tcp_out.c

Go to the documentation of this file.

/*
 * Copyright (c) 2008-2014, Pedigree Developers
 *
 * Please see the CONTRIB file in the root of the source tree for a full
 * list of contributors.
 *
 * Permission to use, copy, modify, and distribute this software for any
 * purpose with or without fee is hereby granted, provided that the above
 * copyright notice and this permission notice appear in all copies.
 *
 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 */

/*
 * Copyright (c) 2001-2004 Swedish Institute of Computer Science.
 * 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.
 *
 * This file is part of the lwIP TCP/IP stack.
 *
 * Author: Adam Dunkels <adam@sics.se>
 *
 */

#include "lwip/opt.h"

#if LWIP_TCP /* don't build if not configured for use in lwipopts.h */

#include "lwip/priv/tcp_priv.h"
#include "lwip/def.h"
#include "lwip/mem.h"
#include "lwip/memp.h"
#include "lwip/ip_addr.h"
#include "lwip/netif.h"
#include "lwip/inet_chksum.h"
#include "lwip/stats.h"
#include "lwip/ip6.h"
#include "lwip/ip6_addr.h"
#if LWIP_TCP_TIMESTAMPS
#include "lwip/sys.h"
#endif

#include <string.h>

/* Define some copy-macros for checksum-on-copy so that the code looks
   nicer by preventing too many ifdef's. */
#if TCP_CHECKSUM_ON_COPY
#define TCP_DATA_COPY(dst, src, len, seg) do { \
  tcp_seg_add_chksum(LWIP_CHKSUM_COPY(dst, src, len), \
                     len, &seg->chksum, &seg->chksum_swapped); \
  seg->flags |= TF_SEG_DATA_CHECKSUMMED; } while(0)
#define TCP_DATA_COPY2(dst, src, len, chksum, chksum_swapped)  \
  tcp_seg_add_chksum(LWIP_CHKSUM_COPY(dst, src, len), len, chksum, chksum_swapped);
#else /* TCP_CHECKSUM_ON_COPY*/
#define TCP_DATA_COPY(dst, src, len, seg)                     MEMCPY(dst, src, len)
#define TCP_DATA_COPY2(dst, src, len, chksum, chksum_swapped) MEMCPY(dst, src, len)
#endif /* TCP_CHECKSUM_ON_COPY*/

#ifndef TCP_CHECKSUM_ON_COPY_SANITY_CHECK
#define TCP_CHECKSUM_ON_COPY_SANITY_CHECK   0
#endif
/* Allow to override the failure of sanity check from warning to e.g. hard failure */
#if TCP_CHECKSUM_ON_COPY_SANITY_CHECK
#ifndef TCP_CHECKSUM_ON_COPY_SANITY_CHECK_FAIL
#define TCP_CHECKSUM_ON_COPY_SANITY_CHECK_FAIL(msg) LWIP_DEBUGF(TCP_DEBUG | LWIP_DBG_LEVEL_WARNING, msg)
#endif
#endif

#if TCP_OVERSIZE

#ifndef TCP_OVERSIZE_CALC_LENGTH
#define TCP_OVERSIZE_CALC_LENGTH(length) ((length) + TCP_OVERSIZE)
#endif
#endif

/* Forward declarations.*/
static err_t tcp_output_segment(struct tcp_seg *seg, struct tcp_pcb *pcb, struct netif *netif);

static struct pbuf *
tcp_output_alloc_header(struct tcp_pcb *pcb, u16_t optlen, u16_t datalen,
                      u32_t seqno_be /* already in network byte order */)
{
  struct tcp_hdr *tcphdr;
  struct pbuf *p = pbuf_alloc(PBUF_IP, TCP_HLEN + optlen + datalen, PBUF_RAM);
  if (p != NULL) {
    LWIP_ASSERT("check that first pbuf can hold struct tcp_hdr",
                 (p->len >= TCP_HLEN + optlen));
    tcphdr = (struct tcp_hdr *)p->payload;
    tcphdr->src = lwip_htons(pcb->local_port);
    tcphdr->dest = lwip_htons(pcb->remote_port);
    tcphdr->seqno = seqno_be;
    tcphdr->ackno = lwip_htonl(pcb->rcv_nxt);
    TCPH_HDRLEN_FLAGS_SET(tcphdr, (5 + optlen / 4), TCP_ACK);
    tcphdr->wnd = lwip_htons(TCPWND_MIN16(RCV_WND_SCALE(pcb, pcb->rcv_ann_wnd)));
    tcphdr->chksum = 0;
    tcphdr->urgp = 0;

    /* If we're sending a packet, update the announced right window edge */
    pcb->rcv_ann_right_edge = pcb->rcv_nxt + pcb->rcv_ann_wnd;
  }
  return p;
}

err_t
tcp_send_fin(struct tcp_pcb *pcb)
{
  /* first, try to add the fin to the last unsent segment */
  if (pcb->unsent != NULL) {
    struct tcp_seg *last_unsent;
    for (last_unsent = pcb->unsent; last_unsent->next != NULL;
         last_unsent = last_unsent->next);

    if ((TCPH_FLAGS(last_unsent->tcphdr) & (TCP_SYN | TCP_FIN | TCP_RST)) == 0) {
      /* no SYN/FIN/RST flag in the header, we can add the FIN flag */
      TCPH_SET_FLAG(last_unsent->tcphdr, TCP_FIN);
      pcb->flags |= TF_FIN;
      return ERR_OK;
    }
  }
  /* no data, no length, flags, copy=1, no optdata */
  return tcp_enqueue_flags(pcb, TCP_FIN);
}

static struct tcp_seg *
tcp_create_segment(struct tcp_pcb *pcb, struct pbuf *p, u8_t flags, u32_t seqno, u8_t optflags)
{
  struct tcp_seg *seg;
  u8_t optlen = LWIP_TCP_OPT_LENGTH(optflags);

  if ((seg = (struct tcp_seg *)memp_malloc(MEMP_TCP_SEG)) == NULL) {
    LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SERIOUS, ("tcp_create_segment: no memory.\n"));
    pbuf_free(p);
    return NULL;
  }
  seg->flags = optflags;
  seg->next = NULL;
  seg->p = p;
  LWIP_ASSERT("p->tot_len >= optlen", p->tot_len >= optlen);
  seg->len = p->tot_len - optlen;
#if TCP_OVERSIZE_DBGCHECK
  seg->oversize_left = 0;
#endif /* TCP_OVERSIZE_DBGCHECK */
#if TCP_CHECKSUM_ON_COPY
  seg->chksum = 0;
  seg->chksum_swapped = 0;
  /* check optflags */
  LWIP_ASSERT("invalid optflags passed: TF_SEG_DATA_CHECKSUMMED",
              (optflags & TF_SEG_DATA_CHECKSUMMED) == 0);
#endif /* TCP_CHECKSUM_ON_COPY */

  /* build TCP header */
  if (pbuf_header(p, TCP_HLEN)) {
    LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SERIOUS, ("tcp_create_segment: no room for TCP header in pbuf.\n"));
    TCP_STATS_INC(tcp.err);
    tcp_seg_free(seg);
    return NULL;
  }
  seg->tcphdr = (struct tcp_hdr *)seg->p->payload;
  seg->tcphdr->src = lwip_htons(pcb->local_port);
  seg->tcphdr->dest = lwip_htons(pcb->remote_port);
  seg->tcphdr->seqno = lwip_htonl(seqno);
  /* ackno is set in tcp_output */
  TCPH_HDRLEN_FLAGS_SET(seg->tcphdr, (5 + optlen / 4), flags);
  /* wnd and chksum are set in tcp_output */
  seg->tcphdr->urgp = 0;
  return seg;
}

#if TCP_OVERSIZE
static struct pbuf *
tcp_pbuf_prealloc(pbuf_layer layer, u16_t length, u16_t max_length,
                  u16_t *oversize, struct tcp_pcb *pcb, u8_t apiflags,
                  u8_t first_seg)
{
  struct pbuf *p;
  u16_t alloc = length;

