/* Caller should be in critical section */
static struct tcpcb *
tcp_timer_persist_handler(struct tcpcb *tp)
{
#ifdef TCPDEBUG
int ostate;
#endif
#ifdef TCPDEBUG
ostate = tp->t_state;
#endif
/*
* Persistance timer into zero window.
* Force a byte to be output, if possible.
*/
tcpstat.tcps_persisttimeo++;
/*
* Hack: if the peer is dead/unreachable, we do not
* time out if the window is closed. After a full
* backoff, drop the connection if the idle time
* (no responses to probes) reaches the maximum
* backoff that we would use if retransmitting.
*/
if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
(ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
tcpstat.tcps_persistdrop++;
tp = tcp_drop(tp, ETIMEDOUT);
goto out;
}
tcp_setpersist(tp);
tp->t_flags |= TF_FORCE;
tcp_output(tp);
tp->t_flags &= ~TF_FORCE;
out:
#ifdef TCPDEBUG
if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
#endif
return tp;
}
int
tcp_output(struct tcpcb * tp)
{
struct socket * so = tp->t_inpcb->inp_socket;
int len;
long win;
int off, flags, error;
struct mbuf * m;
struct tcpiphdr * ti;
unsigned optlen = 0;
int idle, sendalot;
struct mbuf * sendm; /* mbuf which contains data to send */
struct mbuf * tcp_mbuf; /* mbuf containing TCP header */
int bufoff; /* offset of data in sendm->m_data */
#ifdef TCP_SACK
int sack_resend;
int sack_hole = 0; /* next sack hole to fill */
if(tp->t_flags & TF_SACKREPLY)
{
/* we are resending based on a received SACK header */
sack_resend = TRUE;
tp->t_flags &= ~TF_SACKREPLY; /* clear flag */
}
else
sack_resend = FALSE;
#endif /* TCP_SACK */
/*
* Determine length of data that should be transmitted,
* and flags that will be used.
* If there is some data or critical controls (SYN, RST)
* to send, then transmit; otherwise, investigate further.
*/
idle = (tp->snd_max == tp->snd_una);
again:
sendalot = 0;
off = (int)(tp->snd_nxt - tp->snd_una);
win = (long)tp->snd_wnd; /* set basic send window */
if (win > (long)tp->snd_cwnd) /* see if we need congestion control */
{
win = (int)(tp->snd_cwnd & ~(ALIGN_TYPE-1)); /* keep data aligned */
}
/*
* If in persist timeout with window of 0, send 1 byte.
* Otherwise, if window is small but nonzero
* and timer expired, we will send what we can
* and go to transmit state.
*/
if (tp->t_force)
{
if (win == 0)
win = 1;
else
{
tp->t_timer[TCPT_PERSIST] = 0;
tp->t_rxtshift = 0;
}
}
#ifdef TCP_SACK
/* See if we need to adjust the offset for a sack resend */
if(sack_resend)
{
off = (int)(tp->sack_hole_start[sack_hole] - tp->snd_una);
/* if this hole's already been acked then punt and move to next hole */
if(off < 0)
{
/* clear out the acked hole */
tp->sack_hole_start[sack_hole] = tp->sack_hole_end[sack_hole] = 0;
/* see if we're done with SACK hole list (2 tests) */
if(++sack_hole >= SACK_BLOCKS)
return 0;
if(tp->sack_hole_start[sack_hole] == tp->sack_hole_end[sack_hole])
return 0;
goto again;
}
tp->snd_nxt = tp->sack_hole_start[sack_hole];
len = (int)(tp->sack_hole_end[sack_hole] - tp->sack_hole_start[sack_hole]);
len = (int)MIN(len, (int)win);
}
else
#endif /* TCP_SACK */
{
/* set length of packets which are not sack resends */
len = (int)MIN(so->so_snd.sb_cc, (unsigned)win) - off;
}
flags = tcp_outflags[tp->t_state];
/* See if we need to build TCP options field. This test should be fast. */
#if (defined(TCP_TIMESTAMP) | defined(TCP_SACK))
if((flags & TH_SYN) ||
/* !!!??? (so->so_options & SO_TIMESTAMP) || */
(tp->t_flags & TF_SACKNOW)
)
{
optlen = bld_options(tp, &tcp_optionbuf[optlen], flags, so);
}
#else
/* If other options not defined this build then don't bother to call bld_options() except
* on SYN packets
*/
if(flags & TH_SYN)
{
optlen = bld_options(tp, &tcp_optionbuf[optlen], flags, so);
}
#endif
if (len < 0)
{
/*
* If FIN has been sent but not acked,
* but we haven't been called to retransmit,
* len will be -1. Otherwise, window shrank
* after we sent into it. If window shrank to 0,
* cancel pending retransmit and pull snd_nxt
* back to (closed) window. We will enter persist
* state below. If the window didn't close completely,
* just wait for an ACK.
*/
len = 0;
if (win == 0)
{
tp->t_timer[TCPT_REXMT] = 0;
tp->snd_nxt = tp->snd_una;
}
}
if (len > (int)tp->t_maxseg)
{
len = tp->t_maxseg;
sendalot = 1;
}
#ifdef IP_V4
#ifdef IP_PMTU
{
int pmtu = tp->t_inpcb->inp_pmtu - 40;
if (len > pmtu)
{
len = pmtu - 40;
sendalot = 1;
}
}
#endif /* IP_PMTU */
/* We don't need a pmtu test for IPv6. V6 code limits t_maxseg to
* the Path MTU, so the test above the v4 ifdef above covers us.
*/
#endif /* IP_V4 */
if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
flags &= ~TH_FIN;
win = (long)(sbspace(&so->so_rcv));
/*
* If our state indicates that FIN should be sent
* and we have not yet done so, or we're retransmitting the FIN,
* then we need to send.
*/
if ((flags & TH_FIN) &&
(so->so_snd.sb_cc == 0) &&
((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
{
goto send;
}
/*
* Send if we owe peer an ACK.
*/
if (tp->t_flags & TF_ACKNOW)
goto send;
if (flags & (TH_SYN|TH_RST))
goto send;
if (SEQ_GT(tp->snd_up, tp->snd_una))
goto send;
/*
* Sender silly window avoidance. If connection is idle
* and can send all data, a maximum segment,
* at least a maximum default-size segment do it,
* or are forced, do it; otherwise don't bother.
* If peer's buffer is tiny, then send
* when window is at least half open.
* If retransmitting (possibly after persist timer forced us
* to send into a small window), then must resend.
*/
if (len)
{
if (len == (int)tp->t_maxseg)
goto send;
if ((idle || tp->t_flags & TF_NODELAY) &&
len + off >= (int)so->so_snd.sb_cc)
{
goto send;
}
if (tp->t_force)
goto send;
if (len >= (int)(tp->max_sndwnd / 2))
goto send;
if (SEQ_LT(tp->snd_nxt, tp->snd_max))
goto send;
}
/*
* Compare available window to amount of window
* known to peer (as advertised window less
* next expected input). If the difference is at least two
* max size segments or at least 35% of the maximum possible
* window, then want to send a window update to peer.
*/
if (win > 0)
{
int adv = (int)win - (int)(tp->rcv_adv - tp->rcv_nxt);
if (so->so_rcv.sb_cc == 0 && adv >= (int)(tp->t_maxseg * 2))
goto send;
if (100 * (u_int)adv / so->so_rcv.sb_hiwat >= 35)
goto send;
}
/*
* TCP window updates are not reliable, rather a polling protocol
* using ``persist'' packets is used to insure receipt of window
* updates. The three ``states'' for the output side are:
* idle not doing retransmits or persists
* persisting to move a small or zero window
* (re)transmitting and thereby not persisting
*
* tp->t_timer[TCPT_PERSIST]
* is set when we are in persist state.
* tp->t_force
* is set when we are called to send a persist packet.
* tp->t_timer[TCPT_REXMT]
* is set when we are retransmitting
* The output side is idle when both timers are zero.
*
* If send window is too small, there is data to transmit, and no
* retransmit or persist is pending, then go to persist state.
* If nothing happens soon, send when timer expires:
* if window is nonzero, transmit what we can,
* otherwise force out a byte.
*/
if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 &&
tp->t_timer[TCPT_PERSIST] == 0)
{
tp->t_rxtshift = 0;
tcp_setpersist(tp);
}
/*
* No reason to send a segment, just return.
*/
return (0);
send:
ENTER_CRIT_SECTION(tp);
/* Limit send length to the current buffer so as to
* avoid doing the "mbuf shuffle" in m_copy().
*/
bufoff = off;
sendm = so->so_snd.sb_mb;
if (len)
{
/* find mbuf containing data to send (at "off") */
while (sendm) /* loop through socket send list */
{
bufoff -= sendm->m_len;
if (bufoff < 0) /* if off is in this buffer, break */
break;
sendm = sendm->m_next;
}
if (!sendm) { dtrap(); /* shouldn't happen */ }
bufoff += sendm->m_len; /* index to next data to send in msend */
/* if socket has multiple unsent mbufs, set flag for send to loop */
if ((sendm->m_next) && (len > (int)sendm->m_len))
{
flags &= ~TH_FIN; /* don't FIN on segment prior to last */
sendalot = 1; /* set to send more segments */
}
if((flags & TH_FIN) && (so->so_snd.sb_cc > (unsigned)len))
{
/* This can happen on slow links (PPP) which retry the last
* segment - the one with the FIN bit attached to data.
*/
flags &= ~TH_FIN; /* don't FIN on segment prior to last */
}
/* only send the rest of msend */
len = min(len, (int)sendm->m_len);
/* if we're not sending starting at sendm->m_data (in which
* case bufoff != 0), then we will copy the data; else we would
* write IP/TCP headers over sent but un-ack'ed data in sendm.
* Similarly, if sendm->m_data is not aligned with respect to
* sendm->m_base and ALIGN_TYPE, we will copy the data to
* ensure that it (and the then-prepended IP/TCP headers) will
* be aligned according to ALIGN_TYPE.
*/
if ((bufoff != 0) || /* data not front aligned in send mbuf? */
(((sendm->m_data - sendm->m_base) & (ALIGN_TYPE - 1)) != 0))
{
len = min(len, (int)(sendm->m_len - bufoff)); /* limit len again */
/* One more test - if this data is not aligned with the front
* of the m_data buffer then we can't use it in place, else we
* might write the IP/TCP header over data that has not yet
* been acked. In this case we must make sure our send
* fits into a little buffer and send what we can.
*/
if ((len > (int)(lilbufsiz - HDRSLEN)) && /* length is bigger the small buffer? */
(bigfreeq.q_len < 2)) /* and we are low on big buffers */
{
len = lilbufsiz - HDRSLEN;
}
}
}
/* if send data is sufficiently aligned in packet, prepend TCP/IP header
* in the space provided.
*/
if (len && (bufoff == 0) &&
(sendm->pkt->inuse == 1) &&
(((sendm->m_data - sendm->m_base) & (ALIGN_TYPE - 1)) == 0) &&
(optlen == 0))
{
/* get an empty mbuf to "clone" the data */
m = m_getnbuf(MT_TXDATA, 0);
if (!m)
{
EXIT_CRIT_SECTION(tp);
return (ENOBUFS);
}
m->pkt = sendm->pkt; /* copy packet location in new mbuf */
m->pkt->inuse++; /* bump packet's use count */
m->m_base = sendm->m_base; /* clone mbuf members */
m->m_memsz = sendm->m_memsz;
m->m_len = len + TCPIPHDRSZ; /* adjust clone for header */
m->m_data = sendm->m_data - TCPIPHDRSZ;
}
else /* either no data or data is not front aligned in mbuf */
{
/* Grab a header mbuf, attaching a copy of data to be
* transmitted, and initialize the header from
* the template for sends on this connection.
*/
m = m_getwithdata (MT_HEADER, IFNETHDR_SIZE + TCPIPHDRSZ);
if (m ==(struct mbuf *)NULL)
{
EXIT_CRIT_SECTION(tp);
return ENOBUFS;
}
m->m_len = TCPIPHDRSZ;
m->m_data += IFNETHDR_SIZE;/* Move this to sizeof tcpip hdr leave*/
/* 14 bytes for ethernet header */
if (len) /* attach any data to send */
{
m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
if (m->m_next == 0)
{
m_freem(m);
EXIT_CRIT_SECTION(tp);
return ENOBUFS;
}
}
}
EXIT_CRIT_SECTION(tp);
if (len)
{
if (tp->t_force && len == 1)
tcpstat.tcps_sndprobe++;
else if (SEQ_LT(tp->snd_nxt, tp->snd_max))
{
tcpstat.tcps_sndrexmitpack++;
tcpstat.tcps_sndrexmitbyte += len;
#ifdef TCP_SACK
if(sack_resend)
tcpstat.tcps_sackresend++;
#endif
}
else
{
tcpstat.tcps_sndpack++;
tcpstat.tcps_sndbyte += len;
}
}
else if (tp->t_flags & TF_ACKNOW)
{
tcpstat.tcps_sndacks++;
}
else if (flags & (TH_SYN|TH_FIN|TH_RST))
tcpstat.tcps_sndctrl++;
else if (SEQ_GT(tp->snd_up, tp->snd_una))
tcpstat.tcps_sndurg++;
else
tcpstat.tcps_sndwinup++;
ti = (struct tcpiphdr *)(m->m_data+sizeof(struct ip)-sizeof(struct ipovly));
if ((char *)ti < m->pkt->nb_buff)
{
panic("tcp_out- packet ptr underflow\n");
}
tcp_mbuf = m; /* flag TCP header mbuf */
#ifdef IP_V6 /* Dual mode code */
if(so->so_domain == AF_INET6)
{
m = mbuf_prepend(m, sizeof(struct ipv6));
if(m == NULL)
{
/* this can happen when we run out of mbufs or pkt buffers
* That is, mfreeq is empty or (lilfreeq, bigfreeq) are empty.
