]> git.baikalelectronics.ru Git - uboot.git/commitdiff
net: Add TCP protocol
authorYing-Chun Liu (PaulLiu) <paul.liu@linaro.org>
Tue, 8 Nov 2022 06:17:28 +0000 (14:17 +0800)
committerTom Rini <trini@konsulko.com>
Mon, 28 Nov 2022 18:06:39 +0000 (13:06 -0500)
Currently file transfers are done using tftp or NFS both
over udp. This requires a request to be sent from client
(u-boot) to the boot server.

The current standard is TCP with selective acknowledgment.

Signed-off-by: Duncan Hare <DH@Synoia.com>
Signed-off-by: Duncan Hare <DuncanCHare@yahoo.com>
Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
Cc: Christian Gmeiner <christian.gmeiner@gmail.com>
Cc: Joe Hershberger <joe.hershberger@ni.com>
Cc: Michal Simek <michal.simek@xilinx.com>
Cc: Ramon Fried <rfried.dev@gmail.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
include/net.h
include/net/tcp.h [new file with mode: 0644]
net/Kconfig
net/Makefile
net/net.c
net/tcp.c [new file with mode: 0644]

index 32364ed0ced9ee9e05580b0e09dd6cdca7622b98..f4140523c21c3d5b72156f5864a833a9858be281 100644 (file)
@@ -365,6 +365,7 @@ struct vlan_ethernet_hdr {
 #define PROT_NCSI      0x88f8          /* NC-SI control packets        */
 
 #define IPPROTO_ICMP    1      /* Internet Control Message Protocol    */
+#define IPPROTO_TCP    6       /* Transmission Control Protocol        */
 #define IPPROTO_UDP    17      /* User Datagram Protocol               */
 
 /*
@@ -690,19 +691,36 @@ static inline void net_send_packet(uchar *pkt, int len)
        (void) eth_send(pkt, len);
 }
 
-/*
- * Transmit "net_tx_packet" as UDP packet, performing ARP request if needed
- *  (ether will be populated)
- *
- * @param ether Raw packet buffer
- * @param dest IP address to send the datagram to
- * @param dport Destination UDP port
- * @param sport Source UDP port
- * @param payload_len Length of data after the UDP header
+/**
+ * net_send_ip_packet() - Transmit "net_tx_packet" as UDP or TCP packet,
+ *                        send ARP request if needed (ether will be populated)
+ * @ether: Raw packet buffer
+ * @dest: IP address to send the datagram to
+ * @dport: Destination UDP port
+ * @sport: Source UDP port
+ * @payload_len: Length of data after the UDP header
+ * @action: TCP action to be performed
+ * @tcp_seq_num: TCP sequence number of this transmission
+ * @tcp_ack_num: TCP stream acknolegement number
+ *
+ * Return: 0 on success, other value on failure
  */
 int net_send_ip_packet(uchar *ether, struct in_addr dest, int dport, int sport,
                       int payload_len, int proto, u8 action, u32 tcp_seq_num,
                       u32 tcp_ack_num);
+/**
+ * net_send_tcp_packet() - Transmit TCP packet.
+ * @payload_len: length of payload
+ * @dport: Destination TCP port
+ * @sport: Source TCP port
+ * @action: TCP action to be performed
+ * @tcp_seq_num: TCP sequence number of this transmission
+ * @tcp_ack_num: TCP stream acknolegement number
+ *
+ * Return: 0 on success, other value on failure
+ */
+int net_send_tcp_packet(int payload_len, int dport, int sport, u8 action,
+                       u32 tcp_seq_num, u32 tcp_ack_num);
 int net_send_udp_packet(uchar *ether, struct in_addr dest, int dport,
                        int sport, int payload_len);
 
