X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/3a60a9f5b85abb8c2cf24e1926c5c7b3f608a5e2..ebb1b9f42b62218f29061826217bb0f71cd375a6:/bsd/netinet/tcp_input.c diff --git a/bsd/netinet/tcp_input.c b/bsd/netinet/tcp_input.c index e85e6aed5..6f06b2b14 100644 --- a/bsd/netinet/tcp_input.c +++ b/bsd/netinet/tcp_input.c @@ -1,23 +1,29 @@ /* - * Copyright (c) 2000 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2011 Apple Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * - * The contents of this file constitute Original Code as defined in and - * are subject to the Apple Public Source License Version 1.1 (the - * "License"). You may not use this file except in compliance with the - * License. Please obtain a copy of the License at - * http://www.apple.com/publicsource and read it before using this file. + * This file contains Original Code and/or Modifications of Original Code + * as defined in and that are subject to the Apple Public Source License + * Version 2.0 (the 'License'). You may not use this file except in + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. * - * This Original Code and all software distributed under the License are - * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. + * + * The Original Code and all software distributed under the License are + * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the - * License for the specific language governing rights and limitations - * under the License. + * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. + * Please see the License for the specific language governing rights and + * limitations under the License. * - * @APPLE_LICENSE_HEADER_END@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 @@ -54,7 +60,12 @@ * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.16 2001/08/22 00:59:12 silby Exp $ */ - +/* + * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce + * support for mandatory and extensible security protections. This notice + * is included in support of clause 2.2 (b) of the Apple Public License, + * Version 2.0. + */ #include #include @@ -70,9 +81,12 @@ #include /* before tcp_seq.h, for tcp_random18() */ +#include + #include #include #include +#include #include #include @@ -82,6 +96,7 @@ #include /* for ICMP_BANDLIM */ #include #include +#include #if INET6 #include #include @@ -94,6 +109,8 @@ #include #include #include +#include +#include #if INET6 #include #endif @@ -112,91 +129,141 @@ struct tcphdr tcp_savetcp; #include #endif /*IPSEC*/ -#include +#if CONFIG_MACF_NET || CONFIG_MACF_SOCKET +#include +#endif /* CONFIG_MACF_NET || CONFIG_MACF_SOCKET */ -#ifndef __APPLE__ -MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); -#endif +#include #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)) -static int tcprexmtthresh = 3; +static int tcprexmtthresh = 2; tcp_cc tcp_ccgen; -extern int apple_hwcksum_rx; #if IPSEC extern int ipsec_bypass; -extern lck_mtx_t *sadb_mutex; #endif struct tcpstat tcpstat; -SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD, - &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); static int log_in_vain = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW | CTLFLAG_LOCKED, &log_in_vain, 0, "Log all incoming TCP connections"); static int blackhole = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW | CTLFLAG_LOCKED, &blackhole, 0, "Do not send RST when dropping refused connections"); -int tcp_delack_enabled = 1; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, +int tcp_delack_enabled = 3; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_delack_enabled, 0, "Delay ACK to try and piggyback it onto a data packet"); int tcp_lq_overflow = 1; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_lq_overflow, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_lq_overflow, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_lq_overflow, 0, "Listen Queue Overflow"); +int tcp_recv_bg = 0; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbg, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_recv_bg, 0, + "Receive background"); + #if TCP_DROP_SYNFIN static int drop_synfin = 1; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW | CTLFLAG_LOCKED, &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); #endif -SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, +SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW|CTLFLAG_LOCKED, 0, "TCP Segment Reassembly Queue"); __private_extern__ int tcp_reass_maxseg = 0; -SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_reass_maxseg, 0, "Global maximum number of TCP Segments in Reassembly Queue"); __private_extern__ int tcp_reass_qsize = 0; -SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, +SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD | CTLFLAG_LOCKED, &tcp_reass_qsize, 0, "Global number of TCP Segments currently in Reassembly Queue"); static int tcp_reass_overflows = 0; -SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, +SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD | CTLFLAG_LOCKED, &tcp_reass_overflows, 0, "Global number of TCP Segment Reassembly Queue Overflows"); __private_extern__ int slowlink_wsize = 8192; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowlink_wsize, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowlink_wsize, CTLFLAG_RW | CTLFLAG_LOCKED, &slowlink_wsize, 0, "Maximum advertised window size for slowlink"); +int maxseg_unacked = 8; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, maxseg_unacked, CTLFLAG_RW | CTLFLAG_LOCKED, + &maxseg_unacked, 0, "Maximum number of outstanding segments left unacked"); + +int tcp_do_rfc3465 = 1; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_do_rfc3465, 0, ""); + +int tcp_do_rfc3465_lim2 = 1; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465_lim2, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_do_rfc3465_lim2, 0, "Appropriate bytes counting w/ L=2*SMSS"); + +int rtt_samples_per_slot = 20; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, rtt_samples_per_slot, CTLFLAG_RW | CTLFLAG_LOCKED, + &rtt_samples_per_slot, 0, "Number of RTT samples stored for rtt history"); + +int tcp_allowed_iaj = ALLOWED_IAJ; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, recv_allowed_iaj, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_allowed_iaj, 0, "Allowed inter-packet arrival jiter"); + +int tcp_acc_iaj_high_thresh = ACC_IAJ_HIGH_THRESH; +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"); + +#if CONFIG_IFEF_NOWINDOWSCALE +int tcp_obey_ifef_nowindowscale = 0; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, obey_ifef_nowindowscale, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_obey_ifef_nowindowscale, 0, ""); +#endif + +extern int tcp_TCPTV_MIN; +extern int tcp_acc_iaj_high; +extern int tcp_acc_iaj_react_limit; +extern struct zone *tcp_reass_zone; + + +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 */ -u_long tcp_now; struct inpcbhead tcb; #define tcb6 tcb /* for KAME src sync over BSD*'s */ struct inpcbinfo tcbinfo; -static void tcp_dooptions(struct tcpcb *, - u_char *, int, struct tcphdr *, struct tcpopt *); +static void tcp_dooptions(struct tcpcb *, u_char *, int, struct tcphdr *, + 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 *); static