* One solution is to find out which one is getting full and
* then increase them.
*/
dtrap(); /* This is really rare... */
m_freem(tcp_mbuf); /* Free TCP/data chain */
return ENOBUFS;
}
/* strip overlay from front of TCP header */
tcp_mbuf->m_data += sizeof(struct ipovly);
tcp_mbuf->m_len -= sizeof(struct ipovly);
}
#endif /* end IP_V6 */
if (tp->t_template == 0)
panic("tcp_output");
MEMCPY((char*)ti, (char*)tp->t_template, sizeof(struct tcpiphdr));
/*
* Fill in fields, remembering maximum advertised
* window for use in delaying messages about window sizes.
* If resending a FIN, be sure not to use a new sequence number.
*/
if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
tp->snd_nxt == tp->snd_max)
{
tp->snd_nxt--;
}
ti->ti_seq = htonl(tp->snd_nxt);
ti->ti_ack = htonl(tp->rcv_nxt);
/*
* If we're sending a SYN, check the IP address of the interface
* that we will (likely) use to send the IP datagram -- if it's
* changed from what is in the template (as it might if this is
* a retransmission, and the original SYN caused PPP to start
* bringing the interface up, and PPP has got a new IP address
* via IPCP), update the template and the inpcb with the new
* address.
*/
if (flags & TH_SYN)
{
struct inpcb * inp;
inp = (struct inpcb *)so->so_pcb;
switch(so->so_domain)
{
#ifdef IP_V4
case AF_INET:
{
ip_addr src;
#ifdef INCLUDE_PPP
if(((flags & TH_ACK) == 0) && /* SYN only, not SYN/ACK */
(inp->ifp) && /* Make sure we have iface */
(inp->ifp->mib.ifType == PPP)) /* only PPP type */
{
dtrap(); /* remove after confirmed to work in PPP */
src = ip_mymach(ti->ti_dst.s_addr);
if (src != ti->ti_src.s_addr)
{
ti->ti_src.s_addr = src;
tp->t_template->ti_src.s_addr = src;
tp->t_inpcb->inp_laddr.s_addr = src;
}
}
#endif /* INCLUDE_PPP */
/* If this is a SYN (not a SYN/ACK) then set the pmtu */
if((flags & TH_ACK) == 0)
{
#ifdef IP_PMTU
inp->inp_pmtu = pmtucache_get(inp->inp_faddr.s_addr);
#else /* not compiled for pathmtu, guess based on iface */
{
NET ifp;
/* find iface for route. Pass "src" as nexthop return */
ifp = iproute(ti->ti_dst.s_addr, &src);
if(ifp)
inp->inp_pmtu = ifp->n_mtu - (ifp->n_lnh + 40);
else
inp->inp_pmtu = 580; /* Ugh. */
}
#endif /* IP_PMTU */
}
break;
}
#endif /* IP_V4 */
#ifdef IP_V6
case AF_INET6:
{
struct ip6_inaddr * local;
local = ip6_myaddr(&tp->t_inpcb->ip6_faddr, inp->ifp);
/* If we got a local address & it's not the one in the pcb, then
* we assume it changed at the iface and fix it in the pcb. Unlike
* v4, we don't have an IP header yet, not do we have a template
* to worry about.
*/
if((local) &&
(!IP6EQ(&local->addr, &tp->t_inpcb->ip6_laddr)))
{
IP6CPY(&tp->t_inpcb->ip6_laddr, &local->addr);
}
/* If this is a SYN (not a SYN/ACK) then set the pmtu */
if((flags & TH_ACK) == 0)
{
inp->inp_pmtu = ip6_pmtulookup(&inp->ip6_laddr, inp->ifp);
}
break;
}
#endif /* IP_V6 */
default:
dtrap(); /* bad domain setting */
}
}
/* fill in options if any are set */
if (optlen)
{
struct mbuf * mopt;
mopt = m_getwithdata(MT_TXDATA, MAXOPTLEN);
if (mopt == NULL)
{
m_freem(m);
return (ENOBUFS);
}
/* insert options mbuf after after tmp_mbuf */
mopt->m_next = tcp_mbuf->m_next;
tcp_mbuf->m_next = mopt;
/* extend options to aligned address */
while(optlen & 0x03)
tcp_optionbuf[optlen++] = TCPOPT_EOL;
MEMCPY(mtod(mopt, char *), tcp_optionbuf, optlen);
mopt->m_len = optlen;
/* use portable macro to set tcp data offset bits */
SET_TH_OFF(ti->ti_t, ((sizeof (struct tcphdr) + optlen) >> 2));
}
ti->ti_flags = (u_char)flags;
/*
* Calculate receive window. Don't shrink window,
* but avoid silly window syndrome.
*/
if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg)
win = 0;
if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
win = (long)(tp->rcv_adv - tp->rcv_nxt);
/* do check for Iniche buffer limits -JB- */
if (bigfreeq.q_len == 0) /* If queue length is 0, set window to 0 */
{
win = 0;
}
else if(win > (((long)bigfreeq.q_len - 1) * (long)bigbufsiz))
{
win = ((long)bigfreeq.q_len - 1) * bigbufsiz;
}
#ifdef TCP_WIN_SCALE
if(tp->t_flags & TF_WINSCALE)
{
ti->ti_win = htons((u_short)(win >> tp->rcv_wind_scale)); /* apply scale */
}
/*
* Tcp output routine: figure out what should be sent and send it.
*/
int
tcp_output(struct tcpcb *tp)
{
struct inpcb * const inp = tp->t_inpcb;
struct socket *so = inp->inp_socket;
long len, recvwin, sendwin;
int nsacked = 0;
int off, flags, error;
#ifdef TCP_SIGNATURE
int sigoff = 0;
#endif
struct mbuf *m;
struct ip *ip = NULL;
struct ipovly *ipov = NULL;
struct tcphdr *th;
u_char opt[TCP_MAXOLEN];
unsigned int ipoptlen, optlen, hdrlen;
int idle;
boolean_t sendalot;
struct ip6_hdr *ip6 = NULL;
#ifdef INET6
const boolean_t isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
#else
const boolean_t isipv6 = FALSE;
#endif
KKASSERT(so->so_port == &curthread->td_msgport);
/*
* Determine length of data that should be transmitted,
* and flags that will be used.
* If there is some data or critical controls (SYN, RST)
* to send, then transmit; otherwise, investigate further.
*/
/*
* If we have been idle for a while, the send congestion window
* could be no longer representative of the current state of the link.
* So unless we are expecting more acks to come in, slow-start from
* scratch to re-determine the send congestion window.
*/
if (tp->snd_max == tp->snd_una &&
(ticks - tp->t_rcvtime) >= tp->t_rxtcur) {
if (tcp_do_rfc3390) {
int initial_cwnd =
min(4 * tp->t_maxseg, max(2 * tp->t_maxseg, 4380));
tp->snd_cwnd = min(tp->snd_cwnd, initial_cwnd);
} else {
tp->snd_cwnd = tp->t_maxseg;
}
tp->snd_wacked = 0;
}
/*
* Calculate whether the transmit stream was previously idle
* and adjust TF_LASTIDLE for the next time.
*/
idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
if (idle && (tp->t_flags & TF_MORETOCOME))
tp->t_flags |= TF_LASTIDLE;
else
tp->t_flags &= ~TF_LASTIDLE;
if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max &&
!IN_FASTRECOVERY(tp))
nsacked = tcp_sack_bytes_below(&tp->scb, tp->snd_nxt);
again:
/* Make use of SACK information when slow-starting after a RTO. */
if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max &&
!IN_FASTRECOVERY(tp)) {
tcp_seq old_snd_nxt = tp->snd_nxt;
tcp_sack_skip_sacked(&tp->scb, &tp->snd_nxt);
nsacked += tp->snd_nxt - old_snd_nxt;
}
sendalot = FALSE;
off = tp->snd_nxt - tp->snd_una;
sendwin = min(tp->snd_wnd, tp->snd_cwnd + nsacked);
sendwin = min(sendwin, tp->snd_bwnd);
flags = tcp_outflags[tp->t_state];
/*
* Get standard flags, and add SYN or FIN if requested by 'hidden'
* state flags.
*/
if (tp->t_flags & TF_NEEDFIN)
flags |= TH_FIN;
if (tp->t_flags & TF_NEEDSYN)
flags |= TH_SYN;
/*
* If in persist timeout with window of 0, send 1 byte.
* Otherwise, if window is small but nonzero
* and timer expired, we will send what we can
* and go to transmit state.
*/
if (tp->t_flags & TF_FORCE) {
if (sendwin == 0) {
/*
* If we still have some data to send, then
* clear the FIN bit. Usually this would
* happen below when it realizes that we
* aren't sending all the data. However,
* if we have exactly 1 byte of unsent data,
* then it won't clear the FIN bit below,
* and if we are in persist state, we wind
* up sending the packet without recording
* that we sent the FIN bit.
*
* We can't just blindly clear the FIN bit,
* because if we don't have any more data
* to send then the probe will be the FIN
* itself.
*/
if (off < so->so_snd.ssb_cc)
flags &= ~TH_FIN;
sendwin = 1;
} else {
tcp_callout_stop(tp, tp->tt_persist);
tp->t_rxtshift = 0;
}
}
/*
* If snd_nxt == snd_max and we have transmitted a FIN, the
* offset will be > 0 even if so_snd.ssb_cc is 0, resulting in
* a negative length. This can also occur when TCP opens up
* its congestion window while receiving additional duplicate
* acks after fast-retransmit because TCP will reset snd_nxt
* to snd_max after the fast-retransmit.
*
* In the normal retransmit-FIN-only case, however, snd_nxt will
* be set to snd_una, the offset will be 0, and the length may
* wind up 0.
*/
len = (long)ulmin(so->so_snd.ssb_cc, sendwin) - off;
/*
* Lop off SYN bit if it has already been sent. However, if this
* is SYN-SENT state and if segment contains data, suppress sending
* segment (sending the segment would be an option if we still
* did TAO and the remote host supported it).
*/
if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
flags &= ~TH_SYN;
off--, len++;
if (len > 0 && tp->t_state == TCPS_SYN_SENT)
return 0;
}
/*
* Be careful not to send data and/or FIN on SYN segments.
* This measure is needed to prevent interoperability problems
* with not fully conformant TCP implementations.
*/
if (flags & TH_SYN) {
len = 0;
flags &= ~TH_FIN;
}
if (len < 0) {
/*
* If FIN has been sent but not acked,
* but we haven't been called to retransmit,
* len will be < 0. Otherwise, window shrank
* after we sent into it. If window shrank to 0,
* cancel pending retransmit, pull snd_nxt back
* to (closed) window, and set the persist timer
* if it isn't already going. If the window didn't
* close completely, just wait for an ACK.
*/
len = 0;
if (sendwin == 0) {
tcp_callout_stop(tp, tp->tt_rexmt);
tp->t_rxtshift = 0;
tp->snd_nxt = tp->snd_una;
if (!tcp_callout_active(tp, tp->tt_persist))
tcp_setpersist(tp);
}
}
KASSERT(len >= 0, ("%s: len < 0", __func__));
/*
* Automatic sizing of send socket buffer. Often the send buffer
* size is not optimally adjusted to the actual network conditions
* at hand (delay bandwidth product). Setting the buffer size too
* small limits throughput on links with high bandwidth and high
* delay (eg. trans-continental/oceanic links). Setting the
* buffer size too big consumes too much real kernel memory,
* especially with many connections on busy servers.
*
* The criteria to step up the send buffer one notch are:
* 1. receive window of remote host is larger than send buffer
* (with a fudge factor of 5/4th);
* 2. send buffer is filled to 7/8th with data (so we actually
* have data to make use of it);
* 3. send buffer fill has not hit maximal automatic size;
* 4. our send window (slow start and cogestion controlled) is
* larger than sent but unacknowledged data in send buffer.
*
* The remote host receive window scaling factor may limit the
* growing of the send buffer before it reaches its allowed
* maximum.
*
* It scales directly with slow start or congestion window
* and does at most one step per received ACK. This fast
* scaling has the drawback of growing the send buffer beyond
* what is strictly necessary to make full use of a given
* delay*bandwith product. However testing has shown this not
* to be much of an problem. At worst we are trading wasting
* of available bandwith (the non-use of it) for wasting some
* socket buffer memory.
*
* TODO: Shrink send buffer during idle periods together
* with congestion window. Requires another timer. Has to
* wait for upcoming tcp timer rewrite.