#if LWIP_NETIF_TX_SINGLE_PBUF
  LWIP_UNUSED_ARG(max_length);
  LWIP_UNUSED_ARG(pcb);
  LWIP_UNUSED_ARG(apiflags);
  LWIP_UNUSED_ARG(first_seg);
  alloc = max_length;
#else /* LWIP_NETIF_TX_SINGLE_PBUF */
  if (length < max_length) {
    /* Should we allocate an oversized pbuf, or just the minimum
     * length required? If tcp_write is going to be called again
     * before this segment is transmitted, we want the oversized
     * buffer. If the segment will be transmitted immediately, we can
     * save memory by allocating only length. We use a simple
     * heuristic based on the following information:
     *
     * Did the user set TCP_WRITE_FLAG_MORE?
     *
     * Will the Nagle algorithm defer transmission of this segment?
     */
    if ((apiflags & TCP_WRITE_FLAG_MORE) ||
        (!(pcb->flags & TF_NODELAY) &&
         (!first_seg ||
          pcb->unsent != NULL ||
          pcb->unacked != NULL))) {
      alloc = LWIP_MIN(max_length, LWIP_MEM_ALIGN_SIZE(TCP_OVERSIZE_CALC_LENGTH(length)));
    }
  }
#endif /* LWIP_NETIF_TX_SINGLE_PBUF */
  p = pbuf_alloc(layer, alloc, PBUF_RAM);
  if (p == NULL) {
    return NULL;
  }
  LWIP_ASSERT("need unchained pbuf", p->next == NULL);
  *oversize = p->len - length;
  /* trim p->len to the currently used size */
  p->len = p->tot_len = length;
  return p;
}
#else /* TCP_OVERSIZE */
#define tcp_pbuf_prealloc(layer, length, mx, os, pcb, api, fst) pbuf_alloc((layer), (length), PBUF_RAM)
#endif /* TCP_OVERSIZE */

#if TCP_CHECKSUM_ON_COPY

static void
tcp_seg_add_chksum(u16_t chksum, u16_t len, u16_t *seg_chksum,
                   u8_t *seg_chksum_swapped)
{
  u32_t helper;
  /* add chksum to old chksum and fold to u16_t */
  helper = chksum + *seg_chksum;
  chksum = FOLD_U32T(helper);
  if ((len & 1) != 0) {
    *seg_chksum_swapped = 1 - *seg_chksum_swapped;
    chksum = SWAP_BYTES_IN_WORD(chksum);
  }
  *seg_chksum = chksum;
}
#endif /* TCP_CHECKSUM_ON_COPY */

static err_t
tcp_write_checks(struct tcp_pcb *pcb, u16_t len)
{
  /* connection is in invalid state for data transmission? */
  if ((pcb->state != ESTABLISHED) &&
      (pcb->state != CLOSE_WAIT) &&
      (pcb->state != SYN_SENT) &&
      (pcb->state != SYN_RCVD)) {
    LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_STATE | LWIP_DBG_LEVEL_SEVERE, ("tcp_write() called in invalid state\n"));
    return ERR_CONN;
  } else if (len == 0) {
    return ERR_OK;
  }

  /* fail on too much data */
  if (len > pcb->snd_buf) {
    LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SEVERE, ("tcp_write: too much data (len=%"U16_F" > snd_buf=%"TCPWNDSIZE_F")\n",
      len, pcb->snd_buf));
    pcb->flags |= TF_NAGLEMEMERR;
    return ERR_MEM;
  }

  LWIP_DEBUGF(TCP_QLEN_DEBUG, ("tcp_write: queuelen: %"TCPWNDSIZE_F"\n", (tcpwnd_size_t)pcb->snd_queuelen));

  /* If total number of pbufs on the unsent/unacked queues exceeds the
   * configured maximum, return an error */
  /* check for configured max queuelen and possible overflow */
  if ((pcb->snd_queuelen >= TCP_SND_QUEUELEN) || (pcb->snd_queuelen > TCP_SNDQUEUELEN_OVERFLOW)) {
    LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SEVERE, ("tcp_write: too long queue %"U16_F" (max %"U16_F")\n",
      pcb->snd_queuelen, (u16_t)TCP_SND_QUEUELEN));
    TCP_STATS_INC(tcp.memerr);
    pcb->flags |= TF_NAGLEMEMERR;
    return ERR_MEM;
  }
  if (pcb->snd_queuelen != 0) {
    LWIP_ASSERT("tcp_write: pbufs on queue => at least one queue non-empty",
      pcb->unacked != NULL || pcb->unsent != NULL);
  } else {
    LWIP_ASSERT("tcp_write: no pbufs on queue => both queues empty",
      pcb->unacked == NULL && pcb->unsent == NULL);
  }
  return ERR_OK;
}

err_t
tcp_write(struct tcp_pcb *pcb, const void *arg, u16_t len, u8_t apiflags)
{
  struct pbuf *concat_p = NULL;
  struct tcp_seg *last_unsent = NULL, *seg = NULL, *prev_seg = NULL, *queue = NULL;
  u16_t pos = 0; /* position in 'arg' data */
  u16_t queuelen;
  u8_t optlen = 0;
  u8_t optflags = 0;
#if TCP_OVERSIZE
  u16_t oversize = 0;
  u16_t oversize_used = 0;
#if TCP_OVERSIZE_DBGCHECK
  u16_t oversize_add = 0;
#endif /* TCP_OVERSIZE_DBGCHECK*/
#endif /* TCP_OVERSIZE */
  u16_t extendlen = 0;
#if TCP_CHECKSUM_ON_COPY
  u16_t concat_chksum = 0;
  u8_t concat_chksum_swapped = 0;
  u16_t concat_chksummed = 0;
#endif /* TCP_CHECKSUM_ON_COPY */
  err_t err;
  /* don't allocate segments bigger than half the maximum window we ever received */
  u16_t mss_local = LWIP_MIN(pcb->mss, TCPWND_MIN16(pcb->snd_wnd_max/2));
  mss_local = mss_local ? mss_local : pcb->mss;

#if LWIP_NETIF_TX_SINGLE_PBUF
  /* Always copy to try to create single pbufs for TX */
  apiflags |= TCP_WRITE_FLAG_COPY;
#endif /* LWIP_NETIF_TX_SINGLE_PBUF */

  LWIP_DEBUGF(TCP_OUTPUT_DEBUG, ("tcp_write(pcb=%p, data=%p, len=%"U16_F", apiflags=%"U16_F")\n",
    (void *)pcb, arg, len, (u16_t)apiflags));
  LWIP_ERROR("tcp_write: arg == NULL (programmer violates API)",
             arg != NULL, return ERR_ARG;);

  err = tcp_write_checks(pcb, len);
  if (err != ERR_OK) {
    return err;
  }
  queuelen = pcb->snd_queuelen;

#if LWIP_TCP_TIMESTAMPS
  if ((pcb->flags & TF_TIMESTAMP)) {
    /* Make sure the timestamp option is only included in data segments if we
       agreed about it with the remote host. */
    optflags = TF_SEG_OPTS_TS;
    optlen = LWIP_TCP_OPT_LENGTH(TF_SEG_OPTS_TS);
    /* ensure that segments can hold at least one data byte... */
    mss_local = LWIP_MAX(mss_local, LWIP_TCP_OPT_LEN_TS + 1);
  }
#endif /* LWIP_TCP_TIMESTAMPS */


  /*
   * TCP segmentation is done in three phases with increasing complexity:
   *
   * 1. Copy data directly into an oversized pbuf.
   * 2. Chain a new pbuf to the end of pcb->unsent.
   * 3. Create new segments.
   *
   * We may run out of memory at any point. In that case we must
   * return ERR_MEM and not change anything in pcb. Therefore, all
   * changes are recorded in local variables and committed at the end
   * of the function. Some pcb fields are maintained in local copies:
   *
   * queuelen = pcb->snd_queuelen
   * oversize = pcb->unsent_oversize
   *
   * These variables are set consistently by the phases:
   *
   * seg points to the last segment tampered with.
   *
   * pos records progress as data is segmented.
   */

  /* Find the tail of the unsent queue. */
  if (pcb->unsent != NULL) {
    u16_t space;
    u16_t unsent_optlen;

    /* @todo: this could be sped up by keeping last_unsent in the pcb */
    for (last_unsent = pcb->unsent; last_unsent->next != NULL;
         last_unsent = last_unsent->next);

    /* Usable space at the end of the last unsent segment */
    unsent_optlen = LWIP_TCP_OPT_LENGTH(last_unsent->flags);
    LWIP_ASSERT("mss_local is too small", mss_local >= last_unsent->len + unsent_optlen);
    space = mss_local - (last_unsent->len + unsent_optlen);

