SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW,
&ss_fltsz, 1, "Slow start flight size");
-int ss_fltsz_local = TCP_MAXWIN; /* something large */
+int ss_fltsz_local = 4; /* starts with four segments max */
SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW,
&ss_fltsz_local, 1, "Slow start flight size for local networks");
SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno, CTLFLAG_RW, &tcp_do_newreno,
0, "Enable NewReno Algorithms");
-struct mbuf *m_copym_with_hdrs __P((struct mbuf*, int, int, int, struct mbuf**, int*));
+int tcp_packet_chaining = 50;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, packetchain, CTLFLAG_RW, &tcp_packet_chaining,
+ 0, "Enable TCP output packet chaining");
+
+struct mbuf *m_copym_with_hdrs(struct mbuf*, int, int, int, struct mbuf**, int*);
+static long packchain_newlist = 0;
+static long packchain_looped = 0;
+static long packchain_sent = 0;
/* temporary: for testing */
extern int ipsec_bypass;
#endif
+extern int slowlink_wsize; /* window correction for slow links */
+extern u_long route_generation;
+extern int fw_enable; /* firewall is on: disable packet chaining */
+extern int ipsec_bypass;
+
+extern vm_size_t so_cache_zone_element_size;
+
+static __inline__ u_int16_t
+get_socket_id(struct socket * s)
+{
+ u_int16_t val;
+
+ if (so_cache_zone_element_size == 0) {
+ return (0);
+ }
+ val = (u_int16_t)(((u_int32_t)s) / so_cache_zone_element_size);
+ if (val == 0) {
+ val = 0xffff;
+ }
+ return (val);
+}
+
/*
* Tcp output routine: figure out what should be sent and send it.
*/
register struct tcphdr *th;
u_char opt[TCP_MAXOLEN];
unsigned ipoptlen, optlen, hdrlen;
- int idle, sendalot;
+ int idle, sendalot, howmuchsent = 0;
int maxburst = TCP_MAXBURST;
struct rmxp_tao *taop;
struct rmxp_tao tao_noncached;
-#if INET6
- int isipv6;
-#endif
- int last_off;
+ int last_off = 0;
int m_off;
struct mbuf *m_last = 0;
struct mbuf *m_head = 0;
-
-
- KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_START, 0,0,0,0,0);
+ struct mbuf *packetlist = 0;
+ struct mbuf *lastpacket = 0;
#if INET6
- if (isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)) {
-
- KERNEL_DEBUG(DBG_LAYER_BEG,
- ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport),
- (((tp->t_inpcb->in6p_laddr.s6_addr16[0] & 0xffff) << 16) |
- (tp->t_inpcb->in6p_faddr.s6_addr16[0] & 0xffff)),
- 0,0,0);
- }
- else
+ int isipv6 = tp->t_inpcb->inp_vflag & INP_IPV6 ;
#endif
+ short packchain_listadd = 0;
+ u_int16_t socket_id = get_socket_id(so);
+
- {
- KERNEL_DEBUG(DBG_LAYER_BEG,
- ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport),
- (((tp->t_inpcb->inp_laddr.s_addr & 0xffff) << 16) |
- (tp->t_inpcb->inp_faddr.s_addr & 0xffff)),
- 0,0,0);
- }
/*
* Determine length of data that should be transmitted,
* and flags that will be used.
* to send, then transmit; otherwise, investigate further.
*/
idle = (tp->snd_max == tp->snd_una);
-#ifdef __APPLE__
if (idle && tp->t_rcvtime >= tp->t_rxtcur) {
-#else
- if (idle && (ticks - tp->t_rcvtime) >= tp->t_rxtcur) {
-#endif
/*
* We have been idle for "a while" and no acks are
* expected to clock out any data we send --
else
tp->snd_cwnd = tp->t_maxseg * ss_fltsz;
}
+
again:
+ KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_START, 0,0,0,0,0);
+
+#if INET6
+ if (isipv6) {
+
+ KERNEL_DEBUG(DBG_LAYER_BEG,
+ ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport),
+ (((tp->t_inpcb->in6p_laddr.s6_addr16[0] & 0xffff) << 16) |
+ (tp->t_inpcb->in6p_faddr.s6_addr16[0] & 0xffff)),
+ sendalot,0,0);
+ }
+ else
+#endif
+
+ {
+ KERNEL_DEBUG(DBG_LAYER_BEG,
+ ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport),
+ (((tp->t_inpcb->inp_laddr.s_addr & 0xffff) << 16) |
+ (tp->t_inpcb->inp_faddr.s_addr & 0xffff)),
+ sendalot,0,0);
+ /*
+ * If the route generation id changed, we need to check that our
+ * local (source) IP address is still valid. If it isn't either
+ * return error or silently do nothing (assuming the address will
+ * come back before the TCP connection times out).
