/*
- * Copyright (c) 2000-2015 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2016 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
- *
+ *
* This file contains Original Code and/or Modifications of Original Code
* as defined in and that are subject to the Apple Public Source License
* Version 2.0 (the 'License'). You may not use this file except in
* unlawful or unlicensed copies of an Apple operating system, or to
* circumvent, violate, or enable the circumvention or violation of, any
* terms of an Apple operating system software license agreement.
- *
+ *
* Please obtain a copy of the License at
* http://www.opensource.apple.com/apsl/ and read it before using this file.
- *
+ *
* The Original Code and all software distributed under the License are
* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
* Please see the License for the specific language governing rights and
* limitations under the License.
- *
+ *
* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/in_var.h>
+#include <netinet/in_tclass.h>
#include <netinet/ip.h>
#include <netinet/in_pcb.h>
#include <netinet/ip_var.h>
SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_RW | CTLFLAG_LOCKED,
&tcp_do_tso, 0, "Enable TCP Segmentation Offload");
+int tcp_ecn_setup_percentage = 50;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, ecn_setup_percentage,
+ CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_ecn_setup_percentage, 0,
+ "Max ECN setup percentage");
+
static int
sysctl_change_ecn_setting SYSCTL_HANDLER_ARGS
{
return (err);
}
-int tcp_ecn_outbound = 0;
+int tcp_ecn_outbound = 2;
SYSCTL_PROC(_net_inet_tcp, OID_AUTO, ecn_initiate_out,
CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_ecn_outbound, 0,
sysctl_change_ecn_setting, "IU",
"Initiate ECN for outbound connections");
-int tcp_ecn_inbound = 0;
+int tcp_ecn_inbound = 2;
SYSCTL_PROC(_net_inet_tcp, OID_AUTO, ecn_negotiate_in,
CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_ecn_inbound, 0,
sysctl_change_ecn_setting, "IU",
"Use RTT for bg recv algorithm");
uint32_t tcp_recv_throttle_minwin = 16 * 1024;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, recv_throttle_minwin,
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, recv_throttle_minwin,
CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_recv_throttle_minwin, 1,
"Minimum recv win for throttling");
extern int slowlink_wsize; /* window correction for slow links */
#if IPFIREWALL
-extern int fw_enable; /* firewall check for packet chaining */
-extern int fw_bypass; /* firewall check: disable packet chaining if there is rules */
+extern int fw_enable; /* firewall check for packet chaining */
+extern int fw_bypass; /* firewall check: disable packet chaining if there is rules */
#endif /* IPFIREWALL */
extern u_int32_t dlil_filter_disable_tso_count;
extern u_int32_t kipf_count;
-extern int tcp_recv_bg;
-static int tcp_ip_output(struct socket *, struct tcpcb *, struct mbuf *, int,
- struct mbuf *, int, int, int32_t, boolean_t);
+static int tcp_ip_output(struct socket *, struct tcpcb *, struct mbuf *,
+ int, struct mbuf *, int, int, boolean_t);
static struct mbuf* tcp_send_lroacks(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th);
static int tcp_recv_throttle(struct tcpcb *tp);
if (tp->t_flags & TF_NOOPT)
goto fallback;
- if (!tcp_heuristic_do_tfo(tp))
+ if (so->so_flags & SOF1_DATA_AUTHENTICATED)
+ return (len);
+
+ if (!tcp_heuristic_do_tfo(tp)) {
+ tp->t_tfo_stats |= TFO_S_HEURISTICS_DISABLE;
+ tcpstat.tcps_tfo_heuristics_disable++;
goto fallback;
+ }
optlen += TCPOLEN_MAXSEG;
/* No cookie, so we request one */
return (0);
+ /* There is not enough space for the cookie, so we cannot do TFO */
+ if (MAX_TCPOPTLEN - optlen < cookie_len)
+ goto fallback;
+
/* Do not send SYN+data if there is more in the queue than MSS */
if (so->so_snd.sb_cc > (tp->t_maxopd - MAX_TCPOPTLEN))
goto fallback;
}
static unsigned
-tcp_tfo_write_cookie(struct tcpcb *tp, unsigned optlen, int32_t *len,
+tcp_tfo_write_cookie(struct tcpcb *tp, unsigned optlen, int32_t len,
u_char *opt)
{
u_int8_t tfo_len = MAX_TCPOPTLEN - optlen - TCPOLEN_FASTOPEN_REQ;
+ struct socket *so = tp->t_inpcb->inp_socket;
unsigned ret = 0;
int res;
u_char *bp;
+ if (so->so_flags & SOF1_DATA_AUTHENTICATED) {
+ /* If there is some data, let's track it */
+ if (len > 0) {
+ tp->t_tfo_stats |= TFO_S_SYN_DATA_SENT;
+ tcpstat.tcps_tfo_syn_data_sent++;
+ }
+
+ return (0);
+ }
+
bp = opt + optlen;
/*
tp->t_tfo_flags |= TFO_F_COOKIE_SENT;
/* If there is some data, let's track it */
- if (*len) {
+ if (len > 0) {
tp->t_tfo_stats |= TFO_S_SYN_DATA_SENT;
tcpstat.tcps_tfo_syn_data_sent++;
}
check_heuristic:
if (!tcp_heuristic_do_ecn(tp))
tp->ecn_flags &= ~TE_ENABLE_ECN;
+
+ /*
+ * If the interface setting, system-level setting and heuristics
+ * allow to enable ECN, randomly select 5% of connections to
+ * enable it
+ */
+ if ((tp->ecn_flags & (TE_ECN_MODE_ENABLE | TE_ECN_MODE_DISABLE
+ | TE_ENABLE_ECN)) == TE_ENABLE_ECN) {
+ /*
+ * Use the random value in iss for randomizing
+ * this selection
+ */
+ if ((tp->iss % 100) >= tcp_ecn_setup_percentage)
+ tp->ecn_flags &= ~TE_ENABLE_ECN;
+ }
}
/*
struct mbuf *tp_inp_options = inp->inp_depend4.inp4_options;
#if INET6
int isipv6 = inp->inp_vflag & INP_IPV6 ;
+#else
+ int isipv6 = 0;
#endif
short packchain_listadd = 0;
int so_options = so->so_options;
struct rtentry *rt;
- u_int32_t basertt, svc_flags = 0, allocated_len;
+ u_int32_t svc_flags = 0, allocated_len;
u_int32_t lro_ackmore = (tp->t_lropktlen != 0) ? 1 : 0;
struct mbuf *mnext = NULL;
int sackoptlen = 0;
boolean_t wifi = FALSE;
boolean_t wired = FALSE;
boolean_t sack_rescue_rxt = FALSE;
+ int sotc = so->so_traffic_class;
/*
* Determine length of data that should be transmitted,
idle = 0;
}
}
-#if MPTCP
+#if MPTCP
if (tp->t_mpflags & TMPF_RESET) {
tcp_check_timer_state(tp);
- /*
- * Once a RST has been sent for an MPTCP subflow,
+ /*
+ * Once a RST has been sent for an MPTCP subflow,
* the subflow socket stays around until deleted.