*/
if (tcp_do_autosndbuf && so->so_snd.ssb_flags & SSB_AUTOSIZE) {
if ((tp->snd_wnd / 4 * 5) >= so->so_snd.ssb_hiwat &&
so->so_snd.ssb_cc >= (so->so_snd.ssb_hiwat / 8 * 7) &&
so->so_snd.ssb_cc < tcp_autosndbuf_max &&
sendwin >= (so->so_snd.ssb_cc - (tp->snd_nxt - tp->snd_una))) {
u_long newsize;
newsize = ulmin(so->so_snd.ssb_hiwat +
tcp_autosndbuf_inc,
tcp_autosndbuf_max);
if (!ssb_reserve(&so->so_snd, newsize, so, NULL))
atomic_clear_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
if (newsize >= (TCP_MAXWIN << tp->snd_scale))
atomic_clear_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
}
}
/*
* Truncate to the maximum segment length and ensure that FIN is
* removed if the length no longer contains the last data byte.
*/
if (len > tp->t_maxseg) {
len = tp->t_maxseg;
sendalot = TRUE;
}
if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.ssb_cc))
flags &= ~TH_FIN;
recvwin = ssb_space(&so->so_rcv);
/*
* Sender silly window avoidance. We transmit under the following
* conditions when len is non-zero:
*
* - We have a full segment
* - This is the last buffer in a write()/send() and we are
* either idle or running NODELAY
* - we've timed out (e.g. persist timer)
* - we have more then 1/2 the maximum send window's worth of
* data (receiver may be limiting the window size)
* - we need to retransmit
*/
if (len) {
if (len == tp->t_maxseg)
goto send;
/*
* NOTE! on localhost connections an 'ack' from the remote
* end may occur synchronously with the output and cause
* us to flush a buffer queued with moretocome. XXX
*
* note: the len + off check is almost certainly unnecessary.
*/
if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
(idle || (tp->t_flags & TF_NODELAY)) &&
len + off >= so->so_snd.ssb_cc &&
!(tp->t_flags & TF_NOPUSH)) {
goto send;
}
if (tp->t_flags & TF_FORCE) /* typ. timeout case */
goto send;
if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
goto send;
if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
goto send;
}
/*
* Compare available window to amount of window
* known to peer (as advertised window less
* next expected input). If the difference is at least two
* max size segments, or at least 50% of the maximum possible
* window, then want to send a window update to peer.
*/
if (recvwin > 0) {
/*
* "adv" is the amount we can increase the window,
* taking into account that we are limited by
* TCP_MAXWIN << tp->rcv_scale.
*/
long adv = min(recvwin, (long)TCP_MAXWIN << tp->rcv_scale) -
(tp->rcv_adv - tp->rcv_nxt);
long hiwat;
/*
* This ack case typically occurs when the user has drained
* the TCP socket buffer sufficiently to warrent an ack
* containing a 'pure window update'... that is, an ack that
* ONLY updates the tcp window.
*
* It is unclear why we would need to do a pure window update
* past 2 segments if we are going to do one at 1/2 the high
* water mark anyway, especially since under normal conditions
* the user program will drain the socket buffer quickly.
* The 2-segment pure window update will often add a large
* number of extra, unnecessary acks to the stream.
*
* avoid_pure_win_update now defaults to 1.
*/
if (avoid_pure_win_update == 0 ||
(tp->t_flags & TF_RXRESIZED)) {
if (adv >= (long) (2 * tp->t_maxseg)) {
goto send;
}
}
hiwat = (long)(TCP_MAXWIN << tp->rcv_scale);
if (hiwat > (long)so->so_rcv.ssb_hiwat)
hiwat = (long)so->so_rcv.ssb_hiwat;
if (adv >= hiwat / 2)
goto send;
}
/*
* Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
* is also a catch-all for the retransmit timer timeout case.
*/
if (tp->t_flags & TF_ACKNOW)
goto send;
if ((flags & TH_RST) ||
((flags & TH_SYN) && !(tp->t_flags & TF_NEEDSYN)))
goto send;
if (SEQ_GT(tp->snd_up, tp->snd_una))
goto send;
/*
* If our state indicates that FIN should be sent
* and we have not yet done so, then we need to send.
*/
if (flags & TH_FIN &&
(!(tp->t_flags & TF_SENTFIN) || tp->snd_nxt == tp->snd_una))
goto send;
/*
* TCP window updates are not reliable, rather a polling protocol
* using ``persist'' packets is used to insure receipt of window
* updates. The three ``states'' for the output side are:
* idle not doing retransmits or persists
* persisting to move a small or zero window
* (re)transmitting and thereby not persisting
*
* tcp_callout_active(tp, tp->tt_persist)
* is true when we are in persist state.
* The TF_FORCE flag in tp->t_flags
* is set when we are called to send a persist packet.
* tcp_callout_active(tp, tp->tt_rexmt)
* is set when we are retransmitting
* The output side is idle when both timers are zero.
*
* If send window is too small, there is data to transmit, and no
* retransmit or persist is pending, then go to persist state.
* If nothing happens soon, send when timer expires:
* if window is nonzero, transmit what we can,
* otherwise force out a byte.
*/
if (so->so_snd.ssb_cc > 0 &&
!tcp_callout_active(tp, tp->tt_rexmt) &&
!tcp_callout_active(tp, tp->tt_persist)) {
tp->t_rxtshift = 0;
tcp_setpersist(tp);
}
/*
* No reason to send a segment, just return.
*/
return (0);
send:
/*
* Before ESTABLISHED, force sending of initial options
* unless TCP set not to do any options.
* NOTE: we assume that the IP/TCP header plus TCP options
* always fit in a single mbuf, leaving room for a maximum
* link header, i.e.
* max_linkhdr + sizeof(struct tcpiphdr) + optlen <= MCLBYTES
*/
optlen = 0;
if (isipv6)
hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
else
hdrlen = sizeof(struct tcpiphdr);
if (flags & TH_SYN) {
tp->snd_nxt = tp->iss;
if (!(tp->t_flags & TF_NOOPT)) {
u_short mss;
opt[0] = TCPOPT_MAXSEG;
opt[1] = TCPOLEN_MAXSEG;
mss = htons((u_short) tcp_mssopt(tp));
memcpy(opt + 2, &mss, sizeof mss);
optlen = TCPOLEN_MAXSEG;
if ((tp->t_flags & TF_REQ_SCALE) &&
(!(flags & TH_ACK) ||
(tp->t_flags & TF_RCVD_SCALE))) {
*((u_int32_t *)(opt + optlen)) = htonl(
TCPOPT_NOP << 24 |
TCPOPT_WINDOW << 16 |
TCPOLEN_WINDOW << 8 |
tp->request_r_scale);
optlen += 4;
}
if ((tcp_do_sack && !(flags & TH_ACK)) ||
tp->t_flags & TF_SACK_PERMITTED) {
uint32_t *lp = (uint32_t *)(opt + optlen);
*lp = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
optlen += TCPOLEN_SACK_PERMITTED_ALIGNED;
}
}
}
/*
* Send a timestamp and echo-reply if this is a SYN and our side
* wants to use timestamps (TF_REQ_TSTMP is set) or both our side
* and our peer have sent timestamps in our SYN's.
*/
if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
!(flags & TH_RST) &&
(!(flags & TH_ACK) || (tp->t_flags & TF_RCVD_TSTMP))) {
u_int32_t *lp = (u_int32_t *)(opt + optlen);
/* Form timestamp option as shown in appendix A of RFC 1323. */
*lp++ = htonl(TCPOPT_TSTAMP_HDR);
*lp++ = htonl(ticks);
*lp = htonl(tp->ts_recent);
optlen += TCPOLEN_TSTAMP_APPA;
}
/* Set receive buffer autosizing timestamp. */
if (tp->rfbuf_ts == 0 && (so->so_rcv.ssb_flags & SSB_AUTOSIZE))
tp->rfbuf_ts = ticks;
/*
* If this is a SACK connection and we have a block to report,
* fill in the SACK blocks in the TCP options.
*/
if ((tp->t_flags & (TF_SACK_PERMITTED | TF_NOOPT)) ==
TF_SACK_PERMITTED &&
(!LIST_EMPTY(&tp->t_segq) ||
tp->reportblk.rblk_start != tp->reportblk.rblk_end))
tcp_sack_fill_report(tp, opt, &optlen);
#ifdef TCP_SIGNATURE
if (tp->t_flags & TF_SIGNATURE) {
int i;
u_char *bp;
/*
* Initialize TCP-MD5 option (RFC2385)
*/
bp = (u_char *)opt + optlen;
*bp++ = TCPOPT_SIGNATURE;
*bp++ = TCPOLEN_SIGNATURE;
sigoff = optlen + 2;
for (i = 0; i < TCP_SIGLEN; i++)
*bp++ = 0;
optlen += TCPOLEN_SIGNATURE;
/*
* Terminate options list and maintain 32-bit alignment.
*/
*bp++ = TCPOPT_NOP;
*bp++ = TCPOPT_EOL;
optlen += 2;
}
#endif /* TCP_SIGNATURE */
KASSERT(optlen <= TCP_MAXOLEN, ("too many TCP options"));
hdrlen += optlen;
if (isipv6) {
ipoptlen = ip6_optlen(inp);
} else {
if (inp->inp_options) {
ipoptlen = inp->inp_options->m_len -
offsetof(struct ipoption, ipopt_list);
} else {
void
tcp_timer_persist(void *xtp)
{
struct tcpcb *tp = xtp;
struct inpcb *inp;
CURVNET_SET(tp->t_vnet);
#ifdef TCPDEBUG
int ostate;
ostate = tp->t_state;
#endif
inp = tp->t_inpcb;
KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
INP_WLOCK(inp);
if (callout_pending(&tp->t_timers->tt_persist) ||
!callout_active(&tp->t_timers->tt_persist)) {
INP_WUNLOCK(inp);
CURVNET_RESTORE();
return;
}
callout_deactivate(&tp->t_timers->tt_persist);
if ((inp->inp_flags & INP_DROPPED) != 0) {
INP_WUNLOCK(inp);
CURVNET_RESTORE();
return;
}
KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
("%s: tp %p tcpcb can't be stopped here", __func__, tp));
/*
* Persistence timer into zero window.
* Force a byte to be output, if possible.
*/
TCPSTAT_INC(tcps_persisttimeo);
/*
* Hack: if the peer is dead/unreachable, we do not
* time out if the window is closed. After a full
* backoff, drop the connection if the idle time
* (no responses to probes) reaches the maximum
* backoff that we would use if retransmitting.
*/
if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
(ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
TCPSTAT_INC(tcps_persistdrop);
if (tcp_inpinfo_lock_add(inp)) {
tcp_inpinfo_lock_del(inp, tp);
goto out;
}
tp = tcp_drop(tp, ETIMEDOUT);
tcp_inpinfo_lock_del(inp, tp);
goto out;
}
/*
* If the user has closed the socket then drop a persisting
* connection after a much reduced timeout.
*/
if (tp->t_state > TCPS_CLOSE_WAIT &&
(ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
TCPSTAT_INC(tcps_persistdrop);
if (tcp_inpinfo_lock_add(inp)) {
tcp_inpinfo_lock_del(inp, tp);
goto out;
}
tp = tcp_drop(tp, ETIMEDOUT);
tcp_inpinfo_lock_del(inp, tp);
goto out;
}
tcp_setpersist(tp);
tp->t_flags |= TF_FORCEDATA;
(void) tp->t_fb->tfb_tcp_output(tp);
tp->t_flags &= ~TF_FORCEDATA;
#ifdef TCPDEBUG
if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
#endif
TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
INP_WUNLOCK(inp);
out:
CURVNET_RESTORE();
}
void
tcp_timer_persist(void *arg)
{
struct tcpcb *tp = arg;
uint32_t rto;
#ifdef TCP_DEBUG
struct socket *so = NULL;
short ostate;
#endif
mutex_enter(softnet_lock);
if ((tp->t_flags & TF_DEAD) != 0) {
mutex_exit(softnet_lock);
return;
}
if (!callout_expired(&tp->t_timer[TCPT_PERSIST])) {
mutex_exit(softnet_lock);
return;
}
KERNEL_LOCK(1, NULL);
#ifdef TCP_DEBUG
#ifdef INET
if (tp->t_inpcb)
so = tp->t_inpcb->inp_socket;
#endif
#ifdef INET6
if (tp->t_in6pcb)
so = tp->t_in6pcb->in6p_socket;
#endif
ostate = tp->t_state;
#endif /* TCP_DEBUG */
/*
* Persistance timer into zero window.
* Force a byte to be output, if possible.
*/
/*
* Hack: if the peer is dead/unreachable, we do not
* time out if the window is closed. After a full
* backoff, drop the connection if the idle time
* (no responses to probes) reaches the maximum
* backoff that we would use if retransmitting.
