/*
- * Copyright (c) 2000-2014 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2016 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
- *
+ *
* This file contains Original Code and/or Modifications of Original Code
* as defined in and that are subject to the Apple Public Source License
* Version 2.0 (the 'License'). You may not use this file except in
* unlawful or unlicensed copies of an Apple operating system, or to
* circumvent, violate, or enable the circumvention or violation of, any
* terms of an Apple operating system software license agreement.
- *
+ *
* Please obtain a copy of the License at
* http://www.opensource.apple.com/apsl/ and read it before using this file.
- *
+ *
* The Original Code and all software distributed under the License are
* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
* Please see the License for the specific language governing rights and
* limitations under the License.
- *
+ *
* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
#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 <mach/sdt.h>
#include <netinet/mptcp_var.h>
-#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)
-
/* Max number of times a stretch ack can be delayed on a connection */
#define TCP_STRETCHACK_DELAY_THRESHOLD 5
+/*
+ * 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 */
+
/* tcp timer list */
struct tcptimerlist tcp_timer_list;
CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
&tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
-/*
+/*
* 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
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",
+ &tcp_max_persist_timeout, 0, sysctl_msec_to_ticks, "I",
"Maximum persistence timeout for ZWP");
static int always_keepalive = 0;
/*
* 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
+ * 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_IDLERUN_MAX;
* 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");
-
-/* A higher threshold on local connections for disabling RFC 1323 options */
-static int tcp_broken_peer_syn_rxmit_thres_local = 10;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, broken_peer_syn_rexmit_thres_local,
- CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_broken_peer_syn_rxmit_thres_local, 0,
+static int tcp_broken_peer_syn_rxmit_thres = 10;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, broken_peer_syn_rexmit_thres,
+ CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_broken_peer_syn_rxmit_thres, 0,
"Number of retransmitted SYNs before disabling RFC 1323 "
"options on local connections");
static int tcp_resched_timerlist = 0;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_resched_timerlist,
- CTLFLAG_RD | CTLFLAG_LOCKED, &tcp_resched_timerlist, 0,
+ CTLFLAG_RD | CTLFLAG_LOCKED, &tcp_resched_timerlist, 0,
"Number of times timer list was rescheduled as part of processing a packet");
int tcp_pmtud_black_hole_detect = 1 ;
CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_pmtud_black_hole_mss, 0,
"Path MTU Discovery Black Hole Detection lowered MSS");
+static u_int32_t tcp_mss_rec_medium = 1200;
+static u_int32_t tcp_mss_rec_low = 512;
+
+#define TCP_REPORT_STATS_INTERVAL 43200 /* 12 hours, in seconds */
+int tcp_report_stats_interval = TCP_REPORT_STATS_INTERVAL;
+
/* performed garbage collection of "used" sockets */
static boolean_t tcp_gc_done = FALSE;
int tcp_jack_rxmt = TCP_RETRANSHZ / 2;
#endif /* MPTCP */
+static boolean_t tcp_itimer_done = FALSE;
+
static void tcp_remove_timer(struct tcpcb *tp);
static void tcp_sched_timerlist(uint32_t offset);
-static u_int32_t tcp_run_conn_timer(struct tcpcb *tp, u_int16_t *mode);
+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_rexmt_save_state(struct tcpcb *tp);
__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);
+static u_int64_t tcp_last_report_time;
+
/*
- * 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.
+ * Structure to store previously reported stats so that we can send
+ * incremental changes in each report interval.
*/
-inline int32_t
-timer_diff(uint32_t t1, uint32_t toff1, uint32_t t2, uint32_t toff2) {
- return (int32_t)((t1 + toff1) - (t2 + toff2));
+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;
};
-static u_int64_t tcp_last_report_time;
-#define TCP_REPORT_STATS_INTERVAL 345600 /* 4 days, in seconds */
/* Returns true if the timer is on the timer list */
#define TIMER_IS_ON_LIST(tp) ((tp)->t_flags & TF_TIMER_ONLIST)
static void add_to_time_wait_locked(struct tcpcb *tp, uint32_t delay);
static boolean_t tcp_garbage_collect(struct inpcb *, int);
+#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
+{
+#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");
+
+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) */
+
+/*
+ * 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.
