X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/5c9f46613a83ebfc29a5b1f099448259e96a98f0..HEAD:/bsd/netinet/tcp_input.c diff --git a/bsd/netinet/tcp_input.c b/bsd/netinet/tcp_input.c index 139984932..d91a93a61 100644 --- a/bsd/netinet/tcp_input.c +++ b/bsd/netinet/tcp_input.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2017 Apple Inc. All rights reserved. + * Copyright (c) 2000-2020 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -73,17 +73,17 @@ #include #include #include -#include /* for proc0 declaration */ +#include /* for proc0 declaration */ #include #include #include #include #include -#if !CONFIG_EMBEDDED +#if XNU_TARGET_OS_OSX #include -#endif +#endif /* XNU_TARGET_OS_OSX */ #include -#include /* before tcp_seq.h, for tcp_random18() */ +#include /* before tcp_seq.h, for tcp_random18() */ #include @@ -91,24 +91,24 @@ #include #include #include +#include #include +#include #include #include #include #include /* for ICMP_BANDLIM */ #include -#include /* for ICMP_BANDLIM */ +#include /* for ICMP_BANDLIM */ #include #include #include -#if INET6 #include #include #include #include #include -#endif #include #include #include @@ -118,30 +118,22 @@ #include #include #include -#if INET6 #include -#endif #include #if TCPDEBUG #include u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ struct tcphdr tcp_savetcp; #endif /* TCPDEBUG */ +#include #if IPSEC #include -#if INET6 #include -#endif #include #endif /*IPSEC*/ -#if CONFIG_MACF_NET || CONFIG_MACF_SOCKET -#include -#endif /* CONFIG_MACF_NET || CONFIG_MACF_SOCKET */ - #include -#include #if MPTCP #include #include @@ -150,45 +142,51 @@ struct tcphdr tcp_savetcp; #include -#define DBG_LAYER_BEG NETDBG_CODE(DBG_NETTCP, 0) -#define DBG_LAYER_END NETDBG_CODE(DBG_NETTCP, 2) +#define DBG_LAYER_BEG NETDBG_CODE(DBG_NETTCP, 0) +#define DBG_LAYER_END NETDBG_CODE(DBG_NETTCP, 2) #define DBG_FNC_TCP_INPUT NETDBG_CODE(DBG_NETTCP, (3 << 8)) #define DBG_FNC_TCP_NEWCONN NETDBG_CODE(DBG_NETTCP, (7 << 8)) -#define TCP_RTT_HISTORY_EXPIRE_TIME (60 * TCP_RETRANSHZ) -#define TCP_RECV_THROTTLE_WIN (5 * TCP_RETRANSHZ) -#define TCP_STRETCHACK_ENABLE_PKTCNT 2000 +#define TCP_RTT_HISTORY_EXPIRE_TIME (60 * TCP_RETRANSHZ) +#define TCP_RECV_THROTTLE_WIN (5 * TCP_RETRANSHZ) +#define TCP_STRETCHACK_ENABLE_PKTCNT 2000 + +struct tcpstat tcpstat; -struct tcpstat tcpstat; +SYSCTL_SKMEM_TCP_INT(OID_AUTO, flow_control_response, + CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_flow_control_response, 1, + "Improved response to Flow-control events"); static int log_in_vain = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW | CTLFLAG_LOCKED, &log_in_vain, 0, "Log all incoming TCP connections"); +SYSCTL_SKMEM_TCP_INT(OID_AUTO, ack_strategy, + CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_ack_strategy, TCP_ACK_STRATEGY_MODERN, + "Revised TCP ACK-strategy, avoiding stretch-ACK implementation"); + static int blackhole = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW | CTLFLAG_LOCKED, &blackhole, 0, "Do not send RST when dropping refused connections"); +SYSCTL_SKMEM_TCP_INT(OID_AUTO, aggressive_rcvwnd_inc, + CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_aggressive_rcvwnd_inc, 1, + "Be more aggressive about increasing the receive-window."); + SYSCTL_SKMEM_TCP_INT(OID_AUTO, delayed_ack, CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_delack_enabled, 3, "Delay ACK to try and piggyback it onto a data packet"); -SYSCTL_SKMEM_TCP_INT(OID_AUTO, tcp_lq_overflow, - CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_lq_overflow, 1, - "Listen Queue Overflow"); - SYSCTL_SKMEM_TCP_INT(OID_AUTO, recvbg, CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_recv_bg, 0, "Receive background"); -#if TCP_DROP_SYNFIN SYSCTL_SKMEM_TCP_INT(OID_AUTO, drop_synfin, CTLFLAG_RW | CTLFLAG_LOCKED, static int, drop_synfin, 1, "Drop TCP packets with SYN+FIN set"); -#endif -SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW|CTLFLAG_LOCKED, 0, +SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW | CTLFLAG_LOCKED, 0, "TCP Segment Reassembly Queue"); static int tcp_reass_overflows = 0; @@ -198,8 +196,8 @@ SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, SYSCTL_SKMEM_TCP_INT(OID_AUTO, slowlink_wsize, CTLFLAG_RW | CTLFLAG_LOCKED, - __private_extern__ int, slowlink_wsize, 8192, - "Maximum advertised window size for slowlink"); + __private_extern__ int, slowlink_wsize, 8192, + "Maximum advertised window size for slowlink"); SYSCTL_SKMEM_TCP_INT(OID_AUTO, maxseg_unacked, CTLFLAG_RW | CTLFLAG_LOCKED, int, maxseg_unacked, 8, @@ -219,90 +217,48 @@ u_int32_t tcp_autorcvbuf_inc_shift = 3; SYSCTL_SKMEM_TCP_INT(OID_AUTO, recv_allowed_iaj, CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_allowed_iaj, ALLOWED_IAJ, "Allowed inter-packet arrival jiter"); -#if (DEVELOPMENT || DEBUG) -SYSCTL_INT(_net_inet_tcp, OID_AUTO, acc_iaj_high_thresh, - CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_acc_iaj_high_thresh, 0, - "Used in calculating maximum accumulated IAJ"); - -SYSCTL_INT(_net_inet_tcp, OID_AUTO, autorcvbufincshift, - CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_autorcvbuf_inc_shift, 0, - "Shift for increment in receive socket buffer size"); -#endif /* (DEVELOPMENT || DEBUG) */ SYSCTL_SKMEM_TCP_INT(OID_AUTO, doautorcvbuf, CTLFLAG_RW | CTLFLAG_LOCKED, u_int32_t, tcp_do_autorcvbuf, 1, "Enable automatic socket buffer tuning"); +SYSCTL_SKMEM_TCP_INT(OID_AUTO, autotunereorder, + CTLFLAG_RW | CTLFLAG_LOCKED, u_int32_t, tcp_autotune_reorder, 1, + "Enable automatic socket buffer tuning even when reordering is present"); + SYSCTL_SKMEM_TCP_INT(OID_AUTO, autorcvbufmax, - CTLFLAG_RW | CTLFLAG_LOCKED, u_int32_t, tcp_autorcvbuf_max, 512 * 1024, + CTLFLAG_RW | CTLFLAG_LOCKED, u_int32_t, tcp_autorcvbuf_max, 2 * 1024 * 1024, "Maximum receive socket buffer size"); -u_int32_t tcp_autorcvbuf_max_ca = 512 * 1024; -#if (DEBUG || DEVELOPMENT) -SYSCTL_INT(_net_inet_tcp, OID_AUTO, autorcvbufmaxca, - CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_autorcvbuf_max_ca, 0, - "Maximum receive socket buffer size"); -#endif /* (DEBUG || DEVELOPMENT) */ - -#if CONFIG_EMBEDDED -int sw_lro = 1; -#else -int sw_lro = 0; -#endif /* !CONFIG_EMBEDDED */ -SYSCTL_INT(_net_inet_tcp, OID_AUTO, lro, CTLFLAG_RW | CTLFLAG_LOCKED, - &sw_lro, 0, "Used to coalesce TCP packets"); - -int lrodebug = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, lrodbg, - CTLFLAG_RW | CTLFLAG_LOCKED, &lrodebug, 0, - "Used to debug SW LRO"); - -int lro_start = 4; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, lro_startcnt, - CTLFLAG_RW | CTLFLAG_LOCKED, &lro_start, 0, - "Segments for starting LRO computed as power of 2"); - -int limited_txmt = 1; -int early_rexmt = 1; -int sack_ackadv = 1; -int tcp_dsack_enable = 1; - -#if (DEVELOPMENT || DEBUG) -SYSCTL_INT(_net_inet_tcp, OID_AUTO, limited_transmit, - CTLFLAG_RW | CTLFLAG_LOCKED, &limited_txmt, 0, - "Enable limited transmit"); - -SYSCTL_INT(_net_inet_tcp, OID_AUTO, early_rexmt, - CTLFLAG_RW | CTLFLAG_LOCKED, &early_rexmt, 0, - "Enable Early Retransmit"); - -SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack_ackadv, - CTLFLAG_RW | CTLFLAG_LOCKED, &sack_ackadv, 0, - "Use SACK with cumulative ack advancement as a dupack"); - -SYSCTL_INT(_net_inet_tcp, OID_AUTO, dsack_enable, - CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_dsack_enable, 0, - "use DSACK TCP option to report duplicate segments"); - -#endif /* (DEVELOPMENT || DEBUG) */ int tcp_disable_access_to_stats = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_access_to_stats, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_disable_access_to_stats, 0, "Disable access to tcpstat"); +SYSCTL_SKMEM_TCP_INT(OID_AUTO, challengeack_limit, + CTLFLAG_RW | CTLFLAG_LOCKED, uint32_t, tcp_challengeack_limit, 10, + "Maximum number of challenge ACKs per connection per second"); + +/* TO BE REMOVED */ +SYSCTL_SKMEM_TCP_INT(OID_AUTO, do_rfc5961, + CTLFLAG_RW | CTLFLAG_LOCKED, static int, tcp_do_rfc5961, 1, + "Enable/Disable full RFC 5961 compliance"); + +SYSCTL_SKMEM_TCP_INT(OID_AUTO, do_better_lr, + CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_do_better_lr, 1, + "Improved TCP Loss Recovery"); -extern int tcp_TCPTV_MIN; extern int tcp_acc_iaj_high; extern int tcp_acc_iaj_react_limit; int tcprexmtthresh = 3; u_int32_t tcp_now; -struct timeval tcp_uptime; /* uptime when tcp_now was last updated */ -lck_spin_t *tcp_uptime_lock; /* Used to sychronize updates to tcp_now */ +struct timeval tcp_uptime; /* uptime when tcp_now was last updated */ +lck_spin_t *tcp_uptime_lock; /* Used to sychronize updates to tcp_now */ struct inpcbhead tcb; -#define tcb6 tcb /* for KAME src sync over BSD*'s */ +#define tcb6 tcb /* for KAME src sync over BSD*'s */ struct inpcbinfo tcbinfo; static void tcp_dooptions(struct tcpcb *, u_char *, int, struct tcphdr *, @@ -310,8 +266,6 @@ static void tcp_dooptions(struct tcpcb *, u_char *, int, struct tcphdr *, static void tcp_finalize_options(struct tcpcb *, struct tcpopt *, unsigned int); static void tcp_pulloutofband(struct socket *, struct tcphdr *, struct mbuf *, int); -static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, struct mbuf *, - struct ifnet *); static void tcp_xmit_timer(struct tcpcb *, int, u_int32_t, tcp_seq); static inline unsigned int tcp_maxmtu(struct rtentry *); static inline int tcp_stretch_ack_enable(struct tcpcb *tp, int thflags); @@ -320,18 +274,15 @@ static inline void tcp_adaptive_rwtimo_check(struct tcpcb *, int); #if TRAFFIC_MGT static inline void update_iaj_state(struct tcpcb *tp, uint32_t tlen, int reset_size); -void compute_iaj(struct tcpcb *tp, int nlropkts, int lro_delay_factor); -static void compute_iaj_meat(struct tcpcb *tp, uint32_t cur_iaj); +static inline void compute_iaj(struct tcpcb *tp); +static inline void compute_iaj_meat(struct tcpcb *tp, uint32_t cur_iaj); #endif /* TRAFFIC_MGT */ -#if INET6 static inline unsigned int tcp_maxmtu6(struct rtentry *); -#endif - unsigned int get_maxmtu(struct rtentry *); static void tcp_sbrcv_grow(struct tcpcb *tp, struct sockbuf *sb, - struct tcpopt *to, u_int32_t tlen, u_int32_t rcvbuf_max); + struct tcpopt *to, uint32_t tlen); void tcp_sbrcv_trim(struct tcpcb *tp, struct sockbuf *sb); static void tcp_sbsnd_trim(struct sockbuf *sbsnd); static inline void tcp_sbrcv_tstmp_check(struct tcpcb *tp); @@ -357,28 +308,12 @@ static void tcp_bad_rexmt_check(struct tcpcb *tp, struct tcphdr *th, #define TCP_EARLY_REXMT_WIN (60 * TCP_RETRANSHZ) /* 60 seconds */ #define TCP_EARLY_REXMT_LIMIT 10 -extern void ipfwsyslog( int level, const char *format,...); -extern int fw_verbose; - -#if IPFIREWALL -extern void ipfw_stealth_stats_incr_tcp(void); - -#define log_in_vain_log( a ) { \ - if ( (log_in_vain == 3 ) && (fw_verbose == 2)) { /* Apple logging, log to ipfw.log */ \ - ipfwsyslog a ; \ - } else if ( (log_in_vain == 4 ) && (fw_verbose == 2)) { \ - ipfw_stealth_stats_incr_tcp(); \ - } \ - else log a ; \ -} -#else #define log_in_vain_log( a ) { log a; } -#endif int tcp_rcvunackwin = TCPTV_UNACKWIN; int tcp_maxrcvidle = TCPTV_MAXRCVIDLE; SYSCTL_SKMEM_TCP_INT(OID_AUTO, rcvsspktcnt, CTLFLAG_RW | CTLFLAG_LOCKED, - int, tcp_rcvsspktcnt, TCP_RCV_SS_PKTCOUNT, "packets to be seen before receiver stretches acks"); + int, tcp_rcvsspktcnt, TCP_RCV_SS_PKTCOUNT, "packets to be seen before receiver stretches acks"); #define DELAY_ACK(tp, th) \ (CC_ALGO(tp)->delay_ack != NULL && CC_ALGO(tp)->delay_ack(tp, th)) @@ -402,8 +337,9 @@ reset_acc_iaj(struct tcpcb *tp) static inline void update_iaj_state(struct tcpcb *tp, uint32_t size, int rst_size) { - if (rst_size > 0) + if (rst_size > 0) { tp->iaj_size = 0; + } if (tp->iaj_size == 0 || size >= tp->iaj_size) { tp->iaj_size = size; tp->iaj_rcv_ts = tcp_now; @@ -419,8 +355,8 @@ static inline int isqrt(unsigned int val) { unsigned int sqrt_cache[11] = {0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100}; - unsigned int temp, g=0, b=0x8000, bshft=15; - if ( val <= 100) { + unsigned int temp, g = 0, b = 0x8000, bshft = 15; + if (val <= 100) { for (g = 0; g <= 10; ++g) { if (sqrt_cache[g] > val) { g--; @@ -437,48 +373,19 @@ isqrt(unsigned int val) val -= temp; } b >>= 1; - } while ( b > 0 && val > 0); + } while (b > 0 && val > 0); } - return(g); + return g; } -/* -* With LRO, roughly estimate the inter arrival time between -* each sub coalesced packet as an average. Count the delay -* cur_iaj to be the delay between the last packet received -* and the first packet of the LRO stream. Due to round off errors -* cur_iaj may be the same as lro_delay_factor. Averaging has -* round off errors too. lro_delay_factor may be close to 0 -* in steady state leading to lower values fed to compute_iaj_meat. -*/ -void -compute_iaj(struct tcpcb *tp, int nlropkts, int lro_delay_factor) +static inline void +compute_iaj(struct tcpcb *tp) { - uint32_t cur_iaj = tcp_now - tp->iaj_rcv_ts; - uint32_t timediff = 0; - - if (cur_iaj >= lro_delay_factor) { - cur_iaj = cur_iaj - lro_delay_factor; - } - - compute_iaj_meat(tp, cur_iaj); - - if (nlropkts <= 1) - return; - - nlropkts--; - - timediff = lro_delay_factor/nlropkts; - - while (nlropkts > 0) - { - compute_iaj_meat(tp, timediff); - nlropkts--; - } + compute_iaj_meat(tp, (tcp_now - tp->iaj_rcv_ts)); } -static -void compute_iaj_meat(struct tcpcb *tp, uint32_t cur_iaj) +static inline void +compute_iaj_meat(struct tcpcb *tp, uint32_t cur_iaj) { /* When accumulated IAJ reaches MAX_ACC_IAJ in milliseconds, * throttle the receive window to a minimum of MIN_IAJ_WIN packets @@ -499,8 +406,9 @@ void compute_iaj_meat(struct tcpcb *tp, uint32_t cur_iaj) * using standard deviation. */ allowed_iaj = tp->avg_iaj + tp->std_dev_iaj; - if (allowed_iaj < tcp_allowed_iaj) + if (allowed_iaj < tcp_allowed_iaj) { allowed_iaj = tcp_allowed_iaj; + } /* Initially when the connection starts, the senders congestion * window is small. During this period we avoid throttling a @@ -509,18 +417,19 @@ void compute_iaj_meat(struct tcpcb *tp, uint32_t cur_iaj) * the first few packets. */ if (tp->iaj_pktcnt > IAJ_IGNORE_PKTCNT) { - if ( cur_iaj <= allowed_iaj ) { - if (tp->acc_iaj >= 2) + if (cur_iaj <= allowed_iaj) { + if (tp->acc_iaj >= 2) { acc_iaj = tp->acc_iaj - 2; - else + } else { acc_iaj = 0; - + } } else { acc_iaj = tp->acc_iaj + (cur_iaj - allowed_iaj); } - if (acc_iaj > MAX_ACC_IAJ) + if (acc_iaj > MAX_ACC_IAJ) { acc_iaj = MAX_ACC_IAJ; + } tp->acc_iaj = acc_iaj; } @@ -532,25 +441,60 @@ void compute_iaj_meat(struct tcpcb *tp, uint32_t cur_iaj) * instead of round-down */ tp->avg_iaj = (((tp->avg_iaj << IAJ_DIV_SHIFT) - tp->avg_iaj) - + cur_iaj + IAJ_ROUNDUP_CONST) >> IAJ_DIV_SHIFT; + + cur_iaj + IAJ_ROUNDUP_CONST) >> IAJ_DIV_SHIFT; /* Compute Root-mean-square of deviation where mean is a weighted * average as described above. */ temp = tp->std_dev_iaj * tp->std_dev_iaj; mean = (((temp << IAJ_DIV_SHIFT) - temp) - + (cur_iaj_dev * cur_iaj_dev) - + IAJ_ROUNDUP_CONST) >> IAJ_DIV_SHIFT; + + (cur_iaj_dev * cur_iaj_dev) + + IAJ_ROUNDUP_CONST) >> IAJ_DIV_SHIFT; tp->std_dev_iaj = isqrt(mean); DTRACE_TCP3(iaj, struct tcpcb *, tp, uint32_t, cur_iaj, - uint32_t, allowed_iaj); + uint32_t, allowed_iaj); return; } #endif /* TRAFFIC_MGT */ +/* + * Perform rate limit check per connection per second + * tp->t_challengeack_last is the last_time diff was greater than 1sec + * tp->t_challengeack_count is the number of ACKs sent (within 1sec) + * Return TRUE if we shouldn't send the ACK due to rate limitation + * Return FALSE if it is still ok to send challenge ACK + */ +static boolean_t +tcp_is_ack_ratelimited(struct tcpcb *tp) +{ + boolean_t ret = TRUE; + uint32_t now = tcp_now; + int32_t diff = 0; + + diff = timer_diff(now, 0, tp->t_challengeack_last, 0); + /* If it is first time or diff > 1000ms, + * update the challengeack_last and reset the + * current count of ACKs + */ + if (tp->t_challengeack_last == 0 || diff >= 1000) { + tp->t_challengeack_last = now; + tp->t_challengeack_count = 0; + ret = FALSE; + } else if (tp->t_challengeack_count < tcp_challengeack_limit) { + ret = FALSE; + } + + /* Careful about wrap-around */ + if (ret == FALSE && (tp->t_challengeack_count + 1 > 0)) { + tp->t_challengeack_count++; + } + + return ret; +} + /* Check if enough amount of data has been acknowledged since * bw measurement was started */ @@ -560,8 +504,9 @@ tcp_bwmeas_check(struct tcpcb *tp) int32_t bw_meas_bytes; uint32_t bw, bytes, elapsed_time; - if (SEQ_LEQ(tp->snd_una, tp->t_bwmeas->bw_start)) + if (SEQ_LEQ(tp->snd_una, tp->t_bwmeas->bw_start)) { return; + } bw_meas_bytes = tp->snd_una - tp->t_bwmeas->bw_start; if ((tp->t_flagsext & TF_BWMEAS_INPROGRESS) && @@ -570,7 +515,7 @@ tcp_bwmeas_check(struct tcpcb *tp) elapsed_time = tcp_now - tp->t_bwmeas->bw_ts; if (elapsed_time > 0) { bw = bytes / elapsed_time; - if ( bw > 0) { + if (bw > 0) { if (tp->t_bwmeas->bw_sndbw > 0) { tp->t_bwmeas->bw_sndbw = (((tp->t_bwmeas->bw_sndbw << 3) @@ -597,7 +542,7 @@ tcp_bwmeas_check(struct tcpcb *tp) static int tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, - struct ifnet *ifp) + struct ifnet *ifp, int *dowakeup) { struct tseg_qent *q; struct tseg_qent *p = NULL; @@ -606,9 +551,6 @@ tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, struct inpcb *inp = tp->t_inpcb; struct socket *so = inp->inp_socket; int flags = 0; - int dowakeup = 0; - struct mbuf *oodata = NULL; - int copy_oodata = 0; u_int16_t qlimit; boolean_t cell = IFNET_IS_CELLULAR(ifp); boolean_t wifi = (!cell && IFNET_IS_WIFI(ifp)); @@ -619,8 +561,9 @@ tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, * Call with th==0 after become established to * force pre-ESTABLISHED data up to user socket. */ - if (th == NULL) + if (th == NULL) { goto present; + } /* * If the reassembly queue already has entries or if we are going @@ -628,12 +571,29 @@ tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, * Reset the stretch-ack algorithm at this point. */ tcp_reset_stretch_ack(tp); + tp->t_forced_acks = TCP_FORCED_ACKS_COUNT; #if TRAFFIC_MGT - if (tp->acc_iaj > 0) + if (tp->acc_iaj > 0) { reset_acc_iaj(tp); + } #endif /* TRAFFIC_MGT */ + if (th->th_seq != tp->rcv_nxt) { + struct mbuf *tmp = m; + while (tmp != NULL) { + if (mbuf_class_under_pressure(tmp)) { + m_freem(m); + tcp_reass_overflows++; + tcpstat.tcps_rcvmemdrop++; + *tlenp = 0; + return 0; + } + + tmp = tmp->m_next; + } + } + /* * Limit the number of segments in the reassembly queue to prevent * holding on to too many segments (and thus running out of mbufs). @@ -642,14 +602,14 @@ tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, * process the missing segment. */ qlimit = min(max(100, so->so_rcv.sb_hiwat >> 10), - (TCP_AUTORCVBUF_MAX(ifp) >> 10)); + (tcp_autorcvbuf_max >> 10)); if (th->th_seq != tp->rcv_nxt && (tp->t_reassqlen + 1) >= qlimit) { tcp_reass_overflows++; tcpstat.tcps_rcvmemdrop++; m_freem(m); *tlenp = 0; - return (0); + return 0; } /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ @@ -657,7 +617,7 @@ tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, if (te == NULL) { tcpstat.tcps_rcvmemdrop++; m_freem(m); - return (0); + return 0; } tp->t_reassqlen++; @@ -665,8 +625,9 @@ tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, * Find a segment which begins after this one does. */ LIST_FOREACH(q, &tp->t_segq, tqe_q) { - if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) + if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) { break; + } p = q; } @@ -680,7 +641,7 @@ tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, /* conversion to int (in i) handles seq wraparound */ i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; if (i > 0) { - if (TCP_DSACK_ENABLED(tp) && i > 1) { + if (i > 1) { /* * Note duplicate data sequnce numbers * to report in DSACK option @@ -744,8 +705,9 @@ tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, */ while (q) { int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; - if (i <= 0) + if (i <= 0) { break; + } /* * Report only the first part of partial/non-contiguous @@ -753,7 +715,7 @@ tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, * dsack_set will be true if a previous entry has some of * the duplicate sequence space. */ - if (TCP_DSACK_ENABLED(tp) && i > 1 && !dsack_set) { + if (i > 1 && !dsack_set) { if (tp->t_dsack_lseq == 0) { tp->t_dsack_lseq = q->tqe_th->th_seq; tp->t_dsack_rseq = @@ -794,39 +756,17 @@ tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m, LIST_INSERT_AFTER(p, te, tqe_q); } - /* - * New out-of-order data exists, and is pointed to by - * queue entry te. Set copy_oodata to 1 so out-of-order data - * can be copied off to sockbuf after in-order data - * is copied off. - */ - if (!