob offload 1. Add new config CONFIG_NET_LL_GUARDSIZE to isolation of l2 stack, which will benefit l3(IP) layer for multi-MAC(l2) implementation, especially in some NICs such as celluler net driver. new configuration options: CONFIG_NET_LL_GUARDSIZE CONFIG_NET_LL_GUARDSIZE will reserved l2 buffer header size of network buffer to isolate the L2/L3 (MAC/IP) data on network layer, which will be beneficial to L3 network layer protocol transparent transmission and forwarding ------------------------------------------------------------ Layout of frist iob entry: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- iob | Reserved | io_len | ------------------------------------------------- ------------------------------------------------------------- Layout of different NICs implementation: iob_data (aligned by CONFIG_IOB_ALIGNMENT) | | io_offset(CONFIG_NET_LL_GUARDSIZE) | | ------------------------------------------------- Ethernet | Reserved | ETH_HDRLEN | io_len | ---------------------------------|--------------- 8021Q | Reserved | ETH_8021Q_HDRLEN | io_len | ---------------------------------|--------------- ipforward | Reserved | io_len | ------------------------------------------------- -------------------------------------------------------------------- 2. Support iob offload to l2 driver to avoid unnecessary memory copy Support send/receive iob vectors directly between the NICs and l3/l4 stack to avoid unnecessary memory copies, especially on hardware that supports Scatter/gather, which can greatly improve performance. new interface to support iob offload: ------------------------------------------ | IOB version | original | |----------------------------------------| | devif_iob_poll() | devif_poll() | | ... | ... | ------------------------------------------ -------------------------------------------------------------------- 1> NIC hardware support Scatter/gather transfer TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | dev->d_iob: | --------------- --------------- io_data iob1 | | | iob3 | | | \ --------------- --------------- --------------- | --------------- | iob0 | | | | iob2 | | | | --------------- | --------------- | \ | / / \ | / / ---------------------------------------------- NICs io vector | | | | | | | | | | ---------------------------------------------- RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to each iobs) -------------------------------------------------------------------- 2> CONFIG_IOB_BUFSIZE is greater than MTU: TX: "(CONFIG_IOB_BUFSIZE) > (MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE + CONFIG_NET_LL_GUARDSIZE)" tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll("NIC"_txpoll) callback() // "NIC"_txpoll | "NIC"_send() (dev->d_iob->io_data[CONFIG_NET_LL_GUARDSIZE - NET_LL_HDRLEN(dev)]) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | pkt/ipv[4/6]_input()/... | | NICs io vector receive(iov_base to io_data) -------------------------------------------------------------------- 3> Compatible with all old flat buffer NICs TX: tcp_poll()/udp_poll()/pkt_poll()/...(l3|l4) / \ / \ devif_poll_[l3|l4]_connections() devif_iob_send() (nocopy:udp/icmp/...) / \ (copy:tcp) / \ devif_iob_poll(devif_poll_callback()) devif_poll_callback() /* new interface, gather iobs to flat buffer */ / \ / \ devif_poll("NIC"_txpoll) "NIC"_send()(dev->d_buf) RX: [tcp|udp|icmp|...]ipv[4|6]_data_handler()(iob_concat/append to readahead) | | [tcp|udp|icmp|...]_ipv[4|6]_in()/... | | netdev_input() /* new interface, Scatter/gather flat/iob buffer */ | | pkt/ipv[4|6]_input()/... | | NICs io vector receive(Orignal flat buffer) 3. Iperf passthrough on NuttX simulator: ------------------------------------------------- | Protocol | Server | Client | | |-----------------------------------------------| | TCP | 813 | 834 | Mbits/sec | | TCP(Offload) | 1720 | 1100 | Mbits/sec | | UDP | 22 | 757 | Mbits/sec | | UDP(Offload) | 25 | 1250 | Mbits/sec | ------------------------------------------------- Signed-off-by: chao an <anchao@xiaomi.com>
2022-11-23 12:39:40 +01:00
}
#endif /* CONFIG_NET && CONFIG_NET_TCP */