* No packets such as FINs must be sent after RST.
*/
OFFSET_FROM_START(tp, tp->t_rxtcur);
if (tp->t_timer[TCPT_PERSIST]) {
tp->t_timer[TCPT_PERSIST] = 0;
- tp->t_rxtshift = 0;
tp->t_persist_stop = 0;
- tp->t_rxtstart = 0;
+ TCP_RESET_REXMT_STATE(tp);
}
}
TCP_PKTLIST_CLEAR(tp);
/* drop connection if source address isn't available */
- if (so->so_flags & SOF_NOADDRAVAIL) {
+ if (so->so_flags & SOF_NOADDRAVAIL) {
tcp_drop(tp, EADDRNOTAVAIL);
return(EADDRNOTAVAIL);
} else {
* has been disabled)
*/
- if (!path_mtu_discovery || ((rt != NULL) &&
+ if (!path_mtu_discovery || ((rt != NULL) &&
(!(rt->rt_flags & RTF_UP) ||
- (rt->rt_rmx.rmx_locks & RTV_MTU))))
+ (rt->rt_rmx.rmx_locks & RTV_MTU))))
tp->t_flags &= ~TF_PMTUD;
else
tp->t_flags |= TF_PMTUD;
if (SACK_ENABLED(tp) && IN_FASTRECOVERY(tp) &&
(p = tcp_sack_output(tp, &sack_bytes_rxmt))) {
int32_t cwin;
-
+
cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt;
if (cwin < 0)
cwin = 0;
len = ((int32_t)min(cwin, p->end - p->rxmit));
}
if (len > 0) {
- off = p->rxmit - tp->snd_una;
+ off = p->rxmit - tp->snd_una;
sack_rxmit = 1;
sendalot = 1;
tcpstat.tcps_sack_rexmits++;
tcpstat.tcps_sack_rexmit_bytes +=
min(len, tp->t_maxseg);
- if (nstat_collect) {
- nstat_route_tx(inp->inp_route.ro_rt, 1,
- min(len, tp->t_maxseg),
- NSTAT_TX_FLAG_RETRANSMIT);
- INP_ADD_STAT(inp, cell, wifi, wired,
- txpackets, 1);
- INP_ADD_STAT(inp, cell, wifi, wired,
- txbytes, min(len, tp->t_maxseg));
- tp->t_stat.txretransmitbytes += min(len, tp->t_maxseg);
- }
} else {
len = 0;
}
sendwin = 1;
} else {
tp->t_timer[TCPT_PERSIST] = 0;
- tp->t_rxtshift = 0;
- tp->t_rxtstart = 0;
tp->t_persist_stop = 0;
+ TCP_RESET_REXMT_STATE(tp);
}
}
* sending new data, having retransmitted all the
* data possible in the scoreboard.
*/
- len = min(so->so_snd.sb_cc, tp->snd_wnd)
+ len = min(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
+ * 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 ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
if (tp->t_state != TCPS_SYN_RECEIVED || tfo_enabled(tp))
flags &= ~TH_SYN;
- off--, len++;
+ off--;
+ len++;
if (len > 0 && tp->t_state == TCPS_SYN_SENT) {
while (inp->inp_sndinprog_cnt == 0 &&
tp->t_pktlist_head != NULL) {
error = tcp_ip_output(so, tp, packetlist,
packchain_listadd, tp_inp_options,
(so_options & SO_DONTROUTE),
- (sack_rxmit | (sack_bytes_rxmt != 0)), 0,
-#if INET6
+ (sack_rxmit | (sack_bytes_rxmt != 0)),
isipv6);
-#else /* INET6 */
- 0);
-#endif /* !INET6 */
-
-
}
/*
* tcp was closed while we were in ip,
- * resume close
+ * resume close
*/
if (inp->inp_sndinprog_cnt == 0 &&
(tp->t_flags & TF_CLOSING)) {
/*
* The check here used to be (len < 0). Some times len is zero
* when the congestion window is closed and we need to check
- * if persist timer has to be set in that case. But don't set
+ * if persist timer has to be set in that case. But don't set
* persist until connection is established.