*/
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 {
/* Compute the time at which this socket can be closed */
timer = tcp_now + delay;
-
+
/* We will use the TCPT_2MSL timer for tracking this delay */
if (TIMER_IS_ON_LIST(tp))
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)) {
tcp_unlock(tp->t_inpcb->inp_socket, 0, 0);
lck_rw_lock_exclusive(pcbinfo->ipi_lock);
}
/*
- * 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
+ * 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
+ * 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().
*/
lck_mtx_convert_spin(&inp->inpcb_mtx);
/*
- * If this tp still happens to be on the timer list,
+ * If this tp still happens to be on the timer list,
* take it out
*/
if (TIMER_IS_ON_LIST(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
+ * 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.
+ * usecount on it.
*/
- if (tw_tp->t_state == TCPS_CLOSED ||
+ 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);
* 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);
for (i = 0; i < TCPT_NTIMERS; i++)
static int tcp_totbackoff = 511; /* sum of tcp_backoff[] */
-static void tcp_rexmt_save_state(struct tcpcb *tp)
+void
+tcp_rexmt_save_state(struct tcpcb *tp)
{
u_int32_t fsize;
if (TSTMP_SUPPORTED(tp)) {
/*
- * Since timestamps are supported on the connection,
+ * Since timestamps are supported on the connection,
* we can do recovery as described in rfc 4015.
*/
fsize = tp->snd_max - tp->snd_una;
* 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
+ * 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
+ * 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.
*/
* 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)
+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;
+ 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
+ * Reset the slow-start flight size as it
* may depend on the new MSS
*/
if (CC_ALGO(tp)->cwnd_init != NULL)
* TCP timer processing.
*/
struct tcpcb *
-tcp_timers(tp, timer)
- register struct tcpcb *tp;
- int timer;
+tcp_timers(struct tcpcb *tp, int timer)
{
int32_t rexmt, optlen = 0, idle_time = 0;
struct socket *so;
#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 = tp->t_inpcb->inp_socket;
idle_time = tcp_now - tp->t_rcvtime;
if (tp->t_state != TCPS_TIME_WAIT &&
tp->t_state != TCPS_FIN_WAIT_2 &&
((idle_time > 0) && (idle_time < TCP_CONN_MAXIDLE(tp)))) {
- tp->t_timer[TCPT_2MSL] = OFFSET_FROM_START(tp,
+ tp->t_timer[TCPT_2MSL] = OFFSET_FROM_START(tp,
(u_int32_t)TCP_CONN_KEEPINTVL(tp));
} else {
tp = tcp_close(tp);
* to a longer retransmit interval and retransmit one segment.
*/
case TCPT_REXMT:
+ 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
* receiving an ack
*/
if (++tp->t_rxtshift > TCP_MAXRXTSHIFT ||
- (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_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_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++;
}
+ 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;
- postevent(so, 0, EV_TIMEOUT);
- soevent(so,
+ 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);
}
tcpstat.tcps_rexmttimeo++;
+ tp->t_accsleep_ms = accsleep_ms;
- if (tp->t_rxtshift == 1 &&
+ 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
tp->t_flagsext &= ~(TF_DELAY_RECOVERY);
}
+ if (tp->t_state == TCPS_SYN_RECEIVED)
+ tcp_disable_tfo(tp);
+
+ if ((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 > 2)) {
+ /*
+ * 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.
+ */
+ 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++;
+ }
+
if (tp->t_state == TCPS_SYN_SENT) {
rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
tp->t_stat.synrxtshift = tp->t_rxtshift;
+
+ /* When retransmitting, disable TFO */
+ if (tfo_enabled(tp)) {
+ tp->t_flagsext &= ~TF_FASTOPEN;
+ tp->t_tfo_flags |= TFO_F_SYN_LOSS;
+ }
} else {
rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
}
- TCPT_RANGESET(tp->t_rxtcur, rexmt,
- tp->t_rttmin, TCPTV_REXMTMAX,
+ 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);
if (tcp_pmtud_black_hole_detect &&
!(tp->t_flagsext & TF_NOBLACKHOLE_DETECTION) &&
(tp->t_state == TCPS_ESTABLISHED)) {
- if (((tp->t_flags & (TF_PMTUD|TF_MAXSEGSNT))
- == (TF_PMTUD|TF_MAXSEGSNT)) &&
- (tp->t_rxtshift == 2)) {
- /*
+ 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).