void tcp_xmit_timer(struct tcpcb *, int); -static int tcp_newreno __P((struct tcpcb *, struct tcphdr *)); +static inline unsigned int tcp_maxmtu(struct rtentry *); +static inline int tcp_stretch_ack_enable(struct tcpcb *tp); + +#if TRAFFIC_MGT +static inline void update_iaj_state(struct tcpcb *tp, uint32_t tlen, int reset_size); +void compute_iaj(struct tcpcb *tp); +static inline void clear_iaj_state(struct tcpcb *tp); +#endif /* TRAFFIC_MGT */ + +#if INET6 +static inline unsigned int tcp_maxmtu6(struct rtentry *); +#endif /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ #if INET6 @@ -211,42 +278,160 @@ do { \ #define ND6_HINT(tp) #endif -extern u_long *delack_bitmask; +extern void add_to_time_wait(struct tcpcb *, uint32_t delay); +extern void postevent(struct socket *, struct sockbuf *, int); -extern void ipfwsyslog( int level, char *format,...); +extern void ipfwsyslog( int level, const char *format,...); extern int ChkAddressOK( __uint32_t dstaddr, __uint32_t srcaddr ); extern int fw_verbose; +__private_extern__ int tcp_sockthreshold; +__private_extern__ int tcp_win_scale; +#if IPFIREWALL #define log_in_vain_log( a ) { \ if ( (log_in_vain == 3 ) && (fw_verbose == 2)) { /* Apple logging, log to ipfw.log */ \ ipfwsyslog a ; \ } \ else log a ; \ } +#else +#define log_in_vain_log( a ) { log a; } +#endif -/* - * Indicate whether this ack should be delayed. - * We can delay the ack if: - * - delayed acks are enabled (set to 1) and - * - our last ack wasn't a 0-sized window. We never want to delay - * the ack that opens up a 0-sized window. - * - delayed acks are enabled (set to 2, "more compatible") and - * - our last ack wasn't a 0-sized window. - * - if the peer hasn't sent us a TH_PUSH data packet (this solves 3649245) - * - the peer hasn't sent us a TH_PUSH data packet, if he did, take this as a clue that we - * need to ACK with no delay. This helps higher level protocols who won't send - * us more data even if the window is open because their last "segment" hasn't been ACKed - * - * +int tcp_rcvunackwin = TCPTV_UNACKWIN; +int tcp_maxrcvidle = TCPTV_MAXRCVIDLE; +int tcp_rcvsspktcnt = TCP_RCV_SS_PKTCOUNT; + +#define DELAY_ACK(tp, th) (CC_ALGO(tp)->delay_ack != NULL && CC_ALGO(tp)->delay_ack(tp, th)) + +static int tcp_dropdropablreq(struct socket *head); +static void tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th); + +static void update_base_rtt(struct tcpcb *tp, uint32_t rtt); +uint32_t get_base_rtt(struct tcpcb *tp); +void tcp_set_background_cc(struct socket *so); +void tcp_set_foreground_cc(struct socket *so); +static void tcp_set_new_cc(struct socket *so, uint16_t cc_index); + +#if TRAFFIC_MGT +void +reset_acc_iaj(struct tcpcb *tp) +{ + tp->acc_iaj = 0; + tp->iaj_rwintop = 0; + clear_iaj_state(tp); +} + +static inline void +update_iaj_state(struct tcpcb *tp, uint32_t size, int rst_size) +{ + 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; + tp->iaj_small_pkt = 0; + } +} + +static inline void +clear_iaj_state(struct tcpcb *tp) +{ + tp->iaj_rcv_ts = 0; +} + +/* For every 32 bit unsigned integer(v), this function will find the + * largest integer n such that (n*n <= v). This takes at most 16 iterations + * irrespective of the value of v and does not involve multiplications. */ -#define DELAY_ACK(tp) \ - (((tcp_delack_enabled == 1) && ((tp->t_flags & TF_RXWIN0SENT) == 0)) || \ - (((tcp_delack_enabled == 2) && (tp->t_flags & TF_RXWIN0SENT) == 0) && \ - ((thflags & TH_PUSH) == 0) && ((tp->t_flags & TF_DELACK) == 0))) +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) { + for (g = 0; g <= 10; ++g) { + if (sqrt_cache[g] > val) { + g--; + break; + } else if (sqrt_cache[g] == val) { + break; + } + } + } else { + do { + temp = (((g << 1) + b) << (bshft--)); + if (val >= temp) { + g += b; + val -= temp; + } + b >>= 1; + } while ( b > 0 && val > 0); + } + return(g); +} + +void +compute_iaj(struct tcpcb *tp) +{ + /* When accumulated IAJ reaches MAX_ACC_IAJ in milliseconds, throttle the + * receive window to a minimum of MIN_IAJ_WIN packets + */ +#define MAX_ACC_IAJ (tcp_acc_iaj_high_thresh + tcp_acc_iaj_react_limit) + + uint32_t allowed_iaj, acc_iaj = 0; + uint32_t cur_iaj = tcp_now - tp->iaj_rcv_ts; + + uint32_t mean, temp; + int32_t cur_iaj_dev; + cur_iaj_dev = (cur_iaj - tp->avg_iaj); + + /* Allow a jitter of "allowed_iaj" milliseconds. Some connections may have a + * constant jitter more than that. We detect this by using + * standard deviation. + */ + allowed_iaj = tp->avg_iaj + tp->std_dev_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 connection because + * we do not have a good starting point for allowed_iaj. IAJ_IGNORE_PKTCNT + * is used to quietly gloss over the first few packets. + */ + if (tp->iaj_pktcnt > IAJ_IGNORE_PKTCNT) { + if ( cur_iaj <= allowed_iaj ) { + if (tp->acc_iaj >= 2) + acc_iaj = tp->acc_iaj - 2; + else + acc_iaj = 0; + } else { + acc_iaj = tp->acc_iaj + (cur_iaj - allowed_iaj); + } + + if (acc_iaj > MAX_ACC_IAJ) + acc_iaj = MAX_ACC_IAJ; + tp->acc_iaj = acc_iaj; + } + + /* Compute weighted average where the history has a weight of + * 15 out of 16 and the current value has a weight of 1 out of 16. + * This will make the short-term measurements have more weight. + */ + tp->avg_iaj = (((tp->avg_iaj << 4) - tp->avg_iaj) + cur_iaj) >> 4; + /* 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 << 4) - temp) + (cur_iaj_dev * cur_iaj_dev)) >> 4; + + tp->std_dev_iaj = isqrt(mean); -static int tcpdropdropablreq(struct socket *head); + DTRACE_TCP3(iaj, struct tcpcb *, tp, uint32_t, cur_iaj, uint32_t, allowed_iaj); + return; +} +#endif /* TRAFFIC_MGT */ static int tcp_reass(tp, th, tlenp, m) @@ -258,7 +443,7 @@ tcp_reass(tp, th, tlenp, m) struct tseg_qent *q; struct tseg_qent *p = NULL; struct tseg_qent *nq; - struct tseg_qent *te; + struct tseg_qent *te = NULL; struct socket *so = tp->t_inpcb->inp_socket; int flags; int dowakeup = 0; @@ -267,8 +452,29 @@ tcp_reass(tp, th, tlenp, m) * Call with th==0 after become established to * force pre-ESTABLISHED data up to user socket. */ - if (th == 0) + if (th == NULL) goto present; + + /* If the reassembly queue already has entries or if we are going to add + * a new one, then the connection has reached a loss state. + * Reset the stretch-ack algorithm at this point. + */ + if ((tp->t_flags & TF_STRETCHACK) != 0) + tcp_reset_stretch_ack(tp); + + /* When the connection reaches a loss state, we need to send more acks + * for a period of time so that the sender's congestion window will + * open. Wait until we see some packets on the connection before + * stretching acks again. + */ + tp->t_flagsext |= TF_RCVUNACK_WAITSS; + tp->rcv_waitforss = 0; + + +#if TRAFFIC_MGT + if (tp->acc_iaj > 0) + reset_acc_iaj(tp); +#endif /* TRAFFIC_MGT */ /* * Limit the number of segments in the reassembly queue to prevent @@ -282,12 +488,12 @@ tcp_reass(tp, th, tlenp, m) tcp_reass_overflows++; tcpstat.tcps_rcvmemdrop++; m_freem(m); + *tlenp = 0; return (0); } /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ - MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, - M_NOWAIT); + te = (struct tseg_qent *) zalloc_noblock(tcp_reass_zone); if (te == NULL) { tcpstat.tcps_rcvmemdrop++; m_freem(m); @@ -317,8 +523,14 @@ tcp_reass(tp, th, tlenp, m) if (i >= *tlenp) { tcpstat.tcps_rcvduppack++; tcpstat.tcps_rcvdupbyte += *tlenp; + if (nstat_collect) { + nstat_route_rx(tp->t_inpcb->inp_route.ro_rt, 1, *tlenp, NSTAT_RX_FLAG_DUPLICATE); + locked_add_64(&tp->t_inpcb->inp_stat->rxpackets, 1); + locked_add_64(&tp->t_inpcb->inp_stat->rxbytes, *tlenp); + tp->t_stat.rxduplicatebytes += *tlenp; + } m_freem(m); - FREE(te, M_TSEGQ); + zfree(tcp_reass_zone, te); tcp_reass_qsize--; /* * Try to present any queued data @@ -335,6 +547,12 @@ tcp_reass(tp, th, tlenp, m) } tcpstat.tcps_rcvoopack++; tcpstat.tcps_rcvoobyte += *tlenp; + if (nstat_collect) { + nstat_route_rx(tp->t_inpcb->inp_route.ro_rt, 1, *tlenp, NSTAT_RX_FLAG_OUT_OF_ORDER); + locked_add_64(&tp->t_inpcb->inp_stat->rxpackets, 1); + locked_add_64(&tp->t_inpcb->inp_stat->rxbytes, *tlenp); + tp->t_stat.rxoutoforderbytes += *tlenp; + } /* * While we overlap succeeding segments trim them or, @@ -354,7 +572,7 @@ tcp_reass(tp, th, tlenp, m) nq = LIST_NEXT(q, tqe_q); LIST_REMOVE(q, tqe_q); m_freem(q->tqe_m); - FREE(q, M_TSEGQ); + zfree(tcp_reass_zone, q); tcp_reass_qsize--; q = nq; } @@ -388,10 +606,11 @@ present: if (so->so_state & SS_CANTRCVMORE) m_freem(q->tqe_m); else { - if (sbappend(&so->so_rcv, q->tqe_m)) + so_recv_data_stat(so, q->tqe_m, 0); /* XXXX */ + if (sbappendstream(&so->so_rcv, q->tqe_m)) dowakeup = 1; } - FREE(q, M_TSEGQ); + zfree(tcp_reass_zone, q); tcp_reass_qsize--; q = nq; } while (q && q->tqe_th->th_seq == tp->rcv_nxt); @@ -421,6 +640,29 @@ present: } +/* + * Reduce congestion window. + */ +static void +tcp_reduce_congestion_window( + struct tcpcb *tp, struct tcphdr *th) +{ + /* + * 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, th); + ENTER_FASTRECOVERY(tp); + tp->snd_recover = tp->snd_max; + tp->t_timer[TCPT_REXMT] = 0; + tp->t_rtttime = 0; + tp->ecn_flags |= TE_SENDCWR; + tp->snd_cwnd = tp->snd_ssthresh + + tp->t_maxseg * tcprexmtthresh; +} + /* * TCP input routine, follows pages 65-76 of the @@ -428,10 +670,9 @@ present: */ #if INET6 int -tcp6_input(mp, offp) - struct mbuf **mp; - int *offp; +tcp6_input(struct mbuf **mp, int *offp, int proto) { +#pragma unused(proto) register struct mbuf *m = *mp; struct in6_ifaddr *ia6; @@ -442,20 +683,71 @@ tcp6_input(mp, offp) * better place to put this in? */ ia6 = ip6_getdstifaddr(m); - if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { - struct ip6_hdr *ip6; + if (ia6 != NULL) { + IFA_LOCK_SPIN(&ia6->ia_ifa); + if (ia6->ia6_flags & IN6_IFF_ANYCAST) { + struct ip6_hdr *ip6; - ip6 = mtod(m, struct ip6_hdr *); - icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, + IFA_UNLOCK(&ia6->ia_ifa); + IFA_REMREF(&ia6->ia_ifa); + ip6 = mtod(m, struct ip6_hdr *); + icmp6_error(m, ICMP6_DST_UNREACH, + ICMP6_DST_UNREACH_ADDR, (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); - return IPPROTO_DONE; + return (IPPROTO_DONE); + } + IFA_UNLOCK(&ia6->ia_ifa); + IFA_REMREF(&ia6->ia_ifa); } tcp_input(m, *offp); - return IPPROTO_DONE; + return (IPPROTO_DONE); } #endif +/* A receiver will evaluate the flow of packets on a connection + * to see if it can reduce ack traffic. The receiver will start + * stretching acks if all of the following conditions are met: + * 1. tcp_delack_enabled is set to 3 + * 2. If the bytes received in the last 100ms is greater than a threshold + * defined by maxseg_unacked + * 3. If the connection has not been idle for tcp_maxrcvidle period. + * 4. If the connection has seen enough packets to let the slow-start + * finish after connection establishment or after some packet loss. + * + * The receiver will stop stretching acks if there is congestion/reordering + * as indicated by packets on reassembly queue or an ECN. If the delayed-ack + * timer fires while stretching acks, it means that the packet flow has gone + * below the threshold defined by maxseg_unacked and the receiver will stop + * stretching acks. The receiver gets no indication when slow-start is completed + * or when the connection reaches an idle state. That is why we use + * tcp_rcvsspktcnt to cover slow-start and tcp_maxrcvidle to identify idle + * state. + */ + static inline int + tcp_stretch_ack_enable(struct tcpcb *tp) { + if (tp->rcv_by_unackwin >= (maxseg_unacked * tp->t_maxseg) && + TSTMP_GT(tp->rcv_unackwin + tcp_maxrcvidle, tcp_now) && + (((tp->t_flagsext & TF_RCVUNACK_WAITSS) == 0) || + (tp->rcv_waitforss >= tcp_rcvsspktcnt))) { + return(1); + } + return(0); +} + +/* Reset the state related to stretch-ack algorithm. This will make + * the receiver generate an ack every other packet. The receiver + * will start re-evaluating the rate at which packets come to decide + * if it can benefit by lowering the ack traffic. + */ +void +tcp_reset_stretch_ack(struct tcpcb *tp) +{ + tp->t_flags &= ~(TF_STRETCHACK); + tp->rcv_by_unackwin = 0; + tp->rcv_unackwin = tcp_now + tcp_rcvunackwin; +} + void tcp_input(m, off0) struct mbuf *m; @@ -480,18 +772,41 @@ tcp_input(m, off0) int dropsocket = 0; int iss = 0; int nosock = 0; - u_long tiwin; + u_int32_t tiwin; struct tcpopt to; /* options in this segment */ - struct rmxp_tao *taop; /* pointer to our TAO cache entry */ - struct rmxp_tao tao_noncached; /* in case there's no cached entry */ struct sockaddr_in *next_hop = NULL; #if TCPDEBUG short ostate = 0; #endif struct m_tag *fwd_tag; + u_char ip_ecn = IPTOS_ECN_NOTECT; + unsigned int ifscope, nocell = 0; + uint8_t isconnected, isdisconnected; + + /* + * Record the interface where this segment arrived on; this does not + * affect normal data output (for non-detached TCP) as it provides a + * hint about which route and interface to use for sending in the + * absence of a PCB, when scoped routing (and thus source interface + * selection) are enabled. + */ + if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.rcvif != NULL) + ifscope = m->m_pkthdr.rcvif->if_index; + else + ifscope = IFSCOPE_NONE; + + /* Since this is an entry point for input processing of tcp packets, we + * can update the tcp clock here. + */ + calculate_tcp_clock(); /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ - fwd_tag = m_tag_locate(m, KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_IPFORWARD, NULL); + 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); @@ -522,12 +837,29 @@ tcp_input(m, off0) /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ ip6 = mtod(m, struct ip6_hdr *); tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; - if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { - tcpstat.tcps_rcvbadsum++; - goto dropnosock; - } th = (struct tcphdr *)((caddr_t)ip6 + off0); + if ((apple_hwcksum_rx != 0) && (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)) { + if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) + th->th_sum = m->m_pkthdr.csum_data; + else + th->th_sum = in6_cksum_phdr(&ip6->ip6_src, + &ip6->ip6_dst, htonl(sizeof(struct tcphdr)), + htonl(IPPROTO_TCP)); + + th->th_sum ^= 0xffff; + if (th->th_sum) { + tcpstat.tcps_rcvbadsum++; + goto dropnosock; + } + } + else { + if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { + tcpstat.tcps_rcvbadsum++; + 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); @@ -543,6 +875,11 @@ tcp_input(m, off0) /* XXX stat */ goto dropnosock; } + DTRACE_TCP5(receive, sruct mbuf *, m, struct