- */
+ */
if (len <= 0 && !(flags & TH_SYN)) {
/*
* If FIN has been sent but not acked,
if (sendwin == 0) {
tp->t_timer[TCPT_REXMT] = 0;
tp->t_timer[TCPT_PTO] = 0;
- tp->t_rxtshift = 0;
- tp->t_rxtstart = 0;
+ TCP_RESET_REXMT_STATE(tp);
tp->snd_nxt = tp->snd_una;
off = 0;
if (tp->t_timer[TCPT_PERSIST] == 0)
* 3. our send window (slow start and congestion controlled) is
* larger than sent but unacknowledged data in send buffer.
*/
- basertt = get_base_rtt(tp);
if (tcp_do_autosendbuf == 1 &&
!INP_WAIT_FOR_IF_FEEDBACK(inp) && !IN_FASTRECOVERY(tp) &&
(so->so_snd.sb_flags & (SB_AUTOSIZE | SB_TRIM)) == SB_AUTOSIZE &&
tcp_cansbgrow(&so->so_snd)) {
if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat &&
so->so_snd.sb_cc >= (so->so_snd.sb_hiwat / 8 * 7) &&
- sendwin >= (so->so_snd.sb_cc -
- (tp->snd_nxt - tp->snd_una))) {
- /* Also increase the send buffer only if the
- * round-trip time is not increasing because we do
- * not want to contribute to latency by filling
- * buffers.
- * We also do not want to hold onto application's
- * old data for too long. Interactive applications
- * would rather discard old data.
- */
- if (tp->t_rttcur <= (basertt + 25)) {
- if (sbreserve(&so->so_snd,
- min(so->so_snd.sb_hiwat + tcp_autosndbuf_inc,
- tcp_autosndbuf_max)) == 1) {
- so->so_snd.sb_idealsize = so->so_snd.sb_hiwat;
- }
- } else {
- so->so_snd.sb_idealsize =
- max(tcp_sendspace, so->so_snd.sb_hiwat -
- (2 * tcp_autosndbuf_inc));
- so->so_snd.sb_flags |= SB_TRIM;
+ sendwin >= (so->so_snd.sb_cc - (tp->snd_nxt - tp->snd_una))) {
+ if (sbreserve(&so->so_snd,
+ min(so->so_snd.sb_hiwat + tcp_autosndbuf_inc,
+ tcp_autosndbuf_max)) == 1) {
+ so->so_snd.sb_idealsize = so->so_snd.sb_hiwat;
}
}
}
}
#if MPTCP
- if ((so->so_flags & SOF_MP_SUBFLOW) &&
+ if ((so->so_flags & SOF_MP_SUBFLOW) &&
!(tp->t_mpflags & TMPF_TCP_FALLBACK)) {
int newlen = len;
- if (!(tp->t_mpflags & TMPF_PREESTABLISHED) &&
- (tp->t_state > TCPS_CLOSED) &&
+ if ((tp->t_state >= TCPS_ESTABLISHED) &&
((tp->t_mpflags & TMPF_SND_MPPRIO) ||
(tp->t_mpflags & TMPF_SND_REM_ADDR) ||
- (tp->t_mpflags & TMPF_SND_MPFAIL) ||
- (tp->t_mpflags & TMPF_MPCAP_RETRANSMIT))) {
+ (tp->t_mpflags & TMPF_SND_MPFAIL))) {
if (len > 0) {
len = 0;
}
}
/*
* The contiguous bytes in the subflow socket buffer can be
- * discontiguous at the MPTCP level. Since only one DSS
+ * discontiguous at the MPTCP level. Since only one DSS
* option can be sent in one packet, reduce length to match
* the contiguous MPTCP level. Set sendalot to send remainder.
*/
* pull the amount of data that can be sent from the
* unordered priority queues to the serial queue in
* the socket buffer. If bytes are not yet available
- * in the highest priority message, we may not be able
- * to send any new data.
+ * in the highest priority message, we may not be able
+ * to send any new data.
*/
if (so->so_flags & SOF_ENABLE_MSGS) {
if ((off + len) >
sbpull_unordered_data(so, off, len);
/* check if len needs to be modified */
- if ((off + len) >
+ if ((off + len) >
so->so_msg_state->msg_serial_bytes) {
len = so->so_msg_state->msg_serial_bytes - off;
if (len <= 0) {
if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
flags &= ~TH_FIN;
}
-
+ /*
+ * 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 25% 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.