* - Reduce MTU to lower value than what we
tp->t_maxseg = tp->t_maxopd - optlen;
/*
- * Reset the slow-start flight size
+ * 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->snd_cwnd = tp->t_maxseg;
}
/*
* If further retransmissions are still
* MSS and blackhole detection flags.
*/
else {
-
+
if ((tp->t_flags & TF_BLACKHOLE) &&
(tp->t_rxtshift > 4)) {
tcp_pmtud_revert_segment_size(tp);
+ tp->snd_cwnd = tp->t_maxseg;
}
}
}
* Do this only on non-local connections.
*/
if (tp->t_state == TCPS_SYN_SENT &&
- ((!(tp->t_flags & TF_LOCAL) &&
- tp->t_rxtshift == tcp_broken_peer_syn_rxmit_thres) ||
- ((tp->t_flags & TF_LOCAL) &&
- tp->t_rxtshift == tcp_broken_peer_syn_rxmit_thres_local)))
+ tp->t_rxtshift == tcp_broken_peer_syn_rxmit_thres)
tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
/*
* 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)
+ if (tp->t_rxtshift == 1 &&
+ CC_ALGO(tp)->after_timeout != NULL) {
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;
+ }
EXIT_FASTRECOVERY(tp);
- /* CWR notifications are to be sent on new data right after
- * RTOs, Fast Retransmits and ECE notification receipts.
- */
- if ((tp->ecn_flags & TE_ECN_ON) == TE_ECN_ON) {
- tp->ecn_flags |= TE_SENDCWR;
- }
+ /* Exit cwnd non validated phase */
+ tp->t_flagsext &= ~TF_CWND_NONVALIDATED;
+
+
fc_output:
tcp_ccdbg_trace(tp, NULL, TCP_CC_REXMT_TIMEOUT);
* 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 ||
- idle_time >= TCP_REXMTVAL(tp) * tcp_totbackoff)) ||
- ((tp->t_persist_stop != 0) &&
+ idle_time >= TCP_REXMTVAL(tp) * tcp_totbackoff)) ||
+ ((tp->t_persist_stop != 0) &&
TSTMP_LEQ(tp->t_persist_stop, tcp_now))) {
tcpstat.tcps_persistdrop++;
postevent(so, 0, EV_TIMEOUT);
goto dropit;
if ((always_keepalive ||
(tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) ||
- (tp->t_flagsext & TF_DETECT_READSTALL)) &&
+ (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)) {
if (idle_time >= TCP_CONN_KEEPIDLE(tp) + TCP_CONN_MAXIDLE(tp))
goto dropit;
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
tp->t_rtimo_probes++;
}
tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp,
- TCP_CONN_KEEPINTVL(tp));
+ TCP_CONN_KEEPINTVL(tp));
} else {
tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp,
- TCP_CONN_KEEPIDLE(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
+ * 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_rtimo_probes > tp->t_adaptive_rtimo) {
- /* Generate an event */
- soevent(so,
- (SO_FILT_HINT_LOCKED|
- SO_FILT_HINT_ADAPTIVE_RTIMO));
- tcp_keepalive_reset(tp);
+ 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_REXMTVAL(tp));
+ 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_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)) {
/*
* If delayed ack timer fired while stretching
* acks, count the number of times the streaming
- * detection was not correct. If this exceeds a
+ * detection was not correct. If this exceeds a
* threshold, disable strech ack on this
* connection
*
tp->rcv_nostrack_ts = tcp_now;
tcpstat.tcps_nostretchack++;
tp->t_stretchack_delayed = 0;
+ tp->rcv_nostrack_pkts = 0;
}
tcp_reset_stretch_ack(tp);
}
if (++tp->t_mprxtshift > TCP_MAXRXTSHIFT) {
tcpstat.tcps_timeoutdrop++;
postevent(so, 0, EV_TIMEOUT);
- soevent(so,
+ soevent(so,
(SO_FILT_HINT_LOCKED|
SO_FILT_HINT_TIMEOUT));
tp = tcp_drop(tp, tp->t_softerror ?
tp->t_flags |= TF_ACKNOW;
/*
- * No backoff is implemented for simplicity for this
+ * No backoff is implemented for simplicity for this
* corner case.