diff --git a/include/net/tcp.h b/include/net/tcp.h
new file mode 100644 (file)
index 0000000..3225516
--- /dev/null
@@ -0,0 +1,299 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * TCP Support with SACK for file transfer.
+ *
+ * Copyright 2017 Duncan Hare, All rights reserved.
+ */
+
+#define TCP_ACTIVITY 127               /* Number of packets received   */
+                                       /* before console progress mark */
+/**
+ * struct ip_tcp_hdr - IP and TCP header
+ * @ip_hl_v: header length and version
+ * @ip_tos: type of service
+ * @ip_len: total length
+ * @ip_id: identification
+ * @ip_off: fragment offset field
+ * @ip_ttl: time to live
+ * @ip_p: protocol
+ * @ip_sum: checksum
+ * @ip_src: Source IP address
+ * @ip_dst: Destination IP address
+ * @tcp_src: TCP source port
+ * @tcp_dst: TCP destination port
+ * @tcp_seq: TCP sequence number
+ * @tcp_ack: TCP Acknowledgment number
+ * @tcp_hlen: 4 bits TCP header Length/4, 4 bits reserved, 2 more bits reserved
+ * @tcp_flag: flags of TCP
+ * @tcp_win: TCP windows size
+ * @tcp_xsum: Checksum
+ * @tcp_ugr: Pointer to urgent data
+ */
+struct ip_tcp_hdr {
+       u8              ip_hl_v;
+       u8              ip_tos;
+       u16             ip_len;
+       u16             ip_id;
+       u16             ip_off;
+       u8              ip_ttl;
+       u8              ip_p;
+       u16             ip_sum;
+       struct in_addr  ip_src;
+       struct in_addr  ip_dst;
+       u16             tcp_src;
+       u16             tcp_dst;
+       u32             tcp_seq;
+       u32             tcp_ack;
+       u8              tcp_hlen;
+       u8              tcp_flags;
+       u16             tcp_win;
+       u16             tcp_xsum;
+       u16             tcp_ugr;
+} __packed;
+
+#define IP_TCP_HDR_SIZE                (sizeof(struct ip_tcp_hdr))
+#define TCP_HDR_SIZE           (IP_TCP_HDR_SIZE  - IP_HDR_SIZE)
+
+#define TCP_DATA       0x00    /* Data Packet - internal use only      */
+#define TCP_FIN                0x01    /* Finish flag                          */
+#define TCP_SYN                0x02    /* Synch (start) flag                   */
+#define TCP_RST                0x04    /* reset flag                           */
+#define TCP_PUSH       0x08    /* Push - Notify app                    */
+#define TCP_ACK                0x10    /* Acknowledgment of data received      */
+#define TCP_URG                0x20    /* Urgent                               */
+#define TCP_ECE                0x40    /* Congestion control                   */
+#define TCP_CWR                0x80    /* Congestion Control                   */
+
+/*
+ * TCP header options, Seq, MSS, and SACK
+ */
+
+#define TCP_SACK 32                    /* Number of packets analyzed   */
+                                       /* on leading edge of stream    */
+
+#define TCP_O_END      0x00            /* End of option list           */
+#define TCP_1_NOP      0x01            /* Single padding NOP           */
+#define TCP_O_NOP      0x01010101      /* NOPs pad to 32 bit boundary  */
+#define TCP_O_MSS      0x02            /* MSS Size option              */
+#define TCP_O_SCL      0x03            /* Window Scale option          */
+#define TCP_P_SACK     0x04            /* SACK permitted               */
+#define TCP_V_SACK     0x05            /* SACK values                  */
+#define TCP_O_TS       0x08            /* Timestamp option             */
+#define TCP_OPT_LEN_2  0x02
+#define TCP_OPT_LEN_3  0x03
+#define TCP_OPT_LEN_4  0x04
+#define TCP_OPT_LEN_6  0x06
+#define TCP_OPT_LEN_8  0x08
+#define TCP_OPT_LEN_A  0x0a            /* Timestamp Length             */
+#define TCP_MSS                1460            /* Max segment size             */
+#define TCP_SCALE      0x01            /* Scale                        */
+
+/**
+ * struct tcp_mss - TCP option structure for MSS (Max segment size)
+ * @kind: Field ID
+ * @len: Field length
+ * @mss: Segment size value
+ */
+struct tcp_mss {
+       u8      kind;
+       u8      len;
+       u16     mss;
+} __packed;
+
+/**
+ * struct tcp_scale - TCP option structure for Windows scale
+ * @kind: Field ID
+ * @len: Field length
+ * @scale: windows shift value used for networks with many hops.
+ *         Typically 4 or more hops
+ */
+struct tcp_scale {
+       u8      kind;
+       u8      len;
+       u8      scale;
+} __packed;
+
+/**
+ * struct tcp_sack_p - TCP option structure for SACK permitted
+ * @kind: Field ID
+ * @len: Field length
+ */
+struct tcp_sack_p {
+       u8      kind;
+       u8      len;
+} __packed;
+
+/**
+ * struct sack_edges - structure for SACK edges
+ * @l: Left edge of stream
+ * @r: right edge of stream
+ */
+struct sack_edges {
+       u32     l;
+       u32     r;
+} __packed;
+
+#define TCP_SACK_SIZE (sizeof(struct sack_edges))
+
+/*
+ * A TCP stream has holes when packets are missing or disordered.
+ * A hill is the inverse of a hole, and is data received.
+ * TCP received hills (a sequence of data), and inferrs Holes
+ * from the "hills" or packets received.
+ */
+
+#define TCP_SACK_HILLS 4
+
+/**
+ * struct tcp_sack_v - TCP option structure for SACK
+ * @kind: Field ID
+ * @len: Field length
+ * @hill: L & R window edges
+ */
+struct tcp_sack_v {
+       u8      kind;
+       u8      len;
+       struct  sack_edges hill[TCP_SACK_HILLS];
+} __packed;
+
+/**
+ * struct tcp_t_opt - TCP option structure for time stamps
+ * @kind: Field ID
+ * @len: Field length
+ * @t_snd: Sender timestamp
+ * @t_rcv: Receiver timestamp
+ */
+struct tcp_t_opt {
+       u8      kind;
+       u8      len;
+       u32     t_snd;
+       u32     t_rcv;
+} __packed;
+
+#define TCP_TSOPT_SIZE (sizeof(struct tcp_t_opt))
+
+/*
+ * ip tcp  structure with options
+ */
+
+/**
+ * struct ip_tcp_hdr_o - IP + TCP header + TCP options
+ * @hdr: IP + TCP header
+ * @mss: TCP MSS Option
+ * @scale: TCP Windows Scale Option
+ * @sack_p: TCP Sack-Permitted Option
+ * @t_opt: TCP Timestamp Option
+ * @end: end of options
+ */
+struct ip_tcp_hdr_o {
+       struct  ip_tcp_hdr hdr;
+       struct  tcp_mss    mss;
+       struct  tcp_scale  scale;
+       struct  tcp_sack_p sack_p;
+       struct  tcp_t_opt  t_opt;
+       u8      end;
+} __packed;
+
+#define IP_TCP_O_SIZE (sizeof(struct ip_tcp_hdr_o))
+
+/**
+ * struct ip_tcp_hdr_s - IP + TCP header + TCP options
+ * @hdr: IP + TCP header
+ * @t_opt: TCP Timestamp Option
+ * @sack_v: TCP SACK Option
+ * @end: end of options
+ */
+struct ip_tcp_hdr_s {
+       struct  ip_tcp_hdr      hdr;
+       struct  tcp_t_opt       t_opt;
+       struct  tcp_sack_v      sack_v;
+       u8      end;
+} __packed;
+
+#define IP_TCP_SACK_SIZE (sizeof(struct ip_tcp_hdr_s))
+
+/*
+ * TCP pseudo header definitions
+ */
+#define PSEUDO_PAD_SIZE        8
+
+/**
+ * struct pseudo_hdr - Pseudo Header
+ * @padding: pseudo hdr size = ip_tcp hdr size
+ * @p_src: Source IP address
+ * @p_dst: Destination IP address
+ * @rsvd: reserved
+ * @p: protocol
+ * @len: length of header
+ */
+struct pseudo_hdr {
+       u8 padding[PSEUDO_PAD_SIZE];
+       struct in_addr p_src;
+       struct in_addr p_dst;
+       u8      rsvd;
+       u8      p;
+       u16     len;
+} __packed;
+
+#define PSEUDO_HDR_SIZE        (sizeof(struct pseudo_hdr)) - PSEUDO_PAD_SIZE
+
+/**
+ * union tcp_build_pkt - union for building TCP/IP packet.
+ * @ph: pseudo header
+ * @ip: IP and TCP header plus TCP options
+ * @sack: IP and TCP header plus SACK options
+ * @raw: buffer
+ *
+ * Build Pseudo header in packed buffer
+ * first, calculate TCP checksum, then build IP header in packed buffer.
+ *
+ */
+union tcp_build_pkt {
+       struct pseudo_hdr ph;
+       struct ip_tcp_hdr_o ip;
+       struct ip_tcp_hdr_s sack;
+       uchar  raw[1600];
+} __packed;
+
+/**
+ * enum tcp_state - TCP State machine states for connection
+ * @TCP_CLOSED: Need to send SYN to connect
+ * @TCP_SYN_SENT: Trying to connect, waiting for SYN ACK
+ * @TCP_ESTABLISHED: both server & client have a connection
+ * @TCP_CLOSE_WAIT: Rec FIN, passed to app for FIN, ACK rsp
+ * @TCP_CLOSING: Rec FIN, sent FIN, ACK waiting for ACK
+ * @TCP_FIN_WAIT_1: Sent FIN waiting for response
+ * @TCP_FIN_WAIT_2: Rec ACK from FIN sent, waiting for FIN
+ */
+enum tcp_state {
+       TCP_CLOSED,
+       TCP_SYN_SENT,
+       TCP_ESTABLISHED,
+       TCP_CLOSE_WAIT,
+       TCP_CLOSING,
+       TCP_FIN_WAIT_1,
+       TCP_FIN_WAIT_2
+};
+
+enum tcp_state tcp_get_tcp_state(void);
+void tcp_set_tcp_state(enum tcp_state new_state);
+int tcp_set_tcp_header(uchar *pkt, int dport, int sport, int payload_len,
+                      u8 action, u32 tcp_seq_num, u32 tcp_ack_num);
+
+/**
+ * rxhand_tcp() - An incoming packet handler.
+ * @pkt: pointer to the application packet
+ * @dport: destination UDP port
+ * @sip: source IP address
+ * @sport: source UDP port
+ * @len: packet length
+ */
+typedef void rxhand_tcp(uchar *pkt, unsigned int dport,
+                       struct in_addr sip, unsigned int sport,
+                       unsigned int len);
+void tcp_set_tcp_handler(rxhand_tcp *f);
+
+void rxhand_tcp_f(union tcp_build_pkt *b, unsigned int len);
+
+u16 tcp_set_pseudo_header(uchar *pkt, struct in_addr src, struct in_addr dest,
+                         int tcp_len, int pkt_len);
index 52e261884d168b52b88b1d2b90bb96fdabc41c7b..cb600fe5eb482b59f41c564f7741d8adbf0cf7ad 100644 (file)
@@ -174,6 +174,22 @@ config BOOTP_MAX_ROOT_PATH_LEN
        help
          Select maximal length of option 17 root path.
 