+ */
recwin = tcp_sbspace(tp);
+#if MPTCP
+ if (so->so_flags & SOF_MP_SUBFLOW) {
+ struct mptcb *mp_tp = tptomptp(tp);
+
+ if (mp_tp != NULL) {
+ MPT_LOCK(mp_tp);
+ recwin = imin(recwin, (int)mp_tp->mpt_rcvwnd);
+ MPT_UNLOCK(mp_tp);
+ }
+ }
+#endif
+
+ if (recwin < (int32_t)(so->so_rcv.sb_hiwat / 4) &&
+ recwin < (int)tp->t_maxseg)
+ recwin = 0;
+
+#if TRAFFIC_MGT
+ if (tcp_recv_bg == 1 || IS_TCP_RECV_BG(so)) {
+ if (recwin > 0 && tcp_recv_throttle(tp)) {
+ uint32_t min_iaj_win = tcp_min_iaj_win * tp->t_maxseg;
+ uint32_t bg_rwintop = tp->rcv_adv;
+ if (SEQ_LT(bg_rwintop, tp->rcv_nxt + min_iaj_win))
+ bg_rwintop = tp->rcv_nxt + min_iaj_win;
+ recwin = imin((int32_t)(bg_rwintop - tp->rcv_nxt),
+ recwin);
+ if (recwin < 0)
+ recwin = 0;
+ }
+ }
+#endif /* TRAFFIC_MGT */
+
+ if (recwin > (int32_t)(TCP_MAXWIN << tp->rcv_scale))
+ recwin = (int32_t)(TCP_MAXWIN << tp->rcv_scale);
+ if (recwin < (int32_t)(tp->rcv_adv - tp->rcv_nxt))
+ recwin = (int32_t)(tp->rcv_adv - tp->rcv_nxt);
/*
* Sender silly window avoidance. We transmit under the following
if (len >= tp->t_maxseg)
goto send;
if (!(tp->t_flags & TF_MORETOCOME) &&
- (idle || tp->t_flags & TF_NODELAY ||
- tp->t_flags & TF_MAXSEGSNT ||
+ (idle || tp->t_flags & TF_NODELAY ||
+ (tp->t_flags & TF_MAXSEGSNT) ||
ALLOW_LIMITED_TRANSMIT(tp)) &&
(tp->t_flags & TF_NOPUSH) == 0 &&
len + off >= so->so_snd.sb_cc)
}
}
- /*
- * 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 25% 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,
/*
* Update only if the resulting scaled value of
* the window changed, or if there is a change in
- * the sequence since the last ack. This avoids
+ * the sequence since the last ack. This avoids
* what appears as dupe ACKS (see rdar://5640997)
*
* If streaming is detected avoid sending too many
- * window updates. We will depend on the delack
+ * window updates. We will depend on the delack
* timer to send a window update when needed.
*/
if (!(tp->t_flags & TF_STRETCHACK) &&
- (tp->last_ack_sent != tp->rcv_nxt ||
+ (tp->last_ack_sent != tp->rcv_nxt ||
((oldwin + adv) >> tp->rcv_scale) >
(oldwin >> tp->rcv_scale))) {
goto send;
}
- /*
- * Make sure that the delayed ack timer is set if
- * we delayed sending a window update because of
- * streaming detection.
- */
- if ((tp->t_flags & TF_STRETCHACK) &&
- !(tp->t_flags & TF_DELACK)) {
- tp->t_flags |= TF_DELACK;
- tp->t_timer[TCPT_DELACK] =
- OFFSET_FROM_START(tp, tcp_delack);
- }
}
- if (4 * adv >= (int32_t) so->so_rcv.sb_hiwat)
- goto send;
+ if (4 * adv >= (int32_t) so->so_rcv.sb_hiwat)
+ goto send;
+
+ /*
+ * Make sure that the delayed ack timer is set if
+ * we delayed sending a window update because of
+ * streaming detection.
+ */
+ if ((tp->t_flags & TF_STRETCHACK) &&
+ !(tp->t_flags & TF_DELACK)) {
+ tp->t_flags |= TF_DELACK;
+ tp->t_timer[TCPT_DELACK] =
+ OFFSET_FROM_START(tp, tcp_delack);
+ }
}
/*
* after the retransmission timer has been turned off. Make sure
* that the retransmission timer is set.
*/
- if (SACK_ENABLED(tp) && (tp->t_state >= TCPS_ESTABLISHED) &&
+ if (SACK_ENABLED(tp) && (tp->t_state >= TCPS_ESTABLISHED) &&
SEQ_GT(tp->snd_max, tp->snd_una) &&
tp->t_timer[TCPT_REXMT] == 0 &&
tp->t_timer[TCPT_PERSIST] == 0) {
tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp,
tp->t_rxtcur);
goto just_return;
- }
+ }
/*
* TCP window updates are not reliable, rather a polling protocol
* using ``persist'' packets is used to insure receipt of window
*/
if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 &&
tp->t_timer[TCPT_PERSIST] == 0) {
- tp->t_rxtshift = 0;
- tp->t_rxtstart = 0;
+ TCP_RESET_REXMT_STATE(tp);
tcp_setpersist(tp);
}
just_return:
error = tcp_ip_output(so, tp, packetlist,
packchain_listadd,
tp_inp_options, (so_options & SO_DONTROUTE),
- (sack_rxmit | (sack_bytes_rxmt != 0)), recwin,
-#if INET6
- isipv6);
-#else /* INET6 */
- 0);
-#endif /* !INET6 */
+ (sack_rxmit | (sack_bytes_rxmt != 0)), isipv6);
}
/* tcp was closed while we were in ip; resume close */
if (inp->inp_sndinprog_cnt == 0 &&
optlen += 4;
}
#if MPTCP
- if (mptcp_enable) {
+ if (mptcp_enable && (so->so_flags & SOF_MP_SUBFLOW)) {
optlen = mptcp_setup_syn_opts(so, flags, opt,
optlen);
}
tp->rfbuf_ts = tcp_now;
if (SACK_ENABLED(tp) && ((tp->t_flags & TF_NOOPT) == 0)) {
- /*
+ /*
* Tack on the SACK permitted option *last*.
* And do padding of options after tacking this on.