inpcb *, NULL, + struct ip6_hdr *, ip6, struct tcpcb *, NULL, + struct tcphdr *, th); + + ip_ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; } else #endif /* INET6 */ { @@ -568,12 +905,15 @@ tcp_input(m, off0) th = (struct tcphdr *)((caddr_t)ip + off0); tlen = ip->ip_len; + 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); if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { - if (apple_hwcksum_rx && (m->m_pkthdr.csum_flags & CSUM_TCP_SUM16)) { + if (m->m_pkthdr.csum_flags & CSUM_TCP_SUM16) { u_short pseudo; char b[9]; *(uint32_t*)&b[0] = *(uint32_t*)&ipov->ih_x1[0]; @@ -582,7 +922,11 @@ tcp_input(m, off0) bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); ipov->ih_len = (u_short)tlen; + +#if BYTE_ORDER != BIG_ENDIAN HTONS(ipov->ih_len); +#endif + pseudo = in_cksum(m, sizeof (struct ip)); *(uint32_t*)&ipov->ih_x1[0] = *(uint32_t*)&b[0]; @@ -611,12 +955,18 @@ tcp_input(m, off0) len = sizeof (struct ip) + tlen; bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); ipov->ih_len = (u_short)tlen; + +#if BYTE_ORDER != BIG_ENDIAN HTONS(ipov->ih_len); +#endif + th->th_sum = in_cksum(m, len); *(uint32_t*)&ipov->ih_x1[0] = *(uint32_t*)&b[0]; *(uint32_t*)&ipov->ih_x1[4] = *(uint32_t*)&b[4]; *(uint8_t*)&ipov->ih_x1[8] = *(uint8_t*)&b[8]; + + tcp_in_cksum_stats(len); } if (th->th_sum) { tcpstat.tcps_rcvbadsum++; @@ -626,6 +976,7 @@ tcp_input(m, off0) /* Re-initialization for later version check */ ip->ip_v = IPVERSION; #endif + ip_ecn = (ip->ip_tos & IPTOS_ECN_MASK); } /* @@ -646,17 +997,17 @@ tcp_input(m, off0) th = (struct tcphdr *)((caddr_t)ip6 + off0); } else #endif /* INET6 */ - { - if (m->m_len < sizeof(struct ip) + off) { - if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { - tcpstat.tcps_rcvshort++; - return; + { + if (m->m_len < sizeof(struct ip) + off) { + if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { + tcpstat.tcps_rcvshort++; + return; + } + ip = mtod(m, struct ip *); + ipov = (struct ipovly *)ip; + th = (struct tcphdr *)((caddr_t)ip + off0); } - ip = mtod(m, struct ip *); - ipov = (struct ipovly *)ip; - th = (struct tcphdr *)((caddr_t)ip + off0); } - } optlen = off - sizeof (struct tcphdr); optp = (u_char *)(th + 1); /* @@ -667,11 +1018,11 @@ tcp_input(m, off0) * tcp_dooptions(), etc. */ if ((optlen == TCPOLEN_TSTAMP_APPA || - (optlen > TCPOLEN_TSTAMP_APPA && + (optlen > TCPOLEN_TSTAMP_APPA && optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && - *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && - (th->th_flags & TH_SYN) == 0) { - to.to_flag |= TOF_TS; + *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && + (th->th_flags & TH_SYN) == 0) { + to.to_flags |= TOF_TS; to.to_tsval = ntohl(*(u_int32_t *)(optp + 4)); to.to_tsecr = ntohl(*(u_int32_t *)(optp + 8)); optp = NULL; /* we've parsed the options */ @@ -685,7 +1036,7 @@ tcp_input(m, off0) * both the SYN and FIN bits set. This prevents e.g. nmap from * identifying the TCP/IP stack. * - * This is incompatible with RFC1644 extensions (T/TCP). + * This is a violation of the TCP specification. */ if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) goto dropnosock; @@ -694,13 +1045,16 @@ tcp_input(m, off0) /* * 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); +#endif /* - * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, + * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, * until after ip6_savecontrol() is called and before other functions * which don't want those proto headers. * Because ip6_savecontrol() is going to parse the mbuf to @@ -713,10 +1067,14 @@ tcp_input(m, off0) * Locate pcb for segment. */ findpcb: + + isconnected = FALSE; + isdisconnected = FALSE; + #if IPFIREWALL_FORWARD if (next_hop != NULL #if INET6 - && isipv6 == NULL /* IPv6 support is not yet */ + && isipv6 == 0 /* IPv6 support is not yet */ #endif /* INET6 */ ) { /* @@ -755,24 +1113,39 @@ findpcb: ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); } + /* + * Use the interface scope information from the PCB for outbound + * segments. If the PCB isn't present and if scoped routing is + * enabled, tcp_respond will use the scope of the interface where + * the segment arrived on. + */ + if (inp != NULL && (inp->inp_flags & INP_BOUND_IF)) + ifscope = inp->inp_boundif; + /* + * If the PCB is present and the socket isn't allowed to use + * the cellular interface, indicate it as such for tcp_respond. + */ + if (inp != NULL && (inp->inp_flags & INP_NO_IFT_CELLULAR)) + nocell = 1; + #if IPSEC if (ipsec_bypass == 0) { - lck_mtx_lock(sadb_mutex); #if INET6 if (isipv6) { if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { - ipsec6stat.in_polvio++; - lck_mtx_unlock(sadb_mutex); + IPSEC_STAT_INCREMENT(ipsec6stat.in_polvio); + if (in_pcb_checkstate(inp, WNT_RELEASE, 0) == WNT_STOPUSING) + inp = NULL; // pretend we didn't find it goto dropnosock; } } else #endif /* INET6 */ if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { - ipsecstat.in_polvio++; - lck_mtx_unlock(sadb_mutex); + IPSEC_STAT_INCREMENT(ipsecstat.in_polvio); + if (in_pcb_checkstate(inp, WNT_RELEASE, 0) == WNT_STOPUSING) + inp = NULL; // pretend we didn't find it goto dropnosock; } - lck_mtx_unlock(sadb_mutex); } #endif /*IPSEC*/ @@ -853,16 +1226,21 @@ findpcb: } so = inp->inp_socket; if (so == NULL) { - if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) - inp = NULL; // pretend we didn't find it + /* This case shouldn't happen as the socket shouldn't be null + * if inp_state isn't set to INPCB_STATE_DEAD + * But just in case, we pretend we didn't find the socket if we hit this case + * as this isn't cause for a panic (the socket might be leaked however)... + */ + inp = NULL; #if TEMPDEBUG - printf("tcp_input: no more socket for inp=%x\n", inp); + printf("tcp_input: no more socket for inp=%x. This shouldn't happen\n", inp); #endif goto dropnosock; } - tcp_lock(so, 1, 2); + + tcp_lock(so, 1, 0); if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { - tcp_unlock(so, 1, 2); + tcp_unlock(so, 1, (void *)2); inp = NULL; // pretend we didn't find it goto dropnosock; } @@ -875,21 +1253,21 @@ findpcb: if (tp->t_state == TCPS_CLOSED) goto drop; -#ifdef __APPLE__ - /* - * Bogus state when listening port owned by SharedIP with loopback as the - * only configured interface: BlueBox does not filters loopback - */ - if (tp->t_state == TCP_NSTATES) - goto drop; -#endif - /* Unscale the window into a 32-bit value. */ if ((thflags & TH_SYN) == 0) tiwin = th->th_win << tp->snd_scale; else tiwin = th->th_win; +#if CONFIG_MACF_NET + if (mac_inpcb_check_deliver(inp, m, AF_INET, SOCK_STREAM)) + goto drop; +#endif + + /* Radar 7377561: Avoid processing packets while closing a listen socket */ + if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN) == 0) + goto drop; + if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { #if TCPDEBUG if (so->so_options & SO_DEBUG) { @@ -912,28 +1290,71 @@ findpcb: #if INET6 struct inpcb *oinp = sotoinpcb(so); #endif /* INET6 */ - int ogencnt = so->so_gencnt; + unsigned int head_ifscope; + unsigned int head_nocell; + + /* Get listener's bound-to-interface, if any */ + head_ifscope = (inp->inp_flags & INP_BOUND_IF) ? + inp->inp_boundif : IFSCOPE_NONE; + /* Get listener's no-cellular information, if any */ + head_nocell = (inp->inp_flags & INP_NO_IFT_CELLULAR) ? 