]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/netinet/tcp_timer.c
xnu-4903.270.47.tar.gz
[apple/xnu.git] / bsd / netinet / tcp_timer.c
index 706ec823c640ea62fde4e844439b3bd3f499c6fd..fda0f86f674ea97494b32c12a4b9fd1272f4b846 100644 (file)
@@ -1,8 +1,8 @@
 /*
 /*
- * Copyright (c) 2000-2011 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2017 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  *
  * @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
  * 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.
  * 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.
  * 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,
  * 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,
@@ -22,7 +22,7 @@
  * 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.
  * 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@
  */
 /*
  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
 #include <sys/mcache.h>
 #include <sys/queue.h>
 #include <kern/locks.h>
 #include <sys/mcache.h>
 #include <sys/queue.h>
 #include <kern/locks.h>
-
-#include <kern/cpu_number.h>   /* before tcp_seq.h, for tcp_random18() */
+#include <kern/cpu_number.h>    /* before tcp_seq.h, for tcp_random18() */
+#include <mach/boolean.h>
 
 #include <net/route.h>
 
 #include <net/route.h>
+#include <net/if_var.h>
+#include <net/ntstat.h>
 
 #include <netinet/in.h>
 #include <netinet/in_systm.h>
 
 #include <netinet/in.h>
 #include <netinet/in_systm.h>
@@ -87,6 +89,7 @@
 #endif
 #include <netinet/ip_var.h>
 #include <netinet/tcp.h>
 #endif
 #include <netinet/ip_var.h>
 #include <netinet/tcp.h>
+#include <netinet/tcp_cache.h>
 #include <netinet/tcp_fsm.h>
 #include <netinet/tcp_seq.h>
 #include <netinet/tcp_timer.h>
 #include <netinet/tcp_fsm.h>
 #include <netinet/tcp_seq.h>
 #include <netinet/tcp_timer.h>
 #endif
 #include <sys/kdebug.h>
 #include <mach/sdt.h>
 #endif
 #include <sys/kdebug.h>
 #include <mach/sdt.h>
+#include <netinet/mptcp_var.h>
 
 
-extern void postevent(struct socket *, struct sockbuf *,
-                                               int);
-#define DBG_FNC_TCP_FAST       NETDBG_CODE(DBG_NETTCP, (5 << 8))
-#define DBG_FNC_TCP_SLOW       NETDBG_CODE(DBG_NETTCP, (5 << 8) | 1)
-
-#define TIMERENTRY_TO_TP(te) ((struct tcpcb *)((uintptr_t)te - offsetof(struct tcpcb, tentry.le.le_next))) 
+/* Max number of times a stretch ack can be delayed on a connection */
+#define TCP_STRETCHACK_DELAY_THRESHOLD  5
 
 
-#define VERIFY_NEXT_LINK(elm,field) do {       \
-       if (LIST_NEXT((elm),field) != NULL &&   \
-           LIST_NEXT((elm),field)->field.le_prev !=    \
-               &((elm)->field.le_next))        \
-               panic("Bad link elm %p next->prev != elm", (elm));      \
-} while(0)
+/*
+ * If the host processor has been sleeping for too long, this is the threshold
+ * used to avoid sending stale retransmissions.
+ */
+#define TCP_SLEEP_TOO_LONG      (10 * 60 * 1000) /* 10 minutes in ms */
 
 
-#define VERIFY_PREV_LINK(elm,field) do {       \
-       if (*(elm)->field.le_prev != (elm))     \
-               panic("Bad link elm %p prev->next != elm", (elm));      \
-} while(0)
+/* tcp timer list */
+struct tcptimerlist tcp_timer_list;
 
 
-static int     background_io_trigger = 5;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, background_io_trigger, CTLFLAG_RW | CTLFLAG_LOCKED,
-    &background_io_trigger, 0, "Background IO Trigger Setting");
+/* List of pcbs in timewait state, protected by tcbinfo's ipi_lock */
+struct tcptailq tcp_tw_tailq;
 
 static int
 sysctl_msec_to_ticks SYSCTL_HANDLER_ARGS
 {
 
 static int
 sysctl_msec_to_ticks SYSCTL_HANDLER_ARGS
 {
-#pragma unused(arg1, arg2)
+#pragma unused(arg2)
        int error, s, tt;
 
        int error, s, tt;
 
-       tt = *(int *)oidp->oid_arg1;
+       tt = *(int *)arg1;
        s = tt * 1000 / TCP_RETRANSHZ;;
 
        error = sysctl_handle_int(oidp, &s, 0, req);
        s = tt * 1000 / TCP_RETRANSHZ;;
 
        error = sysctl_handle_int(oidp, &s, 0, req);
-       if (error || !req->newptr)
-               return (error);
+       if (error || !req->newptr) {
+               return error;
+       }
 
        tt = s * TCP_RETRANSHZ / 1000;
 
        tt = s * TCP_RETRANSHZ / 1000;
-       if (tt < 1)
-               return (EINVAL);
+       if (tt < 1) {
+               return EINVAL;
+       }
 
 
-       *(int *)oidp->oid_arg1 = tt;
-        return (0);
+       *(int *)arg1 = tt;
+       SYSCTL_SKMEM_UPDATE_AT_OFFSET(arg2, *(int*)arg1);
+       return 0;
 }
 
 }
 
-int    tcp_keepinit;
-SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+#if SYSCTL_SKMEM
+int     tcp_keepinit = TCPTV_KEEP_INIT;
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+    &tcp_keepinit, offsetof(skmem_sysctl, tcp.keepinit),
+    sysctl_msec_to_ticks, "I", "");
+
+int     tcp_keepidle = TCPTV_KEEP_IDLE;
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+    &tcp_keepidle, offsetof(skmem_sysctl, tcp.keepidle),
+    sysctl_msec_to_ticks, "I", "");
+
+int     tcp_keepintvl = TCPTV_KEEPINTVL;
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+    &tcp_keepintvl, offsetof(skmem_sysctl, tcp.keepintvl),
+    sysctl_msec_to_ticks, "I", "");
+
+SYSCTL_SKMEM_TCP_INT(OID_AUTO, keepcnt,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+    int, tcp_keepcnt, TCPTV_KEEPCNT, "number of times to repeat keepalive");
+
+int     tcp_msl = TCPTV_MSL;
+SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+    &tcp_msl, offsetof(skmem_sysctl, tcp.msl),
+    sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
+#else /* SYSCTL_SKMEM */
+int     tcp_keepinit;
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
 
     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
 
-int    tcp_keepidle;
-SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+int     tcp_keepidle;
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
 
     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
 
-int    tcp_keepintvl;
-SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+int     tcp_keepintvl;
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
 
     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
 
-int    tcp_msl;
-SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+int     tcp_keepcnt;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+    &tcp_keepcnt, 0, "number of times to repeat keepalive");
+
+int     tcp_msl;
+SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
+#endif /* SYSCTL_SKMEM */
 
 
-/* 
- * Avoid DoS via TCP Robustness in Persist Condition (see http://www.ietf.org/id/draft-ananth-tcpm-persist-02.txt)
- * by allowing a system wide maximum persistence timeout value when in Zero Window Probe mode.
- * Expressed in milliseconds to be consistent without timeout related values, the TCP socket option is in seconds.
+/*
+ * Avoid DoS via TCP Robustness in Persist Condition
+ * (see http://www.ietf.org/id/draft-ananth-tcpm-persist-02.txt)
+ * by allowing a system wide maximum persistence timeout value when in
+ * Zero Window Probe mode.
+ *
+ * Expressed in milliseconds to be consistent without timeout related
+ * values, the TCP socket option is in seconds.
  */
  */
+#if SYSCTL_SKMEM
 u_int32_t tcp_max_persist_timeout = 0;
 u_int32_t tcp_max_persist_timeout = 0;
-SYSCTL_PROC(_net_inet_tcp, OID_AUTO, max_persist_timeout, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
-    &tcp_max_persist_timeout, 0, sysctl_msec_to_ticks, "I", "Maximum persistence timout for ZWP");
+SYSCTL_PROC(_net_inet_tcp, OID_AUTO, max_persist_timeout,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+    &tcp_max_persist_timeout, offsetof(skmem_sysctl, tcp.max_persist_timeout),
+    sysctl_msec_to_ticks, "I", "Maximum persistence timeout for ZWP");
+#else /* SYSCTL_SKMEM */
+u_int32_t tcp_max_persist_timeout = 0;
+SYSCTL_PROC(_net_inet_tcp, OID_AUTO, max_persist_timeout,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+    &tcp_max_persist_timeout, 0, sysctl_msec_to_ticks, "I",
+    "Maximum persistence timeout for ZWP");
+#endif /* SYSCTL_SKMEM */
 
 
-static int     always_keepalive = 0;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW | CTLFLAG_LOCKED,
-    &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
+SYSCTL_SKMEM_TCP_INT(OID_AUTO, always_keepalive,
+    CTLFLAG_RW | CTLFLAG_LOCKED, static int, always_keepalive, 0,
+    "Assume SO_KEEPALIVE on all TCP connections");
 
 
-/* This parameter determines how long the timer list will stay in fast mode even
- * though all connections are idle. In fast mode, the timer will fire more frequently
- * anticipating new data.
+/*
+ * This parameter determines how long the timer list will stay in fast or
+ * quick mode even though all connections are idle. In this state, the
+ * timer will run more frequently anticipating new data.
  */
  */
-int timer_fastmode_idlemax = TCP_FASTMODE_IDLEGEN_MAX;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_fastmode_idlemax, CTLFLAG_RW | CTLFLAG_LOCKED,
-       &timer_fastmode_idlemax, 0, "Maximum idle generations in fast mode");
+SYSCTL_SKMEM_TCP_INT(OID_AUTO, timer_fastmode_idlemax,
+    CTLFLAG_RW | CTLFLAG_LOCKED, int, timer_fastmode_idlemax,
+    TCP_FASTMODE_IDLERUN_MAX, "Maximum idle generations in fast mode");
 
 /*
  * See tcp_syn_backoff[] for interval values between SYN retransmits;
 
 /*
  * See tcp_syn_backoff[] for interval values between SYN retransmits;
@@ -190,138 +243,331 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_fastmode_idlemax, CTLFLAG_RW | CTLFLAG
  * SYN retransmits.  Setting it to 0 disables the dropping off of those
  * two options.
  */
  * SYN retransmits.  Setting it to 0 disables the dropping off of those
  * two options.
  */
-static int tcp_broken_peer_syn_rxmit_thres = 7;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, broken_peer_syn_rxmit_thres, CTLFLAG_RW | CTLFLAG_LOCKED,
-    &tcp_broken_peer_syn_rxmit_thres, 0, "Number of retransmitted SYNs before "
-    "TCP disables rfc1323 and rfc1644 during the rest of attempts");
+SYSCTL_SKMEM_TCP_INT(OID_AUTO, broken_peer_syn_rexmit_thres,
+    CTLFLAG_RW | CTLFLAG_LOCKED, static int, tcp_broken_peer_syn_rxmit_thres,
+    10, "Number of retransmitted SYNs before disabling RFC 1323 "
+    "options on local connections");
 
 static int tcp_timer_advanced = 0;
 
 static int tcp_timer_advanced = 0;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_timer_advanced, CTLFLAG_RD | CTLFLAG_LOCKED,
-    &tcp_timer_advanced, 0, "Number of times one of the timers was advanced");
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_timer_advanced,
+    CTLFLAG_RD | CTLFLAG_LOCKED, &tcp_timer_advanced, 0,
+    "Number of times one of the timers was advanced");
 
 static int tcp_resched_timerlist = 0;
 
 static int tcp_resched_timerlist = 0;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_resched_timerlist, CTLFLAG_RD | CTLFLAG_LOCKED,
-    &tcp_resched_timerlist, 0, 
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_resched_timerlist,
+    CTLFLAG_RD | CTLFLAG_LOCKED, &tcp_resched_timerlist, 0,
     "Number of times timer list was rescheduled as part of processing a packet");
 
     "Number of times timer list was rescheduled as part of processing a packet");
 
-int    tcp_pmtud_black_hole_detect = 1 ;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection, CTLFLAG_RW | CTLFLAG_LOCKED,
-    &tcp_pmtud_black_hole_detect, 0, "Path MTU Discovery Black Hole Detection");
+SYSCTL_SKMEM_TCP_INT(OID_AUTO, pmtud_blackhole_detection,
+    CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_pmtud_black_hole_detect, 1,
+    "Path MTU Discovery Black Hole Detection");
 
 
-int    tcp_pmtud_black_hole_mss = 1200 ;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss, CTLFLAG_RW | CTLFLAG_LOCKED,
-    &tcp_pmtud_black_hole_mss, 0, "Path MTU Discovery Black Hole Detection lowered MSS");
+SYSCTL_SKMEM_TCP_INT(OID_AUTO, pmtud_blackhole_mss,
+    CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_pmtud_black_hole_mss, 1200,
+    "Path MTU Discovery Black Hole Detection lowered MSS");
 
 
-static int     tcp_keepcnt = TCPTV_KEEPCNT;
-static int     tcp_gc_done = FALSE;    /* perfromed garbage collection of "used" sockets */
-       /* max idle probes */
-int    tcp_maxpersistidle;
-       /* max idle time in persist */
-int    tcp_maxidle;
+static u_int32_t tcp_mss_rec_medium = 1200;
+static u_int32_t tcp_mss_rec_low = 512;
 
 
-/* TCP delack timer is set to 100 ms. Since the processing of timer list in fast
- * mode will happen no faster than 100 ms, the delayed ack timer will fire some where 
- * between 100 and 200 ms.
- */
-int    tcp_delack = TCP_RETRANSHZ / 10;
+#define TCP_REPORT_STATS_INTERVAL       43200 /* 12 hours, in seconds */
+int tcp_report_stats_interval = TCP_REPORT_STATS_INTERVAL;
 
 
-struct inpcbhead       time_wait_slots[N_TIME_WAIT_SLOTS];
-int            cur_tw_slot = 0;
+/* performed garbage collection of "used" sockets */
+static boolean_t tcp_gc_done = FALSE;
 
 
-/* tcp timer list */
-struct tcptimerlist tcp_timer_list;
+/* max idle probes */
+int     tcp_maxpersistidle = TCPTV_KEEP_IDLE;
+
+/*
+ * TCP delack timer is set to 100 ms. Since the processing of timer list
+ * in fast mode will happen no faster than 100 ms, the delayed ack timer
+ * will fire some where between 100 and 200 ms.
+ */
+int     tcp_delack = TCP_RETRANSHZ / 10;
 
 
-/* The frequency of running through the TCP timer list in 
- * fast and slow mode can be configured.
+#if MPTCP
+/*
+ * MP_JOIN retransmission of 3rd ACK will be every 500 msecs without backoff
  */
  */
-SYSCTL_UINT(_net_inet_tcp, OID_AUTO, timer_fastquantum, CTLFLAG_RW | CTLFLAG_LOCKED,
-       &tcp_timer_list.fast_quantum, TCP_FASTTIMER_QUANTUM, 
-       "Frequency of running timer list in fast mode");
+int     tcp_jack_rxmt = TCP_RETRANSHZ / 2;
+#endif /* MPTCP */
 
 
-SYSCTL_UINT(_net_inet_tcp, OID_AUTO, timer_slowquantum, CTLFLAG_RW | CTLFLAG_LOCKED,
-       &tcp_timer_list.slow_quantum, TCP_SLOWTIMER_QUANTUM, 
-       "Frequency of running timer list in slow mode");
+static boolean_t tcp_itimer_done = FALSE;
 
 static void tcp_remove_timer(struct tcpcb *tp);
 static void tcp_sched_timerlist(uint32_t offset);
 
 static void tcp_remove_timer(struct tcpcb *tp);
 static void tcp_sched_timerlist(uint32_t offset);
-static uint32_t tcp_run_conn_timer(struct tcpcb *tp, uint16_t *next_index);
+static u_int32_t tcp_run_conn_timer(struct tcpcb *tp, u_int16_t *mode,
+    u_int16_t probe_if_index);
 static void tcp_sched_timers(struct tcpcb *tp);
 static inline void tcp_set_lotimer_index(struct tcpcb *);
 static void tcp_sched_timers(struct tcpcb *tp);
 static inline void tcp_set_lotimer_index(struct tcpcb *);
+__private_extern__ void tcp_remove_from_time_wait(struct inpcb *inp);
+static inline void tcp_update_mss_core(struct tcpcb *tp, struct ifnet *ifp);
+__private_extern__ void tcp_report_stats(void);
 
 
-/* Macro to compare two timers. If there is a reset of the sign bit, it is 
- * safe to assume that the timer has wrapped around. By doing signed comparision, 
- * we take care of wrap around such that the value with the sign bit reset is 
- * actually ahead of the other.
- */
+static  u_int64_t tcp_last_report_time;
 