*/
rto = TCP_REXMTVAL(tp);
if (rto < tp->t_rttmin)
rto = tp->t_rttmin;
if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
(tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
TCP_STATINC(TCP_STAT_PERSISTDROPS);
tp = tcp_drop(tp, ETIMEDOUT);
goto out;
}
TCP_STATINC(TCP_STAT_PERSISTTIMEO);
tcp_setpersist(tp);
tp->t_force = 1;
(void) tcp_output(tp);
tp->t_force = 0;
out:
#ifdef TCP_DEBUG
if (tp && so->so_options & SO_DEBUG)
tcp_trace(TA_USER, ostate, tp, NULL,
PRU_SLOWTIMO | (TCPT_PERSIST << 8));
#endif
KERNEL_UNLOCK_ONE(NULL);
mutex_exit(softnet_lock);
}
void
tcp_timer_persist(void *xtp)
{
struct tcpcb *tp = xtp;
struct inpcb *inp;
CURVNET_SET(tp->t_vnet);
#ifdef TCPDEBUG
int ostate;
ostate = tp->t_state;
#endif
INP_INFO_WLOCK(&V_tcbinfo);
inp = tp->t_inpcb;
/*
* XXXRW: While this assert is in fact correct, bugs in the tcpcb
* tear-down mean we need it as a work-around for races between
* timers and tcp_discardcb().
*
* KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
*/
if (inp == NULL) {
tcp_timer_race++;
INP_INFO_WUNLOCK(&V_tcbinfo);
CURVNET_RESTORE();
return;
}
INP_WLOCK(inp);
if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_persist)
|| !callout_active(&tp->t_timers->tt_persist)) {
INP_WUNLOCK(inp);
INP_INFO_WUNLOCK(&V_tcbinfo);
CURVNET_RESTORE();
return;
}
callout_deactivate(&tp->t_timers->tt_persist);
/*
* Persistance timer into zero window.
* Force a byte to be output, if possible.
*/
TCPSTAT_INC(tcps_persisttimeo);
/*
* Hack: if the peer is dead/unreachable, we do not
* time out if the window is closed. After a full
* backoff, drop the connection if the idle time
* (no responses to probes) reaches the maximum
* backoff that we would use if retransmitting.
*/
if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
(ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
TCPSTAT_INC(tcps_persistdrop);
tp = tcp_drop(tp, ETIMEDOUT);
goto out;
}
tcp_setpersist(tp);
tp->t_flags |= TF_FORCEDATA;
(void) tcp_output(tp);
tp->t_flags &= ~TF_FORCEDATA;
out:
#ifdef TCPDEBUG
if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
#endif
if (tp != NULL)
INP_WUNLOCK(inp);
INP_INFO_WUNLOCK(&V_tcbinfo);
CURVNET_RESTORE();
}
/*
* Tcp output routine: figure out what should be sent and send it.
*/
int
tcp_output(struct tcpcb *tp)
{
struct inpcb * const inp = tp->t_inpcb;
struct socket *so = inp->inp_socket;
long len, recvwin, sendwin;
int nsacked = 0;
int off, flags, error = 0;
#ifdef TCP_SIGNATURE
int sigoff = 0;
#endif
struct mbuf *m;
struct ip *ip;
struct tcphdr *th;
u_char opt[TCP_MAXOLEN];
unsigned int ipoptlen, optlen, hdrlen;
int idle;
boolean_t sendalot;
struct ip6_hdr *ip6;
#ifdef INET6
const boolean_t isipv6 = INP_ISIPV6(inp);
#else
const boolean_t isipv6 = FALSE;
#endif
boolean_t can_tso = FALSE, use_tso;
boolean_t report_sack, idle_cwv = FALSE;
u_int segsz, tso_hlen, tso_lenmax = 0;
int segcnt = 0;
boolean_t need_sched = FALSE;
KKASSERT(so->so_port == &curthread->td_msgport);
/*
* Determine length of data that should be transmitted,
* and flags that will be used.
* If there is some data or critical controls (SYN, RST)
* to send, then transmit; otherwise, investigate further.
*/
/*
* If we have been idle for a while, the send congestion window
* could be no longer representative of the current state of the
* link; need to validate congestion window. However, we should
* not perform congestion window validation here, since we could
* be asked to send pure ACK.
*/
if (tp->snd_max == tp->snd_una &&
(ticks - tp->snd_last) >= tp->t_rxtcur && tcp_idle_restart)
idle_cwv = TRUE;
/*
* Calculate whether the transmit stream was previously idle
* and adjust TF_LASTIDLE for the next time.
*/
idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
if (idle && (tp->t_flags & TF_MORETOCOME))
tp->t_flags |= TF_LASTIDLE;
else
tp->t_flags &= ~TF_LASTIDLE;
if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max &&
!IN_FASTRECOVERY(tp))
nsacked = tcp_sack_bytes_below(&tp->scb, tp->snd_nxt);
/*
* Find out whether TSO could be used or not
*
* For TSO capable devices, the following assumptions apply to
* the processing of TCP flags:
* - If FIN is set on the large TCP segment, the device must set
* FIN on the last segment that it creates from the large TCP
* segment.
* - If PUSH is set on the large TCP segment, the device must set
* PUSH on the last segment that it creates from the large TCP
* segment.
*/
#if !defined(IPSEC) && !defined(FAST_IPSEC)
if (tcp_do_tso
#ifdef TCP_SIGNATURE
&& (tp->t_flags & TF_SIGNATURE) == 0
#endif
) {
if (!isipv6) {
struct rtentry *rt = inp->inp_route.ro_rt;
if (rt != NULL && (rt->rt_flags & RTF_UP) &&
(rt->rt_ifp->if_hwassist & CSUM_TSO)) {
can_tso = TRUE;
tso_lenmax = rt->rt_ifp->if_tsolen;
}
}
}
#endif /* !IPSEC && !FAST_IPSEC */
again:
m = NULL;
ip = NULL;
th = NULL;
ip6 = NULL;
if ((tp->t_flags & (TF_SACK_PERMITTED | TF_NOOPT)) ==
TF_SACK_PERMITTED &&
(!TAILQ_EMPTY(&tp->t_segq) ||
tp->reportblk.rblk_start != tp->reportblk.rblk_end))
report_sack = TRUE;
else
report_sack = FALSE;
/* Make use of SACK information when slow-starting after a RTO. */
if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max &&
!IN_FASTRECOVERY(tp)) {
tcp_seq old_snd_nxt = tp->snd_nxt;
tcp_sack_skip_sacked(&tp->scb, &tp->snd_nxt);
nsacked += tp->snd_nxt - old_snd_nxt;
}
sendalot = FALSE;
off = tp->snd_nxt - tp->snd_una;
sendwin = min(tp->snd_wnd, tp->snd_cwnd + nsacked);
sendwin = min(sendwin, tp->snd_bwnd);
flags = tcp_outflags[tp->t_state];
/*
* Get standard flags, and add SYN or FIN if requested by 'hidden'
* state flags.
*/
if (tp->t_flags & TF_NEEDFIN)
flags |= TH_FIN;
if (tp->t_flags & TF_NEEDSYN)
flags |= TH_SYN;
/*
* If in persist timeout with window of 0, send 1 byte.
* Otherwise, if window is small but nonzero
* and timer expired, we will send what we can
* and go to transmit state.
*/
if (tp->t_flags & TF_FORCE) {
if (sendwin == 0) {
/*
* If we still have some data to send, then
* clear the FIN bit. Usually this would
* happen below when it realizes that we
* aren't sending all the data. However,
* if we have exactly 1 byte of unsent data,
* then it won't clear the FIN bit below,
* and if we are in persist state, we wind
* up sending the packet without recording
* that we sent the FIN bit.
*
* We can't just blindly clear the FIN bit,
* because if we don't have any more data
* to send then the probe will be the FIN
* itself.
*/
if (off < so->so_snd.ssb_cc)
flags &= ~TH_FIN;
sendwin = 1;
} else {
tcp_callout_stop(tp, tp->tt_persist);
tp->t_rxtshift = 0;
}
}
/*
* If snd_nxt == snd_max and we have transmitted a FIN, the
* offset will be > 0 even if so_snd.ssb_cc is 0, resulting in
* a negative length. This can also occur when TCP opens up
* its congestion window while receiving additional duplicate
* acks after fast-retransmit because TCP will reset snd_nxt
* to snd_max after the fast-retransmit.
*
* A negative length can also occur when we are in the
* TCPS_SYN_RECEIVED state due to a simultanious connect where
* our SYN has not been acked yet.
*
* In the normal retransmit-FIN-only case, however, snd_nxt will
* be set to snd_una, the offset will be 0, and the length may
* wind up 0.
*/
len = (long)ulmin(so->so_snd.ssb_cc, sendwin) - off;
/*
* Lop off SYN bit if it has already been sent. However, if this
* is SYN-SENT state and if segment contains data, suppress sending
* segment (sending the segment would be an option if we still
* did TAO and the remote host supported it).
*/
if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
flags &= ~TH_SYN;
off--, len++;
if (len > 0 && tp->t_state == TCPS_SYN_SENT) {
tp->t_flags &= ~(TF_ACKNOW | TF_XMITNOW);
return 0;
}
}
/*
* Be careful not to send data and/or FIN on SYN segments.
* This measure is needed to prevent interoperability problems
* with not fully conformant TCP implementations.
*/
if (flags & TH_SYN) {
len = 0;
flags &= ~TH_FIN;
}
if (len < 0) {
/*
* A negative len can occur if our FIN has been sent but not
* acked, or if we are in a simultanious connect in the
* TCPS_SYN_RECEIVED state with our SYN sent but not yet
* acked.
*
* If our window has contracted to 0 in the FIN case
* (which can only occur if we have NOT been called to
* retransmit as per code a few paragraphs up) then we
* want to shift the retransmit timer over to the
* persist timer.
*
* However, if we are in the TCPS_SYN_RECEIVED state
* (the SYN case) we will be in a simultanious connect and
* the window may be zero degeneratively. In this case we
* do not want to shift to the persist timer after the SYN
* or the SYN+ACK transmission.
*/
len = 0;
if (sendwin == 0 && tp->t_state != TCPS_SYN_RECEIVED) {
tcp_callout_stop(tp, tp->tt_rexmt);
tp->t_rxtshift = 0;
tp->snd_nxt = tp->snd_una;
if (!tcp_callout_active(tp, tp->tt_persist))
tcp_setpersist(tp);
}
}
KASSERT(len >= 0, ("%s: len < 0", __func__));
/*
* Automatic sizing of send socket buffer. Often the send buffer
* size is not optimally adjusted to the actual network conditions
* at hand (delay bandwidth product). Setting the buffer size too
* small limits throughput on links with high bandwidth and high
* delay (eg. trans-continental/oceanic links). Setting the
* buffer size too big consumes too much real kernel memory,
* especially with many connections on busy servers.
*
* The criteria to step up the send buffer one notch are:
* 1. receive window of remote host is larger than send buffer
* (with a fudge factor of 5/4th);
* 2. hiwat has not significantly exceeded bwnd (inflight)
* (bwnd is a maximal value if inflight is disabled).
* 3. send buffer is filled to 7/8th with data (so we actually
* have data to make use of it);
* 4. hiwat has not hit maximal automatic size;
* 5. our send window (slow start and cogestion controlled) is
* larger than sent but unacknowledged data in send buffer.
*
* The remote host receive window scaling factor may limit the
* growing of the send buffer before it reaches its allowed
* maximum.
*
* It scales directly with slow start or congestion window
* and does at most one step per received ACK. This fast
* scaling has the drawback of growing the send buffer beyond
* what is strictly necessary to make full use of a given
* delay*bandwith product. However testing has shown this not
* to be much of an problem. At worst we are trading wasting
* of available bandwith (the non-use of it) for wasting some
* socket buffer memory.
*
* The criteria for shrinking the buffer is based solely on
* the inflight code (snd_bwnd). If inflight is disabled,
* the buffer will not be shrinked. Note that snd_bwnd already
* has a fudge factor. Our test adds a little hysteresis.
*/
if (tcp_do_autosndbuf && (so->so_snd.ssb_flags & SSB_AUTOSIZE)) {
const int asbinc = tcp_autosndbuf_inc;
const int hiwat = so->so_snd.ssb_hiwat;
const int lowat = so->so_snd.ssb_lowat;
u_long newsize;
if ((tp->snd_wnd / 4 * 5) >= hiwat &&
so->so_snd.ssb_cc >= (hiwat / 8 * 7) &&
hiwat < tp->snd_bwnd + hiwat / 10 &&
hiwat + asbinc < tcp_autosndbuf_max &&
hiwat < (TCP_MAXWIN << tp->snd_scale) &&
sendwin >= (so->so_snd.ssb_cc -
(tp->snd_nxt - tp->snd_una))) {
newsize = ulmin(hiwat + asbinc, tcp_autosndbuf_max);
if (!ssb_reserve(&so->so_snd, newsize, so, NULL))
atomic_clear_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
#if 0
if (newsize >= (TCP_MAXWIN << tp->snd_scale))
atomic_clear_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
#endif
} else if ((long)tp->snd_bwnd <
(long)(hiwat * 3 / 4 - lowat - asbinc) &&
hiwat > tp->t_maxseg * 2 + asbinc &&
hiwat + asbinc >= tcp_autosndbuf_min &&
tcp_do_autosndbuf == 1) {
newsize = ulmax(hiwat - asbinc, tp->t_maxseg * 2);
ssb_reserve(&so->so_snd, newsize, so, NULL);
}
}
/*
* Don't use TSO, if:
* - Congestion window needs validation
* - There are SACK blocks to report
* - RST or SYN flags is set
* - URG will be set
*
* XXX
* Checking for SYN|RST looks overkill, just to be safe than sorry
*/
use_tso = can_tso;
if (report_sack || idle_cwv || (flags & (TH_RST | TH_SYN)))
use_tso = FALSE;
if (use_tso) {
tcp_seq ugr_nxt = tp->snd_nxt;
if ((flags & TH_FIN) && (tp->t_flags & TF_SENTFIN) &&
tp->snd_nxt == tp->snd_max)
--ugr_nxt;
if (SEQ_GT(tp->snd_up, ugr_nxt))
use_tso = FALSE;
}
if (use_tso) {
/*
* Find out segment size and header length for TSO
*/
error = tcp_tso_getsize(tp, &segsz, &tso_hlen);
if (error)
use_tso = FALSE;
}
if (!use_tso) {
segsz = tp->t_maxseg;
tso_hlen = 0; /* not used */
}
/*
* Truncate to the maximum segment length if not TSO, and ensure that
* FIN is removed if the length no longer contains the last data byte.