1 : 0; -#if !IPSEC /* - * Current IPsec implementation makes incorrect IPsec - * cache if this check is done here. - * So delay this until duplicated socket is created. + * If the state is LISTEN then ignore segment if it contains an RST. + * If the segment contains an ACK then it is bad and send a RST. + * 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) { - /* - * Note: dropwithreset makes sure we don't - * send a RST in response to a RST. - */ + if (thflags & TH_RST) { + goto drop; + } if (thflags & TH_ACK) { + tp = NULL; tcpstat.tcps_badsyn++; rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } + + /* We come here if there is no SYN set */ + tcpstat.tcps_badsyn++; goto drop; } -#endif 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) + goto drop; + } + /* + * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN + * in_broadcast() should never return true on a received + * packet with M_BCAST not set. + * + * Packets with a multicast source address should also + * be discarded. + */ + if (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)) + 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)) + goto drop; + #if INET6 /* @@ -969,15 +1390,23 @@ findpcb: if (isipv6 && !ip6_use_deprecated) { struct in6_ifaddr *ia6; - if ((ia6 = ip6_getdstifaddr(m)) && - (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { - tp = NULL; - rstreason = BANDLIM_RST_OPENPORT; - goto dropwithreset; + ia6 = ip6_getdstifaddr(m); + if (ia6 != NULL) { + IFA_LOCK_SPIN(&ia6->ia_ifa); + if (ia6->ia6_flags & IN6_IFF_DEPRECATED) { + IFA_UNLOCK(&ia6->ia_ifa); + IFA_REMREF(&ia6->ia_ifa); + tp = NULL; + rstreason = BANDLIM_RST_OPENPORT; + goto dropwithreset; + } + IFA_UNLOCK(&ia6->ia_ifa); + IFA_REMREF(&ia6->ia_ifa); } } #endif if (so->so_filt) { +#if INET6 if (isipv6) { struct sockaddr_in6 *sin6 = (struct sockaddr_in6*)&from; @@ -987,7 +1416,10 @@ findpcb: sin6->sin6_flowinfo = 0; sin6->sin6_addr = ip6->ip6_src; sin6->sin6_scope_id = 0; - } else { + } + else +#endif + { struct sockaddr_in *sin = (struct sockaddr_in*)&from; sin->sin_len = sizeof(*sin); @@ -1001,7 +1433,7 @@ findpcb: } if (so2 == 0) { tcpstat.tcps_listendrop++; - if (tcpdropdropablreq(so)) { + if (tcp_dropdropablreq(so)) { if (so->so_filt) so2 = sonewconn(so, 0, (struct sockaddr*)&from); else @@ -1010,12 +1442,10 @@ findpcb: if (!so2) goto drop; } - /* - * Make sure listening socket did not get closed during socket allocation, - * not only this is incorrect but it is know to cause panic - */ - if (so->so_gencnt != ogencnt) - goto drop; + + /* Point "inp" and "tp" in tandem to new socket */ + inp = (struct inpcb *)so2->so_pcb; + tp = intotcpcb(inp); oso = so; tcp_unlock(so, 0, 0); /* Unlock but keep a reference on listener for now */ @@ -1023,8 +1453,6 @@ findpcb: so = so2; tcp_lock(so, 1, 0); /* - * This is ugly, but .... - * * Mark socket as temporary until we're * committed to keeping it. The code at * ``drop'' and ``dropwithreset'' check the @@ -1032,9 +1460,26 @@ findpcb: * socket created here should be discarded. * We mark the socket as discardable until * we're committed to it below in TCPS_LISTEN. + * There are some error conditions in which we + * have to drop the temporary socket. */ dropsocket++; - inp = (struct inpcb *)so->so_pcb; + /* + * Inherit INP_BOUND_IF from listener; testing if + * head_ifscope is non-zero is sufficient, since it + * can only be set to a non-zero value earlier if + * the listener has such a flag set. + */ + if (head_ifscope != IFSCOPE_NONE) { + inp->inp_flags |= INP_BOUND_IF; + inp->inp_boundif = head_ifscope; + } + /* + * Inherit INP_NO_IFT_CELLULAR from listener. + */ + if (head_nocell) { + inp->inp_flags |= INP_NO_IFT_CELLULAR; + } #if INET6 if (isipv6) inp->in6p_laddr = ip6->ip6_dst; @@ -1042,7 +1487,7 @@ findpcb: inp->inp_vflag &= ~INP_IPV6; inp->inp_vflag |= INP_IPV4; #endif /* INET6 */ - inp->inp_laddr = ip->ip_dst; + inp->inp_laddr = ip->ip_dst; #if INET6 } #endif /* INET6 */ @@ -1063,30 +1508,6 @@ findpcb: tcp_unlock(oso, 1, 0); goto drop; } -#if IPSEC - /* - * To avoid creating incorrectly cached IPsec - * association, this is need to be done here. - * - * Subject: (KAME-snap 748) - * From: Wayne Knowles - * ftp://ftp.kame.net/pub/mail-list/snap-users/748 - */ - if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { - /* - * Note: dropwithreset makes sure we don't - * send a RST in response to a RST. - */ - tcp_lock(oso, 0, 0); /* release ref on parent */ - tcp_unlock(oso, 1, 0); - if (thflags & TH_ACK) { - tcpstat.tcps_badsyn++; - rstreason = BANDLIM_RST_OPENPORT; - goto dropwithreset; - } - goto drop; - } -#endif #if INET6 if (isipv6) { /* @@ -1117,33 +1538,42 @@ findpcb: if (sotoinpcb(oso)->inp_sp) { int error = 0; - lck_mtx_lock(sadb_mutex); /* Is it a security hole here to silently fail to copy the policy? */ 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)) printf("tcp_input: could not copy policy\n"); - lck_mtx_unlock(sadb_mutex); } #endif - tcp_unlock(oso, 1, 0); /* now drop the reference on the listener */ - tp = intotcpcb(inp); + /* inherit states from the listener */ + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + 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); + tp->t_keepinit = tp0->t_keepinit; tp->t_inpcb->inp_ip_ttl = tp0->t_inpcb->inp_ip_ttl; + + /* now drop the reference on the listener */ + tcp_unlock(oso, 1, 0); + /* Compute proper scaling value from buffer space */ - while (tp->request_r_scale < TCP_MAX_WINSHIFT && - TCP_MAXWIN << tp->request_r_scale < - so->so_rcv.sb_hiwat) - tp->request_r_scale++; + if (inp->inp_pcbinfo->ipi_count < tcp_sockthreshold) { + tp->request_r_scale = max(tcp_win_scale, tp->request_r_scale); + so->so_rcv.sb_hiwat = imin(TCP_MAXWIN << tp->request_r_scale, (sb_max / (MSIZE+MCLBYTES)) * MCLBYTES); + } + else { + while (tp->request_r_scale < TCP_MAX_WINSHIFT && + TCP_MAXWIN << tp->request_r_scale < + so->so_rcv.sb_hiwat) + tp->request_r_scale++; + } KERNEL_DEBUG(DBG_FNC_TCP_NEWCONN | DBG_FUNC_END,0,0,0,0,0); } } + lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); -#if 1 - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); -#endif /* * Radar 3529618 * This is the second part of the MSS DoS prevention code (after @@ -1171,18 +1601,23 @@ findpcb: * * Account for packet if payload packet, skip over ACK, etc. */ - if (tcp_minmss && tcp_minmssoverload && - tp->t_state == TCPS_ESTABLISHED && tlen > 0) { - if (tp->rcv_reset > tcp_now) { + if (tp->t_state == TCPS_ESTABLISHED && tlen > 0) { + if (TSTMP_GT(tp->rcv_reset, tcp_now)) { tp->rcv_pps++; tp->rcv_byps += tlen + off; - if (tp->rcv_pps > tcp_minmssoverload) { + if (tp->rcv_byps > tp->rcv_maxbyps) + tp->rcv_maxbyps = tp->rcv_byps; + /* + * Setting either tcp_minmssoverload or tcp_minmss to "0" disables + * the check. + */ + if (tcp_minmss && tcp_minmssoverload && tp->rcv_pps > tcp_minmssoverload) { if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { char ipstrbuf[MAX_IPv6_STR_LEN]; printf("too many small tcp packets from " - "%s:%u, av. %lubyte/packet, " + "%s:%u, av. %ubyte/packet, " "dropping connection\n", -#ifdef INET6 +#if INET6 isipv6 ? inet_ntop(AF_INET6, &inp->in6p_faddr, ipstrbuf, sizeof(ipstrbuf)) : @@ -1197,56 +1632,185 @@ findpcb: } } } else { - tp->rcv_reset = tcp_now + PR_SLOWHZ; + tp->rcv_reset = tcp_now + TCP_RETRANSHZ; tp->rcv_pps = 1; tp->rcv_byps = tlen + off; } + + /* Evaluate the rate of arrival of packets to see if the + * receiver can reduce the ack traffic. The algorithm to + * stretch acks will be enabled if the connection meets + * certain criteria defined in tcp_stretch_ack_enable function. + */ + if ((tp->t_flagsext & TF_RCVUNACK_WAITSS) != 0) { + tp->rcv_waitforss++; + } + if (tcp_stretch_ack_enable(tp)) { + tp->t_flags |= TF_STRETCHACK; + tp->t_flagsext &= ~(TF_RCVUNACK_WAITSS); + tp->rcv_waitforss = 0; + } else { + tp->t_flags &= ~(TF_STRETCHACK); + } + if (TSTMP_GT(tp->rcv_unackwin, tcp_now)) { + tp->rcv_by_unackwin += (tlen + off); + } else { + tp->rcv_unackwin = tcp_now + tcp_rcvunackwin; + tp->rcv_by_unackwin = tlen + off; + } + } + + /* + 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 + + 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 && + (tp->ecn_flags & (TE_SETUPSENT | TE_SETUPRECEIVED)) == + (TE_SETUPSENT | TE_SETUPRECEIVED) && tlen > 0 && + SEQ_GEQ(th->th_seq, tp->last_ack_sent) && + SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { + 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) == TH_CWR) { + tp->ecn_flags &= ~TE_SENDECE; } + + /* 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. + */ + if (tp->t_state == TCPS_ESTABLISHED && (tp->t_flags & TF_STRETCHACK) != 0 && + ((ip_ecn == IPTOS_ECN_CE) || ((thflags & TH_CWR) == TH_CWR))) + tcp_reset_stretch_ack(tp); /* * Segment received on connection. * Reset idle time and keep-alive timer. */ - tp->t_rcvtime = 0; + tp->t_rcvtime = tcp_now; if (TCPS_HAVEESTABLISHED(tp->t_state)) - tp->t_timer[TCPT_KEEP] = TCP_KEEPIDLE(tp); + tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, TCP_KEEPIDLE(tp)); /* * Process options if not in LISTEN state, * else do it below (after getting remote address). */ if (tp->t_state != TCPS_LISTEN && optp) - tcp_dooptions(tp, optp, optlen, th, &to); + tcp_dooptions(tp, optp, optlen, th, &to, ifscope); - /* - * Header prediction: check for the two common cases - * of a uni-directional data xfer. If the packet has - * no control flags, is in-sequence, the window didn't - * change and we're not retransmitting, it's a - * candidate. If the length is zero and the ack moved - * forward, we're the sender side of the xfer. Just - * free the data acked & wake any higher level process - * that was blocked waiting for space. If the length - * is non-zero and the ack didn't move, we're the - * receiver side. If we're getting packets in-order - * (the reassembly queue is empty), add the data to + if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { + if (to.to_flags & TOF_SCALE) { + tp->t_flags |= TF_RCVD_SCALE; + tp->requested_s_scale = to.to_requested_s_scale; + tp->snd_wnd = th->th_win << tp->snd_scale; + tiwin = tp->snd_wnd; + } + if (to.to_flags & TOF_TS) { + tp->t_flags |= TF_RCVD_TSTMP; + tp->ts_recent = to.to_tsval; + tp->ts_recent_age = tcp_now; + } + if (to.to_flags & TOF_MSS) + tcp_mss(tp, to.to_mss, ifscope); + if (tp->sack_enable) { + if (!(to.to_flags & TOF_SACK)) + tp->sack_enable = 0; + else + tp->t_flags |= TF_SACK_PERMIT; + } + } + +#if TRAFFIC_MGT + /* Compute inter-packet arrival jitter. According to RFC 3550, inter-packet + * arrival jitter is defined as the difference in packet spacing at the + * receiver compared to the sender for a pair of packets. When two packets + * of maximum segment size come one after the other with consecutive + * sequence numbers, we consider them as packets sent together at the + * sender and use them as a pair to compute inter-packet arrival jitter. + * This metric indicates the delay induced by the network components due + * 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) && + ((to.to_flags & TOF_TS) == 0 || + TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && + th->th_seq == tp->rcv_nxt && + LIST_EMPTY(&tp->t_segq)) { + if (tp->iaj_pktcnt <= IAJ_IGNORE_PKTCNT) { + tp->iaj_pktcnt++; + } + + if ( tp->iaj_size == 0 || tlen > tp->iaj_size || + (tlen == tp->iaj_size && tp->iaj_rcv_ts == 0)) { + /* State related to inter-arrival jitter is uninitialized + * or we are trying to find a good first packet to start + * computing the metric + */ + update_iaj_state(tp, tlen, 0); + } else { + if (tlen == tp->iaj_size) { + /* Compute inter-arrival jitter taking this packet + * as the second packet + */ + compute_iaj(tp); + } + if (tlen < tp->iaj_size) { + /* There is a smaller packet in the stream. + * Some times the maximum size supported on a path can + * change if there is a new link with smaller MTU. + * The receiver will not know about this change. + * If there are too many packets smaller than iaj_size, + * we try to learn the iaj_size again. + */ + tp->iaj_small_pkt++; + if (tp->iaj_small_pkt > RESET_IAJ_SIZE_THRESH) { + update_iaj_state(tp, tlen, 1); + } else { + clear_iaj_state(tp); + } + } else { + update_iaj_state(tp, tlen, 0); + } + } + } else { + clear_iaj_state(tp); + } +#endif /* TRAFFIC_MGT */ + + /* + * Header prediction: check for the two common cases + * of a uni-directional data xfer. If the packet has + * no control flags, is in-sequence, the window didn't + * change and we're not retransmitting, it's a + * candidate. If the length is zero and the ack moved + * forward, we're the sender side of the xfer. Just + * free the data acked & wake any higher level process + * that was blocked waiting for space. If the length + * is non-zero and the ack didn't move, we're the + * receiver side. If we're getting packets in-order + * (the reassembly queue is empty), add the data to * the socket buffer and note that we need a delayed ack. * Make sure that the hidden state-flags are also off. * Since we check for TCPS_ESTABLISHED above, it can only * be TH_NEEDSYN. */ if (tp->t_state == TCPS_ESTABLISHED && - (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && + (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK|TH_ECE)) == TH_ACK && ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && - ((to.to_flag & TOF_TS) == 0 || + ((to.to_flags & TOF_TS) == 0 || TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && - /* - * Using the CC option is compulsory if once started: - * the segment is OK if no T/TCP was negotiated or - * if the segment has a CC option equal to CCrecv - */ - ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || - ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && tp->snd_nxt == tp->snd_max) { @@ -1257,17 +1821,25 @@ findpcb: * NOTE that the test is modified according to the latest * proposal of the tcplw@cray.com list (Braden 1993/04/26). */ - if ((to.to_flag & TOF_TS) != 0 && + if ((to.to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { tp->ts_recent_age = tcp_now; tp->ts_recent = to.to_tsval; } + /* Force acknowledgment if we received a FIN */ + + if (thflags & TH_FIN) + tp->t_flags |= TF_ACKNOW; + if (tlen == 0) { if (SEQ_GT(th->th_ack, tp->snd_una) && SEQ_LEQ(th->th_ack, tp->snd_max) && - tp->snd_cwnd >= tp->snd_wnd && - tp->t_dupacks < tcprexmtthresh) { + tp->snd_cwnd >= tp->snd_ssthresh && + (!IN_FASTRECOVERY(tp) && + ((!tp->sack_enable && tp->t_dupacks < tcprexmtthresh) || + (tp->sack_enable && to.to_nsacks == 0 && + TAILQ_EMPTY(&tp->snd_holes))))) { /* * this is a pure ack for outstanding data. */ @@ -1276,24 +1848,64 @@ findpcb: * "bad retransmit" recovery */ if (tp->t_rxtshift == 1 && - tcp_now < tp->t_badrxtwin) { + TSTMP_LT(tcp_now, tp->t_badrxtwin)) { + ++tcpstat.tcps_sndrexmitbad; tp->snd_cwnd = tp->snd_cwnd_prev; tp->snd_ssthresh = tp->snd_ssthresh_prev; + tp->snd_recover = tp->snd_recover_prev; + if (tp->t_flags & TF_WASFRECOVERY) + ENTER_FASTRECOVERY(tp); tp->snd_nxt = tp->snd_max; tp->t_badrxtwin = 0; + tp->t_rxtshift = 0; + tp->rxt_start = 0; + DTRACE_TCP5(cc, void, NULL, struct inpcb *, tp->t_inpcb, + struct tcpcb *, tp, struct tcphdr *, th, + int32_t, TCP_CC_BAD_REXMT_RECOVERY); } - if (((to.to_flag & TOF_TS) != 0) && (to.to_tsecr != 0)) /* Makes sure we already have a TS */ + /* + * Recalculate the transmit timer / rtt. + * + * Some boxes send broken timestamp replies + * during the SYN+ACK phase, ignore + * timestamps of 0 or we could calculate a + * huge RTT and blow up the retransmit timer. + */ + if (((to.to_flags & TOF_TS) != 0) && (to.to_tsecr != 0) && + TSTMP_GEQ(tcp_now, to.to_tsecr)) { tcp_xmit_timer(tp, - tcp_now - to.to_tsecr + 1); - else if (tp->t_rtttime && - SEQ_GT(th->th_ack, tp->t_rtseq)) - tcp_xmit_timer(tp, tp->t_rtttime); + tcp_now - to.to_tsecr); + } else if (tp->t_rtttime && + SEQ_GT(th->th_ack, tp->t_rtseq)) { + tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); + } acked = th->th_ack - tp->snd_una; tcpstat.tcps_rcvackpack++; tcpstat.tcps_rcvackbyte += acked; + + /* Handle an ack that is in sequence during congestion + * avoidance phase. The calculations in this function + * assume that snd_una is not updated yet. + */ + if (CC_ALGO(tp)->inseq_ack_rcvd != NULL) + CC_ALGO(tp)->inseq_ack_rcvd(tp, th); + + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, th, + int32_t, TCP_CC_INSEQ_ACK_RCVD); + sbdrop(&so->so_snd, acked); + if (SEQ_GT(tp->snd_una, tp->snd_recover) && + SEQ_LEQ(th->th_ack, tp->snd_recover)) + tp->snd_recover = th->th_ack - 1; tp->snd_una = th->th_ack; + /* + * pull snd_wl2 up to prevent seq wrap relative + * to th_ack. + */ + tp->snd_wl2 = th->th_ack; + tp->t_dupacks = 0; m_freem(m); ND6_HINT(tp); /* some progress has been done */ @@ -1309,33 +1921,54 @@ findpcb: if (tp->snd_una == tp->snd_max) tp->t_timer[TCPT_REXMT] = 0; else if (tp->t_timer[TCPT_PERSIST] == 0) - tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; + tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp, tp->t_rxtcur); sowwakeup(so); /* has to be done with socket lock held */ - if ((so->so_snd.sb_cc) || (tp->t_flags & TF_ACKNOW)) + if ((so->so_snd.sb_cc) || (tp->t_flags & TF_ACKNOW)) { (void) tcp_output(tp); + } + + tcp_check_timer_state(tp); tcp_unlock(so, 1, 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) && - tlen <= sbspace(&so->so_rcv)) { + tlen <= tcp_sbspace(tp)) { /* * this is a pure, in-sequence data packet * with nothing on the reassembly queue and * we have enough buffer space to take it. */ + /* Clean receiver SACK report if present */ + if (tp->sack_enable && tp->rcv_numsacks) + tcp_clean_sackreport(tp); ++tcpstat.tcps_preddat; tp->rcv_nxt += tlen; + /* + * Pull snd_wl1 up to prevent seq wrap relative to + * th_seq. + */ + tp->snd_wl1 = th->th_seq; + /* + * Pull rcv_up up to prevent seq wrap relative to + * rcv_nxt. + */ + tp->rcv_up = tp->rcv_nxt; tcpstat.tcps_rcvpack++; tcpstat.tcps_rcvbyte += tlen; + if (nstat_collect) { + locked_add_64(&inp->inp_stat->rxpackets, 1); + locked_add_64(&inp->inp_stat->rxbytes, tlen); + } ND6_HINT(tp); /* some progress has been done */ /* * Add data to socket buffer. */ + so_recv_data_stat(so, m, 0); m_adj(m, drop_hdrlen); /* delayed header drop */ - if (sbappend(&so->so_rcv, m)) + if (sbappendstream(&so->so_rcv, m)) sorwakeup(so); #if INET6 if (isipv6) { @@ -1350,12 +1983,17 @@ findpcb: (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)), th->th_seq, th->th_ack, th->th_win); } - if (DELAY_ACK(tp)) { - tp->t_flags |= TF_DELACK; + if (DELAY_ACK(tp, th)) { + if ((tp->t_flags & TF_DELACK) == 0) { + tp->t_flags |= TF_DELACK; + tp->t_timer[TCPT_DELACK] = OFFSET_FROM_START(tp, tcp_delack); + } + tp->t_unacksegs++; } else { tp->t_flags |= TF_ACKNOW; tcp_output(tp); } + tcp_check_timer_state(tp); tcp_unlock(so, 1, 0); KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); return; @@ -1368,12 +2006,12 @@ findpcb: * Receive window is amount of space in rcv queue, * but not less than advertised window. */ -#if 1 - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); -#endif + lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); + { int win; - win = sbspace(&so->so_rcv); + win = tcp_sbspace(tp); + if (win < 0) win = 0; else { /* clip rcv window to 4K for modems */ @@ -1386,12 +2024,7 @@ findpcb: switch (tp->t_state) { /* - * If the state is LISTEN then ignore segment if it contains an RST. - * If the segment contains an ACK then it is bad and send a RST. - * 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. - * Don't bother responding if the destination was a broadcast. - * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial + * 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. @@ -1405,50 +2038,7 @@ findpcb: register struct sockaddr_in6 *sin6; #endif -#if 1 - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); -#endif - if (thflags & TH_RST) - goto drop; - if (thflags & TH_ACK) { - rstreason = BANDLIM_RST_OPENPORT; - goto dropwithreset; - } - if ((thflags & TH_SYN) == 0) - goto drop; - 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) - goto drop; - } - /* - * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN - * in_broadcast() should never return true on a received - * packet with M_BCAST not set. - * - * Packets with a multicast source address should also - * be discarded. - */ - if (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)) - 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)) - goto drop; + lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); #if INET6 if (isipv6) { MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6, @@ -1473,9 +2063,7 @@ findpcb: } else #endif { -#if 1 - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); -#endif + lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, M_NOWAIT); if (sin == NULL) @@ -1488,18 +2076,23 @@ findpcb: laddr = inp->inp_laddr; 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, proc0, NULL)) { inp->inp_laddr = laddr; FREE(sin, M_SONAME); goto drop; } FREE(sin, M_SONAME); } - if ((taop = tcp_gettaocache(inp)) == NULL) { - taop = &tao_noncached; - bzero(taop, sizeof(*taop)); + + tcp_dooptions(tp, optp, optlen, th, &to, ifscope); + + if (tp->sack_enable) { + if (!