 
-static inline int32_t
-timer_diff(uint32_t t1, uint32_t toff1, uint32_t t2, uint32_t toff2) { 
-       return (int32_t)((t1 + toff1) - (t2 + toff2));
+/*
+ * Structure to store previously reported stats so that we can send
+ * incremental changes in each report interval.
+ */
+struct tcp_last_report_stats {
+       u_int32_t       tcps_connattempt;
+       u_int32_t       tcps_accepts;
+       u_int32_t       tcps_ecn_client_setup;
+       u_int32_t       tcps_ecn_server_setup;
+       u_int32_t       tcps_ecn_client_success;
+       u_int32_t       tcps_ecn_server_success;
+       u_int32_t       tcps_ecn_not_supported;
+       u_int32_t       tcps_ecn_lost_syn;
+       u_int32_t       tcps_ecn_lost_synack;
+       u_int32_t       tcps_ecn_recv_ce;
+       u_int32_t       tcps_ecn_recv_ece;
+       u_int32_t       tcps_ecn_sent_ece;
+       u_int32_t       tcps_ecn_conn_recv_ce;
+       u_int32_t       tcps_ecn_conn_recv_ece;
+       u_int32_t       tcps_ecn_conn_plnoce;
+       u_int32_t       tcps_ecn_conn_pl_ce;
+       u_int32_t       tcps_ecn_conn_nopl_ce;
+       u_int32_t       tcps_ecn_fallback_synloss;
+       u_int32_t       tcps_ecn_fallback_reorder;
+       u_int32_t       tcps_ecn_fallback_ce;
+
+       /* TFO-related statistics */
+       u_int32_t       tcps_tfo_syn_data_rcv;
+       u_int32_t       tcps_tfo_cookie_req_rcv;
+       u_int32_t       tcps_tfo_cookie_sent;
+       u_int32_t       tcps_tfo_cookie_invalid;
+       u_int32_t       tcps_tfo_cookie_req;
+       u_int32_t       tcps_tfo_cookie_rcv;
+       u_int32_t       tcps_tfo_syn_data_sent;
+       u_int32_t       tcps_tfo_syn_data_acked;
+       u_int32_t       tcps_tfo_syn_loss;
+       u_int32_t       tcps_tfo_blackhole;
+       u_int32_t       tcps_tfo_cookie_wrong;
+       u_int32_t       tcps_tfo_no_cookie_rcv;
+       u_int32_t       tcps_tfo_heuristics_disable;
+       u_int32_t       tcps_tfo_sndblackhole;
+
+       /* MPTCP-related statistics */
+       u_int32_t       tcps_mptcp_handover_attempt;
+       u_int32_t       tcps_mptcp_interactive_attempt;
+       u_int32_t       tcps_mptcp_aggregate_attempt;
+       u_int32_t       tcps_mptcp_fp_handover_attempt;
+       u_int32_t       tcps_mptcp_fp_interactive_attempt;
+       u_int32_t       tcps_mptcp_fp_aggregate_attempt;
+       u_int32_t       tcps_mptcp_heuristic_fallback;
+       u_int32_t       tcps_mptcp_fp_heuristic_fallback;
+       u_int32_t       tcps_mptcp_handover_success_wifi;
+       u_int32_t       tcps_mptcp_handover_success_cell;
+       u_int32_t       tcps_mptcp_interactive_success;
+       u_int32_t       tcps_mptcp_aggregate_success;
+       u_int32_t       tcps_mptcp_fp_handover_success_wifi;
+       u_int32_t       tcps_mptcp_fp_handover_success_cell;
+       u_int32_t       tcps_mptcp_fp_interactive_success;
+       u_int32_t       tcps_mptcp_fp_aggregate_success;
+       u_int32_t       tcps_mptcp_handover_cell_from_wifi;
+       u_int32_t       tcps_mptcp_handover_wifi_from_cell;
+       u_int32_t       tcps_mptcp_interactive_cell_from_wifi;
+       u_int64_t       tcps_mptcp_handover_cell_bytes;
+       u_int64_t       tcps_mptcp_interactive_cell_bytes;
+       u_int64_t       tcps_mptcp_aggregate_cell_bytes;
+       u_int64_t       tcps_mptcp_handover_all_bytes;
+       u_int64_t       tcps_mptcp_interactive_all_bytes;
+       u_int64_t       tcps_mptcp_aggregate_all_bytes;
+       u_int32_t       tcps_mptcp_back_to_wifi;
+       u_int32_t       tcps_mptcp_wifi_proxy;
+       u_int32_t       tcps_mptcp_cell_proxy;
+       u_int32_t       tcps_mptcp_triggered_cell;
 };
 
 };
 
+
 /* Returns true if the timer is on the timer list */
 #define TIMER_IS_ON_LIST(tp) ((tp)->t_flags & TF_TIMER_ONLIST)
 
 /* Returns true if the timer is on the timer list */
 #define TIMER_IS_ON_LIST(tp) ((tp)->t_flags & TF_TIMER_ONLIST)
 
+/* Run the TCP timerlist atleast once every hour */
+#define TCP_TIMERLIST_MAX_OFFSET (60 * 60 * TCP_RETRANSHZ)
 
 
-void   add_to_time_wait_locked(struct tcpcb *tp, uint32_t delay);
-void   add_to_time_wait(struct tcpcb *tp, uint32_t delay) ;
 
 
-static void tcp_garbage_collect(struct inpcb *, int);
+static void add_to_time_wait_locked(struct tcpcb *tp, uint32_t delay);
+static boolean_t tcp_garbage_collect(struct inpcb *, int);
 
 
-void   add_to_time_wait_locked(struct tcpcb *tp, uint32_t delay) 
+#define TIMERENTRY_TO_TP(te) ((struct tcpcb *)((uintptr_t)te - offsetof(struct tcpcb, tentry.le.le_next)))
+
+#define VERIFY_NEXT_LINK(elm, field) do {       \
+       if (LIST_NEXT((elm),field) != NULL &&   \
+           LIST_NEXT((elm),field)->field.le_prev !=    \
+               &((elm)->field.le_next))        \
+               panic("Bad link elm %p next->prev != elm", (elm));      \
+} while(0)
+
+#define VERIFY_PREV_LINK(elm, field) do {       \
+       if (*(elm)->field.le_prev != (elm))     \
+               panic("Bad link elm %p prev->next != elm", (elm));      \
+} while(0)
+
+#define TCP_SET_TIMER_MODE(mode, i) do { \
+       if (IS_TIMER_HZ_10MS(i)) \
+               (mode) |= TCP_TIMERLIST_10MS_MODE; \
+       else if (IS_TIMER_HZ_100MS(i)) \
+               (mode) |= TCP_TIMERLIST_100MS_MODE; \
+       else \
+               (mode) |= TCP_TIMERLIST_500MS_MODE; \
+} while(0)
+
+#if (DEVELOPMENT || DEBUG)
+SYSCTL_UINT(_net_inet_tcp, OID_AUTO, mss_rec_medium,
+    CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_mss_rec_medium, 0,
+    "Medium MSS based on recommendation in link status report");
+SYSCTL_UINT(_net_inet_tcp, OID_AUTO, mss_rec_low,
+    CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_mss_rec_low, 0,
+    "Low MSS based on recommendation in link status report");
+
+static int32_t tcp_change_mss_recommended = 0;
+static int
+sysctl_change_mss_recommended SYSCTL_HANDLER_ARGS
 {
 {
-       int             tw_slot;
-       struct inpcbinfo *pcbinfo       = &tcbinfo;
-       uint32_t timer;
+#pragma unused(oidp, arg1, arg2)
+       int i, err = 0, changed = 0;
+       struct ifnet *ifp;
+       struct if_link_status ifsr;
+       struct if_cellular_status_v1 *new_cell_sr;
+       err = sysctl_io_number(req, tcp_change_mss_recommended,
+           sizeof(int32_t), &i, &changed);
+       if (changed) {
+               ifnet_head_lock_shared();
+               TAILQ_FOREACH(ifp, &ifnet_head, if_link) {
+                       if (IFNET_IS_CELLULAR(ifp)) {
+                               bzero(&ifsr, sizeof(ifsr));
+                               new_cell_sr = &ifsr.ifsr_u.ifsr_cell.if_cell_u.if_status_v1;
+                               ifsr.ifsr_version = IF_CELLULAR_STATUS_REPORT_CURRENT_VERSION;
+                               ifsr.ifsr_len = sizeof(*new_cell_sr);
+
+                               /* Set MSS recommended */
+                               new_cell_sr->valid_bitmask |= IF_CELL_UL_MSS_RECOMMENDED_VALID;
+                               new_cell_sr->mss_recommended = i;
+                               err = ifnet_link_status_report(ifp, new_cell_sr, sizeof(new_cell_sr));
+                               if (err == 0) {
+                                       tcp_change_mss_recommended = i;
+                               } else {
+                                       break;
+                               }
+                       }
+               }
+               ifnet_head_done();
+       }
+       return err;
+}
+
+SYSCTL_PROC(_net_inet_tcp, OID_AUTO, change_mss_recommended,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_change_mss_recommended,
+    0, sysctl_change_mss_recommended, "IU", "Change MSS recommended");
 
 
-       /* pcb list should be locked when we get here */        
-       lck_rw_assert(pcbinfo->mtx, LCK_RW_ASSERT_EXCLUSIVE);
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, report_stats_interval,
+    CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_report_stats_interval, 0,
+    "Report stats interval");
+#endif /* (DEVELOPMENT || DEBUG) */
 
 
-       LIST_REMOVE(tp->t_inpcb, inp_list);
+/*
+ * Macro to compare two timers. If there is a reset of the sign bit,
+ * it is safe to assume that the timer has wrapped around. By doing
+ * signed comparision, we take care of wrap around such that the value
+ * with the sign bit reset is actually ahead of the other.
+ */
+inline int32_t
+timer_diff(uint32_t t1, uint32_t toff1, uint32_t t2, uint32_t toff2)
+{
+       return (int32_t)((t1 + toff1) - (t2 + toff2));
+};
+
+/*
+ * Add to tcp timewait list, delay is given in milliseconds.
+ */
+static void
+add_to_time_wait_locked(struct tcpcb *tp, uint32_t delay)
+{
+       struct inpcbinfo *pcbinfo = &tcbinfo;
+       struct inpcb *inp = tp->t_inpcb;
+       uint32_t timer;
 
 
-       /* if (tp->t_timer[TCPT_2MSL] <= 0) 
-           tp->t_timer[TCPT_2MSL] = 1; */
+       /* pcb list should be locked when we get here */
+       LCK_RW_ASSERT(pcbinfo->ipi_lock, LCK_RW_ASSERT_EXCLUSIVE);
 
 
-       /*
-        * Because we're pulling this pcb out of the main TCP pcb list,
-        * we need to recalculate the TCPT_2MSL timer value for tcp_slowtimo
-        * higher timer granularity.
-        */
+       /* We may get here multiple times, so check */
+       if (!(inp->inp_flags2 & INP2_TIMEWAIT)) {
+               pcbinfo->ipi_twcount++;
+               inp->inp_flags2 |= INP2_TIMEWAIT;
+
+               /* Remove from global inp list */
+               LIST_REMOVE(inp, inp_list);
+       } else {
+               TAILQ_REMOVE(&tcp_tw_tailq, tp, t_twentry);
+       }
 
 
-       timer = (delay / TCP_RETRANSHZ) * PR_SLOWHZ;
-       tp->t_rcvtime = (tp->t_rcvtime / TCP_RETRANSHZ) * PR_SLOWHZ;
+       /* Compute the time at which this socket can be closed */
+       timer = tcp_now + delay;
 
 
-       tp->t_rcvtime += timer & (N_TIME_WAIT_SLOTS - 1); 
+       /* We will use the TCPT_2MSL timer for tracking this delay */
 
 
-       tw_slot = (timer & (N_TIME_WAIT_SLOTS - 1)) + cur_tw_slot; 
-       if (tw_slot >= N_TIME_WAIT_SLOTS)
-           tw_slot -= N_TIME_WAIT_SLOTS;
+       if (TIMER_IS_ON_LIST(tp)) {
+               tcp_remove_timer(tp);
+       }
+       tp->t_timer[TCPT_2MSL] = timer;
 
 
-       LIST_INSERT_HEAD(&time_wait_slots[tw_slot], tp->t_inpcb, inp_list);
+       TAILQ_INSERT_TAIL(&tcp_tw_tailq, tp, t_twentry);
 }
 
 }
 
-void   add_to_time_wait(struct tcpcb *tp, uint32_t delay) 
+void
+add_to_time_wait(struct tcpcb *tp, uint32_t delay)
 {
 {
-       struct inpcbinfo *pcbinfo               = &tcbinfo;
-       
-       if (!lck_rw_try_lock_exclusive(pcbinfo->mtx)) {
-               tcp_unlock(tp->t_inpcb->inp_socket, 0, 0);
-               lck_rw_lock_exclusive(pcbinfo->mtx);
-               tcp_lock(tp->t_inpcb->inp_socket, 0, 0);
+       struct inpcbinfo *pcbinfo = &tcbinfo;
+       if (tp->t_inpcb->inp_socket->so_options & SO_NOWAKEFROMSLEEP) {
+               socket_post_kev_msg_closed(tp->t_inpcb->inp_socket);
+       }
+
+       /* 19182803: Notify nstat that connection is closing before waiting. */
+       nstat_pcb_detach(tp->t_inpcb);
+
+       if (!lck_rw_try_lock_exclusive(pcbinfo->ipi_lock)) {
+               socket_unlock(tp->t_inpcb->inp_socket, 0);
+               lck_rw_lock_exclusive(pcbinfo->ipi_lock);
+               socket_lock(tp->t_inpcb->inp_socket, 0);
        }
        add_to_time_wait_locked(tp, delay);
        }
        add_to_time_wait_locked(tp, delay);
-       lck_rw_done(pcbinfo->mtx);
+       lck_rw_done(pcbinfo->ipi_lock);
+
+       inpcb_gc_sched(pcbinfo, INPCB_TIMER_LAZY);
 }
 
 }
 
-static void
+/* If this is on time wait queue, remove it. */
+void
+tcp_remove_from_time_wait(struct inpcb *inp)
+{
+       struct tcpcb *tp = intotcpcb(inp);
+       if (inp->inp_flags2 & INP2_TIMEWAIT) {
+               TAILQ_REMOVE(&tcp_tw_tailq, tp, t_twentry);
+       }
+}
+
+static boolean_t
 tcp_garbage_collect(struct inpcb *inp, int istimewait)
 {
 tcp_garbage_collect(struct inpcb *inp, int istimewait)
 {
-       struct socket *so;
+       boolean_t active = FALSE;
+       struct socket *so, *mp_so = NULL;
        struct tcpcb *tp;
 
        so = inp->inp_socket;
        tp = intotcpcb(inp);
 
        struct tcpcb *tp;
 
        so = inp->inp_socket;
        tp = intotcpcb(inp);
 
+       if (so->so_flags & SOF_MP_SUBFLOW) {
+               mp_so = mptetoso(tptomptp(tp)->mpt_mpte);
+               if (!socket_try_lock(mp_so)) {
+                       mp_so = NULL;
+                       active = TRUE;
+                       goto out;
+               }
+               mp_so->so_usecount++;
+       }
+
        /*
         * Skip if still in use or busy; it would have been more efficient
         * if we were to test so_usecount against 0, but this isn't possible
        /*
         * Skip if still in use or busy; it would have been more efficient
         * if we were to test so_usecount against 0, but this isn't possible
@@ -329,13 +575,25 @@ tcp_garbage_collect(struct inpcb *inp, int istimewait)
         * overflow sockets that are eligible for garbage collection have
         * their usecounts set to 1.
         */
         * overflow sockets that are eligible for garbage collection have
         * their usecounts set to 1.
         */
-       if (so->so_usecount > 1 || !lck_mtx_try_lock_spin(&inp->inpcb_mtx))
-               return;
+       if (!lck_mtx_try_lock_spin(&inp->inpcb_mtx)) {
+               active = TRUE;
+               goto out;
+       }
 
        /* Check again under the lock */
        if (so->so_usecount > 1) {
 
        /* Check again under the lock */
        if (so->so_usecount > 1) {
+               if (inp->inp_wantcnt == WNT_STOPUSING) {
+                       active = TRUE;
+               }
                lck_mtx_unlock(&inp->inpcb_mtx);
                lck_mtx_unlock(&inp->inpcb_mtx);
-               return;
+               goto out;
+       }
+
+       if (istimewait && TSTMP_GEQ(tcp_now, tp->t_timer[TCPT_2MSL]) &&
+           tp->t_state != TCPS_CLOSED) {
+               /* Become a regular mutex */
+               lck_mtx_convert_spin(&inp->inpcb_mtx);
+               tcp_close(tp);
        }
 
        /*
        }
 
        /*
@@ -343,196 +601,285 @@ tcp_garbage_collect(struct inpcb *inp, int istimewait)
         * only if we are called to clean up the time wait slots, since
         * tcp_dropdropablreq() considers a socket to have been fully
         * dropped after add_to_time_wait() is finished.
         * only if we are called to clean up the time wait slots, since
         * tcp_dropdropablreq() considers a socket to have been fully
         * dropped after add_to_time_wait() is finished.
-        * Also handle the case of connections getting closed by the peer while in the queue as
-        * seen with rdar://6422317
-        * 
+        * Also handle the case of connections getting closed by the peer
+        * while in the queue as seen with rdar://6422317
+        *
         */
         */
-       if (so->so_usecount == 1 && 
+       if (so->so_usecount == 1 &&
            ((istimewait && (so->so_flags & SOF_OVERFLOW)) ||
            ((istimewait && (so->so_flags & SOF_OVERFLOW)) ||
-           ((tp != NULL) && (tp->t_state == TCPS_CLOSED) && (so->so_head != NULL)
-                && ((so->so_state & (SS_INCOMP|SS_CANTSENDMORE|SS_CANTRCVMORE)) ==
-                        (SS_INCOMP|SS_CANTSENDMORE|SS_CANTRCVMORE))))) {
-
+           ((tp != NULL) && (tp->t_state == TCPS_CLOSED) &&
+           (so->so_head != NULL) &&
+           ((so->so_state & (SS_INCOMP | SS_CANTSENDMORE | SS_CANTRCVMORE)) ==
+           (SS_INCOMP | SS_CANTSENDMORE | SS_CANTRCVMORE))))) {
                if (inp->inp_state != INPCB_STATE_DEAD) {
                        /* Become a regular mutex */
                        lck_mtx_convert_spin(&inp->inpcb_mtx);
 #if INET6
                if (inp->inp_state != INPCB_STATE_DEAD) {
                        /* Become a regular mutex */
                        lck_mtx_convert_spin(&inp->inpcb_mtx);
 #if INET6
-                       if (INP_CHECK_SOCKAF(so, AF_INET6))
+                       if (SOCK_CHECK_DOM(so, PF_INET6)) {
                                in6_pcbdetach(inp);
                                in6_pcbdetach(inp);
-                       else
+                       else
 #endif /* INET6 */
                        in_pcbdetach(inp);
                }
 #endif /* INET6 */
                        in_pcbdetach(inp);
                }
+               VERIFY(so->so_usecount > 0);
                so->so_usecount--;
                so->so_usecount--;
+               if (inp->inp_wantcnt == WNT_STOPUSING) {
+                       active = TRUE;
+               }
                lck_mtx_unlock(&inp->inpcb_mtx);
                lck_mtx_unlock(&inp->inpcb_mtx);
-               return;
+               goto out;
        } else if (inp->inp_wantcnt != WNT_STOPUSING) {
                lck_mtx_unlock(&inp->inpcb_mtx);
        } else if (inp->inp_wantcnt != WNT_STOPUSING) {
                lck_mtx_unlock(&inp->inpcb_mtx);
-               return;
+               active = FALSE;
+               goto out;
        }
 