    /*
     * Phase 1: Copy data directly into an oversized pbuf.
     *
     * The number of bytes copied is recorded in the oversize_used
     * variable. The actual copying is done at the bottom of the
     * function.
     */
#if TCP_OVERSIZE
#if TCP_OVERSIZE_DBGCHECK
    /* check that pcb->unsent_oversize matches last_unsent->oversize_left */
    LWIP_ASSERT("unsent_oversize mismatch (pcb vs. last_unsent)",
                pcb->unsent_oversize == last_unsent->oversize_left);
#endif /* TCP_OVERSIZE_DBGCHECK */
    oversize = pcb->unsent_oversize;
    if (oversize > 0) {
      LWIP_ASSERT("inconsistent oversize vs. space", oversize <= space);
      seg = last_unsent;
      oversize_used = LWIP_MIN(space, LWIP_MIN(oversize, len));
      pos += oversize_used;
      oversize -= oversize_used;
      space -= oversize_used;
    }
    /* now we are either finished or oversize is zero */
    LWIP_ASSERT("inconsistent oversize vs. len", (oversize == 0) || (pos == len));
#endif /* TCP_OVERSIZE */

    /*
     * Phase 2: Chain a new pbuf to the end of pcb->unsent.
     *
     * As an exception when NOT copying the data, if the given data buffer
     * directly follows the last unsent data buffer in memory, extend the last
     * ROM pbuf reference to the buffer, thus saving a ROM pbuf allocation.
     *
     * We don't extend segments containing SYN/FIN flags or options
     * (len==0). The new pbuf is kept in concat_p and pbuf_cat'ed at
     * the end.
     */
    if ((pos < len) && (space > 0) && (last_unsent->len > 0)) {
      u16_t seglen = LWIP_MIN(space, len - pos);
      seg = last_unsent;

      /* Create a pbuf with a copy or reference to seglen bytes. We
       * can use PBUF_RAW here since the data appears in the middle of
       * a segment. A header will never be prepended. */
      if (apiflags & TCP_WRITE_FLAG_COPY) {
        /* Data is copied */
        if ((concat_p = tcp_pbuf_prealloc(PBUF_RAW, seglen, space, &oversize, pcb, apiflags, 1)) == NULL) {
          LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SERIOUS,
                      ("tcp_write : could not allocate memory for pbuf copy size %"U16_F"\n",
                       seglen));
          goto memerr;
        }
#if TCP_OVERSIZE_DBGCHECK
        oversize_add = oversize;
#endif /* TCP_OVERSIZE_DBGCHECK */
        TCP_DATA_COPY2(concat_p->payload, (const u8_t*)arg + pos, seglen, &concat_chksum, &concat_chksum_swapped);
#if TCP_CHECKSUM_ON_COPY
        concat_chksummed += seglen;
#endif /* TCP_CHECKSUM_ON_COPY */
        queuelen += pbuf_clen(concat_p);
      } else {
        /* Data is not copied */
        /* If the last unsent pbuf is of type PBUF_ROM, try to extend it. */
        struct pbuf *p;
        for (p = last_unsent->p; p->next != NULL; p = p->next);
        if (p->type == PBUF_ROM && (const u8_t *)p->payload + p->len == (const u8_t *)arg) {
          LWIP_ASSERT("tcp_write: ROM pbufs cannot be oversized", pos == 0);
          extendlen = seglen;
        } else {
          if ((concat_p = pbuf_alloc(PBUF_RAW, seglen, PBUF_ROM)) == NULL) {
            LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SERIOUS,
                        ("tcp_write: could not allocate memory for zero-copy pbuf\n"));
            goto memerr;
          }
          /* reference the non-volatile payload data */
          ((struct pbuf_rom*)concat_p)->payload = (const u8_t*)arg + pos;
          queuelen += pbuf_clen(concat_p);
        }
#if TCP_CHECKSUM_ON_COPY
        /* calculate the checksum of nocopy-data */
        tcp_seg_add_chksum(~inet_chksum((const u8_t*)arg + pos, seglen), seglen,
          &concat_chksum, &concat_chksum_swapped);
        concat_chksummed += seglen;
#endif /* TCP_CHECKSUM_ON_COPY */
      }

      pos += seglen;
    }
  } else {
#if TCP_OVERSIZE
    LWIP_ASSERT("unsent_oversize mismatch (pcb->unsent is NULL)",
                pcb->unsent_oversize == 0);
#endif /* TCP_OVERSIZE */
  }

  /*
   * Phase 3: Create new segments.
   *
   * The new segments are chained together in the local 'queue'
   * variable, ready to be appended to pcb->unsent.
   */
  while (pos < len) {
    struct pbuf *p;
    u16_t left = len - pos;
    u16_t max_len = mss_local - optlen;
    u16_t seglen = LWIP_MIN(left, max_len);
#if TCP_CHECKSUM_ON_COPY
    u16_t chksum = 0;
    u8_t chksum_swapped = 0;
#endif /* TCP_CHECKSUM_ON_COPY */

    if (apiflags & TCP_WRITE_FLAG_COPY) {
      /* If copy is set, memory should be allocated and data copied
       * into pbuf */
      if ((p = tcp_pbuf_prealloc(PBUF_TRANSPORT, seglen + optlen, mss_local, &oversize, pcb, apiflags, queue == NULL)) == NULL) {
        LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SERIOUS, ("tcp_write : could not allocate memory for pbuf copy size %"U16_F"\n", seglen));
        goto memerr;
      }
      LWIP_ASSERT("tcp_write: check that first pbuf can hold the complete seglen",
                  (p->len >= seglen));
      TCP_DATA_COPY2((char *)p->payload + optlen, (const u8_t*)arg + pos, seglen, &chksum, &chksum_swapped);
    } else {
      /* Copy is not set: First allocate a pbuf for holding the data.
       * Since the referenced data is available at least until it is
       * sent out on the link (as it has to be ACKed by the remote
       * party) we can safely use PBUF_ROM instead of PBUF_REF here.
       */
      struct pbuf *p2;
#if TCP_OVERSIZE
      LWIP_ASSERT("oversize == 0", oversize == 0);
#endif /* TCP_OVERSIZE */
      if ((p2 = pbuf_alloc(PBUF_TRANSPORT, seglen, PBUF_ROM)) == NULL) {
        LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SERIOUS, ("tcp_write: could not allocate memory for zero-copy pbuf\n"));
        goto memerr;
      }
#if TCP_CHECKSUM_ON_COPY
      /* calculate the checksum of nocopy-data */
      chksum = ~inet_chksum((const u8_t*)arg + pos, seglen);
      if (seglen & 1) {
        chksum_swapped = 1;
        chksum = SWAP_BYTES_IN_WORD(chksum);
      }
#endif /* TCP_CHECKSUM_ON_COPY */
      /* reference the non-volatile payload data */
      ((struct pbuf_rom*)p2)->payload = (const u8_t*)arg + pos;

      /* Second, allocate a pbuf for the headers. */
      if ((p = pbuf_alloc(PBUF_TRANSPORT, optlen, PBUF_RAM)) == NULL) {
        /* If allocation fails, we have to deallocate the data pbuf as
         * well. */
        pbuf_free(p2);
        LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SERIOUS, ("tcp_write: could not allocate memory for header pbuf\n"));
        goto memerr;
      }
      /* Concatenate the headers and data pbufs together. */
      pbuf_cat(p/*header*/, p2/*data*/);
    }

    queuelen += pbuf_clen(p);

    /* Now that there are more segments queued, we check again if the
     * length of the queue exceeds the configured maximum or
     * overflows. */
    if ((queuelen > TCP_SND_QUEUELEN) || (queuelen > TCP_SNDQUEUELEN_OVERFLOW)) {
      LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SERIOUS, ("tcp_write: queue too long %"U16_F" (%d)\n",
        queuelen, (int)TCP_SND_QUEUELEN));
      pbuf_free(p);
      goto memerr;
    }

    if ((seg = tcp_create_segment(pcb, p, 0, pcb->snd_lbb + pos, optflags)) == NULL) {
      goto memerr;
    }
#if TCP_OVERSIZE_DBGCHECK
    seg->oversize_left = oversize;
#endif /* TCP_OVERSIZE_DBGCHECK */
#if TCP_CHECKSUM_ON_COPY
    seg->chksum = chksum;
    seg->chksum_swapped = chksum_swapped;
    seg->flags |= TF_SEG_DATA_CHECKSUMMED;
#endif /* TCP_CHECKSUM_ON_COPY */