* This is because of MSS, TS, WinScale and Signatures are
if (tfo_enabled(tp) && !(tp->t_flags & TF_NOOPT) &&
(flags & (TH_SYN | TH_ACK)) == TH_SYN)
- optlen += tcp_tfo_write_cookie(tp, optlen, &len, opt);
+ optlen += tcp_tfo_write_cookie(tp, optlen, len, opt);
if (tfo_enabled(tp) &&
(flags & (TH_SYN | TH_ACK)) == (TH_SYN | TH_ACK) &&
panic("tcphdr too big");
/* Check if there is enough data in the send socket
- * buffer to start measuring bw
+ * buffer to start measuring bw
*/
if ((tp->t_flagsext & TF_MEASURESNDBW) != 0 &&
(tp->t_bwmeas != NULL) &&
(tp->t_flagsext & TF_BWMEAS_INPROGRESS) == 0 &&
- (so->so_snd.sb_cc - (tp->snd_max - tp->snd_una)) >=
+ (so->so_snd.sb_cc - (tp->snd_max - tp->snd_una)) >=
tp->t_bwmeas->bw_minsize) {
tp->t_bwmeas->bw_size = min(
(so->so_snd.sb_cc - (tp->snd_max - tp->snd_una)),
INP_ADD_STAT(inp, cell, wifi, wired,
txbytes, len);
tp->t_stat.txretransmitbytes += len;
+ tp->t_stat.rxmitpkts++;
}
} else {
tcpstat.tcps_sndpack++;
tcpstat.tcps_sndbyte += len;
-
+
if (nstat_collect) {
INP_ADD_STAT(inp, cell, wifi, wired,
txpackets, 1);
INP_ADD_STAT(inp, cell, wifi, wired,
txbytes, len);
}
+ inp_decr_sndbytes_unsent(so, len);
}
#if MPTCP
if (tp->t_mpflags & TMPF_MPTCP_TRUE) {
}
#endif /* MPTCP */
/*
- * try to use the new interface that allocates all
- * the necessary mbuf hdrs under 1 mbuf lock and
- * avoids rescanning the socket mbuf list if
+ * try to use the new interface that allocates all
+ * the necessary mbuf hdrs under 1 mbuf lock and
+ * avoids rescanning the socket mbuf list if
* certain conditions are met. This routine can't
* be used in the following cases...
* 1) the protocol headers exceed the capacity of
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
}
svc_flags |= PKT_SCF_IPV6;
#if PF_ECN
- m->m_pkthdr.pf_mtag.pftag_hdr = (void *)ip6;
- m->m_pkthdr.pf_mtag.pftag_flags |= PF_TAG_HDR_INET6;
+ m_pftag(m)->pftag_hdr = (void *)ip6;
+ m_pftag(m)->pftag_flags |= PF_TAG_HDR_INET6;
#endif /* PF_ECN */
} else
#endif /* INET6 */
ip->ip_tos |= IPTOS_ECN_ECT0;
}
#if PF_ECN
- m->m_pkthdr.pf_mtag.pftag_hdr = (void *)ip;
- m->m_pkthdr.pf_mtag.pftag_flags |= PF_TAG_HDR_INET;
+ m_pftag(m)->pftag_hdr = (void *)ip;
+ m_pftag(m)->pftag_flags |= PF_TAG_HDR_INET;
#endif /* PF_ECN */
}
if (len || (flags & (TH_SYN|TH_FIN)) ||
tp->t_timer[TCPT_PERSIST]) {
th->th_seq = htonl(tp->snd_nxt);
+ if (len > 0) {
+ m->m_pkthdr.tx_start_seq = tp->snd_nxt;
+ m->m_pkthdr.pkt_flags |= PKTF_START_SEQ;
+ }
if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
if (SACK_ENABLED(tp) && len > 1) {
tcp_rxtseg_insert(tp, tp->snd_nxt,
}
} else {
th->th_seq = htonl(p->rxmit);
+ if (len > 0) {
+ m->m_pkthdr.pkt_flags |=
+ (PKTF_TCP_REXMT | PKTF_START_SEQ);
+ m->m_pkthdr.tx_start_seq = p->rxmit;
+ }
tcp_rxtseg_insert(tp, p->rxmit, (p->rxmit + len - 1));
p->rxmit += len;
tp->sackhint.sack_bytes_rexmit += len;
- if (len > 0)
- m->m_pkthdr.pkt_flags |= PKTF_TCP_REXMT;
}
th->th_ack = htonl(tp->rcv_nxt);
tp->last_ack_sent = tp->rcv_nxt;
th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
}
th->th_flags = flags;
- /*
- * Calculate receive window. Don't shrink window,
- * but avoid silly window syndrome.
- */
- if (recwin < (int32_t)(so->so_rcv.sb_hiwat / 4) && recwin < (int)tp->t_maxseg)
- recwin = 0;
- if (recwin < (int32_t)(tp->rcv_adv - tp->rcv_nxt))
- recwin = (int32_t)(tp->rcv_adv - tp->rcv_nxt);
- if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0) {
- if (recwin > (int32_t)slowlink_wsize)
- recwin = slowlink_wsize;
- }
-
-#if TRAFFIC_MGT
- if (tcp_recv_bg == 1 || IS_TCP_RECV_BG(so)) {
- if (tcp_recv_throttle(tp)) {
- uint32_t min_iaj_win =
- tcp_min_iaj_win * tp->t_maxseg;
- if (tp->iaj_rwintop == 0 ||
- SEQ_LT(tp->iaj_rwintop, tp->rcv_adv))
- tp->iaj_rwintop = tp->rcv_adv;
- if (SEQ_LT(tp->iaj_rwintop,
- tp->rcv_nxt + min_iaj_win))
- tp->iaj_rwintop = tp->rcv_nxt + min_iaj_win;
- recwin = min(tp->iaj_rwintop - tp->rcv_nxt, recwin);
- }
- }
-#endif /* TRAFFIC_MGT */
-
- if (recwin > (int32_t)(TCP_MAXWIN << tp->rcv_scale))
- recwin = (int32_t)(TCP_MAXWIN << tp->rcv_scale);
th->th_win = htons((u_short) (recwin>>tp->rcv_scale));
+ if (recwin > 0 && SEQ_LT(tp->rcv_adv, tp->rcv_nxt + recwin))
+ tp->rcv_adv = tp->rcv_nxt + recwin;
/*
* Adjust the RXWIN0SENT flag - indicate that we have advertised
* checksum extended header and data.
