X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/813fb2f63a553c957e917ede5f119b021d6ce391..0a7de7458d150b5d4dffc935ba399be265ef0a1a:/bsd/netinet/tcp_timer.c diff --git a/bsd/netinet/tcp_timer.c b/bsd/netinet/tcp_timer.c index 5c87a2d18..fda0f86f6 100644 --- a/bsd/netinet/tcp_timer.c +++ b/bsd/netinet/tcp_timer.c @@ -74,7 +74,7 @@ #include #include #include -#include /* before tcp_seq.h, for tcp_random18() */ +#include /* before tcp_seq.h, for tcp_random18() */ #include #include @@ -107,13 +107,13 @@ #include /* Max number of times a stretch ack can be delayed on a connection */ -#define TCP_STRETCHACK_DELAY_THRESHOLD 5 +#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 */ +#define TCP_SLEEP_TOO_LONG (10 * 60 * 1000) /* 10 minutes in ms */ /* tcp timer list */ struct tcptimerlist tcp_timer_list; @@ -124,48 +124,81 @@ struct tcptailq tcp_tw_tailq; static int sysctl_msec_to_ticks SYSCTL_HANDLER_ARGS { -#pragma unused(arg1, arg2) +#pragma unused(arg2) 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); - if (error || !req->newptr) - return (error); + if (error || !req->newptr) { + return error; + } 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; +#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", ""); -int tcp_keepidle; +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", ""); -int tcp_keepintvl; +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", ""); -int tcp_keepcnt; +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; +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"); +#endif /* SYSCTL_SKMEM */ /* * Avoid DoS via TCP Robustness in Persist Condition @@ -176,26 +209,32 @@ SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, * 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; +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 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_IDLERUN_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; @@ -204,10 +243,9 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_fastmode_idlemax, * SYN retransmits. Setting it to 0 disables the dropping off of those * two options. */ -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 " +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; @@ -220,40 +258,38 @@ 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"); -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, +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, +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 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 */ +#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; /* max idle probes */ -int tcp_maxpersistidle; +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; +int tcp_delack = TCP_RETRANSHZ / 10; #if MPTCP /* * MP_JOIN retransmission of 3rd ACK will be every 500 msecs without backoff */ -int tcp_jack_rxmt = TCP_RETRANSHZ / 2; +int tcp_jack_rxmt = TCP_RETRANSHZ / 2; #endif /* MPTCP */ static boolean_t tcp_itimer_done = FALSE; @@ -268,49 +304,80 @@ __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; +static u_int64_t tcp_last_report_time; /* * 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; + 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; + 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; }; @@ -326,25 +393,25 @@ 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)); \ +#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)); \ +#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; \ + (mode) |= TCP_TIMERLIST_10MS_MODE; \ else if (IS_TIMER_HZ_100MS(i)) \ - (mode) |= TCP_TIMERLIST_100MS_MODE; \ + (mode) |= TCP_TIMERLIST_100MS_MODE; \ else \ - (mode) |= TCP_TIMERLIST_500MS_MODE; \ + (mode) |= TCP_TIMERLIST_500MS_MODE; \ } while(0) #if (DEVELOPMENT || DEBUG) @@ -365,12 +432,12 @@ sysctl_change_mss_recommended SYSCTL_HANDLER_ARGS 