+config PROT_TCP
+       bool "TCP stack"
+       help
+         Enable a generic tcp framework that allows defining a custom
+         handler for tcp protocol.
+
+config PROT_TCP_SACK
+       bool "TCP SACK support"
+       depends on PROT_TCP
+       help
+         TCP protocol with SACK. SACK means selective acknowledgements.
+         By turning this option on TCP will learn what segments are already
+         received. So that it improves TCP's retransmission efficiency.
+         This option should be turn on if you want to achieve the fastest
+         file transfer possible.
+
 endif   # if NET
 
 config SYS_RX_ETH_BUFFER
index 6c812502d3ea5bb6b9523c3fa0267ed67859c167..d131d1cb1a48d05599f9b2adbf85520e547343ae 100644 (file)
@@ -30,6 +30,7 @@ obj-$(CONFIG_CMD_TFTPBOOT) += tftp.o
 obj-$(CONFIG_UDP_FUNCTION_FASTBOOT)  += fastboot.o
 obj-$(CONFIG_CMD_WOL)  += wol.o
 obj-$(CONFIG_PROT_UDP) += udp.o
+obj-$(CONFIG_PROT_TCP) += tcp.o
 
 # Disable this warning as it is triggered by:
 # sprintf(buf, index ? "foo%d" : "foo", index)
index 6f0a48361c96e48e8080b78ddccad4d79a6b5702..9cb2aab09ded2cc28e63e6066463d4263342dbbf 100644 (file)
--- a/net/net.c
+++ b/net/net.c
 #if defined(CONFIG_CMD_WOL)
 #include "wol.h"
 #endif
+#include <net/tcp.h>
 
 /** BOOTP EXTENTIONS **/
 
@@ -387,6 +388,8 @@ int net_init(void)
 
                /* Only need to setup buffer pointers once. */
                first_call = 0;
+               if (IS_ENABLED(CONFIG_PROT_TCP))
+                       tcp_set_tcp_state(TCP_CLOSED);
        }
 
        return net_init_loop();
@@ -833,6 +836,16 @@ int net_send_udp_packet(uchar *ether, struct in_addr dest, int dport, int sport,
                                  IPPROTO_UDP, 0, 0, 0);
 }
 