*/
if (len > segsz) {
if (!use_tso) {
len = segsz;
++segcnt;
} else {
int nsegs;
if (__predict_false(tso_lenmax < segsz))
tso_lenmax = segsz << 1;
/*
* Truncate TSO transfers to (IP_MAXPACKET - iphlen -
* thoff), and make sure that we send equal size
* transfers down the stack (rather than big-small-
* big-small-...).
*/
len = min(len, tso_lenmax);
nsegs = min(len, (IP_MAXPACKET - tso_hlen)) / segsz;
KKASSERT(nsegs > 0);
len = nsegs * segsz;
if (len <= segsz) {
use_tso = FALSE;
++segcnt;
} else {
segcnt += nsegs;
}
}
sendalot = TRUE;
} else {
use_tso = FALSE;
if (len > 0)
++segcnt;
}
if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.ssb_cc))
flags &= ~TH_FIN;
recvwin = ssb_space(&so->so_rcv);
/*
* Sender silly window avoidance. We transmit under the following
* conditions when len is non-zero:
*
* - We have a full segment
* - This is the last buffer in a write()/send() and we are
* either idle or running NODELAY
* - we've timed out (e.g. persist timer)
* - we have more then 1/2 the maximum send window's worth of
* data (receiver may be limiting the window size)
* - we need to retransmit
*/
if (len) {
if (len >= segsz)
goto send;
/*
* NOTE! on localhost connections an 'ack' from the remote
* end may occur synchronously with the output and cause
* us to flush a buffer queued with moretocome. XXX
*
* note: the len + off check is almost certainly unnecessary.
*/
if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
(idle || (tp->t_flags & TF_NODELAY)) &&
len + off >= so->so_snd.ssb_cc &&
!(tp->t_flags & TF_NOPUSH)) {
goto send;
}
if (tp->t_flags & TF_FORCE) /* typ. timeout case */
goto send;
if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
goto send;
if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
goto send;
if (tp->t_flags & TF_XMITNOW)
goto send;
}
/*
* Compare available window to amount of window
* known to peer (as advertised window less
* next expected input). If the difference is at least two
* max size segments, or at least 50% of the maximum possible
* window, then want to send a window update to peer.
*/
if (recvwin > 0) {
/*
* "adv" is the amount we can increase the window,
* taking into account that we are limited by
* TCP_MAXWIN << tp->rcv_scale.
*/
long adv = min(recvwin, (long)TCP_MAXWIN << tp->rcv_scale) -
(tp->rcv_adv - tp->rcv_nxt);
long hiwat;
/*
* This ack case typically occurs when the user has drained
* the TCP socket buffer sufficiently to warrent an ack
* containing a 'pure window update'... that is, an ack that
* ONLY updates the tcp window.
*
* It is unclear why we would need to do a pure window update
* past 2 segments if we are going to do one at 1/2 the high
* water mark anyway, especially since under normal conditions
* the user program will drain the socket buffer quickly.
* The 2-segment pure window update will often add a large
* number of extra, unnecessary acks to the stream.
*
* avoid_pure_win_update now defaults to 1.
*/
if (avoid_pure_win_update == 0 ||
(tp->t_flags & TF_RXRESIZED)) {
if (adv >= (long) (2 * segsz)) {
goto send;
}
}
hiwat = (long)(TCP_MAXWIN << tp->rcv_scale);
if (hiwat > (long)so->so_rcv.ssb_hiwat)
hiwat = (long)so->so_rcv.ssb_hiwat;
if (adv >= hiwat / 2)
goto send;
}
/*
* Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
* is also a catch-all for the retransmit timer timeout case.
*/
if (tp->t_flags & TF_ACKNOW)
goto send;
if ((flags & TH_RST) ||
((flags & TH_SYN) && !(tp->t_flags & TF_NEEDSYN)))
goto send;
if (SEQ_GT(tp->snd_up, tp->snd_una))
goto send;
/*
* If our state indicates that FIN should be sent
* and we have not yet done so, then we need to send.
*/
if ((flags & TH_FIN) &&
(!(tp->t_flags & TF_SENTFIN) || tp->snd_nxt == tp->snd_una))
goto send;
/*
* TCP window updates are not reliable, rather a polling protocol
* using ``persist'' packets is used to insure receipt of window
* updates. The three ``states'' for the output side are:
* idle not doing retransmits or persists
* persisting to move a small or zero window
* (re)transmitting and thereby not persisting
*
* tcp_callout_active(tp, tp->tt_persist)
* is true when we are in persist state.
* The TF_FORCE flag in tp->t_flags
* is set when we are called to send a persist packet.
* tcp_callout_active(tp, tp->tt_rexmt)
* is set when we are retransmitting
* The output side is idle when both timers are zero.
*
* If send window is too small, there is data to transmit, and no
* retransmit or persist is pending, then go to persist state.
*
* If nothing happens soon, send when timer expires:
* if window is nonzero, transmit what we can, otherwise force out
* a byte.
*
* Don't try to set the persist state if we are in TCPS_SYN_RECEIVED
* with data pending. This situation can occur during a
* simultanious connect.
*/
if (so->so_snd.ssb_cc > 0 &&
tp->t_state != TCPS_SYN_RECEIVED &&
!tcp_callout_active(tp, tp->tt_rexmt) &&
!tcp_callout_active(tp, tp->tt_persist)) {
tp->t_rxtshift = 0;
tcp_setpersist(tp);
}
/*
* No reason to send a segment, just return.
*/
tp->t_flags &= ~TF_XMITNOW;
return (0);
send:
if (need_sched && len > 0) {
tcp_output_sched(tp);
return 0;
}
/*
* Before ESTABLISHED, force sending of initial options
* unless TCP set not to do any options.
* NOTE: we assume that the IP/TCP header plus TCP options
* always fit in a single mbuf, leaving room for a maximum
* link header, i.e.
* max_linkhdr + sizeof(struct tcpiphdr) + optlen <= MCLBYTES
*/
optlen = 0;
if (isipv6)
hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
else
hdrlen = sizeof(struct tcpiphdr);
if (flags & TH_SYN) {
tp->snd_nxt = tp->iss;
if (!(tp->t_flags & TF_NOOPT)) {
u_short mss;
opt[0] = TCPOPT_MAXSEG;
opt[1] = TCPOLEN_MAXSEG;
mss = htons((u_short) tcp_mssopt(tp));
memcpy(opt + 2, &mss, sizeof mss);
optlen = TCPOLEN_MAXSEG;
if ((tp->t_flags & TF_REQ_SCALE) &&
(!(flags & TH_ACK) ||
(tp->t_flags & TF_RCVD_SCALE))) {
*((u_int32_t *)(opt + optlen)) = htonl(
TCPOPT_NOP << 24 |
TCPOPT_WINDOW << 16 |
TCPOLEN_WINDOW << 8 |
tp->request_r_scale);
optlen += 4;
}
if ((tcp_do_sack && !(flags & TH_ACK)) ||
tp->t_flags & TF_SACK_PERMITTED) {
uint32_t *lp = (uint32_t *)(opt + optlen);
*lp = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
optlen += TCPOLEN_SACK_PERMITTED_ALIGNED;
}
}
}
/*
* Send a timestamp and echo-reply if this is a SYN and our side
* wants to use timestamps (TF_REQ_TSTMP is set) or both our side
* and our peer have sent timestamps in our SYN's.
*/
if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
!(flags & TH_RST) &&
(!(flags & TH_ACK) || (tp->t_flags & TF_RCVD_TSTMP))) {
u_int32_t *lp = (u_int32_t *)(opt + optlen);
/* Form timestamp option as shown in appendix A of RFC 1323. */
*lp++ = htonl(TCPOPT_TSTAMP_HDR);
*lp++ = htonl(ticks);
*lp = htonl(tp->ts_recent);
optlen += TCPOLEN_TSTAMP_APPA;
}
/* Set receive buffer autosizing timestamp. */
if (tp->rfbuf_ts == 0 && (so->so_rcv.ssb_flags & SSB_AUTOSIZE))
tp->rfbuf_ts = ticks;
/*
* If this is a SACK connection and we have a block to report,
* fill in the SACK blocks in the TCP options.
*/
if (report_sack)
tcp_sack_fill_report(tp, opt, &optlen);
#ifdef TCP_SIGNATURE
if (tp->t_flags & TF_SIGNATURE) {
int i;
u_char *bp;
/*
* Initialize TCP-MD5 option (RFC2385)
*/
bp = (u_char *)opt + optlen;
*bp++ = TCPOPT_SIGNATURE;
*bp++ = TCPOLEN_SIGNATURE;
sigoff = optlen + 2;
for (i = 0; i < TCP_SIGLEN; i++)
*bp++ = 0;
optlen += TCPOLEN_SIGNATURE;
/*
* Terminate options list and maintain 32-bit alignment.
*/
*bp++ = TCPOPT_NOP;
*bp++ = TCPOPT_EOL;
optlen += 2;
}
#endif /* TCP_SIGNATURE */
KASSERT(optlen <= TCP_MAXOLEN, ("too many TCP options"));
hdrlen += optlen;
if (isipv6) {
ipoptlen = ip6_optlen(inp);
} else {
if (inp->inp_options) {
ipoptlen = inp->inp_options->m_len -
offsetof(struct ipoption, ipopt_list);
} else {
//-------------------------------------------------------------------------//
// TCP timer processing.
//-------------------------------------------------------------------------//
static
struct tcpcb * tcp_timers(
Node *node,
struct tcpcb *tp,
int timer,
UInt32 tcp_now,
struct tcpstat *tcp_stat)
{
int rexmt;
TransportDataTcp *tcpLayer = (TransportDataTcp *)
node->transportData.tcp;
switch (timer) {
//
// 2 MSL timeout in shutdown went off. If we're closed but
// still waiting for peer to close and connection has been idle
// too long, or if 2MSL time is up from TIME_WAIT, delete connection
// control block. Otherwise, check again in a bit.
//
case TCPT_2MSL:
if (tp->t_state != TCPS_TIME_WAIT &&
tp->t_idle <= TCPTV_MAXIDLE)
tp->t_timer[TCPT_2MSL] = TCPTV_KEEPINTVL;
else {
// printf("TCP: Connection closed by timer\n");
tp = tcp_close(node, tp, tcp_stat);
}
break;
//
// Retransmission timer went off. Message has not
// been acked within retransmit interval. Back off
// to a longer retransmit interval and retransmit one segment.
//
case TCPT_REXMT:
if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
tp->t_rxtshift = TCP_MAXRXTSHIFT;
//if (tcp_stat)
//tcp_stat->tcps_timeoutdrop++;
printf("TCP: Retransmission timer went off\n");
tp = tcp_drop(node, tp, tcp_now, tcp_stat);
break;
}
//if (tcp_stat)
//tcp_stat->tcps_rexmttimeo++;
rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
TCPT_RANGESET(tp->t_rxtcur, rexmt,
tp->t_rttmin, TCPTV_REXMTMAX);
tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
//
// If we backed off this far,
// our srtt estimate is probably bogus. Clobber it
// so we'll take the next rtt measurement as our srtt;
// move the current srtt into rttvar to keep the current
// retransmit times until then.
//
if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
tp->t_rttvar +=
(tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
tp->t_srtt = 0;
}
tp->snd_nxt = tp->snd_una;
if (TCP_VARIANT_IS_SACK(tp) && tp->isSackFastRextOn) {
TransportTcpSackRextTimeoutInit(tp);
TransportTcpTrace(node, 0, 0, "Faxt: timeout");
}
// Force a segment to be sent.
tp->t_flags |= TF_ACKNOW;
// If timing a segment in this window, stop the timer.
// The retransmitted segment shouldn't be timed.
tp->t_rtt = 0;
//
// Close the congestion window down to one segment
// (we'll open it by one segment for each ack we get).
// Since we probably have a window's worth of unacked
// data accumulated, this "slow start" keeps us from
// dumping all that data as back-to-back packets (which
// might overwhelm an intermediate gateway).
//
// There are two phases to the opening: Initially we
// open by one mss on each ack. This makes the window
// size increase exponentially with time. If the
// window is larger than the path can handle, this
// exponential growth results in dropped packet(s)
// almost immediately. To get more time between
// drops but still "push" the network to take advantage
// of improving conditions, we switch from exponential
// to linear window opening at some threshhold size.