*/
(void) tcp_output(tp);
case TCPT_PTO:
{
- tcp_seq old_snd_nxt;
int32_t snd_len;
- boolean_t rescue_rxt = FALSE;
-
tp->t_flagsext &= ~(TF_SENT_TLPROBE);
/*
* send a probe
*/
if (tp->t_state != TCPS_ESTABLISHED ||
- tp->t_rxtshift > 0 || tp->snd_max == tp->snd_una ||
- !SACK_ENABLED(tp) || TAILQ_EMPTY(&tp->snd_holes) ||
- (IN_FASTRECOVERY(tp) &&
- (SEQ_GEQ(tp->snd_fack, tp->snd_recover) ||
- SEQ_GT(tp->snd_nxt, tp->sack_newdata))))
+ (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;
-
- if (IN_FASTRECOVERY(tp)) {
- /*
- * Send a probe to detect tail loss in a
- * recovery window when the connection is in
- * fast_recovery.
- */
- old_snd_nxt = tp->snd_nxt;
- rescue_rxt = TRUE;
- VERIFY(SEQ_GEQ(tp->snd_fack, tp->snd_una));
- snd_len = min((tp->snd_recover - tp->snd_fack),
- tp->t_maxseg);
- tp->snd_nxt = tp->snd_recover - snd_len;
- tcpstat.tcps_pto_in_recovery++;
- tcp_ccdbg_trace(tp, NULL, TCP_CC_TLP_IN_FASTRECOVERY);
- } else {
- /*
- * 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;
- }
- }
-
/* 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;
tp->t_tlphighrxt = tp->snd_nxt;
-
- /*
- * If a tail loss probe was sent after entering recovery,
- * restore the old snd_nxt value so that other packets
- * will get retransmitted correctly.
- */
- if (rescue_rxt)
- tp->snd_nxt = old_snd_nxt;
break;
}
case TCPT_DELAYFR:
break;
VERIFY(SACK_ENABLED(tp));
- if (CC_ALGO(tp)->pre_fr != NULL)
+ 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);
- if ((tp->ecn_flags & TE_ECN_ON) == TE_ECN_ON)
- tp->ecn_flags |= TE_SENDCWR;
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);
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;
}
-
+
if (listp->next_te != NULL && listp->next_te == &tp->tentry)
listp->next_te = LIST_NEXT(&tp->tentry, le);
}
void
-tcp_sched_timerlist(uint32_t offset)
+tcp_sched_timerlist(uint32_t offset)
{
-
uint64_t deadline = 0;
struct tcptimerlist *listp = &tcp_timer_list;
/*
* 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.
*
* te_mode is an out parameter that indicates the modes of active
* timers for this connection.
*/
u_int32_t
-tcp_run_conn_timer(struct tcpcb *tp, u_int16_t *te_mode) {
-
+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;
tcp_lock(tp->t_inpcb->inp_socket, 1, 0);
so = tp->t_inpcb->inp_socket;
- /* Release the want count on inp */
+ /* 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);
}
- /* 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;
}
+ /*
+ * 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;
+ }
+
/*
* Since the timer thread needs to wait for tcp lock, it may race
* with another thread that can cancel or reschedule the timer
*/
if ((index = tp->tentry.index) == TCPT_NONE)
goto done;
-
+
timer_val = tp->t_timer[index];
diff = timer_diff(tp->tentry.runtime, 0, tcp_now, 0);
if (tp == NULL)
goto done;
}
-
+
/*
* Check if there are any other timers that need to be run.
* While doing it, adjust the timer values wrt tcp_now.
}
}
}
-
+
tp->tentry.timer_start = tcp_now;
tp->tentry.index = lo_index;
VERIFY(tp->tentry.index == TCPT_NONE || tp->tentry.mode > 0);
}
void
-tcp_run_timerlist(void * arg1, void * arg2) {
+tcp_run_timerlist(void * arg1, void * arg2)
+{
#pragma unused(arg1, arg2)
struct tcptimerentry *te, *next_te;
struct tcptimerlist *listp = &tcp_timer_list;
lck_mtx_lock(listp->mtx);
listp->running = TRUE;
-
+
LIST_FOREACH_SAFE(te, &listp->lhead, le, next_te) {
uint32_t offset = 0;
uint32_t runtime = te->runtime;
* 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
+ * protected by the timer list lock, we can
* do it here without the socket lock.