*/
m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
+
+ /*
+ * If this is potentially the last packet on the stream, then mark
+ * it in order to enable some optimizations in the underlying
+ * layers
+ */
+ if (tp->t_state != TCPS_ESTABLISHED &&
+ (tp->t_state == TCPS_CLOSING || tp->t_state == TCPS_TIME_WAIT
+ || tp->t_state == TCPS_LAST_ACK || (th->th_flags & TH_RST)))
+ m->m_pkthdr.pkt_flags |= PKTF_LAST_PKT;
+
#if INET6
if (isipv6) {
/*
m->m_pkthdr.csum_flags = CSUM_TCPIPV6;
m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
if (len + optlen)
- th->th_sum = in_addword(th->th_sum,
+ th->th_sum = in_addword(th->th_sum,
htons((u_short)(optlen + len)));
}
else
m->m_pkthdr.csum_flags = CSUM_TCP;
m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
if (len + optlen)
- th->th_sum = in_addword(th->th_sum,
+ th->th_sum = in_addword(th->th_sum,
htons((u_short)(optlen + len)));
}
if (flags & (TH_SYN|TH_FIN)) {
if (flags & TH_SYN)
tp->snd_nxt++;
- if ((flags & TH_FIN) &&
+ if ((flags & TH_FIN) &&
!(tp->t_flags & TF_SENTFIN)) {
tp->snd_nxt++;
tp->t_flags |= TF_SENTFIN;
tp->snd_nxt != tp->snd_una || (flags & TH_FIN))) {
if (tp->t_timer[TCPT_PERSIST]) {
tp->t_timer[TCPT_PERSIST] = 0;
- tp->t_rxtshift = 0;
- tp->t_rxtstart = 0;
tp->t_persist_stop = 0;
+ TCP_RESET_REXMT_STATE(tp);
}
tp->t_timer[TCPT_REXMT] =
OFFSET_FROM_START(tp, tp->t_rxtcur);
/*
* Using SRTT alone to set PTO can cause spurious
* retransmissions on wireless networks where there
- * is a lot of variance in RTT. Taking variance
+ * is a lot of variance in RTT. Taking variance
* into account will avoid this.
*/
srtt = tp->t_srtt >> TCP_RTT_SHIFT;
int xlen = len;
if (flags & TH_SYN)
++xlen;
- if ((flags & TH_FIN) &&
+ if ((flags & TH_FIN) &&
!(tp->t_flags & TF_SENTFIN)) {
++xlen;
tp->t_flags |= TF_SENTFIN;
error = EHOSTUNREACH;
goto out;
}
-
necp_mark_packet_from_socket(m, inp, policy_id, route_rule_id);
+
+ if (net_qos_policy_restricted != 0) {
+ necp_socket_update_qos_marking(inp, inp->inp_route.ro_rt,
+ NULL, route_rule_id);
+ }
}
#endif /* NECP */
!(inp->inp_last_outifp->if_flags & IFF_LOOPBACK)) {
/* Hint to prioritize this packet if
* 1. if the packet has no data
- * 2. the interface supports transmit-start model and did
+ * 2. the interface supports transmit-start model and did
* not disable ACK prioritization.
* 3. Only ACK flag is set.
* 4. there is no outstanding data on this connection.
*/
if (tcp_prioritize_acks != 0 && len == 0 &&
- (inp->inp_last_outifp->if_eflags &
- (IFEF_TXSTART | IFEF_NOACKPRI)) == IFEF_TXSTART &&
- th->th_flags == TH_ACK && tp->snd_una == tp->snd_max &&
- tp->t_timer[TCPT_REXMT] == 0) {
- svc_flags |= PKT_SCF_TCP_ACK;
- }
- set_packet_service_class(m, so, MBUF_SC_UNSPEC, svc_flags);
+ (inp->inp_last_outifp->if_eflags &
+ (IFEF_TXSTART | IFEF_NOACKPRI)) == IFEF_TXSTART) {
+ if (th->th_flags == TH_ACK &&
+ tp->snd_una == tp->snd_max &&
+ tp->t_timer[TCPT_REXMT] == 0)
+ svc_flags |= PKT_SCF_TCP_ACK;
+ if (th->th_flags & TH_SYN)
+ svc_flags |= PKT_SCF_TCP_SYN;
+ }
+ set_packet_service_class(m, so, sotc, svc_flags);
+ } else {
+ /*
+ * Optimization for loopback just set the mbuf
+ * service class
+ */
+ (void) m_set_service_class(m, so_tc2msc(sotc));
}
tp->t_pktlist_sentlen += len;
if ((lro_ackmore) && (!sackoptlen) && (!tp->t_timer[TCPT_PERSIST]) &&
((th->th_flags & TH_ACK) == TH_ACK) && (!len) &&
(tp->t_state == TCPS_ESTABLISHED)) {
- /* For a pure ACK, see if you need to send more of them */
+ /* For a pure ACK, see if you need to send more of them */
mnext = tcp_send_lroacks(tp, m, th);
if (mnext) {
tp->t_pktlist_tail->m_nextpkt = mnext;
error = tcp_ip_output(so, tp, packetlist,
packchain_listadd, tp_inp_options,
(so_options & SO_DONTROUTE),
- (sack_rxmit | (sack_bytes_rxmt != 0)), recwin,
-#if INET6
- isipv6);
-#else /* INET6 */
- 0);
-#endif /* !INET6 */
+ (sack_rxmit | (sack_bytes_rxmt != 0)), isipv6);
if (error) {
/*
* Take into account the rest of unsent
if (error == ENOBUFS) {
if (!tp->t_timer[TCPT_REXMT] &&
- !tp->t_timer[TCPT_PERSIST])
- tp->t_timer[TCPT_REXMT] =
+ !tp->t_timer[TCPT_PERSIST] &&
+ (SEQ_GT(tp->snd_max, tp->snd_una) ||
+ so->so_snd.sb_cc > 0))
+ tp->t_timer[TCPT_REXMT] =
OFFSET_FROM_START(tp, tp->t_rxtcur);
tp->snd_cwnd = tp->t_maxseg;
tp->t_bytes_acked = 0;
* treat EHOSTUNREACH/ENETDOWN as a soft error.