// For a threshhold, we use half the current window
// size, truncated to a multiple of the mss.
//
// (the minimum cwnd that will give us exponential
// growth is 2 mss. We don't allow the threshhold
// to go below this.)
//
{
unsigned int win;
win = MIN(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
if (win < 2)
win = 2;
tp->snd_cwnd = tp->t_maxseg;
tp->snd_ssthresh = win * tp->t_maxseg;
tp->t_partialacks = -1;
tp->t_dupacks = 0;
}
tp->t_ecnFlags |= TF_CWND_REDUCED;
TransportTcpTrace(node, 0, 0, "Rext: timeout");
//
// To eliminates the problem of multiple Fast Retransmits we uses this
// new variable "send_high", whose initial value is the initial send
// sequence number. After each retransmit timeout, the highest sequence
// numbers transmitted so far is recorded in the variable "send_high".
//
if (TCP_VARIANT_IS_NEWRENO(tp)) {
tp->send_high = tp->snd_max;
}
tcp_output(node, tp, tcp_now, tcp_stat);
break;
//
// Persistance timer into zero window.
// Force a byte to be output, if possible.
//
case TCPT_PERSIST:
//if (tcp_stat)
//tcp_stat->tcps_persisttimeo++;
//
// Hack: if the peer is dead/unreachable, we do not
// time out if the window is closed. After a full
// backoff, drop the connection if the idle time
// (no responses to probes) reaches the maximum
// backoff that we would use if retransmitting.
//
if (tp->t_rxtshift == TCP_MAXRXTSHIFT) {
UInt32 maxidle = TCP_REXMTVAL(tp);
if (maxidle < tp->t_rttmin)
maxidle = tp->t_rttmin;
maxidle *= tcp_totbackoff;
if (tp->t_idle >= TCPTV_KEEP_IDLE ||
tp->t_idle >= maxidle) {
//if (tcp_stat)
//tcp_stat->tcps_persistdrop++;
printf("TCP: Idle timer went off\n");
tp = tcp_drop(node, tp, tcp_now, tcp_stat);
break;
}
}
tcp_setpersist(tp);
tp->t_force = 1;
tcp_output(node, tp, tcp_now, tcp_stat);
tp->t_force = 0;
break;
//
// Keep-alive timer went off; send something
// or drop connection if idle for too long.
//
case TCPT_KEEP:
//if (tcp_stat)
//tcp_stat->tcps_keeptimeo++;
if (tp->t_state < TCPS_ESTABLISHED)
printf("TCP: Keep-alive timer went off before established\n");
goto dropit;
if (tcpLayer->tcpUseKeepAliveProbes && tp->t_state <= TCPS_CLOSING) {
//
// If the connection has been idle for more than the sum of
// TCPTV_KEEP_IDLE (set to 2 hours) and TCPTV_MAXIDLE
// (set to the total time taken to send all the probes),
// it's time to drop the connection.
//
if (tp->t_idle >= TCPTV_KEEP_IDLE + TCPTV_MAXIDLE)
printf("TCP: Keep-alive timer went off\n");
goto dropit;
//
// Send a packet designed to force a response
// if the peer is up and reachable:
// either an ACK if the connection is still alive,
// or an RST if the peer has closed the connection
// due to timeout or reboot.
// Using sequence number tp->snd_una-1
// causes the transmitted zero-length segment
// to lie outside the receive window;
// by the protocol spec, this requires the
// correspondent TCP to respond.
//
//if (tcp_stat)
//tcp_stat->tcps_keepprobe++;
tcp_respond(node, tp, tp->t_template,
0, tp->rcv_nxt, tp->snd_una - 1,
0, tcp_stat);
tp->t_timer[TCPT_KEEP] = TCPTV_KEEPINTVL;
} else {
//
// If the tcpUseKeepAliveProbes is FALSE
// or the connection state is greater than TCPS_CLOSING,
// reset the keepalive timer to TCPTV_KEEP_IDLE.
//
tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_IDLE;
}
break;
dropit:
//if (tcp_stat) {
//
// Note that this counter counts connection drops due to
// failure in connection establishment and the keepalive
// timer timeouts
//
//tcp_stat->tcps_keepdrops++;
//}
// printf("TCP: Unknown timer went off\n");
tp = tcp_drop(node, tp, tcp_now, tcp_stat);
break;
}
/*
* Tcp output routine: figure out what should be sent and send it.
*/
int
tcp_output(struct tcpcb *tp)
{
struct socket *so = tp->t_inpcb->inp_socket;
long len, recwin, sendwin;
int off, flags, error;
#ifdef TCP_SIGNATURE
int sigoff = 0;
#endif
struct mbuf *m;
struct ip *ip = NULL;
struct ipovly *ipov = NULL;
struct tcphdr *th;
u_char opt[TCP_MAXOLEN];
unsigned ipoptlen, optlen, hdrlen;
int idle, sendalot;
int i, sack_rxmit;
int sack_bytes_rxmt;
struct sackhole *p;
#if 0
int maxburst = TCP_MAXBURST;
#endif
struct rmxp_tao tao;
#ifdef INET6
struct ip6_hdr *ip6 = NULL;
int isipv6;
bzero(&tao, sizeof(tao));
isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
#endif
#ifdef TCP_ECN
int needect;
#endif
INP_LOCK_ASSERT(tp->t_inpcb);
/*
* Determine length of data that should be transmitted,
* and flags that will be used.
* If there is some data or critical controls (SYN, RST)
* to send, then transmit; otherwise, investigate further.
*/
idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
if (idle && (ticks - tp->t_rcvtime) >= tp->t_rxtcur) {
/*
* We have been idle for "a while" and no acks are
* expected to clock out any data we send --
* slow start to get ack "clock" running again.
*
* Set the slow-start flight size depending on whether
* this is a local network or not.
*/
int ss = ss_fltsz;
#ifdef INET6
if (isipv6) {
if (in6_localaddr(&tp->t_inpcb->in6p_faddr))
ss = ss_fltsz_local;
} else
#endif
if (in_localaddr(tp->t_inpcb->inp_faddr))
ss = ss_fltsz_local;
tp->snd_cwnd = tp->t_maxseg * ss;
}
tp->t_flags &= ~TF_LASTIDLE;
if (idle) {
if (tp->t_flags & TF_MORETOCOME) {
tp->t_flags |= TF_LASTIDLE;
idle = 0;
}
}
again:
/*
* If we've recently taken a timeout, snd_max will be greater than
* snd_nxt. There may be SACK information that allows us to avoid
* resending already delivered data. Adjust snd_nxt accordingly.
*/
if (tp->sack_enable && SEQ_LT(tp->snd_nxt, tp->snd_max))
tcp_sack_adjust(tp);
sendalot = 0;
off = tp->snd_nxt - tp->snd_una;
sendwin = min(tp->snd_wnd, tp->snd_cwnd);
sendwin = min(sendwin, tp->snd_bwnd);
flags = tcp_outflags[tp->t_state];
/*
* Send any SACK-generated retransmissions. If we're explicitly trying
* to send out new data (when sendalot is 1), bypass this function.
* If we retransmit in fast recovery mode, decrement snd_cwnd, since
* we're replacing a (future) new transmission with a retransmission
* now, and we previously incremented snd_cwnd in tcp_input().
*/
/*
* Still in sack recovery , reset rxmit flag to zero.
*/
sack_rxmit = 0;
sack_bytes_rxmt = 0;
len = 0;
p = NULL;
if (tp->sack_enable && IN_FASTRECOVERY(tp) &&
(p = tcp_sack_output(tp, &sack_bytes_rxmt))) {
long cwin;
cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt;
if (cwin < 0)
cwin = 0;
/* Do not retransmit SACK segments beyond snd_recover */
if (SEQ_GT(p->end, tp->snd_recover)) {
/*
* (At least) part of sack hole extends beyond
* snd_recover. Check to see if we can rexmit data
* for this hole.
*/
if (SEQ_GEQ(p->rxmit, tp->snd_recover)) {
/*
* Can't rexmit any more data for this hole.
* That data will be rexmitted in the next
* sack recovery episode, when snd_recover
* moves past p->rxmit.
*/
p = NULL;
goto after_sack_rexmit;
} else
/* Can rexmit part of the current hole */
len = ((long)ulmin(cwin,
tp->snd_recover - p->rxmit));
} else
len = ((long)ulmin(cwin, p->end - p->rxmit));
off = p->rxmit - tp->snd_una;
KASSERT(off >= 0,("%s: sack block to the left of una : %d",
__func__, off));
if (len > 0) {
sack_rxmit = 1;
sendalot = 1;
tcpstat.tcps_sack_rexmits++;
tcpstat.tcps_sack_rexmit_bytes +=
min(len, tp->t_maxseg);
}
}
after_sack_rexmit:
/*
* Get standard flags, and add SYN or FIN if requested by 'hidden'
* state flags.
*/
if (tp->t_flags & TF_NEEDFIN)
flags |= TH_FIN;
if (tp->t_flags & TF_NEEDSYN)
flags |= TH_SYN;
SOCKBUF_LOCK(&so->so_snd);
/*
* If in persist timeout with window of 0, send 1 byte.
* Otherwise, if window is small but nonzero
* and timer expired, we will send what we can
* and go to transmit state.
*/
if (tp->t_force) {
if (sendwin == 0) {
/*
* If we still have some data to send, then
* clear the FIN bit. Usually this would
* happen below when it realizes that we
* aren't sending all the data. However,
* if we have exactly 1 byte of unsent data,
* then it won't clear the FIN bit below,
* and if we are in persist state, we wind
* up sending the packet without recording
* that we sent the FIN bit.
*
* We can't just blindly clear the FIN bit,
* because if we don't have any more data
* to send then the probe will be the FIN
* itself.
*/
if (off < so->so_snd.sb_cc)
flags &= ~TH_FIN;
sendwin = 1;
} else {
callout_stop(tp->tt_persist);
tp->t_rxtshift = 0;
}
}
/*
* If snd_nxt == snd_max and we have transmitted a FIN, the
* offset will be > 0 even if so_snd.sb_cc is 0, resulting in
* a negative length. This can also occur when TCP opens up
* its congestion window while receiving additional duplicate
* acks after fast-retransmit because TCP will reset snd_nxt
* to snd_max after the fast-retransmit.
*
* In the normal retransmit-FIN-only case, however, snd_nxt will
* be set to snd_una, the offset will be 0, and the length may
* wind up 0.
*
* If sack_rxmit is true we are retransmitting from the scoreboard
* in which case len is already set.
*/
if (sack_rxmit == 0) {
if (sack_bytes_rxmt == 0)
len = ((long)ulmin(so->so_snd.sb_cc, sendwin) - off);
else {
long cwin;
/*
* We are inside of a SACK recovery episode and are
* sending new data, having retransmitted all the
* data possible in the scoreboard.
*/
len = ((long)ulmin(so->so_snd.sb_cc, tp->snd_wnd)
- off);
/*
* Don't remove this (len > 0) check !
* We explicitly check for len > 0 here (although it
* isn't really necessary), to work around a gcc
* optimization issue - to force gcc to compute
* len above. Without this check, the computation
* of len is bungled by the optimizer.
*/
if (len > 0) {
cwin = tp->snd_cwnd -
(tp->snd_nxt - tp->sack_newdata) -
sack_bytes_rxmt;
if (cwin < 0)
cwin = 0;
len = lmin(len, cwin);
}
}
}
/*
* Lop off SYN bit if it has already been sent. However, if this
* is SYN-SENT state and if segment contains data and if we don't
* know that foreign host supports TAO, suppress sending segment.
*/
if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
flags &= ~TH_SYN;
off--, len++;
if (tcp_do_rfc1644)
tcp_hc_gettao(&tp->t_inpcb->inp_inc, &tao);
if (len > 0 && tp->t_state == TCPS_SYN_SENT &&
tao.tao_ccsent == 0)
goto just_return;
}
/*
* Be careful not to send data and/or FIN on SYN segments
* in cases when no CC option will be sent.
* This measure is needed to prevent interoperability problems
* with not fully conformant TCP implementations.
*/
if ((flags & TH_SYN) &&
((tp->t_flags & TF_NOOPT) || !(tp->t_flags & TF_REQ_CC) ||
((flags & TH_ACK) && !(tp->t_flags & TF_RCVD_CC)))) {
len = 0;
flags &= ~TH_FIN;
}
if (len < 0) {
/*
* If FIN has been sent but not acked,
* but we haven't been called to retransmit,
* len will be < 0. Otherwise, window shrank
* after we sent into it. If window shrank to 0,
* cancel pending retransmit, pull snd_nxt back
* to (closed) window, and set the persist timer
* if it isn't already going. If the window didn't
* close completely, just wait for an ACK.
*/
len = 0;
if (sendwin == 0) {
callout_stop(tp->tt_rexmt);
tp->t_rxtshift = 0;
tp->snd_nxt = tp->snd_una;
if (!callout_active(tp->tt_persist))
tcp_setpersist(tp);
}
}
/*
* len will be >= 0 after this point. Truncate to the maximum
* segment length and ensure that FIN is removed if the length
* no longer contains the last data byte.