+#if defined(CONFIG_PROT_TCP)
+int net_send_tcp_packet(int payload_len, int dport, int sport, u8 action,
+                       u32 tcp_seq_num, u32 tcp_ack_num)
+{
+       return net_send_ip_packet(net_server_ethaddr, net_server_ip, dport,
+                                 sport, payload_len, IPPROTO_TCP, action,
+                                 tcp_seq_num, tcp_ack_num);
+}
+#endif
+
 int net_send_ip_packet(uchar *ether, struct in_addr dest, int dport, int sport,
                       int payload_len, int proto, u8 action, u32 tcp_seq_num,
                       u32 tcp_ack_num)
@@ -864,6 +877,14 @@ int net_send_ip_packet(uchar *ether, struct in_addr dest, int dport, int sport,
                                   payload_len);
                pkt_hdr_size = eth_hdr_size + IP_UDP_HDR_SIZE;
                break;
+#if defined(CONFIG_PROT_TCP)
+       case IPPROTO_TCP:
+               pkt_hdr_size = eth_hdr_size
+                       + tcp_set_tcp_header(pkt + eth_hdr_size, dport, sport,
+                                            payload_len, action, tcp_seq_num,
+                                            tcp_ack_num);
+               break;
+#endif
        default:
                return -EINVAL;
        }
@@ -1289,6 +1310,15 @@ void net_process_received_packet(uchar *in_packet, int len)
                if (ip->ip_p == IPPROTO_ICMP) {
                        receive_icmp(ip, len, src_ip, et);
                        return;
+#if defined(CONFIG_PROT_TCP)
+               } else if (ip->ip_p == IPPROTO_TCP) {
+                       debug_cond(DEBUG_DEV_PKT,
+                                  "TCP PH (to=%pI4, from=%pI4, len=%d)\n",
+                                  &dst_ip, &src_ip, len);
+
+                       rxhand_tcp_f((union tcp_build_pkt *)ip, len);
+                       return;
+#endif
                } else if (ip->ip_p != IPPROTO_UDP) {   /* Only UDP packets */
                        return;
                }
diff --git a/net/tcp.c b/net/tcp.c
new file mode 100644 (file)
index 0000000..8d338c7
--- /dev/null
+++ b/net/tcp.c
@@ -0,0 +1,720 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright 2017 Duncan Hare, all rights reserved.
+ */
+
+/*
+ * General Desription:
+ *
+ * TCP support for the wget command, for fast file downloading.
+ *
+ * HTTP/TCP Receiver:
+ *
+ *      Prerequisites:  - own ethernet address
+ *                      - own IP address
+ *                      - Server IP address
+ *                      - Server with TCP
+ *                      - TCP application (eg wget)
+ *      Next Step       HTTPS?
+ */
+#include <common.h>
+#include <command.h>
+#include <console.h>
+#include <env_internal.h>
+#include <errno.h>
+#include <net.h>
+#include <net/tcp.h>
+
+/*
+ * TCP sliding window  control used by us to request re-TX
+ */
+static struct tcp_sack_v tcp_lost;
+
+/* TCP option timestamp */
+static u32 loc_timestamp;
+static u32 rmt_timestamp;
+
+static u32 tcp_seq_init;
+static u32 tcp_ack_edge;
+static u32 tcp_seq_max;
+
+static int tcp_activity_count;
+
+/*
+ * Search for TCP_SACK and review the comments before the code section
+ * TCP_SACK is the number of packets at the front of the stream
+ */
+
+enum pkt_state {PKT, NOPKT};
+struct sack_r {
+       struct sack_edges se;
+       enum pkt_state st;
+};
+
+static struct sack_r edge_a[TCP_SACK];
+static unsigned int sack_idx;
+static unsigned int prev_len;
+
+/*
+ * TCP lengths are stored as a rounded up number of 32 bit words.
+ * Add 3 to length round up, rounded, then divided into the
+ * length in 32 bit words.
+ */
+#define LEN_B_TO_DW(x) ((x) >> 2)
+#define ROUND_TCPHDR_LEN(x) (LEN_B_TO_DW((x) + 3))
+#define SHIFT_TO_TCPHDRLEN_FIELD(x) ((x) << 4)
+#define GET_TCP_HDR_LEN_IN_BYTES(x) ((x) >> 2)
+
+/* TCP connection state */
+static enum tcp_state current_tcp_state;
+
+/* Current TCP RX packet handler */
+static rxhand_tcp *tcp_packet_handler;
+
+/**
+ * tcp_get_tcp_state() - get current TCP state
+ *
+ * Return: Current TCP state
+ */
+enum tcp_state tcp_get_tcp_state(void)
+{
+       return current_tcp_state;
+}
+
+/**
+ * tcp_set_tcp_state() - set current TCP state
+ * @new_state: new TCP state
+ */
+void tcp_set_tcp_state(enum tcp_state new_state)
+{
+       current_tcp_state = new_state;
+}
+
+static void dummy_handler(uchar *pkt, unsigned int dport,
+                         struct in_addr sip, unsigned int sport,
+                         unsigned int len)
+{
+}
+
+/**
+ * tcp_set_tcp_handler() - set a handler to receive data
+ * @f: handler
+ */
+void tcp_set_tcp_handler(rxhand_tcp *f)
+{
+       debug_cond(DEBUG_INT_STATE, "--- net_loop TCP handler set (%p)\n", f);
+       if (!f)
+               tcp_packet_handler = dummy_handler;
+       else
+               tcp_packet_handler = f;
+}
+
+/**
+ * tcp_set_pseudo_header() - set TCP pseudo header
+ * @pkt: the packet
+ * @src: source IP address
+ * @dest: destinaion IP address
+ * @tcp_len: tcp length
+ * @pkt_len: packet length
+ *
+ * Return: the checksum of the packet
+ */
+u16 tcp_set_pseudo_header(uchar *pkt, struct in_addr src, struct in_addr dest,
+                         int tcp_len, int pkt_len)
+{
+       union tcp_build_pkt *b = (union tcp_build_pkt *)pkt;
+       int checksum_len;
+
+       /*
+        * Pseudo header
+        *
+        * Zero the byte after the last byte so that the header checksum
+        * will always work.
+        */
+       pkt[pkt_len] = 0;
+
+       net_copy_ip((void *)&b->ph.p_src, &src);
+       net_copy_ip((void *)&b->ph.p_dst, &dest);
+       b->ph.rsvd = 0;
+       b->ph.p = IPPROTO_TCP;
+       b->ph.len = htons(tcp_len);
+       checksum_len = tcp_len + PSEUDO_HDR_SIZE;
+
+       debug_cond(DEBUG_DEV_PKT,
+                  "TCP Pesudo  Header  (to=%pI4, from=%pI4, Len=%d)\n",
+                  &b->ph.p_dst, &b->ph.p_src, checksum_len);