(so->so_state & SS_CANTRCVMORE)) - copy_oodata = 1; - present: /* * Present data to user, advancing rcv_nxt through * completed sequence space. */ - if (!TCPS_HAVEESTABLISHED(tp->t_state)) - return (0); + if (!TCPS_HAVEESTABLISHED(tp->t_state)) { + return 0; + } q = LIST_FIRST(&tp->t_segq); if (!q || q->tqe_th->th_seq != tp->rcv_nxt) { - /* Stop using LRO once out of order packets arrive */ - if (tp->t_flagsext & TF_LRO_OFFLOADED) { - tcp_lro_remove_state(inp->inp_laddr, inp->inp_faddr, - th->th_dport, th->th_sport); - tp->t_flagsext &= ~TF_LRO_OFFLOADED; - } - - /* - * continue processing if out-of-order data - * can be delivered - */ - if (q && (so->so_flags & SOF_ENABLE_MSGS)) - goto msg_unordered_delivery; - - return (0); + return 0; } /* @@ -834,8 +774,9 @@ present: * connection, it is better to let it finish the job -- * (radar 16316196) */ - if (tp->t_flagsext & TF_REASS_INPROG) - return (0); + if (tp->t_flagsext & TF_REASS_INPROG) { + return 0; + } tp->t_flagsext |= TF_REASS_INPROG; /* lost packet was recovered, so ooo data can be returned */ @@ -848,28 +789,26 @@ present: if (so->so_state & SS_CANTRCVMORE) { m_freem(q->tqe_m); } else { + /* + * The mbuf may be freed after it has been added to the + * receive socket buffer so we reinitialize th to point + * to a safe copy of the TCP header + */ + struct tcphdr saved_tcphdr = {}; + so_recv_data_stat(so, q->tqe_m, 0); /* XXXX */ - if (so->so_flags & SOF_ENABLE_MSGS) { - /* - * Append the inorder data as a message to the - * receive socket buffer. Also check to see if - * the data we are about to deliver is the same - * data that we wanted to pass up to the user - * out of order. If so, reset copy_oodata -- - * the received data filled a gap, and - * is now in order! - */ - if (q == te) - copy_oodata = 0; + memcpy(&saved_tcphdr, th, sizeof(struct tcphdr)); + + if (q->tqe_th->th_flags & TH_PUSH) { + tp->t_flagsext |= TF_LAST_IS_PSH; + } else { + tp->t_flagsext &= ~TF_LAST_IS_PSH; } - if (sbappendstream_rcvdemux(so, q->tqe_m, - q->tqe_th->th_seq - (tp->irs + 1), 0)) - dowakeup = 1; - if (tp->t_flagsext & TF_LRO_OFFLOADED) { - tcp_update_lro_seq(tp->rcv_nxt, - inp->inp_laddr, inp->inp_faddr, - th->th_dport, th->th_sport); + + if (sbappendstream_rcvdemux(so, q->tqe_m)) { + *dowakeup = 1; } + th = &saved_tcphdr; } zfree(tcp_reass_zone, q); tp->t_reassqlen--; @@ -877,47 +816,21 @@ present: } while (q && q->tqe_th->th_seq == tp->rcv_nxt); tp->t_flagsext &= ~TF_REASS_INPROG; -#if INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { - KERNEL_DEBUG(DBG_LAYER_BEG, - ((inp->inp_fport << 16) | inp->inp_lport), - (((inp->in6p_laddr.s6_addr16[0] & 0xffff) << 16) | - (inp->in6p_faddr.s6_addr16[0] & 0xffff)), - 0,0,0); - } - else -#endif - { + ((inp->inp_fport << 16) | inp->inp_lport), + (((inp->in6p_laddr.s6_addr16[0] & 0xffff) << 16) | + (inp->in6p_faddr.s6_addr16[0] & 0xffff)), + 0, 0, 0); + } else { KERNEL_DEBUG(DBG_LAYER_BEG, - ((inp->inp_fport << 16) | inp->inp_lport), - (((inp->inp_laddr.s_addr & 0xffff) << 16) | - (inp->inp_faddr.s_addr & 0xffff)), - 0,0,0); - } - -msg_unordered_delivery: - /* Deliver out-of-order data as a message */ - if (te && (so->so_flags & SOF_ENABLE_MSGS) && copy_oodata && te->tqe_len) { - /* - * make a copy of the mbuf to be delivered up to - * the user, and add it to the sockbuf - */ - oodata = m_copym(te->tqe_m, 0, M_COPYALL, M_DONTWAIT); - if (oodata != NULL) { - if (sbappendmsgstream_rcv(&so->so_rcv, oodata, - te->tqe_th->th_seq - (tp->irs + 1), 1)) { - dowakeup = 1; - tcpstat.tcps_msg_unopkts++; - } else { - tcpstat.tcps_msg_unoappendfail++; - } - } + ((inp->inp_fport << 16) | inp->inp_lport), + (((inp->inp_laddr.s_addr & 0xffff) << 16) | + (inp->inp_faddr.s_addr & 0xffff)), + 0, 0, 0); } - if (dowakeup) - sorwakeup(so); /* done with socket lock held */ - return (flags); + return flags; } /* @@ -925,21 +838,22 @@ msg_unordered_delivery: * probe recovers the last packet. */ static void -tcp_reduce_congestion_window( - struct tcpcb *tp) +tcp_reduce_congestion_window(struct tcpcb *tp) { /* * If the current tcp cc module has * defined a hook for tasks to run * before entering FR, call it */ - if (CC_ALGO(tp)->pre_fr != NULL) + if (CC_ALGO(tp)->pre_fr != NULL) { CC_ALGO(tp)->pre_fr(tp); + } ENTER_FASTRECOVERY(tp); - if (tp->t_flags & TF_SENTFIN) + if (tp->t_flags & TF_SENTFIN) { tp->snd_recover = tp->snd_max - 1; - else + } else { tp->snd_recover = tp->snd_max; + } tp->t_timer[TCPT_REXMT] = 0; tp->t_timer[TCPT_PTO] = 0; tp->t_rtttime = 0; @@ -973,7 +887,7 @@ tcp_adaptive_rwtimo_check(struct tcpcb *tp, int tlen) && tlen > 0 && tp->t_state == TCPS_ESTABLISHED) { tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, - (TCP_REXMTVAL(tp) << 1)); + (TCP_REXMTVAL(tp) << 1)); tp->t_flagsext |= TF_DETECT_READSTALL; tp->t_rtimo_probes = 0; } @@ -983,7 +897,7 @@ inline void tcp_keepalive_reset(struct tcpcb *tp) { tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, - TCP_CONN_KEEPIDLE(tp)); + TCP_CONN_KEEPIDLE(tp)); tp->t_flagsext &= ~(TF_DETECT_READSTALL); tp->t_rtimo_probes = 0; } @@ -992,7 +906,6 @@ tcp_keepalive_reset(struct tcpcb *tp) * TCP input routine, follows pages 65-76 of the * protocol specification dated September, 1981 very closely. */ -#if INET6 int tcp6_input(struct mbuf **mp, int *offp, int proto) { @@ -1021,14 +934,13 @@ tcp6_input(struct mbuf **mp, int *offp, int proto) IF_TCP_STATINC(ifp, icmp6unreach); - return (IPPROTO_DONE); + return IPPROTO_DONE; } } tcp_input(m, *offp); - return (IPPROTO_DONE); + return IPPROTO_DONE; } -#endif /* Depending on the usage of mbuf space in the system, this function * will return true or false. This is used to determine if a socket @@ -1050,17 +962,17 @@ tcp_cansbgrow(struct sockbuf *sb) u_int32_t sbspacelim = ((nmbclusters >> 4) << MCLSHIFT); if ((total_sbmb_cnt < mblim) && - (sb->sb_hiwat < sbspacelim)) { - return(1); + (sb->sb_hiwat < sbspacelim)) { + return 1; } else { OSIncrementAtomic64(&sbmb_limreached); } - return(0); + return 0; } static void tcp_sbrcv_reserve(struct tcpcb *tp, struct sockbuf *sbrcv, - u_int32_t newsize, u_int32_t idealsize, u_int32_t rcvbuf_max) + u_int32_t newsize, u_int32_t idealsize, u_int32_t rcvbuf_max) { /* newsize should not exceed max */ newsize = min(newsize, rcvbuf_max); @@ -1073,7 +985,7 @@ tcp_sbrcv_reserve(struct tcpcb *tp, struct sockbuf *sbrcv, /* Set new socket buffer size */ if (newsize > sbrcv->sb_hiwat && - (sbreserve(sbrcv, newsize) == 1)) { + (sbreserve(sbrcv, newsize) == 1)) { sbrcv->sb_idealsize = min(max(sbrcv->sb_idealsize, (idealsize != 0) ? idealsize : newsize), rcvbuf_max); @@ -1081,7 +993,7 @@ tcp_sbrcv_reserve(struct tcpcb *tp, struct sockbuf *sbrcv, * window scale */ sbrcv->sb_idealsize = min(sbrcv->sb_idealsize, - TCP_MAXWIN << tp->rcv_scale); + TCP_MAXWIN << tp->rcv_scale); } } @@ -1092,7 +1004,7 @@ tcp_sbrcv_reserve(struct tcpcb *tp, struct sockbuf *sbrcv, */ static void tcp_sbrcv_grow(struct tcpcb *tp, struct sockbuf *sbrcv, - struct tcpopt *to, u_int32_t pktlen, u_int32_t rcvbuf_max) + struct tcpopt *to, uint32_t pktlen) { struct socket *so = sbrcv->sb_so; @@ -1102,30 +1014,18 @@ tcp_sbrcv_grow(struct tcpcb *tp, struct sockbuf *sbrcv, * - the high water mark already reached the maximum * - the stream is in background and receive side is being * throttled - * - if there are segments in reassembly queue indicating loss, - * do not need to increase recv window during recovery as more - * data is not going to be sent. A duplicate ack sent during - * recovery should not change the receive window */ if (tcp_do_autorcvbuf == 0 || - (sbrcv->sb_flags & SB_AUTOSIZE) == 0 || - tcp_cansbgrow(sbrcv) == 0 || - sbrcv->sb_hiwat >= rcvbuf_max || - (tp->t_flagsext & TF_RECV_THROTTLE) || - (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) || - !LIST_EMPTY(&tp->t_segq)) { + (sbrcv->sb_flags & SB_AUTOSIZE) == 0 || + tcp_cansbgrow(sbrcv) == 0 || + sbrcv->sb_hiwat >= tcp_autorcvbuf_max || + (tp->t_flagsext & TF_RECV_THROTTLE) || + (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) || + (!tcp_autotune_reorder && !LIST_EMPTY(&tp->t_segq))) { /* Can not resize the socket buffer, just return */ goto out; } - if (TSTMP_GT(tcp_now, - tp->rfbuf_ts + TCPTV_RCVBUFIDLE)) { - /* If there has been an idle period in the - * connection, just restart the measurement - */ - goto out; - } - if (!TSTMP_SUPPORTED(tp)) { /* * Timestamp option is not supported on this connection. @@ -1134,8 +1034,8 @@ tcp_sbrcv_grow(struct tcpcb *tp, struct sockbuf *sbrcv, * the high water mark. */ if (TSTMP_GEQ(tcp_now, - tp->rfbuf_ts + TCPTV_RCVNOTS_QUANTUM)) { - if (tp->rfbuf_cnt >= TCP_RCVNOTS_BYTELEVEL) { + tp->rfbuf_ts + TCPTV_RCVNOTS_QUANTUM)) { + if (tp->rfbuf_cnt + pktlen >= TCP_RCVNOTS_BYTELEVEL) { tcp_sbrcv_reserve(tp, sbrcv, tcp_autorcvbuf_max, 0, tcp_autorcvbuf_max); @@ -1156,37 +1056,67 @@ tcp_sbrcv_grow(struct tcpcb *tp, struct sockbuf *sbrcv, * on the link. */ if (TSTMP_GEQ(to->to_tsecr, tp->rfbuf_ts)) { - if (tp->rfbuf_cnt > (sbrcv->sb_hiwat - - (sbrcv->sb_hiwat >> 1))) { - int32_t rcvbuf_inc, min_incr; - /* - * Increment the receive window by a - * multiple of maximum sized segments. - * This will prevent a connection from - * sending smaller segments on wire if it - * is limited by the receive window. - * - * Set the ideal size based on current - * bandwidth measurements. We set the - * ideal size on receive socket buffer to - * be twice the bandwidth delay product. - */ - rcvbuf_inc = (tp->rfbuf_cnt << 1) - - sbrcv->sb_hiwat; + if (tcp_aggressive_rcvwnd_inc) { + tp->rfbuf_cnt += pktlen; + } - /* - * Make the increment equal to 8 segments - * at least - */ - min_incr = tp->t_maxseg << tcp_autorcvbuf_inc_shift; - if (rcvbuf_inc < min_incr) - rcvbuf_inc = min_incr; + if ((tcp_aggressive_rcvwnd_inc == 0 && + tp->rfbuf_cnt + pktlen > (sbrcv->sb_hiwat - + (sbrcv->sb_hiwat >> 1))) || + (tcp_aggressive_rcvwnd_inc && + tp->rfbuf_cnt > tp->rfbuf_space)) { + int32_t rcvbuf_inc; + uint32_t idealsize; - rcvbuf_inc = - (rcvbuf_inc / tp->t_maxseg) * tp->t_maxseg; - tcp_sbrcv_reserve(tp, sbrcv, - sbrcv->sb_hiwat + rcvbuf_inc, - (tp->rfbuf_cnt * 2), rcvbuf_max); + if (tcp_aggressive_rcvwnd_inc == 0) { + int32_t min_incr; + + tp->rfbuf_cnt += pktlen; + /* + * Increment the receive window by a + * multiple of maximum sized segments. + * This will prevent a connection from + * sending smaller segments on wire if it + * is limited by the receive window. + * + * Set the ideal size based on current + * bandwidth measurements. We set the + * ideal size on receive socket buffer to + * be twice the bandwidth delay product. + */ + rcvbuf_inc = (tp->rfbuf_cnt << 1) + - sbrcv->sb_hiwat; + + /* + * Make the increment equal to 8 segments + * at least + */ + min_incr = tp->t_maxseg << tcp_autorcvbuf_inc_shift; + if (rcvbuf_inc < min_incr) { + rcvbuf_inc = min_incr; + } + + idealsize = (tp->rfbuf_cnt << 1); + } else { + if (tp->rfbuf_cnt > tp->rfbuf_space + (tp->rfbuf_space >> 1)) { + rcvbuf_inc = (tp->rfbuf_cnt << 2) - sbrcv->sb_hiwat; + idealsize = (tp->rfbuf_cnt << 2); + } else { + rcvbuf_inc = (tp->rfbuf_cnt << 1) - sbrcv->sb_hiwat; + idealsize = (tp->rfbuf_cnt << 1); + } + } + + tp->rfbuf_space = tp->rfbuf_cnt; + + if (rcvbuf_inc > 0) { + rcvbuf_inc = + (rcvbuf_inc / tp->t_maxseg) * tp->t_maxseg; + + tcp_sbrcv_reserve(tp, sbrcv, + sbrcv->sb_hiwat + rcvbuf_inc, + idealsize, tcp_autorcvbuf_max); + } } /* Measure instantaneous receive bandwidth */ if (tp->t_bwmeas != NULL && tp->rfbuf_cnt > 0 && @@ -1198,7 +1128,7 @@ tcp_sbrcv_grow(struct tcpcb *tp, struct sockbuf *sbrcv, tp->t_bwmeas->bw_rcvbw_max = rcv_bw; } else { tp->t_bwmeas->bw_rcvbw_max = max( - tp->t_bwmeas->bw_rcvbw_max, rcv_bw); + tp->t_bwmeas->bw_rcvbw_max, rcv_bw); } } goto out; @@ -1209,7 +1139,7 @@ tcp_sbrcv_grow(struct tcpcb *tp, struct sockbuf *sbrcv, } out: /* Restart the measurement */ - tp->rfbuf_ts = 0; + tp->rfbuf_ts = tcp_now; tp->rfbuf_cnt = 0; return; } @@ -1223,7 +1153,7 @@ void tcp_sbrcv_trim(struct tcpcb *tp, struct sockbuf *sbrcv) { if (tcp_do_autorcvbuf == 1 && sbrcv->sb_idealsize > 0 && - sbrcv->sb_hiwat > sbrcv->sb_idealsize) { + sbrcv->sb_hiwat > sbrcv->sb_idealsize) { int32_t trim; /* compute the difference between ideal and current sizes */ u_int32_t diff = sbrcv->sb_hiwat - sbrcv->sb_idealsize; @@ -1241,16 +1171,18 @@ tcp_sbrcv_trim(struct tcpcb *tp, struct sockbuf *sbrcv) u_int32_t leave = max(advwin, (sbrcv->sb_idealsize >> 4)); /* Sometimes leave can be zero, in that case leave at least - * a few segments worth of space. + * a few segments worth of space. */ - if (leave == 0) + if (leave == 0) { leave = tp->t_maxseg << tcp_autorcvbuf_inc_shift; + } trim = sbrcv->sb_hiwat - (sbrcv->sb_cc + leave); trim = imin(trim, (int32_t)diff); - if (trim > 0) + if (trim > 0) { sbreserve(sbrcv, (sbrcv->sb_hiwat - trim)); + } } } @@ -1263,11 +1195,10 @@ tcp_sbrcv_trim(struct tcpcb *tp, struct sockbuf *sbrcv) void tcp_sbsnd_trim(struct sockbuf *sbsnd) { - if (tcp_do_autosendbuf == 1 && - ((sbsnd->sb_flags & (SB_AUTOSIZE | SB_TRIM)) == - (SB_AUTOSIZE | SB_TRIM)) && - (sbsnd->sb_idealsize > 0) && - (sbsnd->sb_hiwat > sbsnd->sb_idealsize)) { + if (((sbsnd->sb_flags & (SB_AUTOSIZE | SB_TRIM)) == + (SB_AUTOSIZE | SB_TRIM)) && + (sbsnd->sb_idealsize > 0) && + (sbsnd->sb_hiwat > sbsnd->sb_idealsize)) { u_int32_t trim = 0; if (sbsnd->sb_cc <= sbsnd->sb_idealsize) { trim = sbsnd->sb_hiwat - sbsnd->sb_idealsize; @@ -1276,8 +1207,9 @@ tcp_sbsnd_trim(struct sockbuf *sbsnd) } sbreserve(sbsnd, (sbsnd->sb_hiwat - trim)); } - if (sbsnd->sb_hiwat <= sbsnd->sb_idealsize) + if (sbsnd->sb_hiwat <= sbsnd->sb_idealsize) { sbsnd->sb_flags &= ~(SB_TRIM); + } } /* @@ -1295,8 +1227,8 @@ tcp_sbrcv_tstmp_check(struct tcpcb *tp) struct sockbuf *sbrcv = &so->so_rcv; if ((tp->t_flags & (TF_REQ_TSTMP | TF_RCVD_TSTMP)) != - (TF_REQ_TSTMP | TF_RCVD_TSTMP) && - (sbrcv->sb_flags & SB_AUTOSIZE) != 0) { + (TF_REQ_TSTMP | TF_RCVD_TSTMP) && + (sbrcv->sb_flags & SB_AUTOSIZE) != 0) { tcp_sbrcv_reserve(tp, sbrcv, newsize, 0, newsize); } } @@ -1324,21 +1256,24 @@ static inline int tcp_stretch_ack_enable(struct tcpcb *tp, int thflags) { if (tp->rcv_by_unackwin >= (maxseg_unacked * tp->t_maxseg) && - TSTMP_GEQ(tp->rcv_unackwin, tcp_now)) + TSTMP_GEQ(tp->rcv_unackwin, tcp_now)) { tp->t_flags |= TF_STREAMING_ON; - else + } else { tp->t_flags &= ~TF_STREAMING_ON; + } /* If there has been an idle time, reset streaming detection */ - if (TSTMP_GT(tcp_now, tp->rcv_unackwin + tcp_maxrcvidle)) + if (TSTMP_GT(tcp_now, tp->rcv_unackwin + tcp_maxrcvidle)) { tp->t_flags &= ~TF_STREAMING_ON; + } /* * If there are flags other than TH_ACK set, reset streaming * detection */ - if (thflags & ~TH_ACK) + if (thflags & ~TH_ACK) { tp->t_flags &= ~TF_STREAMING_ON; + } if (tp->t_flagsext & TF_DISABLE_STRETCHACK) { if (tp->rcv_nostrack_pkts >= TCP_STRETCHACK_ENABLE_PKTCNT) { @@ -1350,14 +1285,14 @@ tcp_stretch_ack_enable(struct tcpcb *tp, int thflags) } } - if (!(tp->t_flagsext & (TF_NOSTRETCHACK|TF_DISABLE_STRETCHACK)) && + if (!(tp->t_flagsext & (TF_NOSTRETCHACK | TF_DISABLE_STRETCHACK)) && (tp->t_flags & TF_STREAMING_ON) && (!(tp->t_flagsext & TF_RCVUNACK_WAITSS) || (tp->rcv_waitforss >= tcp_rcvsspktcnt))) { - return(1); + return 1; } - return(0); + return 0; } /* @@ -1369,8 +1304,9 @@ tcp_stretch_ack_enable(struct tcpcb *tp, int thflags) void tcp_reset_stretch_ack(struct tcpcb *tp) { - tp->t_flags &= ~(TF_STRETCHACK|TF_STREAMING_ON); + tp->t_flags &= ~(TF_STRETCHACK | TF_STREAMING_ON); tp->rcv_by_unackwin = 0; + tp->rcv_by_unackhalfwin = 0; tp->rcv_unackwin = tcp_now + tcp_rcvunackwin; /* @@ -1397,29 +1333,30 @@ tcp_reset_stretch_ack(struct tcpcb *tp) */ int tcp_detect_bad_rexmt(struct tcpcb *tp, struct tcphdr *th, - struct tcpopt *to, u_int32_t rxtime) + struct tcpopt *to, u_int32_t rxtime) { int32_t tdiff, bad_rexmt_win; bad_rexmt_win = (tp->t_srtt >> (TCP_RTT_SHIFT + 1)); /* If the ack has ECN CE bit, then cwnd has to be adjusted */ - if (TCP_ECN_ENABLED(tp) && (th->th_flags & TH_ECE)) - return (0); + if (TCP_ECN_ENABLED(tp) && (th->th_flags & TH_ECE)) { + return 0; + } if (TSTMP_SUPPORTED(tp)) { - if (rxtime > 0 && (to->to_flags & TOF_TS) - && to->to_tsecr != 0 - && TSTMP_LT(to->to_tsecr, rxtime)) - return (1); + if (rxtime > 0 && (to->to_flags & TOF_TS) && to->to_tsecr != 0 && + TSTMP_LT(to->to_tsecr, rxtime)) { + return 1; + } } else { - if ((tp->t_rxtshift == 1 - || (tp->t_flagsext & TF_SENT_TLPROBE)) - && rxtime > 0) { + if ((tp->t_rxtshift == 1 || (tp->t_flagsext & TF_SENT_TLPROBE)) && + rxtime > 0) { tdiff = (int32_t)(tcp_now - rxtime); - if (tdiff < bad_rexmt_win) - return(1); + if (tdiff < bad_rexmt_win) { + return 1; + } } } - return(0); + return 0; } @@ -1442,21 +1379,22 @@ tcp_bad_rexmt_restore_state(struct tcpcb *tp, struct tcphdr *th) tp->snd_ssthresh = tp->snd_ssthresh_prev; /* Initialize cwnd to the initial window */ - if (CC_ALGO(tp)->cwnd_init != NULL) + if (CC_ALGO(tp)->cwnd_init != NULL) { CC_ALGO(tp)->cwnd_init(tp); + } tp->snd_cwnd = fsize + min(acked, tp->snd_cwnd); - } else { tp->snd_cwnd = tp->snd_cwnd_prev; tp->snd_ssthresh = tp->snd_ssthresh_prev; - if (tp->t_flags & TF_WASFRECOVERY) + if (tp->t_flags & TF_WASFRECOVERY) { ENTER_FASTRECOVERY(tp); + } /* Do not use the loss flight size in this case */ tp->t_lossflightsize = 0; } - tp->snd_cwnd = max(tp->snd_cwnd, TCP_CC_CWND_INIT_BYTES); + tp->snd_cwnd = max(tp->snd_cwnd, tcp_initial_cwnd(tp)); tp->snd_recover = tp->snd_recover_prev; tp->snd_nxt = tp->snd_max; @@ -1492,10 +1430,9 @@ tcp_bad_rexmt_check(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) ++tcpstat.tcps_sndrexmitbad; tcp_bad_rexmt_restore_state(tp, th); tcp_ccdbg_trace(tp, th, TCP_CC_BAD_REXMT_RECOVERY); - } else if ((tp->t_flagsext & TF_SENT_TLPROBE) - && tp->t_tlphighrxt > 0 - && SEQ_GEQ(th->th_ack, tp->t_tlphighrxt) - && !tcp_detect_bad_rexmt(tp, th, to, tp->t_tlpstart)) { + } else if ((tp->t_flagsext & TF_SENT_TLPROBE) && tp->t_tlphighrxt > 0 && + SEQ_GEQ(th->th_ack, tp->t_tlphighrxt) && + !tcp_detect_bad_rexmt(tp, th, to, tp->t_tlpstart)) { /* * check DSACK information also to make sure that * the TLP was indeed needed @@ -1548,8 +1485,11 @@ out: tcp_pmtud_revert_segment_size(tp); } } - if (tp->t_pmtud_start_ts > 0) + if (tp->t_pmtud_start_ts > 0) { tp->t_pmtud_start_ts = 0; + } + + tp->t_pmtud_lastseg_size = 0; } /* @@ -1565,18 +1505,15 @@ out: * packet reordering. */ static void -tcp_early_rexmt_check (struct tcpcb *tp, struct tcphdr *th) +tcp_early_rexmt_check(struct tcpcb *tp, struct tcphdr *th) { u_int32_t obytes, snd_off; int32_t snd_len; struct socket *so = tp->t_inpcb->inp_socket; - if (early_rexmt && (SACK_ENABLED(tp) || - tp->t_early_rexmt_count < TCP_EARLY_REXMT_LIMIT) && + if ((SACK_ENABLED(tp) || tp->t_early_rexmt_count < TCP_EARLY_REXMT_LIMIT) && SEQ_GT(tp->snd_max, tp->snd_una) && - (tp->t_dupacks == 1 || - (SACK_ENABLED(tp) && - !TAILQ_EMPTY(&tp->snd_holes)))) { + (tp->t_dupacks == 1 || (SACK_ENABLED(tp) && !TAILQ_EMPTY(&tp->snd_holes)))) { /* * If there are only a few outstanding * segments on the connection, we might need @@ -1587,7 +1524,7 @@ tcp_early_rexmt_check (struct tcpcb *tp, struct tcphdr *th) if (SACK_ENABLED(tp) && !TAILQ_EMPTY(&tp->snd_holes)) { obytes = (tp->snd_max - tp->snd_fack) + - tp->sackhint.sack_bytes_rexmit; + tp->sackhint.sack_bytes_rexmit; } else { obytes = (tp->snd_max - tp->snd_una); } @@ -1608,8 +1545,9 @@ tcp_early_rexmt_check (struct tcpcb *tp, struct tcphdr *th) u_int32_t osegs; osegs = obytes / tp->t_maxseg; - if ((osegs * tp->t_maxseg) < obytes) + if ((osegs * tp->t_maxseg) < obytes) { osegs++; + } /* * Since the connection might have already @@ -1632,8 +1570,9 @@ tcp_early_rexmt_check (struct tcpcb *tp, struct tcphdr *th) tp->t_rexmtthresh = max(tp->t_rexmtthresh, tp->t_dupacks); - if (tp->t_early_rexmt_count == 0) + if (tp->t_early_rexmt_count == 0) { tp->t_early_rexmt_win = tcp_now; + } if (tp->t_flagsext & TF_SENT_TLPROBE) { tcpstat.tcps_tlp_recovery++; @@ -1667,15 +1606,16 @@ tcp_tfo_syn(struct tcpcb *tp, struct tcpopt *to) unsigned char len; if (!(to->to_flags & (TOF_TFO | TOF_TFOREQ)) || - !(tcp_fastopen & TCP_FASTOPEN_SERVER)) - return (FALSE); + !(tcp_fastopen & TCP_FASTOPEN_SERVER)) { + return FALSE; + } if ((to->to_flags & TOF_TFOREQ)) { tp->t_tfo_flags |= TFO_F_OFFER_COOKIE; tp->t_tfo_stats |= TFO_S_COOKIEREQ_RECV; tcpstat.tcps_tfo_cookie_req_rcv++; - return (FALSE); + return FALSE; } /* Ok, then it must be an offered cookie. We need to check that ... */ @@ -1689,13 +1629,13 @@ tcp_tfo_syn(struct tcpcb *tp, struct tcpopt *to) tp->t_tfo_stats |= TFO_S_COOKIE_INVALID; tcpstat.tcps_tfo_cookie_invalid++; - return (FALSE); + return FALSE; } if (OSIncrementAtomic(&tcp_tfo_halfcnt) >= tcp_tfo_backlog) { /* Need to decrement again as we just increased it... */ OSDecrementAtomic(&tcp_tfo_halfcnt); - return (FALSE); + return FALSE; } tp->t_tfo_flags |= TFO_F_COOKIE_VALID; @@ -1703,7 +1643,7 @@ tcp_tfo_syn(struct tcpcb *tp, struct tcpopt *to) tp->t_tfo_stats |= TFO_S_SYNDATA_RCV; tcpstat.tcps_tfo_syn_data_rcv++; - return (TRUE); + return TRUE; } static void @@ -1736,7 +1676,8 @@ tcp_tfo_synack(struct tcpcb *tp, struct tcpopt *to) * rexmit the SYN. If that's the case, it's better to start * backing of TFO-cookie requests. */ - if (tp->t_tfo_flags & TFO_F_SYN_LOSS) { + if (!