    /* first segment of to-be-queued data? */
    if (queue == NULL) {
      queue = seg;
    } else {
      /* Attach the segment to the end of the queued segments */
      LWIP_ASSERT("prev_seg != NULL", prev_seg != NULL);
      prev_seg->next = seg;
    }
    /* remember last segment of to-be-queued data for next iteration */
    prev_seg = seg;

    LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_TRACE, ("tcp_write: queueing %"U32_F":%"U32_F"\n",
      lwip_ntohl(seg->tcphdr->seqno),
      lwip_ntohl(seg->tcphdr->seqno) + TCP_TCPLEN(seg)));

    pos += seglen;
  }

  /*
   * All three segmentation phases were successful. We can commit the
   * transaction.
   */
#if TCP_OVERSIZE_DBGCHECK
  if ((last_unsent != NULL) && (oversize_add != 0)) {
    last_unsent->oversize_left += oversize_add;
  }
#endif /* TCP_OVERSIZE_DBGCHECK */

  /*
   * Phase 1: If data has been added to the preallocated tail of
   * last_unsent, we update the length fields of the pbuf chain.
   */
#if TCP_OVERSIZE
  if (oversize_used > 0) {
    struct pbuf *p;
    /* Bump tot_len of whole chain, len of tail */
    for (p = last_unsent->p; p; p = p->next) {
      p->tot_len += oversize_used;
      if (p->next == NULL) {
        TCP_DATA_COPY((char *)p->payload + p->len, arg, oversize_used, last_unsent);
        p->len += oversize_used;
      }
    }
    last_unsent->len += oversize_used;
#if TCP_OVERSIZE_DBGCHECK
    LWIP_ASSERT("last_unsent->oversize_left >= oversize_used",
                last_unsent->oversize_left >= oversize_used);
    last_unsent->oversize_left -= oversize_used;
#endif /* TCP_OVERSIZE_DBGCHECK */
  }
  pcb->unsent_oversize = oversize;
#endif /* TCP_OVERSIZE */

  /*
   * Phase 2: concat_p can be concatenated onto last_unsent->p, unless we
   * determined that the last ROM pbuf can be extended to include the new data.
   */
  if (concat_p != NULL) {
    LWIP_ASSERT("tcp_write: cannot concatenate when pcb->unsent is empty",
      (last_unsent != NULL));
    pbuf_cat(last_unsent->p, concat_p);
    last_unsent->len += concat_p->tot_len;
  } else if (extendlen > 0) {
    struct pbuf *p;
    LWIP_ASSERT("tcp_write: extension of reference requires reference",
      last_unsent != NULL && last_unsent->p != NULL);
    for (p = last_unsent->p; p->next != NULL; p = p->next) {
      p->tot_len += extendlen;
    }
    p->tot_len += extendlen;
    p->len += extendlen;
    last_unsent->len += extendlen;
  }

#if TCP_CHECKSUM_ON_COPY
  if (concat_chksummed) {
    LWIP_ASSERT("tcp_write: concat checksum needs concatenated data",
        concat_p != NULL || extendlen > 0);
    /*if concat checksumm swapped - swap it back */
    if (concat_chksum_swapped) {
      concat_chksum = SWAP_BYTES_IN_WORD(concat_chksum);
    }
    tcp_seg_add_chksum(concat_chksum, concat_chksummed, &last_unsent->chksum,
      &last_unsent->chksum_swapped);
    last_unsent->flags |= TF_SEG_DATA_CHECKSUMMED;
  }
#endif /* TCP_CHECKSUM_ON_COPY */

  /*
   * Phase 3: Append queue to pcb->unsent. Queue may be NULL, but that
   * is harmless
   */
  if (last_unsent == NULL) {
    pcb->unsent = queue;
  } else {
    last_unsent->next = queue;
  }

  /*
   * Finally update the pcb state.
   */
  pcb->snd_lbb += len;
  pcb->snd_buf -= len;
  pcb->snd_queuelen = queuelen;

  LWIP_DEBUGF(TCP_QLEN_DEBUG, ("tcp_write: %"S16_F" (after enqueued)\n",
    pcb->snd_queuelen));
  if (pcb->snd_queuelen != 0) {
    LWIP_ASSERT("tcp_write: valid queue length",
                pcb->unacked != NULL || pcb->unsent != NULL);
  }

  /* Set the PSH flag in the last segment that we enqueued. */
  if (seg != NULL && seg->tcphdr != NULL && ((apiflags & TCP_WRITE_FLAG_MORE)==0)) {
    TCPH_SET_FLAG(seg->tcphdr, TCP_PSH);
  }

  return ERR_OK;
memerr:
  pcb->flags |= TF_NAGLEMEMERR;
  TCP_STATS_INC(tcp.memerr);

  if (concat_p != NULL) {
    pbuf_free(concat_p);
  }
  if (queue != NULL) {
    tcp_segs_free(queue);
  }
  if (pcb->snd_queuelen != 0) {
    LWIP_ASSERT("tcp_write: valid queue length", pcb->unacked != NULL ||
      pcb->unsent != NULL);
  }
  LWIP_DEBUGF(TCP_QLEN_DEBUG | LWIP_DBG_STATE, ("tcp_write: %"S16_F" (with mem err)\n", pcb->snd_queuelen));
  return ERR_MEM;
}

err_t
tcp_enqueue_flags(struct tcp_pcb *pcb, u8_t flags)
{
  struct pbuf *p;
  struct tcp_seg *seg;
  u8_t optflags = 0;
  u8_t optlen = 0;

  LWIP_DEBUGF(TCP_QLEN_DEBUG, ("tcp_enqueue_flags: queuelen: %"U16_F"\n", (u16_t)pcb->snd_queuelen));

  LWIP_ASSERT("tcp_enqueue_flags: need either TCP_SYN or TCP_FIN in flags (programmer violates API)",
              (flags & (TCP_SYN | TCP_FIN)) != 0);

  /* check for configured max queuelen and possible overflow (FIN flag should always come through!) */
  if (((pcb->snd_queuelen >= TCP_SND_QUEUELEN) || (pcb->snd_queuelen > TCP_SNDQUEUELEN_OVERFLOW)) &&
      ((flags & TCP_FIN) == 0)) {
    LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_LEVEL_SEVERE, ("tcp_enqueue_flags: too long queue %"U16_F" (max %"U16_F")\n",
                                       pcb->snd_queuelen, (u16_t)TCP_SND_QUEUELEN));
    TCP_STATS_INC(tcp.memerr);
    pcb->flags |= TF_NAGLEMEMERR;
    return ERR_MEM;
  }

  if (flags & TCP_SYN) {
    optflags = TF_SEG_OPTS_MSS;
#if LWIP_WND_SCALE
    if ((pcb->state != SYN_RCVD) || (pcb->flags & TF_WND_SCALE)) {
      /* In a <SYN,ACK> (sent in state SYN_RCVD), the window scale option may only
         be sent if we received a window scale option from the remote host. */
      optflags |= TF_SEG_OPTS_WND_SCALE;
    }
#endif /* LWIP_WND_SCALE */
  }
#if LWIP_TCP_TIMESTAMPS
  if ((pcb->flags & TF_TIMESTAMP)) {
    /* Make sure the timestamp option is only included in data segments if we
       agreed about it with the remote host. */
    optflags |= TF_SEG_OPTS_TS;
  }
#endif /* LWIP_TCP_TIMESTAMPS */
  optlen = LWIP_TCP_OPT_LENGTH(optflags);

  /* Allocate pbuf with room for TCP header + options */
  if ((p = pbuf_alloc(PBUF_TRANSPORT, optlen, PBUF_RAM)) == NULL) {
    pcb->flags |= TF_NAGLEMEMERR;
    TCP_STATS_INC(tcp.memerr);
    return ERR_MEM;
  }
  LWIP_ASSERT("tcp_enqueue_flags: check that first pbuf can hold optlen",
              (p->len >= optlen));