        /*
        }
 
        /*
-        * We get here because the PCB is no longer searchable (WNT_STOPUSING);
-        * detach (if needed) and dispose if it is dead (usecount is 0).  This
-        * covers all cases, including overflow sockets and those that are
-        * considered as "embryonic", i.e. created by sonewconn() in TCP input
-        * path, and have not yet been committed.  For the former, we reduce
-        * the usecount to 0 as done by the code above.  For the latter, the
-        * usecount would have reduced to 0 as part calling soabort() when the
+        * We get here because the PCB is no longer searchable
+        * (WNT_STOPUSING); detach (if needed) and dispose if it is dead
+        * (usecount is 0).  This covers all cases, including overflow
+        * sockets and those that are considered as "embryonic",
+        * i.e. created by sonewconn() in TCP input path, and have
+        * not yet been committed.  For the former, we reduce the usecount
+        *  to 0 as done by the code above.  For the latter, the usecount
+        * would have reduced to 0 as part calling soabort() when the
         * socket is dropped at the end of tcp_input().
         */
        if (so->so_usecount == 0) {
                DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
         * socket is dropped at the end of tcp_input().
         */
        if (so->so_usecount == 0) {
                DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
-                       struct tcpcb *, tp, int32_t, TCPS_CLOSED);
+                   struct tcpcb *, tp, int32_t, TCPS_CLOSED);
                /* Become a regular mutex */
                lck_mtx_convert_spin(&inp->inpcb_mtx);
                /* Become a regular mutex */
                lck_mtx_convert_spin(&inp->inpcb_mtx);
+
+               /*
+                * If this tp still happens to be on the timer list,
+                * take it out
+                */
+               if (TIMER_IS_ON_LIST(tp)) {
+                       tcp_remove_timer(tp);
+               }
+
                if (inp->inp_state != INPCB_STATE_DEAD) {
 #if INET6
                if (inp->inp_state != INPCB_STATE_DEAD) {
 #if INET6
-                       if (INP_CHECK_SOCKAF(so, AF_INET6))
+                       if (SOCK_CHECK_DOM(so, PF_INET6)) {
                                in6_pcbdetach(inp);
                                in6_pcbdetach(inp);
-                       else
+                       else
 #endif /* INET6 */
                        in_pcbdetach(inp);
                }
 #endif /* INET6 */
                        in_pcbdetach(inp);
                }
+
+               if (mp_so) {
+                       mptcp_subflow_del(tptomptp(tp)->mpt_mpte, tp->t_mpsub);
+
+                       /* so is now unlinked from mp_so - let's drop the lock */
+                       socket_unlock(mp_so, 1);
+                       mp_so = NULL;
+               }
+
                in_pcbdispose(inp);
                in_pcbdispose(inp);
-       } else {
-               lck_mtx_unlock(&inp->inpcb_mtx);
+               active = FALSE;
+               goto out;
+       }
+
+       lck_mtx_unlock(&inp->inpcb_mtx);
+       active = TRUE;
+
+out:
+       if (mp_so) {
+               socket_unlock(mp_so, 1);
        }
        }
+
+       return active;
 }
 
 }
 
+/*
+ * TCP garbage collector callback (inpcb_timer_func_t).
+ *
+ * Returns the number of pcbs that will need to be gc-ed soon,
+ * returnining > 0 will keep timer active.
+ */
 void
 void
-tcp_slowtimo(void)
+tcp_gc(struct inpcbinfo *ipi)
 {
        struct inpcb *inp, *nxt;
 {
        struct inpcb *inp, *nxt;
-       struct tcpcb *tp;
+       struct tcpcb *tw_tp, *tw_ntp;
 #if TCPDEBUG
        int ostate;
 #endif
 #if TCPDEBUG
        int ostate;
 #endif
-
 #if  KDEBUG
        static int tws_checked = 0;
 #endif
 
 #if  KDEBUG
        static int tws_checked = 0;
 #endif
 
-       struct inpcbinfo *pcbinfo               = &tcbinfo;
+       KERNEL_DEBUG(DBG_FNC_TCP_SLOW | DBG_FUNC_START, 0, 0, 0, 0, 0);
 
 
-       KERNEL_DEBUG(DBG_FNC_TCP_SLOW | DBG_FUNC_START, 0,0,0,0,0);
-
-       tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
-
-       /* Update tcp_now here as it may get used while processing the slow timer */
+       /*
+        * Update tcp_now here as it may get used while
+        * processing the slow timer.
+        */
        calculate_tcp_clock();
 
        calculate_tcp_clock();
 
-       /* Garbage collect socket/tcpcb: We need to acquire the list lock 
+       /*
+        * Garbage collect socket/tcpcb: We need to acquire the list lock
         * exclusively to do this
         */
 
         * exclusively to do this
         */
 
-       if (lck_rw_try_lock_exclusive(pcbinfo->mtx) == FALSE) {
-               if (tcp_gc_done == TRUE) {      /* don't sweat it this time. cleanup was done last time */
+       if (lck_rw_try_lock_exclusive(ipi->ipi_lock) == FALSE) {
+               /* don't sweat it this time; cleanup was done last time */
+               if (tcp_gc_done == TRUE) {
                        tcp_gc_done = FALSE;
                        tcp_gc_done = FALSE;
-                       KERNEL_DEBUG(DBG_FNC_TCP_SLOW | DBG_FUNC_END, tws_checked, cur_tw_slot,0,0,0);
-                       return; /* Upgrade failed and lost lock - give up this time. */
+                       KERNEL_DEBUG(DBG_FNC_TCP_SLOW | DBG_FUNC_END,
+                           tws_checked, cur_tw_slot, 0, 0, 0);
+                       /* Lock upgrade failed, give up this round */
+                       atomic_add_32(&ipi->ipi_gc_req.intimer_fast, 1);
+                       return;
                }
                }
-               lck_rw_lock_exclusive(pcbinfo->mtx);    /* Upgrade failed, lost lock now take it again exclusive */
+               /* Upgrade failed, lost lock now take it again exclusive */
+               lck_rw_lock_exclusive(ipi->ipi_lock);
        }
        tcp_gc_done = TRUE;
 
        }
        tcp_gc_done = TRUE;
 
-       /*
-        * Process the items in the current time-wait slot
-        */
-#if  KDEBUG
-       tws_checked = 0;
-#endif
-       KERNEL_DEBUG(DBG_FNC_TCP_SLOW | DBG_FUNC_NONE, tws_checked,0,0,0,0);
-
-       LIST_FOREACH(inp, &time_wait_slots[cur_tw_slot], inp_list) {
-#if KDEBUG
-               tws_checked++;
-#endif
-
-               if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) 
-                       continue;
-
-               tcp_lock(inp->inp_socket, 1, 0);
-
-               if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) 
-                       goto twunlock;
-
-               tp = intotcpcb(inp);
-               if (tp == NULL)  /* tp already closed, remove from list */
-                       goto twunlock;
-
-               if (tp->t_timer[TCPT_2MSL] >= N_TIME_WAIT_SLOTS) {
-                   tp->t_timer[TCPT_2MSL] -= N_TIME_WAIT_SLOTS;
-                   tp->t_rcvtime += N_TIME_WAIT_SLOTS;
+       LIST_FOREACH_SAFE(inp, &tcb, inp_list, nxt) {
+               if (tcp_garbage_collect(inp, 0)) {
+                       atomic_add_32(&ipi->ipi_gc_req.intimer_fast, 1);
                }
                }
-               else
-                   tp->t_timer[TCPT_2MSL] = 0;
-
-               if (tp->t_timer[TCPT_2MSL] == 0)  {
+       }
 
 
-                       /* That pcb is ready for a close */     
-                       tcp_free_sackholes(tp);
-                       tp = tcp_close(tp);
+       /* Now cleanup the time wait ones */
+       TAILQ_FOREACH_SAFE(tw_tp, &tcp_tw_tailq, t_twentry, tw_ntp) {
+               /*
+                * We check the timestamp here without holding the
+                * socket lock for better performance. If there are
+                * any pcbs in time-wait, the timer will get rescheduled.
+                * Hence some error in this check can be tolerated.
+                *
+                * Sometimes a socket on time-wait queue can be closed if
+                * 2MSL timer expired but the application still has a
+                * usecount on it.
+                */
+               if (tw_tp->t_state == TCPS_CLOSED ||
+                   TSTMP_GEQ(tcp_now, tw_tp->t_timer[TCPT_2MSL])) {
+                       if (tcp_garbage_collect(tw_tp->t_inpcb, 1)) {
+                               atomic_add_32(&ipi->ipi_gc_req.intimer_lazy, 1);
+                       }
                }
                }
-twunlock:
-               tcp_unlock(inp->inp_socket, 1, 0);
        }
 
        }
 
+       /* take into account pcbs that are still in time_wait_slots */
+       atomic_add_32(&ipi->ipi_gc_req.intimer_lazy, ipi->ipi_twcount);
 
 
-       LIST_FOREACH_SAFE(inp, &tcb, inp_list, nxt) {
-               tcp_garbage_collect(inp, 0);
-       }
+       lck_rw_done(ipi->ipi_lock);
 
 
-       /* Now cleanup the time wait ones */
-       LIST_FOREACH_SAFE(inp, &time_wait_slots[cur_tw_slot], inp_list, nxt) {
-               tcp_garbage_collect(inp, 1);
+       /* Clean up the socache while we are here */
+       if (so_cache_timer()) {
+               atomic_add_32(&ipi->ipi_gc_req.intimer_lazy, 1);
        }
 
        }
 
-       if (++cur_tw_slot >= N_TIME_WAIT_SLOTS)
-               cur_tw_slot = 0;
-       
-       lck_rw_done(pcbinfo->mtx);
-       KERNEL_DEBUG(DBG_FNC_TCP_SLOW | DBG_FUNC_END, tws_checked, cur_tw_slot,0,0,0);
+       KERNEL_DEBUG(DBG_FNC_TCP_SLOW | DBG_FUNC_END, tws_checked,
+           cur_tw_slot, 0, 0, 0);
+
+       return;
 }
 
 /*
  * Cancel all timers for TCP tp.
  */
 void
 }
 
 /*
  * Cancel all timers for TCP tp.
  */
 void
-tcp_canceltimers(tp)
-       struct tcpcb *tp;
+tcp_canceltimers(struct tcpcb *tp)
 {
 {
-       register int i;
+       int i;
 
        tcp_remove_timer(tp);
 
        tcp_remove_timer(tp);
-       for (i = 0; i < TCPT_NTIMERS; i++)
+       for (i = 0; i < TCPT_NTIMERS; i++) {
                tp->t_timer[i] = 0;
                tp->t_timer[i] = 0;
+       }
        tp->tentry.timer_start = tcp_now;
        tp->tentry.index = TCPT_NONE;
 }
 
        tp->tentry.timer_start = tcp_now;
        tp->tentry.index = TCPT_NONE;
 }
 
-int    tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
-    { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
+int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
+{ 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
+
+int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
+{ 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
+
+static int tcp_totbackoff = 511;        /* sum of tcp_backoff[] */
+
+void
+tcp_rexmt_save_state(struct tcpcb *tp)
+{
+       u_int32_t fsize;
+       if (TSTMP_SUPPORTED(tp)) {
+               /*
+                * Since timestamps are supported on the connection,
+                * we can do recovery as described in rfc 4015.
+                */
+               fsize = tp->snd_max - tp->snd_una;
+               tp->snd_ssthresh_prev = max(fsize, tp->snd_ssthresh);
+               tp->snd_recover_prev = tp->snd_recover;
+       } else {
+               /*
+                * Timestamp option is not supported on this connection.
+                * Record ssthresh and cwnd so they can
+                * be recovered if this turns out to be a "bad" retransmit.
+                * A retransmit is considered "bad" if an ACK for this
+                * segment is received within RTT/2 interval; the assumption
+                * here is that the ACK was already in flight.  See
+                * "On Estimating End-to-End Network Path Properties" by
+                * Allman and Paxson for more details.
+                */
+               tp->snd_cwnd_prev = tp->snd_cwnd;
+               tp->snd_ssthresh_prev = tp->snd_ssthresh;
+               tp->snd_recover_prev = tp->snd_recover;
+               if (IN_FASTRECOVERY(tp)) {
+                       tp->t_flags |= TF_WASFRECOVERY;
+               } else {
+                       tp->t_flags &= ~TF_WASFRECOVERY;
+               }
+       }
+       tp->t_srtt_prev = (tp->t_srtt >> TCP_RTT_SHIFT) + 2;
+       tp->t_rttvar_prev = (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
+       tp->t_flagsext &= ~(TF_RECOMPUTE_RTT);
+}
+
+/*
+ * Revert to the older segment size if there is an indication that PMTU
+ * blackhole detection was not needed.
+ */
+void
+tcp_pmtud_revert_segment_size(struct tcpcb *tp)
+{
+       int32_t optlen;
+
+       VERIFY(tp->t_pmtud_saved_maxopd > 0);
+       tp->t_flags |= TF_PMTUD;
+       tp->t_flags &= ~TF_BLACKHOLE;
+       optlen = tp->t_maxopd - tp->t_maxseg;
+       tp->t_maxopd = tp->t_pmtud_saved_maxopd;
+       tp->t_maxseg = tp->t_maxopd - optlen;
 
 
-int    tcp_backoff[TCP_MAXRXTSHIFT + 1] =
-    { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
+       /*
+        * Reset the slow-start flight size as it
+        * may depend on the new MSS
+        */
+       if (CC_ALGO(tp)->cwnd_init != NULL) {
+               CC_ALGO(tp)->cwnd_init(tp);
+       }
+       tp->t_pmtud_start_ts = 0;
+       tcpstat.tcps_pmtudbh_reverted++;
 
 
-static int tcp_totbackoff = 511;       /* sum of tcp_backoff[] */
+       /* change MSS according to recommendation, if there was one */
+       tcp_update_mss_locked(tp->t_inpcb->inp_socket, NULL);
+}
 
 /*
  * TCP timer processing.
  */
 struct tcpcb *
 
 /*
  * TCP timer processing.
  */
 struct tcpcb *
-tcp_timers(tp, timer)
-       register struct tcpcb *tp;
-       int timer;
+tcp_timers(struct tcpcb *tp, int timer)
 {
 {
-       register int rexmt;
-       struct socket *so_tmp;
+       int32_t rexmt, optlen = 0, idle_time = 0;
+       struct socket *so;
        struct tcptemp *t_template;
        struct tcptemp *t_template;
-       int optlen = 0;
-       int idle_time = 0;
-
 #if TCPDEBUG
        int ostate;
 #endif
 #if TCPDEBUG
        int ostate;
 #endif
@@ -540,12 +887,13 @@ tcp_timers(tp, timer)
 #if INET6
        int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV4) == 0;
 #endif /* INET6 */
 #if INET6
        int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV4) == 0;
 #endif /* INET6 */
+       u_int64_t accsleep_ms;
+       u_int32_t last_sleep_ms = 0;
 
 
-       so_tmp = tp->t_inpcb->inp_socket;
+       so = tp->t_inpcb->inp_socket;
        idle_time = tcp_now - tp->t_rcvtime;
 
        switch (timer) {
        idle_time = tcp_now - tp->t_rcvtime;
 
        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
        /*
         * 2 MSL timeout in shutdown went off.  If we're closed but
         * still waiting for peer to close and connection has been idle
@@ -558,12 +906,12 @@ tcp_timers(tp, timer)
                tcp_free_sackholes(tp);
                if (tp->t_state != TCPS_TIME_WAIT &&
                    tp->t_state != TCPS_FIN_WAIT_2 &&
                tcp_free_sackholes(tp);
                if (tp->t_state != TCPS_TIME_WAIT &&
                    tp->t_state != TCPS_FIN_WAIT_2 &&
-                   ((idle_time > 0) && (idle_time < tcp_maxidle))) {
-                       tp->t_timer[TCPT_2MSL] = OFFSET_FROM_START(tp, (u_int32_t)tcp_keepintvl);
-               }
-               else {
+                   ((idle_time > 0) && (idle_time < TCP_CONN_MAXIDLE(tp)))) {
+                       tp->t_timer[TCPT_2MSL] = OFFSET_FROM_START(tp,
+                           (u_int32_t)TCP_CONN_KEEPINTVL(tp));
+               else {
                        tp = tcp_close(tp);
                        tp = tcp_close(tp);
-                       return(tp);
+                       return tp;
                }
                break;
 
                }
                break;
 