(tp->t_flagsext & TF_FASTOPEN_FORCE_ENABLE) && + tp->t_tfo_flags & TFO_F_SYN_LOSS) { tp->t_tfo_stats |= TFO_S_SYN_LOSS; tcpstat.tcps_tfo_syn_loss++; @@ -1755,8 +1696,9 @@ tcp_tfo_synack(struct tcpcb *tp, struct tcpopt *to) static void tcp_tfo_rcv_probe(struct tcpcb *tp, int tlen) { - if (tlen != 0) + if (tlen != 0) { return; + } tp->t_tfo_probe_state = TFO_PROBE_PROBING; @@ -1771,8 +1713,9 @@ static void tcp_tfo_rcv_data(struct tcpcb *tp) { /* Transition from PROBING to NONE as data has been received */ - if (tp->t_tfo_probe_state >= TFO_PROBE_PROBING) + if (tp->t_tfo_probe_state >= TFO_PROBE_PROBING) { tp->t_tfo_probe_state = TFO_PROBE_NONE; + } } static void @@ -1806,24 +1749,72 @@ tcp_update_window(struct tcpcb *tp, int thflags, struct tcphdr * th, (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { /* keep track of pure window updates */ if (tlen == 0 && - tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) + tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) { tcpstat.tcps_rcvwinupd++; + } tp->snd_wnd = tiwin; tp->snd_wl1 = th->th_seq; tp->snd_wl2 = th->th_ack; - if (tp->snd_wnd > tp->max_sndwnd) + if (tp->snd_wnd > tp->max_sndwnd) { tp->max_sndwnd = tp->snd_wnd; + } + + if (tp->t_inpcb->inp_socket->so_flags & SOF_MP_SUBFLOW) { + mptcp_update_window_wakeup(tp); + } + return true; + } + return false; +} + +static void +tcp_handle_wakeup(struct socket *so, int read_wakeup, int write_wakeup) +{ + if (read_wakeup != 0) { + sorwakeup(so); + } + if (write_wakeup != 0) { + sowwakeup(so); + } +} + +static void +tcp_update_snd_una(struct tcpcb *tp, uint32_t ack) +{ + tp->snd_una = ack; + if (SACK_ENABLED(tp) && SEQ_LT(tp->send_highest_sack, tp->snd_una)) { + tp->send_highest_sack = tp->snd_una; + + /* If we move our marker, we need to start fresh */ + tp->t_new_dupacks = 0; + } +} + +static bool +tcp_syn_data_valid(struct tcpcb *tp, struct tcphdr *tcp_hdr, int tlen) +{ + /* No data? */ + if (tlen <= 0) { + return false; + } - if (tp->t_inpcb->inp_socket->so_flags & SOF_MP_SUBFLOW) - mptcp_update_window_fallback(tp); - return (true); + /* Not the right sequence-number? */ + if (tcp_hdr->th_seq != tp->irs) { + return false; } - return (false); + + /* We could have wrapped around, check that */ + if (tp->t_inpcb->inp_stat->rxbytes > INT32_MAX) { + return false; + } + + return true; } void tcp_input(struct mbuf *m, int off0) { + int exiting_fr = 0; struct tcphdr *th; struct ip *ip = NULL; struct inpcb *inp; @@ -1835,28 +1826,24 @@ tcp_input(struct mbuf *m, int off0) int thflags; struct socket *so = 0; int todrop, acked, ourfinisacked, needoutput = 0; + int read_wakeup = 0; + int write_wakeup = 0; struct in_addr laddr; -#if INET6 struct in6_addr laddr6; -#endif int dropsocket = 0; int iss = 0, nosock = 0; - u_int32_t tiwin, sack_bytes_acked = 0; - struct tcpopt to; /* options in this segment */ + u_int32_t tiwin, sack_bytes_acked = 0, sack_bytes_newly_acked = 0; + struct tcpopt to; /* options in this segment */ #if TCPDEBUG short ostate = 0; #endif -#if IPFIREWALL - struct sockaddr_in *next_hop = NULL; - struct m_tag *fwd_tag; -#endif /* IPFIREWALL */ u_char ip_ecn = IPTOS_ECN_NOTECT; unsigned int ifscope; uint8_t isconnected, isdisconnected; struct ifnet *ifp = m->m_pkthdr.rcvif; - int pktf_sw_lro_pkt = (m->m_pkthdr.pkt_flags & PKTF_SW_LRO_PKT) ? 1 : 0; - int nlropkts = (pktf_sw_lro_pkt == 1) ? m->m_pkthdr.lro_npkts : 1; - int turnoff_lro = 0, win; + int segment_count = m->m_pkthdr.seg_cnt ? : 1; + int win; + u_int16_t pf_tag = 0; #if MPTCP struct mptcb *mp_tp = NULL; #endif /* MPTCP */ @@ -1865,44 +1852,45 @@ tcp_input(struct mbuf *m, int off0) boolean_t wired = (!wifi && IFNET_IS_WIRED(ifp)); boolean_t recvd_dsack = FALSE; struct tcp_respond_args tra; + int prev_t_state; + boolean_t check_cfil = cfil_filter_present(); + bool findpcb_iterated = false; + /* + * The mbuf may be freed after it has been added to the receive socket + * buffer or the reassembly queue, so we reinitialize th to point to a + * safe copy of the TCP header + */ + struct tcphdr saved_tcphdr = {}; + /* + * Save copy of the IPv4/IPv6 header. + * Note: use array of uint32_t to silence compiler warning when casting + * to a struct ip6_hdr pointer. + */ +#define MAX_IPWORDS ((sizeof(struct ip) + MAX_IPOPTLEN) / sizeof(uint32_t)) + uint32_t saved_hdr[MAX_IPWORDS]; -#define TCP_INC_VAR(stat, npkts) do { \ - stat += npkts; \ +#define TCP_INC_VAR(stat, npkts) do { \ + stat += npkts; \ } while (0) - TCP_INC_VAR(tcpstat.tcps_rcvtotal, nlropkts); -#if IPFIREWALL - /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ - if (!SLIST_EMPTY(&m->m_pkthdr.tags)) { - fwd_tag = m_tag_locate(m, KERNEL_MODULE_TAG_ID, - KERNEL_TAG_TYPE_IPFORWARD, NULL); - } else { - fwd_tag = NULL; - } - if (fwd_tag != NULL) { - struct ip_fwd_tag *ipfwd_tag = - (struct ip_fwd_tag *)(fwd_tag+1); - - next_hop = ipfwd_tag->next_hop; - m_tag_delete(m, fwd_tag); + if (tcp_ack_strategy == TCP_ACK_STRATEGY_LEGACY) { + segment_count = 1; } -#endif /* IPFIREWALL */ + TCP_INC_VAR(tcpstat.tcps_rcvtotal, segment_count); -#if INET6 struct ip6_hdr *ip6 = NULL; int isipv6; -#endif /* INET6 */ - int rstreason; /* For badport_bandlim accounting purposes */ - struct proc *proc0=current_proc(); + struct proc *kernel_proc = current_proc(); - KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_START,0,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_START, 0, 0, 0, 0, 0); -#if INET6 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; -#endif bzero((char *)&to, sizeof(to)); -#if INET6 + if (m->m_flags & M_PKTHDR) { + pf_tag = m_pftag(m)->pftag_tag; + } + if (isipv6) { /* * Expect 32-bit aligned data pointer on @@ -1915,12 +1903,14 @@ tcp_input(struct mbuf *m, int off0) tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; th = (struct tcphdr *)(void *)((caddr_t)ip6 + off0); - if (tcp_input_checksum(AF_INET6, m, th, off0, tlen)) + if (tcp_input_checksum(AF_INET6, m, th, off0, tlen)) { + TCP_LOG_DROP_PKT(ip6, th, ifp, "IPv6 bad tcp checksum"); goto dropnosock; + } KERNEL_DEBUG(DBG_LAYER_BEG, ((th->th_dport << 16) | th->th_sport), - (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])), - th->th_seq, th->th_ack, th->th_win); + (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])), + th->th_seq, th->th_ack, th->th_win); /* * Be proactive about unspecified IPv6 address in source. * As we use all-zero to indicate unbounded/unconnected pcb, @@ -1932,78 +1922,76 @@ tcp_input(struct mbuf *m, int off0) if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { /* XXX stat */ IF_TCP_STATINC(ifp, unspecv6); + TCP_LOG_DROP_PKT(ip6, th, ifp, "src IPv6 address unspecified"); goto dropnosock; } DTRACE_TCP5(receive, struct mbuf *, m, struct inpcb *, NULL, - struct ip6_hdr *, ip6, struct tcpcb *, NULL, - struct tcphdr *, th); + struct ip6_hdr *, ip6, struct tcpcb *, NULL, + struct tcphdr *, th); ip_ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; - } else -#endif /* INET6 */ - { - /* - * Get IP and TCP header together in first mbuf. - * Note: IP leaves IP header in first mbuf. - */ - if (off0 > sizeof (struct ip)) { - ip_stripoptions(m); - off0 = sizeof(struct ip); - } - if (m->m_len < sizeof (struct tcpiphdr)) { - if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { - tcpstat.tcps_rcvshort++; - return; + } else { + /* + * Get IP and TCP header together in first mbuf. + * Note: IP leaves IP header in first mbuf. + */ + if (off0 > sizeof(struct ip)) { + ip_stripoptions(m); + off0 = sizeof(struct ip); + } + if (m->m_len < sizeof(struct tcpiphdr)) { + if ((m = m_pullup(m, sizeof(struct tcpiphdr))) == 0) { + tcpstat.tcps_rcvshort++; + return; + } } - } - - /* Expect 32-bit aligned data pointer on strict-align platforms */ - MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m); - ip = mtod(m, struct ip *); - th = (struct tcphdr *)(void *)((caddr_t)ip + off0); - tlen = ip->ip_len; + /* Expect 32-bit aligned data pointer on strict-align platforms */ + MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m); - if (tcp_input_checksum(AF_INET, m, th, off0, tlen)) - goto dropnosock; + ip = mtod(m, struct ip *); + th = (struct tcphdr *)(void *)((caddr_t)ip + off0); + tlen = ip->ip_len; -#if INET6 - /* Re-initialization for later version check */ - ip->ip_v = IPVERSION; -#endif - ip_ecn = (ip->ip_tos & IPTOS_ECN_MASK); + if (tcp_input_checksum(AF_INET, m, th, off0, tlen)) { + TCP_LOG_DROP_PKT(ip, th, ifp, "IPv4 bad tcp checksum"); + goto dropnosock; + } - DTRACE_TCP5(receive, struct mbuf *, m, struct inpcb *, NULL, - struct ip *, ip, struct tcpcb *, NULL, struct tcphdr *, th); + /* Re-initialization for later version check */ + ip->ip_v = IPVERSION; + ip_ecn = (ip->ip_tos & IPTOS_ECN_MASK); - KERNEL_DEBUG(DBG_LAYER_BEG, ((th->th_dport << 16) | th->th_sport), - (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)), - th->th_seq, th->th_ack, th->th_win); + DTRACE_TCP5(receive, struct mbuf *, m, struct inpcb *, NULL, + struct ip *, ip, struct tcpcb *, NULL, struct tcphdr *, th); + KERNEL_DEBUG(DBG_LAYER_BEG, ((th->th_dport << 16) | th->th_sport), + (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)), + th->th_seq, th->th_ack, th->th_win); } +#define TCP_LOG_HDR (isipv6 ? (void *)ip6 : (void *)ip) + /* * Check that TCP offset makes sense, * pull out TCP options and adjust length. */ off = th->th_off << 2; - if (off < sizeof (struct tcphdr) || off > tlen) { + if (off < sizeof(struct tcphdr) || off > tlen) { tcpstat.tcps_rcvbadoff++; IF_TCP_STATINC(ifp, badformat); + TCP_LOG_DROP_PKT(TCP_LOG_HDR, th, ifp, "bad tcp offset"); goto dropnosock; } - tlen -= off; /* tlen is used instead of ti->ti_len */ - if (off > sizeof (struct tcphdr)) { -#if INET6 + tlen -= off; /* tlen is used instead of ti->ti_len */ + if (off > sizeof(struct tcphdr)) { if (isipv6) { - IP6_EXTHDR_CHECK(m, off0, off, return); + IP6_EXTHDR_CHECK(m, off0, off, return ); ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)(void *)((caddr_t)ip6 + off0); - } else -#endif /* INET6 */ - { + } else { if (m->m_len < sizeof(struct ip) + off) { - if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { + if ((m = m_pullup(m, sizeof(struct ip) + off)) == 0) { tcpstat.tcps_rcvshort++; return; } @@ -2011,7 +1999,7 @@ tcp_input(struct mbuf *m, int off0) th = (struct tcphdr *)(void *)((caddr_t)ip + off0); } } - optlen = off - sizeof (struct tcphdr); + optlen = off - sizeof(struct tcphdr); optp = (u_char *)(th + 1); /* * Do quick retrieval of timestamp options ("options @@ -2021,31 +2009,29 @@ tcp_input(struct mbuf *m, int off0) * tcp_dooptions(), etc. */ if ((optlen == TCPOLEN_TSTAMP_APPA || - (optlen > TCPOLEN_TSTAMP_APPA && - optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && - *(u_int32_t *)(void *)optp == htonl(TCPOPT_TSTAMP_HDR) && - (th->th_flags & TH_SYN) == 0) { + (optlen > TCPOLEN_TSTAMP_APPA && + optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && + *(u_int32_t *)(void *)optp == htonl(TCPOPT_TSTAMP_HDR) && + (th->th_flags & TH_SYN) == 0) { to.to_flags |= TOF_TS; to.to_tsval = ntohl(*(u_int32_t *)(void *)(optp + 4)); to.to_tsecr = ntohl(*(u_int32_t *)(void *)(optp + 8)); - optp = NULL; /* we've parsed the options */ + optp = NULL; /* we've parsed the options */ } } thflags = th->th_flags; -#if TCP_DROP_SYNFIN /* - * If the drop_synfin option is enabled, drop all packets with - * both the SYN and FIN bits set. This prevents e.g. nmap from - * identifying the TCP/IP stack. + * Drop all packets with both the SYN and FIN bits set. + * This prevents e.g. nmap from identifying the TCP/IP stack. * * This is a violation of the TCP specification. */ - if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) { + if ((thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN)) { IF_TCP_STATINC(ifp, synfin); + TCP_LOG_DROP_PKT(TCP_LOG_HDR, th, ifp, "drop SYN FIN"); goto dropnosock; } -#endif /* * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, @@ -2069,20 +2055,21 @@ tcp_input(struct mbuf *m, int off0) * absence of a PCB, when scoped routing (and thus source interface * selection) are enabled. */ - if ((m->m_pkthdr.pkt_flags & PKTF_LOOP) || m->m_pkthdr.rcvif == NULL) + if ((m->m_pkthdr.pkt_flags & PKTF_LOOP) || m->m_pkthdr.rcvif == NULL) { ifscope = IFSCOPE_NONE; - else + } else { ifscope = m->m_pkthdr.rcvif->if_index; + } - /* - * Convert TCP protocol specific fields to host format. - */ + /* + * Convert TCP protocol specific fields to host format. + */ #if BYTE_ORDER != BIG_ENDIAN - NTOHL(th->th_seq); - NTOHL(th->th_ack); - NTOHS(th->th_win); - NTOHS(th->th_urp); + NTOHL(th->th_seq); + NTOHL(th->th_ack); + NTOHS(th->th_win); + NTOHS(th->th_urp); #endif /* @@ -2093,47 +2080,14 @@ findpcb: isconnected = FALSE; isdisconnected = FALSE; -#if IPFIREWALL_FORWARD - if (next_hop != NULL -#if INET6 - && isipv6 == 0 /* IPv6 support is not yet */ -#endif /* INET6 */ - ) { - /* - * Diverted. Pretend to be the destination. - * already got one like this? - */ - inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, - ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif); - if (!inp) { - /* - * No, then it's new. Try find the ambushing socket - */ - if (!next_hop->sin_port) { - inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, - th->th_sport, next_hop->sin_addr, - th->th_dport, 1, m->m_pkthdr.rcvif); - } else { - inp = in_pcblookup_hash(&tcbinfo, - ip->ip_src, th->th_sport, - next_hop->sin_addr, - ntohs(next_hop->sin_port), 1, - m->m_pkthdr.rcvif); - } - } - } else -#endif /* IPFIREWALL_FORWARD */ - { -#if INET6 - if (isipv6) + if (isipv6) { inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport, - &ip6->ip6_dst, th->th_dport, 1, - m->m_pkthdr.rcvif); - else -#endif /* INET6 */ - inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, - ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); - } + &ip6->ip6_dst, th->th_dport, 1, + m->m_pkthdr.rcvif); + } else { + inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, + ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); + } /* * Use the interface scope information from the PCB for outbound @@ -2141,8 +2095,9 @@ findpcb: * enabled, tcp_respond will use the scope of the interface where * the segment arrived on. */ - if (inp != NULL && (inp->inp_flags & INP_BOUND_IF)) + if (inp != NULL && (inp->inp_flags & INP_BOUND_IF)) { ifscope = inp->inp_boundifp->if_index; + } /* * If the state is CLOSED (i.e., TCB does not exist) then @@ -2152,74 +2107,68 @@ findpcb: */ if (inp == NULL) { if (log_in_vain) { -#if INET6 char dbuf[MAX_IPv6_STR_LEN], sbuf[MAX_IPv6_STR_LEN]; -#else /* INET6 */ - char dbuf[MAX_IPv4_STR_LEN], sbuf[MAX_IPv4_STR_LEN]; -#endif /* INET6 */ -#if INET6 if (isipv6) { inet_ntop(AF_INET6, &ip6->ip6_dst, dbuf, sizeof(dbuf)); inet_ntop(AF_INET6, &ip6->ip6_src, sbuf, sizeof(sbuf)); - } else -#endif - { + } else { inet_ntop(AF_INET, &ip->ip_dst, dbuf, sizeof(dbuf)); inet_ntop(AF_INET, &ip->ip_src, sbuf, sizeof(sbuf)); } switch (log_in_vain) { case 1: - if(thflags & TH_SYN) + if (thflags & TH_SYN) { log(LOG_INFO, - "Connection attempt to TCP %s:%d from %s:%d\n", - dbuf, ntohs(th->th_dport), - sbuf, - ntohs(th->th_sport)); + "Connection attempt to TCP %s:%d from %s:%d\n", + dbuf, ntohs(th->th_dport), + sbuf, + ntohs(th->th_sport)); + } break; case 2: log(LOG_INFO, - "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", - dbuf, ntohs(th->th_dport), sbuf, - ntohs(th->th_sport), thflags); + "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", + dbuf, ntohs(th->th_dport), sbuf, + ntohs(th->th_sport), thflags); break; case 3: case 4: if ((thflags & TH_SYN) && !(thflags & TH_ACK) && - !(m->m_flags & (M_BCAST | M_MCAST)) && -#if INET6 - ((isipv6 && !IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) || - (!isipv6 && ip->ip_dst.s_addr != ip->ip_src.s_addr)) -#else - ip->ip_dst.s_addr != ip->ip_src.s_addr -#endif - ) + !(m->m_flags & (M_BCAST | M_MCAST)) && + ((isipv6 && !IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) || + (!isipv6 && ip->ip_dst.s_addr != ip->ip_src.s_addr))) { log_in_vain_log((LOG_INFO, - "Stealth Mode connection attempt to TCP %s:%d from %s:%d\n", - dbuf, ntohs(th->th_dport), - sbuf, - ntohs(th->th_sport))); + "Stealth Mode connection attempt to TCP %s:%d from %s:%d\n", + dbuf, ntohs(th->th_dport), + sbuf, + ntohs(th->th_sport))); + } break; default: break; } } if (blackhole) { - if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type != IFT_LOOP) - + if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type != IFT_LOOP) { switch (blackhole) { case 1: - if (thflags & TH_SYN) + if (thflags & TH_SYN) { + TCP_LOG_DROP_PKT(TCP_LOG_HDR, th, ifp, "blackhole 1 syn for closed port"); goto dropnosock; + } break; case 2: + TCP_LOG_DROP_PKT(TCP_LOG_HDR, th, ifp, "blackhole 2 closed port"); goto dropnosock; default: + TCP_LOG_DROP_PKT(TCP_LOG_HDR, th, ifp, "blackhole closed port"); goto dropnosock; } + } } - rstreason = BANDLIM_RST_CLOSEDPORT; IF_TCP_STATINC(ifp, noconnnolist); + TCP_LOG_DROP_PKT(TCP_LOG_HDR, th, ifp, "closed port"); goto dropwithresetnosock; } so = inp->inp_socket; @@ -2233,39 +2182,109 @@ findpcb: #if TEMPDEBUG printf("tcp_input: no more socket for inp=%x. This shouldn't happen\n", inp); #endif + TCP_LOG_DROP_PKT(TCP_LOG_HDR, th, ifp, "inp_socket NULL"); goto dropnosock; } socket_lock(so, 1); if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { socket_unlock(so, 1); - inp = NULL; // pretend we didn't find it + inp = NULL; // pretend we didn't find it + TCP_LOG_DROP_PKT(TCP_LOG_HDR, th, ifp, "inp state WNT_STOPUSING"); goto dropnosock; } + if (!isipv6 && inp->inp_faddr.s_addr != INADDR_ANY) { + if (inp->inp_faddr.s_addr != ip->ip_src.s_addr || + inp->inp_laddr.s_addr != ip->ip_dst.s_addr || + inp->inp_fport != th->th_sport || + inp->inp_lport != th->th_dport) { + os_log_error(OS_LOG_DEFAULT, "%s 5-tuple does not match: %u:%u %u:%u\n", + __func__, + ntohs(inp->inp_fport), ntohs(th->th_sport), + ntohs(inp->inp_lport), ntohs(th->th_dport)); + if (findpcb_iterated) { + goto drop; + } + findpcb_iterated = true; + socket_unlock(so, 1); + inp = NULL; + goto findpcb; + } + } else if (isipv6 && !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) { + if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr, &ip6->ip6_src) || + !IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, &ip6->ip6_dst) || + inp->inp_fport != th->th_sport || + inp->inp_lport != th->th_dport) { + os_log_error(OS_LOG_DEFAULT, "%s 5-tuple does not match: %u:%u %u:%u\n", + __func__, + ntohs(inp->inp_fport), ntohs(th->th_sport), + ntohs(inp->inp_lport), ntohs(th->th_dport)); + if (findpcb_iterated) { + goto drop; + } + findpcb_iterated = true; + socket_unlock(so, 1); + inp = NULL; + goto findpcb; + } + } + + tp = intotcpcb(inp); + if (tp == NULL) { + IF_TCP_STATINC(ifp, noconnlist); + TCP_LOG_DROP_PKT(TCP_LOG_HDR, th, ifp, "tp is NULL"); + goto dropwithreset; + } + + /* Now that we found the tcpcb, we can adjust the TCP timestamp */ + if (to.to_flags & TOF_TS) { + to.to_tsecr -= tp->t_ts_offset; + } + + TCP_LOG_TH_FLAGS(TCP_LOG_HDR, th, tp, false, ifp); + + if (tp->t_state == TCPS_CLOSED) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "tp state TCPS_CLOSED"); + goto drop; + } + #if NECP if (so->so_state & SS_ISCONNECTED) { // Connected TCP sockets have a fully-bound local and remote, // so the policy check doesn't need to override addresses - if (!necp_socket_is_allowed_to_send_recv(inp, NULL, NULL)) { + if (!necp_socket_is_allowed_to_send_recv(inp, ifp, pf_tag, NULL, NULL, NULL, NULL)) { + TCP_LOG_DROP_NECP(TCP_LOG_HDR, th, intotcpcb(inp), false); IF_TCP_STATINC(ifp, badformat); goto drop; } } else { -#if INET6 + /* + * If the proc_uuid_policy table has been updated since the last use + * of the listening socket (i.e., the proc_uuid_policy_table_gencount + * has been updated), the flags in the socket may be out of date. + * If INP2_WANT_APP_POLICY is stale, inbound packets may + * be dropped by NECP if the socket should now match a per-app + * exception policy. + * In order to avoid this refresh the proc_uuid_policy state to + * potentially recalculate the socket's flags before checking + * with NECP. + */ + (void) inp_update_policy(inp); + if (isipv6) { if (!necp_socket_is_allowed_to_send_recv_v6(inp, - th->th_dport, th->th_sport, &ip6->ip6_dst, - &ip6->ip6_src, ifp, NULL, NULL)) { + th->th_dport, th->th_sport, &ip6->ip6_dst, + &ip6->ip6_src, ifp, pf_tag, NULL, NULL, NULL, NULL)) { + TCP_LOG_DROP_NECP(TCP_LOG_HDR, th, intotcpcb(inp), false); IF_TCP_STATINC(ifp, badformat); goto drop; } - } else -#endif - { + } else { if (!necp_socket_is_allowed_to_send_recv_v4(inp, - th->th_dport, th->th_sport, &ip->ip_dst, &ip->ip_src, - ifp, NULL, NULL)) { + th->th_dport, th->th_sport, &ip->ip_dst, &ip->ip_src, + ifp, pf_tag, NULL, NULL, NULL, NULL)) { + TCP_LOG_DROP_NECP(TCP_LOG_HDR, th, intotcpcb(inp), false); IF_TCP_STATINC(ifp, badformat); goto drop; } @@ -2273,42 +2292,38 @@ findpcb: } #endif /* NECP */ - tp = intotcpcb(inp); - if (tp == 0) { - rstreason = BANDLIM_RST_CLOSEDPORT; - IF_TCP_STATINC(ifp, noconnlist); - goto dropwithreset; - } - if (tp->t_state == TCPS_CLOSED) + prev_t_state = tp->t_state; + + /* If none of the FIN|SYN|RST|ACK flag is set, drop */ + if ((thflags & TH_ACCEPT) == 0) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "rfc5961 TH_ACCEPT == 0"); goto drop; + } /* Unscale the window into a 32-bit value. */ - if ((thflags & TH_SYN) == 0) + if ((thflags & TH_SYN) == 0) { tiwin = th->th_win << tp->snd_scale; - else + } else { tiwin = th->th_win; - -#if CONFIG_MACF_NET - if (mac_inpcb_check_deliver(inp, m, AF_INET, SOCK_STREAM)) - goto drop; -#endif + } /* Avoid processing packets while closing a listen socket */ if (tp->t_state == TCPS_LISTEN && - (so->so_options & SO_ACCEPTCONN) == 0) + (so->so_options & SO_ACCEPTCONN) == 0) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "closing a listening socket"); goto drop; + } - if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { + if (so->so_options & (SO_DEBUG | SO_ACCEPTCONN)) { #if TCPDEBUG if (so->so_options & SO_DEBUG) { ostate = tp->t_state; -#if INET6 - if (isipv6) + if (isipv6) { bcopy((char *)ip6, (char *)tcp_saveipgen, - sizeof(*ip6)); - else -#endif /* INET6 */ - bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); + sizeof(*ip6)); + } else { + bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); + } tcp_savetcp = *th; } #endif @@ -2317,13 +2332,13 @@ findpcb: struct socket *so2; struct socket *oso; struct sockaddr_storage from; -#if INET6 + struct sockaddr_storage to2; struct inpcb *oinp = sotoinpcb(so); -#endif /* INET6 */ struct ifnet *head_ifscope; unsigned int