  /* Allocate memory for tcp_seg, and fill in fields. */
  if ((seg = tcp_create_segment(pcb, p, flags, pcb->snd_lbb, optflags)) == NULL) {
    pcb->flags |= TF_NAGLEMEMERR;
    TCP_STATS_INC(tcp.memerr);
    return ERR_MEM;
  }
  LWIP_ASSERT("seg->tcphdr not aligned", ((mem_ptr_t)seg->tcphdr % LWIP_MIN(MEM_ALIGNMENT, 4)) == 0);
  LWIP_ASSERT("tcp_enqueue_flags: invalid segment length", seg->len == 0);

  LWIP_DEBUGF(TCP_OUTPUT_DEBUG | LWIP_DBG_TRACE,
              ("tcp_enqueue_flags: queueing %"U32_F":%"U32_F" (0x%"X16_F")\n",
               lwip_ntohl(seg->tcphdr->seqno),
               lwip_ntohl(seg->tcphdr->seqno) + TCP_TCPLEN(seg),
               (u16_t)flags));

  /* Now append seg to pcb->unsent queue */
  if (pcb->unsent == NULL) {
    pcb->unsent = seg;
  } else {
    struct tcp_seg *useg;
    for (useg = pcb->unsent; useg->next != NULL; useg = useg->next);
    useg->next = seg;
  }
#if TCP_OVERSIZE
  /* The new unsent tail has no space */
  pcb->unsent_oversize = 0;
#endif /* TCP_OVERSIZE */

  /* SYN and FIN bump the sequence number */
  if ((flags & TCP_SYN) || (flags & TCP_FIN)) {
    pcb->snd_lbb++;
    /* optlen does not influence snd_buf */
  }
  if (flags & TCP_FIN) {
    pcb->flags |= TF_FIN;
  }

  /* update number of segments on the queues */
  pcb->snd_queuelen += pbuf_clen(seg->p);
  LWIP_DEBUGF(TCP_QLEN_DEBUG, ("tcp_enqueue_flags: %"S16_F" (after enqueued)\n", pcb->snd_queuelen));
  if (pcb->snd_queuelen != 0) {
    LWIP_ASSERT("tcp_enqueue_flags: invalid queue length",
      pcb->unacked != NULL || pcb->unsent != NULL);
  }

  return ERR_OK;
}

#if LWIP_TCP_TIMESTAMPS
/* Build a timestamp option (12 bytes long) at the specified options pointer)
 *
 * @param pcb tcp_pcb
 * @param opts option pointer where to store the timestamp option
 */
static void
tcp_build_timestamp_option(struct tcp_pcb *pcb, u32_t *opts)
{
  /* Pad with two NOP options to make everything nicely aligned */
  opts[0] = PP_HTONL(0x0101080A);
  opts[1] = lwip_htonl(sys_now());
  opts[2] = lwip_htonl(pcb->ts_recent);
}
#endif

#if LWIP_WND_SCALE

static void
tcp_build_wnd_scale_option(u32_t *opts)
{
  /* Pad with one NOP option to make everything nicely aligned */
  opts[0] = PP_HTONL(0x01030300 | TCP_RCV_SCALE);
}
#endif

err_t
tcp_send_empty_ack(struct tcp_pcb *pcb)
{
  err_t err;
  struct pbuf *p;
  u8_t optlen = 0;
  struct netif *netif;
#if LWIP_TCP_TIMESTAMPS || CHECKSUM_GEN_TCP
  struct tcp_hdr *tcphdr;
#endif /* LWIP_TCP_TIMESTAMPS || CHECKSUM_GEN_TCP */

#if LWIP_TCP_TIMESTAMPS
  if (pcb->flags & TF_TIMESTAMP) {
    optlen = LWIP_TCP_OPT_LENGTH(TF_SEG_OPTS_TS);
  }
#endif

  p = tcp_output_alloc_header(pcb, optlen, 0, lwip_htonl(pcb->snd_nxt));
  if (p == NULL) {
    /* let tcp_fasttmr retry sending this ACK */
    pcb->flags |= (TF_ACK_DELAY | TF_ACK_NOW);
    LWIP_DEBUGF(TCP_OUTPUT_DEBUG, ("tcp_output: (ACK) could not allocate pbuf\n"));
    return ERR_BUF;
  }
#if LWIP_TCP_TIMESTAMPS || CHECKSUM_GEN_TCP
  tcphdr = (struct tcp_hdr *)p->payload;
#endif /* LWIP_TCP_TIMESTAMPS || CHECKSUM_GEN_TCP */
  LWIP_DEBUGF(TCP_OUTPUT_DEBUG,
              ("tcp_output: sending ACK for %"U32_F"\n", pcb->rcv_nxt));

  /* NB. MSS and window scale options are only sent on SYNs, so ignore them here */
#if LWIP_TCP_TIMESTAMPS
  pcb->ts_lastacksent = pcb->rcv_nxt;

  if (pcb->flags & TF_TIMESTAMP) {
    tcp_build_timestamp_option(pcb, (u32_t *)(tcphdr + 1));
  }
#endif

  netif = ip_route(&pcb->local_ip, &pcb->remote_ip);
  if (netif == NULL) {
    err = ERR_RTE;
  } else {
#if CHECKSUM_GEN_TCP
    IF__NETIF_CHECKSUM_ENABLED(netif, NETIF_CHECKSUM_GEN_TCP) {
      tcphdr->chksum = ip_chksum_pseudo(p, IP_PROTO_TCP, p->tot_len,
        &pcb->local_ip, &pcb->remote_ip);
    }
#endif
    NETIF_SET_HWADDRHINT(netif, &(pcb->addr_hint));
    err = ip_output_if(p, &pcb->local_ip, &pcb->remote_ip,
      pcb->ttl, pcb->tos, IP_PROTO_TCP, netif);
    NETIF_SET_HWADDRHINT(netif, NULL);
  }
  pbuf_free(p);

  if (err != ERR_OK) {
    /* let tcp_fasttmr retry sending this ACK */
    pcb->flags |= (TF_ACK_DELAY | TF_ACK_NOW);
  } else {
    /* remove ACK flags from the PCB, as we sent an empty ACK now */
    pcb->flags &= ~(TF_ACK_DELAY | TF_ACK_NOW);
  }

  return err;
}

err_t
tcp_output(struct tcp_pcb *pcb)
{
  struct tcp_seg *seg, *useg;
  u32_t wnd, snd_nxt;
  err_t err;
  struct netif *netif;
#if TCP_CWND_DEBUG
  s16_t i = 0;
#endif /* TCP_CWND_DEBUG */

  /* pcb->state LISTEN not allowed here */
  LWIP_ASSERT("don't call tcp_output for listen-pcbs",
    pcb->state != LISTEN);

  /* First, check if we are invoked by the TCP input processing
     code. If so, we do not output anything. Instead, we rely on the
     input processing code to call us when input processing is done
     with. */
  if (tcp_input_pcb == pcb) {
    return ERR_OK;
  }

  wnd = LWIP_MIN(pcb->snd_wnd, pcb->cwnd);

  seg = pcb->unsent;

  /* If the TF_ACK_NOW flag is set and no data will be sent (either
   * because the ->unsent queue is empty or because the window does
   * not allow it), construct an empty ACK segment and send it.
   *
   * If data is to be sent, we will just piggyback the ACK (see below).
   */
  if (pcb->flags & TF_ACK_NOW &&
     (seg == NULL ||
      lwip_ntohl(seg->tcphdr->seqno) - pcb->lastack + seg->len > wnd)) {
     return tcp_send_empty_ack(pcb);
  }

  /* useg should point to last segment on unacked queue */
  useg = pcb->unacked;
  if (useg != NULL) {
    for (; useg->next != NULL; useg = useg->next);
  }

  netif = ip_route(&pcb->local_ip, &pcb->remote_ip);
  if (netif == NULL) {
    return ERR_RTE;
  }

  /* If we don't have a local IP address, we get one from netif */
  if (ip_addr_isany(&pcb->local_ip)) {
    const ip_addr_t *local_ip = ip_netif_get_local_ip(netif, &pcb->remote_ip);
    if (local_ip == NULL) {
      return ERR_RTE;
    }
    ip_addr_copy(pcb->local_ip, *local_ip);
  }