@@ -573,133 +921,276 @@ tcp_timers(tp, timer)
         * to a longer retransmit interval and retransmit one segment.
         */
        case TCPT_REXMT:
         * to a longer retransmit interval and retransmit one segment.
         */
        case TCPT_REXMT:
-               tcp_free_sackholes(tp);
-               /* Drop a connection in the retransmit timer
-                * 1. If we have retransmitted more than TCP_MAXRXTSHIFT times
-                * 2. If the time spent in this retransmission episode is more than
-                *    the time limit set with TCP_RXT_CONNDROPTIME socket option
-                * 3. If TCP_RXT_FINDROP socket option was set and we have already
-                *    retransmitted the FIN 3 times without receiving an ack
+               absolutetime_to_nanoseconds(mach_absolutetime_asleep,
+                   &accsleep_ms);
+               accsleep_ms = accsleep_ms / 1000000UL;
+               if (accsleep_ms > tp->t_accsleep_ms) {
+                       last_sleep_ms = accsleep_ms - tp->t_accsleep_ms;
+               }
+               /*
+                * Drop a connection in the retransmit timer
+                * 1. If we have retransmitted more than TCP_MAXRXTSHIFT
+                * times
+                * 2. If the time spent in this retransmission episode is
+                * more than the time limit set with TCP_RXT_CONNDROPTIME
+                * socket option
+                * 3. If TCP_RXT_FINDROP socket option was set and
+                * we have already retransmitted the FIN 3 times without
+                * receiving an ack
                 */
                if (++tp->t_rxtshift > TCP_MAXRXTSHIFT ||
                 */
                if (++tp->t_rxtshift > TCP_MAXRXTSHIFT ||
-                       (tp->rxt_conndroptime > 0 && tp->rxt_start > 0 && 
-                       (tcp_now - tp->rxt_start) >= tp->rxt_conndroptime) ||
-                       ((tp->t_flagsext & TF_RXTFINDROP) != 0 &&
-                       (tp->t_flags & TF_SENTFIN) != 0 &&
-                       tp->t_rxtshift >= 4)) {
-
+                   (tp->t_rxt_conndroptime > 0 && tp->t_rxtstart > 0 &&
+                   (tcp_now - tp->t_rxtstart) >= tp->t_rxt_conndroptime) ||
+                   ((tp->t_flagsext & TF_RXTFINDROP) != 0 &&
+                   (tp->t_flags & TF_SENTFIN) != 0 && tp->t_rxtshift >= 4) ||
+                   (tp->t_rxtshift > 4 && last_sleep_ms >= TCP_SLEEP_TOO_LONG)) {
+                       if (tp->t_state == TCPS_ESTABLISHED &&
+                           tp->t_rxt_minimum_timeout > 0) {
+                               /*
+                                * Avoid dropping a connection if minimum
+                                * timeout is set and that time did not
+                                * pass. We will retry sending
+                                * retransmissions at the maximum interval
+                                */
+                               if (TSTMP_LT(tcp_now, (tp->t_rxtstart +
+                                   tp->t_rxt_minimum_timeout))) {
+                                       tp->t_rxtshift = TCP_MAXRXTSHIFT - 1;
+                                       goto retransmit_packet;
+                               }
+                       }
                        if ((tp->t_flagsext & TF_RXTFINDROP) != 0) {
                                tcpstat.tcps_rxtfindrop++;
                        if ((tp->t_flagsext & TF_RXTFINDROP) != 0) {
                                tcpstat.tcps_rxtfindrop++;
+                       } else if (last_sleep_ms >= TCP_SLEEP_TOO_LONG) {
+                               tcpstat.tcps_drop_after_sleep++;
                        } else {
                                tcpstat.tcps_timeoutdrop++;
                        }
                        } else {
                                tcpstat.tcps_timeoutdrop++;
                        }
+                       if (tp->t_rxtshift >= TCP_MAXRXTSHIFT) {
+                               if (TCP_ECN_ENABLED(tp)) {
+                                       INP_INC_IFNET_STAT(tp->t_inpcb,
+                                           ecn_on.rxmit_drop);
+                               } else {
+                                       INP_INC_IFNET_STAT(tp->t_inpcb,
+                                           ecn_off.rxmit_drop);
+                               }
+                       }
                        tp->t_rxtshift = TCP_MAXRXTSHIFT;
                        tp->t_rxtshift = TCP_MAXRXTSHIFT;
+                       postevent(so, 0, EV_TIMEOUT);
+                       soevent(so,
+                           (SO_FILT_HINT_LOCKED | SO_FILT_HINT_TIMEOUT));
+
+                       if (TCP_ECN_ENABLED(tp) &&
+                           tp->t_state == TCPS_ESTABLISHED) {
+                               tcp_heuristic_ecn_droprxmt(tp);
+                       }
+
                        tp = tcp_drop(tp, tp->t_softerror ?
                            tp->t_softerror : ETIMEDOUT);
                        tp = tcp_drop(tp, tp->t_softerror ?
                            tp->t_softerror : ETIMEDOUT);
-                       postevent(so_tmp, 0, EV_TIMEOUT);                       
+
                        break;
                }
                        break;
                }
+retransmit_packet:
+               tcpstat.tcps_rexmttimeo++;
+               tp->t_accsleep_ms = accsleep_ms;
+
+               if (tp->t_rxtshift == 1 &&
+                   tp->t_state == TCPS_ESTABLISHED) {
+                       /* Set the time at which retransmission started. */
+                       tp->t_rxtstart = tcp_now;
+
+                       /*
+                        * if this is the first retransmit timeout, save
+                        * the state so that we can recover if the timeout
+                        * is spurious.
+                        */
+                       tcp_rexmt_save_state(tp);
+               }
+#if MPTCP
+               if ((tp->t_rxtshift >= mptcp_fail_thresh) &&
+                   (tp->t_state == TCPS_ESTABLISHED) &&
+                   (tp->t_mpflags & TMPF_MPTCP_TRUE)) {
+                       mptcp_act_on_txfail(so);
+               }
+
+               if (so->so_flags & SOF_MP_SUBFLOW) {
+                       struct mptses *mpte = tptomptp(tp)->mpt_mpte;
+
+                       mptcp_check_subflows_and_add(mpte);
+               }
+#endif /* MPTCP */
+
+               if (tp->t_adaptive_wtimo > 0 &&
+                   tp->t_rxtshift > tp->t_adaptive_wtimo &&
+                   TCPS_HAVEESTABLISHED(tp->t_state)) {
+                       /* Send an event to the application */
+                       soevent(so,
+                           (SO_FILT_HINT_LOCKED |
+                           SO_FILT_HINT_ADAPTIVE_WTIMO));
+               }
+
+               /*
+                * If this is a retransmit timeout after PTO, the PTO
+                * was not effective
+                */
+               if (tp->t_flagsext & TF_SENT_TLPROBE) {
+                       tp->t_flagsext &= ~(TF_SENT_TLPROBE);
+                       tcpstat.tcps_rto_after_pto++;
+               }
 
 
-               if (tp->t_rxtshift == 1) {
+               if (tp->t_flagsext & TF_DELAY_RECOVERY) {
                        /*
                        /*
-                        * first retransmit; record ssthresh and cwnd so they can
-                        * be recovered if this turns out to be a "bad" retransmit.
-                        * A retransmit is considered "bad" if an ACK for this 
-                        * segment is received within RTT/2 interval; the assumption
-                        * here is that the ACK was already in flight.  See 
-                        * "On Estimating End-to-End Network Path Properties" by
-                        * Allman and Paxson for more details.
+                        * Retransmit timer fired before entering recovery
+                        * on a connection with packet re-ordering. This
+                        * suggests that the reordering metrics computed
+                        * are not accurate.
                         */
                         */
-                       tp->snd_cwnd_prev = tp->snd_cwnd;
-                       tp->snd_ssthresh_prev = tp->snd_ssthresh;
-                       tp->snd_recover_prev = tp->snd_recover;
-                       if (IN_FASTRECOVERY(tp))
-                                 tp->t_flags |= TF_WASFRECOVERY;
-                       else
-                                 tp->t_flags &= ~TF_WASFRECOVERY;
-                       tp->t_badrxtwin = tcp_now  + (tp->t_srtt >> (TCP_RTT_SHIFT)); 
-
-                       /* Set the time at which retransmission on this 
-                        * connection started
+                       tp->t_reorderwin = 0;
+                       tp->t_timer[TCPT_DELAYFR] = 0;
+                       tp->t_flagsext &= ~(TF_DELAY_RECOVERY);
+               }
+
+               if (tp->t_state == TCPS_SYN_RECEIVED) {
+                       tcp_disable_tfo(tp);
+               }
+
+               if (!(tp->t_tfo_flags & TFO_F_HEURISTIC_DONE) &&
+                   (tp->t_tfo_stats & TFO_S_SYN_DATA_SENT) &&
+                   !(tp->t_tfo_flags & TFO_F_NO_SNDPROBING) &&
+                   ((tp->t_state != TCPS_SYN_SENT && tp->t_rxtshift > 1) ||
+                   tp->t_rxtshift > 4)) {
+                       /*
+                        * For regular retransmissions, a first one is being
+                        * done for tail-loss probe.
+                        * Thus, if rxtshift > 1, this means we have sent the segment
+                        * a total of 3 times.
+                        *
+                        * If we are in SYN-SENT state, then there is no tail-loss
+                        * probe thus we have to let rxtshift go up to 3.
                         */
                         */
-                       tp->rxt_start = tcp_now;
+                       tcp_heuristic_tfo_middlebox(tp);
+
+                       so->so_error = ENODATA;
+                       sorwakeup(so);
+                       sowwakeup(so);
+
+                       tp->t_tfo_stats |= TFO_S_SEND_BLACKHOLE;
+                       tcpstat.tcps_tfo_sndblackhole++;
                }
                }
-               tcpstat.tcps_rexmttimeo++;
-               if (tp->t_state == TCPS_SYN_SENT)
+
+               if (!(tp->t_tfo_flags & TFO_F_HEURISTIC_DONE) &&
+                   (tp->t_tfo_stats & TFO_S_SYN_DATA_ACKED) &&
+                   tp->t_rxtshift > 3) {
+                       if (TSTMP_GT(tp->t_sndtime - 10 * TCP_RETRANSHZ, tp->t_rcvtime)) {
+                               tcp_heuristic_tfo_middlebox(tp);
+
+                               so->so_error = ENODATA;
+                               sorwakeup(so);
+                               sowwakeup(so);
+                       }
+               }
+
+               if (tp->t_state == TCPS_SYN_SENT) {
                        rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
                        rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
-               else
+                       tp->t_stat.synrxtshift = tp->t_rxtshift;
+
+                       /* When retransmitting, disable TFO */
+                       if (tfo_enabled(tp) &&
+                           (!(so->so_flags1 & SOF1_DATA_AUTHENTICATED) ||
+                           (tp->t_flagsext & TF_FASTOPEN_HEUR))) {
+                               tp->t_flagsext &= ~TF_FASTOPEN;
+                               tp->t_tfo_flags |= TFO_F_SYN_LOSS;
+                       }
+               } else {
                        rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
                        rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
-               TCPT_RANGESET(tp->t_rxtcur, rexmt,
-                       tp->t_rttmin, TCPTV_REXMTMAX, 
-                       TCP_ADD_REXMTSLOP(tp));
+               }
+
+               TCPT_RANGESET(tp->t_rxtcur, rexmt, tp->t_rttmin, TCPTV_REXMTMAX,
+                   TCP_ADD_REXMTSLOP(tp));
                tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp, tp->t_rxtcur);
 
                tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp, tp->t_rxtcur);
 
+               if (INP_WAIT_FOR_IF_FEEDBACK(tp->t_inpcb)) {
+                       goto fc_output;
+               }
+
+               tcp_free_sackholes(tp);
                /*
                /*
-                * Check for potential Path MTU Discovery Black Hole 
+                * Check for potential Path MTU Discovery Black Hole
                 */
                 */
-
-               if (tcp_pmtud_black_hole_detect && (tp->t_state == TCPS_ESTABLISHED)) {
-                       if (((tp->t_flags & (TF_PMTUD|TF_MAXSEGSNT)) == (TF_PMTUD|TF_MAXSEGSNT)) && (tp->t_rxtshift == 2)) {
-                               /* 
+               if (tcp_pmtud_black_hole_detect &&
+                   !(tp->t_flagsext & TF_NOBLACKHOLE_DETECTION) &&
+                   (tp->t_state == TCPS_ESTABLISHED)) {
+                       if ((tp->t_flags & TF_PMTUD) &&
+                           ((tp->t_flags & TF_MAXSEGSNT)
+                           || tp->t_pmtud_lastseg_size > tcp_pmtud_black_hole_mss) &&
+                           tp->t_rxtshift == 2) {
+                               /*
                                 * Enter Path MTU Black-hole Detection mechanism:
                                 * - Disable Path MTU Discovery (IP "DF" bit).
                                 * Enter Path MTU Black-hole Detection mechanism:
                                 * - Disable Path MTU Discovery (IP "DF" bit).
-                                * - Reduce MTU to lower value than what we negociated with peer.
+                                * - Reduce MTU to lower value than what we
+                                * negotiated with the peer.
                                 */
                                 */
-
-                               tp->t_flags &= ~TF_PMTUD; /* Disable Path MTU Discovery for now */
-                               tp->t_flags |= TF_BLACKHOLE; /* Record that we may have found a black hole */
+                               /* Disable Path MTU Discovery for now */
+                               tp->t_flags &= ~TF_PMTUD;
+                               /* Record that we may have found a black hole */
+                               tp->t_flags |= TF_BLACKHOLE;
                                optlen = tp->t_maxopd - tp->t_maxseg;
                                optlen = tp->t_maxopd - tp->t_maxseg;
-                               tp->t_pmtud_saved_maxopd = tp->t_maxopd; /* Keep track of previous MSS */
-                               if (tp->t_maxopd > tcp_pmtud_black_hole_mss)
-                                       tp->t_maxopd = tcp_pmtud_black_hole_mss; /* Reduce the MSS to intermediary value */
-                               else {
-                                       tp->t_maxopd =  /* use the default MSS */
+                               /* Keep track of previous MSS */
+                               tp->t_pmtud_saved_maxopd = tp->t_maxopd;
+                               tp->t_pmtud_start_ts = tcp_now;
+                               if (tp->t_pmtud_start_ts == 0) {
+                                       tp->t_pmtud_start_ts++;
+                               }
+                               /* Reduce the MSS to intermediary value */
+                               if (tp->t_maxopd > tcp_pmtud_black_hole_mss) {
+                                       tp->t_maxopd = tcp_pmtud_black_hole_mss;
+                               } else {
+                                       tp->t_maxopd = /* use the default MSS */
 #if INET6
 #if INET6
-                                               isipv6 ? tcp_v6mssdflt :
+                                           isipv6 ? tcp_v6mssdflt :
 #endif /* INET6 */
 #endif /* INET6 */
-                                                       tcp_mssdflt;
+                                           tcp_mssdflt;
                                }
                                tp->t_maxseg = tp->t_maxopd - optlen;
 
                                /*
                                }
                                tp->t_maxseg = tp->t_maxopd - optlen;
 
                                /*
-                                * Reset the slow-start flight size as it may depends on the new MSS
-                                */
-                               if (CC_ALGO(tp)->cwnd_init != NULL)
+                                * Reset the slow-start flight size
+                                * as it may depend on the new MSS
+                                */
+                               if (CC_ALGO(tp)->cwnd_init != NULL) {
                                        CC_ALGO(tp)->cwnd_init(tp);
                                        CC_ALGO(tp)->cwnd_init(tp);
+                               }
+                               tp->snd_cwnd = tp->t_maxseg;
                        }
                        /*
                        }
                        /*
-                        * If further retransmissions are still unsuccessful with a lowered MTU,
-                        * maybe this isn't a Black Hole and we restore the previous MSS and
-                        * blackhole detection flags.
+                        * If further retransmissions are still
+                        * unsuccessful with a lowered MTU, maybe this
+                        * isn't a Black Hole and we restore the previous
+                        * MSS and blackhole detection flags.
                         */
                        else {
                         */
                        else {
-       
-                               if ((tp->t_flags & TF_BLACKHOLE) && (tp->t_rxtshift > 4)) {
-                                       tp->t_flags |= TF_PMTUD; 
-                                       tp->t_flags &= ~TF_BLACKHOLE; 
-                                       optlen = tp->t_maxopd - tp->t_maxseg;
-                                       tp->t_maxopd = tp->t_pmtud_saved_maxopd;
-                                       tp->t_maxseg = tp->t_maxopd - optlen;
-                                       /*
-                                       * Reset the slow-start flight size as it may depends on the new MSS
-                                       */
-                                       if (CC_ALGO(tp)->cwnd_init != NULL)
-                                               CC_ALGO(tp)->cwnd_init(tp);
+                               if ((tp->t_flags & TF_BLACKHOLE) &&
+                                   (tp->t_rxtshift > 4)) {
+                                       tcp_pmtud_revert_segment_size(tp);
+                                       tp->snd_cwnd = tp->t_maxseg;
                                }
                        }
                }
 
 
                /*
                                }
                        }
                }
 
 
                /*
-                * Disable rfc1323 and rfc1644 if we haven't got any response to
-                * our SYN (after we reach the threshold) to work-around some
-                * broken terminal servers (most of which have hopefully been
-                * retired) that have bad VJ header compression code which
-                * trashes TCP segments containing unknown-to-them TCP options.
+                * Disable rfc1323 and rfc1644 if we haven't got any
+                * response to our SYN (after we reach the threshold)
+                * to work-around some broken terminal servers (most of
+                * which have hopefully been retired) that have bad VJ
+                * header compression code which trashes TCP segments
+                * containing unknown-to-them TCP options.
+                * Do this only on non-local connections.
                 */
                 */
-               if ((tp->t_state == TCPS_SYN_SENT) &&
-                   (tp->t_rxtshift == tcp_broken_peer_syn_rxmit_thres))
-                       tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
+               if (tp->t_state == TCPS_SYN_SENT &&
+                   tp->t_rxtshift == tcp_broken_peer_syn_rxmit_thres) {
+                       tp->t_flags &= ~(TF_REQ_SCALE | TF_REQ_TSTMP | TF_REQ_CC);
+               }
+
                /*
                 * If losing, let the lower level know and try for
                 * a better route.  Also, if we backed off this far,
                /*
                 * If losing, let the lower level know and try for
                 * a better route.  Also, if we backed off this far,
@@ -710,9 +1201,9 @@ tcp_timers(tp, timer)
                 */
                if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 #if INET6
                 */
                if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
 #if INET6
-                       if (isipv6)
+                       if (isipv6) {
                                in6_losing(tp->t_inpcb);
                                in6_losing(tp->t_inpcb);
-                       else
+                       else
 #endif /* INET6 */
                        in_losing(tp->t_inpcb);
                        tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 #endif /* INET6 */
                        in_losing(tp->t_inpcb);
                        tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
@@ -728,20 +1219,42 @@ tcp_timers(tp, timer)
                 * Force a segment to be sent.
                 */
                tp->t_flags |= TF_ACKNOW;
                 * Force a segment to be sent.
                 */
                tp->t_flags |= TF_ACKNOW;
-               /*
-                * If timing a segment in this window, stop the timer.
-                */
+
+               /* If timing a segment in this window, stop the timer */
                tp->t_rtttime = 0;
 
                tp->t_rtttime = 0;
 
-               if (CC_ALGO(tp)->after_timeout != NULL)
+               if (!IN_FASTRECOVERY(tp) && tp->t_rxtshift == 1) {
+                       tcpstat.tcps_tailloss_rto++;
+               }
+
+
+               /*
+                * RFC 5681 says: when a TCP sender detects segment loss
+                * using retransmit timer and the given segment has already
+                * been retransmitted by way of the retransmission timer at
+                * least once, the value of ssthresh is held constant
+                */
+               if (tp->t_rxtshift == 1 &&
+                   CC_ALGO(tp)->after_timeout != NULL) {
                        CC_ALGO(tp)->after_timeout(tp);
                        CC_ALGO(tp)->after_timeout(tp);
+                       /*
+                        * CWR notifications are to be sent on new data
+                        * right after Fast Retransmits and ECE
+                        * notification receipts.
+                        */
+                       if (TCP_ECN_ENABLED(tp)) {
+                               tp->ecn_flags |= TE_SENDCWR;
+                       }
+               }
 
 
-               tp->t_dupacks = 0;
                EXIT_FASTRECOVERY(tp);
 
                EXIT_FASTRECOVERY(tp);
 
-               DTRACE_TCP5(cc, void, NULL, struct inpcb *, tp->t_inpcb,
-                       struct tcpcb *, tp, struct tcphdr *, NULL,
-                       int32_t, TCP_CC_REXMT_TIMEOUT);
+               /* Exit cwnd non validated phase */
+               tp->t_flagsext &= ~TF_CWND_NONVALIDATED;
+
+
+fc_output:
+               tcp_ccdbg_trace(tp, NULL, TCP_CC_REXMT_TIMEOUT);
 
                (void) tcp_output(tp);
                break;
 