*/
if ((error == EHOSTUNREACH || error == ENETDOWN) &&
- TCPS_HAVERCVDSYN(tp->t_state) &&
+ TCPS_HAVERCVDSYN(tp->t_state) &&
!inp_restricted_send(inp, inp->inp_last_outifp)) {
tp->t_softerror = error;
error = 0;
static int
tcp_ip_output(struct socket *so, struct tcpcb *tp, struct mbuf *pkt,
- int cnt, struct mbuf *opt, int flags, int sack_in_progress, int recwin,
- boolean_t isipv6)
+ int cnt, struct mbuf *opt, int flags, int sack_in_progress, boolean_t isipv6)
{
int error = 0;
boolean_t chain;
boolean_t ifdenied = FALSE;
struct inpcb *inp = tp->t_inpcb;
struct ip_out_args ipoa =
- { IFSCOPE_NONE, { 0 }, IPOAF_SELECT_SRCIF|IPOAF_BOUND_SRCADDR, 0 };
+ { IFSCOPE_NONE, { 0 }, IPOAF_SELECT_SRCIF|IPOAF_BOUND_SRCADDR, 0,
+ SO_TC_UNSPEC, _NET_SERVICE_TYPE_UNSPEC };
struct route ro;
struct ifnet *outif = NULL;
#if INET6
struct ip6_out_args ip6oa =
- { IFSCOPE_NONE, { 0 }, IP6OAF_SELECT_SRCIF|IP6OAF_BOUND_SRCADDR, 0 };
+ { IFSCOPE_NONE, { 0 }, IP6OAF_SELECT_SRCIF|IP6OAF_BOUND_SRCADDR, 0,
+ SO_TC_UNSPEC, _NET_SERVICE_TYPE_UNSPEC };
struct route_in6 ro6;
struct flowadv *adv =
(isipv6 ? &ip6oa.ip6oa_flowadv : &ipoa.ipoa_flowadv);
else
#endif /* INET6 */
ipoa.ipoa_flags |= IPOAF_NO_CELLULAR;
- }
+ }
if (INP_NO_EXPENSIVE(inp)) {
#if INET6
if (isipv6)
else
#endif /* INET6 */
ipoa.ipoa_flags |= IPOAF_NO_EXPENSIVE;
-
+
}
if (INP_AWDL_UNRESTRICTED(inp)) {
#if INET6
else
#endif /* INET6 */
ipoa.ipoa_flags |= IPOAF_AWDL_UNRESTRICTED;
-
+
+ }
+#if INET6
+ if (INP_INTCOPROC_ALLOWED(inp) && isipv6) {
+ ip6oa.ip6oa_flags |= IP6OAF_INTCOPROC_ALLOWED;
+ }
+ if (isipv6) {
+ ip6oa.ip6oa_sotc = so->so_traffic_class;
+ ip6oa.ip6oa_netsvctype = so->so_netsvctype;
+ } else
+#endif /* INET6 */
+ {
+ ipoa.ipoa_sotc = so->so_traffic_class;
+ ipoa.ipoa_netsvctype = so->so_netsvctype;
+ }
+ if ((so->so_flags1 & SOF1_QOSMARKING_ALLOWED)) {
+#if INET6
+ if (isipv6)
+ ip6oa.ip6oa_flags |= IP6OAF_QOSMARKING_ALLOWED;
+ else
+#endif /* INET6 */
+ ipoa.ipoa_flags |= IPOAF_QOSMARKING_ALLOWED;
}
#if INET6
if (isipv6)
inp_route_copyout(inp, &ro);
/*
- * Data sent (as far as we can tell).
- * If this advertises a larger window than any other segment,
- * then remember the size of the advertised window.
* Make sure ACK/DELACK conditions are cleared before
* we unlock the socket.
*/
- if (recwin > 0 && SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv))
- tp->rcv_adv = tp->rcv_nxt + recwin;
tp->last_ack_sent = tp->rcv_nxt;
tp->t_flags &= ~(TF_ACKNOW | TF_DELACK);
tp->t_timer[TCPT_DELACK] = 0;
unlocked = TRUE;
socket_unlock(so, 0);
}
-
+
/*
* Don't send down a chain of packets when:
* - TCP chaining is disabled
if (unlocked)
socket_lock(so, 0);
- /*
+ /*
* Enter flow controlled state if the connection is established
* and is not in recovery.