#if TCP_OUTPUT_DEBUG
  if (seg == NULL) {
    LWIP_DEBUGF(TCP_OUTPUT_DEBUG, ("tcp_output: nothing to send (%p)\n",
                                   (void*)pcb->unsent));
  }
#endif /* TCP_OUTPUT_DEBUG */
#if TCP_CWND_DEBUG
  if (seg == NULL) {
    LWIP_DEBUGF(TCP_CWND_DEBUG, ("tcp_output: snd_wnd %"TCPWNDSIZE_F
                                 ", cwnd %"TCPWNDSIZE_F", wnd %"U32_F
                                 ", seg == NULL, ack %"U32_F"\n",
                                 pcb->snd_wnd, pcb->cwnd, wnd, pcb->lastack));
  } else {
    LWIP_DEBUGF(TCP_CWND_DEBUG,
                ("tcp_output: snd_wnd %"TCPWNDSIZE_F", cwnd %"TCPWNDSIZE_F", wnd %"U32_F
                 ", effwnd %"U32_F", seq %"U32_F", ack %"U32_F"\n",
                 pcb->snd_wnd, pcb->cwnd, wnd,
                 lwip_ntohl(seg->tcphdr->seqno) - pcb->lastack + seg->len,
                 lwip_ntohl(seg->tcphdr->seqno), pcb->lastack));
  }
#endif /* TCP_CWND_DEBUG */
  /* Check if we need to start the persistent timer when the next unsent segment
   * does not fit within the remaining send window and RTO timer is not running (we
   * have no in-flight data). A traditional approach would fill the remaining window
   * with part of the unsent segment (which will engage zero-window probing upon
   * reception of the zero window update from the receiver). This ensures the
   * subsequent window update is reliably received. With the goal of being lightweight,
   * we avoid splitting the unsent segment and treat the window as already zero.
   */
  if (seg != NULL &&
      lwip_ntohl(seg->tcphdr->seqno) - pcb->lastack + seg->len > wnd &&
      wnd > 0 && wnd == pcb->snd_wnd && pcb->unacked == NULL) {
    /* Start the persist timer */
    if (pcb->persist_backoff == 0) {
      pcb->persist_cnt = 0;
      pcb->persist_backoff = 1;
    }
    goto output_done;
  }
  /* data available and window allows it to be sent? */
  while (seg != NULL &&
         lwip_ntohl(seg->tcphdr->seqno) - pcb->lastack + seg->len <= wnd) {
    LWIP_ASSERT("RST not expected here!",
                (TCPH_FLAGS(seg->tcphdr) & TCP_RST) == 0);
    /* Stop sending if the nagle algorithm would prevent it
     * Don't stop:
     * - if tcp_write had a memory error before (prevent delayed ACK timeout) or
     * - if FIN was already enqueued for this PCB (SYN is always alone in a segment -
     *   either seg->next != NULL or pcb->unacked == NULL;
     *   RST is no sent using tcp_write/tcp_output.
     */
    if ((tcp_do_output_nagle(pcb) == 0) &&
      ((pcb->flags & (TF_NAGLEMEMERR | TF_FIN)) == 0)) {
      break;
    }
#if TCP_CWND_DEBUG
    LWIP_DEBUGF(TCP_CWND_DEBUG, ("tcp_output: snd_wnd %"TCPWNDSIZE_F", cwnd %"TCPWNDSIZE_F", wnd %"U32_F", effwnd %"U32_F", seq %"U32_F", ack %"U32_F", i %"S16_F"\n",
                            pcb->snd_wnd, pcb->cwnd, wnd,
                            lwip_ntohl(seg->tcphdr->seqno) + seg->len -
                            pcb->lastack,
                            lwip_ntohl(seg->tcphdr->seqno), pcb->lastack, i));
    ++i;
#endif /* TCP_CWND_DEBUG */

    if (pcb->state != SYN_SENT) {
      TCPH_SET_FLAG(seg->tcphdr, TCP_ACK);
    }

#if TCP_OVERSIZE_DBGCHECK
    seg->oversize_left = 0;
#endif /* TCP_OVERSIZE_DBGCHECK */
    err = tcp_output_segment(seg, pcb, netif);
    if (err != ERR_OK) {
      /* segment could not be sent, for whatever reason */
      pcb->flags |= TF_NAGLEMEMERR;
      return err;
    }
    pcb->unsent = seg->next;
    if (pcb->state != SYN_SENT) {
      pcb->flags &= ~(TF_ACK_DELAY | TF_ACK_NOW);
    }
    snd_nxt = lwip_ntohl(seg->tcphdr->seqno) + TCP_TCPLEN(seg);
    if (TCP_SEQ_LT(pcb->snd_nxt, snd_nxt)) {
      pcb->snd_nxt = snd_nxt;
    }
    /* put segment on unacknowledged list if length > 0 */
    if (TCP_TCPLEN(seg) > 0) {
      seg->next = NULL;
      /* unacked list is empty? */
      if (pcb->unacked == NULL) {
        pcb->unacked = seg;
        useg = seg;
      /* unacked list is not empty? */
      } else {
        /* In the case of fast retransmit, the packet should not go to the tail
         * of the unacked queue, but rather somewhere before it. We need to check for
         * this case. -STJ Jul 27, 2004 */
        if (TCP_SEQ_LT(lwip_ntohl(seg->tcphdr->seqno), lwip_ntohl(useg->tcphdr->seqno))) {
          /* add segment to before tail of unacked list, keeping the list sorted */
          struct tcp_seg **cur_seg = &(pcb->unacked);
          while (*cur_seg &&
            TCP_SEQ_LT(lwip_ntohl((*cur_seg)->tcphdr->seqno), lwip_ntohl(seg->tcphdr->seqno))) {
              cur_seg = &((*cur_seg)->next );
          }
          seg->next = (*cur_seg);
          (*cur_seg) = seg;
        } else {
          /* add segment to tail of unacked list */
          useg->next = seg;
          useg = useg->next;
        }
      }
    /* do not queue empty segments on the unacked list */
    } else {
      tcp_seg_free(seg);
    }
    seg = pcb->unsent;
  }
output_done:
#if TCP_OVERSIZE
  if (pcb->unsent == NULL) {
    /* last unsent has been removed, reset unsent_oversize */
    pcb->unsent_oversize = 0;
  }
#endif /* TCP_OVERSIZE */

  pcb->flags &= ~TF_NAGLEMEMERR;
  return ERR_OK;
}

static err_t
tcp_output_segment(struct tcp_seg *seg, struct tcp_pcb *pcb, struct netif *netif)
{
  err_t err;
  u16_t len;
  u32_t *opts;

  if (seg->p->ref != 1) {
    /* This can happen if the pbuf of this segment is still referenced by the
       netif driver due to deferred transmission. Since this function modifies
       p->len, we must not continue in this case. */
    return ERR_OK;
  }

  /* The TCP header has already been constructed, but the ackno and
   wnd fields remain. */
  seg->tcphdr->ackno = lwip_htonl(pcb->rcv_nxt);

  /* advertise our receive window size in this TCP segment */
#if LWIP_WND_SCALE
  if (seg->flags & TF_SEG_OPTS_WND_SCALE) {
    /* The Window field in a SYN segment itself (the only type where we send
       the window scale option) is never scaled. */
    seg->tcphdr->wnd = lwip_htons(TCPWND_MIN16(pcb->rcv_ann_wnd));
  } else
#endif /* LWIP_WND_SCALE */
  {
    seg->tcphdr->wnd = lwip_htons(TCPWND_MIN16(RCV_WND_SCALE(pcb, pcb->rcv_ann_wnd)));
  }

  pcb->rcv_ann_right_edge = pcb->rcv_nxt + pcb->rcv_ann_wnd;

  /* Add any requested options.  NB MSS option is only set on SYN
     packets, so ignore it here */
  /* cast through void* to get rid of alignment warnings */
  opts = (u32_t *)(void *)(seg->tcphdr + 1);
  if (seg->flags & TF_SEG_OPTS_MSS) {
    u16_t mss;
#if TCP_CALCULATE_EFF_SEND_MSS
    mss = tcp_eff_send_mss(TCP_MSS, &pcb->local_ip, &pcb->remote_ip);
#else /* TCP_CALCULATE_EFF_SEND_MSS */
    mss = TCP_MSS;
#endif /* TCP_CALCULATE_EFF_SEND_MSS */
    *opts = TCP_BUILD_MSS_OPTION(mss);
    opts += 1;
  }
#if LWIP_TCP_TIMESTAMPS
  pcb->ts_lastacksent = pcb->rcv_nxt;

  if (seg->flags & TF_SEG_OPTS_TS) {
    tcp_build_timestamp_option(pcb, opts);
    opts += 3;
  }
#endif
#if LWIP_WND_SCALE
  if (seg->flags & TF_SEG_OPTS_WND_SCALE) {
    tcp_build_wnd_scale_option(opts);
    opts += 1;
  }
#endif