+
+       return compute_ip_checksum(pkt + PSEUDO_PAD_SIZE, checksum_len);
+}
+
+/**
+ * net_set_ack_options() - set TCP options in acknowledge packets
+ * @b: the packet
+ *
+ * Return: TCP header length
+ */
+int net_set_ack_options(union tcp_build_pkt *b)
+{
+       b->sack.hdr.tcp_hlen = SHIFT_TO_TCPHDRLEN_FIELD(LEN_B_TO_DW(TCP_HDR_SIZE));
+
+       b->sack.t_opt.kind = TCP_O_TS;
+       b->sack.t_opt.len = TCP_OPT_LEN_A;
+       b->sack.t_opt.t_snd = htons(loc_timestamp);
+       b->sack.t_opt.t_rcv = rmt_timestamp;
+       b->sack.sack_v.kind = TCP_1_NOP;
+       b->sack.sack_v.len = 0;
+
+       if (IS_ENABLED(CONFIG_PROT_TCP_SACK)) {
+               if (tcp_lost.len > TCP_OPT_LEN_2) {
+                       debug_cond(DEBUG_DEV_PKT, "TCP ack opt lost.len %x\n",
+                                  tcp_lost.len);
+                       b->sack.sack_v.len = tcp_lost.len;
+                       b->sack.sack_v.kind = TCP_V_SACK;
+                       b->sack.sack_v.hill[0].l = htonl(tcp_lost.hill[0].l);
+                       b->sack.sack_v.hill[0].r = htonl(tcp_lost.hill[0].r);
+
+                       /*
+                        * These SACK structures are initialized with NOPs to
+                        * provide TCP header alignment padding. There are 4
+                        * SACK structures used for both header padding and
+                        * internally.
+                        */
+                       b->sack.sack_v.hill[1].l = htonl(tcp_lost.hill[1].l);
+                       b->sack.sack_v.hill[1].r = htonl(tcp_lost.hill[1].r);
+                       b->sack.sack_v.hill[2].l = htonl(tcp_lost.hill[2].l);
+                       b->sack.sack_v.hill[2].r = htonl(tcp_lost.hill[2].r);
+                       b->sack.sack_v.hill[3].l = TCP_O_NOP;
+                       b->sack.sack_v.hill[3].r = TCP_O_NOP;
+               }
+
+               b->sack.hdr.tcp_hlen = SHIFT_TO_TCPHDRLEN_FIELD(ROUND_TCPHDR_LEN(TCP_HDR_SIZE +
+                                                                                TCP_TSOPT_SIZE +
+                                                                                tcp_lost.len));
+       } else {
+               b->sack.sack_v.kind = 0;
+               b->sack.hdr.tcp_hlen = SHIFT_TO_TCPHDRLEN_FIELD(ROUND_TCPHDR_LEN(TCP_HDR_SIZE +
+                                                                                TCP_TSOPT_SIZE));
+       }
+
+       /*
+        * This returns the actual rounded up length of the
+        * TCP header to add to the total packet length
+        */
+
+       return GET_TCP_HDR_LEN_IN_BYTES(b->sack.hdr.tcp_hlen);
+}
+
+/**
+ * net_set_ack_options() - set TCP options in SYN packets
+ * @b: the packet
+ */
+void net_set_syn_options(union tcp_build_pkt *b)
+{
+       if (IS_ENABLED(CONFIG_PROT_TCP_SACK))
+               tcp_lost.len = 0;
+
+       b->ip.hdr.tcp_hlen = 0xa0;
+
+       b->ip.mss.kind = TCP_O_MSS;
+       b->ip.mss.len = TCP_OPT_LEN_4;
+       b->ip.mss.mss = htons(TCP_MSS);
+       b->ip.scale.kind = TCP_O_SCL;
+       b->ip.scale.scale = TCP_SCALE;
+       b->ip.scale.len = TCP_OPT_LEN_3;
+       if (IS_ENABLED(CONFIG_PROT_TCP_SACK)) {
+               b->ip.sack_p.kind = TCP_P_SACK;
+               b->ip.sack_p.len = TCP_OPT_LEN_2;
+       } else {
+               b->ip.sack_p.kind = TCP_1_NOP;
+               b->ip.sack_p.len = TCP_1_NOP;
+       }
+       b->ip.t_opt.kind = TCP_O_TS;
+       b->ip.t_opt.len = TCP_OPT_LEN_A;
+       loc_timestamp = get_ticks();
+       rmt_timestamp = 0;
+       b->ip.t_opt.t_snd = 0;
+       b->ip.t_opt.t_rcv = 0;
+       b->ip.end = TCP_O_END;
+}
+
+int tcp_set_tcp_header(uchar *pkt, int dport, int sport, int payload_len,
+                      u8 action, u32 tcp_seq_num, u32 tcp_ack_num)
+{
+       union tcp_build_pkt *b = (union tcp_build_pkt *)pkt;
+       int pkt_hdr_len;
+       int pkt_len;
+       int tcp_len;
+
+       /*
+        * Header: 5 32 bit words. 4 bits TCP header Length,
+        *         4 bits reserved options
+        */
+       b->ip.hdr.tcp_flags = action;
+       pkt_hdr_len = IP_TCP_HDR_SIZE;
+       b->ip.hdr.tcp_hlen = SHIFT_TO_TCPHDRLEN_FIELD(LEN_B_TO_DW(TCP_HDR_SIZE));
+
+       switch (action) {
+       case TCP_SYN:
+               debug_cond(DEBUG_DEV_PKT,
+                          "TCP Hdr:SYN (%pI4, %pI4, sq=%d, ak=%d)\n",
+                          &net_server_ip, &net_ip,
+                          tcp_seq_num, tcp_ack_num);
+               tcp_activity_count = 0;
+               net_set_syn_options(b);
+               tcp_seq_num = 0;
+               tcp_ack_num = 0;
+               pkt_hdr_len = IP_TCP_O_SIZE;
+               if (current_tcp_state == TCP_SYN_SENT) {  /* Too many SYNs */
+                       action = TCP_FIN;
+                       current_tcp_state = TCP_FIN_WAIT_1;
+               } else {
+                       current_tcp_state = TCP_SYN_SENT;
+               }
+               break;
+       case TCP_ACK:
+               pkt_hdr_len = IP_HDR_SIZE + net_set_ack_options(b);
+               b->ip.hdr.tcp_flags = action;
+               debug_cond(DEBUG_DEV_PKT,
+                          "TCP Hdr:ACK (%pI4, %pI4, s=%d, a=%d, A=%x)\n",
+                          &net_server_ip, &net_ip, tcp_seq_num, tcp_ack_num,
+                          action);
+               break;
+       case TCP_FIN:
+               debug_cond(DEBUG_DEV_PKT,
+                          "TCP Hdr:FIN  (%pI4, %pI4, s=%d, a=%d)\n",
+                          &net_server_ip, &net_ip, tcp_seq_num, tcp_ack_num);
+               payload_len = 0;
+               pkt_hdr_len = IP_TCP_HDR_SIZE;
+               current_tcp_state = TCP_FIN_WAIT_1;
+               break;
+
+       /* Notify connection closing */
+
+       case (TCP_FIN | TCP_ACK):