                (void) tcp_output(tp);
                break;
@@ -758,25 +1271,27 @@ tcp_timers(tp, timer)
                 * backoff, drop the connection if the idle time
                 * (no responses to probes) reaches the maximum
                 * backoff that we would use if retransmitting.
                 * backoff, drop the connection if the idle time
                 * (no responses to probes) reaches the maximum
                 * backoff that we would use if retransmitting.
-                * 
-                * Drop the connection if we reached the maximum allowed time for 
-                * Zero Window Probes without a non-zero update from the peer. 
+                *
+                * Drop the connection if we reached the maximum allowed time for
+                * Zero Window Probes without a non-zero update from the peer.
                 * See rdar://5805356
                 */
                if ((tp->t_rxtshift == TCP_MAXRXTSHIFT &&
                    (idle_time >= tcp_maxpersistidle ||
                 * See rdar://5805356
                 */
                if ((tp->t_rxtshift == TCP_MAXRXTSHIFT &&
                    (idle_time >= tcp_maxpersistidle ||
-                   idle_time >= TCP_REXMTVAL(tp) * tcp_totbackoff)) || 
-                   ((tp->t_persist_stop != 0) && (tp->t_persist_stop <= tcp_now))) {
+                   idle_time >= TCP_REXMTVAL(tp) * tcp_totbackoff)) ||
+                   ((tp->t_persist_stop != 0) &&
+                   TSTMP_LEQ(tp->t_persist_stop, tcp_now))) {
                        tcpstat.tcps_persistdrop++;
                        tcpstat.tcps_persistdrop++;
-                       so_tmp = tp->t_inpcb->inp_socket;
+                       postevent(so, 0, EV_TIMEOUT);
+                       soevent(so,
+                           (SO_FILT_HINT_LOCKED | SO_FILT_HINT_TIMEOUT));
                        tp = tcp_drop(tp, ETIMEDOUT);
                        tp = tcp_drop(tp, ETIMEDOUT);
-                       postevent(so_tmp, 0, EV_TIMEOUT);
                        break;
                }
                tcp_setpersist(tp);
                        break;
                }
                tcp_setpersist(tp);
-               tp->t_force = 1;
+               tp->t_flagsext |= TF_FORCE;
                (void) tcp_output(tp);
                (void) tcp_output(tp);
-               tp->t_force = 0;
+               tp->t_flagsext &= ~TF_FORCE;
                break;
 
        /*
                break;
 
        /*
@@ -785,13 +1300,32 @@ tcp_timers(tp, timer)
         */
        case TCPT_KEEP:
                tcpstat.tcps_keeptimeo++;
         */
        case TCPT_KEEP:
                tcpstat.tcps_keeptimeo++;
-               if (tp->t_state < TCPS_ESTABLISHED)
+#if MPTCP
+               /*
+                * Regular TCP connections do not send keepalives after closing
+                * MPTCP must not also, after sending Data FINs.
+                */
+               struct mptcb *mp_tp = tptomptp(tp);
+               if ((tp->t_mpflags & TMPF_MPTCP_TRUE) &&
+                   (tp->t_state > TCPS_ESTABLISHED)) {
+                       goto dropit;
+               } else if (mp_tp != NULL) {
+                       if ((mptcp_ok_to_keepalive(mp_tp) == 0)) {
+                               goto dropit;
+                       }
+               }
+#endif /* MPTCP */
+               if (tp->t_state < TCPS_ESTABLISHED) {
                        goto dropit;
                        goto dropit;
+               }
                if ((always_keepalive ||
                if ((always_keepalive ||
-                   tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
+                   (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) ||
+                   (tp->t_flagsext & TF_DETECT_READSTALL) ||
+                   (tp->t_tfo_probe_state == TFO_PROBE_PROBING)) &&
                    (tp->t_state <= TCPS_CLOSING || tp->t_state == TCPS_FIN_WAIT_2)) {
                    (tp->t_state <= TCPS_CLOSING || tp->t_state == TCPS_FIN_WAIT_2)) {
-                       if (idle_time >= TCP_KEEPIDLE(tp) + (u_int32_t)tcp_maxidle)
+                       if (idle_time >= TCP_CONN_KEEPIDLE(tp) + TCP_CONN_MAXIDLE(tp)) {
                                goto dropit;
                                goto dropit;
+                       }
                        /*
                         * Send a packet designed to force a response
                         * if the peer is up and reachable:
                        /*
                         * Send a packet designed to force a response
                         * if the peer is up and reachable:
@@ -807,29 +1341,93 @@ tcp_timers(tp, timer)
                        tcpstat.tcps_keepprobe++;
                        t_template = tcp_maketemplate(tp);
                        if (t_template) {
                        tcpstat.tcps_keepprobe++;
                        t_template = tcp_maketemplate(tp);
                        if (t_template) {
-                               unsigned int ifscope, nocell = 0;
-
-                               if (tp->t_inpcb->inp_flags & INP_BOUND_IF)
-                                       ifscope = tp->t_inpcb->inp_boundif;
-                               else
-                                       ifscope = IFSCOPE_NONE;
-
-                               /*
-                                * If the socket isn't allowed to use the
-                                * cellular interface, indicate it as such.
-                                */
-                               if (tp->t_inpcb->inp_flags & INP_NO_IFT_CELLULAR)
-                                       nocell = 1;
-
+                               struct inpcb *inp = tp->t_inpcb;
+                               struct tcp_respond_args tra;
+
+                               bzero(&tra, sizeof(tra));
+                               tra.nocell = INP_NO_CELLULAR(inp);
+                               tra.noexpensive = INP_NO_EXPENSIVE(inp);
+                               tra.awdl_unrestricted = INP_AWDL_UNRESTRICTED(inp);
+                               tra.intcoproc_allowed = INP_INTCOPROC_ALLOWED(inp);
+                               if (tp->t_inpcb->inp_flags & INP_BOUND_IF) {
+                                       tra.ifscope = tp->t_inpcb->inp_boundifp->if_index;
+                               } else {
+                                       tra.ifscope = IFSCOPE_NONE;
+                               }
                                tcp_respond(tp, t_template->tt_ipgen,
                                    &t_template->tt_t, (struct mbuf *)NULL,
                                tcp_respond(tp, t_template->tt_ipgen,
                                    &t_template->tt_t, (struct mbuf *)NULL,
-                                   tp->rcv_nxt, tp->snd_una - 1, 0, ifscope,
-                                   nocell);
+                                   tp->rcv_nxt, tp->snd_una - 1, 0, &tra);
                                (void) m_free(dtom(t_template));
                                (void) m_free(dtom(t_template));
+                               if (tp->t_flagsext & TF_DETECT_READSTALL) {
+                                       tp->t_rtimo_probes++;
+                               }
                        }
                        }
-                       tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, tcp_keepintvl);
-               } else
-                       tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, TCP_KEEPIDLE(tp));
+                       tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp,
+                           TCP_CONN_KEEPINTVL(tp));
+               } else {
+                       tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp,
+                           TCP_CONN_KEEPIDLE(tp));
+               }
+               if (tp->t_flagsext & TF_DETECT_READSTALL) {
+                       struct ifnet *outifp = tp->t_inpcb->inp_last_outifp;
+                       bool reenable_probe = false;
+                       /*
+                        * The keep alive packets sent to detect a read
+                        * stall did not get a response from the
+                        * peer. Generate more keep-alives to confirm this.
+                        * If the number of probes sent reaches the limit,
+                        * generate an event.
+                        */
+                       if (tp->t_adaptive_rtimo > 0) {
+                               if (tp->t_rtimo_probes > tp->t_adaptive_rtimo) {
+                                       /* Generate an event */
+                                       soevent(so,
+                                           (SO_FILT_HINT_LOCKED |
+                                           SO_FILT_HINT_ADAPTIVE_RTIMO));
+                                       tcp_keepalive_reset(tp);
+                               } else {
+                                       reenable_probe = true;
+                               }
+                       } else if (outifp != NULL &&
+                           (outifp->if_eflags & IFEF_PROBE_CONNECTIVITY) &&
+                           tp->t_rtimo_probes <= TCP_CONNECTIVITY_PROBES_MAX) {
+                               reenable_probe = true;
+                       } else {
+                               tp->t_flagsext &= ~TF_DETECT_READSTALL;
+                       }
+                       if (reenable_probe) {
+                               int ind = min(tp->t_rtimo_probes,
+                                   TCP_MAXRXTSHIFT);
+                               tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(
+                                       tp, tcp_backoff[ind] * TCP_REXMTVAL(tp));
+                       }
+               }
+               if (tp->t_tfo_probe_state == TFO_PROBE_PROBING) {
+                       int ind;
+
+                       tp->t_tfo_probes++;
+                       ind = min(tp->t_tfo_probes, TCP_MAXRXTSHIFT);
+
+                       /*
+                        * We take the minimum among the time set by true
+                        * keepalive (see above) and the backoff'd RTO. That
+                        * way we backoff in case of packet-loss but will never
+                        * timeout slower than regular keepalive due to the
+                        * backing off.
+                        */
+                       tp->t_timer[TCPT_KEEP] = min(OFFSET_FROM_START(
+                                   tp, tcp_backoff[ind] * TCP_REXMTVAL(tp)),
+                           tp->t_timer[TCPT_KEEP]);
+               } else if (!(tp->t_tfo_flags & TFO_F_HEURISTIC_DONE) &&
+                   tp->t_tfo_probe_state == TFO_PROBE_WAIT_DATA) {
+                       /* Still no data! Let's assume a TFO-error and err out... */
+                       tcp_heuristic_tfo_middlebox(tp);
+
+                       so->so_error = ENODATA;
+                       sorwakeup(so);
+                       tp->t_tfo_stats |= TFO_S_RECV_BLACKHOLE;
+                       tcpstat.tcps_tfo_blackhole++;
+               }
                break;
        case TCPT_DELACK:
                if (tcp_delack_enabled && (tp->t_flags & TF_DELACK)) {
                break;
        case TCPT_DELACK:
                if (tcp_delack_enabled && (tp->t_flags & TF_DELACK)) {
@@ -837,29 +1435,186 @@ tcp_timers(tp, timer)
                        tp->t_timer[TCPT_DELACK] = 0;
                        tp->t_flags |= TF_ACKNOW;
 
                        tp->t_timer[TCPT_DELACK] = 0;
                        tp->t_flags |= TF_ACKNOW;
 
-                       /* If delayed ack timer fired while we are stretching acks, 
-                        * go back to acking every other packet
+                       /*
+                        * If delayed ack timer fired while stretching
+                        * acks, count the number of times the streaming
+                        * detection was not correct. If this exceeds a
+                        * threshold, disable strech ack on this
+                        * connection
+                        *
+                        * Also, go back to acking every other packet.
                         */
                         */
-                       if ((tp->t_flags & TF_STRETCHACK) != 0)
+                       if ((tp->t_flags & TF_STRETCHACK)) {
+                               if (tp->t_unacksegs > 1 &&
+                                   tp->t_unacksegs < maxseg_unacked) {
+                                       tp->t_stretchack_delayed++;
+                               }
+
+                               if (tp->t_stretchack_delayed >
+                                   TCP_STRETCHACK_DELAY_THRESHOLD) {
+                                       tp->t_flagsext |= TF_DISABLE_STRETCHACK;
+                                       /*
+                                        * Note the time at which stretch
+                                        * ack was disabled automatically
+                                        */
+                                       tp->rcv_nostrack_ts = tcp_now;
+                                       tcpstat.tcps_nostretchack++;
+                                       tp->t_stretchack_delayed = 0;
+                                       tp->rcv_nostrack_pkts = 0;
+                               }
                                tcp_reset_stretch_ack(tp);
                                tcp_reset_stretch_ack(tp);
+                       }
+
+                       /*
+                        * If we are measuring inter packet arrival jitter
+                        * for throttling a connection, this delayed ack
+                        * might be the reason for accumulating some
+                        * jitter. So let's restart the measurement.
+                        */
+                       CLEAR_IAJ_STATE(tp);
 
                        tcpstat.tcps_delack++;
                        (void) tcp_output(tp);
                }
                break;
 
 
                        tcpstat.tcps_delack++;
                        (void) tcp_output(tp);
                }
                break;
 
-#if TCPDEBUG
-       if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
-               tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
-                         PRU_SLOWTIMO);
-#endif
-       dropit:
+#if MPTCP
+       case TCPT_JACK_RXMT:
+               if ((tp->t_state == TCPS_ESTABLISHED) &&
+                   (tp->t_mpflags & TMPF_PREESTABLISHED) &&
+                   (tp->t_mpflags & TMPF_JOINED_FLOW)) {
+                       if (++tp->t_mprxtshift > TCP_MAXRXTSHIFT) {
+                               tcpstat.tcps_timeoutdrop++;
+                               postevent(so, 0, EV_TIMEOUT);
+                               soevent(so,
+                                   (SO_FILT_HINT_LOCKED |
+                                   SO_FILT_HINT_TIMEOUT));
+                               tp = tcp_drop(tp, tp->t_softerror ?
+                                   tp->t_softerror : ETIMEDOUT);
+                               break;
+                       }
+                       tcpstat.tcps_join_rxmts++;
+                       tp->t_mpflags |= TMPF_SND_JACK;
+                       tp->t_flags |= TF_ACKNOW;
+
+                       /*
+                        * No backoff is implemented for simplicity for this
+                        * corner case.
+                        */
+                       (void) tcp_output(tp);
+               }
+               break;
+#endif /* MPTCP */
+
+       case TCPT_PTO:
+       {
+               int32_t snd_len;
+               tp->t_flagsext &= ~(TF_SENT_TLPROBE);
+
+               /*
+                * Check if the connection is in the right state to
+                * send a probe
+                */
+               if (tp->t_state != TCPS_ESTABLISHED ||
+                   (tp->t_rxtshift > 0 && !(tp->t_flagsext & TF_PROBING)) ||
+                   tp->snd_max == tp->snd_una ||
+                   !SACK_ENABLED(tp) ||
+                   !TAILQ_EMPTY(&tp->snd_holes) ||
+                   IN_FASTRECOVERY(tp)) {
+                       break;
+               }
+
+               /*
+                * If there is no new data to send or if the
+                * connection is limited by receive window then
+                * retransmit the last segment, otherwise send
+                * new data.
+                */
+               snd_len = min(so->so_snd.sb_cc, tp->snd_wnd)
+                   - (tp->snd_max - tp->snd_una);
+               if (snd_len > 0) {
+                       tp->snd_nxt = tp->snd_max;
+               } else {
+                       snd_len = min((tp->snd_max - tp->snd_una),
+                           tp->t_maxseg);
+                       tp->snd_nxt = tp->snd_max - snd_len;
+               }
+
+               tcpstat.tcps_pto++;
+               if (tp->t_flagsext & TF_PROBING) {
+                       tcpstat.tcps_probe_if++;
+               }
+
+               /* If timing a segment in this window, stop the timer */
+               tp->t_rtttime = 0;
+               /* Note that tail loss probe is being sent */
+               tp->t_flagsext |= TF_SENT_TLPROBE;
+               tp->t_tlpstart = tcp_now;
+
+               tp->snd_cwnd += tp->t_maxseg;
+
+               /*
+                * When tail-loss-probe fires, we reset the RTO timer, because
+                * a probe just got sent, so we are good to push out the timer.
+                *
+                * Set to 0 to ensure that tcp_output() will reschedule it
+                */
+               tp->t_timer[TCPT_REXMT] = 0;
+
+               (void)tcp_output(tp);
+               tp->snd_cwnd -= tp->t_maxseg;
+
+               tp->t_tlphighrxt = tp->snd_nxt;
+               break;
+       }
+       case TCPT_DELAYFR:
+               tp->t_flagsext &= ~TF_DELAY_RECOVERY;
+
+               /*
+                * Don't do anything if one of the following is true:
+                * - the connection is already in recovery
+                * - sequence until snd_recover has been acknowledged.
+                * - retransmit timeout has fired
+                */
+               if (IN_FASTRECOVERY(tp) ||
+                   SEQ_GEQ(tp->snd_una, tp->snd_recover) ||
+                   tp->t_rxtshift > 0) {
+                       break;
+               }
+
+               VERIFY(SACK_ENABLED(tp));
+               tcp_rexmt_save_state(tp);
+               if (CC_ALGO(tp)->pre_fr != NULL) {
+                       CC_ALGO(tp)->pre_fr(tp);
+                       if (TCP_ECN_ENABLED(tp)) {
+                               tp->ecn_flags |= TE_SENDCWR;
+                       }
+               }
+               ENTER_FASTRECOVERY(tp);
+
+               tp->t_timer[TCPT_REXMT] = 0;
+               tcpstat.tcps_sack_recovery_episode++;
+               tp->t_sack_recovery_episode++;
+               tp->sack_newdata = tp->snd_nxt;
+               tp->snd_cwnd = tp->t_maxseg;
+               tcp_ccdbg_trace(tp, NULL, TCP_CC_ENTER_FASTRECOVERY);
+               (void) tcp_output(tp);
+               break;
+dropit:
                tcpstat.tcps_keepdrops++;
                tcpstat.tcps_keepdrops++;
+               postevent(so, 0, EV_TIMEOUT);
+               soevent(so,
+                   (SO_FILT_HINT_LOCKED | SO_FILT_HINT_TIMEOUT));
                tp = tcp_drop(tp, ETIMEDOUT);
                tp = tcp_drop(tp, ETIMEDOUT);
-               postevent(so_tmp, 0, EV_TIMEOUT);
                break;
        }
                break;
        }
-       return (tp);
+#if TCPDEBUG
+       if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) {
+               tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
+                   PRU_SLOWTIMO);
+       }
+#endif
+       return tp;
 }
 
 /* Remove a timer entry from timer list */
 }
 
 /* Remove a timer entry from timer list */
@@ -868,134 +1623,169 @@ tcp_remove_timer(struct tcpcb *tp)
 {
        struct tcptimerlist *listp = &tcp_timer_list;
 
 {
        struct tcptimerlist *listp = &tcp_timer_list;
 
-       lck_mtx_assert(&tp->t_inpcb->inpcb_mtx, LCK_MTX_ASSERT_OWNED);
+       socket_lock_assert_owned(tp->t_inpcb->inp_socket);
        if (!(TIMER_IS_ON_LIST(tp))) {
                return;
        }
        lck_mtx_lock(listp->mtx);
        if (!(TIMER_IS_ON_LIST(tp))) {
                return;
        }
        lck_mtx_lock(listp->mtx);
-       
+
        /* Check if pcb is on timer list again after acquiring the lock */
        if (!(TIMER_IS_ON_LIST(tp))) {
                lck_mtx_unlock(listp->mtx);
                return;
        }
        /* Check if pcb is on timer list again after acquiring the lock */
        if (!(TIMER_IS_ON_LIST(tp))) {
                lck_mtx_unlock(listp->mtx);
                return;
        }
-       
-       if (listp->next_te != NULL && listp->next_te == &tp->tentry)
+
+       if (listp->next_te != NULL && listp->next_te == &tp->tentry) {
                listp->next_te = LIST_NEXT(&tp->tentry, le);
                listp->next_te = LIST_NEXT(&tp->tentry, le);
+       }
 
        LIST_REMOVE(&tp->tentry, le);
        tp->t_flags &= ~(TF_TIMER_ONLIST);
 
        listp->entries--;
 