  /* Set retransmission timer running if it is not currently enabled
     This must be set before checking the route. */
  if (pcb->rtime < 0) {
    pcb->rtime = 0;
  }

  if (pcb->rttest == 0) {
    pcb->rttest = tcp_ticks;
    pcb->rtseq = lwip_ntohl(seg->tcphdr->seqno);

    LWIP_DEBUGF(TCP_RTO_DEBUG, ("tcp_output_segment: rtseq %"U32_F"\n", pcb->rtseq));
  }
  LWIP_DEBUGF(TCP_OUTPUT_DEBUG, ("tcp_output_segment: %"U32_F":%"U32_F"\n",
          lwip_htonl(seg->tcphdr->seqno), lwip_htonl(seg->tcphdr->seqno) +
          seg->len));

  len = (u16_t)((u8_t *)seg->tcphdr - (u8_t *)seg->p->payload);
  if (len == 0) {
    MIB2_STATS_INC(mib2.tcpoutsegs);
  }

  seg->p->len -= len;
  seg->p->tot_len -= len;

  seg->p->payload = seg->tcphdr;

  seg->tcphdr->chksum = 0;
#if CHECKSUM_GEN_TCP
  IF__NETIF_CHECKSUM_ENABLED(netif, NETIF_CHECKSUM_GEN_TCP) {
#if TCP_CHECKSUM_ON_COPY
    u32_t acc;
#if TCP_CHECKSUM_ON_COPY_SANITY_CHECK
    u16_t chksum_slow = ip_chksum_pseudo(seg->p, IP_PROTO_TCP,
      seg->p->tot_len, &pcb->local_ip, &pcb->remote_ip);
#endif /* TCP_CHECKSUM_ON_COPY_SANITY_CHECK */
    if ((seg->flags & TF_SEG_DATA_CHECKSUMMED) == 0) {
      LWIP_ASSERT("data included but not checksummed",
        seg->p->tot_len == (TCPH_HDRLEN(seg->tcphdr) * 4));
    }

    /* rebuild TCP header checksum (TCP header changes for retransmissions!) */
    acc = ip_chksum_pseudo_partial(seg->p, IP_PROTO_TCP,
      seg->p->tot_len, TCPH_HDRLEN(seg->tcphdr) * 4, &pcb->local_ip, &pcb->remote_ip);
    /* add payload checksum */
    if (seg->chksum_swapped) {
      seg->chksum = SWAP_BYTES_IN_WORD(seg->chksum);
      seg->chksum_swapped = 0;
    }
    acc += (u16_t)~(seg->chksum);
    seg->tcphdr->chksum = FOLD_U32T(acc);
#if TCP_CHECKSUM_ON_COPY_SANITY_CHECK
    if (chksum_slow != seg->tcphdr->chksum) {
      TCP_CHECKSUM_ON_COPY_SANITY_CHECK_FAIL(
                  ("tcp_output_segment: calculated checksum is %"X16_F" instead of %"X16_F"\n",
                  seg->tcphdr->chksum, chksum_slow));
      seg->tcphdr->chksum = chksum_slow;
    }
#endif /* TCP_CHECKSUM_ON_COPY_SANITY_CHECK */
#else /* TCP_CHECKSUM_ON_COPY */
    seg->tcphdr->chksum = ip_chksum_pseudo(seg->p, IP_PROTO_TCP,
      seg->p->tot_len, &pcb->local_ip, &pcb->remote_ip);
#endif /* TCP_CHECKSUM_ON_COPY */
  }
#endif /* CHECKSUM_GEN_TCP */
  TCP_STATS_INC(tcp.xmit);

  NETIF_SET_HWADDRHINT(netif, &(pcb->addr_hint));
  err = ip_output_if(seg->p, &pcb->local_ip, &pcb->remote_ip, pcb->ttl,
    pcb->tos, IP_PROTO_TCP, netif);
  NETIF_SET_HWADDRHINT(netif, NULL);
  return err;
}

void
tcp_rst(u32_t seqno, u32_t ackno,
  const ip_addr_t *local_ip, const ip_addr_t *remote_ip,
  u16_t local_port, u16_t remote_port)
{
  struct pbuf *p;
  struct tcp_hdr *tcphdr;
  struct netif *netif;
  p = pbuf_alloc(PBUF_IP, TCP_HLEN, PBUF_RAM);
  if (p == NULL) {
    LWIP_DEBUGF(TCP_DEBUG, ("tcp_rst: could not allocate memory for pbuf\n"));
    return;
  }
  LWIP_ASSERT("check that first pbuf can hold struct tcp_hdr",
              (p->len >= sizeof(struct tcp_hdr)));

  tcphdr = (struct tcp_hdr *)p->payload;
  tcphdr->src = lwip_htons(local_port);
  tcphdr->dest = lwip_htons(remote_port);
  tcphdr->seqno = lwip_htonl(seqno);
  tcphdr->ackno = lwip_htonl(ackno);
  TCPH_HDRLEN_FLAGS_SET(tcphdr, TCP_HLEN/4, TCP_RST | TCP_ACK);
#if LWIP_WND_SCALE
  tcphdr->wnd = PP_HTONS(((TCP_WND >> TCP_RCV_SCALE) & 0xFFFF));
#else
  tcphdr->wnd = PP_HTONS(TCP_WND);
#endif
  tcphdr->chksum = 0;
  tcphdr->urgp = 0;

  TCP_STATS_INC(tcp.xmit);
  MIB2_STATS_INC(mib2.tcpoutrsts);

  netif = ip_route(local_ip, remote_ip);
  if (netif != NULL) {
#if CHECKSUM_GEN_TCP
    IF__NETIF_CHECKSUM_ENABLED(netif, NETIF_CHECKSUM_GEN_TCP) {
      tcphdr->chksum = ip_chksum_pseudo(p, IP_PROTO_TCP, p->tot_len,
                                        local_ip, remote_ip);
    }
#endif
    /* Send output with hardcoded TTL/HL since we have no access to the pcb */
    ip_output_if(p, local_ip, remote_ip, TCP_TTL, 0, IP_PROTO_TCP, netif);
  }
  pbuf_free(p);
  LWIP_DEBUGF(TCP_RST_DEBUG, ("tcp_rst: seqno %"U32_F" ackno %"U32_F".\n", seqno, ackno));
}

void
tcp_rexmit_rto(struct tcp_pcb *pcb)
{
  struct tcp_seg *seg;

  if (pcb->unacked == NULL) {
    return;
  }

  /* Move all unacked segments to the head of the unsent queue */
  for (seg = pcb->unacked; seg->next != NULL; seg = seg->next);
  /* concatenate unsent queue after unacked queue */
  seg->next = pcb->unsent;
#if TCP_OVERSIZE_DBGCHECK
  /* if last unsent changed, we need to update unsent_oversize */
  if (pcb->unsent == NULL) {
    pcb->unsent_oversize = seg->oversize_left;
  }
#endif /* TCP_OVERSIZE_DBGCHECK */
  /* unsent queue is the concatenated queue (of unacked, unsent) */
  pcb->unsent = pcb->unacked;
  /* unacked queue is now empty */
  pcb->unacked = NULL;

  /* increment number of retransmissions */
  if (pcb->nrtx < 0xFF) {
    ++pcb->nrtx;
  }

  /* Don't take any RTT measurements after retransmitting. */
  pcb->rttest = 0;

  /* Do the actual retransmission */
  tcp_output(pcb);
}

void
tcp_rexmit(struct tcp_pcb *pcb)
{
  struct tcp_seg *seg;
  struct tcp_seg **cur_seg;

  if (pcb->unacked == NULL) {
    return;
  }

  /* Move the first unacked segment to the unsent queue */
  /* Keep the unsent queue sorted. */
  seg = pcb->unacked;
  pcb->unacked = seg->next;

  cur_seg = &(pcb->unsent);
  while (*cur_seg &&
    TCP_SEQ_LT(lwip_ntohl((*cur_seg)->tcphdr->seqno), lwip_ntohl(seg->tcphdr->seqno))) {
      cur_seg = &((*cur_seg)->next );
  }
  seg->next = *cur_seg;
  *cur_seg = seg;
#if TCP_OVERSIZE
  if (seg->next == NULL) {
    /* the retransmitted segment is last in unsent, so reset unsent_oversize */
    pcb->unsent_oversize = 0;
  }
#endif /* TCP_OVERSIZE */

  if (pcb->nrtx < 0xFF) {
    ++pcb->nrtx;
  }

  /* Don't take any rtt measurements after retransmitting. */
  pcb->rttest = 0;