+       case (TCP_FIN | TCP_ACK | TCP_PUSH):
+               if (current_tcp_state == TCP_CLOSE_WAIT)
+                       current_tcp_state = TCP_CLOSING;
+
+               tcp_ack_edge++;
+               debug_cond(DEBUG_DEV_PKT,
+                          "TCP Hdr:FIN ACK PSH(%pI4, %pI4, s=%d, a=%d, A=%x)\n",
+                          &net_server_ip, &net_ip,
+                          tcp_seq_num, tcp_ack_edge, action);
+               fallthrough;
+       default:
+               pkt_hdr_len = IP_HDR_SIZE + net_set_ack_options(b);
+               b->ip.hdr.tcp_flags = action | TCP_PUSH | TCP_ACK;
+               debug_cond(DEBUG_DEV_PKT,
+                          "TCP Hdr:dft  (%pI4, %pI4, s=%d, a=%d, A=%x)\n",
+                          &net_server_ip, &net_ip,
+                          tcp_seq_num, tcp_ack_num, action);
+       }
+
+       pkt_len = pkt_hdr_len + payload_len;
+       tcp_len = pkt_len - IP_HDR_SIZE;
+
+       /* TCP Header */
+       b->ip.hdr.tcp_ack = htonl(tcp_ack_edge);
+       b->ip.hdr.tcp_src = htons(sport);
+       b->ip.hdr.tcp_dst = htons(dport);
+       b->ip.hdr.tcp_seq = htonl(tcp_seq_num);
+       tcp_seq_num = tcp_seq_num + payload_len;
+
+       /*
+        * TCP window size - TCP header variable tcp_win.
+        * Change tcp_win only if you have an understanding of network
+        * overrun, congestion, TCP segment sizes, TCP windows, TCP scale,
+        * queuing theory  and packet buffering. If there are too few buffers,
+        * there will be data loss, recovery may work or the sending TCP,
+        * the server, could abort the stream transmission.
+        * MSS is governed by maximum Ethernet frame length.
+        * The number of buffers is governed by the desire to have a queue of
+        * full buffers to be processed at the destination to maximize
+        * throughput. Temporary memory use for the boot phase on modern
+        * SOCs is may not be considered a constraint to buffer space, if
+        * it is, then the u-boot tftp or nfs kernel netboot should be
+        * considered.
+        */
+       b->ip.hdr.tcp_win = htons(PKTBUFSRX * TCP_MSS >> TCP_SCALE);
+
+       b->ip.hdr.tcp_xsum = 0;
+       b->ip.hdr.tcp_ugr = 0;
+
+       b->ip.hdr.tcp_xsum = tcp_set_pseudo_header(pkt, net_ip, net_server_ip,
+                                                  tcp_len, pkt_len);
+
+       net_set_ip_header((uchar *)&b->ip, net_server_ip, net_ip,
+                         pkt_len, IPPROTO_TCP);
+
+       return pkt_hdr_len;
+}
+
+/**
+ * tcp_hole() - Selective Acknowledgment (Essential for fast stream transfer)
+ * @tcp_seq_num: TCP sequence start number
+ * @len: the length of sequence numbers
+ * @tcp_seq_max: maximum of sequence numbers
+ */
+void tcp_hole(u32 tcp_seq_num, u32 len, u32 tcp_seq_max)
+{
+       u32 idx_sack, sack_in;
+       u32 sack_end = TCP_SACK - 1;
+       u32 hill = 0;
+       enum pkt_state expect = PKT;
+       u32 seq = tcp_seq_num - tcp_seq_init;
+       u32 hol_l = tcp_ack_edge - tcp_seq_init;
+       u32 hol_r = 0;
+
+       /* Place new seq number in correct place in receive array */
+       if (prev_len == 0)
+               prev_len = len;
+
+       idx_sack = sack_idx + ((tcp_seq_num - tcp_ack_edge) / prev_len);
+       if (idx_sack < TCP_SACK) {
+               edge_a[idx_sack].se.l = tcp_seq_num;
+               edge_a[idx_sack].se.r = tcp_seq_num + len;
+               edge_a[idx_sack].st = PKT;
+
+               /*
+                * The fin (last) packet is not the same length as data
+                * packets, and if it's length is recorded and used for
+                *  array index calculation, calculation breaks.
+                */
+               if (prev_len < len)
+                       prev_len = len;
+       }
+
+       debug_cond(DEBUG_DEV_PKT,
+                  "TCP 1 seq %d, edg %d, len %d, sack_idx %d, sack_end %d\n",
+                   seq, hol_l, len, sack_idx, sack_end);
+
+       /* Right edge of contiguous stream, is the left edge of first hill */
+       hol_l = tcp_seq_num - tcp_seq_init;
+       hol_r = hol_l + len;
+
+       if (IS_ENABLED(CONFIG_PROT_TCP_SACK))
+               tcp_lost.len = TCP_OPT_LEN_2;
+
+       debug_cond(DEBUG_DEV_PKT,
+                  "TCP 1 in %d, seq %d, pkt_l %d, pkt_r %d, sack_idx %d, sack_end %d\n",
+                  idx_sack, seq, hol_l, hol_r, sack_idx, sack_end);
+
+       for (sack_in = sack_idx; sack_in < sack_end && hill < TCP_SACK_HILLS;
+            sack_in++)  {
+               switch (expect) {
+               case NOPKT:
+                       switch (edge_a[sack_in].st) {
+                       case NOPKT:
+                               debug_cond(DEBUG_INT_STATE, "N");
+                               break;
+                       case PKT:
+                               debug_cond(DEBUG_INT_STATE, "n");
+                               if (IS_ENABLED(CONFIG_PROT_TCP_SACK)) {
+                                       tcp_lost.hill[hill].l =
+                                               edge_a[sack_in].se.l;
+                                       tcp_lost.hill[hill].r =
+                                               edge_a[sack_in].se.r;
+                               }
+                               expect = PKT;
+                               break;
+                       }
+                       break;
+               case PKT:
+                       switch (edge_a[sack_in].st) {
+                       case NOPKT:
+                               debug_cond(DEBUG_INT_STATE, "p");
+                               if (sack_in > sack_idx &&
+                                   hill < TCP_SACK_HILLS) {
+                                       hill++;
+                                       if (IS_ENABLED(CONFIG_PROT_TCP_SACK))
+                                               tcp_lost.len += TCP_OPT_LEN_8;