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); + 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)); + 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); @@ -378,7 +445,7 @@ sysctl_change_mss_recommended SYSCTL_HANDLER_ARGS /* 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)); + err = ifnet_link_status_report(ifp, new_cell_sr, sizeof(new_cell_sr)); if (err == 0) { tcp_change_mss_recommended = i; } else { @@ -388,7 +455,7 @@ sysctl_change_mss_recommended SYSCTL_HANDLER_ARGS } ifnet_head_done(); } - return (err); + return err; } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, change_mss_recommended, @@ -407,7 +474,8 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, report_stats_interval, * 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) { +timer_diff(uint32_t t1, uint32_t toff1, uint32_t t2, uint32_t toff2) +{ return (int32_t)((t1 + toff1) - (t2 + toff2)); }; @@ -422,7 +490,7 @@ add_to_time_wait_locked(struct tcpcb *tp, uint32_t delay) uint32_t timer; /* pcb list should be locked when we get here */ - lck_rw_assert(pcbinfo->ipi_lock, LCK_RW_ASSERT_EXCLUSIVE); + LCK_RW_ASSERT(pcbinfo->ipi_lock, LCK_RW_ASSERT_EXCLUSIVE); /* We may get here multiple times, so check */ if (!(inp->inp_flags2 & INP2_TIMEWAIT)) { @@ -440,8 +508,9 @@ add_to_time_wait_locked(struct tcpcb *tp, uint32_t delay) /* We will use the TCPT_2MSL timer for tracking this delay */ - if (TIMER_IS_ON_LIST(tp)) + if (TIMER_IS_ON_LIST(tp)) { tcp_remove_timer(tp); + } tp->t_timer[TCPT_2MSL] = timer; TAILQ_INSERT_TAIL(&tcp_tw_tailq, tp, t_twentry); @@ -451,16 +520,17 @@ void add_to_time_wait(struct tcpcb *tp, uint32_t delay) { struct inpcbinfo *pcbinfo = &tcbinfo; - if (tp->t_inpcb->inp_socket->so_options & SO_NOWAKEFROMSLEEP) + 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); + socket_unlock(tp->t_inpcb->inp_socket, 0); lck_rw_lock_exclusive(pcbinfo->ipi_lock); - tcp_lock(tp->t_inpcb->inp_socket, 0, 0); + socket_lock(tp->t_inpcb->inp_socket, 0); } add_to_time_wait_locked(tp, delay); lck_rw_done(pcbinfo->ipi_lock); @@ -473,20 +543,31 @@ void tcp_remove_from_time_wait(struct inpcb *inp) { struct tcpcb *tp = intotcpcb(inp); - if (inp->inp_flags2 & INP2_TIMEWAIT) + 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) { boolean_t active = FALSE; - struct socket *so; + struct socket *so, *mp_so = NULL; 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 @@ -494,20 +575,22 @@ tcp_garbage_collect(struct inpcb *inp, int istimewait) * overflow sockets that are eligible for garbage collection have * their usecounts set to 1. */ - if (!lck_mtx_try_lock_spin(&inp->inpcb_mtx)) - return (TRUE); + if (!lck_mtx_try_lock_spin(&inp->inpcb_mtx)) { + active = TRUE; + goto out; + } /* Check again under the lock */ if (so->so_usecount > 1) { - if (inp->inp_wantcnt == WNT_STOPUSING) + if (inp->inp_wantcnt == WNT_STOPUSING) { active = TRUE; + } lck_mtx_unlock(&inp->inpcb_mtx); - return (active); + goto out; } - if (istimewait && - TSTMP_GEQ(tcp_now, tp->t_timer[TCPT_2MSL]) && - tp->t_state != TCPS_CLOSED) { + 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); @@ -526,28 +609,29 @@ tcp_garbage_collect(struct inpcb *inp, int istimewait) ((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))))) { - + ((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 (SOCK_CHECK_DOM(so, PF_INET6)) + if (SOCK_CHECK_DOM(so, PF_INET6)) { in6_pcbdetach(inp); - else + } else #endif /* INET6 */ - in_pcbdetach(inp); + in_pcbdetach(inp); } VERIFY(so->so_usecount > 0); so->so_usecount--; - if (inp->inp_wantcnt == WNT_STOPUSING) + if (inp->inp_wantcnt == WNT_STOPUSING) { active = TRUE; + } lck_mtx_unlock(&inp->inpcb_mtx); - return (active); + goto out; } else if (inp->inp_wantcnt != WNT_STOPUSING) { lck_mtx_unlock(&inp->inpcb_mtx); - return (FALSE); + active = FALSE; + goto out; } /* @@ -563,7 +647,7 @@ tcp_garbage_collect(struct inpcb *inp, int istimewait) */ 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); @@ -577,18 +661,35 @@ tcp_garbage_collect(struct inpcb *inp, int istimewait) if (inp->inp_state != INPCB_STATE_DEAD) { #if INET6 - if (SOCK_CHECK_DOM(so, PF_INET6)) + if (SOCK_CHECK_DOM(so, PF_INET6)) { in6_pcbdetach(inp); - else + } else #endif /* INET6 */ - in_pcbdetach(inp); + 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); - return (FALSE); + active = FALSE; + goto out; } lck_mtx_unlock(&inp->inpcb_mtx); - return (TRUE); + active = TRUE; + +out: + if (mp_so) { + socket_unlock(mp_so, 1); + } + + return active; } /* @@ -638,8 +739,9 @@ tcp_gc(struct inpcbinfo *ipi) tcp_gc_done = TRUE; LIST_FOREACH_SAFE(inp, &tcb, inp_list, nxt) { - if (tcp_garbage_collect(inp, 0)) + if (tcp_garbage_collect(inp, 0)) { atomic_add_32(&ipi->ipi_gc_req.intimer_fast, 1); + } } /* Now cleanup the time wait ones */ @@ -656,8 +758,9 @@ tcp_gc(struct inpcbinfo *ipi) */ 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)) + if (tcp_garbage_collect(tw_tp->t_inpcb, 1)) { atomic_add_32(&ipi->ipi_gc_req.intimer_lazy, 1); + } } } @@ -667,8 +770,9 @@ tcp_gc(struct inpcbinfo *ipi) lck_rw_done(ipi->ipi_lock); /* Clean up the socache while we are here */ - if (so_cache_timer()) + if (so_cache_timer()) { atomic_add_32(&ipi->ipi_gc_req.intimer_lazy, 1); + } KERNEL_DEBUG(DBG_FNC_TCP_SLOW | DBG_FUNC_END, tws_checked, cur_tw_slot, 0, 0, 0); @@ -685,19 +789,20 @@ tcp_canceltimers(struct tcpcb *tp) int i; tcp_remove_timer(tp); - for (i = 0; i < TCPT_NTIMERS; i++) + for (i = 0; i < TCPT_NTIMERS; i++) { tp->t_timer[i] = 0; + } 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 }; +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[] */ +static int tcp_totbackoff = 511; /* sum of tcp_backoff[] */ void tcp_rexmt_save_state(struct tcpcb *tp) @@ -725,10 +830,11 @@ tcp_rexmt_save_state(struct tcpcb *tp) 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)) + if (IN_FASTRECOVERY(tp)) { tp->t_flags |= TF_WASFRECOVERY; - else + } 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); @@ -750,14 +856,19 @@ tcp_pmtud_revert_segment_size(struct tcpcb *tp) 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 depend on the new MSS */ - if (CC_ALGO(tp)->cwnd_init != NULL) + if (CC_ALGO(tp)->cwnd_init != NULL) { CC_ALGO(tp)->cwnd_init(tp); + } tp->t_pmtud_start_ts = 0; tcpstat.tcps_pmtudbh_reverted++; + + /* change MSS according to recommendation, if there was one */ + tcp_update_mss_locked(tp->t_inpcb->inp_socket, NULL); } /* @@ -783,7 +894,6 @@ tcp_timers(struct tcpcb *tp, int 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 @@ -798,10 +908,10 @@ tcp_timers(struct tcpcb *tp, int timer) 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, - (u_int32_t)TCP_CONN_KEEPINTVL(tp)); + (u_int32_t)TCP_CONN_KEEPINTVL(tp)); } else { tp = tcp_close(tp); - return(tp); + return tp; } break; @@ -814,8 +924,9 @@ tcp_timers(struct tcpcb *tp, int timer) absolutetime_to_nanoseconds(mach_absolutetime_asleep, &accsleep_ms); accsleep_ms = accsleep_ms / 1000000UL; - if (accsleep_ms > tp->t_accsleep_ms) + 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 @@ -833,6 +944,20 @@ tcp_timers(struct tcpcb *tp, int timer) ((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++; } else if (last_sleep_ms >= TCP_SLEEP_TOO_LONG) { @@ -852,23 +977,24 @@ tcp_timers(struct tcpcb *tp, int timer) tp->t_rxtshift = TCP_MAXRXTSHIFT; postevent(so, 0, EV_TIMEOUT); soevent(so, - (SO_FILT_HINT_LOCKED|SO_FILT_HINT_TIMEOUT)); + (SO_FILT_HINT_LOCKED | SO_FILT_HINT_TIMEOUT)); if (TCP_ECN_ENABLED(tp) && - tp->t_state == TCPS_ESTABLISHED) + tp->t_state == TCPS_ESTABLISHED) { tcp_heuristic_ecn_droprxmt(tp); + } tp = tcp_drop(tp, tp->t_softerror ? tp->t_softerror : ETIMEDOUT); break; } - +retransmit_packet: tcpstat.tcps_rexmttimeo++; tp->t_accsleep_ms = accsleep_ms; if (tp->t_rxtshift == 1 && - tp->t_state == TCPS_ESTABLISHED) { + tp->t_state == TCPS_ESTABLISHED) { /* Set the time at which retransmission started. */ tp->t_rxtstart = tcp_now; @@ -884,17 +1010,22 @@ tcp_timers(struct tcpcb *tp, int timer) (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)) { + 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)); + (SO_FILT_HINT_LOCKED | + SO_FILT_HINT_ADAPTIVE_WTIMO)); } /* @@ -918,13 +1049,15 @@ tcp_timers(struct tcpcb *tp, int timer) tp->t_flagsext &= ~(TF_DELAY_RECOVERY); } - if (tp->t_state == TCPS_SYN_RECEIVED) + if (tp->t_state == TCPS_SYN_RECEIVED) { tcp_disable_tfo(tp); + } - if ((tp->t_tfo_stats & TFO_S_SYN_DATA_SENT) && + 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 > 2)) { + tp->t_rxtshift > 4)) { /* * For regular retransmissions, a first one is being * done for tail-loss probe. @@ -944,12 +1077,26 @@ tcp_timers(struct tcpcb *tp, int timer) tcpstat.tcps_tfo_sndblackhole++; } + 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]; tp->t_stat.synrxtshift = tp->t_rxtshift; /* When retransmitting, disable TFO */ - if (tfo_enabled(tp) && !(so->so_flags & SOF1_DATA_AUTHENTICATED)) { + 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; } @@ -958,19 +1105,20 @@ tcp_timers(struct tcpcb *tp, int timer) } TCPT_RANGESET(tp->t_rxtcur, rexmt, tp->t_rttmin, TCPTV_REXMTMAX, - TCP_ADD_REXMTSLOP(tp)); + TCP_ADD_REXMTSLOP(tp)); tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp, tp->t_rxtcur); - if (INP_WAIT_FOR_IF_FEEDBACK(tp->t_inpcb)) + if (INP_WAIT_FOR_IF_FEEDBACK(tp->t_inpcb)) { goto fc_output; + } tcp_free_sackholes(tp); /* * Check for potential Path MTU Discovery Black Hole */ if (tcp_pmtud_black_hole_detect && - !(tp->t_flagsext & TF_NOBLACKHOLE_DETECTION) && - (tp->t_state == TCPS_ESTABLISHED)) { + !(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) && @@ -989,26 +1137,28 @@ tcp_timers(struct tcpcb *tp, int timer) /* 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) + 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 */ + tp->t_maxopd = /* use the default MSS */ #if INET6 - isipv6 ? tcp_v6mssdflt : + isipv6 ? tcp_v6mssdflt : #endif /* INET6 */ - tcp_mssdflt; + tcp_mssdflt; } 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) + */ + if (CC_ALGO(tp)->cwnd_init != NULL) { CC_ALGO(tp)->cwnd_init(tp); + } tp->snd_cwnd = tp->t_maxseg; } /* @@ -1018,7 +1168,6 @@ tcp_timers(struct tcpcb *tp, int timer) * MSS and blackhole detection flags. */ else { - if ((tp->t_flags & TF_BLACKHOLE) && (tp->t_rxtshift > 4)) { tcp_pmtud_revert_segment_size(tp); @@ -1038,8 +1187,9 @@ tcp_timers(struct tcpcb *tp, int timer) * 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); + 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 @@ -1051,9 +1201,9 @@ tcp_timers(struct tcpcb *tp, int timer) */ if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { #if INET6 - if (isipv6) + if (isipv6) { in6_losing(tp->t_inpcb); - else + } else #endif /* INET6 */ in_losing(tp->t_inpcb); tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); @@ -1073,8 +1223,9 @@ tcp_timers(struct tcpcb *tp, int timer) /* If timing a segment in this window, stop the timer */ tp->t_rtttime = 0; - if (!IN_FASTRECOVERY(tp) && tp->t_rxtshift == 1) + if (!IN_FASTRECOVERY(tp) && tp->t_rxtshift == 1) { tcpstat.tcps_tailloss_rto++; + } /* @@ -1091,8 +1242,9 @@ tcp_timers(struct tcpcb *tp, int timer) * right after Fast Retransmits and ECE * notification receipts. */ - if (TCP_ECN_ENABLED(tp)) + if (TCP_ECN_ENABLED(tp)) { tp->ecn_flags |= TE_SENDCWR; + } } EXIT_FASTRECOVERY(tp); @@ -1128,11 +1280,11 @@ fc_output: (idle_time >= tcp_maxpersistidle || idle_time >= TCP_REXMTVAL(tp) * tcp_totbackoff)) || ((tp->t_persist_stop != 