  /* Do the actual retransmission. */
  MIB2_STATS_INC(mib2.tcpretranssegs);
  /* No need to call tcp_output: we are always called from tcp_input()
     and thus tcp_output directly returns. */
}


void
tcp_rexmit_fast(struct tcp_pcb *pcb)
{
  if (pcb->unacked != NULL && !(pcb->flags & TF_INFR)) {
    /* This is fast retransmit. Retransmit the first unacked segment. */
    LWIP_DEBUGF(TCP_FR_DEBUG,
                ("tcp_receive: dupacks %"U16_F" (%"U32_F
                 "), fast retransmit %"U32_F"\n",
                 (u16_t)pcb->dupacks, pcb->lastack,
                 lwip_ntohl(pcb->unacked->tcphdr->seqno)));
    tcp_rexmit(pcb);

    /* Set ssthresh to half of the minimum of the current
     * cwnd and the advertised window */
    pcb->ssthresh = LWIP_MIN(pcb->cwnd, pcb->snd_wnd) / 2;

    /* The minimum value for ssthresh should be 2 MSS */
    if (pcb->ssthresh < (2U * pcb->mss)) {
      LWIP_DEBUGF(TCP_FR_DEBUG,
                  ("tcp_receive: The minimum value for ssthresh %"TCPWNDSIZE_F
                   " should be min 2 mss %"U16_F"...\n",
                   pcb->ssthresh, (u16_t)(2*pcb->mss)));
      pcb->ssthresh = 2*pcb->mss;
    }

    pcb->cwnd = pcb->ssthresh + 3 * pcb->mss;
    pcb->flags |= TF_INFR;

    /* Reset the retransmission timer to prevent immediate rto retransmissions */
    pcb->rtime = 0;
  }
}


err_t
tcp_keepalive(struct tcp_pcb *pcb)
{
  err_t err;
  struct pbuf *p;
  struct netif *netif;

  LWIP_DEBUGF(TCP_DEBUG, ("tcp_keepalive: sending KEEPALIVE probe to "));
  ip_addr_debug_print(TCP_DEBUG, &pcb->remote_ip);
  LWIP_DEBUGF(TCP_DEBUG, ("\n"));

  LWIP_DEBUGF(TCP_DEBUG, ("tcp_keepalive: tcp_ticks %"U32_F"   pcb->tmr %"U32_F" pcb->keep_cnt_sent %"U16_F"\n",
                          tcp_ticks, pcb->tmr, (u16_t)pcb->keep_cnt_sent));

  p = tcp_output_alloc_header(pcb, 0, 0, lwip_htonl(pcb->snd_nxt - 1));
  if (p == NULL) {
    LWIP_DEBUGF(TCP_DEBUG,
                ("tcp_keepalive: could not allocate memory for pbuf\n"));
    return ERR_MEM;
  }
  netif = ip_route(&pcb->local_ip, &pcb->remote_ip);
  if (netif == NULL) {
    err = ERR_RTE;
  } else {
#if CHECKSUM_GEN_TCP
    IF__NETIF_CHECKSUM_ENABLED(netif, NETIF_CHECKSUM_GEN_TCP) {
      struct tcp_hdr *tcphdr = (struct tcp_hdr *)p->payload;
      tcphdr->chksum = ip_chksum_pseudo(p, IP_PROTO_TCP, p->tot_len,
        &pcb->local_ip, &pcb->remote_ip);
    }
#endif /* CHECKSUM_GEN_TCP */
    TCP_STATS_INC(tcp.xmit);

    /* Send output to IP */
    NETIF_SET_HWADDRHINT(netif, &(pcb->addr_hint));
    err = ip_output_if(p, &pcb->local_ip, &pcb->remote_ip, pcb->ttl, 0, IP_PROTO_TCP, netif);
    NETIF_SET_HWADDRHINT(netif, NULL);
  }
  pbuf_free(p);

  LWIP_DEBUGF(TCP_DEBUG, ("tcp_keepalive: seqno %"U32_F" ackno %"U32_F" err %d.\n",
                          pcb->snd_nxt - 1, pcb->rcv_nxt, (int)err));
  return err;
}


err_t
tcp_zero_window_probe(struct tcp_pcb *pcb)
{
  err_t err;
  struct pbuf *p;
  struct tcp_hdr *tcphdr;
  struct tcp_seg *seg;
  u16_t len;
  u8_t is_fin;
  u32_t snd_nxt;
  struct netif *netif;

  LWIP_DEBUGF(TCP_DEBUG, ("tcp_zero_window_probe: sending ZERO WINDOW probe to "));
  ip_addr_debug_print(TCP_DEBUG, &pcb->remote_ip);
  LWIP_DEBUGF(TCP_DEBUG, ("\n"));

  LWIP_DEBUGF(TCP_DEBUG,
              ("tcp_zero_window_probe: tcp_ticks %"U32_F
               "   pcb->tmr %"U32_F" pcb->keep_cnt_sent %"U16_F"\n",
               tcp_ticks, pcb->tmr, (u16_t)pcb->keep_cnt_sent));

  seg = pcb->unacked;

  if (seg == NULL) {
    seg = pcb->unsent;
  }
  if (seg == NULL) {
    /* nothing to send, zero window probe not needed */
    return ERR_OK;
  }

  is_fin = ((TCPH_FLAGS(seg->tcphdr) & TCP_FIN) != 0) && (seg->len == 0);
  /* we want to send one seqno: either FIN or data (no options) */
  len = is_fin ? 0 : 1;

  p = tcp_output_alloc_header(pcb, 0, len, seg->tcphdr->seqno);
  if (p == NULL) {
    LWIP_DEBUGF(TCP_DEBUG, ("tcp_zero_window_probe: no memory for pbuf\n"));
    return ERR_MEM;
  }
  tcphdr = (struct tcp_hdr *)p->payload;

  if (is_fin) {
    /* FIN segment, no data */
    TCPH_FLAGS_SET(tcphdr, TCP_ACK | TCP_FIN);
  } else {
    /* Data segment, copy in one byte from the head of the unacked queue */
    char *d = ((char *)p->payload + TCP_HLEN);
    /* Depending on whether the segment has already been sent (unacked) or not
       (unsent), seg->p->payload points to the IP header or TCP header.
       Ensure we copy the first TCP data byte: */
    pbuf_copy_partial(seg->p, d, 1, seg->p->tot_len - seg->len);
  }

  /* The byte may be acknowledged without the window being opened. */
  snd_nxt = lwip_ntohl(seg->tcphdr->seqno) + 1;
  if (TCP_SEQ_LT(pcb->snd_nxt, snd_nxt)) {
    pcb->snd_nxt = snd_nxt;
  }

  netif = ip_route(&pcb->local_ip, &pcb->remote_ip);
  if (netif == NULL) {
    err = ERR_RTE;
  } else {
#if CHECKSUM_GEN_TCP
    IF__NETIF_CHECKSUM_ENABLED(netif, NETIF_CHECKSUM_GEN_TCP) {
      tcphdr->chksum = ip_chksum_pseudo(p, IP_PROTO_TCP, p->tot_len,
        &pcb->local_ip, &pcb->remote_ip);
    }
#endif
    TCP_STATS_INC(tcp.xmit);

    /* Send output to IP */
    NETIF_SET_HWADDRHINT(netif, &(pcb->addr_hint));
    err = ip_output_if(p, &pcb->local_ip, &pcb->remote_ip, pcb->ttl,
      0, IP_PROTO_TCP, netif);
    NETIF_SET_HWADDRHINT(netif, NULL);
  }

  pbuf_free(p);

  LWIP_DEBUGF(TCP_DEBUG, ("tcp_zero_window_probe: seqno %"U32_F
                          " ackno %"U32_F" err %d.\n",
                          pcb->snd_nxt - 1, pcb->rcv_nxt, (int)err));
  return err;
}
#endif /* LWIP_TCP */