+                               }
+                               expect = NOPKT;
+                               break;
+                       case PKT:
+                               debug_cond(DEBUG_INT_STATE, "P");
+
+                               if (tcp_ack_edge == edge_a[sack_in].se.l) {
+                                       tcp_ack_edge = edge_a[sack_in].se.r;
+                                       edge_a[sack_in].st = NOPKT;
+                                       sack_idx++;
+                               } else {
+                                       if (IS_ENABLED(CONFIG_PROT_TCP_SACK) &&
+                                           hill < TCP_SACK_HILLS)
+                                               tcp_lost.hill[hill].r =
+                                                       edge_a[sack_in].se.r;
+                               if (IS_ENABLED(CONFIG_PROT_TCP_SACK) &&
+                                   sack_in == sack_end - 1)
+                                       tcp_lost.hill[hill].r =
+                                               edge_a[sack_in].se.r;
+                               }
+                               break;
+                       }
+                       break;
+               }
+       }
+       debug_cond(DEBUG_INT_STATE, "\n");
+       if (!IS_ENABLED(CONFIG_PROT_TCP_SACK) || tcp_lost.len <= TCP_OPT_LEN_2)
+               sack_idx = 0;
+}
+
+/**
+ * tcp_parse_options() - parsing TCP options
+ * @o: pointer to the option field.
+ * @o_len: length of the option field.
+ */
+void tcp_parse_options(uchar *o, int o_len)
+{
+       struct tcp_t_opt  *tsopt;
+       uchar *p = o;
+
+       /*
+        * NOPs are options with a zero length, and thus are special.
+        * All other options have length fields.
+        */
+       for (p = o; p < (o + o_len); p = p + p[1]) {
+               if (!p[1])
+                       return; /* Finished processing options */
+
+               switch (p[0]) {
+               case TCP_O_END:
+                       return;
+               case TCP_O_MSS:
+               case TCP_O_SCL:
+               case TCP_P_SACK:
+               case TCP_V_SACK:
+                       break;
+               case TCP_O_TS:
+                       tsopt = (struct tcp_t_opt *)p;
+                       rmt_timestamp = tsopt->t_snd;
+                       return;
+               }
+
+               /* Process optional NOPs */
+               if (p[0] == TCP_O_NOP)
+                       p++;
+       }
+}
+
+static u8 tcp_state_machine(u8 tcp_flags, u32 *tcp_seq_num, int payload_len)
+{
+       u8 tcp_fin = tcp_flags & TCP_FIN;
+       u8 tcp_syn = tcp_flags & TCP_SYN;
+       u8 tcp_rst = tcp_flags & TCP_RST;
+       u8 tcp_push = tcp_flags & TCP_PUSH;
+       u8 tcp_ack = tcp_flags & TCP_ACK;
+       u8 action = TCP_DATA;
+       int i;
+
+       /*
+        * tcp_flags are examined to determine TX action in a given state
+        * tcp_push is interpreted to mean "inform the app"
+        * urg, ece, cer and nonce flags are not supported.
+        *
+        * exe and crw are use to signal and confirm knowledge of congestion.
+        * This TCP only sends a file request and acks. If it generates
+        * congestion, the network is broken.
+        */
+       debug_cond(DEBUG_INT_STATE, "TCP STATE ENTRY %x\n", action);
+       if (tcp_rst) {
+               action = TCP_DATA;
+               current_tcp_state = TCP_CLOSED;
+               net_set_state(NETLOOP_FAIL);
+               debug_cond(DEBUG_INT_STATE, "TCP Reset %x\n", tcp_flags);
+               return TCP_RST;
+       }
+
+       switch  (current_tcp_state) {
+       case TCP_CLOSED:
+               debug_cond(DEBUG_INT_STATE, "TCP CLOSED %x\n", tcp_flags);
+               if (tcp_ack)
+                       action = TCP_DATA;
+               else if (tcp_syn)
+                       action = TCP_RST;
+               else if (tcp_fin)
+                       action = TCP_DATA;
+               break;
+       case TCP_SYN_SENT:
+               debug_cond(DEBUG_INT_STATE, "TCP_SYN_SENT %x, %d\n",
+                          tcp_flags, *tcp_seq_num);
+               if (tcp_fin) {
+                       action = action | TCP_PUSH;
+                       current_tcp_state = TCP_CLOSE_WAIT;
+               }
+               if (tcp_syn) {
+                       action = action | TCP_ACK | TCP_PUSH;
+                       if (tcp_ack) {
+                               tcp_seq_init = *tcp_seq_num;
+                               *tcp_seq_num = *tcp_seq_num + 1;
+                               tcp_seq_max = *tcp_seq_num;
+                               tcp_ack_edge = *tcp_seq_num;
+                               sack_idx = 0;
+                               edge_a[sack_idx].se.l = *tcp_seq_num;
+                               edge_a[sack_idx].se.r = *tcp_seq_num;
+                               prev_len = 0;
+                               current_tcp_state = TCP_ESTABLISHED;
+                               for (i = 0; i < TCP_SACK; i++)
+                                       edge_a[i].st = NOPKT;
+                       }
+               } else if (tcp_ack) {
+                       action = TCP_DATA;
+               }
+
+               break;
+       case TCP_ESTABLISHED:
+               debug_cond(DEBUG_INT_STATE, "TCP_ESTABLISHED %x\n", tcp_flags);
+               if (*tcp_seq_num > tcp_seq_max)
+                       tcp_seq_max = *tcp_seq_num;
+               if (payload_len > 0) {
+                       tcp_hole(*tcp_seq_num, payload_len, tcp_seq_max);
+                       tcp_fin = TCP_DATA;  /* cause standalone FIN */
+               }
+
+               if ((tcp_fin) &&
+                   (!IS_ENABLED(CONFIG_PROT_TCP_SACK) ||
+                    tcp_lost.len <= TCP_OPT_LEN_2)) {