*/
if (len > tp->t_maxseg) {
len = tp->t_maxseg;
sendalot = 1;
}
if (sack_rxmit) {
if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc))
flags &= ~TH_FIN;
} else {
if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
flags &= ~TH_FIN;
}
recwin = sbspace(&so->so_rcv);
/*
* Sender silly window avoidance. We transmit under the following
* conditions when len is non-zero:
*
* - We have a full segment
* - This is the last buffer in a write()/send() and we are
* either idle or running NODELAY
* - we've timed out (e.g. persist timer)
* - we have more then 1/2 the maximum send window's worth of
* data (receiver may be limited the window size)
* - we need to retransmit
*/
if (len) {
if (len == tp->t_maxseg)
goto send;
/*
* NOTE! on localhost connections an 'ack' from the remote
* end may occur synchronously with the output and cause
* us to flush a buffer queued with moretocome. XXX
*
* note: the len + off check is almost certainly unnecessary.
*/
if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
(idle || (tp->t_flags & TF_NODELAY)) &&
len + off >= so->so_snd.sb_cc &&
(tp->t_flags & TF_NOPUSH) == 0) {
goto send;
}
if (tp->t_force) /* typ. timeout case */
goto send;
if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
goto send;
if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
goto send;
if (sack_rxmit)
goto send;
}
/*
* Compare available window to amount of window
* known to peer (as advertised window less
* next expected input). If the difference is at least two
* max size segments, or at least 50% of the maximum possible
* window, then want to send a window update to peer.
* Skip this if the connection is in T/TCP half-open state.
*/
if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN)) {
/*
* "adv" is the amount we can increase the window,
* taking into account that we are limited by
* TCP_MAXWIN << tp->rcv_scale.
*/
long adv = min(recwin, (long)TCP_MAXWIN << tp->rcv_scale) -
(tp->rcv_adv - tp->rcv_nxt);
if (adv >= (long) (2 * tp->t_maxseg))
goto send;
if (2 * adv >= (long) so->so_rcv.sb_hiwat)
goto send;
}
/*
* Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
* is also a catch-all for the retransmit timer timeout case.
*/
if (tp->t_flags & TF_ACKNOW)
goto send;
if ((flags & TH_RST) ||
((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0))
goto send;
if (SEQ_GT(tp->snd_up, tp->snd_una))
goto send;
/*
* If our state indicates that FIN should be sent
* and we have not yet done so, then we need to send.
*/
if (flags & TH_FIN &&
((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
goto send;
/*
* In SACK, it is possible for tcp_output to fail to send a segment
* after the retransmission timer has been turned off. Make sure
* that the retransmission timer is set.
*/
if (tp->sack_enable && SEQ_GT(tp->snd_max, tp->snd_una) &&
!callout_active(tp->tt_rexmt) &&
!callout_active(tp->tt_persist)) {
callout_reset(tp->tt_rexmt, tp->t_rxtcur,
tcp_timer_rexmt, tp);
goto just_return;
}
/*
* TCP window updates are not reliable, rather a polling protocol
* using ``persist'' packets is used to insure receipt of window
* updates. The three ``states'' for the output side are:
* idle not doing retransmits or persists
* persisting to move a small or zero window
* (re)transmitting and thereby not persisting
*
* callout_active(tp->tt_persist)
* is true when we are in persist state.
* tp->t_force
* is set when we are called to send a persist packet.
* callout_active(tp->tt_rexmt)
* is set when we are retransmitting
* The output side is idle when both timers are zero.
*
* If send window is too small, there is data to transmit, and no
* retransmit or persist is pending, then go to persist state.
* If nothing happens soon, send when timer expires:
* if window is nonzero, transmit what we can,
* otherwise force out a byte.
*/
if (so->so_snd.sb_cc && !callout_active(tp->tt_rexmt) &&
!callout_active(tp->tt_persist)) {
tp->t_rxtshift = 0;
tcp_setpersist(tp);
}
/*
* No reason to send a segment, just return.
*/
just_return:
SOCKBUF_UNLOCK(&so->so_snd);
return (0);
send:
SOCKBUF_LOCK_ASSERT(&so->so_snd);
/*
* Before ESTABLISHED, force sending of initial options
* unless TCP set not to do any options.
* NOTE: we assume that the IP/TCP header plus TCP options
* always fit in a single mbuf, leaving room for a maximum
* link header, i.e.
* max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES
*/
optlen = 0;
#ifdef INET6
if (isipv6)
hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
else
#endif
hdrlen = sizeof (struct tcpiphdr);
if (flags & TH_SYN) {
tp->snd_nxt = tp->iss;
if ((tp->t_flags & TF_NOOPT) == 0) {
u_short mss;
opt[0] = TCPOPT_MAXSEG;
opt[1] = TCPOLEN_MAXSEG;
mss = htons((u_short) tcp_mssopt(&tp->t_inpcb->inp_inc));
(void)memcpy(opt + 2, &mss, sizeof(mss));
optlen = TCPOLEN_MAXSEG;
/*
* If this is the first SYN of connection (not a SYN
* ACK), include SACK_PERMIT_HDR option. If this is a
* SYN ACK, include SACK_PERMIT_HDR option if peer has
* already done so. This is only for active connect,
* since the syncache takes care of the passive connect.
*/
if (tp->sack_enable && ((flags & TH_ACK) == 0 ||
(tp->t_flags & TF_SACK_PERMIT))) {
*((u_int32_t *) (opt + optlen)) =
htonl(TCPOPT_SACK_PERMIT_HDR);
optlen += 4;
}
if ((tp->t_flags & TF_REQ_SCALE) &&
((flags & TH_ACK) == 0 ||
(tp->t_flags & TF_RCVD_SCALE))) {
*((u_int32_t *)(opt + optlen)) = htonl(
TCPOPT_NOP << 24 |
TCPOPT_WINDOW << 16 |
TCPOLEN_WINDOW << 8 |
tp->request_r_scale);
optlen += 4;
}
}
}
/*
* Send a timestamp and echo-reply if this is a SYN and our side
* wants to use timestamps (TF_REQ_TSTMP is set) or both our side
* and our peer have sent timestamps in our SYN's.
*/
if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
(flags & TH_RST) == 0 &&
((flags & TH_ACK) == 0 ||
(tp->t_flags & TF_RCVD_TSTMP))) {
u_int32_t *lp = (u_int32_t *)(opt + optlen);
/* Form timestamp option as shown in appendix A of RFC 1323. */
*lp++ = htonl(TCPOPT_TSTAMP_HDR);
*lp++ = htonl(ticks);
*lp = htonl(tp->ts_recent);
optlen += TCPOLEN_TSTAMP_APPA;
}
/*
* Send SACKs if necessary. This should be the last option processed.
* Only as many SACKs are sent as are permitted by the maximum options
* size. No more than three SACKs are sent.
*/
if (tp->sack_enable && tp->t_state == TCPS_ESTABLISHED &&
(tp->t_flags & (TF_SACK_PERMIT|TF_NOOPT)) == TF_SACK_PERMIT &&
tp->rcv_numsacks) {
u_int32_t *lp = (u_int32_t *)(opt + optlen);
u_int32_t *olp = lp++;
int count = 0; /* actual number of SACKs inserted */
int maxsack = (MAX_TCPOPTLEN - (optlen + 4))/TCPOLEN_SACK;
tcpstat.tcps_sack_send_blocks++;
maxsack = min(maxsack, TCP_MAX_SACK);
for (i = 0; (i < tp->rcv_numsacks && count < maxsack); i++) {
struct sackblk sack = tp->sackblks[i];
if (sack.start == 0 && sack.end == 0)
continue;
*lp++ = htonl(sack.start);
*lp++ = htonl(sack.end);
count++;
}
*olp = htonl(TCPOPT_SACK_HDR|(TCPOLEN_SACK*count+2));
optlen += TCPOLEN_SACK*count + 4; /* including leading NOPs */
}
/*
* Send `CC-family' options if our side wants to use them (TF_REQ_CC),
* options are allowed (!TF_NOOPT) and it's not a RST.
*/
if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
(flags & TH_RST) == 0) {
switch (flags & (TH_SYN|TH_ACK)) {
/*
* This is a normal ACK, send CC if we received CC before
* from our peer.
*/
case TH_ACK:
if (!(tp->t_flags & TF_RCVD_CC))
break;
/*FALLTHROUGH*/
/*
* We can only get here in T/TCP's SYN_SENT* state, when
* we're a sending a non-SYN segment without waiting for
* the ACK of our SYN. A check above assures that we only
* do this if our peer understands T/TCP.
*/
case 0:
opt[optlen++] = TCPOPT_NOP;
opt[optlen++] = TCPOPT_NOP;
opt[optlen++] = TCPOPT_CC;
opt[optlen++] = TCPOLEN_CC;
*(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
optlen += 4;
break;
/*
* This is our initial SYN, check whether we have to use
* CC or CC.new.
*/
case TH_SYN:
opt[optlen++] = TCPOPT_NOP;
opt[optlen++] = TCPOPT_NOP;
opt[optlen++] = tp->t_flags & TF_SENDCCNEW ?
TCPOPT_CCNEW : TCPOPT_CC;
opt[optlen++] = TCPOLEN_CC;
*(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
optlen += 4;
break;
/*
* This is a SYN,ACK; send CC and CC.echo if we received
* CC from our peer.
*/
case (TH_SYN|TH_ACK):
if (tp->t_flags & TF_RCVD_CC) {
opt[optlen++] = TCPOPT_NOP;
opt[optlen++] = TCPOPT_NOP;
opt[optlen++] = TCPOPT_CC;
opt[optlen++] = TCPOLEN_CC;
*(u_int32_t *)&opt[optlen] =
htonl(tp->cc_send);
optlen += 4;
opt[optlen++] = TCPOPT_NOP;
opt[optlen++] = TCPOPT_NOP;
opt[optlen++] = TCPOPT_CCECHO;
opt[optlen++] = TCPOLEN_CC;
*(u_int32_t *)&opt[optlen] =
htonl(tp->cc_recv);
optlen += 4;
}
break;
}
}
#ifdef TCP_SIGNATURE
#ifdef INET6
if (!isipv6)
#endif
if (tp->t_flags & TF_SIGNATURE) {
int i;
u_char *bp;
/* Initialize TCP-MD5 option (RFC2385) */
bp = (u_char *)opt + optlen;
*bp++ = TCPOPT_SIGNATURE;
*bp++ = TCPOLEN_SIGNATURE;
sigoff = optlen + 2;
for (i = 0; i < TCP_SIGLEN; i++)
*bp++ = 0;
optlen += TCPOLEN_SIGNATURE;
/* Terminate options list and maintain 32-bit alignment. */
*bp++ = TCPOPT_NOP;
*bp++ = TCPOPT_EOL;
optlen += 2;
}
#endif /* TCP_SIGNATURE */
hdrlen += optlen;
#ifdef INET6
if (isipv6)
ipoptlen = ip6_optlen(tp->t_inpcb);
else
#endif
if (tp->t_inpcb->inp_options)
ipoptlen = tp->t_inpcb->inp_options->m_len -
offsetof(struct ipoption, ipopt_list);
else
/*
* Tcp output routine: figure out what should be sent and send it.
*/
int
tcp_output(PNATState pData, register struct tcpcb *tp)
{
register struct socket *so = tp->t_socket;
register long len, win;
int off, flags, error;
register struct mbuf *m = NULL;
register struct tcpiphdr *ti;
u_char opt[MAX_TCPOPTLEN];
unsigned optlen, hdrlen;
int idle, sendalot;
int size = 0;
LogFlowFunc(("ENTER: tcp_output: tp = %R[tcpcb793]\n", tp));
/*
* Determine length of data that should be transmitted,
* and flags that will be used.
* If there is some data or critical controls (SYN, RST)
* to send, then transmit; otherwise, investigate further.
*/
idle = (tp->snd_max == tp->snd_una);
if (idle && tp->t_idle >= tp->t_rxtcur)
/*
* We have been idle for "a while" and no acks are
* expected to clock out any data we send --
* slow start to get ack "clock" running again.
*/
tp->snd_cwnd = tp->t_maxseg;
again:
sendalot = 0;
off = tp->snd_nxt - tp->snd_una;
win = min(tp->snd_wnd, tp->snd_cwnd);
flags = tcp_outflags[tp->t_state];
Log2((" --- tcp_output flags = 0x%x\n", flags));
/*
* If in persist timeout with window of 0, send 1 byte.
* Otherwise, if window is small but nonzero
* and timer expired, we will send what we can
* and go to transmit state.
*/
if (tp->t_force)
{
if (win == 0)
{
/*
* If we still have some data to send, then
* clear the FIN bit. Usually this would
* happen below when it realizes that we
* aren't sending all the data. However,
* if we have exactly 1 byte of unset data,
* then it won't clear the FIN bit below,
* and if we are in persist state, we wind
* up sending the packet without recording
* that we sent the FIN bit.
*
* We can't just blindly clear the FIN bit,
* because if we don't have any more data
* to send then the probe will be the FIN
* itself.