head_nocell, head_recvanyif, - head_noexpensive, head_awdl_unrestricted, - head_intcoproc_allowed; + head_noexpensive, head_awdl_unrestricted, + head_intcoproc_allowed, head_external_port, + head_noconstrained; /* Get listener's bound-to-interface, if any */ head_ifscope = (inp->inp_flags & INP_BOUND_IF) ? @@ -2334,8 +2349,10 @@ findpcb: head_recvanyif = (inp->inp_flags & INP_RECV_ANYIF); /* Get listener's no-expensive information, if any */ head_noexpensive = INP_NO_EXPENSIVE(inp); + head_noconstrained = INP_NO_CONSTRAINED(inp); head_awdl_unrestricted = INP_AWDL_UNRESTRICTED(inp); head_intcoproc_allowed = INP_INTCOPROC_ALLOWED(inp); + head_external_port = (inp->inp_flags2 & INP2_EXTERNAL_PORT); /* * If the state is LISTEN then ignore segment if it contains an RST. @@ -2343,34 +2360,37 @@ findpcb: * If it does not contain a SYN then it is not interesting; drop it. * If it is from this socket, drop it, it must be forged. */ - if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { + if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) { IF_TCP_STATINC(ifp, listbadsyn); if (thflags & TH_RST) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SYN with RST"); goto drop; } if (thflags & TH_ACK) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SYN with ACK"); tp = NULL; tcpstat.tcps_badsyn++; - rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } /* We come here if there is no SYN set */ tcpstat.tcps_badsyn++; + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "bad SYN"); goto drop; } - KERNEL_DEBUG(DBG_FNC_TCP_NEWCONN | DBG_FUNC_START,0,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_TCP_NEWCONN | DBG_FUNC_START, 0, 0, 0, 0, 0); if (th->th_dport == th->th_sport) { -#if INET6 if (isipv6) { if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, - &ip6->ip6_src)) - goto drop; - } else -#endif /* INET6 */ - if (ip->ip_dst.s_addr == ip->ip_src.s_addr) + &ip6->ip6_src)) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "bad tuple same port"); goto drop; + } + } else if (ip->ip_dst.s_addr == ip->ip_src.s_addr) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "bad tuple same IPv4 address"); + goto drop; + } } /* * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN @@ -2380,23 +2400,25 @@ findpcb: * Packets with a multicast source address should also * be discarded. */ - if (m->m_flags & (M_BCAST|M_MCAST)) + if (m->m_flags & (M_BCAST | M_MCAST)) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "mbuf M_BCAST | M_MCAST"); goto drop; -#if INET6 + } if (isipv6) { if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || - IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) + IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "IN6_IS_ADDR_MULTICAST"); goto drop; - } else -#endif - if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || - IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || - ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || - in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) + } + } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || + IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || + ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || + in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "multicast or broadcast address"); goto drop; + } -#if INET6 /* * If deprecated address is forbidden, * we do not accept SYN to deprecated interface @@ -2419,17 +2441,15 @@ findpcb: &ia6_flags) == 0) { if (ia6_flags & IN6_IFF_DEPRECATED) { tp = NULL; - rstreason = BANDLIM_RST_OPENPORT; IF_TCP_STATINC(ifp, deprecate6); + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "deprecated IPv6 address"); goto dropwithreset; } } } -#endif - if (so->so_filt) { -#if INET6 + if (so->so_filt || check_cfil) { if (isipv6) { - struct sockaddr_in6 *sin6 = (struct sockaddr_in6*)&from; + struct sockaddr_in6 *sin6 = (struct sockaddr_in6*)&from; sin6->sin6_len = sizeof(*sin6); sin6->sin6_family = AF_INET6; @@ -2437,17 +2457,33 @@ findpcb: sin6->sin6_flowinfo = 0; sin6->sin6_addr = ip6->ip6_src; sin6->sin6_scope_id = 0; - } - else -#endif - { + + sin6 = (struct sockaddr_in6*)&to2; + + sin6->sin6_len = sizeof(struct sockaddr_in6); + sin6->sin6_family = AF_INET6; + sin6->sin6_port = th->th_dport; + sin6->sin6_flowinfo = 0; + sin6->sin6_addr = ip6->ip6_dst; + sin6->sin6_scope_id = 0; + } else { struct sockaddr_in *sin = (struct sockaddr_in*)&from; sin->sin_len = sizeof(*sin); sin->sin_family = AF_INET; sin->sin_port = th->th_sport; sin->sin_addr = ip->ip_src; + + sin = (struct sockaddr_in*)&to2; + + sin->sin_len = sizeof(struct sockaddr_in); + sin->sin_family = AF_INET; + sin->sin_port = th->th_dport; + sin->sin_addr = ip->ip_dst; } + } + + if (so->so_filt) { so2 = sonewconn(so, 0, (struct sockaddr*)&from); } else { so2 = sonewconn(so, 0, NULL); @@ -2455,13 +2491,16 @@ findpcb: if (so2 == 0) { tcpstat.tcps_listendrop++; if (tcp_dropdropablreq(so)) { - if (so->so_filt) + if (so->so_filt) { so2 = sonewconn(so, 0, (struct sockaddr*)&from); - else + } else { so2 = sonewconn(so, 0, NULL); + } } - if (!so2) + if (!so2) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, " listen drop"); goto drop; + } } /* Point "inp" and "tp" in tandem to new socket */ @@ -2500,50 +2539,58 @@ findpcb: /* * Inherit restrictions from listener. */ - if (head_nocell) + if (head_nocell) { inp_set_nocellular(inp); - if (head_noexpensive) + } + if (head_noexpensive) { inp_set_noexpensive(inp); - if (head_awdl_unrestricted) + } + if (head_noconstrained) { + inp_set_noconstrained(inp); + } + if (head_awdl_unrestricted) { inp_set_awdl_unrestricted(inp); - if (head_intcoproc_allowed) + } + if (head_intcoproc_allowed) { inp_set_intcoproc_allowed(inp); + } /* * Inherit {IN,IN6}_RECV_ANYIF from listener. */ - if (head_recvanyif) + if (head_recvanyif) { inp->inp_flags |= INP_RECV_ANYIF; - else + } else { inp->inp_flags &= ~INP_RECV_ANYIF; -#if INET6 - if (isipv6) + } + + if (head_external_port) { + inp->inp_flags2 |= INP2_EXTERNAL_PORT; + } + if (isipv6) { inp->in6p_laddr = ip6->ip6_dst; - else { + } else { inp->inp_vflag &= ~INP_IPV6; inp->inp_vflag |= INP_IPV4; -#endif /* INET6 */ inp->inp_laddr = ip->ip_dst; -#if INET6 } -#endif /* INET6 */ inp->inp_lport = th->th_dport; if (in_pcbinshash(inp, 0) != 0) { /* * Undo the assignments above if we failed to * put the PCB on the hash lists. */ -#if INET6 - if (isipv6) + if (isipv6) { inp->in6p_laddr = in6addr_any; - else -#endif /* INET6 */ + } else { inp->inp_laddr.s_addr = INADDR_ANY; + } inp->inp_lport = 0; - socket_lock(oso, 0); /* release ref on parent */ + socket_lock(oso, 0); /* release ref on parent */ socket_unlock(oso, 1); + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, " in_pcbinshash failed"); goto drop; } -#if INET6 + socket_lock(oso, 0); if (isipv6) { /* * Inherit socket options from the listening @@ -2559,57 +2606,75 @@ findpcb: * calling the accept system call. */ inp->inp_flags |= - oinp->inp_flags & INP_CONTROLOPTS; - if (oinp->in6p_outputopts) + oinp->inp_flags & INP_CONTROLOPTS; + if (oinp->in6p_outputopts) { inp->in6p_outputopts = - ip6_copypktopts(oinp->in6p_outputopts, - M_NOWAIT); - } else -#endif /* INET6 */ - { + ip6_copypktopts(oinp->in6p_outputopts, + M_NOWAIT); + } + } else { inp->inp_options = ip_srcroute(); inp->inp_ip_tos = oinp->inp_ip_tos; } - socket_lock(oso, 0); #if IPSEC /* copy old policy into new socket's */ - if (sotoinpcb(oso)->inp_sp) - { + if (sotoinpcb(oso)->inp_sp) { int error = 0; /* Is it a security hole here to silently fail to copy the policy? */ - if (inp->inp_sp != NULL) + if (inp->inp_sp != NULL) { error = ipsec_init_policy(so, &inp->inp_sp); - if (error != 0 || ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) + } + if (error != 0 || ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) { printf("tcp_input: could not copy policy\n"); + } } #endif /* inherit states from the listener */ DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, - struct tcpcb *, tp, int32_t, TCPS_LISTEN); + struct tcpcb *, tp, int32_t, TCPS_LISTEN); tp->t_state = TCPS_LISTEN; - tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT|TF_NODELAY); - tp->t_flagsext |= (tp0->t_flagsext & (TF_RXTFINDROP|TF_NOTIMEWAIT|TF_FASTOPEN)); + tp->t_flags |= tp0->t_flags & (TF_NOPUSH | TF_NOOPT | TF_NODELAY); + tp->t_flagsext |= (tp0->t_flagsext & (TF_RXTFINDROP | TF_NOTIMEWAIT | TF_FASTOPEN)); tp->t_keepinit = tp0->t_keepinit; tp->t_keepcnt = tp0->t_keepcnt; tp->t_keepintvl = tp0->t_keepintvl; tp->t_adaptive_wtimo = tp0->t_adaptive_wtimo; tp->t_adaptive_rtimo = tp0->t_adaptive_rtimo; tp->t_inpcb->inp_ip_ttl = tp0->t_inpcb->inp_ip_ttl; - if ((so->so_flags & SOF_NOTSENT_LOWAT) != 0) + if ((so->so_flags & SOF_NOTSENT_LOWAT) != 0) { tp->t_notsent_lowat = tp0->t_notsent_lowat; + } tp->t_inpcb->inp_flags2 |= tp0->t_inpcb->inp_flags2 & INP2_KEEPALIVE_OFFLOAD; /* now drop the reference on the listener */ socket_unlock(oso, 1); - tcp_set_max_rwinscale(tp, so, TCP_AUTORCVBUF_MAX(ifp)); + tcp_set_max_rwinscale(tp, so); - KERNEL_DEBUG(DBG_FNC_TCP_NEWCONN | DBG_FUNC_END,0,0,0,0,0); +#if CONTENT_FILTER + if (check_cfil) { + int error = cfil_sock_attach(so2, (struct sockaddr*)&to2, (struct sockaddr*)&from, + CFS_CONNECTION_DIR_IN); + if (error != 0) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, " cfil_sock_attach failed"); + goto drop; + } + } +#endif /* CONTENT_FILTER */ + + KERNEL_DEBUG(DBG_FNC_TCP_NEWCONN | DBG_FUNC_END, 0, 0, 0, 0, 0); } } socket_lock_assert_owned(so); + if (net_mpklog_enabled && (m->m_pkthdr.rcvif->if_xflags & IFXF_MPK_LOG)) { + MPKL_TCP_INPUT(tcp_mpkl_log_object, + ntohs(tp->t_inpcb->inp_lport), ntohs(tp->t_inpcb->inp_fport), + th->th_seq, th->th_ack, tlen, thflags, + so->last_pid, so->so_log_seqn++); + } + if (tp->t_state == TCPS_ESTABLISHED && tlen > 0) { /* * Evaluate the rate of arrival of packets to see if the @@ -2618,7 +2683,7 @@ findpcb: * certain criteria defined in tcp_stretch_ack_enable function. */ if ((tp->t_flagsext & TF_RCVUNACK_WAITSS) != 0) { - TCP_INC_VAR(tp->rcv_waitforss, nlropkts); + TCP_INC_VAR(tp->rcv_waitforss, segment_count); } if (tcp_stretch_ack_enable(tp, thflags)) { tp->t_flags |= TF_STRETCHACK; @@ -2627,29 +2692,34 @@ findpcb: } else { tp->t_flags &= ~(TF_STRETCHACK); } - if (TSTMP_GT(tp->rcv_unackwin, tcp_now)) { + if (TSTMP_GT(tp->rcv_unackwin - (tcp_rcvunackwin >> 1), tcp_now)) { + tp->rcv_by_unackhalfwin += (tlen + off); tp->rcv_by_unackwin += (tlen + off); } else { tp->rcv_unackwin = tcp_now + tcp_rcvunackwin; - tp->rcv_by_unackwin = tlen + off; + tp->rcv_by_unackwin = tp->rcv_by_unackhalfwin + tlen + off; + tp->rcv_by_unackhalfwin = tlen + off; } } /* - * Keep track of how many bytes were received in the LRO packet + * Clear TE_SENDECE if TH_CWR is set. This is harmless, so we don't + * bother doing extensive checks for state and whatnot. */ - if ((pktf_sw_lro_pkt) && (nlropkts > 2)) { - tp->t_lropktlen += tlen; + if (thflags & TH_CWR) { + tp->ecn_flags &= ~TE_SENDECE; + tp->t_ecn_recv_cwr++; } + /* * Explicit Congestion Notification - Flag that we need to send ECT if - * + The IP Congestion experienced flag was set. - * + Socket is in established state - * + We negotiated ECN in the TCP setup - * + This isn't a pure ack (tlen > 0) - * + The data is in the valid window + * + The IP Congestion experienced flag was set. + * + Socket is in established state + * + We negotiated ECN in the TCP setup + * + This isn't a pure ack (tlen > 0) + * + The data is in the valid window * - * TE_SENDECE will be cleared when we receive a packet with TH_CWR set. + * TE_SENDECE will be cleared when we receive a packet with TH_CWR set. */ if (ip_ecn == IPTOS_ECN_CE && tp->t_state == TCPS_ESTABLISHED && TCP_ECN_ENABLED(tp) && tlen > 0 && @@ -2663,25 +2733,17 @@ findpcb: tp->ecn_flags |= TE_SENDECE; } - /* - * Clear TE_SENDECE if TH_CWR is set. This is harmless, so we don't - * bother doing extensive checks for state and whatnot. - */ - if (thflags & TH_CWR) { - tp->ecn_flags &= ~TE_SENDECE; - tp->t_ecn_recv_cwr++; - } - /* * If we received an explicit notification of congestion in * ip tos ecn bits or by the CWR bit in TCP header flags, reset - * the ack-strteching state. We need to handle ECN notification if + * the ack-stretching state. We need to handle ECN notification if * an ECN setup SYN was sent even once. */ - if (tp->t_state == TCPS_ESTABLISHED - && (tp->ecn_flags & TE_SETUPSENT) - && (ip_ecn == IPTOS_ECN_CE || (thflags & TH_CWR))) { + if (tp->t_state == TCPS_ESTABLISHED && + (tp->ecn_flags & TE_SETUPSENT) && + (ip_ecn == IPTOS_ECN_CE || (thflags & TH_CWR))) { tcp_reset_stretch_ack(tp); + tp->t_forced_acks = TCP_FORCED_ACKS_COUNT; CLEAR_IAJ_STATE(tp); } @@ -2700,7 +2762,7 @@ findpcb: tcpstat.tcps_ecn_fallback_ce++; tcp_heuristic_ecn_aggressive(tp); tp->ecn_flags |= TE_CEHEURI_SET; - INP_INC_IFNET_STAT(inp,ecn_fallback_ce); + INP_INC_IFNET_STAT(inp, ecn_fallback_ce); } else { /* We tracked the first ECN_MIN_CE_PROBES segments, we * now know that the path is good. @@ -2709,18 +2771,6 @@ findpcb: } } - /* - * Try to determine if we are receiving a packet after a long time. - * Use our own approximation of idletime to roughly measure remote - * end's idle time. Since slowstart is used after an idle period - * we want to avoid doing LRO if the remote end is not up to date - * on initial window support and starts with 1 or 2 packets as its IW. - */ - if (sw_lro && (tp->t_flagsext & TF_LRO_OFFLOADED) && - ((tcp_now - tp->t_rcvtime) >= (TCP_IDLETIMEOUT(tp)))) { - turnoff_lro = 1; - } - /* Update rcvtime as a new segment was received on the connection */ tp->t_rcvtime = tcp_now; @@ -2731,8 +2781,9 @@ findpcb: if (TCPS_HAVEESTABLISHED(tp->t_state)) { tcp_keepalive_reset(tp); - if (tp->t_mpsub) + if (tp->t_mpsub) { mptcp_reset_keepalive(tp); + } } /* @@ -2749,16 +2800,15 @@ findpcb: (void) tcp_output(tp); tcp_check_timer_state(tp); socket_unlock(so, 1); - KERNEL_DEBUG(DBG_FNC_TCP_INPUT | - DBG_FUNC_END,0,0,0,0,0); return; } #endif /* MPTCP */ if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { if (!(thflags & TH_ACK) || (SEQ_GT(th->th_ack, tp->iss) && - SEQ_LEQ(th->th_ack, tp->snd_max))) + SEQ_LEQ(th->th_ack, tp->snd_max))) { tcp_finalize_options(tp, &to, ifscope); + } } #if TRAFFIC_MGT @@ -2774,21 +2824,18 @@ findpcb: * to queuing in edge/access routers. */ if (tp->t_state == TCPS_ESTABLISHED && - (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK|TH_ECE|TH_PUSH)) == TH_ACK && - ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && + (thflags & (TH_SYN | TH_FIN | TH_RST | TH_URG | TH_ACK | TH_ECE | TH_PUSH)) == TH_ACK && + ((tp->t_flags & TF_NEEDFIN) == 0) && ((to.to_flags & TOF_TS) == 0 || - TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && + TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && th->th_seq == tp->rcv_nxt && LIST_EMPTY(&tp->t_segq)) { int seg_size = tlen; if (tp->iaj_pktcnt <= IAJ_IGNORE_PKTCNT) { - TCP_INC_VAR(tp->iaj_pktcnt, nlropkts); + TCP_INC_VAR(tp->iaj_pktcnt, segment_count); } - if (m->m_pkthdr.pkt_flags & PKTF_SW_LRO_PKT) { - seg_size = m->m_pkthdr.lro_pktlen; - } - if ( tp->iaj_size == 0 || seg_size > tp->iaj_size || - (seg_size == tp->iaj_size && tp->iaj_rcv_ts == 0)) { + if (tp->iaj_size == 0 || seg_size > tp->iaj_size || + (seg_size == tp->iaj_size && tp->iaj_rcv_ts == 0)) { /* * State related to inter-arrival jitter is * uninitialized or we are trying to find a good @@ -2801,13 +2848,9 @@ findpcb: * Compute inter-arrival jitter taking * this packet as the second packet */ - if (pktf_sw_lro_pkt) - compute_iaj(tp, nlropkts, - m->m_pkthdr.lro_elapsed); - else - compute_iaj(tp, 1, 0); + compute_iaj(tp); } - if (seg_size < tp->iaj_size) { + if (seg_size < tp->iaj_size) { /* * There is a smaller packet in the stream. * Some times the maximum size supported @@ -2818,7 +2861,7 @@ findpcb: * iaj_size, we try to learn the iaj_size * again. */ - TCP_INC_VAR(tp->iaj_small_pkt, nlropkts); + TCP_INC_VAR(tp->iaj_small_pkt, segment_count); if (tp->iaj_small_pkt > RESET_IAJ_SIZE_THRESH) { update_iaj_state(tp, seg_size, 1); } else { @@ -2851,14 +2894,14 @@ findpcb: * be TH_NEEDSYN. */ if (tp->t_state == TCPS_ESTABLISHED && - (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK|TH_ECE|TH_CWR)) == TH_ACK && - ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && + !(so->so_state & SS_CANTRCVMORE) && + (thflags & TH_FLAGS) == TH_ACK && + ((tp->t_flags & TF_NEEDFIN) == 0) && ((to.to_flags & TOF_TS) == 0 || - TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && + TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && tp->snd_nxt == tp->snd_max) { - /* * If last ACK falls within this segment's sequence numbers, * record the timestamp. @@ -2866,7 +2909,7 @@ findpcb: * proposal of the tcplw@cray.com list (Braden 1993/04/26). */ if ((to.to_flags & TOF_TS) != 0 && - SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { + SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { tp->ts_recent_age = tcp_now; tp->ts_recent = to.to_tsval; } @@ -2901,20 +2944,19 @@ findpcb: * calculations in this function * assume that snd_una is not updated yet. */ - if (CC_ALGO(tp)->congestion_avd != NULL) + if (CC_ALGO(tp)->congestion_avd != NULL) { CC_ALGO(tp)->congestion_avd(tp, th); + } tcp_ccdbg_trace(tp, th, TCP_CC_INSEQ_ACK_RCVD); sbdrop(&so->so_snd, acked); - if (so->so_flags & SOF_ENABLE_MSGS) { - VERIFY(acked <= so->so_msg_state->msg_serial_bytes); - so->so_msg_state->msg_serial_bytes -= acked; - } tcp_sbsnd_trim(&so->so_snd); if (SEQ_GT(tp->snd_una, tp->snd_recover) && - SEQ_LEQ(th->th_ack, tp->snd_recover)) + SEQ_LEQ(th->th_ack, tp->snd_recover)) { tp->snd_recover = th->th_ack - 1; - tp->snd_una = th->th_ack; + } + + tcp_update_snd_una(tp, th->th_ack); TCP_RESET_REXMT_STATE(tp); @@ -2927,9 +2969,10 @@ findpcb: if (tp->t_dupacks > 0) { tp->t_dupacks = 0; tp->t_rexmtthresh = tcprexmtthresh; + tp->t_new_dupacks = 0; } - m_freem(m); + tp->sackhint.sack_bytes_acked = 0; /* * If all outstanding data are acked, stop @@ -2944,22 +2987,23 @@ findpcb: tp->t_timer[TCPT_REXMT] = 0; tp->t_timer[TCPT_PTO] = 0; } else if (tp->t_timer[TCPT_PERSIST] == 0) { - tp->t_timer[TCPT_REXMT] = - OFFSET_FROM_START(tp, - tp->t_rxtcur); + tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp, tp->t_rxtcur); } if (!SLIST_EMPTY(&tp->t_rxt_segments) && !TCP_DSACK_SEQ_IN_WINDOW(tp, - tp->t_dsack_lastuna, tp->snd_una)) + tp->t_dsack_lastuna, tp->snd_una)) { tcp_rxtseg_clean(tp); + } if ((tp->t_flagsext & TF_MEASURESNDBW) != 0 && - tp->t_bwmeas != NULL) + tp->t_bwmeas != NULL) { tcp_bwmeas_check(tp); + } - sowwakeup(so); /* has to be done with socket lock held */ - if (!SLIST_EMPTY(&tp->t_notify_ack)) + write_wakeup = 1; + if (!SLIST_EMPTY(&tp->t_notify_ack)) { tcp_notify_acknowledgement(tp, so); + } if ((so->so_snd.sb_cc) || (tp->t_flags & TF_ACKNOW)) { (void) tcp_output(tp); @@ -2967,13 +3011,17 @@ findpcb: tcp_tfo_rcv_ack(tp, th); + m_freem(m); + tcp_check_timer_state(tp); + + tcp_handle_wakeup(so, read_wakeup, write_wakeup); + socket_unlock(so, 1); - KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END, 0, 0, 0, 0, 0); return; } - } else if (th->th_ack == tp->snd_una && - LIST_EMPTY(&tp->t_segq) && + } else if (th->th_ack == tp->snd_una && LIST_EMPTY(&tp->t_segq) && tlen <= tcp_sbspace(tp)) { /* * this is a pure, in-sequence data packet @@ -2981,33 +3029,10 @@ findpcb: * we have enough buffer space to take it. */ - /* - * If this is a connection in steady state, start - * coalescing packets belonging to this flow. - */ - if (turnoff_lro) { - tcp_lro_remove_state(tp->t_inpcb->inp_laddr, - tp->t_inpcb->inp_faddr, - tp->t_inpcb->inp_lport, - tp->t_inpcb->inp_fport); - tp->t_flagsext &= ~TF_LRO_OFFLOADED; - tp->t_idleat = tp->rcv_nxt; - } else if (sw_lro && !pktf_sw_lro_pkt && !isipv6 && - (so->so_flags & SOF_USELRO) && - !IFNET_IS_CELLULAR(m->m_pkthdr.rcvif) && - (m->m_pkthdr.rcvif->if_type != IFT_LOOP) && - ((th->th_seq - tp->irs) > - (tp->t_maxseg << lro_start)) && - ((tp->t_idleat == 0) || ((th->th_seq - - tp->t_idleat) > (tp->t_maxseg << lro_start)))) { - tp->t_flagsext |= TF_LRO_OFFLOADED; - tcp_start_coalescing(ip, th, tlen); - tp->t_idleat = 0; - } - /* Clean receiver SACK report if present */ - if (SACK_ENABLED(tp) && tp->rcv_numsacks) + if (SACK_ENABLED(tp) && tp->rcv_numsacks) { tcp_clean_sackreport(tp); + } ++tcpstat.tcps_preddat; tp->rcv_nxt += tlen; /* @@ -3020,17 +3045,12 @@ findpcb: * rcv_nxt. */ tp->rcv_up = tp->rcv_nxt; - TCP_INC_VAR(tcpstat.tcps_rcvpack, nlropkts); + TCP_INC_VAR(tcpstat.tcps_rcvpack, segment_count); tcpstat.tcps_rcvbyte += tlen; if (nstat_collect) { - if (m->m_pkthdr.pkt_flags & PKTF_SW_LRO_PKT) { - INP_ADD_STAT(inp, cell, wifi, wired, - rxpackets, m->m_pkthdr.lro_npkts); - } else { - INP_ADD_STAT(inp, cell, wifi, wired, - rxpackets, 1); - } - INP_ADD_STAT(inp, cell, wifi, wired,rxbytes, + INP_ADD_STAT(inp, cell, wifi, wired, + rxpackets, 1); + INP_ADD_STAT(inp, cell, wifi, wired, rxbytes, tlen); inp_set_activity_bitmap(inp); } @@ -3040,44 +3060,52 @@ findpcb: * connection is in streaming mode and the last * packet was not an end-of-write */ - if (tp->t_flags & TF_STREAMING_ON) + if (tp->t_flags & TF_STREAMING_ON) { tcp_compute_rtt(tp, &to, th); + } - tcp_sbrcv_grow(tp, &so->so_rcv, &to, tlen, - TCP_AUTORCVBUF_MAX(ifp)); + tcp_sbrcv_grow(tp, &so->so_rcv, &to, tlen); /* * Add data to socket buffer. */ so_recv_data_stat(so, m, 0); - m_adj(m, drop_hdrlen); /* delayed header drop */ + m_adj(m, drop_hdrlen); /* delayed header drop */ - /* - * If message delivery (SOF_ENABLE_MSGS) is enabled on - * this socket, deliver the packet received