0) && - TSTMP_LEQ(tp->t_persist_stop, tcp_now))) { + TSTMP_LEQ(tp->t_persist_stop, tcp_now))) { tcpstat.tcps_persistdrop++; postevent(so, 0, EV_TIMEOUT); soevent(so, - (SO_FILT_HINT_LOCKED|SO_FILT_HINT_TIMEOUT)); + (SO_FILT_HINT_LOCKED | SO_FILT_HINT_TIMEOUT)); tp = tcp_drop(tp, ETIMEDOUT); break; } @@ -1153,24 +1305,27 @@ fc_output: * Regular TCP connections do not send keepalives after closing * MPTCP must not also, after sending Data FINs. */ - struct mptcb *mp_tp = tp->t_mptcb; + 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)) + if ((mptcp_ok_to_keepalive(mp_tp) == 0)) { goto dropit; + } } #endif /* MPTCP */ - if (tp->t_state < TCPS_ESTABLISHED) + if (tp->t_state < TCPS_ESTABLISHED) { goto dropit; + } if ((always_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)) { - if (idle_time >= TCP_CONN_KEEPIDLE(tp) + TCP_CONN_MAXIDLE(tp)) + if (idle_time >= TCP_CONN_KEEPIDLE(tp) + TCP_CONN_MAXIDLE(tp)) { goto dropit; + } /* * Send a packet designed to force a response * if the peer is up and reachable: @@ -1194,16 +1349,18 @@ fc_output: 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) + if (tp->t_inpcb->inp_flags & INP_BOUND_IF) { tra.ifscope = tp->t_inpcb->inp_boundifp->if_index; - else + } else { tra.ifscope = IFSCOPE_NONE; + } tcp_respond(tp, t_template->tt_ipgen, &t_template->tt_t, (struct mbuf *)NULL, tp->rcv_nxt, tp->snd_una - 1, 0, &tra); (void) m_free(dtom(t_template)); - if (tp->t_flagsext & TF_DETECT_READSTALL) + if (tp->t_flagsext & TF_DETECT_READSTALL) { tp->t_rtimo_probes++; + } } tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, TCP_CONN_KEEPINTVL(tp)); @@ -1242,7 +1399,7 @@ fc_output: int ind = min(tp->t_rtimo_probes, TCP_MAXRXTSHIFT); tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START( - tp, tcp_backoff[ind] * TCP_REXMTVAL(tp)); + tp, tcp_backoff[ind] * TCP_REXMTVAL(tp)); } } if (tp->t_tfo_probe_state == TFO_PROBE_PROBING) { @@ -1259,9 +1416,10 @@ fc_output: * backing off. */ tp->t_timer[TCPT_KEEP] = min(OFFSET_FROM_START( - tp, tcp_backoff[ind] * TCP_REXMTVAL(tp)), + tp, tcp_backoff[ind] * TCP_REXMTVAL(tp)), tp->t_timer[TCPT_KEEP]); - } else if (tp->t_tfo_probe_state == TFO_PROBE_WAIT_DATA) { + } 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); @@ -1288,11 +1446,12 @@ fc_output: */ if ((tp->t_flags & TF_STRETCHACK)) { if (tp->t_unacksegs > 1 && - tp->t_unacksegs < maxseg_unacked) + tp->t_unacksegs < maxseg_unacked) { tp->t_stretchack_delayed++; + } if (tp->t_stretchack_delayed > - TCP_STRETCHACK_DELAY_THRESHOLD) { + TCP_STRETCHACK_DELAY_THRESHOLD) { tp->t_flagsext |= TF_DISABLE_STRETCHACK; /* * Note the time at which stretch @@ -1328,13 +1487,14 @@ fc_output: tcpstat.tcps_timeoutdrop++; postevent(so, 0, EV_TIMEOUT); soevent(so, - (SO_FILT_HINT_LOCKED| + (SO_FILT_HINT_LOCKED | SO_FILT_HINT_TIMEOUT)); tp = tcp_drop(tp, tp->t_softerror ? - tp->t_softerror : ETIMEDOUT); + tp->t_softerror : ETIMEDOUT); break; } tcpstat.tcps_join_rxmts++; + tp->t_mpflags |= TMPF_SND_JACK; tp->t_flags |= TF_ACKNOW; /* @@ -1356,11 +1516,13 @@ fc_output: * 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)) + (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 @@ -1379,8 +1541,9 @@ fc_output: } tcpstat.tcps_pto++; - if (tp->t_flagsext & TF_PROBING) + if (tp->t_flagsext & TF_PROBING) { tcpstat.tcps_probe_if++; + } /* If timing a segment in this window, stop the timer */ tp->t_rtttime = 0; @@ -1389,7 +1552,16 @@ fc_output: tp->t_tlpstart = tcp_now; tp->snd_cwnd += tp->t_maxseg; - (void )tcp_output(tp); + + /* + * 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; @@ -1406,15 +1578,17 @@ fc_output: */ if (IN_FASTRECOVERY(tp) || SEQ_GEQ(tp->snd_una, tp->snd_recover) || - tp->t_rxtshift > 0) + 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)) + if (TCP_ECN_ENABLED(tp)) { tp->ecn_flags |= TE_SENDCWR; + } } ENTER_FASTRECOVERY(tp); @@ -1426,20 +1600,21 @@ fc_output: tcp_ccdbg_trace(tp, NULL, TCP_CC_ENTER_FASTRECOVERY); (void) tcp_output(tp); break; - dropit: +dropit: tcpstat.tcps_keepdrops++; postevent(so, 0, EV_TIMEOUT); soevent(so, - (SO_FILT_HINT_LOCKED|SO_FILT_HINT_TIMEOUT)); + (SO_FILT_HINT_LOCKED | SO_FILT_HINT_TIMEOUT)); tp = tcp_drop(tp, ETIMEDOUT); break; } #if TCPDEBUG - if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) + if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG) { tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, - PRU_SLOWTIMO); + PRU_SLOWTIMO); + } #endif - return (tp); + return tp; } /* Remove a timer entry from timer list */ @@ -1448,7 +1623,7 @@ tcp_remove_timer(struct tcpcb *tp) { 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; } @@ -1460,8 +1635,9 @@ tcp_remove_timer(struct tcpcb *tp) 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); + } LIST_REMOVE(&tp->tentry, le); tp->t_flags &= ~(TF_TIMER_ONLIST); @@ -1490,29 +1666,34 @@ need_to_resched_timerlist(u_int32_t runtime, u_int16_t mode) * 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); + 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); + return FALSE; } else { if (mode & TCP_TIMERLIST_10MS_MODE) { - if (diff <= TCP_TIMER_10MS_QUANTUM) - return (FALSE); + if (diff <= TCP_TIMER_10MS_QUANTUM) { + return FALSE; + } } else if (mode & TCP_TIMERLIST_100MS_MODE) { - if (diff <= TCP_TIMER_100MS_QUANTUM) - return (FALSE); + if (diff <= TCP_TIMER_100MS_QUANTUM) { + return FALSE; + } } else { - if (diff <= TCP_TIMER_500MS_QUANTUM) - return (FALSE); + if (diff <= TCP_TIMER_500MS_QUANTUM) { + return FALSE; + } } } - return (TRUE); + return TRUE; } void @@ -1521,10 +1702,11 @@ tcp_sched_timerlist(uint32_t offset) 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->schedtime = tcp_now; if (listp->runtime == 0) { listp->runtime++; offset++; @@ -1547,7 +1729,7 @@ tcp_sched_timerlist(uint32_t offset) */ u_int32_t tcp_run_conn_timer(struct tcpcb *tp, u_int16_t *te_mode, - u_int16_t probe_if_index) + u_int16_t probe_if_index) { struct socket *so; u_int16_t i = 0, index = TCPT_NONE, lo_index = TCPT_NONE; @@ -1560,12 +1742,12 @@ tcp_run_conn_timer(struct tcpcb *tp, u_int16_t *te_mode, 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) { + == WNT_STOPUSING) { if (TIMER_IS_ON_LIST(tp)) { tcp_remove_timer(tp); } @@ -1593,8 +1775,9 @@ tcp_run_conn_timer(struct tcpcb *tp, u_int16_t *te_mode, * 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) + if ((index = tp->tentry.index) == TCPT_NONE) { goto done; + } timer_val = tp->t_timer[index]; @@ -1610,8 +1793,9 @@ tcp_run_conn_timer(struct tcpcb *tp, u_int16_t *te_mode, tp->t_timer[index] = 0; if (timer_val > 0) { tp = tcp_timers(tp, index); - if (tp == NULL) + if (tp == NULL) { goto done; + } } /* @@ -1622,7 +1806,7 @@ tcp_run_conn_timer(struct tcpcb *tp, u_int16_t *te_mode, 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); + tp->t_timer[i], tcp_now, 0); if (diff <= 0) { needtorun[i] = TRUE; count++; @@ -1644,9 +1828,10 @@ tcp_run_conn_timer(struct tcpcb *tp, u_int16_t *te_mode, 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->t_timer[tp->tentry.index]; + if (tp->tentry.runtime == 0) { tp->tentry.runtime++; + } } if (count > 0) { @@ -1676,8 +1861,8 @@ done: offset = 0; } - tcp_unlock(so, 1, 0); - return(offset); + socket_unlock(so, 1); + return offset; } void @@ -1688,7 +1873,7 @@ tcp_run_timerlist(void * arg1, void * arg2) struct tcptimerlist *listp = &tcp_timer_list; struct tcpcb *tp; 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 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; @@ -1696,6 +1881,27 @@ tcp_run_timerlist(void * arg1, void * arg2) 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; LIST_FOREACH_SAFE(te, &listp->lhead, le, next_te) { @@ -1761,21 +1967,23 @@ tcp_run_timerlist(void * arg1, void * arg2) if (offset > 0 && te_mode != 0) { list_mode |= te_mode; - if (next_timer == 0 || offset < next_timer) + if (next_timer == 0 || offset < next_timer) { next_timer = offset; + } } } if (!LIST_EMPTY(&listp->lhead)) { u_int16_t next_mode = 0; if ((list_mode & TCP_TIMERLIST_10MS_MODE) || - (listp->pref_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)) + } else if ((list_mode & TCP_TIMERLIST_100MS_MODE) || + (listp->pref_mode & TCP_TIMERLIST_100MS_MODE)) { next_mode = TCP_TIMERLIST_100MS_MODE; - else + } else { next_mode = TCP_TIMERLIST_500MS_MODE; + } if (next_mode != TCP_TIMERLIST_500MS_MODE) { listp->idleruns = 0; @@ -1798,12 +2006,14 @@ tcp_run_timerlist(void * arg1, void * arg2) } } listp->mode = next_mode; - if (listp->pref_offset != 0) + if (listp->pref_offset != 0) { next_timer = min(listp->pref_offset, next_timer); + } - if (listp->mode == TCP_TIMERLIST_500MS_MODE) + if (listp->mode == TCP_TIMERLIST_500MS_MODE) { next_timer = max(next_timer, - TCP_TIMER_500MS_QUANTUM); + TCP_TIMER_500MS_QUANTUM); + } tcp_sched_timerlist(next_timer); } else { @@ -1871,12 +2081,14 @@ tcp_sched_timers(struct tcpcb *tp) tp->t_flags |= TF_TIMER_ONLIST; listp->entries++; - if (listp->entries > listp->maxentries) + if (listp->entries > listp->maxentries) { listp->maxentries = listp->entries; + } /* if the list is not scheduled, just schedule it */ - if (!listp->scheduled) + if (!listp->scheduled) { goto schedule; + } } } @@ -1898,7 +2110,7 @@ tcp_sched_timers(struct tcpcb *tp) if (listp->running) { listp->pref_mode |= mode; if (listp->pref_offset == 0 || - offset < listp->pref_offset) { + offset < listp->pref_offset) { listp->pref_offset = offset; } } else { @@ -1910,10 +2122,11 @@ tcp_sched_timers(struct tcpcb *tp) int32_t diff; diff = timer_diff(listp->runtime, 0, tcp_now, offset); - if (diff <= 0) + if (diff <= 0) { goto done; - else + } else { goto schedule; + } } else { goto schedule; } @@ -1932,16 +2145,18 @@ schedule: listp->idleruns = 0; offset = min(offset, TCP_TIMER_10MS_QUANTUM); } else if (mode & TCP_TIMERLIST_100MS_MODE) { - if (listp->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: - if (list_locked) + if (list_locked) { lck_mtx_unlock(listp->mtx); + } return; } @@ -1967,18 +2182,20 @@ tcp_set_lotimer_index(struct tcpcb *tp) if (tp->tentry.index != TCPT_NONE) { tp->tentry.runtime = tp->tentry.timer_start + tp->t_timer[tp->tentry.index]; - if (tp->tentry.runtime == 0) + if (tp->tentry.runtime == 0) { tp->tentry.runtime++; + } } } void tcp_check_timer_state(struct tcpcb *tp) { - lck_mtx_assert(&tp->t_inpcb->inpcb_mtx, LCK_MTX_ASSERT_OWNED); + socket_lock_assert_owned(tp->t_inpcb->inp_socket); - if (tp->t_inpcb->inp_flags2 & INP2_TIMEWAIT) + if (tp->t_inpcb->inp_flags2 & INP2_TIMEWAIT) { return; + } tcp_set_lotimer_index(tp); @@ -1991,10 +2208,25 @@ 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) + if (diff > 0) { *dest = diff; - else + } 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; } @@ -2009,10 +2241,11 @@ tcp_report_stats(void) static struct tcp_last_report_stats prev; u_int64_t var, uptime; -#define stat data.u.tcp_stats +#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; @@ -2026,12 +2259,12 @@ tcp_report_stats(void) /* ipv4 avg rtt */ lck_mtx_lock(rnh_lock); rt = rt_lookup(TRUE, (struct sockaddr *)&dst, NULL, - rt_tables[AF_INET], IFSCOPE_NONE); + 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) { + rt->rt_stats != NULL) { stat.ipv4_avgrtt = rt->rt_stats->nstat_avg_rtt; } RT_UNLOCK(rt); @@ -2045,13 +2278,13 @@ tcp_report_stats(void) dst6.sin6_family = AF_INET6; lck_mtx_lock(rnh_lock); - rt = rt_lookup(TRUE,(struct sockaddr *)&dst6, NULL, - rt_tables[AF_INET6], IFSCOPE_NONE); + 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) { + rt->rt_stats != NULL) { stat.ipv6_avgrtt = rt->rt_stats->nstat_avg_rtt; } RT_UNLOCK(rt); @@ -2076,20 +2309,22 @@ tcp_report_stats(void) && tcpstat.tcps_tailloss_rto > 0) { var = tcpstat.tcps_tailloss_rto << 10; stat.send_tlrto_rate = - (var * 100) / tcpstat.tcps_sndrexmitpack; + (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; + (var * 100) / tcpstat.tcps_sndpack; } - if (tcp_ecn_outbound == 1) + if (tcp_ecn_outbound == 1) { stat.ecn_client_enabled = 1; - if (tcp_ecn_inbound == 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, @@ -2164,11 +2399,68 @@ tcp_report_stats(void) &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 +#undef stat } void @@ -2187,8 +2479,9 @@ tcp_interface_send_probe(u_int16_t probe_if_index) } listp->probe_if_index = probe_if_index; - if (listp->running) + if (listp->running) { goto done; + } /* * Reschedule the timerlist to run within the next 10ms, which is @@ -2246,8 +2539,9 @@ tcp_enable_read_probe(struct tcpcb *tp, struct ifnet *ifp) tp->tentry.index = TCPT_KEEP; tp->tentry.runtime = tcp_now + TCP_TIMER_10MS_QUANTUM; - if (tp->tentry.runtime == 0) + if (tp->tentry.runtime == 0) { tp->tentry.runtime++; + } } } } @@ -2263,6 +2557,10 @@ tcp_disable_read_probe(struct tcpcb *tp) ((tp->t_flagsext & TF_DETECT_READSTALL) || tp->t_rtimo_probes > 0)) { tcp_keepalive_reset(tp); + + if (tp->t_mpsub) { + mptcp_reset_keepalive(tp); + } } } @@ -2278,8 +2576,9 @@ tcp_probe_connectivity(struct ifnet *ifp, u_int32_t enable) struct inpcbinfo *pcbinfo = &tcbinfo; struct inpcb *inp, *nxt; - if (ifp == NULL) + if (ifp == NULL) { return; + } /* update clock */ calculate_tcp_clock(); @@ -2293,25 +2592,27 @@ tcp_probe_connectivity(struct ifnet *ifp, u_int32_t enable) LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, nxt) { struct tcpcb *tp = NULL; if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == - WNT_STOPUSING) + WNT_STOPUSING) { continue; + } /* Acquire lock to look at the state of the connection */ - tcp_lock(inp->inp_socket, 1, 0); + socket_lock(inp->inp_socket, 1); /* 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); + socket_unlock(inp->inp_socket, 1); continue; } tp = intotcpcb(inp); - if (enable) + if (enable) { tcp_enable_read_probe(tp, ifp); - else + } else { tcp_disable_read_probe(tp); + } - tcp_unlock(inp->inp_socket, 1, 0); + socket_unlock(inp->inp_socket, 1); } lck_rw_done(pcbinfo->ipi_lock); @@ -2373,8 +2674,9 @@ tcp_update_mss_core(struct tcpcb *tp, struct ifnet *ifp) /* * clear the cached value if it is same as the current */ - if (tp->t_maxopd == tp->t_cached_maxopd) + if (tp->t_maxopd == tp->t_cached_maxopd) { tp->t_cached_maxopd = 0; + } } } @@ -2384,11 +2686,9 @@ 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) + if (ifp == NULL && (ifp = inp->inp_last_outifp) == NULL) { return; - - if (ifp == NULL) - ifp = inp->inp_last_outifp; + } if (!IFNET_IS_CELLULAR(ifp)) { /* @@ -2397,15 +2697,17 @@ tcp_update_mss_locked(struct socket *so, struct ifnet *ifp) */ return; } - if ( tp->t_state <= TCPS_CLOSE_WAIT) { + 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) + if (tp->t_flags & TF_BLACKHOLE) { return; - if (ifp->if_link_status == NULL) + } + if (ifp->if_link_status == NULL) { return; + } tcp_update_mss_core(tp, ifp); } } @@ -2428,23 +2730,42 @@ tcp_itimer(struct inpcbinfo *ipi) 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) + in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) { continue; + } so = inp->inp_socket; - tcp_lock(so, 1, 0); + ifp = inp->inp_last_outifp; + socket_lock(so, 1); if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { - tcp_unlock(so, 1, 0); + 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); } - tcp_unlock(so, 1, 0); + 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; + ipi->ipi_flags &= ~(INPCBINFO_UPDATE_MSS | INPCBINFO_HANDLE_LQM_ABORT); lck_rw_done(ipi->ipi_lock); }