+ */
+
+ if ((tp->t_inpcb->inp_route.ro_rt != NULL &&
+ (tp->t_inpcb->inp_route.ro_rt->generation_id != route_generation)) || (tp->t_inpcb->inp_route.ro_rt == NULL)) {
+ /* check that the source address is still valid */
+ if (ifa_foraddr(tp->t_inpcb->inp_laddr.s_addr) == 0) {
+ if (tp->t_state >= TCPS_CLOSE_WAIT) {
+ tcp_close(tp);
+ return(EADDRNOTAVAIL);
+ }
+
+ /* set Retransmit timer if it wasn't set
+ * reset Persist timer and shift register as the
+ * adversed peer window may not be valid anymore
+ */
+
+ if (!tp->t_timer[TCPT_REXMT]) {
+ tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
+ if (tp->t_timer[TCPT_PERSIST]) {
+ tp->t_timer[TCPT_PERSIST] = 0;
+ tp->t_rxtshift = 0;
+ }
+ }
+
+ if (packetlist) {
+ error = ip_output_list(packetlist, packchain_listadd, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
+ (so->so_options & SO_DONTROUTE), 0);
+ tp->t_lastchain = 0;
+ }
+ if (so->so_flags & SOF_NOADDRAVAIL)
+ return(EADDRNOTAVAIL);
+ else
+ return(0); /* silently ignore and keep data in socket */
+ }
+ }
+ }
sendalot = 0;
off = tp->snd_nxt - tp->snd_una;
win = min(tp->snd_wnd, tp->snd_cwnd);
+ if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0)
+ win = min(win, slowlink_wsize);
flags = tcp_outflags[tp->t_state];
/*
off--, len++;
if (len > 0 && tp->t_state == TCPS_SYN_SENT &&
taop->tao_ccsent == 0) {
+ if (packetlist) {
+ error = ip_output_list(packetlist, packchain_listadd, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
+ (so->so_options & SO_DONTROUTE), 0);
+ tp->t_lastchain = 0;
+ }
KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0);
return 0;
}
}
if (len > tp->t_maxseg) {
len = tp->t_maxseg;
+ howmuchsent += len;
sendalot = 1;
}
if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
flags &= ~TH_FIN;
- win = sbspace(&so->so_rcv);
+ if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0 ) /* Clips window size for slow links */
+ win = min(sbspace(&so->so_rcv), slowlink_wsize);
+ else
+ win = sbspace(&so->so_rcv);
/*
* Sender silly window avoidance. If connection is idle
}
/*
- * No reason to send a segment, just return.
+ * If there is no reason to send a segment, just return.
+ * but if there is some packets left in the packet list, send them now.
*/
+ if (packetlist) {
+ error = ip_output_list(packetlist, packchain_listadd, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
+ (so->so_options & SO_DONTROUTE), 0);
+ }
KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0);
return (0);
*/
flags &= ~TH_FIN;
len = tp->t_maxopd - optlen - ipoptlen;
+ howmuchsent += len;
sendalot = 1;
}
m->m_data += max_linkhdr;
m->m_len = hdrlen;
}
+ /* makes sure we still have data left to be sent at this point */
+ if (so->so_snd.sb_mb == NULL || off == -1) {
+ if (m != NULL) m_freem(m);
+ error = 0; /* should we return an error? */
+ goto out;
+ }
m_copydata(so->so_snd.sb_mb, off, (int) len,
mtod(m, caddr_t) + hdrlen);
m->m_len += len;
m_last = NULL;
last_off = off + len;
m_head = so->so_snd.sb_mb;
-
+
+ /* makes sure we still have data left to be sent at this point */
+ if (m_head == NULL) {
+ error = 0; /* should we return an error? */
+ goto out;
+ }
+
/*
* m_copym_with_hdrs will always return the last mbuf pointer and the offset into it that
* it acted on to fullfill the current request, whether a valid 'hint' was passed in or not
m->m_data += max_linkhdr;
m->m_len = hdrlen;
}
- m->m_pkthdr.rcvif = (struct ifnet *)0;
+ m->m_pkthdr.rcvif = 0;
#if INET6
if (isipv6) {
ip6 = mtod(m, struct ip6_hdr *);
win = 0;
if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
win = (long)(tp->rcv_adv - tp->rcv_nxt);
- if (win > (long)TCP_MAXWIN << tp->rcv_scale)
+ if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0) {
+ if (win > (long)slowlink_wsize)
+ win = slowlink_wsize;
+ th->th_win = htons((u_short) (win>>tp->rcv_scale));
+ }
+ else {
+
+ if (win > (long)TCP_MAXWIN << tp->rcv_scale)
win = (long)TCP_MAXWIN << tp->rcv_scale;
- th->th_win = htons((u_short) (win>>tp->rcv_scale));
+ th->th_win = htons((u_short) (win>>tp->rcv_scale));
+ }
+
+ /*
+ * Adjust the RXWIN0SENT flag - indicate that we have advertised
+ * a 0 window. This may cause the remote transmitter to stall. This
+ * flag tells soreceive() to disable delayed acknowledgements when
+ * draining the buffer. This can occur if the receiver is attempting
+ * to read more data then can be buffered prior to transmitting on
+ * the connection.