+                       action = action | TCP_FIN | TCP_PUSH | TCP_ACK;
+                       current_tcp_state = TCP_CLOSE_WAIT;
+               } else if (tcp_ack) {
+                       action = TCP_DATA;
+               }
+
+               if (tcp_syn)
+                       action = TCP_ACK + TCP_RST;
+               else if (tcp_push)
+                       action = action | TCP_PUSH;
+               break;
+       case TCP_CLOSE_WAIT:
+               debug_cond(DEBUG_INT_STATE, "TCP_CLOSE_WAIT (%x)\n", tcp_flags);
+               action = TCP_DATA;
+               break;
+       case TCP_FIN_WAIT_2:
+               debug_cond(DEBUG_INT_STATE, "TCP_FIN_WAIT_2 (%x)\n", tcp_flags);
+               if (tcp_ack) {
+                       action = TCP_PUSH | TCP_ACK;
+                       current_tcp_state = TCP_CLOSED;
+                       puts("\n");
+               } else if (tcp_syn) {
+                       action = TCP_DATA;
+               } else if (tcp_fin) {
+                       action = TCP_DATA;
+               }
+               break;
+       case TCP_FIN_WAIT_1:
+               debug_cond(DEBUG_INT_STATE, "TCP_FIN_WAIT_1 (%x)\n", tcp_flags);
+               if (tcp_fin) {
+                       action = TCP_ACK | TCP_FIN;
+                       current_tcp_state = TCP_FIN_WAIT_2;
+               }
+               if (tcp_syn)
+                       action = TCP_RST;
+               if (tcp_ack) {
+                       current_tcp_state = TCP_CLOSED;
+                       tcp_seq_num = tcp_seq_num + 1;
+               }
+               break;
+       case TCP_CLOSING:
+               debug_cond(DEBUG_INT_STATE, "TCP_CLOSING (%x)\n", tcp_flags);
+               if (tcp_ack) {
+                       action = TCP_PUSH;
+                       current_tcp_state = TCP_CLOSED;
+                       puts("\n");
+               } else if (tcp_syn) {
+                       action = TCP_RST;
+               } else if (tcp_fin) {
+                       action = TCP_DATA;
+               }
+               break;
+       }
+       return action;
+}
+
+/**
+ * rxhand_tcp_f() - process receiving data and call data handler.
+ * @b: the packet
+ * @pkt_len: the length of packet.
+ */
+void rxhand_tcp_f(union tcp_build_pkt *b, unsigned int pkt_len)
+{
+       int tcp_len = pkt_len - IP_HDR_SIZE;
+       u16 tcp_rx_xsum = b->ip.hdr.ip_sum;
+       u8  tcp_action = TCP_DATA;
+       u32 tcp_seq_num, tcp_ack_num;
+       struct in_addr action_and_state;
+       int tcp_hdr_len, payload_len;
+
+       /* Verify IP header */
+       debug_cond(DEBUG_DEV_PKT,
+                  "TCP RX in RX Sum (to=%pI4, from=%pI4, len=%d)\n",
+                  &b->ip.hdr.ip_src, &b->ip.hdr.ip_dst, pkt_len);
+
+       b->ip.hdr.ip_src = net_server_ip;
+       b->ip.hdr.ip_dst = net_ip;
+       b->ip.hdr.ip_sum = 0;
+       if (tcp_rx_xsum != compute_ip_checksum(b, IP_HDR_SIZE)) {
+               debug_cond(DEBUG_DEV_PKT,
+                          "TCP RX IP xSum Error (%pI4, =%pI4, len=%d)\n",
+                          &net_ip, &net_server_ip, pkt_len);
+               return;
+       }
+
+       /* Build pseudo header and verify TCP header */
+       tcp_rx_xsum = b->ip.hdr.tcp_xsum;
+       b->ip.hdr.tcp_xsum = 0;
+       if (tcp_rx_xsum != tcp_set_pseudo_header((uchar *)b, b->ip.hdr.ip_src,
+                                                b->ip.hdr.ip_dst, tcp_len,
+                                                pkt_len)) {
+               debug_cond(DEBUG_DEV_PKT,
+                          "TCP RX TCP xSum Error (%pI4, %pI4, len=%d)\n",
+                          &net_ip, &net_server_ip, tcp_len);
+               return;
+       }
+
+       tcp_hdr_len = GET_TCP_HDR_LEN_IN_BYTES(b->ip.hdr.tcp_hlen);
+       payload_len = tcp_len - tcp_hdr_len;
+
+       if (tcp_hdr_len > TCP_HDR_SIZE)
+               tcp_parse_options((uchar *)b + IP_TCP_HDR_SIZE,
+                                 tcp_hdr_len - TCP_HDR_SIZE);
+       /*
+        * Incoming sequence and ack numbers are server's view of the numbers.
+        * The app must swap the numbers when responding.
+        */
+       tcp_seq_num = ntohl(b->ip.hdr.tcp_seq);
+       tcp_ack_num = ntohl(b->ip.hdr.tcp_ack);
+
+       /* Packets are not ordered. Send to app as received. */
+       tcp_action = tcp_state_machine(b->ip.hdr.tcp_flags,
+                                      &tcp_seq_num, payload_len);
+
+       tcp_activity_count++;
+       if (tcp_activity_count > TCP_ACTIVITY) {
+               puts("| ");
+               tcp_activity_count = 0;
+       }
+
+       if ((tcp_action & TCP_PUSH) || payload_len > 0) {
+               debug_cond(DEBUG_DEV_PKT,
+                          "TCP Notify (action=%x, Seq=%d,Ack=%d,Pay%d)\n",
+                          tcp_action, tcp_seq_num, tcp_ack_num, payload_len);
+
+               action_and_state.s_addr = tcp_action;
+               (*tcp_packet_handler) ((uchar *)b + pkt_len - payload_len,
+                                      tcp_seq_num, action_and_state,
+                                      tcp_ack_num, payload_len);
+
+       } else if (tcp_action != TCP_DATA) {
+               debug_cond(DEBUG_DEV_PKT,
+                          "TCP Action (action=%x,Seq=%d,Ack=%d,Pay=%d)\n",
+                          tcp_action, tcp_seq_num, tcp_ack_num, payload_len);
+
+               /*
+                * Warning: Incoming Ack & Seq sequence numbers are transposed
+                * here to outgoing Seq & Ack sequence numbers
+                */
+               net_send_tcp_packet(0, ntohs(b->ip.hdr.tcp_src),
+                                   ntohs(b->ip.hdr.tcp_dst),
+                                   (tcp_action & (~TCP_PUSH)),
+                                   tcp_seq_num, tcp_ack_num);
+       }
+}