as an - * in-order message with sequence number attached to it. - */ - if (sbappendstream_rcvdemux(so, m, - th->th_seq - (tp->irs + 1), 0)) { - sorwakeup(so); + if (isipv6) { + memcpy(&saved_hdr, ip6, sizeof(struct ip6_hdr)); + ip6 = (struct ip6_hdr *)&saved_hdr[0]; + } else { + memcpy(&saved_hdr, ip, ip->ip_hl << 2); + ip = (struct ip *)&saved_hdr[0]; } -#if INET6 + memcpy(&saved_tcphdr, th, sizeof(struct tcphdr)); + + if (th->th_flags & TH_PUSH) { + tp->t_flagsext |= TF_LAST_IS_PSH; + } else { + tp->t_flagsext &= ~TF_LAST_IS_PSH; + } + + if (sbappendstream_rcvdemux(so, m)) { + mptcp_handle_input(so); + read_wakeup = 1; + } + th = &saved_tcphdr; + if (isipv6) { KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport), - (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])), - th->th_seq, th->th_ack, th->th_win); - } - else -#endif - { + (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])), + th->th_seq, th->th_ack, th->th_win); + } else { KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport), - (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)), - th->th_seq, th->th_ack, th->th_win); + (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)), + th->th_seq, th->th_ack, th->th_win); } - TCP_INC_VAR(tp->t_unacksegs, nlropkts); - if (DELAY_ACK(tp, th)) { + TCP_INC_VAR(tp->t_unacksegs, segment_count); + if (DELAY_ACK(tp, th)) { if ((tp->t_flags & TF_DELACK) == 0) { - tp->t_flags |= TF_DELACK; + tp->t_flags |= TF_DELACK; tp->t_timer[TCPT_DELACK] = OFFSET_FROM_START(tp, tcp_delack); } } else { @@ -3087,12 +3115,16 @@ findpcb: tcp_adaptive_rwtimo_check(tp, tlen); - if (tlen > 0) + if (tlen > 0) { tcp_tfo_rcv_data(tp); + } tcp_check_timer_state(tp); + + tcp_handle_wakeup(so, read_wakeup, write_wakeup); + socket_unlock(so, 1); - KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END, 0, 0, 0, 0, 0); return; } } @@ -3105,11 +3137,12 @@ findpcb: */ socket_lock_assert_owned(so); win = tcp_sbspace(tp); - if (win < 0) + if (win < 0) { win = 0; - else { /* clip rcv window to 4K for modems */ - if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0) + } else { /* clip rcv window to 4K for modems */ + if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0) { win = min(win, slowlink_wsize); + } } tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); #if MPTCP @@ -3117,22 +3150,23 @@ findpcb: * Ensure that the subflow receive window isn't greater * than the connection level receive window. */ - if ((tp->t_mpflags & TMPF_MPTCP_TRUE) && - (mp_tp = tptomptp(tp))) { - mpte_lock_assert_held(mp_tp->mpt_mpte); - if (tp->rcv_wnd > mp_tp->mpt_rcvwnd) { - tp->rcv_wnd = imax(mp_tp->mpt_rcvwnd, (int)(tp->rcv_adv - tp->rcv_nxt)); + if ((tp->t_mpflags & TMPF_MPTCP_TRUE) && (mp_tp = tptomptp(tp))) { + socket_lock_assert_owned(mptetoso(mp_tp->mpt_mpte)); + int64_t recwin_conn = (int64_t)(mp_tp->mpt_rcvadv - mp_tp->mpt_rcvnxt); + + VERIFY(recwin_conn < INT32_MAX && recwin_conn > INT32_MIN); + if (recwin_conn > 0 && tp->rcv_wnd > (uint32_t)recwin_conn) { + tp->rcv_wnd = (uint32_t)recwin_conn; tcpstat.tcps_mp_reducedwin++; } } #endif /* MPTCP */ switch (tp->t_state) { - /* * Initialize tp->rcv_nxt, and tp->irs, select an initial * tp->iss, and send a segment: - * + * * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. * Fill in remote peer address fields if not previously specified. * Enter SYN_RECEIVED state, and process any other fields of this @@ -3140,52 +3174,60 @@ findpcb: */ case TCPS_LISTEN: { struct sockaddr_in *sin; -#if INET6 struct sockaddr_in6 *sin6; -#endif socket_lock_assert_owned(so); -#if INET6 + + /* Clear the logging flags inherited from the listening socket */ + tp->t_log_flags = 0; + tp->t_flagsext &= ~TF_LOGGED_CONN_SUMMARY; + if (isipv6) { MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6, - M_SONAME, M_NOWAIT); - if (sin6 == NULL) + M_SONAME, M_NOWAIT); + if (sin6 == NULL) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "LISTEN malloc M_SONAME failed"); goto drop; + } bzero(sin6, sizeof(*sin6)); sin6->sin6_family = AF_INET6; sin6->sin6_len = sizeof(*sin6); sin6->sin6_addr = ip6->ip6_src; sin6->sin6_port = th->th_sport; laddr6 = inp->in6p_laddr; - if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) + if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { inp->in6p_laddr = ip6->ip6_dst; + } if (in6_pcbconnect(inp, (struct sockaddr *)sin6, - proc0)) { + kernel_proc)) { inp->in6p_laddr = laddr6; FREE(sin6, M_SONAME); + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, " LISTEN in6_pcbconnect failed"); goto drop; } FREE(sin6, M_SONAME); - } else -#endif - { + } else { socket_lock_assert_owned(so); MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, M_NOWAIT); - if (sin == NULL) + if (sin == NULL) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "LISTEN malloc M_SONAME failed"); goto drop; + } sin->sin_family = AF_INET; sin->sin_len = sizeof(*sin); sin->sin_addr = ip->ip_src; sin->sin_port = th->th_sport; bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); laddr = inp->inp_laddr; - if (inp->inp_laddr.s_addr == INADDR_ANY) + if (inp->inp_laddr.s_addr == INADDR_ANY) { inp->inp_laddr = ip->ip_dst; - if (in_pcbconnect(inp, (struct sockaddr *)sin, proc0, + } + if (in_pcbconnect(inp, (struct sockaddr *)sin, kernel_proc, IFSCOPE_NONE, NULL)) { inp->inp_laddr = laddr; FREE(sin, M_SONAME); + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, " LISTEN in_pcbconnect failed"); goto drop; } FREE(sin, M_SONAME); @@ -3194,14 +3236,15 @@ findpcb: tcp_dooptions(tp, optp, optlen, th, &to); tcp_finalize_options(tp, &to, ifscope); - if (tfo_enabled(tp) && tcp_tfo_syn(tp, &to)) + if (tfo_enabled(tp) && tcp_tfo_syn(tp, &to)) { isconnected = TRUE; + } - if (iss) + if (iss) { tp->iss = iss; - else { + } else { tp->iss = tcp_new_isn(tp); - } + } tp->irs = th->th_seq; tcp_sendseqinit(tp); tcp_rcvseqinit(tp); @@ -3211,27 +3254,28 @@ findpcb: * set SND.WND = SEG.WND, * initialize CCsend and CCrecv. */ - tp->snd_wnd = tiwin; /* initial send-window */ + tp->snd_wnd = tiwin; /* initial send-window */ + tp->max_sndwnd = tp->snd_wnd; tp->t_flags |= TF_ACKNOW; tp->t_unacksegs = 0; DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, - struct tcpcb *, tp, int32_t, TCPS_SYN_RECEIVED); + struct tcpcb *, tp, int32_t, TCPS_SYN_RECEIVED); tp->t_state = TCPS_SYN_RECEIVED; tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, - TCP_CONN_KEEPINIT(tp)); - dropsocket = 0; /* committed to socket */ + TCP_CONN_KEEPINIT(tp)); + tp->t_connect_time = tcp_now; + dropsocket = 0; /* committed to socket */ - if (inp->inp_flowhash == 0) + if (inp->inp_flowhash == 0) { inp->inp_flowhash = inp_calc_flowhash(inp); -#if INET6 + } /* update flowinfo - RFC 6437 */ if (inp->inp_flow == 0 && inp->in6p_flags & IN6P_AUTOFLOWLABEL) { inp->inp_flow &= ~IPV6_FLOWLABEL_MASK; inp->inp_flow |= - (htonl(inp->inp_flowhash) & IPV6_FLOWLABEL_MASK); + (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); } -#endif /* INET6 */ /* reset the incomp processing flag */ so->so_flags &= ~(SOF_INCOMP_INPROGRESS); @@ -3241,8 +3285,15 @@ findpcb: tp->ecn_flags |= (TE_SETUPRECEIVED | TE_SENDIPECT); } + /* + * The address and connection state are finalized + */ + TCP_LOG_CONNECT(tp, false, 0); + + tcp_add_fsw_flow(tp, ifp); + goto trimthenstep6; - } + } /* * If the state is SYN_RECEIVED and the seg contains an ACK, @@ -3251,10 +3302,10 @@ findpcb: case TCPS_SYN_RECEIVED: if ((thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->snd_una) || - SEQ_GT(th->th_ack, tp->snd_max))) { - rstreason = BANDLIM_RST_OPENPORT; - IF_TCP_STATINC(ifp, ooopacket); - goto dropwithreset; + SEQ_GT(th->th_ack, tp->snd_max))) { + IF_TCP_STATINC(ifp, ooopacket); + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SYN_RECEIVED bad ACK"); + goto dropwithreset; } /* @@ -3265,8 +3316,9 @@ findpcb: */ if ((thflags & TH_SYN) && (tp->irs == th->th_seq) && - !(to.to_flags & TOF_SCALE)) + !(to.to_flags & TOF_SCALE)) { tp->t_flags &= ~TF_RCVD_SCALE; + } break; /* @@ -3284,15 +3336,17 @@ findpcb: case TCPS_SYN_SENT: if ((thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->iss) || - SEQ_GT(th->th_ack, tp->snd_max))) { - rstreason = BANDLIM_UNLIMITED; + SEQ_GT(th->th_ack, tp->snd_max))) { IF_TCP_STATINC(ifp, ooopacket); + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SYN_SENT bad ACK"); goto dropwithreset; } if (thflags & TH_RST) { if ((thflags & TH_ACK) != 0) { - if (tfo_enabled(tp)) + if (tfo_enabled(tp) && + !(tp->t_flagsext & TF_FASTOPEN_FORCE_ENABLE)) { tcp_heuristic_tfo_rst(tp); + } if ((tp->ecn_flags & (TE_SETUPSENT | TE_RCVD_SYN_RST)) == TE_SETUPSENT) { /* * On local connections, send @@ -3310,13 +3364,16 @@ findpcb: (SO_FILT_HINT_LOCKED | SO_FILT_HINT_CONNRESET)); tp = tcp_drop(tp, ECONNREFUSED); - postevent(so, 0, EV_RESET); } + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SYN_SENT got RST"); goto drop; } - if ((thflags & TH_SYN) == 0) + if ((thflags & TH_SYN) == 0) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SYN_SENT no SYN"); goto drop; - tp->snd_wnd = th->th_win; /* initial send window */ + } + tp->snd_wnd = th->th_win; /* initial send window */ + tp->max_sndwnd = tp->snd_wnd; tp->irs = th->th_seq; tcp_rcvseqinit(tp); @@ -3337,18 +3394,14 @@ findpcb: tcpstat.tcps_ecn_not_supported++; } if (tp->ecn_flags & TE_SETUPSENT && - tp->t_rxtshift > 0) + tp->t_rxtshift > 0) { tcp_heuristic_ecn_loss(tp); + } /* non-ECN-setup SYN-ACK */ tp->ecn_flags &= ~TE_SENDIPECT; } -#if CONFIG_MACF_NET && CONFIG_MACF_SOCKET - /* XXXMAC: recursive lock: SOCK_LOCK(so); */ - mac_socketpeer_label_associate_mbuf(m, so); - /* XXXMAC: SOCK_UNLOCK(so); */ -#endif /* Do window scaling on this connection? */ if (TCP_WINDOW_SCALE_ENABLED(tp)) { tp->snd_scale = tp->requested_s_scale; @@ -3356,9 +3409,10 @@ findpcb: } tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN << tp->rcv_scale); - tp->snd_una++; /* SYN is acked */ - if (SEQ_LT(tp->snd_nxt, tp->snd_una)) + tp->snd_una++; /* SYN is acked */ + if (SEQ_LT(tp->snd_nxt, tp->snd_una)) { tp->snd_nxt = tp->snd_una; + } /* * We have sent more in the SYN than what is being @@ -3372,12 +3426,15 @@ findpcb: * There is a middlebox that acks all but one * byte and still drops the data. */ - if ((tp->t_tfo_stats & TFO_S_SYN_DATA_SENT) && + if (!(tp->t_flagsext & TF_FASTOPEN_FORCE_ENABLE) && + (tp->t_tfo_stats & TFO_S_SYN_DATA_SENT) && tp->snd_max == th->th_ack + 1 && tp->snd_max > tp->snd_una + 1) { tcp_heuristic_tfo_middlebox(tp); so->so_error = ENODATA; + soevent(so, + (SO_FILT_HINT_LOCKED | SO_FILT_HINT_MP_SUB_ERROR)); tp->t_tfo_stats |= TFO_S_ONE_BYTE_PROXY; } @@ -3389,14 +3446,13 @@ findpcb: * If there's data, delay ACK; if there's also a FIN * ACKNOW will be turned on later. */ - TCP_INC_VAR(tp->t_unacksegs, nlropkts); - if (DELAY_ACK(tp, th) && tlen != 0 ) { + TCP_INC_VAR(tp->t_unacksegs, segment_count); + if (DELAY_ACK(tp, th) && tlen != 0) { if ((tp->t_flags & TF_DELACK) == 0) { tp->t_flags |= TF_DELACK; tp->t_timer[TCPT_DELACK] = OFFSET_FROM_START(tp, tcp_delack); } - } - else { + } else { tp->t_flags |= TF_ACKNOW; } /* @@ -3415,6 +3471,8 @@ findpcb: tp->t_state = TCPS_FIN_WAIT_1; tp->t_flags &= ~TF_NEEDFIN; thflags &= ~TH_SYN; + + TCP_LOG_CONNECTION_SUMMARY(tp); } else { DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, struct tcpcb *, @@ -3423,9 +3481,10 @@ findpcb: tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, TCP_CONN_KEEPIDLE(tp)); - if (nstat_collect) + if (nstat_collect) { nstat_route_connect_success( - inp->inp_route.ro_rt); + inp->inp_route.ro_rt); + } /* * The SYN is acknowledged but una is not * updated yet. So pass the value of @@ -3433,6 +3492,7 @@ findpcb: */ inp_count_sndbytes(inp, th->th_ack); } + tp->t_forced_acks = TCP_FORCED_ACKS_COUNT; #if MPTCP /* * Do not send the connect notification for additional @@ -3443,7 +3503,7 @@ findpcb: isconnected = FALSE; } else #endif /* MPTCP */ - isconnected = TRUE; + isconnected = TRUE; if ((tp->t_tfo_flags & (TFO_F_COOKIE_REQ | TFO_F_COOKIE_SENT)) || (tp->t_tfo_stats & TFO_S_SYN_DATA_SENT)) { @@ -3454,8 +3514,9 @@ findpcb: tp->t_tfo_stats |= TFO_S_SYN_DATA_ACKED; tcpstat.tcps_tfo_syn_data_acked++; #if MPTCP - if (so->so_flags & SOF_MP_SUBFLOW) + if (so->so_flags & SOF_MP_SUBFLOW) { so->so_flags1 |= SOF1_TFO_REWIND; + } #endif tcp_tfo_rcv_probe(tp, tlen); } @@ -3463,16 +3524,15 @@ findpcb: } else { /* * Received initial SYN in SYN-SENT[*] state => simul- - * taneous open. If segment contains CC option and there is - * a cached CC, apply TAO test; if it succeeds, connection is - * half-synchronized. Otherwise, do 3-way handshake: + * taneous open. + * Do 3-way handshake: * SYN-SENT -> SYN-RECEIVED * SYN-SENT* -> SYN-RECEIVED* */ tp->t_flags |= TF_ACKNOW; tp->t_timer[TCPT_REXMT] = 0; DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, - struct tcpcb *, tp, int32_t, TCPS_SYN_RECEIVED); + struct tcpcb *, tp, int32_t, TCPS_SYN_RECEIVED); tp->t_state = TCPS_SYN_RECEIVED; /* @@ -3507,8 +3567,9 @@ trimthenstep6: * processing in the middle of step 5, ack processing. * Otherwise, goto step 6. */ - if (thflags & TH_ACK) + if (thflags & TH_ACK) { goto process_ACK; + } goto step6; /* * If the state is LAST_ACK or CLOSING or TIME_WAIT: @@ -3519,16 +3580,27 @@ trimthenstep6: case TCPS_LAST_ACK: case TCPS_CLOSING: case TCPS_TIME_WAIT: - break; /* continue normal processing */ + break; /* continue normal processing */ /* Received a SYN while connection is already established. * This is a "half open connection and other anomalies" described * in RFC793 page 34, send an ACK so the remote reset the connection - * or recovers by adjusting its sequence numberering + * or recovers by adjusting its sequence numbering. Sending an ACK is + * in accordance with RFC 5961 Section 4.2 */ case TCPS_ESTABLISHED: - if (thflags & TH_SYN) - goto dropafterack; + if (thflags & TH_SYN && tlen <= 0) { + /* Drop the packet silently if we have reached the limit */ + if (tcp_is_ack_ratelimited(tp)) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "ESTABLISHED rfc5961 rate limited"); + goto drop; + } else { + /* Send challenge ACK */ + tcpstat.tcps_synchallenge++; + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "ESTABLISHED rfc5961 challenge ACK"); + goto dropafterack; + } + } break; } @@ -3566,6 +3638,11 @@ trimthenstep6: * only accepting RSTs where the sequence number is equal to * last_ack_sent. In all other states (the states in which a * RST is more likely), the more permissive check is used. + * RFC 5961 Section 3.2: if the RST bit is set, sequence # is + * within the receive window and last_ack_sent == seq, + * then reset the connection. Otherwise if the seq doesn't + * match last_ack_sent, TCP must send challenge ACK. Perform + * rate limitation when sending the challenge ACK. * If we have multiple segments in flight, the intial reset * segment sequence numbers will be to the left of last_ack_sent, * but they will eventually catch up. @@ -3598,60 +3675,66 @@ trimthenstep6: * there is already a RST in flight from the peer. * In that case, accept the RST for non-established * state if it's one off from last_ack_sent. - + * */ if (thflags & TH_RST) { if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || (tp->rcv_wnd == 0 && ((tp->last_ack_sent == th->th_seq) || - ((tp->last_ack_sent -1) == th->th_seq)))) { - switch (tp->t_state) { + ((tp->last_ack_sent - 1) == th->th_seq)))) { + if (tp->last_ack_sent == th->th_seq) { + switch (tp->t_state) { + case TCPS_SYN_RECEIVED: + IF_TCP_STATINC(ifp, rstinsynrcv); + so->so_error = ECONNREFUSED; + goto close; + + case TCPS_ESTABLISHED: + if (TCP_ECN_ENABLED(tp) && + tp->snd_una == tp->iss + 1 && + SEQ_GT(tp->snd_max, tp->snd_una)) { + /* + * If the first data packet on an + * ECN connection, receives a RST + * increment the heuristic + */ + tcp_heuristic_ecn_droprst(tp); + } + OS_FALLTHROUGH; + case TCPS_FIN_WAIT_1: + case TCPS_CLOSE_WAIT: + case TCPS_FIN_WAIT_2: + so->so_error = ECONNRESET; +close: + soevent(so, + (SO_FILT_HINT_LOCKED | + SO_FILT_HINT_CONNRESET)); + + tcpstat.tcps_drops++; + tp = tcp_close(tp); + break; - case TCPS_SYN_RECEIVED: - IF_TCP_STATINC(ifp, rstinsynrcv); - so->so_error = ECONNREFUSED; - goto close; + case TCPS_CLOSING: + case TCPS_LAST_ACK: + tp = tcp_close(tp); + break; - case TCPS_ESTABLISHED: - if (tp->last_ack_sent != th->th_seq) { - tcpstat.tcps_badrst++; - goto drop; + case TCPS_TIME_WAIT: + break; } - if (TCP_ECN_ENABLED(tp) && - tp->snd_una == tp->iss + 1 && - SEQ_GT(tp->snd_max, tp->snd_una)) { - /* - * If the first data packet on an - * ECN connection, receives a RST - * increment the heuristic - */ - tcp_heuristic_ecn_droprst(tp); + } else { + tcpstat.tcps_badrst++; + /* Drop if we have reached the ACK limit */ + if (tcp_is_ack_ratelimited(tp)) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "ESTABLISHED rfc5961 rate limited"); + goto drop; + } else { + /* Send challenge ACK */ + tcpstat.tcps_rstchallenge++; + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "ESTABLISHED rfc5961 challenge ACK"); + goto dropafterack; } - case TCPS_FIN_WAIT_1: - case TCPS_CLOSE_WAIT: - /* - Drop through ... - */ - case TCPS_FIN_WAIT_2: - so->so_error = ECONNRESET; - close: - postevent(so, 0, EV_RESET); - soevent(so, - (SO_FILT_HINT_LOCKED | - SO_FILT_HINT_CONNRESET)); - - tcpstat.tcps_drops++; - tp = tcp_close(tp); - break; - - case TCPS_CLOSING: - case TCPS_LAST_ACK: - tp = tcp_close(tp); - break; - - case TCPS_TIME_WAIT: - break; } } goto drop; @@ -3663,7 +3746,6 @@ trimthenstep6: */ if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { - /* Check to see if ts_recent is over 24 days old. */ if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { /* @@ -3701,7 +3783,7 @@ trimthenstep6: if (nstat_collect) { nstat_route_rx(tp->t_inpcb->inp_route.ro_rt, - 1, tlen, NSTAT_RX_FLAG_DUPLICATE); + 1, tlen, NSTAT_RX_FLAG_DUPLICATE); INP_ADD_STAT(inp, cell, wifi, wired, rxpackets, 1); INP_ADD_STAT(inp, cell, wifi, wired, @@ -3709,8 +3791,9 @@ trimthenstep6: tp->t_stat.rxduplicatebytes += tlen; inp_set_activity_bitmap(inp); } - if (tlen > 0) + if (tlen > 0) { goto dropafterack; + } goto drop; } } @@ -3723,24 +3806,34 @@ trimthenstep6: * for the "LAND" DoS attack. */ if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { - rstreason = BANDLIM_RST_OPENPORT; IF_TCP_STATINC(ifp, dospacket); + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SYN_RECEIVED bad SEQ"); goto dropwithreset; } + /* + * Check if there is old data at the beginning of the window + * i.e. the sequence number is before rcv_nxt + */ todrop = tp->rcv_nxt - th->th_seq; if (todrop > 0) { + boolean_t is_syn_set = FALSE; + if (thflags & TH_SYN) { + is_syn_set = TRUE; thflags &= ~TH_SYN; th->th_seq++; - if (th->th_urp > 1) + if (th->th_urp > 1) { th->th_urp--; - else + } else { thflags &= ~TH_URG; + } todrop--; } /* * Following if statement from Stevens, vol. 2, p. 960. + * The amount of duplicate data is greater than or equal + * to the size of the segment - entire segment is duplicate */ if (todrop > tlen || (todrop == tlen && (thflags & TH_FIN) == 0)) { @@ -3754,12 +3847,23 @@ trimthenstep6: /* * Send an ACK to resynchronize and drop any data. * But keep on processing for RST or ACK. + * + * If the SYN bit was originally set, then only send + * an ACK if we are not rate-limiting this connection. */ - tp->t_flags |= TF_ACKNOW; + if (is_syn_set) { + if (!tcp_is_ack_ratelimited(tp)) { + tcpstat.tcps_synchallenge++; + tp->t_flags |= TF_ACKNOW; + } + } else { + tp->t_flags |= TF_ACKNOW; + } + if (todrop == 1) { /* This could be a keepalive */ soevent(so, SO_FILT_HINT_LOCKED | - SO_FILT_HINT_KEEPALIVE); + SO_FILT_HINT_KEEPALIVE); } todrop = tlen; tcpstat.tcps_rcvduppack++; @@ -3769,7 +3873,7 @@ trimthenstep6: tcpstat.tcps_rcvpartdupbyte += todrop; } - if (TCP_DSACK_ENABLED(tp) && todrop > 1) { + if (todrop > 1) { /* * Note the duplicate data sequence space so that * it can be reported in DSACK option. @@ -3780,18 +3884,18 @@ trimthenstep6: } if (nstat_collect) { nstat_route_rx(tp->t_inpcb->inp_route.ro_rt, 1, - todrop, NSTAT_RX_FLAG_DUPLICATE); + todrop, NSTAT_RX_FLAG_DUPLICATE); INP_ADD_STAT(inp, cell, wifi, wired, rxpackets, 1); INP_ADD_STAT(inp, cell, wifi, wired, rxbytes, todrop); tp->t_stat.rxduplicatebytes += todrop; inp_set_activity_bitmap(inp); } - drop_hdrlen += todrop; /* drop from the top afterwards */ + drop_hdrlen += todrop; /* drop from the top afterwards */ th->th_seq += todrop; tlen -= todrop; - if (th->th_urp > todrop) + if (th->th_urp > todrop) { th->th_urp -= todrop; - else { + } else { thflags &= ~TH_URG; th->th_urp = 0; } @@ -3810,20 +3914,30 @@ trimthenstep6: boolean_t close_it = FALSE; if (!