        LIST_REMOVE(&tp->tentry, le);
        tp->t_flags &= ~(TF_TIMER_ONLIST);
 
        listp->entries--;
-       lck_mtx_unlock(listp->mtx);
 
        tp->tentry.le.le_next = NULL;
        tp->tentry.le.le_prev = NULL;
 
        tp->tentry.le.le_next = NULL;
        tp->tentry.le.le_prev = NULL;
+       lck_mtx_unlock(listp->mtx);
 }
 
 }
 
-/* Function to check if the timerlist needs to be rescheduled to run
+/*
+ * Function to check if the timerlist needs to be rescheduled to run
  * the timer entry correctly. Basically, this is to check if we can avoid
  * taking the list lock.
  */
 
 static boolean_t
  * the timer entry correctly. Basically, this is to check if we can avoid
  * taking the list lock.
  */
 
 static boolean_t
-need_to_resched_timerlist(uint32_t runtime, uint16_t index) {
+need_to_resched_timerlist(u_int32_t runtime, u_int16_t mode)
+{
        struct tcptimerlist *listp = &tcp_timer_list;
        int32_t diff;
        struct tcptimerlist *listp = &tcp_timer_list;
        int32_t diff;
-       boolean_t is_fast;
 
 
-       if (runtime == 0 || index == TCPT_NONE)
-               return FALSE;
-       is_fast = !(IS_TIMER_SLOW(index));
-
-       /* If the list is being processed then the state of the list is in flux.
-        * In this case always acquire the lock and set the state correctly.
+       /*
+        * If the list is being processed then the state of the list is
+        * in flux. In this case always acquire the lock and set the state
+        * correctly.
         */
        if (listp->running) {
                return TRUE;
        }
 
         */
        if (listp->running) {
                return TRUE;
        }
 
+       if (!listp->scheduled) {
+               return TRUE;
+       }
+
        diff = timer_diff(listp->runtime, 0, runtime, 0);
        if (diff <= 0) {
                /* The list is going to run before this timer */
                return FALSE;
        } else {
        diff = timer_diff(listp->runtime, 0, runtime, 0);
        if (diff <= 0) {
                /* The list is going to run before this timer */
                return FALSE;
        } else {
-               if (is_fast) {
-                       if (diff <= listp->fast_quantum)
+               if (mode & TCP_TIMERLIST_10MS_MODE) {
+                       if (diff <= TCP_TIMER_10MS_QUANTUM) {
                                return FALSE;
                                return FALSE;
+                       }
+               } else if (mode & TCP_TIMERLIST_100MS_MODE) {
+                       if (diff <= TCP_TIMER_100MS_QUANTUM) {
+                               return FALSE;
+                       }
                } else {
                } else {
-                       if (diff <= listp->slow_quantum)
+                       if (diff <= TCP_TIMER_500MS_QUANTUM) {
                                return FALSE;
                                return FALSE;
+                       }
                }
        }
        return TRUE;
 }
 
 void
                }
        }
        return TRUE;
 }
 
 void
-tcp_sched_timerlist(uint32_t offset) 
+tcp_sched_timerlist(uint32_t offset)
 {
 {
-
        uint64_t deadline = 0;
        struct tcptimerlist *listp = &tcp_timer_list;
 
        uint64_t deadline = 0;
        struct tcptimerlist *listp = &tcp_timer_list;
 
-       lck_mtx_assert(listp->mtx, LCK_MTX_ASSERT_OWNED);
+       LCK_MTX_ASSERT(listp->mtx, LCK_MTX_ASSERT_OWNED);
 
 
+       offset = min(offset, TCP_TIMERLIST_MAX_OFFSET);
        listp->runtime = tcp_now + offset;
        listp->runtime = tcp_now + offset;
+       listp->schedtime = tcp_now;
+       if (listp->runtime == 0) {
+               listp->runtime++;
+               offset++;
+       }
 
 
-       clock_interval_to_deadline(offset, NSEC_PER_SEC / TCP_RETRANSHZ,
-               &deadline);
+       clock_interval_to_deadline(offset, USEC_PER_SEC, &deadline);
 
        thread_call_enter_delayed(listp->call, deadline);
 
        thread_call_enter_delayed(listp->call, deadline);
+       listp->scheduled = TRUE;
 }
 
 }
 
-/* Function to run the timers for a connection.
+/*
+ * Function to run the timers for a connection.
  *
  *
- * Returns the offset of next timer to be run for this connection which 
+ * Returns the offset of next timer to be run for this connection which
  * can be used to reschedule the timerlist.
  * can be used to reschedule the timerlist.
+ *
+ * te_mode is an out parameter that indicates the modes of active
+ * timers for this connection.
  */
  */
-uint32_t
-tcp_run_conn_timer(struct tcpcb *tp, uint16_t *next_index) {
-
-        struct socket *so;
-        uint16_t i = 0, index = TCPT_NONE, lo_index = TCPT_NONE;
-        uint32_t timer_val, offset = 0, lo_timer = 0;
+u_int32_t
+tcp_run_conn_timer(struct tcpcb *tp, u_int16_t *te_mode,
+    u_int16_t probe_if_index)
+{
+       struct socket *so;
+       u_int16_t i = 0, index = TCPT_NONE, lo_index = TCPT_NONE;
+       u_int32_t timer_val, offset = 0, lo_timer = 0;
        int32_t diff;
        boolean_t needtorun[TCPT_NTIMERS];
        int count = 0;
 
        int32_t diff;
        boolean_t needtorun[TCPT_NTIMERS];
        int count = 0;
 
-        VERIFY(tp != NULL);
-        bzero(needtorun, sizeof(needtorun));
+       VERIFY(tp != NULL);
+       bzero(needtorun, sizeof(needtorun));
+       *te_mode = 0;
 
 
-        tcp_lock(tp->t_inpcb->inp_socket, 1, 0);
+       socket_lock(tp->t_inpcb->inp_socket, 1);
 
 
-        so = tp->t_inpcb->inp_socket;
-       /* Release the want count on inp */ 
-       if (in_pcb_checkstate(tp->t_inpcb, WNT_RELEASE, 1) == WNT_STOPUSING) {
+       so = tp->t_inpcb->inp_socket;
+       /* Release the want count on inp */
+       if (in_pcb_checkstate(tp->t_inpcb, WNT_RELEASE, 1)
+           == WNT_STOPUSING) {
                if (TIMER_IS_ON_LIST(tp)) {
                        tcp_remove_timer(tp);
                }
 
                if (TIMER_IS_ON_LIST(tp)) {
                        tcp_remove_timer(tp);
                }
 
-               /* Looks like the TCP connection got closed while we 
+               /* Looks like the TCP connection got closed while we
                 * were waiting for the lock.. Done
                 */
                goto done;
        }
 
                 * were waiting for the lock.. Done
                 */
                goto done;
        }
 
-        /* Since the timer thread needs to wait for tcp lock, it may race
-         * with another thread that can cancel or reschedule the timer that is
-         * about to run. Check if we need to run anything.
-         */
-       index = tp->tentry.index;
-       timer_val = tp->t_timer[index];
+       /*
+        * If this connection is over an interface that needs to
+        * be probed, send probe packets to reinitiate communication.
+        */
+       if (probe_if_index > 0 && tp->t_inpcb->inp_last_outifp != NULL &&
+           tp->t_inpcb->inp_last_outifp->if_index == probe_if_index) {
+               tp->t_flagsext |= TF_PROBING;
+               tcp_timers(tp, TCPT_PTO);
+               tp->t_timer[TCPT_PTO] = 0;
+               tp->t_flagsext &= ~TF_PROBING;
+       }
 
 
-        if (index == TCPT_NONE || tp->tentry.runtime == 0) 
+       /*
+        * Since the timer thread needs to wait for tcp lock, it may race
+        * with another thread that can cancel or reschedule the timer
+        * that is about to run. Check if we need to run anything.
+        */
+       if ((index = tp->tentry.index) == TCPT_NONE) {
                goto done;
                goto done;
+       }
+
+       timer_val = tp->t_timer[index];
 
        diff = timer_diff(tp->tentry.runtime, 0, tcp_now, 0);
        if (diff > 0) {
                if (tp->tentry.index != TCPT_NONE) {
                        offset = diff;
 
        diff = timer_diff(tp->tentry.runtime, 0, tcp_now, 0);
        if (diff > 0) {
                if (tp->tentry.index != TCPT_NONE) {
                        offset = diff;
-                       *(next_index) = tp->tentry.index;
+                       *(te_mode) = tp->tentry.mode;
                }
                goto done;
        }
                }
                goto done;
        }
@@ -1003,18 +1793,21 @@ tcp_run_conn_timer(struct tcpcb *tp, uint16_t *next_index) {
        tp->t_timer[index] = 0;
        if (timer_val > 0) {
                tp = tcp_timers(tp, index);
        tp->t_timer[index] = 0;
        if (timer_val > 0) {
                tp = tcp_timers(tp, index);
-               if (tp == NULL) 
+               if (tp == NULL) {
                        goto done;
                        goto done;
+               }
        }
        }
-       
-       /* Check if there are any other timers that need to be run. While doing it,
-        * adjust the timer values wrt tcp_now.
+
+       /*
+        * Check if there are any other timers that need to be run.
+        * While doing it, adjust the timer values wrt tcp_now.
         */
         */
+       tp->tentry.mode = 0;
        for (i = 0; i < TCPT_NTIMERS; ++i) {
                if (tp->t_timer[i] != 0) {
        for (i = 0; i < TCPT_NTIMERS; ++i) {
                if (tp->t_timer[i] != 0) {
-                       diff = timer_diff(tp->tentry.timer_start, tp->t_timer[i], tcp_now, 0);
+                       diff = timer_diff(tp->tentry.timer_start,
+                           tp->t_timer[i], tcp_now, 0);
                        if (diff <= 0) {
                        if (diff <= 0) {
-                               tp->t_timer[i] = 0;
                                needtorun[i] = TRUE;
                                count++;
                        } else {
                                needtorun[i] = TRUE;
                                count++;
                        } else {
@@ -1024,26 +1817,34 @@ tcp_run_conn_timer(struct tcpcb *tp, uint16_t *next_index) {
                                        lo_timer = diff;
                                        lo_index = i;
                                }
                                        lo_timer = diff;
                                        lo_index = i;
                                }
+                               TCP_SET_TIMER_MODE(tp->tentry.mode, i);
                        }
                }
        }
                        }
                }
        }
-       
+
        tp->tentry.timer_start = tcp_now;
        tp->tentry.index = lo_index;
        tp->tentry.timer_start = tcp_now;
        tp->tentry.index = lo_index;
-       if (lo_index != TCPT_NONE) {
-               tp->tentry.runtime = tp->tentry.timer_start + tp->t_timer[lo_index];
-       } else {
-               tp->tentry.runtime = 0;
+       VERIFY(tp->tentry.index == TCPT_NONE || tp->tentry.mode > 0);
+
+       if (tp->tentry.index != TCPT_NONE) {
+               tp->tentry.runtime = tp->tentry.timer_start +
+                   tp->t_timer[tp->tentry.index];
+               if (tp->tentry.runtime == 0) {
+                       tp->tentry.runtime++;
+               }
        }
 
        if (count > 0) {
        }
 
        if (count > 0) {
-               /* run any other timers that are also outstanding at this time. */
+               /* run any other timers outstanding at this time. */
                for (i = 0; i < TCPT_NTIMERS; ++i) {
                        if (needtorun[i]) {
                                tp->t_timer[i] = 0;
                                tp = tcp_timers(tp, i);
                for (i = 0; i < TCPT_NTIMERS; ++i) {
                        if (needtorun[i]) {
                                tp->t_timer[i] = 0;
                                tp = tcp_timers(tp, i);
-                               if (tp == NULL) 
+                               if (tp == NULL) {
+                                       offset = 0;
+                                       *(te_mode) = 0;
                                        goto done;
                                        goto done;
+                               }
                        }
                }
                tcp_set_lotimer_index(tp);
                        }
                }
                tcp_set_lotimer_index(tp);
@@ -1051,58 +1852,84 @@ tcp_run_conn_timer(struct tcpcb *tp, uint16_t *next_index) {
 
        if (tp->tentry.index < TCPT_NONE) {
                offset = tp->t_timer[tp->tentry.index];
 
        if (tp->tentry.index < TCPT_NONE) {
                offset = tp->t_timer[tp->tentry.index];
-               *(next_index) = tp->tentry.index;
+               *(te_mode) = tp->tentry.mode;
        }
 
 done:
        if (tp != NULL && tp->tentry.index == TCPT_NONE) {
                tcp_remove_timer(tp);
        }
 
 done:
        if (tp != NULL && tp->tentry.index == TCPT_NONE) {
                tcp_remove_timer(tp);
+               offset = 0;
        }
        }
-        tcp_unlock(so, 1, 0);
-        return offset;
+
+       socket_unlock(so, 1);
+       return offset;
 }
 
 void
 }
 
 void
-tcp_run_timerlist(void * arg1, void * arg2) {
-
+tcp_run_timerlist(void * arg1, void * arg2)
+{
 #pragma unused(arg1, arg2)
 #pragma unused(arg1, arg2)
-       
        struct tcptimerentry *te, *next_te;
        struct tcptimerlist *listp = &tcp_timer_list;
        struct tcpcb *tp;
        struct tcptimerentry *te, *next_te;
        struct tcptimerlist *listp = &tcp_timer_list;
        struct tcpcb *tp;
-       uint32_t next_timer = 0;
-       uint16_t index = TCPT_NONE;
-       boolean_t need_fast = FALSE;
+       uint32_t next_timer = 0; /* offset of the next timer on the list */
+       u_int16_t te_mode = 0;  /* modes of all active timers in a tcpcb */
+       u_int16_t list_mode = 0; /* cumulative of modes of all tcpcbs */
        uint32_t active_count = 0;
        uint32_t active_count = 0;
-       uint32_t mode = TCP_TIMERLIST_FASTMODE;
 
        calculate_tcp_clock();
 
        lck_mtx_lock(listp->mtx);
 
 
        calculate_tcp_clock();
 
        lck_mtx_lock(listp->mtx);
 
+       int32_t drift = tcp_now - listp->runtime;
+       if (drift <= 1) {
+               tcpstat.tcps_timer_drift_le_1_ms++;
+       } else if (drift <= 10) {
+               tcpstat.tcps_timer_drift_le_10_ms++;
+       } else if (drift <= 20) {
+               tcpstat.tcps_timer_drift_le_20_ms++;
+       } else if (drift <= 50) {
+               tcpstat.tcps_timer_drift_le_50_ms++;
+       } else if (drift <= 100) {
+               tcpstat.tcps_timer_drift_le_100_ms++;
+       } else if (drift <= 200) {
+               tcpstat.tcps_timer_drift_le_200_ms++;
+       } else if (drift <= 500) {
+               tcpstat.tcps_timer_drift_le_500_ms++;
+       } else if (drift <= 1000) {
+               tcpstat.tcps_timer_drift_le_1000_ms++;
+       } else {
+               tcpstat.tcps_timer_drift_gt_1000_ms++;
+       }
+
        listp->running = TRUE;
        listp->running = TRUE;
-       
+
        LIST_FOREACH_SAFE(te, &listp->lhead, le, next_te) {
                uint32_t offset = 0;
                uint32_t runtime = te->runtime;
        LIST_FOREACH_SAFE(te, &listp->lhead, le, next_te) {
                uint32_t offset = 0;
                uint32_t runtime = te->runtime;
-               if (TSTMP_GT(runtime, tcp_now)) {
+               if (te->index < TCPT_NONE && TSTMP_GT(runtime, tcp_now)) {
                        offset = timer_diff(runtime, 0, tcp_now, 0);
                        if (next_timer == 0 || offset < next_timer) {
                                next_timer = offset;
                        }
                        offset = timer_diff(runtime, 0, tcp_now, 0);
                        if (next_timer == 0 || offset < next_timer) {
                                next_timer = offset;
                        }
+                       list_mode |= te->mode;
                        continue;
                }
                        continue;
                }
-               active_count++;
 
                tp = TIMERENTRY_TO_TP(te);
 
 
                tp = TIMERENTRY_TO_TP(te);
 
-               /* Acquire an inp wantcnt on the inpcb so that the socket won't get
-                * detached even if tcp_close is called
+               /*
+                * Acquire an inp wantcnt on the inpcb so that the socket
+                * won't get detached even if tcp_close is called
                 */
                 */
-               if (in_pcb_checkstate(tp->t_inpcb, WNT_ACQUIRE, 0) == WNT_STOPUSING) {
-                       /* Some how this pcb went into dead state while on the timer list,
-                        * just take it off the list. Since the timer list entry pointers 
-                        * are protected by the timer list lock, we can do it here
+               if (in_pcb_checkstate(tp->t_inpcb, WNT_ACQUIRE, 0)
+                   == WNT_STOPUSING) {
+                       /*
+                        * Some how this pcb went into dead state while
+                        * on the timer list, just take it off the list.
+                        * Since the timer list entry pointers are
+                        * protected by the timer list lock, we can
+                        * do it here without the socket lock.
                         */
                        if (TIMER_IS_ON_LIST(tp)) {
                                tp->t_flags &= ~(TF_TIMER_ONLIST);
                         */
                        if (TIMER_IS_ON_LIST(tp)) {
                                tp->t_flags &= ~(TF_TIMER_ONLIST);
@@ -1114,94 +1941,110 @@ tcp_run_timerlist(void * arg1, void * arg2) {
                        }
                        continue;
                }
                        }
                        continue;
                }
+               active_count++;
 
 
-               /* Store the next timerentry pointer before releasing the list lock.
-                * If that entry has to be removed when we release the lock, this
-                * pointer will be updated to the element after that.
+               /*
+                * Store the next timerentry pointer before releasing the
+                * list lock. If that entry has to be removed when we
+                * release the lock, this pointer will be updated to the
+                * element after that.
                 */
                 */
-               listp->next_te = next_te; 
+               listp->next_te = next_te;
 
                VERIFY_NEXT_LINK(&tp->tentry, le);
                VERIFY_PREV_LINK(&tp->tentry, le);
 
                lck_mtx_unlock(listp->mtx);
 
 
                VERIFY_NEXT_LINK(&tp->tentry, le);
                VERIFY_PREV_LINK(&tp->tentry, le);
 
                lck_mtx_unlock(listp->mtx);
 
-               index = TCPT_NONE;
-               offset = tcp_run_conn_timer(tp, &index);
-               
+               offset = tcp_run_conn_timer(tp, &te_mode,
+                   listp->probe_if_index);
+
                lck_mtx_lock(listp->mtx);
 
                next_te = listp->next_te;
                listp->next_te = NULL;
 
                lck_mtx_lock(listp->mtx);
 
                next_te = listp->next_te;
                listp->next_te = NULL;
 