*
- * A connection will enter suspended state even if it is in
+ * A connection will enter suspended state even if it is in
* recovery.
*/
if (((adv->code == FADV_FLOW_CONTROLLED && !IN_FASTRECOVERY(tp)) ||
- adv->code == FADV_SUSPENDED) &&
+ adv->code == FADV_SUSPENDED) &&
!(tp->t_flags & TF_CLOSING) &&
tp->t_state == TCPS_ESTABLISHED) {
int rc;
rc = inp_set_fc_state(inp, adv->code);
- if (rc == 1)
- tcp_ccdbg_trace(tp, NULL,
+ if (rc == 1)
+ tcp_ccdbg_trace(tp, NULL,
((adv->code == FADV_FLOW_CONTROLLED) ?
TCP_CC_FLOW_CONTROL : TCP_CC_SUSPEND));
}
- /*
+ /*
* When an interface queue gets suspended, some of the
* packets are dropped. Return ENOBUFS, to update the
* pcb state.
#if INET6
if (isipv6) {
- if (ro6.ro_rt != NULL && (outif = ro6.ro_rt->rt_ifp) !=
- inp->in6p_last_outifp)
- inp->in6p_last_outifp = outif;
+ if (ro6.ro_rt != NULL)
+ outif = ro6.ro_rt->rt_ifp;
} else
#endif /* INET6 */
- if (ro.ro_rt != NULL && (outif = ro.ro_rt->rt_ifp) !=
- inp->inp_last_outifp)
- inp->inp_last_outifp = outif;
+ if (ro.ro_rt != NULL)
+ outif = ro.ro_rt->rt_ifp;
- if (error != 0 && ifdenied &&
+ if (outif != NULL && outif != inp->inp_last_outifp &&
+ so->so_snd.sb_cc > 0) {
+ /* Update the send byte count */
+ if (so->so_snd.sb_flags & SB_SNDBYTE_CNT) {
+ inp_decr_sndbytes_total(so, so->so_snd.sb_cc);
+ inp_decr_sndbytes_allunsent(so, tp->snd_una);
+ so->so_snd.sb_flags &= ~SB_SNDBYTE_CNT;
+ }
+ inp->inp_last_outifp = outif;
+ }
+
+ if (error != 0 && ifdenied &&
(INP_NO_CELLULAR(inp) || INP_NO_EXPENSIVE(inp)))
- soevent(inp->inp_socket,
+ soevent(so,
(SO_FILT_HINT_LOCKED|SO_FILT_HINT_IFDENIED));
/* Synchronize cached PCB route & options */
#endif /* INET6 */
inp_route_copyin(inp, &ro);
- if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift == 0 &&
+ if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift == 0 &&
tp->t_inpcb->inp_route.ro_rt != NULL) {
/* If we found the route and there is an rtt on it
* reset the retransmit timer
}
void
-tcp_setpersist(tp)
- register struct tcpcb *tp;
+tcp_setpersist(struct tcpcb *tp)
{
int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
int count = 0;
tcp_seq org_ack = ntohl(th->th_ack);
tcp_seq prev_ack = 0;
- int tack_offset = 28; /* XXX IPv6 and IP options not supported */
- int twin_offset = 34; /* XXX IPv6 and IP options not supported */
+ int tack_offset = 28; /* IPv6 and IP options not supported */
+ int twin_offset = 34; /* IPv6 and IP options not supported */
int ack_size = (tp->t_flags & TF_STRETCHACK) ?
(maxseg_unacked * tp->t_maxseg) : (tp->t_maxseg << 1);
int segs_acked = (tp->t_flags & TF_STRETCHACK) ? maxseg_unacked : 2;
tcp_recv_throttle (struct tcpcb *tp)
{
uint32_t base_rtt, newsize;
- int32_t qdelay;
struct sockbuf *sbrcv = &tp->t_inpcb->inp_socket->so_rcv;
if (tcp_use_rtt_recvbg == 1 &&
TSTMP_SUPPORTED(tp)) {
- /*
+ /*
* Timestamps are supported on this connection. Use
* RTT to look for an increase in latency.
*/
- /*
+ /*
* If the connection is already being throttled, leave it
* in that state until rtt comes closer to base rtt
*/
return (1);
base_rtt = get_base_rtt(tp);
-
+
if (base_rtt != 0 && tp->t_rttcur != 0) {
- qdelay = tp->t_rttcur - base_rtt;
/*
* if latency increased on a background flow,
* return 1 to start throttling.
*/
- if (qdelay > target_qdelay) {
+ if (tp->t_rttcur > (base_rtt + target_qdelay)) {
tp->t_flagsext |= TF_RECV_THROTTLE;
-
+ if (tp->t_recv_throttle_ts == 0)
+ tp->t_recv_throttle_ts = tcp_now;
/*
* Reduce the recv socket buffer size to
* minimize latecy.
*/
- if (sbrcv->sb_idealsize >
+ if (sbrcv->sb_idealsize >
tcp_recv_throttle_minwin) {
newsize = sbrcv->sb_idealsize >> 1;
/* Set a minimum of 16 K */
- newsize =
- max(newsize,
+ newsize =
+ max(newsize,
tcp_recv_throttle_minwin);
sbrcv->sb_idealsize = newsize;
}
*/
if (tp->acc_iaj > tcp_acc_iaj_react_limit)
return (1);
-
+
return (0);
}
projects / apple / xnu.git / blobdiff