(so->so_flags & SOF_MP_SUBFLOW) && (so->so_state & SS_NOFDREF) && - tp->t_state > TCPS_CLOSE_WAIT) + tp->t_state > TCPS_CLOSE_WAIT) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SS_NOFDREF"); close_it = TRUE; + } if ((so->so_flags & SOF_MP_SUBFLOW) && (mptetoso(tptomptp(tp)->mpt_mpte)->so_state & SS_NOFDREF) && - tp->t_state > TCPS_CLOSE_WAIT) + tp->t_state > TCPS_CLOSE_WAIT) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SOF_MP_SUBFLOW SS_NOFDREF"); + close_it = TRUE; + } + + if ((so->so_flags & SOF_DEFUNCT) && tp->t_state > TCPS_FIN_WAIT_1) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SOF_DEFUNCT"); close_it = TRUE; + } - if ((so->so_flags & SOF_DEFUNCT) && tp->t_state > TCPS_FIN_WAIT_1) + if (so->so_state & SS_CANTRCVMORE) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "SS_CANTRCVMORE"); close_it = TRUE; + } if (close_it) { tp = tcp_close(tp); tcpstat.tcps_rcvafterclose++; - rstreason = BANDLIM_UNLIMITED; IF_TCP_STATINC(ifp, cleanup); goto dropwithreset; } @@ -3833,7 +3947,7 @@ trimthenstep6: * If segment ends after window, drop trailing data * (and PUSH and FIN); if nothing left, just ACK. */ - todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); + todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); if (todrop > 0) { tcpstat.tcps_rcvpackafterwin++; if (todrop >= tlen) { @@ -3862,13 +3976,15 @@ trimthenstep6: if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { tp->t_flags |= TF_ACKNOW; tcpstat.tcps_rcvwinprobe++; - } else + } else { goto dropafterack; - } else + } + } else { tcpstat.tcps_rcvbyteafterwin += todrop; + } m_adj(m, -todrop); tlen -= todrop; - thflags &= ~(TH_PUSH|TH_FIN); + thflags &= ~(TH_PUSH | TH_FIN); } /* @@ -3892,21 +4008,40 @@ trimthenstep6: if ((to.to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + - ((thflags & (TH_SYN|TH_FIN)) != 0))) { + ((thflags & (TH_SYN | TH_FIN)) != 0))) { tp->ts_recent_age = tcp_now; tp->ts_recent = to.to_tsval; } /* - * If a SYN is in the window, then this is an + * Stevens: If a SYN is in the window, then this is an * error and we send an RST and drop the connection. + * + * RFC 5961 Section 4.2 + * Send challenge ACK for any SYN in synchronized state + * Perform rate limitation in doing so. */ if (thflags & TH_SYN) { - tp = tcp_drop(tp, ECONNRESET); - rstreason = BANDLIM_UNLIMITED; - postevent(so, 0, EV_RESET); - IF_TCP_STATINC(ifp, synwindow); - goto dropwithreset; + if (!tcp_syn_data_valid(tp, th, tlen)) { + tcpstat.tcps_badsyn++; + /* Drop if we have reached ACK limit */ + if (tcp_is_ack_ratelimited(tp)) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "rfc5961 bad SYN rate limited"); + goto drop; + } else { + /* Send challenge ACK */ + tcpstat.tcps_synchallenge++; + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "rfc5961 bad SYN challenge ack"); + goto dropafterack; + } + } else { + /* + * Received SYN (/ACK) with data. + * Move sequence number along to process the data. + */ + th->th_seq++; + thflags &= ~TH_SYN; + } } /* @@ -3915,13 +4050,11 @@ trimthenstep6: * later processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { - if (tp->t_state == TCPS_SYN_RECEIVED || - (tp->t_flags & TF_NEEDSYN)) { + if (tp->t_state == TCPS_SYN_RECEIVED) { if ((tfo_enabled(tp))) { /* * So, we received a valid segment while in - * SYN-RECEIVED (TF_NEEDSYN is actually never - * set, so this is dead code). + * SYN-RECEIVED. * As this cannot be an RST (see that if a bit * higher), and it does not have the ACK-flag * set, we want to retransmit the SYN/ACK. @@ -3938,15 +4071,19 @@ trimthenstep6: * We need to make absolutely sure that we are * going to reply upon a duplicate SYN-segment. */ - if (th->th_flags & TH_SYN) + if (th->th_flags & TH_SYN) { needoutput = 1; + } } goto step6; - } else if (tp->t_flags & TF_ACKNOW) + } else if (tp->t_flags & TF_ACKNOW) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "bad ACK"); goto dropafterack; - else + } else { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "bad ACK"); goto drop; + } } /* @@ -3954,7 +4091,6 @@ trimthenstep6: */ switch (tp->t_state) { - /* * In SYN_RECEIVED state, the ack ACKs our SYN, so enter * ESTABLISHED state and continue processing. @@ -3969,6 +4105,7 @@ trimthenstep6: tp->snd_scale = tp->requested_s_scale; tp->rcv_scale = tp->request_r_scale; tp->snd_wnd = th->th_win << tp->snd_scale; + tp->max_sndwnd = tp->snd_wnd; tiwin = tp->snd_wnd; } /* @@ -3984,16 +4121,20 @@ trimthenstep6: struct tcpcb *, tp, int32_t, TCPS_FIN_WAIT_1); tp->t_state = TCPS_FIN_WAIT_1; tp->t_flags &= ~TF_NEEDFIN; + + TCP_LOG_CONNECTION_SUMMARY(tp); } else { DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, struct tcpcb *, tp, int32_t, TCPS_ESTABLISHED); tp->t_state = TCPS_ESTABLISHED; tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, - TCP_CONN_KEEPIDLE(tp)); - if (nstat_collect) + TCP_CONN_KEEPIDLE(tp)); + if (nstat_collect) { nstat_route_connect_success( - tp->t_inpcb->inp_route.ro_rt); + tp->t_inpcb->inp_route.ro_rt); + } + /* * The SYN is acknowledged but una is not updated * yet. So pass the value of ack to compute @@ -4001,13 +4142,24 @@ trimthenstep6: */ inp_count_sndbytes(inp, th->th_ack); } + tp->t_forced_acks = TCP_FORCED_ACKS_COUNT; /* * If segment contains data or ACK, will call tcp_reass() * later; if not, do so now to pass queued data to user. */ - if (tlen == 0 && (thflags & TH_FIN) == 0) + if (tlen == 0 && (thflags & TH_FIN) == 0) { + if (isipv6) { + memcpy(&saved_hdr, ip6, sizeof(struct ip6_hdr)); + ip6 = (struct ip6_hdr *)&saved_hdr[0]; + } else { + memcpy(&saved_hdr, ip, ip->ip_hl << 2); + ip = (struct ip *)&saved_hdr[0]; + } + memcpy(&saved_tcphdr, th, sizeof(struct tcphdr)); (void) tcp_reass(tp, (struct tcphdr *)0, &tlen, - NULL, ifp); + NULL, ifp, &read_wakeup); + th = &saved_tcphdr; + } tp->snd_wl1 = th->th_seq - 1; #if MPTCP @@ -4020,7 +4172,7 @@ trimthenstep6: isconnected = FALSE; } else #endif /* MPTCP */ - isconnected = TRUE; + isconnected = TRUE; if ((tp->t_tfo_flags & TFO_F_COOKIE_VALID)) { /* Done this when receiving the SYN */ isconnected = FALSE; @@ -4046,9 +4198,10 @@ trimthenstep6: * handled by the label "dodata" right after step6. */ if (so->so_snd.sb_cc) { - tp->snd_una++; /* SYN is acked */ - if (SEQ_LT(tp->snd_nxt, tp->snd_una)) + tp->snd_una++; /* SYN is acked */ + if (SEQ_LT(tp->snd_nxt, tp->snd_una)) { tp->snd_nxt = tp->snd_una; + } /* * No duplicate-ACK handling is needed. So, we @@ -4061,15 +4214,14 @@ trimthenstep6: * includes a SACK with its ACK. */ if (SACK_ENABLED(tp) && - (to.to_nsacks > 0 || - !TAILQ_EMPTY(&tp->snd_holes))) - tcp_sack_doack(tp, &to, th, - &sack_bytes_acked); + (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes))) { + tcp_sack_doack(tp, &to, th, &sack_bytes_acked, &sack_bytes_newly_acked); + } goto process_ACK; } - /* FALLTHROUGH */ + OS_FALLTHROUGH; /* * In ESTABLISHED state: drop duplicate ACKs; ACK out of range @@ -4088,7 +4240,20 @@ trimthenstep6: case TCPS_TIME_WAIT: if (SEQ_GT(th->th_ack, tp->snd_max)) { tcpstat.tcps_rcvacktoomuch++; - goto dropafterack; + if (tcp_is_ack_ratelimited(tp)) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "rfc5961 rcvacktoomuch"); + goto drop; + } else { + goto dropafterack; + } + } + if (SEQ_LT(th->th_ack, tp->snd_una - tp->max_sndwnd)) { + if (tcp_is_ack_ratelimited(tp)) { + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "rfc5961 bad ACK"); + goto drop; + } else { + goto dropafterack; + } } if (SACK_ENABLED(tp) && to.to_nsacks > 0) { recvd_dsack = tcp_sack_process_dsack(tp, &to, th); @@ -4106,8 +4271,9 @@ trimthenstep6: } if (SACK_ENABLED(tp) && - (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes))) - tcp_sack_doack(tp, &to, th, &sack_bytes_acked); + (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes))) { + tcp_sack_doack(tp, &to, th, &sack_bytes_acked, &sack_bytes_newly_acked); + } #if MPTCP if (tp->t_mpuna && SEQ_GEQ(th->th_ack, tp->t_mpuna)) { @@ -4120,11 +4286,6 @@ trimthenstep6: ~TMPF_PREESTABLISHED; tp->t_mpflags |= TMPF_MPTCP_TRUE; - so->so_flags |= SOF_MPTCP_TRUE; - mptcplog((LOG_DEBUG, "MPTCP " - "Sockets: %s \n",__func__), - MPTCP_SOCKET_DBG, - MPTCP_LOGLVL_LOG); tp->t_timer[TCPT_JACK_RXMT] = 0; tp->t_mprxtshift = 0; @@ -4153,6 +4314,7 @@ trimthenstep6: if (SEQ_LEQ(th->th_ack, tp->snd_una)) { if (tlen == 0 && (tiwin == tp->snd_wnd || (to.to_nsacks > 0 && sack_bytes_acked > 0))) { + uint32_t old_dupacks; /* * If both ends send FIN at the same time, * then the ack will be a duplicate ack @@ -4161,9 +4323,11 @@ trimthenstep6: * instead of the dupack */ if ((thflags & TH_FIN) && - !TCPS_HAVERCVDFIN(tp->t_state)) + !TCPS_HAVERCVDFIN(tp->t_state)) { break; + } process_dupack: + old_dupacks = tp->t_dupacks; #if MPTCP /* * MPTCP options that are ignored must @@ -4188,11 +4352,55 @@ process_dupack: * so that we can process partial acks * correctly */ - if (tp->ecn_flags & TE_INRECOVERY) + if (tp->ecn_flags & TE_INRECOVERY) { tp->ecn_flags &= ~TE_INRECOVERY; + } tcpstat.tcps_rcvdupack++; - ++tp->t_dupacks; + if (SACK_ENABLED(tp) && tcp_do_better_lr) { + tp->t_dupacks += max(1, sack_bytes_acked / tp->t_maxseg); + } else { + ++tp->t_dupacks; + } + + tp->sackhint.sack_bytes_acked += sack_bytes_acked; + + if (SACK_ENABLED(tp) && tcp_do_better_lr) { + tp->t_new_dupacks += (sack_bytes_newly_acked / tp->t_maxseg); + + if (tp->t_new_dupacks >= tp->t_rexmtthresh && IN_FASTRECOVERY(tp)) { + /* Let's restart the retransmission */ + tcp_sack_lost_rexmit(tp); + + /* + * If the current tcp cc module has + * defined a hook for tasks to run + * before entering FR, call it + */ + if (CC_ALGO(tp)->pre_fr != NULL) { + CC_ALGO(tp)->pre_fr(tp); + } + + ENTER_FASTRECOVERY(tp); + + if (tp->t_flags & TF_SENTFIN) { + tp->snd_recover = tp->snd_max - 1; + } else { + tp->snd_recover = tp->snd_max; + } + tp->t_rtttime = 0; + + if (TCP_ECN_ENABLED(tp)) { + tp->ecn_flags |= TE_SENDCWR; + } + + if (tp->t_flagsext & TF_CWND_NONVALIDATED) { + tcp_cc_adjust_nonvalidated_cwnd(tp); + } else { + tp->snd_cwnd = tp->snd_ssthresh; + } + } + } /* * Check if we need to reset the limit on @@ -4201,16 +4409,18 @@ process_dupack: if (tp->t_early_rexmt_count > 0 && TSTMP_GEQ(tcp_now, (tp->t_early_rexmt_win + - TCP_EARLY_REXMT_WIN))) + TCP_EARLY_REXMT_WIN))) { tp->t_early_rexmt_count = 0; + } /* * Is early retransmit needed? We check for * this when the connection is waiting for * duplicate acks to enter fast recovery. */ - if (!IN_FASTRECOVERY(tp)) + if (!IN_FASTRECOVERY(tp)) { tcp_early_rexmt_check(tp, th); + } /* * If we've seen exactly rexmt threshold @@ -4233,13 +4443,12 @@ process_dupack: * network. */ if (tp->t_timer[TCPT_REXMT] == 0 || - (th->th_ack != tp->snd_una - && sack_bytes_acked == 0)) { + (th->th_ack != tp->snd_una && sack_bytes_acked == 0)) { tp->t_dupacks = 0; tp->t_rexmtthresh = tcprexmtthresh; - } else if (tp->t_dupacks > tp->t_rexmtthresh || - IN_FASTRECOVERY(tp)) { - + tp->t_new_dupacks = 0; + } else if ((tp->t_dupacks > tp->t_rexmtthresh && (!tcp_do_better_lr || old_dupacks >= tp->t_rexmtthresh)) || + IN_FASTRECOVERY(tp)) { /* * If this connection was seeing packet * reordering, then recovery might be @@ -4248,8 +4457,8 @@ process_dupack: */ if (SACK_ENABLED(tp) && !IN_FASTRECOVERY(tp) && (tp->t_flagsext & - (TF_PKTS_REORDERED|TF_DELAY_RECOVERY)) == - (TF_PKTS_REORDERED|TF_DELAY_RECOVERY)) { + (TF_PKTS_REORDERED | TF_DELAY_RECOVERY)) == + (TF_PKTS_REORDERED | TF_DELAY_RECOVERY)) { /* * Since the SACK information is already * updated, this ACK will be dropped @@ -4257,38 +4466,46 @@ process_dupack: break; } - if (SACK_ENABLED(tp) - && IN_FASTRECOVERY(tp)) { + /* + * Dup acks mean that packets have left the + * network (they're now cached at the receiver) + * so bump cwnd by the amount in the receiver + * to keep a constant cwnd packets in the + * network. + */ + if (SACK_ENABLED(tp) && IN_FASTRECOVERY(tp)) { int awnd; /* * Compute the amount of data in flight first. * We can inject new data into the pipe iff - * we have less than 1/2 the original window's - * worth of data in flight. + * we have less than snd_ssthres worth of data in + * flight. */ - awnd = (tp->snd_nxt - tp->snd_fack) + - tp->sackhint.sack_bytes_rexmit; + awnd = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; if (awnd < tp->snd_ssthresh) { tp->snd_cwnd += tp->t_maxseg; - if (tp->snd_cwnd > tp->snd_ssthresh) + if (tp->snd_cwnd > tp->snd_ssthresh) { tp->snd_cwnd = tp->snd_ssthresh; + } } } else { tp->snd_cwnd += tp->t_maxseg; } /* Process any window updates */ - if (tiwin > tp->snd_wnd) + if (tiwin > tp->snd_wnd) { tcp_update_window(tp, thflags, th, tiwin, tlen); + } tcp_ccdbg_trace(tp, th, TCP_CC_IN_FASTRECOVERY); (void) tcp_output(tp); goto drop; - } else if (tp->t_dupacks == tp->t_rexmtthresh) { + } else if ((!tcp_do_better_lr && tp->t_dupacks == tp->t_rexmtthresh) || + (tcp_do_better_lr && tp->t_dupacks >= tp->t_rexmtthresh)) { tcp_seq onxt = tp->snd_nxt; /* @@ -4306,16 +4523,16 @@ process_dupack: break; } } else { - if (SEQ_LEQ(th->th_ack, - tp->snd_recover)) { + if (SEQ_LEQ(th->th_ack, tp->snd_recover)) { tp->t_dupacks = 0; break; } } - if (tp->t_flags & TF_SENTFIN) + if (tp->t_flags & TF_SENTFIN) { tp->snd_recover = tp->snd_max - 1; - else + } else { tp->snd_recover = tp->snd_max; + } tp->t_timer[TCPT_PTO] = 0; tp->t_rtttime = 0; @@ -4327,7 +4544,7 @@ process_dupack: */ if (SACK_ENABLED(tp) && (tp->t_flagsext & - (TF_PKTS_REORDERED|TF_DELAY_RECOVERY)) + (TF_PKTS_REORDERED | TF_DELAY_RECOVERY)) == TF_PKTS_REORDERED && !IN_FASTRECOVERY(tp) && tp->t_reorderwin > 0 && @@ -4349,29 +4566,32 @@ process_dupack: * defined a hook for tasks to run * before entering FR, call it */ - if (CC_ALGO(tp)->pre_fr != NULL) + if (CC_ALGO(tp)->pre_fr != NULL) { CC_ALGO(tp)->pre_fr(tp); + } ENTER_FASTRECOVERY(tp); tp->t_timer[TCPT_REXMT] = 0; - if (TCP_ECN_ENABLED(tp)) + if (TCP_ECN_ENABLED(tp)) { tp->ecn_flags |= TE_SENDCWR; + } if (SACK_ENABLED(tp)) { tcpstat.tcps_sack_recovery_episode++; tp->t_sack_recovery_episode++; tp->sack_newdata = tp->snd_nxt; - tp->snd_cwnd = tp->t_maxseg; - tp->t_flagsext &= - ~TF_CWND_NONVALIDATED; + if (tcp_do_better_lr) { + tp->snd_cwnd = tp->snd_ssthresh; + } else { + tp->snd_cwnd = tp->t_maxseg; + } + tp->t_flagsext &= ~TF_CWND_NONVALIDATED; /* Process any window updates */ - if (tiwin > tp->snd_wnd) - tcp_update_window( - tp, thflags, - th, tiwin, tlen); + if (tiwin > tp->snd_wnd) { + tcp_update_window(tp, thflags, th, tiwin, tlen); + } - tcp_ccdbg_trace(tp, th, - TCP_CC_ENTER_FASTRECOVERY); + tcp_ccdbg_trace(tp, th, TCP_CC_ENTER_FASTRECOVERY); (void) tcp_output(tp); goto drop; } @@ -4379,28 +4599,25 @@ process_dupack: tp->snd_cwnd = tp->t_maxseg; /* Process any window updates */ - if (tiwin > tp->snd_wnd) - tcp_update_window(tp, - thflags, - th, tiwin, tlen); + if (tiwin > tp->snd_wnd) { + tcp_update_window(tp, thflags, th, tiwin, tlen); + } (void) tcp_output(tp); if (tp->t_flagsext & TF_CWND_NONVALIDATED) { tcp_cc_adjust_nonvalidated_cwnd(tp); } else { - tp->snd_cwnd = tp->snd_ssthresh + - tp->t_maxseg * tp->t_dupacks; + tp->snd_cwnd = tp->snd_ssthresh + tp->t_maxseg * tp->t_dupacks; } - if (SEQ_GT(onxt, tp->snd_nxt)) + if (SEQ_GT(onxt, tp->snd_nxt)) { tp->snd_nxt = onxt; + } - tcp_ccdbg_trace(tp, th, - TCP_CC_ENTER_FASTRECOVERY); + tcp_ccdbg_trace(tp, th, TCP_CC_ENTER_FASTRECOVERY); goto drop; - } else if (limited_txmt && - ALLOW_LIMITED_TRANSMIT(tp) && - (!(SACK_ENABLED(tp)) || sack_bytes_acked > 0) && - (so->so_snd.sb_cc - (tp->snd_max - tp->snd_una)) > 0) { + } else if (ALLOW_LIMITED_TRANSMIT(tp) && + (!(SACK_ENABLED(tp)) || sack_bytes_acked > 0) && + (so->so_snd.sb_cc - (tp->snd_max - tp->snd_una)) > 0) { u_int32_t incr = (tp->t_maxseg * tp->t_dupacks); /* Use Limited Transmit algorithm on the first two @@ -4429,26 +4646,32 @@ process_dupack: * recovery, the subsequent ACKs should * not be treated as partial acks. */ - if (tp->ecn_flags & TE_INRECOVERY) + if (tp->ecn_flags & TE_INRECOVERY) { goto process_ACK; + } - if (SACK_ENABLED(tp)) + if (SACK_ENABLED(tp)) { tcp_sack_partialack(tp, th); - else + } else { tcp_newreno_partial_ack(tp, th); + } tcp_ccdbg_trace(tp, th, TCP_CC_PARTIAL_ACK); } else { + if (tcp_cubic_minor_fixes) { + exiting_fr = 1; + } EXIT_FASTRECOVERY(tp); - if (CC_ALGO(tp)->post_fr != NULL) + if (CC_ALGO(tp)->post_fr != NULL) { CC_ALGO(tp)->post_fr(tp, th); + } tp->t_pipeack = 0; tcp_clear_pipeack_state(tp); tcp_ccdbg_trace(tp, th, TCP_CC_EXIT_FASTRECOVERY); } } else if ((tp->t_flagsext & - (TF_PKTS_REORDERED|TF_DELAY_RECOVERY)) - == (TF_PKTS_REORDERED|TF_DELAY_RECOVERY)) { + (TF_PKTS_REORDERED | TF_DELAY_RECOVERY)) + == (TF_PKTS_REORDERED | TF_DELAY_RECOVERY)) { /* * If the ack acknowledges upto snd_recover or if * it acknowledges all the snd holes, exit @@ -4467,33 +4690,12 @@ process_dupack: } } else { /* - * We were not in fast recovery. Reset the - * duplicate ack counter. - */ - tp->t_dupacks = 0; - tp->t_rexmtthresh = tcprexmtthresh; - } - - - /* - * If we reach this point, ACK is not a duplicate, - * i.e., it ACKs something we sent. - */ - if (tp->t_flags & TF_NEEDSYN) { - /* - * T/TCP: Connection was half-synchronized, and our - * SYN has been ACK'd (so connection is now fully - * synchronized). Go to non-starred state, - * increment snd_una for ACK of SYN, and check if - * we can do window scaling. - */ - tp->t_flags &= ~TF_NEEDSYN; - tp->snd_una++; - /* Do window scaling? */ - if (TCP_WINDOW_SCALE_ENABLED(tp)) { - tp->snd_scale = tp->requested_s_scale; - tp->rcv_scale = tp->request_r_scale; - } + * We were not in fast recovery. Reset the + * duplicate ack counter. + */ + tp->t_dupacks = 0; + tp->t_rexmtthresh = tcprexmtthresh; + tp->t_new_dupacks = 0; } process_ACK: @@ -4529,16 +4731,24 @@ process_ACK: tp->t_timer[TCPT_REXMT] = 0; tp->t_timer[TCPT_PTO] = 0; needoutput = 1; - } else if (tp->t_timer[TCPT_PERSIST] == 0) + } else if (tp->t_timer[TCPT_PERSIST] == 0) { tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp, tp->t_rxtcur); + } + + if ((prev_t_state == TCPS_SYN_SENT || + prev_t_state == TCPS_SYN_RECEIVED) && + tp->t_state == TCPS_ESTABLISHED) { + TCP_LOG_RTT_INFO(tp); + } /* * If no data (only SYN) was ACK'd, skip rest of ACK * processing. */ - if (acked == 0) + if (acked == 0) { goto step6; + } /* * When outgoing data has been acked (except the SYN+data), we @@ -4546,8 +4756,9 @@ process_ACK: */ if ((tp->t_tfo_stats & TFO_S_SYN_DATA_SENT) && !(tp->t_tfo_flags & TFO_F_NO_SNDPROBING) && - !(th->th_flags & TH_SYN)) + !(th->th_flags & TH_SYN)) { tp->t_tfo_flags |= TFO_F_NO_SNDPROBING; + } /* * If TH_ECE is received, make sure that ECN is enabled @@ -4561,7 +4772,7 @@ process_ACK: */ if (!IN_FASTRECOVERY(tp)) { tcp_reduce_congestion_window(tp); - tp->ecn_flags |= (TE_INRECOVERY|TE_SENDCWR); + tp->ecn_flags |= (TE_INRECOVERY | TE_SENDCWR); /* * Also note that the connection received * ECE atleast once @@ -4581,67 +4792,61 @@ process_ACK: * The calculations in this function assume that snd_una is * not updated yet. */ - if (!IN_FASTRECOVERY(tp)) { - if (CC_ALGO(tp)->ack_rcvd != NULL) + if (!IN_FASTRECOVERY(tp) && !exiting_fr) { + if (CC_ALGO(tp)->ack_rcvd != NULL) { CC_ALGO(tp)->ack_rcvd(tp, th); + } tcp_ccdbg_trace(tp, th, TCP_CC_ACK_RCVD); } if (acked > so->so_snd.sb_cc) { tp->snd_wnd -= so->so_snd.sb_cc; sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); - if (so->so_flags & SOF_ENABLE_MSGS) { - so->so_msg_state->msg_serial_bytes -= - (int)so->so_snd.sb_cc; - } ourfinisacked = 1; } else { sbdrop(&so->so_snd, acked); - if (so->so_flags & SOF_ENABLE_MSGS) { - so->so_msg_state->msg_serial_bytes -= - acked; - } tcp_sbsnd_trim(&so->so_snd); tp->snd_wnd -= acked; ourfinisacked = 0; } /* detect una wraparound */ - if ( !IN_FASTRECOVERY(tp) && + if (!IN_FASTRECOVERY(tp) && SEQ_GT(tp->snd_una, tp->snd_recover) && - SEQ_LEQ(th->th_ack, tp->snd_recover)) + SEQ_LEQ(th->th_ack, tp->snd_recover)) { tp->snd_recover = th->th_ack - 1; + } if (IN_FASTRECOVERY(tp) && - SEQ_GEQ(th->th_ack, tp->snd_recover)) + SEQ_GEQ(th->th_ack, tp->snd_recover)) { EXIT_FASTRECOVERY(tp); + } - tp->snd_una = th->th_ack; + tcp_update_snd_una(tp, th->th_ack); if (SACK_ENABLED(tp)) { - if (SEQ_GT(tp->snd_una, tp->snd_recover)) + if (SEQ_GT(tp->snd_una, tp->snd_recover)) { tp->snd_recover = tp->snd_una; + } } - if (SEQ_LT(tp->snd_nxt, tp->snd_una)) + if (SEQ_LT(tp->snd_nxt, tp->snd_una)) { tp->snd_nxt = tp->snd_una; + } if (!SLIST_EMPTY(&tp->t_rxt_segments) && !TCP_DSACK_SEQ_IN_WINDOW(tp, tp->t_dsack_lastuna, - tp->snd_una)) + tp->snd_una)) { tcp_rxtseg_clean(tp); + } if ((tp->t_flagsext & TF_MEASURESNDBW) != 0 && - tp->t_bwmeas != NULL) + tp->t_bwmeas != NULL) { tcp_bwmeas_check(tp); + } - /* - * sowwakeup must happen after snd_una, et al. are - * updated so that the sequence numbers are in sync with - * so_snd - */ - sowwakeup(so); + write_wakeup = 1; - if (!SLIST_EMPTY(&tp->t_notify_ack)) + if (!SLIST_EMPTY(&tp->t_notify_ack)) { tcp_notify_acknowledgement(tp, so); + } switch (tp->t_state) { - /* * In FIN_WAIT_1 STATE in addition to the processing * for the ESTABLISHED state if our FIN is now acknowledged @@ -4658,14 +4863,14 @@ process_ACK: */ if (so->so_state & SS_CANTRCVMORE) { tp->t_timer[TCPT_2MSL] = OFFSET_FROM_START(tp, - TCP_CONN_MAXIDLE(tp)); + TCP_CONN_MAXIDLE(tp)); isconnected = FALSE; isdisconnected = TRUE; } DTRACE_TCP4(state__change, void, NULL, - struct inpcb *, inp, - struct tcpcb *, tp, - int32_t, TCPS_FIN_WAIT_2); + struct inpcb *, inp, + struct tcpcb *, tp, + int32_t, TCPS_FIN_WAIT_2); tp->t_state = TCPS_FIN_WAIT_2; /* fall through and make sure we also recognize * data ACKed with the FIN @@ -4673,7 +4878,7 @@ process_ACK: } break; - /* + /* * In CLOSING STATE in addition to the processing for * the ESTABLISHED state if the ACK acknowledges our FIN * then enter the TIME-WAIT state, otherwise ignore @@ -4682,9 +4887,9 @@ process_ACK: case TCPS_CLOSING: if (ourfinisacked) { DTRACE_TCP4(state__change, void, NULL, - struct inpcb *, inp, - struct tcpcb *, tp, - int32_t, TCPS_TIME_WAIT); + struct inpcb *, inp, + struct tcpcb *, tp, + int32_t, TCPS_TIME_WAIT); tp->t_state = TCPS_TIME_WAIT; tcp_canceltimers(tp); if (tp->t_flagsext & TF_NOTIMEWAIT) { @@ -4730,8 +4935,7 @@ process_ACK: * a SACK option. So counting it as one duplicate * ack is ok. */ - if (sack_ackadv == 1 && - tp->t_state == TCPS_ESTABLISHED && + if (tp->t_state == TCPS_ESTABLISHED && SACK_ENABLED(tp) && sack_bytes_acked > 0 && to.to_nsacks > 0 && tp->t_dupacks == 0 && SEQ_LEQ(th->th_ack, tp->snd_una) && tlen == 0 && @@ -4745,8 +4949,9 @@ step6: /* * Update window information. */ - if (tcp_update_window(tp, thflags, th, tiwin, tlen)) + if (tcp_update_window(tp, thflags, th, tiwin, tlen)) { needoutput = 1; + } /* * Process segments with URG. @@ -4760,9 +4965,9 @@ step6: * actually wanting to send this much urgent data. */ if (th->th_urp + so->so_rcv.sb_cc > sb_max) { - th->th_urp = 0; /* XXX */ - thflags &= ~TH_URG; /* XXX */ - goto dodata; /* XXX */ + th->th_urp = 0; /* XXX */ + thflags &= ~TH_URG; /* XXX */ + goto dodata; /* XXX */ } /* * If this segment advances the known urgent pointer, @@ -4778,13 +4983,12 @@ step6: * of data past the urgent section as the original * spec states (in one of two places). */ - if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { + if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) { tp->rcv_up = th->th_seq + th->th_urp; so->so_oobmark = so->so_rcv.sb_cc + (tp->rcv_up - tp->rcv_nxt) - 1; if (so->so_oobmark == 0) { so->so_state |= SS_RCVATMARK; - postevent(so, 0, EV_OOB); } sohasoutofband(so); tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); @@ -4797,19 +5001,21 @@ step6: */ if (th->th_urp <= (u_int32_t)tlen #if SO_OOBINLINE - && (so->so_options & SO_OOBINLINE) == 0 + && (so->so_options & SO_OOBINLINE) == 0 #endif - ) + ) { tcp_pulloutofband(so, th, m, - drop_hdrlen); /* hdr drop is delayed */ + drop_hdrlen); /* hdr drop is delayed */ + } } else { /* * If no out of band data is expected, * pull receive urgent pointer along * with the receive window. */ - if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) + if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) { tp->rcv_up = tp->rcv_nxt; + } } dodata: @@ -4828,6 +5034,7 @@ dodata: */ if (inp->inp_state == INPCB_STATE_DEAD) { /* Just drop the packet that we are processing and return */ + TCP_LOG_DROP_PCB(TCP_LOG_HDR, th, tp, false, "INPCB_STATE_DEAD"); goto drop; } @@ -4845,11 +5052,11 @@ dodata: if ((tlen || (thflags & TH_FIN)) && TCPS_HAVERCVDFIN(tp->t_state) == 0 && (TCPS_HAVEESTABLISHED(tp->t_state) || - (tp->t_state == TCPS_SYN_RECEIVED && - (tp->t_tfo_flags & TFO_F_COOKIE_VALID)))) { + (tp->t_state == TCPS_SYN_RECEIVED && + (tp->t_tfo_flags & TFO_F_COOKIE_VALID)))) { tcp_seq save_start = th->th_seq; tcp_seq save_end = th->th_seq + tlen; - m_adj(m, drop_hdrlen); /* delayed header drop */ + m_adj(m, drop_hdrlen); /* delayed header drop */ /* * Insert segment which includes th into TCP reassembly queue * with control block tp. Set thflags to whether reassembly now @@ -4863,87 +5070,112 @@ dodata: * fast retransmit can work). */ if (th->th_seq == tp->rcv_nxt && LIST_EMPTY(&tp->t_segq)) { - TCP_INC_VAR(tp->t_unacksegs, nlropkts); + TCP_INC_VAR(tp->t_unacksegs, segment_count); /* * Calculate the RTT on the receiver only if the * connection is in streaming mode and the last * packet was not an end-of-write */ - if (tp->t_flags & TF_STREAMING_ON) + if (tp->t_flags & TF_STREAMING_ON) { tcp_compute_rtt(tp, &to, th); + } if (DELAY_ACK(tp, th) && - ((tp->t_flags & TF_ACKNOW) == 0) ) { + ((tp->t_flags & TF_ACKNOW) == 0)) { if ((tp->t_flags & TF_DELACK) == 0) { tp->t_flags |= TF_DELACK; tp->t_timer[TCPT_DELACK] = - OFFSET_FROM_START(tp, tcp_delack); + OFFSET_FROM_START(tp, tcp_delack); } - } - else { + } else { tp->t_flags |= TF_ACKNOW; } tp->rcv_nxt += tlen; thflags = th->th_flags & TH_FIN; - TCP_INC_VAR(tcpstat.tcps_rcvpack, nlropkts); + TCP_INC_VAR(tcpstat.tcps_rcvpack, segment_count); tcpstat.tcps_rcvbyte += tlen; if (nstat_collect) { - if (m->m_pkthdr.pkt_flags & PKTF_SW_LRO_PKT) { - INP_ADD_STAT(inp, cell, wifi, wired, - rxpackets, m->m_pkthdr.lro_npkts); - } else { - INP_ADD_STAT(inp, cell, wifi, wired, - rxpackets, 1); - } + INP_ADD_STAT(inp, cell, wifi, wired, + rxpackets, 1); INP_ADD_STAT(inp, cell, wifi, wired, rxbytes, tlen); inp_set_activity_bitmap(inp); } - tcp_sbrcv_grow(tp, &so->so_rcv, &to, tlen, - TCP_AUTORCVBUF_MAX(ifp)); + tcp_sbrcv_grow(tp, &so->so_rcv, &to, tlen); so_recv_data_stat(so, m, drop_hdrlen); - if (sbappendstream_rcvdemux(so, m, - th->th_seq - (tp->irs + 1), 0)) { - sorwakeup(so); + if (isipv6) { + memcpy(&saved_hdr, ip6, sizeof(struct ip6_hdr)); + ip6 = (struct ip6_hdr *)&saved_hdr[0]; + } else { + memcpy(&saved_hdr, ip, ip->ip_hl << 2); + ip = (struct ip *)&saved_hdr[0]; + } + memcpy(&saved_tcphdr, th, sizeof(struct tcphdr)); + + if (th->th_flags & TH_PUSH) { + tp->t_flagsext |= TF_LAST_IS_PSH; + } else { + tp->t_flagsext &= ~TF_LAST_IS_PSH; + } + + if (sbappendstream_rcvdemux(so, m)) { + read_wakeup = 1; } + th = &saved_tcphdr; } else { - thflags = tcp_reass(tp, th, &tlen, m, ifp); + if (isipv6) { + memcpy(&saved_hdr, ip6, sizeof(struct ip6_hdr)); + ip6 = (struct ip6_hdr *)&saved_hdr[0]; + } else { + memcpy(&saved_hdr, ip, ip->ip_hl << 2); + ip = (struct ip *)&saved_hdr[0]; + } + + if (tcp_autotune_reorder) { + tcp_sbrcv_grow(tp, &so->so_rcv, &to, tlen); + } + + memcpy(&saved_tcphdr, th, sizeof(struct tcphdr)); + thflags = tcp_reass(tp, th, &tlen, m, ifp, &read_wakeup); + th = &saved_tcphdr; tp->t_flags |= TF_ACKNOW; } if ((tlen > 0 || (th->th_flags & TH_FIN)) && SACK_ENABLED(tp)) { - if (th->th_flags & TH_FIN) + if (th->th_flags & TH_FIN) { save_end++; + } tcp_update_sack_list(tp, save_start, save_end); } tcp_adaptive_rwtimo_check(tp, tlen); - if (tlen > 0) + if (tlen > 0) { tcp_tfo_rcv_data(tp); + } - if (tp->t_flags & TF_DELACK) - { -#if INET6 + if (tp->t_flags & TF_DELACK) { if (isipv6) { KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport), - (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])), - th->th_seq, th->th_ack, th->th_win); - } - else -#endif - { + (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])), + th->th_seq, th->th_ack, th->th_win); + } else { KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport), - (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)), - th->th_seq, th->th_ack, th->th_win); + (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)), + th->th_seq, th->th_ack, th->th_win); } - } } else { if ((so->so_flags & SOF_MP_SUBFLOW) && tlen == 0 && - (m->m_pkthdr.pkt_flags & PKTF_MPTCP_DFIN)) { - m_adj(m, drop_hdrlen); /* delayed header drop */ + (m->m_pkthdr.pkt_flags & PKTF_MPTCP_DFIN) && + (m->m_pkthdr.pkt_flags & PKTF_MPTCP)) { + m_adj(m, drop_hdrlen); /* delayed header drop */ + /* + * 0-length DATA_FIN. The rlen is actually 0. We special-case the + * byte consumed by the dfin in mptcp_input and mptcp_reass_present + */ + m->m_pkthdr.mp_rlen = 0; mptcp_input(tptomptp(tp)->mpt_mpte, m); tp->t_flags |= TF_ACKNOW; } else { @@ -4959,7 +5191,6 @@ dodata: if (thflags & TH_FIN) { if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { socantrcvmore(so); - postevent(so, 0, EV_FIN); /* * If connection is half-synchronized * (ie NEEDSYN flag on) then delay ACK, @@ -4967,51 +5198,44 @@ dodata: * Otherwise, since we received a FIN then no * more input can be expected, send ACK now. */ - TCP_INC_VAR(tp->t_unacksegs, nlropkts); - if (DELAY_ACK(tp, th) && (tp->t_flags & TF_NEEDSYN)) { - if ((tp->t_flags & TF_DELACK) == 0) { - tp->t_flags |= TF_DELACK; - tp->t_timer[TCPT_DELACK] = OFFSET_FROM_START(tp, tcp_delack); - } - } else { - tp->t_flags |= TF_ACKNOW; - } + TCP_INC_VAR(tp->t_unacksegs, segment_count); + tp->t_flags |= TF_ACKNOW; tp->rcv_nxt++; } switch (tp->t_state) { - - /* + /* * In SYN_RECEIVED and ESTABLISHED STATES * enter the CLOSE_WAIT state. */ case TCPS_SYN_RECEIVED: tp->t_starttime = tcp_now; + OS_FALLTHROUGH; case TCPS_ESTABLISHED: DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, - struct tcpcb *, tp, int32_t, TCPS_CLOSE_WAIT); + struct tcpcb *, tp, int32_t, TCPS_CLOSE_WAIT); tp->t_state = TCPS_CLOSE_WAIT; break; - /* + /* * If still in FIN_WAIT_1 STATE FIN has not been acked so * enter the CLOSING state. */ case TCPS_FIN_WAIT_1: DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, - struct tcpcb *, tp, int32_t, TCPS_CLOSING); + struct tcpcb *, tp, int32_t, TCPS_CLOSING); tp->t_state = TCPS_CLOSING; break; - /* + /* * In FIN_WAIT_2 state enter the TIME_WAIT state, * starting the time-wait timer, turning off the other * standard timers. */ case TCPS_FIN_WAIT_2: DTRACE_TCP4(state__change, void, NULL, - struct inpcb *, inp, - struct tcpcb *, tp, - int32_t, TCPS_TIME_WAIT); + struct inpcb *, inp, + struct tcpcb *, tp, + int32_t, TCPS_TIME_WAIT); tp->t_state = TCPS_TIME_WAIT; tcp_canceltimers(tp); tp->t_flags |= TF_ACKNOW; @@ -5032,11 +5256,16 @@ dodata: } } #if TCPDEBUG - if (so->so_options & SO_DEBUG) + if (so->so_options & SO_DEBUG) { tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, - &tcp_savetcp, 0); + &tcp_savetcp, 0); + } #endif + if (read_wakeup) { + mptcp_handle_input(so); + } + /* * Return any desired output. */ @@ -5046,9 +5275,10 @@ dodata: tcp_check_timer_state(tp); + tcp_handle_wakeup(so, read_wakeup, write_wakeup); socket_unlock(so, 1); - KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END, 0, 0, 0, 0, 0); return; dropafterack: @@ -5069,23 +5299,26 @@ dropafterack: */ if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && (SEQ_GT(tp->snd_una, th->th_ack) || - SEQ_GT(th->th_ack, tp->snd_max)) ) { - rstreason = BANDLIM_RST_OPENPORT; + SEQ_GT(th->th_ack, tp->snd_max))) { IF_TCP_STATINC(ifp, dospacket); goto dropwithreset; } #if TCPDEBUG - if (so->so_options & SO_DEBUG) + if (so->so_options & SO_DEBUG) { tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, - &tcp_savetcp, 0); + &tcp_savetcp, 0); + } #endif m_freem(m); tp->t_flags |= TF_ACKNOW; + (void) tcp_output(tp); + tcp_handle_wakeup(so, read_wakeup, write_wakeup); + /* Don't need to check timer state as we should have done it during tcp_output */ socket_unlock(so, 1); - KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END, 0, 0, 0, 0, 0); return; dropwithresetnosock: nosock = 1; @@ -5095,58 +5328,54 @@ dropwithreset: * Make ACK acceptable to originator of segment. * Don't bother to respond if destination was broadcast/multicast. */ - if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) + if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST)) { goto drop; -#if INET6 + } if (isipv6) { if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || - IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) + IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { goto drop; - } else -#endif /* INET6 */ - if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || + } + } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || - in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) + in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { goto drop; + } /* IPv6 anycast check is done at tcp6_input() */ - /* - * Perform bandwidth limiting. - */ -#if ICMP_BANDLIM - if (badport_bandlim(rstreason) < 0) - goto drop; -#endif - #if TCPDEBUG - if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) + if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) { tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, - &tcp_savetcp, 0); + &tcp_savetcp, 0); + } #endif bzero(&tra, sizeof(tra)); tra.ifscope = ifscope; tra.awdl_unrestricted = 1; tra.intcoproc_allowed = 1; - if (thflags & TH_ACK) + if (thflags & TH_ACK) { /* mtod() below is safe as long as hdr dropping is delayed */ tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, TH_RST, &tra); - else { - if (thflags & TH_SYN) + } else { + if (thflags & TH_SYN) { tlen++; + } /* mtod() below is safe as long as hdr dropping is delayed */ - tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, - (tcp_seq)0, TH_RST|TH_ACK, &tra); + tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen, + (tcp_seq)0, TH_RST | TH_ACK, &tra); } /* destroy temporarily created socket */ if (dropsocket) { (void) soabort(so); socket_unlock(so, 1); } else if ((inp != NULL) && (nosock == 0)) { + tcp_handle_wakeup(so, read_wakeup, write_wakeup); + socket_unlock(so, 1); } - KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END, 0, 0, 0, 0, 0); return; dropnosock: nosock = 1; @@ -5155,20 +5384,22 @@ drop: * Drop space held by incoming segment and return. */ #if TCPDEBUG - if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) + if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) { tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, - &tcp_savetcp, 0); + &tcp_savetcp, 0); + } #endif m_freem(m); /* destroy temporarily created socket */ if (dropsocket) { (void) soabort(so); socket_unlock(so, 1); - } - else if (nosock == 0) { + } else if (nosock == 0) { + tcp_handle_wakeup(so, read_wakeup, write_wakeup); + socket_unlock(so, 1); } - KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END, 0, 0, 0, 0, 0); return; } @@ -5184,27 +5415,31 @@ tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, for (; cnt > 0; cnt -= optlen, cp += optlen) { opt = cp[0]; - if (opt == TCPOPT_EOL) + if (opt == TCPOPT_EOL) { break; - if (opt == TCPOPT_NOP) + } + if (opt == TCPOPT_NOP) { optlen = 1; - else { - if (cnt < 2) + } else { + if (cnt < 2) { break; + } optlen = cp[1]; - if (optlen < 2 || optlen > cnt) + if (optlen < 2 || optlen > cnt) { break; + } } switch (opt) { - default: continue; case TCPOPT_MAXSEG: - if (optlen != TCPOLEN_MAXSEG) + if (optlen != TCPOLEN_MAXSEG) { continue; - if (!(th->th_flags & TH_SYN)) + } + if (!(th->th_flags & TH_SYN)) { continue; + } bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); NTOHS(mss); to->to_mss = mss; @@ -5212,17 +5447,20 @@ tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, break; case TCPOPT_WINDOW: - if (optlen != TCPOLEN_WINDOW) + if (optlen != TCPOLEN_WINDOW) { continue; - if (!(th->th_flags & TH_SYN)) + } + if (!(th->th_flags & TH_SYN)) { continue; + } to->to_flags |= TOF_SCALE; to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); break; case TCPOPT_TIMESTAMP: - if (optlen != TCPOLEN_TIMESTAMP) + if (optlen != TCPOLEN_TIMESTAMP) { continue; + } to->to_flags |= TOF_TS; bcopy((char *)cp + 2, (char *)&to->to_tsval, sizeof(to->to_tsval)); @@ -5230,21 +5468,24 @@ tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, bcopy((char *)cp + 6, (char *)&to->to_tsecr, sizeof(to->to_tsecr)); NTOHL(to->to_tsecr); + to->to_tsecr -= tp->t_ts_offset; /* Re-enable sending Timestamps if we received them */ - if (!(tp->t_flags & TF_REQ_TSTMP) && - tcp_do_rfc1323 == 1) + if (!(tp->t_flags & TF_REQ_TSTMP)) { tp->t_flags |= TF_REQ_TSTMP; + } break; case TCPOPT_SACK_PERMITTED: - if (!tcp_do_sack || - optlen != TCPOLEN_SACK_PERMITTED) + if (optlen != TCPOLEN_SACK_PERMITTED) { continue; - if (th->th_flags & TH_SYN) + } + if (th->th_flags & TH_SYN) { to->to_flags |= TOF_SACK; + } break; case TCPOPT_SACK: - if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) + if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) { continue; + } to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; to->to_sacks = cp + 2; tcpstat.tcps_sack_rcv_blocks++; @@ -5252,18 +5493,21 @@ tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, break; case TCPOPT_FASTOPEN: if (optlen == TCPOLEN_FASTOPEN_REQ) { - if (tp->t_state != TCPS_LISTEN) + if (tp->t_state != TCPS_LISTEN) { continue; + } to->to_flags |= TOF_TFOREQ; } else { if (optlen < TCPOLEN_FASTOPEN_REQ || (optlen - TCPOLEN_FASTOPEN_REQ) > TFO_COOKIE_LEN_MAX || - (optlen - TCPOLEN_FASTOPEN_REQ) < TFO_COOKIE_LEN_MIN) + (optlen - TCPOLEN_FASTOPEN_REQ) < TFO_COOKIE_LEN_MIN) { continue; + } if (tp->t_state != TCPS_LISTEN && - tp->t_state != TCPS_SYN_SENT) + tp->t_state != TCPS_SYN_SENT) { continue; + } to->to_flags |= TOF_TFO; to->to_tfo = cp + 1; @@ -5286,23 +5530,25 @@ tcp_finalize_options(struct tcpcb *tp, struct tcpopt *to, unsigned int ifscope) tp->t_flags |= TF_RCVD_TSTMP; tp->ts_recent = to->to_tsval; tp->ts_recent_age = tcp_now; - } - if (to->to_flags & TOF_MSS) + if (to->to_flags & TOF_MSS) { tcp_mss(tp, to->to_mss, ifscope); + } if (SACK_ENABLED(tp)) { - if (!(to->to_flags & TOF_SACK)) + if (!(to->to_flags & TOF_SACK)) { tp->t_flagsext &= ~(TF_SACK_ENABLE); - else + } else { tp->t_flags |= TF_SACK_PERMIT; + } } if (to->to_flags & TOF_SCALE) { tp->t_flags |= TF_RCVD_SCALE; tp->requested_s_scale = to->to_requested_s_scale; /* Re-enable window scaling, if the option is received */ - if (tp->request_r_scale > 0) + if (tp->request_r_scale > 0) { tp->t_flags |= TF_REQ_SCALE; + } } } @@ -5326,16 +5572,18 @@ tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off) tp->t_iobc = *cp; tp->t_oobflags |= TCPOOB_HAVEDATA; - bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); + bcopy(cp + 1, cp, (unsigned)(m->m_len - cnt - 1)); m->m_len--; - if (m->m_flags & M_PKTHDR) + if (m->m_flags & M_PKTHDR) { m->m_pkthdr.len--; + } return; } cnt -= m->m_len; m = m->m_next; - if (m == 0) + if (m == 0) { break; + } } panic("tcp_pulloutofband"); } @@ -5344,7 +5592,7 @@ uint32_t get_base_rtt(struct tcpcb *tp) { struct rtentry *rt = tp->t_inpcb->inp_route.ro_rt; - return ((rt == NULL) ? 0 : rt->rtt_min); + return (rt == NULL) ? 0 : rt->rtt_min; } /* Each value of RTT base represents the minimum RTT seen in a minute. @@ -5356,8 +5604,9 @@ update_base_rtt(struct tcpcb *tp, uint32_t rtt) u_int32_t base_rtt, i; struct rtentry *rt; - if ((rt = tp->t_inpcb->inp_route.ro_rt) == NULL) + if ((rt = tp->t_inpcb->inp_route.ro_rt) == NULL) { return; + } if (rt->rtt_expire_ts == 0) { RT_LOCK_SPIN(rt); if (rt->rtt_expire_ts != 0) { @@ -5394,8 +5643,9 @@ update: if ((int)(tcp_now - rt->rtt_expire_ts) >= TCP_RTT_HISTORY_EXPIRE_TIME) { rt->rtt_index++; - if (rt->rtt_index >= NRTT_HIST) + if (rt->rtt_index >= NRTT_HIST) { rt->rtt_index = 0; + } rt->rtt_hist[rt->rtt_index] = rtt; rt->rtt_expire_ts = tcp_now; } else { @@ -5407,17 +5657,19 @@ update: for (i = 0; i < NRTT_HIST; ++i) { if (rt->rtt_hist[i] != 0 && (rt->rtt_min == 0 || - rt->rtt_hist[i] < rt->rtt_min)) + rt->rtt_hist[i] < rt->rtt_min)) { rt->rtt_min = rt->rtt_hist[i]; + } } RT_UNLOCK(rt); } else { rt->rtt_hist[rt->rtt_index] = min(rt->rtt_hist[rt->rtt_index], rtt); - if (rt->rtt_min == 0) + if (rt->rtt_min == 0) { rt->rtt_min = rtt; - else + } else { rt->rtt_min = min(rt->rtt_min, rtt); + } } } @@ -5455,17 +5707,17 @@ tcp_compute_rtt(struct tcpcb *tp, struct tcpopt *to, struct tcphdr *th) /* Compute the max of the pipeack samples */ pipe_ack_val = tcp_get_max_pipeack(tp); tp->t_pipeack = (pipe_ack_val > - TCP_CC_CWND_INIT_BYTES) ? - pipe_ack_val : 0; + tcp_initial_cwnd(tp)) ? + pipe_ack_val : 0; } /* start another measurement */ tp->t_rtttime = 0; } if (((to->to_flags & TOF_TS) != 0) && - (to->to_tsecr != 0) && - TSTMP_GEQ(tcp_now, to->to_tsecr)) { + (to->to_tsecr != 0) && + TSTMP_GEQ(tcp_now, to->to_tsecr)) { tcp_xmit_timer(tp, (tcp_now - to->to_tsecr), - to->to_tsecr, th->th_ack); + to->to_tsecr, th->th_ack); } else if (rtt > 0) { tcp_xmit_timer(tp, rtt, 0, th->th_ack); } @@ -5477,9 +5729,12 @@ tcp_compute_rtt(struct tcpcb *tp, struct tcpopt *to, struct tcphdr *th) */ static void tcp_xmit_timer(struct tcpcb *tp, int rtt, - u_int32_t tsecr, tcp_seq th_ack) + u_int32_t tsecr, tcp_seq th_ack) { int delta; + int old_srtt = tp->t_srtt; + int old_rttvar = tp->t_rttvar; + bool log_rtt = false; /* * On AWDL interface, the initial RTT measurement on SYN @@ -5489,8 +5744,9 @@ tcp_xmit_timer(struct tcpcb *tp, int rtt, */ if (tp->t_inpcb->inp_last_outifp != NULL && (tp->t_inpcb->inp_last_outifp->if_eflags & IFEF_AWDL) && - th_ack == tp->iss + 1) + th_ack == tp->iss + 1) { return; + } if (tp->t_flagsext & TF_RECOMPUTE_RTT) { if (SEQ_GT(th_ack, tp->snd_una) && @@ -5498,7 +5754,7 @@ tcp_xmit_timer(struct tcpcb *tp, int rtt, (tsecr == 0 || TSTMP_GEQ(tsecr, tp->t_badrexmt_time))) { /* - * We received a new ACk after a + * We received a new ACK after a * spurious timeout. Adapt retransmission * timer as described in rfc 4015. */ @@ -5509,8 +5765,9 @@ tcp_xmit_timer(struct tcpcb *tp, int rtt, tp->t_rttvar = max(tp->t_rttvar_prev, (rtt >> 1)); tp->t_rttvar = tp->t_rttvar << TCP_RTTVAR_SHIFT; - if (tp->t_rttbest > (tp->t_srtt + tp->t_rttvar)) + if (tp->t_rttbest > (tp->t_srtt + tp->t_rttvar)) { tp->t_rttbest = tp->t_srtt + tp->t_rttvar; + } goto compute_rto; } else { @@ -5541,10 +5798,11 @@ tcp_xmit_timer(struct tcpcb *tp, int rtt, * tcp_now, this extra adjustment is not needed. */ delta = (rtt << TCP_DELTA_SHIFT) - - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); + - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); - if ((tp->t_srtt += delta) <= 0) + if ((tp->t_srtt += delta) <= 0) { tp->t_srtt = 1; + } /* * We accumulate a smoothed rtt variance (actually, a @@ -5556,14 +5814,17 @@ tcp_xmit_timer(struct tcpcb *tp, int rtt, * (rttvar = rttvar*3/4 + |delta| / 4). This replaces * rfc793's wired-in beta. */ - if (delta < 0) + if (delta < 0) { delta = -delta; + } delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); - if ((tp->t_rttvar += delta) <= 0) + if ((tp->t_rttvar += delta) <= 0) { tp->t_rttvar = 1; - if (tp->t_rttbest == 0 || - tp->t_rttbest > (tp->t_srtt + tp->t_rttvar)) + } + if (tp->t_rttbest == 0 || + tp->t_rttbest > (tp->t_srtt + tp->t_rttvar)) { tp->t_rttbest = tp->t_srtt + tp->t_rttvar; + } } else { /* * No rtt measurement yet - use the unsmoothed rtt. @@ -5572,11 +5833,12 @@ tcp_xmit_timer(struct tcpcb *tp, int rtt, */ tp->t_srtt = rtt << TCP_RTT_SHIFT; tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); + tp->t_rttbest = tp->t_srtt + tp->t_rttvar; } compute_rto: nstat_route_rtt(tp->t_inpcb->inp_route.ro_rt, tp->t_srtt, - tp->t_rttvar); + tp->t_rttvar); /* * the retransmit should happen at rtt + 4 * rttvar. @@ -5590,8 +5852,8 @@ compute_rto: * the minimum feasible timer (which is 2 ticks). */ TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), - max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX, - TCP_ADD_REXMTSLOP(tp)); + max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX, + TCP_ADD_REXMTSLOP(tp)); /* * We received an ack for a packet that wasn't retransmitted; @@ -5601,23 +5863,39 @@ compute_rto: * and the return path might not be symmetrical). */ tp->t_softerror = 0; + + if (log_rtt) { + TCP_LOG_RTT_INFO(tp); + } + + TCP_LOG_RTT_CHANGE(tp, old_srtt, old_rttvar); } static inline unsigned int tcp_maxmtu(struct rtentry *rt) { unsigned int maxmtu; + int interface_mtu = 0; RT_LOCK_ASSERT_HELD(rt); - if (rt->rt_rmx.rmx_mtu == 0) - maxmtu = rt->rt_ifp->if_mtu; - else - maxmtu = MIN(rt->rt_rmx.rmx_mtu, rt->rt_ifp->if_mtu); + interface_mtu = rt->rt_ifp->if_mtu; + + if (rt_key(rt)->sa_family == AF_INET && + INTF_ADJUST_MTU_FOR_CLAT46(rt->rt_ifp)) { + interface_mtu = IN6_LINKMTU(rt->rt_ifp); + /* Further adjust the size for CLAT46 expansion */ + interface_mtu -= CLAT46_HDR_EXPANSION_OVERHD; + } + + if (rt->rt_rmx.rmx_mtu == 0) { + maxmtu = interface_mtu; + } else { + maxmtu = MIN(rt->rt_rmx.rmx_mtu, interface_mtu); + } - return (maxmtu); + return maxmtu; } -#if INET6 static inline unsigned int tcp_maxmtu6(struct rtentry *rt) { @@ -5625,20 +5903,23 @@ tcp_maxmtu6(struct rtentry *rt) struct nd_ifinfo *ndi = NULL; RT_LOCK_ASSERT_HELD(rt); - if ((ndi = ND_IFINFO(rt->rt_ifp)) != NULL && !ndi->initialized) + if ((ndi = ND_IFINFO(rt->rt_ifp)) != NULL && !ndi->initialized) { ndi = NULL; - if (ndi != NULL) + } + if (ndi != NULL) { lck_mtx_lock(&ndi->lock); - if (rt->rt_rmx.rmx_mtu == 0) + } + if (rt->rt_rmx.rmx_mtu == 0) { maxmtu = IN6_LINKMTU(rt->rt_ifp); - else + } else { maxmtu = MIN(rt->rt_rmx.rmx_mtu, IN6_LINKMTU(rt->rt_ifp)); - if (ndi != NULL) + } + if (ndi != NULL) { lck_mtx_unlock(&ndi->lock); + } - return (maxmtu); + return maxmtu; } -#endif unsigned int get_maxmtu(struct rtentry *rt) @@ -5657,7 +5938,7 @@ get_maxmtu(struct rtentry *rt) RT_UNLOCK(rt); - return (maxmtu); + return maxmtu; } /* @@ -5694,41 +5975,35 @@ tcp_mss(struct tcpcb *tp, int offer, unsigned int input_ifscope) u_int32_t bufsize; struct inpcb *inp; struct socket *so; - struct rmxp_tao *taop; int origoffer = offer; u_int32_t sb_max_corrected; int isnetlocal = 0; -#if INET6 int isipv6; int min_protoh; -#endif inp = tp->t_inpcb; -#if INET6 + + so = inp->inp_socket; + /* + * Nothing left to send after the socket is defunct or TCP is in the closed state + */ + if ((so->so_state & SS_DEFUNCT) || tp->t_state == TCPS_CLOSED) { + return; + } + isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; - min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) - : sizeof (struct tcpiphdr); -#else -#define min_protoh (sizeof (struct tcpiphdr)) -#endif + min_protoh = isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + : sizeof(struct tcpiphdr); -#if INET6 if (isipv6) { rt = tcp_rtlookup6(inp, input_ifscope); - } - else -#endif /* INET6 */ - { + } else { rt = tcp_rtlookup(inp, input_ifscope); } isnetlocal = (tp->t_flags & TF_LOCAL); if (rt == NULL) { - tp->t_maxopd = tp->t_maxseg = -#if INET6 - isipv6 ? tcp_v6mssdflt : -#endif /* INET6 */ - tcp_mssdflt; + tp->t_maxopd = tp->t_maxseg = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; return; } ifp = rt->rt_ifp; @@ -5743,26 +6018,20 @@ tcp_mss(struct tcpcb *tp, int offer, unsigned int input_ifscope) ifp->if_baudrate > 9600 && ifp->if_baudrate <= 128000) { tp->t_flags |= TF_SLOWLINK; } - so = inp->inp_socket; - taop = rmx_taop(rt->rt_rmx); /* - * Offer == -1 means that we didn't receive SYN yet, - * use cached value in that case; + * Offer == -1 means that we didn't receive SYN yet. Use 0 then. */ - if (offer == -1) - offer = taop->tao_mssopt; + if (offer == -1) { + offer = rt->rt_rmx.rmx_filler[0]; + } /* * Offer == 0 means that there was no MSS on the SYN segment, * in this case we use tcp_mssdflt. */ - if (offer == 0) - offer = -#if INET6 - isipv6 ? tcp_v6mssdflt : -#endif /* INET6 */ - tcp_mssdflt; - else { + if (offer == 0) { + offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; + } else { /* * Prevent DoS attack with too small MSS. Round up * to at least minmss. @@ -5776,7 +6045,7 @@ tcp_mss(struct tcpcb *tp, int offer, unsigned int input_ifscope) */ offer = max(offer, 64); } - taop->tao_mssopt = offer; + rt->rt_rmx.rmx_filler[0] = offer; /* * While we're here, check if there's an initial rtt @@ -5789,11 +6058,7 @@ tcp_mss(struct tcpcb *tp, int offer, unsigned int input_ifscope) tp->t_rttmin = isnetlocal ? tcp_TCPTV_MIN : TCPTV_REXMTMIN; } -#if INET6 mss = (isipv6 ? tcp_maxmtu6(rt) : tcp_maxmtu(rt)); -#else - mss = tcp_maxmtu(rt); -#endif #if NECP // At this point, the mss is just the MTU. Adjust if necessary. @@ -5803,14 +6068,13 @@ tcp_mss(struct tcpcb *tp, int offer, unsigned int input_ifscope) mss -= min_protoh; if (rt->rt_rmx.rmx_mtu == 0) { -#if INET6 if (isipv6) { - if (!isnetlocal) + if (!isnetlocal) { mss = min(mss, tcp_v6mssdflt); - } else -#endif /* INET6 */ - if (!isnetlocal) + } + } else if (!isnetlocal) { mss = min(mss, tcp_mssdflt); + } } mss = min(mss, offer); @@ -5827,10 +6091,11 @@ tcp_mss(struct tcpcb *tp, int offer, unsigned int input_ifscope) * origoffer==-1 indicates, that no segments were received yet. * In this case we just guess. */ - if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && + if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP && (origoffer == -1 || - (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) + (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) { mss -= TCPOLEN_TSTAMP_APPA; + } #if MPTCP mss -= mptcp_adj_mss(tp, FALSE); @@ -5854,17 +6119,20 @@ tcp_mss(struct tcpcb *tp, int offer, unsigned int input_ifscope) bufsize = rt->rt_rmx.rmx_sendpipe; if (bufsize < so->so_snd.sb_hiwat) #endif - bufsize = so->so_snd.sb_hiwat; - if (bufsize < mss) + bufsize = so->so_snd.sb_hiwat; + if (bufsize < mss) { mss = bufsize; - else { + } else { bufsize = (((bufsize + (u_int64_t)mss - 1) / (u_int64_t)mss) * (u_int64_t)mss); - if (bufsize > sb_max_corrected) + if (bufsize > sb_max_corrected) { bufsize = sb_max_corrected; + } (void)sbreserve(&so->so_snd, bufsize); } tp->t_maxseg = mss; + ASSERT(tp->t_maxseg); + /* * Update MSS using recommendation from link status report. This is * temporary @@ -5875,11 +6143,12 @@ tcp_mss(struct tcpcb *tp, int offer, unsigned int input_ifscope) bufsize = rt->rt_rmx.rmx_recvpipe; if (bufsize < so->so_rcv.sb_hiwat) #endif - bufsize = so->so_rcv.sb_hiwat; + bufsize = so->so_rcv.sb_hiwat; if (bufsize > mss) { bufsize = (((bufsize + (u_int64_t)mss - 1) / (u_int64_t)mss) * (u_int64_t)mss); - if (bufsize > sb_max_corrected) + if (bufsize > sb_max_corrected) { bufsize = sb_max_corrected; + } (void)sbreserve(&so->so_rcv, bufsize); } @@ -5902,8 +6171,9 @@ tcp_mss(struct tcpcb *tp, int offer, unsigned int input_ifscope) * Set the slow-start flight size depending on whether this * is a local network or not. */ - if (CC_ALGO(tp)->cwnd_init != NULL) + if (CC_ALGO(tp)->cwnd_init != NULL) { CC_ALGO(tp)->cwnd_init(tp); + } tcp_ccdbg_trace(tp, NULL, TCP_CC_CWND_INIT); @@ -5919,31 +6189,20 @@ tcp_mssopt(struct tcpcb *tp) { struct rtentry *rt; int mss; -#if INET6 int isipv6; int min_protoh; -#endif -#if INET6 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; - min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) - : sizeof (struct tcpiphdr); -#else -#define min_protoh (sizeof (struct tcpiphdr)) -#endif + min_protoh = isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + : sizeof(struct tcpiphdr); -#if INET6 - if (isipv6) + if (isipv6) { rt = tcp_rtlookup6(tp->t_inpcb, IFSCOPE_NONE); - else -#endif /* INET6 */ - rt = tcp_rtlookup(tp->t_inpcb, IFSCOPE_NONE); + } else { + rt = tcp_rtlookup(tp->t_inpcb, IFSCOPE_NONE); + } if (rt == NULL) { - return ( -#if INET6 - isipv6 ? tcp_v6mssdflt : -#endif /* INET6 */ - tcp_mssdflt); + return isipv6 ? tcp_v6mssdflt : tcp_mssdflt; } /* * Slower link window correction: @@ -5956,11 +6215,7 @@ tcp_mssopt(struct tcpcb *tp) tp->t_flags |= TF_SLOWLINK; } -#if INET6 mss = (isipv6 ? tcp_maxmtu6(rt) : tcp_maxmtu(rt)); -#else - mss = tcp_maxmtu(rt); -#endif /* Route locked during lookup above */ RT_UNLOCK(rt); @@ -5969,7 +6224,7 @@ tcp_mssopt(struct tcpcb *tp) mss = necp_socket_get_effective_mtu(tp->t_inpcb, mss); #endif /* NECP */ - return (mss - min_protoh); + return mss - min_protoh; } /* @@ -5981,32 +6236,33 @@ tcp_mssopt(struct tcpcb *tp) static void tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) { - tcp_seq onxt = tp->snd_nxt; - u_int32_t ocwnd = tp->snd_cwnd; - tp->t_timer[TCPT_REXMT] = 0; - tp->t_timer[TCPT_PTO] = 0; - tp->t_rtttime = 0; - tp->snd_nxt = th->th_ack; - /* - * Set snd_cwnd to one segment beyond acknowledged offset - * (tp->snd_una has not yet been updated when this function - * is called) - */ - tp->snd_cwnd = tp->t_maxseg + BYTES_ACKED(th, tp); - tp->t_flags |= TF_ACKNOW; - (void) tcp_output(tp); - tp->snd_cwnd = ocwnd; - if (SEQ_GT(onxt, tp->snd_nxt)) - tp->snd_nxt = onxt; - /* - * Partial window deflation. Relies on fact that tp->snd_una - * not updated yet. - */ - if (tp->snd_cwnd > BYTES_ACKED(th, tp)) - tp->snd_cwnd -= BYTES_ACKED(th, tp); - else - tp->snd_cwnd = 0; - tp->snd_cwnd += tp->t_maxseg; + tcp_seq onxt = tp->snd_nxt; + u_int32_t ocwnd = tp->snd_cwnd; + tp->t_timer[TCPT_REXMT] = 0; + tp->t_timer[TCPT_PTO] = 0; + tp->t_rtttime = 0; + tp->snd_nxt = th->th_ack; + /* + * Set snd_cwnd to one segment beyond acknowledged offset + * (tp->snd_una has not yet been updated when this function + * is called) + */ + tp->snd_cwnd = tp->t_maxseg + BYTES_ACKED(th, tp); + (void) tcp_output(tp); + tp->snd_cwnd = ocwnd; + if (SEQ_GT(onxt, tp->snd_nxt)) { + tp->snd_nxt = onxt; + } + /* + * Partial window deflation. Relies on fact that tp->snd_una + * not updated yet. + */ + if (tp->snd_cwnd > BYTES_ACKED(th, tp)) { + tp->snd_cwnd -= BYTES_ACKED(th, tp); + } else { + tp->snd_cwnd = 0; + } + tp->snd_cwnd += tp->t_maxseg; } /* @@ -6030,11 +6286,13 @@ tcp_dropdropablreq(struct socket *head) struct inpcb *inp = NULL; struct tcpcb *tp; - if ((head->so_options & SO_ACCEPTCONN) == 0) - return (0); + if ((head->so_options & SO_ACCEPTCONN) == 0) { + return 0; + } - if (TAILQ_EMPTY(&head->so_incomp)) - return (0); + if (TAILQ_EMPTY(&head->so_incomp)) { + return 0; + } so_acquire_accept_list(head, NULL); socket_unlock(head, 0); @@ -6050,8 +6308,8 @@ tcp_dropdropablreq(struct socket *head) tp = intotcpcb(inp); if (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)) { /* * The listen socket is already locked but we * can lock this socket here without lock ordering @@ -6077,15 +6335,17 @@ tcp_dropdropablreq(struct socket *head) } qlen = head->so_incqlen; - if (rnd == 0) + if (rnd == 0) { rnd = RandomULong(); + } if (++cur_cnt > qlen || old_cnt > qlen) { rnd = (314159 * rnd + 66329) & 0xffff; j = ((qlen + 1) * rnd) >> 16; - while (j-- && so) + while (j-- && so) { so = TAILQ_NEXT(so, so_list); + } } /* Find a connection that is not already closing (or being served) */ while (so) { @@ -6105,7 +6365,7 @@ tcp_dropdropablreq(struct socket *head) so->so_usecount++; if ((so->so_usecount == 2) && (so->so_state & SS_INCOMP) && - !(so->so_flags & SOF_INCOMP_INPROGRESS)) { + !(so->so_flags & SOF_INCOMP_INPROGRESS)) { break; } else { /* @@ -6130,7 +6390,7 @@ tcp_dropdropablreq(struct socket *head) if (so == NULL) { socket_lock(head, 0); so_release_accept_list(head); - return (0); + return 0; } /* Makes sure socket is still in the right state to be discarded */ @@ -6139,7 +6399,7 @@ tcp_dropdropablreq(struct socket *head) socket_unlock(so, 1); socket_lock(head, 0); so_release_accept_list(head); - return (0); + return 0; } found_victim: @@ -6148,7 +6408,7 @@ found_victim: socket_unlock(so, 1); socket_lock(head, 0); so_release_accept_list(head); - return (0); + return 0; } socket_lock(head, 0); @@ -6213,7 +6473,7 @@ found_victim: tcpstat.tcps_drops++; socket_lock(head, 0); - return(1); + return 1; } /* Set background congestion control on a socket */ @@ -6227,10 +6487,11 @@ tcp_set_background_cc(struct socket *so) void tcp_set_foreground_cc(struct socket *so) { - if (tcp_use_newreno) + if (tcp_use_newreno) { tcp_set_new_cc(so, TCP_CC_ALGO_NEWRENO_INDEX); - else + } else { tcp_set_new_cc(so, TCP_CC_ALGO_CUBIC_INDEX); + } } static void @@ -6240,17 +6501,18 @@ tcp_set_new_cc(struct socket *so, uint16_t cc_index) struct tcpcb *tp = intotcpcb(inp); u_char old_cc_index = 0; if (tp->tcp_cc_index != cc_index) { - old_cc_index = tp->tcp_cc_index; - if (CC_ALGO(tp)->cleanup != NULL) + if (CC_ALGO(tp)->cleanup != NULL) { CC_ALGO(tp)->cleanup(tp); + } tp->tcp_cc_index = cc_index; tcp_cc_allocate_state(tp); - if (CC_ALGO(tp)->switch_to != NULL) + if (CC_ALGO(tp)->switch_to != NULL) { CC_ALGO(tp)->switch_to(tp, old_cc_index); + } tcp_ccdbg_trace(tp, NULL, TCP_CC_CHANGE_ALGO); } @@ -6259,36 +6521,75 @@ tcp_set_new_cc(struct socket *so, uint16_t cc_index) void tcp_set_recv_bg(struct socket *so) { - if (!IS_TCP_RECV_BG(so)) + if (!IS_TCP_RECV_BG(so)) { so->so_flags1 |= SOF1_TRAFFIC_MGT_TCP_RECVBG; - - /* Unset Large Receive Offload on background sockets */ - so_set_lro(so, SO_TC_BK); + } } void tcp_clear_recv_bg(struct socket *so) { - if (IS_TCP_RECV_BG(so)) + if (IS_TCP_RECV_BG(so)) { so->so_flags1 &= ~(SOF1_TRAFFIC_MGT_TCP_RECVBG); + } +} + +void +inp_fc_throttle_tcp(struct inpcb *inp) +{ + struct tcpcb *tp = inp->inp_ppcb; + + if (!tcp_flow_control_response) { + return; + } /* - * Set/unset use of Large Receive Offload depending on - * the traffic class + * Back off the slow-start threshold and enter + * congestion avoidance phase */ - so_set_lro(so, so->so_traffic_class); + if (CC_ALGO(tp)->pre_fr != NULL) { + CC_ALGO(tp)->pre_fr(tp); + } } void inp_fc_unthrottle_tcp(struct inpcb *inp) { struct tcpcb *tp = inp->inp_ppcb; + + if (tcp_flow_control_response) { + if (CC_ALGO(tp)->post_fr != NULL) { + CC_ALGO(tp)->post_fr(tp, NULL); + } + + tp->t_bytes_acked = 0; + + /* + * Reset retransmit shift as we know that the reason + * for delay in sending a packet is due to flow + * control on the outgoing interface. There is no need + * to backoff retransmit timer. + */ + TCP_RESET_REXMT_STATE(tp); + + tp->t_flagsext &= ~TF_CWND_NONVALIDATED; + + /* + * Start the output stream again. Since we are + * not retransmitting data, do not reset the + * retransmit timer or rtt calculation. + */ + tcp_output(tp); + return; + } + /* * Back off the slow-start threshold and enter * congestion avoidance phase */ - if (CC_ALGO(tp)->pre_fr != NULL) + if (CC_ALGO(tp)->pre_fr != NULL) { CC_ALGO(tp)->pre_fr(tp); + } tp->snd_cwnd = tp->snd_ssthresh; tp->t_flagsext &= ~TF_CWND_NONVALIDATED; @@ -6321,57 +6622,24 @@ tcp_getstat SYSCTL_HANDLER_ARGS int error; struct tcpstat *stat; stat = &tcpstat; -#if !CONFIG_EMBEDDED - proc_t caller = PROC_NULL; - proc_t caller_parent = PROC_NULL; - char command_name[MAXCOMLEN + 1] = ""; - char parent_name[MAXCOMLEN + 1] = ""; +#if XNU_TARGET_OS_OSX struct tcpstat zero_stat; - if ((caller = proc_self()) != PROC_NULL) { - /* get process name */ - strlcpy(command_name, caller->p_comm, sizeof(command_name)); - - /* get parent process name if possible */ - if ((caller_parent = proc_find(caller->p_ppid)) != PROC_NULL) { - strlcpy(parent_name, caller_parent->p_comm, - sizeof(parent_name)); - proc_rele(caller_parent); - } - - if ((escape_str(command_name, strlen(command_name), - sizeof(command_name)) == 0) && - (escape_str(parent_name, strlen(parent_name), - sizeof(parent_name)) == 0)) { - kern_asl_msg(LOG_DEBUG, "messagetracer", - 5, - "com.apple.message.domain", - "com.apple.kernel.tcpstat", /* 1 */ - "com.apple.message.signature", - "tcpstat", /* 2 */ - "com.apple.message.signature2", command_name, /* 3 */ - "com.apple.message.signature3", parent_name, /* 4 */ - "com.apple.message.summarize", "YES", /* 5 */ - NULL); - } - } - if (caller != PROC_NULL) - proc_rele(caller); + if (tcp_disable_access_to_stats && !kauth_cred_issuser(kauth_cred_get())) { bzero(&zero_stat, sizeof(zero_stat)); stat = &zero_stat; } -#endif /* !CONFIG_EMBEDDED */ +#endif /* XNU_TARGET_OS_OSX */ if (req->oldptr == 0) { - req->oldlen= (size_t)sizeof(struct tcpstat); + req->oldlen = (size_t)sizeof(struct tcpstat); } - error = SYSCTL_OUT(req, stat, MIN(sizeof (tcpstat), req->oldlen)); - - return (error); + error = SYSCTL_OUT(req, stat, MIN(sizeof(tcpstat), req->oldlen)); + return error; } /* @@ -6387,11 +6655,8 @@ tcp_input_checksum(int af, struct mbuf *m, struct tcphdr *th, int off, int tlen) struct ip *ip = mtod(m, struct ip *); struct ipovly *ipov = (struct ipovly *)ip; - if (m->m_pkthdr.pkt_flags & PKTF_SW_LRO_DID_CSUM) - return (0); - /* ip_stripoptions() must have been called before we get here */ - ASSERT((ip->ip_hl << 2) == sizeof (*ip)); + ASSERT((ip->ip_hl << 2) == sizeof(*ip)); if ((hwcksum_rx || (ifp->if_flags & IFF_LOOPBACK) || (m->m_pkthdr.pkt_flags & PKTF_LOOP)) && @@ -6423,7 +6688,7 @@ tcp_input_checksum(int af, struct mbuf *m, struct tcphdr *th, int off, int tlen) uint32_t swbytes = (uint32_t)trailer; if (start < off) { - ip->ip_len += sizeof (*ip); + ip->ip_len += sizeof(*ip); #if BYTE_ORDER != BIG_ENDIAN HTONS(ip->ip_len); HTONS(ip->ip_off); @@ -6432,23 +6697,26 @@ tcp_input_checksum(int af, struct mbuf *m, struct tcphdr *th, int off, int tlen) /* callee folds in sum */ sum = m_adj_sum16(m, start, off, tlen, sum); - if (off > start) + if (off > start) { swbytes += (off - start); - else + } else { swbytes += (start - off); + } if (start < off) { #if BYTE_ORDER != BIG_ENDIAN NTOHS(ip->ip_off); NTOHS(ip->ip_len); #endif /* BYTE_ORDER != BIG_ENDIAN */ - ip->ip_len -= sizeof (*ip); + ip->ip_len -= sizeof(*ip); } - if (swbytes != 0) + if (swbytes != 0) { tcp_in_cksum_stats(swbytes); - if (trailer != 0) + } + if (trailer != 0) { m_adj(m, -trailer); + } } /* callee folds in sum */ @@ -6462,29 +6730,25 @@ tcp_input_checksum(int af, struct mbuf *m, struct tcphdr *th, int off, int tlen) int len; char b[9]; - bcopy(ipov->ih_x1, b, sizeof (ipov->ih_x1)); - bzero(ipov->ih_x1, sizeof (ipov->ih_x1)); + bcopy(ipov->ih_x1, b, sizeof(ipov->ih_x1)); + bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); ip_sum = ipov->ih_len; ipov->ih_len = (u_short)tlen; #if BYTE_ORDER != BIG_ENDIAN HTONS(ipov->ih_len); #endif - len = sizeof (struct ip) + tlen; + len = sizeof(struct ip) + tlen; th->th_sum = in_cksum(m, len); - bcopy(b, ipov->ih_x1, sizeof (ipov->ih_x1)); + bcopy(b, ipov->ih_x1, sizeof(ipov->ih_x1)); ipov->ih_len = ip_sum; tcp_in_cksum_stats(len); } break; } -#if INET6 case AF_INET6: { struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); - if (m->m_pkthdr.pkt_flags & PKTF_SW_LRO_DID_CSUM) - return (0); - if ((hwcksum_rx || (ifp->if_flags & IFF_LOOPBACK) || (m->m_pkthdr.pkt_flags & PKTF_LOOP)) && (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)) { @@ -6510,7 +6774,7 @@ tcp_input_checksum(int af, struct mbuf *m, struct tcphdr *th, int off, int tlen) if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) { s = ip6->ip6_src.s6_addr16[1]; - ip6->ip6_src.s6_addr16[1] = 0 ; + ip6->ip6_src.s6_addr16[1] = 0; } if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) { d = ip6->ip6_dst.s6_addr16[1]; @@ -6520,25 +6784,30 @@ tcp_input_checksum(int af, struct mbuf *m, struct tcphdr *th, int off, int tlen) /* callee folds in sum */ sum = m_adj_sum16(m, start, off, tlen, sum); - if (off > start) + if (off > start) { swbytes += (off - start); - else + } else { swbytes += (start - off); + } - if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) + if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) { ip6->ip6_src.s6_addr16[1] = s; - if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) + } + if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) { ip6->ip6_dst.s6_addr16[1] = d; + } - if (swbytes != 0) + if (swbytes != 0) { tcp_in6_cksum_stats(swbytes); - if (trailer != 0) + } + if (trailer != 0) { m_adj(m, -trailer); + } } th->th_sum = in6_pseudo( - &ip6->ip6_src, &ip6->ip6_dst, - sum + htonl(tlen + IPPROTO_TCP)); + &ip6->ip6_src, &ip6->ip6_dst, + sum + htonl(tlen + IPPROTO_TCP)); } th->th_sum ^= 0xffff; } else { @@ -6547,7 +6816,6 @@ tcp_input_checksum(int af, struct mbuf *m, struct tcphdr *th, int off, int tlen) } break; } -#endif /* INET6 */ default: VERIFY(0); /* NOTREACHED */ @@ -6556,12 +6824,13 @@ tcp_input_checksum(int af, struct mbuf *m, struct tcphdr *th, int off, int tlen) if (th->th_sum != 0) { tcpstat.tcps_rcvbadsum++; IF_TCP_STATINC(ifp, badformat); - return (-1); + return -1; } - return (0); + return 0; } + SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, tcp_getstat, "S,tcpstat", "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); @@ -6574,8 +6843,9 @@ sysctl_rexmtthresh SYSCTL_HANDLER_ARGS int error, val = tcprexmtthresh; error = sysctl_handle_int(oidp, &val, 0, req); - if (error || !req->newptr) - return (error); + if (error || !req->newptr) { + return error; + } /* * Constrain the number of duplicate ACKs @@ -6583,14 +6853,15 @@ sysctl_rexmtthresh SYSCTL_HANDLER_ARGS * to either 2 or 3 */ - if (val < 2 || val > 3) - return (EINVAL); + if (val < 2 || val > 3) { + return EINVAL; + } - tcprexmtthresh = val; + tcprexmtthresh = val; - return (0); + return 0; } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmt_thresh, CTLTYPE_INT | CTLFLAG_RW | - CTLFLAG_LOCKED, &tcprexmtthresh, 0, &sysctl_rexmtthresh, "I", - "Duplicate ACK Threshold for Fast Retransmit"); + CTLFLAG_LOCKED, &tcprexmtthresh, 0, &sysctl_rexmtthresh, "I", + "Duplicate ACK Threshold for Fast Retransmit");