*/
if (TIMER_IS_ON_LIST(tp)) {
* 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);
- offset = tcp_run_conn_timer(tp, &te_mode);
-
+ offset = tcp_run_conn_timer(tp, &te_mode,
+ listp->probe_if_index);
+
lck_mtx_lock(listp->mtx);
next_te = listp->next_te;
listp->running = FALSE;
listp->pref_mode = 0;
listp->pref_offset = 0;
+ listp->probe_if_index = 0;
lck_mtx_unlock(listp->mtx);
}
/*
- * Function to check if the timerlist needs to be reschduled to run this
+ * Function to check if the timerlist needs to be rescheduled to run this
* connection's timers correctly.
*/
-void
-tcp_sched_timers(struct tcpcb *tp)
+void
+tcp_sched_timers(struct tcpcb *tp)
{
struct tcptimerentry *te = &tp->tentry;
u_int16_t index = te->index;
*/
if (need_to_resched_timerlist(te->runtime, mode)) {
tcp_resched_timerlist++;
-
+
if (!list_locked) {
lck_mtx_lock(listp->mtx);
list_locked = TRUE;
return;
}
-
+
static inline void
-tcp_set_lotimer_index(struct tcpcb *tp) {
+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) {
VERIFY(tp->tentry.index == TCPT_NONE || tp->tentry.mode > 0);
if (tp->tentry.index != TCPT_NONE) {
- tp->tentry.runtime = tp->tentry.timer_start
+ tp->tentry.runtime = tp->tentry.timer_start
+ tp->t_timer[tp->tentry.index];
if (tp->tentry.runtime == 0)
tp->tentry.runtime++;
}
void
-tcp_check_timer_state(struct tcpcb *tp) {
-
+tcp_check_timer_state(struct tcpcb *tp)
+{
lck_mtx_assert(&tp->t_inpcb->inpcb_mtx, LCK_MTX_ASSERT_OWNED);
if (tp->t_inpcb->inp_flags2 & INP2_TIMEWAIT)
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;
+}
+
__private_extern__ void
tcp_report_stats(void)
{
struct sockaddr_in dst;
struct sockaddr_in6 dst6;
struct rtentry *rt = NULL;
- u_int64_t var, uptime;
+ 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)
+ tcp_report_stats_interval)
return;
tcp_last_report_time = uptime;
}
/* RTO after tail loss, shift by 10 for precision */
- if (tcpstat.tcps_sndrexmitpack > 0
+ 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;
(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);
+
+
+
+
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);
+ }
+}
+
+/*
+ * 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 */
+ tcp_lock(inp->inp_socket, 1, 0);
+
+ /* Release the want count */
+ if (inp->inp_ppcb == NULL ||
+ (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING)) {
+ tcp_unlock(inp->inp_socket, 1, 0);
+ continue;
+ }
+ tp = intotcpcb(inp);
+ if (enable)
+ tcp_enable_read_probe(tp, ifp);
+ else
+ tcp_disable_read_probe(tp);
+
+ tcp_unlock(inp->inp_socket, 1, 0);
+ }
+ 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 && inp->inp_last_outifp == NULL)
+ return;
+
+ if (ifp == NULL)
+ ifp = inp->inp_last_outifp;
+
+ 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;
+
+ if (inp->inp_ppcb == NULL ||
+ in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING)
+ continue;
+ so = inp->inp_socket;
+ tcp_lock(so, 1, 0);
+ if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
+ tcp_unlock(so, 1, 0);
+ continue;
+ }
+ so_check_extended_bk_idle_time(so);
+ if (ipi->ipi_flags & INPCBINFO_UPDATE_MSS) {
+ tcp_update_mss_locked(so, NULL);
+ }
+ tcp_unlock(so, 1, 0);
+ }
+
+ ipi->ipi_flags &= ~INPCBINFO_UPDATE_MSS;
+ lck_rw_done(ipi->ipi_lock);
+}
projects / apple / xnu.git / blobdiff