-               if (offset > 0) {
-                       if (index < TCPT_NONE) {
-                               /* Check if this is a fast_timer. */
-                               if (!need_fast && !(IS_TIMER_SLOW(index))) {
-                                       need_fast = TRUE;
-                               }
+               if (offset > 0 && te_mode != 0) {
+                       list_mode |= te_mode;
 
 
-                               if (next_timer == 0 || offset < next_timer) {
-                                       next_timer = offset;
-                               }
+                       if (next_timer == 0 || offset < next_timer) {
+                               next_timer = offset;
                        }
                }
        }
 
        if (!LIST_EMPTY(&listp->lhead)) {
                        }
                }
        }
 
        if (!LIST_EMPTY(&listp->lhead)) {
-               if (listp->mode == TCP_TIMERLIST_FASTMODE) {
-                       if (need_fast || active_count > 0 || 
-                               listp->pref_mode == TCP_TIMERLIST_FASTMODE) {
-                               listp->idlegen = 0;
-                       } else {
-                               listp->idlegen++;
-                               if (listp->idlegen > timer_fastmode_idlemax) {
-                                       mode = TCP_TIMERLIST_SLOWMODE;
-                                       listp->idlegen = 0;
-                               }
-                       }
+               u_int16_t next_mode = 0;
+               if ((list_mode & TCP_TIMERLIST_10MS_MODE) ||
+                   (listp->pref_mode & TCP_TIMERLIST_10MS_MODE)) {
+                       next_mode = TCP_TIMERLIST_10MS_MODE;
+               } else if ((list_mode & TCP_TIMERLIST_100MS_MODE) ||
+                   (listp->pref_mode & TCP_TIMERLIST_100MS_MODE)) {
+                       next_mode = TCP_TIMERLIST_100MS_MODE;
                } else {
                } else {
-                       if (!need_fast) {
-                               mode = TCP_TIMERLIST_SLOWMODE;
-                       }
+                       next_mode = TCP_TIMERLIST_500MS_MODE;
                }
 
                }
 
-               if (mode == TCP_TIMERLIST_FASTMODE || 
-                       listp->pref_mode == TCP_TIMERLIST_FASTMODE) {
-                       next_timer = listp->fast_quantum;
+               if (next_mode != TCP_TIMERLIST_500MS_MODE) {
+                       listp->idleruns = 0;
                } else {
                } else {
-                       if (listp->pref_offset != 0 && 
-                               listp->pref_offset < next_timer)
-                               next_timer = listp->pref_offset;
-                       if (next_timer < listp->slow_quantum)
-                               next_timer = listp->slow_quantum;
+                       /*
+                        * the next required mode is slow mode, but if
+                        * the last one was a faster mode and we did not
+                        * have enough idle runs, repeat the last mode.
+                        *
+                        * We try to keep the timer list in fast mode for
+                        * some idle time in expectation of new data.
+                        */
+                       if (listp->mode != next_mode &&
+                           listp->idleruns < timer_fastmode_idlemax) {
+                               listp->idleruns++;
+                               next_mode = listp->mode;
+                               next_timer = TCP_TIMER_100MS_QUANTUM;
+                       } else {
+                               listp->idleruns = 0;
+                       }
+               }
+               listp->mode = next_mode;
+               if (listp->pref_offset != 0) {
+                       next_timer = min(listp->pref_offset, next_timer);
                }
 
                }
 
-               listp->mode = mode;
+               if (listp->mode == TCP_TIMERLIST_500MS_MODE) {
+                       next_timer = max(next_timer,
+                           TCP_TIMER_500MS_QUANTUM);
+               }
 
                tcp_sched_timerlist(next_timer);
        } else {
 
                tcp_sched_timerlist(next_timer);
        } else {
-               /* No need to reschedule this timer */
-               listp->runtime = 0;
+               /*
+                * No need to reschedule this timer, but always run
+                * periodically at a much higher granularity.
+                */
+               tcp_sched_timerlist(TCP_TIMERLIST_MAX_OFFSET);
        }
 
        listp->running = FALSE;
        listp->pref_mode = 0;
        listp->pref_offset = 0;
        }
 
        listp->running = FALSE;
        listp->pref_mode = 0;
        listp->pref_offset = 0;
+       listp->probe_if_index = 0;
 
        lck_mtx_unlock(listp->mtx);
 }
 
 
        lck_mtx_unlock(listp->mtx);
 }
 
-/* Function to verify if a change in timer state is required for a connection */
-void 
-tcp_sched_timers(struct tcpcb *tp) 
+/*
+ * Function to check if the timerlist needs to be rescheduled to run this
+ * connection's timers correctly.
+ */
+void
+tcp_sched_timers(struct tcpcb *tp)
 {
        struct tcptimerentry *te = &tp->tentry;
 {
        struct tcptimerentry *te = &tp->tentry;
-       uint16_t index = te->index;
+       u_int16_t index = te->index;
+       u_int16_t mode = te->mode;
        struct tcptimerlist *listp = &tcp_timer_list;
        struct tcptimerlist *listp = &tcp_timer_list;
-       uint32_t offset = 0;
-       boolean_t is_fast;
-       int list_locked = 0;
+       int32_t offset = 0;
+       boolean_t list_locked = FALSE;
 
        if (tp->t_inpcb->inp_state == INPCB_STATE_DEAD) {
                /* Just return without adding the dead pcb to the list */
 
        if (tp->t_inpcb->inp_state == INPCB_STATE_DEAD) {
                /* Just return without adding the dead pcb to the list */
@@ -1212,64 +2055,78 @@ tcp_sched_timers(struct tcpcb *tp)
        }
 
        if (index == TCPT_NONE) {
        }
 
        if (index == TCPT_NONE) {
+               /* Nothing to run */
                tcp_remove_timer(tp);
                return;
        }
 
                tcp_remove_timer(tp);
                return;
        }
 
-       is_fast = !(IS_TIMER_SLOW(index));
-       offset = te->runtime - tcp_now;
-       if (offset == 0) {
+       /*
+        * compute the offset at which the next timer for this connection
+        * has to run.
+        */
+       offset = timer_diff(te->runtime, 0, tcp_now, 0);
+       if (offset <= 0) {
                offset = 1;
                tcp_timer_advanced++;
        }
                offset = 1;
                tcp_timer_advanced++;
        }
-       if (is_fast)
-               offset = listp->fast_quantum;
 
        if (!TIMER_IS_ON_LIST(tp)) {
                if (!list_locked) {
                        lck_mtx_lock(listp->mtx);
 
        if (!TIMER_IS_ON_LIST(tp)) {
                if (!list_locked) {
                        lck_mtx_lock(listp->mtx);
-                       list_locked = 1;
+                       list_locked = TRUE;
                }
 
                }
 
-               LIST_INSERT_HEAD(&listp->lhead, te, le);
-               tp->t_flags |= TF_TIMER_ONLIST;
-
-               listp->entries++;
-               if (listp->entries > listp->maxentries)
-                       listp->maxentries = listp->entries;
+               if (!TIMER_IS_ON_LIST(tp)) {
+                       LIST_INSERT_HEAD(&listp->lhead, te, le);
+                       tp->t_flags |= TF_TIMER_ONLIST;
 
 
-               /* if the list is not scheduled, just schedule it */
-               if (listp->runtime == 0)
-                       goto schedule;
+                       listp->entries++;
+                       if (listp->entries > listp->maxentries) {
+                               listp->maxentries = listp->entries;
+                       }
 
 
+                       /* if the list is not scheduled, just schedule it */
+                       if (!listp->scheduled) {
+                               goto schedule;
+                       }
+               }
        }
 
        }
 
-
-       /* timer entry is currently on the list */
-       if (need_to_resched_timerlist(te->runtime, index)) {
+       /*
+        * Timer entry is currently on the list, check if the list needs
+        * to be rescheduled.
+        */
+       if (need_to_resched_timerlist(te->runtime, mode)) {
                tcp_resched_timerlist++;
                tcp_resched_timerlist++;
-       
+
                if (!list_locked) {
                        lck_mtx_lock(listp->mtx);
                if (!list_locked) {
                        lck_mtx_lock(listp->mtx);
-                       list_locked = 1;
+                       list_locked = TRUE;
                }
 
                VERIFY_NEXT_LINK(te, le);
                VERIFY_PREV_LINK(te, le);
 
                if (listp->running) {
                }
 
                VERIFY_NEXT_LINK(te, le);
                VERIFY_PREV_LINK(te, le);
 
                if (listp->running) {
-                       if (is_fast) {
-                               listp->pref_mode = TCP_TIMERLIST_FASTMODE;
-                       } else if (listp->pref_offset == 0 ||
-                               ((int)offset) < listp->pref_offset) {
+                       listp->pref_mode |= mode;
+                       if (listp->pref_offset == 0 ||
+                           offset < listp->pref_offset) {
                                listp->pref_offset = offset;
                        }
                } else {
                                listp->pref_offset = offset;
                        }
                } else {
-                       int32_t diff;
-                       diff = timer_diff(listp->runtime, 0, tcp_now, offset);
-                       if (diff <= 0) {
-                               /* The list is going to run before this timer */
-                               goto done;
+                       /*
+                        * The list could have got rescheduled while
+                        * this thread was waiting for the lock
+                        */
+                       if (listp->scheduled) {
+                               int32_t diff;
+                               diff = timer_diff(listp->runtime, 0,
+                                   tcp_now, offset);
+                               if (diff <= 0) {
+                                       goto done;
+                               } else {
+                                       goto schedule;
+                               }
                        } else {
                                goto schedule;
                        }
                        } else {
                                goto schedule;
                        }
@@ -1278,45 +2135,637 @@ tcp_sched_timers(struct tcpcb *tp)
        goto done;
 
 schedule:
        goto done;
 
 schedule:
-       if (is_fast) {
-               listp->mode = TCP_TIMERLIST_FASTMODE;
-               listp->idlegen = 0;
+       /*
+        * Since a connection with timers is getting scheduled, the timer
+        * list moves from idle to active state and that is why idlegen is
+        * reset
+        */
+       if (mode & TCP_TIMERLIST_10MS_MODE) {
+               listp->mode = TCP_TIMERLIST_10MS_MODE;
+               listp->idleruns = 0;
+               offset = min(offset, TCP_TIMER_10MS_QUANTUM);
+       } else if (mode & TCP_TIMERLIST_100MS_MODE) {
+               if (listp->mode > TCP_TIMERLIST_100MS_MODE) {
+                       listp->mode = TCP_TIMERLIST_100MS_MODE;
+               }
+               listp->idleruns = 0;
+               offset = min(offset, TCP_TIMER_100MS_QUANTUM);
        }
        tcp_sched_timerlist(offset);
 
 done:
        }
        tcp_sched_timerlist(offset);
 
 done:
-       if (list_locked)
+       if (list_locked) {
                lck_mtx_unlock(listp->mtx);
                lck_mtx_unlock(listp->mtx);
+       }
 
        return;
 }
 
        return;
 }
-               
-void
-tcp_set_lotimer_index(struct tcpcb *tp) {
-       uint16_t i, lo_index = TCPT_NONE;
+
+static inline void
+tcp_set_lotimer_index(struct tcpcb *tp)
+{
+       uint16_t i, lo_index = TCPT_NONE, mode = 0;
        uint32_t lo_timer = 0;
        for (i = 0; i < TCPT_NTIMERS; ++i) {
        uint32_t lo_timer = 0;
        for (i = 0; i < TCPT_NTIMERS; ++i) {
-               if (tp->t_timer[i] != 0 &&
-                       (lo_timer == 0 || tp->t_timer[i] < lo_timer)) {
-                       lo_timer = tp->t_timer[i];
-                       lo_index = i;
+               if (tp->t_timer[i] != 0) {
+                       TCP_SET_TIMER_MODE(mode, i);
+                       if (lo_timer == 0 || tp->t_timer[i] < lo_timer) {
+                               lo_timer = tp->t_timer[i];
+                               lo_index = i;
+                       }
                }
        }
        tp->tentry.index = lo_index;
                }
        }
        tp->tentry.index = lo_index;
-       if (lo_index != TCPT_NONE) {
-               tp->tentry.runtime = tp->tentry.timer_start + tp->t_timer[lo_index];
-       } else {
-               tp->tentry.runtime = 0;
+       tp->tentry.mode = mode;
+       VERIFY(tp->tentry.index == TCPT_NONE || tp->tentry.mode > 0);
+
+       if (tp->tentry.index != TCPT_NONE) {
+               tp->tentry.runtime = tp->tentry.timer_start
+                   + tp->t_timer[tp->tentry.index];
+               if (tp->tentry.runtime == 0) {
+                       tp->tentry.runtime++;
+               }
        }
 }
 
 void
        }
 }
 
 void
-tcp_check_timer_state(struct tcpcb *tp) {
+tcp_check_timer_state(struct tcpcb *tp)
+{
+       socket_lock_assert_owned(tp->t_inpcb->inp_socket);
 
 
-       lck_mtx_assert(&tp->t_inpcb->inpcb_mtx, LCK_MTX_ASSERT_OWNED);
+       if (tp->t_inpcb->inp_flags2 & INP2_TIMEWAIT) {
+               return;
+       }
 
        tcp_set_lotimer_index(tp);
 
        tcp_sched_timers(tp);
        return;
 }
 
        tcp_set_lotimer_index(tp);
 