*/
if (off < SBUF_LEN(&so->so_snd))
flags &= ~TH_FIN;
win = 1;
}
else
{
tp->t_timer[TCPT_PERSIST] = 0;
tp->t_rxtshift = 0;
}
}
len = min(SBUF_LEN(&so->so_snd), win) - off;
if (len < 0)
{
/*
* If FIN has been sent but not acked,
* but we haven't been called to retransmit,
* len will be -1. Otherwise, window shrank
* after we sent into it. If window shrank to 0,
* cancel pending retransmit and pull snd_nxt
* back to (closed) window. We will enter persist
* state below. If the window didn't close completely,
* just wait for an ACK.
*/
len = 0;
if (win == 0)
{
tp->t_timer[TCPT_REXMT] = 0;
tp->snd_nxt = tp->snd_una;
}
}
if (len > tp->t_maxseg)
{
len = tp->t_maxseg;
sendalot = 1;
}
if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + SBUF_LEN(&so->so_snd)))
flags &= ~TH_FIN;
win = sbspace(&so->so_rcv);
/*
* Sender silly window avoidance. If connection is idle
* and can send all data, a maximum segment,
* at least a maximum default-size segment do it,
* or are forced, do it; otherwise don't bother.
* If peer's buffer is tiny, then send
* when window is at least half open.
* If retransmitting (possibly after persist timer forced us
* to send into a small window), then must resend.
*/
if (len)
{
if (len == tp->t_maxseg)
goto send;
if ((1 || idle || tp->t_flags & TF_NODELAY) &&
len + off >= SBUF_LEN(&so->so_snd))
goto send;
if (tp->t_force)
goto send;
if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
goto send;
if (SEQ_LT(tp->snd_nxt, tp->snd_max))
goto send;
}
/*
* Compare available window to amount of window
* known to peer (as advertised window less
* next expected input). If the difference is at least two
* max size segments, or at least 50% of the maximum possible
* window, then want to send a window update to peer.
*/
if (win > 0)
{
/*
* "adv" is the amount we can increase the window,
* taking into account that we are limited by
* TCP_MAXWIN << tp->rcv_scale.
*/
long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale);
if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
adv -= tp->rcv_adv - tp->rcv_nxt;
if (adv >= (long) (2 * tp->t_maxseg))
goto send;
if (2 * adv >= (long) SBUF_SIZE(&so->so_rcv))
goto send;
}
/*
* Send if we owe peer an ACK.
*/
if (tp->t_flags & TF_ACKNOW)
goto send;
if (flags & (TH_SYN|TH_RST))
goto send;
if (SEQ_GT(tp->snd_up, tp->snd_una))
goto send;
/*
* If our state indicates that FIN should be sent
* and we have not yet done so, or we're retransmitting the FIN,
* then we need to send.
*/
if ( flags & TH_FIN
&& ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
goto send;
/*
* TCP window updates are not reliable, rather a polling protocol
* using ``persist'' packets is used to insure receipt of window
* updates. The three ``states'' for the output side are:
* idle not doing retransmits or persists
* persisting to move a small or zero window
* (re)transmitting and thereby not persisting
*
* tp->t_timer[TCPT_PERSIST]
* is set when we are in persist state.
* tp->t_force
* is set when we are called to send a persist packet.
* tp->t_timer[TCPT_REXMT]
* is set when we are retransmitting
* The output side is idle when both timers are zero.
*
* If send window is too small, there is data to transmit, and no
* retransmit or persist is pending, then go to persist state.
* If nothing happens soon, send when timer expires:
* if window is nonzero, transmit what we can,
* otherwise force out a byte.
*/
if ( SBUF_LEN(&so->so_snd)
&& tp->t_timer[TCPT_REXMT] == 0
&& tp->t_timer[TCPT_PERSIST] == 0)
{
tp->t_rxtshift = 0;
tcp_setpersist(tp);
}
/*
* No reason to send a segment, just return.
*/
tcpstat.tcps_didnuttin++;
LogFlowFuncLeave();
return (0);
send:
LogFlowFunc(("send\n"));
/*
* Before ESTABLISHED, force sending of initial options
* unless TCP set not to do any options.
* NOTE: we assume that the IP/TCP header plus TCP options
* always fit in a single mbuf, leaving room for a maximum
* link header, i.e.
* max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
*/
optlen = 0;
hdrlen = sizeof (struct tcpiphdr);
if (flags & TH_SYN)
{
tp->snd_nxt = tp->iss;
if ((tp->t_flags & TF_NOOPT) == 0)
{
u_int16_t mss;
opt[0] = TCPOPT_MAXSEG;
opt[1] = 4;
mss = RT_H2N_U16((u_int16_t) tcp_mss(pData, tp, 0));
memcpy((caddr_t)(opt + 2), (caddr_t)&mss, sizeof(mss));
optlen = 4;
#if 0
if ( (tp->t_flags & TF_REQ_SCALE)
&& ( (flags & TH_ACK) == 0
|| (tp->t_flags & TF_RCVD_SCALE)))
{
*((u_int32_t *) (opt + optlen)) = RT_H2N_U32( TCPOPT_NOP << 24
| TCPOPT_WINDOW << 16
| TCPOLEN_WINDOW << 8
| tp->request_r_scale);
optlen += 4;
}
#endif
}
}
/*
* Send a timestamp and echo-reply if this is a SYN and our side
* wants to use timestamps (TF_REQ_TSTMP is set) or both our side
* and our peer have sent timestamps in our SYN's.
*/
#if 0
if ( (tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP
&& (flags & TH_RST) == 0
&& ( (flags & (TH_SYN|TH_ACK)) == TH_SYN
|| (tp->t_flags & TF_RCVD_TSTMP)))
{
u_int32_t *lp = (u_int32_t *)(opt + optlen);
/* Form timestamp option as shown in appendix A of RFC 1323. */
*lp++ = RT_H2N_U32_C(TCPOPT_TSTAMP_HDR);
*lp++ = RT_H2N_U32(tcp_now);
*lp = RT_H2N_U32(tp->ts_recent);
optlen += TCPOLEN_TSTAMP_APPA;
}
#endif
hdrlen += optlen;
/*
* Adjust data length if insertion of options will
* bump the packet length beyond the t_maxseg length.
*/
if (len > tp->t_maxseg - optlen)
{
len = tp->t_maxseg - optlen;
sendalot = 1;
}
/*
* Grab a header mbuf, attaching a copy of data to
* be transmitted, and initialize the header from
* the template for sends on this connection.
*/
if (len)
{
if (tp->t_force && len == 1)
tcpstat.tcps_sndprobe++;
else if (SEQ_LT(tp->snd_nxt, tp->snd_max))
{
tcpstat.tcps_sndrexmitpack++;
tcpstat.tcps_sndrexmitbyte += len;
}
else
{
tcpstat.tcps_sndpack++;
tcpstat.tcps_sndbyte += len;
}
size = MCLBYTES;
if ((len + hdrlen + ETH_HLEN) < MSIZE)
size = MCLBYTES;
else if ((len + hdrlen + ETH_HLEN) < MCLBYTES)
size = MCLBYTES;
else if((len + hdrlen + ETH_HLEN) < MJUM9BYTES)
size = MJUM9BYTES;
else if ((len + hdrlen + ETH_HLEN) < MJUM16BYTES)
size = MJUM16BYTES;
else
AssertMsgFailed(("Unsupported size"));
m = m_getjcl(pData, M_NOWAIT, MT_HEADER, M_PKTHDR, size);
if (m == NULL)
{
/* error = ENOBUFS; */
error = 1;
goto out;
}
m->m_data += if_maxlinkhdr;
m->m_pkthdr.header = mtod(m, void *);
m->m_len = hdrlen;
/*
* This will always succeed, since we make sure our mbufs
* are big enough to hold one MSS packet + header + ... etc.
*/
#if 0
if (len <= MHLEN - hdrlen - max_linkhdr)
{
#endif
sbcopy(&so->so_snd, off, (int) len, mtod(m, caddr_t) + hdrlen);
m->m_len += len;
#if 0
}
else
{
m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
if (m->m_next == 0)
len = 0;
}
#endif
/*
* If we're sending everything we've got, set PUSH.
* (This will keep happy those implementations which only
* give data to the user when a buffer fills or
* a PUSH comes in.)
*/
if (off + len == SBUF_LEN(&so->so_snd))
flags |= TH_PUSH;
}
else
{
/*
* TCP timer processing.
*/
struct tcpcb*
tcp_timers(struct tcpcb *tp, int timer)
{
int rexmt;
switch (timer) {
/*
* 2 MSL timeout in shutdown went off. If we're closed but
* still waiting for peer to close and connection has been idle
* too long, or if 2MSL time is up from TIME_WAIT, delete connection
* control block. Otherwise, check again in a bit.
* If TIME_WAIT is not set, this is FIN_WAIT_2 timer.
*/
case TCPT_2MSL:
if (tp->t_state != TCPS_TIME_WAIT &&
tp->t_idle <= g_tcp_maxidle)
tp->t_timer[TCPT_2MSL] = g_tcp_keepintvl;
else
tp = tcp_close(tp);
break;
/*
* Retransmission timer went off. Message has not
* been acked within retransmit interval. Back off
* to a longer retransmit interval and retransmit one segment.
*/
case TCPT_REXMT:
if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
tp->t_rxtshift = TCP_MAXRXTSHIFT;
g_tcpstat.tcps_timeoutdrop++;
tp = tcp_drop(tp, tp->t_softerror ?
tp->t_softerror : ETIMEDOUT);
break;
}
g_tcpstat.tcps_rexmttimeo++;
rexmt = TCP_REXMTVAL(tp) * g_tcp_backoff[tp->t_rxtshift];
TCPT_RANGESET(tp->t_rxtcur, rexmt,
tp->t_rttmin, TCPTV_REXMTMAX);
tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
/*
* If losing, let the lower level know and try for
* a better route. Also, if we backed off this far,
* our srtt estimate is probably bogus. Clobber it
* so we'll take the next rtt measurement as our srtt;
* move the current srtt into rttvar to keep the current
* retransmit times until then.
*/
if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
in_losing(tp->t_inpcb);
tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
tp->t_srtt = 0;
}
tp->snd_nxt = tp->snd_una;
/*
* If timing a segment in this window, stop the timer.
*/
tp->t_rtt = 0;
/*
* Close the congestion window down to one segment
* (we'll open it by one segment for each ack we get).
* Since we probably have a window's worth of unacked
* data accumulated, this "slow start" keeps us from
* dumping all that data as back-to-back packets (which
* might overwhelm an intermediate gateway).
*
* There are two phases to the opening: Initially we
* open by one mss on each ack. This makes the window
* size increase exponentially with time. If the
* window is larger than the path can handle, this
* exponential growth results in dropped packet(s)
* almost immediately. To get more time between
* drops but still "push" the network to take advantage
* of improving conditions, we switch from exponential
* to linear window opening at some threshhold size.
* For a threshhold, we use half the current window
* size, truncated to a multiple of the mss.
*
* (the minimum cwnd that will give us exponential
* growth is 2 mss. We don't allow the threshhold
* to go below this.)
*/
{
u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
if (win < 2)
win = 2;
tp->snd_cwnd = tp->t_maxseg;
tp->snd_ssthresh = win * tp->t_maxseg;
tp->t_dupacks = 0;
}
(void) tcp_output(tp);
break;
/*
* Persistance timer into zero window.
* Force a byte to be output, if possible.
*/
case TCPT_PERSIST:
g_tcpstat.tcps_persisttimeo++;
/*
* Hack: if the peer is dead/unreachable, we do not
* time out if the window is closed. After a full
* backoff, drop the connection if the idle time
* (no responses to probes) reaches the maximum
* backoff that we would use if retransmitting.
*/
if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
(tp->t_idle >= g_tcp_maxpersistidle ||
tp->t_idle >= TCP_REXMTVAL(tp) * g_tcp_totbackoff)) {
g_tcpstat.tcps_persistdrop++;
tp = tcp_drop(tp, ETIMEDOUT);
break;
}
tcp_setpersist(tp);
tp->t_force = 1;
(void) tcp_output(tp);
tp->t_force = 0;
break;
/*
* Keep-alive timer went off; send something
* or drop connection if idle for too long.
*/
case TCPT_KEEP:
g_tcpstat.tcps_keeptimeo++;
if (tp->t_state < TCPS_ESTABLISHED) // connection-establishment timer.
goto dropit;
if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
tp->t_state <= TCPS_CLOSE_WAIT) { // keepalive timer.
if (tp->t_idle >= g_tcp_keepidle + g_tcp_maxidle)
goto dropit;
/*
* Send a packet designed to force a response
* if the peer is up and reachable:
* either an ACK if the connection is still alive,
* or an RST if the peer has closed the connection
* due to timeout or reboot.
* Using sequence number tp->snd_una-1
* causes the transmitted zero-length segment
* to lie outside the receive window;
* by the protocol spec, this requires the
* correspondent TCP to respond.
*/
g_tcpstat.tcps_keepprobe++;
tcp_respond(tp, tp->t_template, (usn_mbuf_t *)NULL,
tp->rcv_nxt, tp->snd_una - 1, 0);
tp->t_timer[TCPT_KEEP] = g_tcp_keepintvl;
} else
tp->t_timer[TCPT_KEEP] = g_tcp_keepidle;
break;
dropit:
g_tcpstat.tcps_keepdrops++;
tp = tcp_drop(tp, ETIMEDOUT);
break;
}