+ */
+ if (win == 0)
+ tp->t_flags |= TF_RXWIN0SENT;
+ else
+ tp->t_flags &= ~TF_RXWIN0SENT;
+
if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
th->th_flags |= TH_URG;
goto out;
}
#endif /*IPSEC*/
+ m->m_pkthdr.socket_id = socket_id;
error = ip6_output(m,
tp->t_inpcb->in6p_outputopts,
&tp->t_inpcb->in6p_route,
- (so->so_options & SO_DONTROUTE), NULL, NULL);
+ (so->so_options & SO_DONTROUTE), NULL, NULL, 0);
} else
#endif /* INET6 */
{
struct rtentry *rt;
ip->ip_len = m->m_pkthdr.len;
#if INET6
- if (INP_CHECK_SOCKAF(so, AF_INET6))
+ if (isipv6)
ip->ip_ttl = in6_selecthlim(tp->t_inpcb,
tp->t_inpcb->in6p_route.ro_rt ?
tp->t_inpcb->in6p_route.ro_rt->rt_ifp
if (ipsec_bypass == 0)
ipsec_setsocket(m, so);
#endif /*IPSEC*/
- error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
- (so->so_options & SO_DONTROUTE), 0);
- }
+
+ /*
+ * The socket is kept locked while sending out packets in ip_output, even if packet chaining is not active.
+ */
+
+ m->m_pkthdr.socket_id = socket_id;
+ if (packetlist) {
+ m->m_nextpkt = NULL;
+ lastpacket->m_nextpkt = m;
+ lastpacket = m;
+ packchain_listadd++;
+ }
+ else {
+ m->m_nextpkt = NULL;
+ packchain_newlist++;
+ packetlist = lastpacket = m;
+ packchain_listadd=0;
+ }
+
+ if ((ipsec_bypass == 0) || fw_enable || sendalot == 0 || (tp->t_state != TCPS_ESTABLISHED) ||
+ (tp->snd_cwnd <= (tp->snd_wnd / 4)) ||
+ (tp->t_flags & (TH_PUSH | TF_ACKNOW)) || tp->t_force != 0 ||
+ packchain_listadd >= tcp_packet_chaining) {
+ lastpacket->m_nextpkt = 0;
+ error = ip_output_list(packetlist, packchain_listadd, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
+ (so->so_options & SO_DONTROUTE), 0);
+ tp->t_lastchain = packchain_listadd;
+ packchain_sent++;
+ packetlist = NULL;
+ if (error == 0)
+ howmuchsent = 0;
+ }
+ else {
+ error = 0;
+ packchain_looped++;
+ tcpstat.tcps_sndtotal++;
+ if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
+ tp->rcv_adv = tp->rcv_nxt + win;
+ tp->last_ack_sent = tp->rcv_nxt;
+ tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
+ goto again;
+ }
+ }
if (error) {
/*
* No need to check for TH_FIN here because
* the TF_SENTFIN flag handles that case.
*/
- if ((flags & TH_SYN) == 0)
- tp->snd_nxt -= len;
+ if ((flags & TH_SYN) == 0)
+ tp->snd_nxt -= howmuchsent;
}
+ howmuchsent = 0;
out:
if (error == ENOBUFS) {
if (!tp->t_timer[TCPT_REXMT] &&
!tp->t_timer[TCPT_PERSIST])
tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
tcp_quench(tp->t_inpcb, 0);
+ if (packetlist)
+ m_freem_list(packetlist);
+ tp->t_lastchain = 0;
KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0);
return (0);
}
* not do so here.
*/
tcp_mtudisc(tp->t_inpcb, 0);
+ if (packetlist)
+ m_freem_list(packetlist);
+ tp->t_lastchain = 0;
KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0);
return 0;
}
if ((error == EHOSTUNREACH || error == ENETDOWN)
&& TCPS_HAVERCVDSYN(tp->t_state)) {
tp->t_softerror = error;
+ if (packetlist)
+ m_freem_list(packetlist);
+ tp->t_lastchain = 0;
KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0);
return (0);
}
+ if (packetlist)
+ m_freem_list(packetlist);
+ tp->t_lastchain = 0;
KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0);
return (error);
}
+sentit:
tcpstat.tcps_sndtotal++;
/*
tp->rcv_adv = tp->rcv_nxt + win;
tp->last_ack_sent = tp->rcv_nxt;
tp->t_flags &= ~(TF_ACKNOW|TF_DELACK);
- if (sendalot)
+
+ KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END,0,0,0,0,0);
+ if (sendalot && (!tcp_do_newreno || --maxburst))
goto again;
- KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0);
return (0);
}
register struct tcpcb *tp;
{
int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
- int tt;
if (tp->t_timer[TCPT_REXMT])
panic("tcp_setpersist: retransmit pending");