        tcp_sched_timers(tp);
        return;
 }
+
+static inline void
+tcp_cumulative_stat(u_int32_t cur, u_int32_t *prev, u_int32_t *dest)
+{
+       /* handle wrap around */
+       int32_t diff = (int32_t) (cur - *prev);
+       if (diff > 0) {
+               *dest = diff;
+       } else {
+               *dest = 0;
+       }
+       *prev = cur;
+       return;
+}
+
+static inline void
+tcp_cumulative_stat64(u_int64_t cur, u_int64_t *prev, u_int64_t *dest)
+{
+       /* handle wrap around */
+       int64_t diff = (int64_t) (cur - *prev);
+       if (diff > 0) {
+               *dest = diff;
+       } else {
+               *dest = 0;
+       }
+       *prev = cur;
+       return;
+}
+
+__private_extern__ void
+tcp_report_stats(void)
+{
+       struct nstat_sysinfo_data data;
+       struct sockaddr_in dst;
+       struct sockaddr_in6 dst6;
+       struct rtentry *rt = NULL;
+       static struct tcp_last_report_stats prev;
+       u_int64_t var, uptime;
+
+#define stat    data.u.tcp_stats
+       if (((uptime = net_uptime()) - tcp_last_report_time) <
+           tcp_report_stats_interval) {
+               return;
+       }
+
+       tcp_last_report_time = uptime;
+
+       bzero(&data, sizeof(data));
+       data.flags = NSTAT_SYSINFO_TCP_STATS;
+
+       bzero(&dst, sizeof(dst));
+       dst.sin_len = sizeof(dst);
+       dst.sin_family = AF_INET;
+
+       /* ipv4 avg rtt */
+       lck_mtx_lock(rnh_lock);
+       rt =  rt_lookup(TRUE, (struct sockaddr *)&dst, NULL,
+           rt_tables[AF_INET], IFSCOPE_NONE);
+       lck_mtx_unlock(rnh_lock);
+       if (rt != NULL) {
+               RT_LOCK(rt);
+               if (rt_primary_default(rt, rt_key(rt)) &&
+                   rt->rt_stats != NULL) {
+                       stat.ipv4_avgrtt = rt->rt_stats->nstat_avg_rtt;
+               }
+               RT_UNLOCK(rt);
+               rtfree(rt);
+               rt = NULL;
+       }
+
+       /* ipv6 avg rtt */
+       bzero(&dst6, sizeof(dst6));
+       dst6.sin6_len = sizeof(dst6);
+       dst6.sin6_family = AF_INET6;
+
+       lck_mtx_lock(rnh_lock);
+       rt = rt_lookup(TRUE, (struct sockaddr *)&dst6, NULL,
+           rt_tables[AF_INET6], IFSCOPE_NONE);
+       lck_mtx_unlock(rnh_lock);
+       if (rt != NULL) {
+               RT_LOCK(rt);
+               if (rt_primary_default(rt, rt_key(rt)) &&
+                   rt->rt_stats != NULL) {
+                       stat.ipv6_avgrtt = rt->rt_stats->nstat_avg_rtt;
+               }
+               RT_UNLOCK(rt);
+               rtfree(rt);
+               rt = NULL;
+       }
+
+       /* send packet loss rate, shift by 10 for precision */
+       if (tcpstat.tcps_sndpack > 0 && tcpstat.tcps_sndrexmitpack > 0) {
+               var = tcpstat.tcps_sndrexmitpack << 10;
+               stat.send_plr = (var * 100) / tcpstat.tcps_sndpack;
+       }
+
+       /* recv packet loss rate, shift by 10 for precision */
+       if (tcpstat.tcps_rcvpack > 0 && tcpstat.tcps_recovered_pkts > 0) {
+               var = tcpstat.tcps_recovered_pkts << 10;
+               stat.recv_plr = (var * 100) / tcpstat.tcps_rcvpack;
+       }
+
+       /* RTO after tail loss, shift by 10 for precision */
+       if (tcpstat.tcps_sndrexmitpack > 0
+           && tcpstat.tcps_tailloss_rto > 0) {
+               var = tcpstat.tcps_tailloss_rto << 10;
+               stat.send_tlrto_rate =
+                   (var * 100) / tcpstat.tcps_sndrexmitpack;
+       }
+
+       /* packet reordering */
+       if (tcpstat.tcps_sndpack > 0 && tcpstat.tcps_reordered_pkts > 0) {
+               var = tcpstat.tcps_reordered_pkts << 10;
+               stat.send_reorder_rate =
+                   (var * 100) / tcpstat.tcps_sndpack;
+       }
+
+       if (tcp_ecn_outbound == 1) {
+               stat.ecn_client_enabled = 1;
+       }
+       if (tcp_ecn_inbound == 1) {
+               stat.ecn_server_enabled = 1;
+       }
+       tcp_cumulative_stat(tcpstat.tcps_connattempt,
+           &prev.tcps_connattempt, &stat.connection_attempts);
+       tcp_cumulative_stat(tcpstat.tcps_accepts,
+           &prev.tcps_accepts, &stat.connection_accepts);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_client_setup,
+           &prev.tcps_ecn_client_setup, &stat.ecn_client_setup);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_server_setup,
+           &prev.tcps_ecn_server_setup, &stat.ecn_server_setup);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_client_success,
+           &prev.tcps_ecn_client_success, &stat.ecn_client_success);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_server_success,
+           &prev.tcps_ecn_server_success, &stat.ecn_server_success);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_not_supported,
+           &prev.tcps_ecn_not_supported, &stat.ecn_not_supported);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_lost_syn,
+           &prev.tcps_ecn_lost_syn, &stat.ecn_lost_syn);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_lost_synack,
+           &prev.tcps_ecn_lost_synack, &stat.ecn_lost_synack);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_recv_ce,
+           &prev.tcps_ecn_recv_ce, &stat.ecn_recv_ce);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_recv_ece,
+           &prev.tcps_ecn_recv_ece, &stat.ecn_recv_ece);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_recv_ece,
+           &prev.tcps_ecn_recv_ece, &stat.ecn_recv_ece);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_sent_ece,
+           &prev.tcps_ecn_sent_ece, &stat.ecn_sent_ece);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_sent_ece,
+           &prev.tcps_ecn_sent_ece, &stat.ecn_sent_ece);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_conn_recv_ce,
+           &prev.tcps_ecn_conn_recv_ce, &stat.ecn_conn_recv_ce);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_conn_recv_ece,
+           &prev.tcps_ecn_conn_recv_ece, &stat.ecn_conn_recv_ece);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_conn_plnoce,
+           &prev.tcps_ecn_conn_plnoce, &stat.ecn_conn_plnoce);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_conn_pl_ce,
+           &prev.tcps_ecn_conn_pl_ce, &stat.ecn_conn_pl_ce);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_conn_nopl_ce,
+           &prev.tcps_ecn_conn_nopl_ce, &stat.ecn_conn_nopl_ce);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_fallback_synloss,
+           &prev.tcps_ecn_fallback_synloss, &stat.ecn_fallback_synloss);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_fallback_reorder,
+           &prev.tcps_ecn_fallback_reorder, &stat.ecn_fallback_reorder);
+       tcp_cumulative_stat(tcpstat.tcps_ecn_fallback_ce,
+           &prev.tcps_ecn_fallback_ce, &stat.ecn_fallback_ce);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_syn_data_rcv,
+           &prev.tcps_tfo_syn_data_rcv, &stat.tfo_syn_data_rcv);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_cookie_req_rcv,
+           &prev.tcps_tfo_cookie_req_rcv, &stat.tfo_cookie_req_rcv);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_cookie_sent,
+           &prev.tcps_tfo_cookie_sent, &stat.tfo_cookie_sent);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_cookie_invalid,
+           &prev.tcps_tfo_cookie_invalid, &stat.tfo_cookie_invalid);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_cookie_req,
+           &prev.tcps_tfo_cookie_req, &stat.tfo_cookie_req);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_cookie_rcv,
+           &prev.tcps_tfo_cookie_rcv, &stat.tfo_cookie_rcv);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_syn_data_sent,
+           &prev.tcps_tfo_syn_data_sent, &stat.tfo_syn_data_sent);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_syn_data_acked,
+           &prev.tcps_tfo_syn_data_acked, &stat.tfo_syn_data_acked);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_syn_loss,
+           &prev.tcps_tfo_syn_loss, &stat.tfo_syn_loss);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_blackhole,
+           &prev.tcps_tfo_blackhole, &stat.tfo_blackhole);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_cookie_wrong,
+           &prev.tcps_tfo_cookie_wrong, &stat.tfo_cookie_wrong);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_no_cookie_rcv,
+           &prev.tcps_tfo_no_cookie_rcv, &stat.tfo_no_cookie_rcv);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_heuristics_disable,
+           &prev.tcps_tfo_heuristics_disable, &stat.tfo_heuristics_disable);
+       tcp_cumulative_stat(tcpstat.tcps_tfo_sndblackhole,
+           &prev.tcps_tfo_sndblackhole, &stat.tfo_sndblackhole);
+
+
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_handover_attempt,
+           &prev.tcps_mptcp_handover_attempt, &stat.mptcp_handover_attempt);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_interactive_attempt,
+           &prev.tcps_mptcp_interactive_attempt, &stat.mptcp_interactive_attempt);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_aggregate_attempt,
+           &prev.tcps_mptcp_aggregate_attempt, &stat.mptcp_aggregate_attempt);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_fp_handover_attempt,
+           &prev.tcps_mptcp_fp_handover_attempt, &stat.mptcp_fp_handover_attempt);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_fp_interactive_attempt,
+           &prev.tcps_mptcp_fp_interactive_attempt, &stat.mptcp_fp_interactive_attempt);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_fp_aggregate_attempt,
+           &prev.tcps_mptcp_fp_aggregate_attempt, &stat.mptcp_fp_aggregate_attempt);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_heuristic_fallback,
+           &prev.tcps_mptcp_heuristic_fallback, &stat.mptcp_heuristic_fallback);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_fp_heuristic_fallback,
+           &prev.tcps_mptcp_fp_heuristic_fallback, &stat.mptcp_fp_heuristic_fallback);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_handover_success_wifi,
+           &prev.tcps_mptcp_handover_success_wifi, &stat.mptcp_handover_success_wifi);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_handover_success_cell,
+           &prev.tcps_mptcp_handover_success_cell, &stat.mptcp_handover_success_cell);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_interactive_success,
+           &prev.tcps_mptcp_interactive_success, &stat.mptcp_interactive_success);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_aggregate_success,
+           &prev.tcps_mptcp_aggregate_success, &stat.mptcp_aggregate_success);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_fp_handover_success_wifi,
+           &prev.tcps_mptcp_fp_handover_success_wifi, &stat.mptcp_fp_handover_success_wifi);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_fp_handover_success_cell,
+           &prev.tcps_mptcp_fp_handover_success_cell, &stat.mptcp_fp_handover_success_cell);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_fp_interactive_success,
+           &prev.tcps_mptcp_fp_interactive_success, &stat.mptcp_fp_interactive_success);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_fp_aggregate_success,
+           &prev.tcps_mptcp_fp_aggregate_success, &stat.mptcp_fp_aggregate_success);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_handover_cell_from_wifi,
+           &prev.tcps_mptcp_handover_cell_from_wifi, &stat.mptcp_handover_cell_from_wifi);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_handover_wifi_from_cell,
+           &prev.tcps_mptcp_handover_wifi_from_cell, &stat.mptcp_handover_wifi_from_cell);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_interactive_cell_from_wifi,
+           &prev.tcps_mptcp_interactive_cell_from_wifi, &stat.mptcp_interactive_cell_from_wifi);
+       tcp_cumulative_stat64(tcpstat.tcps_mptcp_handover_cell_bytes,
+           &prev.tcps_mptcp_handover_cell_bytes, &stat.mptcp_handover_cell_bytes);
+       tcp_cumulative_stat64(tcpstat.tcps_mptcp_interactive_cell_bytes,
+           &prev.tcps_mptcp_interactive_cell_bytes, &stat.mptcp_interactive_cell_bytes);
+       tcp_cumulative_stat64(tcpstat.tcps_mptcp_aggregate_cell_bytes,
+           &prev.tcps_mptcp_aggregate_cell_bytes, &stat.mptcp_aggregate_cell_bytes);
+       tcp_cumulative_stat64(tcpstat.tcps_mptcp_handover_all_bytes,
+           &prev.tcps_mptcp_handover_all_bytes, &stat.mptcp_handover_all_bytes);
+       tcp_cumulative_stat64(tcpstat.tcps_mptcp_interactive_all_bytes,
+           &prev.tcps_mptcp_interactive_all_bytes, &stat.mptcp_interactive_all_bytes);
+       tcp_cumulative_stat64(tcpstat.tcps_mptcp_aggregate_all_bytes,
+           &prev.tcps_mptcp_aggregate_all_bytes, &stat.mptcp_aggregate_all_bytes);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_back_to_wifi,
+           &prev.tcps_mptcp_back_to_wifi, &stat.mptcp_back_to_wifi);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_wifi_proxy,
+           &prev.tcps_mptcp_wifi_proxy, &stat.mptcp_wifi_proxy);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_cell_proxy,
+           &prev.tcps_mptcp_cell_proxy, &stat.mptcp_cell_proxy);
+       tcp_cumulative_stat(tcpstat.tcps_mptcp_triggered_cell,
+           &prev.tcps_mptcp_triggered_cell, &stat.mptcp_triggered_cell);
+
+       nstat_sysinfo_send_data(&data);
+
+#undef  stat
+}
+
+void
+tcp_interface_send_probe(u_int16_t probe_if_index)
+{
+       int32_t offset = 0;
+       struct tcptimerlist *listp = &tcp_timer_list;
+
+       /* Make sure TCP clock is up to date */
+       calculate_tcp_clock();
+
+       lck_mtx_lock(listp->mtx);
+       if (listp->probe_if_index > 0) {
+               tcpstat.tcps_probe_if_conflict++;
+               goto done;
+       }
+
+       listp->probe_if_index = probe_if_index;
+       if (listp->running) {
+               goto done;
+       }
+
+       /*
+        * Reschedule the timerlist to run within the next 10ms, which is
+        * the fastest that we can do.
+        */
+       offset = TCP_TIMER_10MS_QUANTUM;
+       if (listp->scheduled) {
+               int32_t diff;
+               diff = timer_diff(listp->runtime, 0, tcp_now, offset);
+               if (diff <= 0) {
+                       /* The timer will fire sooner than what's needed */
+                       goto done;
+               }
+       }
+       listp->mode = TCP_TIMERLIST_10MS_MODE;
+       listp->idleruns = 0;
+
+       tcp_sched_timerlist(offset);
+
+done:
+       lck_mtx_unlock(listp->mtx);
+       return;
+}
+
+/*
+ * Enable read probes on this connection, if:
+ * - it is in established state
+ * - doesn't have any data outstanding
+ * - the outgoing ifp matches
+ * - we have not already sent any read probes
+ */
+static void
+tcp_enable_read_probe(struct tcpcb *tp, struct ifnet *ifp)
+{
+       if (tp->t_state == TCPS_ESTABLISHED &&
+           tp->snd_max == tp->snd_una &&
+           tp->t_inpcb->inp_last_outifp == ifp &&
+           !(tp->t_flagsext & TF_DETECT_READSTALL) &&
+           tp->t_rtimo_probes == 0) {
+               tp->t_flagsext |= TF_DETECT_READSTALL;
+               tp->t_rtimo_probes = 0;
+               tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp,
+                   TCP_TIMER_10MS_QUANTUM);
+               if (tp->tentry.index == TCPT_NONE) {
+                       tp->tentry.index = TCPT_KEEP;
+                       tp->tentry.runtime = tcp_now +
+                           TCP_TIMER_10MS_QUANTUM;
+               } else {
+                       int32_t diff = 0;
+
+                       /* Reset runtime to be in next 10ms */
+                       diff = timer_diff(tp->tentry.runtime, 0,
+                           tcp_now, TCP_TIMER_10MS_QUANTUM);
+                       if (diff > 0) {
+                               tp->tentry.index = TCPT_KEEP;
+                               tp->tentry.runtime = tcp_now +
+                                   TCP_TIMER_10MS_QUANTUM;
+                               if (tp->tentry.runtime == 0) {
+                                       tp->tentry.runtime++;
+                               }
+                       }
+               }
+       }
+}
+
+/*
+ * Disable read probe and reset the keep alive timer
+ */
+static void
+tcp_disable_read_probe(struct tcpcb *tp)
+{
+       if (tp->t_adaptive_rtimo == 0 &&
+           ((tp->t_flagsext & TF_DETECT_READSTALL) ||
+           tp->t_rtimo_probes > 0)) {
+               tcp_keepalive_reset(tp);
+
+               if (tp->t_mpsub) {
+                       mptcp_reset_keepalive(tp);
+               }
+       }
+}
+
+/*
+ * Reschedule the tcp timerlist in the next 10ms to re-enable read/write
+ * probes on connections going over a particular interface.
+ */
+void
+tcp_probe_connectivity(struct ifnet *ifp, u_int32_t enable)
+{
+       int32_t offset;
+       struct tcptimerlist *listp = &tcp_timer_list;
+       struct inpcbinfo *pcbinfo = &tcbinfo;
+       struct inpcb *inp, *nxt;
+
+       if (ifp == NULL) {
+               return;
+       }
+
+       /* update clock */
+       calculate_tcp_clock();
+
+       /*
+        * Enable keep alive timer on all connections that are
+        * active/established on this interface.
+        */
+       lck_rw_lock_shared(pcbinfo->ipi_lock);
+
+       LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, nxt) {
+               struct tcpcb *tp = NULL;
+               if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) ==
+                   WNT_STOPUSING) {
+                       continue;
+               }
+
+               /* Acquire lock to look at the state of the connection */
+               socket_lock(inp->inp_socket, 1);
+
+               /* Release the want count */
+               if (inp->inp_ppcb == NULL ||
+                   (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING)) {
+                       socket_unlock(inp->inp_socket, 1);
+                       continue;
+               }
+               tp = intotcpcb(inp);
+               if (enable) {
+                       tcp_enable_read_probe(tp, ifp);
+               } else {
+                       tcp_disable_read_probe(tp);
+               }
+
+               socket_unlock(inp->inp_socket, 1);
+       }
+       lck_rw_done(pcbinfo->ipi_lock);
+
+       lck_mtx_lock(listp->mtx);
+       if (listp->running) {
+               listp->pref_mode |= TCP_TIMERLIST_10MS_MODE;
+               goto done;
+       }
+
+       /* Reschedule within the next 10ms */
+       offset = TCP_TIMER_10MS_QUANTUM;
+       if (listp->scheduled) {
+               int32_t diff;
+               diff = timer_diff(listp->runtime, 0, tcp_now, offset);
+               if (diff <= 0) {
+                       /* The timer will fire sooner than what's needed */
+                       goto done;
+               }
+       }
+       listp->mode = TCP_TIMERLIST_10MS_MODE;
+       listp->idleruns = 0;
+
+       tcp_sched_timerlist(offset);
+done:
+       lck_mtx_unlock(listp->mtx);
+       return;
+}
+
+inline void
+tcp_update_mss_core(struct tcpcb *tp, struct ifnet *ifp)
+{
+       struct if_cellular_status_v1 *ifsr;
+       u_int32_t optlen;
+       ifsr = &ifp->if_link_status->ifsr_u.ifsr_cell.if_cell_u.if_status_v1;
+       if (ifsr->valid_bitmask & IF_CELL_UL_MSS_RECOMMENDED_VALID) {
+               optlen = tp->t_maxopd - tp->t_maxseg;
+
+               if (ifsr->mss_recommended ==
+                   IF_CELL_UL_MSS_RECOMMENDED_NONE &&
+                   tp->t_cached_maxopd > 0 &&
+                   tp->t_maxopd < tp->t_cached_maxopd) {
+                       tp->t_maxopd = tp->t_cached_maxopd;
+                       tcpstat.tcps_mss_to_default++;
+               } else if (ifsr->mss_recommended ==
+                   IF_CELL_UL_MSS_RECOMMENDED_MEDIUM &&
+                   tp->t_maxopd > tcp_mss_rec_medium) {
+                       tp->t_cached_maxopd = tp->t_maxopd;
+                       tp->t_maxopd = tcp_mss_rec_medium;
+                       tcpstat.tcps_mss_to_medium++;
+               } else if (ifsr->mss_recommended ==
+                   IF_CELL_UL_MSS_RECOMMENDED_LOW &&
+                   tp->t_maxopd > tcp_mss_rec_low) {
+                       tp->t_cached_maxopd = tp->t_maxopd;
+                       tp->t_maxopd = tcp_mss_rec_low;
+                       tcpstat.tcps_mss_to_low++;
+               }
+               tp->t_maxseg = tp->t_maxopd - optlen;
+
+               /*
+                * clear the cached value if it is same as the current
+                */
+               if (tp->t_maxopd == tp->t_cached_maxopd) {
+                       tp->t_cached_maxopd = 0;
+               }
+       }
+}
+
+void
+tcp_update_mss_locked(struct socket *so, struct ifnet *ifp)
+{
+       struct inpcb *inp = sotoinpcb(so);
+       struct tcpcb *tp = intotcpcb(inp);
+
+       if (ifp == NULL && (ifp = inp->inp_last_outifp) == NULL) {
+               return;
+       }
+
+       if (!IFNET_IS_CELLULAR(ifp)) {
+               /*
+                * This optimization is implemented for cellular
+                * networks only
+                */
+               return;
+       }
+       if (tp->t_state <= TCPS_CLOSE_WAIT) {
+               /*
+                * If the connection is currently doing or has done PMTU
+                * blackhole detection, do not change the MSS
+                */
+               if (tp->t_flags & TF_BLACKHOLE) {
+                       return;
+               }
+               if (ifp->if_link_status == NULL) {
+                       return;
+               }
+               tcp_update_mss_core(tp, ifp);
+       }
+}
+
+void
+tcp_itimer(struct inpcbinfo *ipi)
+{
+       struct inpcb *inp, *nxt;
+
+       if (lck_rw_try_lock_exclusive(ipi->ipi_lock) == FALSE) {
+               if (tcp_itimer_done == TRUE) {
+                       tcp_itimer_done = FALSE;
+                       atomic_add_32(&ipi->ipi_timer_req.intimer_fast, 1);
+                       return;
+               }
+               /* Upgrade failed, lost lock now take it again exclusive */
+               lck_rw_lock_exclusive(ipi->ipi_lock);
+       }
+       tcp_itimer_done = TRUE;
+
+       LIST_FOREACH_SAFE(inp, &tcb, inp_list, nxt) {
+               struct socket *so;
+               struct ifnet *ifp;
+
+               if (inp->inp_ppcb == NULL ||
+                   in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) {
+                       continue;
+               }
+               so = inp->inp_socket;
+               ifp = inp->inp_last_outifp;
+               socket_lock(so, 1);
+               if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
+                       socket_unlock(so, 1);
+                       continue;
+               }
+               so_check_extended_bk_idle_time(so);
+               if (ipi->ipi_flags & INPCBINFO_UPDATE_MSS) {
+                       tcp_update_mss_locked(so, NULL);
+               }
+               socket_unlock(so, 1);
+
+               /*
+                * Defunct all system-initiated background sockets if the
+                * socket is using the cellular interface and the interface
+                * has its LQM set to abort.
+                */
+               if ((ipi->ipi_flags & INPCBINFO_HANDLE_LQM_ABORT) &&
+                   IS_SO_TC_BACKGROUNDSYSTEM(so->so_traffic_class) &&
+                   ifp != NULL && IFNET_IS_CELLULAR(ifp) &&
+                   (ifp->if_interface_state.valid_bitmask &
+                   IF_INTERFACE_STATE_LQM_STATE_VALID) &&
+                   ifp->if_interface_state.lqm_state ==
+                   IFNET_LQM_THRESH_ABORT) {
+                       socket_defunct(current_proc(), so,
+                           SHUTDOWN_SOCKET_LEVEL_DISCONNECT_ALL);
+               }
+       }
+
+       ipi->ipi_flags &= ~(INPCBINFO_UPDATE_MSS | INPCBINFO_HANDLE_LQM_ABORT);
+       lck_rw_done(ipi->ipi_lock);
+}