(to.to_flags & TOF_SACK)) + tp->sack_enable = 0; + else + tp->t_flags |= TF_SACK_PERMIT; } - tcp_dooptions(tp, optp, optlen, th, &to); + if (iss) tp->iss = iss; else { @@ -1515,87 +2108,29 @@ findpcb: * initialize CCsend and CCrecv. */ tp->snd_wnd = tiwin; /* initial send-window */ - tp->cc_send = CC_INC(tcp_ccgen); - tp->cc_recv = to.to_cc; - /* - * Perform TAO test on incoming CC (SEG.CC) option, if any. - * - compare SEG.CC against cached CC from the same host, - * if any. - * - if SEG.CC > chached value, SYN must be new and is accepted - * immediately: save new CC in the cache, mark the socket - * connected, enter ESTABLISHED state, turn on flag to - * send a SYN in the next segment. - * A virtual advertised window is set in rcv_adv to - * initialize SWS prevention. Then enter normal segment - * processing: drop SYN, process data and FIN. - * - otherwise do a normal 3-way handshake. - */ - if ((to.to_flag & TOF_CC) != 0) { - if (((tp->t_flags & TF_NOPUSH) != 0) && - taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { - - taop->tao_cc = to.to_cc; - - tp->t_state = TCPS_ESTABLISHED; - - /* - * If there is a FIN, or if there is data and the - * connection is local, then delay SYN,ACK(SYN) in - * the hope of piggy-backing it on a response - * segment. Otherwise must send ACK now in case - * the other side is slow starting. - */ - if (DELAY_ACK(tp) && ((thflags & TH_FIN) || - (tlen != 0 && -#if INET6 - (isipv6 && in6_localaddr(&inp->in6p_faddr)) - || - (!isipv6 && -#endif /* INET6 */ - in_localaddr(inp->inp_faddr) -#if INET6 - ) -#endif /* INET6 */ - ))) { - tp->t_flags |= (TF_DELACK | TF_NEEDSYN); - } - else { - tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); - } - - /* - * Limit the `virtual advertised window' to TCP_MAXWIN - * here. Even if we requested window scaling, it will - * become effective only later when our SYN is acked. - */ - if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0) /* clip window size for for slow link */ - tp->rcv_adv += min(tp->rcv_wnd, slowlink_wsize); - else - tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); - tcpstat.tcps_connects++; - soisconnected(so); - tp->t_timer[TCPT_KEEP] = tcp_keepinit; - dropsocket = 0; /* committed to socket */ - tcpstat.tcps_accepts++; - goto trimthenstep6; - } - /* else do standard 3-way handshake */ - } else { - /* - * No CC option, but maybe CC.NEW: - * invalidate cached value. - */ - taop->tao_cc = 0; - } - /* - * TAO test failed or there was no CC option, - * do a standard 3-way handshake. - */ 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); tp->t_state = TCPS_SYN_RECEIVED; - tp->t_timer[TCPT_KEEP] = tcp_keepinit; + tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, + tp->t_keepinit ? tp->t_keepinit : tcp_keepinit); dropsocket = 0; /* committed to socket */ + + /* reset the incomp processing flag */ + so->so_flags &= ~(SOF_INCOMP_INPROGRESS); tcpstat.tcps_accepts++; + if ((thflags & (TH_ECE | TH_CWR)) == (TH_ECE | TH_CWR)) { + /* ECN-setup SYN */ + tp->ecn_flags |= (TE_SETUPRECEIVED | TE_SENDIPECT); + } +#if CONFIG_IFEF_NOWINDOWSCALE + if (tcp_obey_ifef_nowindowscale && m->m_pkthdr.rcvif != NULL && + (m->m_pkthdr.rcvif->if_eflags & IFEF_NOWINDOWSCALE)) { + /* Window scaling is not enabled on this interface */ + tp->t_flags &= ~TF_REQ_SCALE; + } +#endif goto trimthenstep6; } @@ -1625,83 +2160,60 @@ findpcb: * continue processing rest of data/controls, beginning with URG */ case TCPS_SYN_SENT: - if ((taop = tcp_gettaocache(inp)) == NULL) { - taop = &tao_noncached; - bzero(taop, sizeof(*taop)); - } - if ((thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { - /* - * If we have a cached CCsent for the remote host, - * hence we haven't just crashed and restarted, - * do not send a RST. This may be a retransmission - * from the other side after our earlier ACK was lost. - * Our new SYN, when it arrives, will serve as the - * needed ACK. - */ - if (taop->tao_ccsent != 0) - goto drop; - else { - rstreason = BANDLIM_UNLIMITED; - goto dropwithreset; - } + rstreason = BANDLIM_UNLIMITED; + goto dropwithreset; } if (thflags & TH_RST) { - if (thflags & TH_ACK) { + if ((thflags & TH_ACK) != 0) { tp = tcp_drop(tp, ECONNREFUSED); postevent(so, 0, EV_RESET); - } + } goto drop; } if ((thflags & TH_SYN) == 0) goto drop; tp->snd_wnd = th->th_win; /* initial send window */ - tp->cc_recv = to.to_cc; /* foreign CC */ tp->irs = th->th_seq; tcp_rcvseqinit(tp); if (thflags & TH_ACK) { - /* - * Our SYN was acked. If segment contains CC.ECHO - * option, check it to make sure this segment really - * matches our SYN. If not, just drop it as old - * duplicate, but send an RST if we're still playing - * by the old rules. If no CC.ECHO option, make sure - * we don't get fooled into using T/TCP. - */ - if (to.to_flag & TOF_CCECHO) { - if (tp->cc_send != to.to_ccecho) { - if (taop->tao_ccsent != 0) - goto drop; - else { - rstreason = BANDLIM_UNLIMITED; - goto dropwithreset; - } - } - } else - tp->t_flags &= ~TF_RCVD_CC; tcpstat.tcps_connects++; - soisconnected(so); + + if ((thflags & (TH_ECE | TH_CWR)) == (TH_ECE)) { + /* ECN-setup SYN-ACK */ + tp->ecn_flags |= TE_SETUPRECEIVED; + } + else { + /* 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 ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->snd_scale = tp->requested_s_scale; tp->rcv_scale = tp->request_r_scale; } - /* Segment is acceptable, update cache if undefined. */ - if (taop->tao_ccsent == 0) - taop->tao_ccsent = to.to_ccecho; - tp->rcv_adv += tp->rcv_wnd; tp->snd_una++; /* SYN is acked */ /* * If there's data, delay ACK; if there's also a FIN * ACKNOW will be turned on later. */ - if (DELAY_ACK(tp) && tlen != 0) { - tp->t_flags |= TF_DELACK; + 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); + } + tp->t_unacksegs++; } else { tp->t_flags |= TF_ACKNOW; @@ -1712,50 +2224,37 @@ findpcb: * SYN_SENT --> ESTABLISHED * SYN_SENT* --> FIN_WAIT_1 */ + tp->t_starttime = tcp_now; if (tp->t_flags & TF_NEEDFIN) { + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_FIN_WAIT_1); tp->t_state = TCPS_FIN_WAIT_1; tp->t_flags &= ~TF_NEEDFIN; thflags &= ~TH_SYN; } 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] = TCP_KEEPIDLE(tp); + tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, TCP_KEEPIDLE(tp)); + if (nstat_collect) + nstat_route_connect_success(tp->t_inpcb->inp_route.ro_rt); } + isconnected = TRUE; } 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: - * SYN-SENT -> SYN-RECEIVED - * SYN-SENT* -> SYN-RECEIVED* - * If there was no CC option, clear cached CC value. - */ + /* + * 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: + * SYN-SENT -> SYN-RECEIVED + * SYN-SENT* -> SYN-RECEIVED* + */ tp->t_flags |= TF_ACKNOW; tp->t_timer[TCPT_REXMT] = 0; - if (to.to_flag & TOF_CC) { - if (taop->tao_cc != 0 && - CC_GT(to.to_cc, taop->tao_cc)) { - /* - * update cache and make transition: - * SYN-SENT -> ESTABLISHED* - * SYN-SENT* -> FIN-WAIT-1* - */ - taop->tao_cc = to.to_cc; - if (tp->t_flags & TF_NEEDFIN) { - tp->t_state = TCPS_FIN_WAIT_1; - tp->t_flags &= ~TF_NEEDFIN; - } else { - tp->t_state = TCPS_ESTABLISHED; - tp->t_timer[TCPT_KEEP] = TCP_KEEPIDLE(tp); - } - tp->t_flags |= TF_NEEDSYN; - } else - tp->t_state = TCPS_SYN_RECEIVED; - } else { - /* CC.NEW or no option => invalidate cache */ - taop->tao_cc = 0; - tp->t_state = TCPS_SYN_RECEIVED; - } + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_SYN_RECEIVED); + tp->t_state = TCPS_SYN_RECEIVED; + } trimthenstep6: @@ -1787,36 +2286,13 @@ trimthenstep6: goto step6; /* * If the state is LAST_ACK or CLOSING or TIME_WAIT: - * if segment contains a SYN and CC [not CC.NEW] option: - * if state == TIME_WAIT and connection duration > MSL, - * drop packet and send RST; + * do normal processing. * - * if SEG.CC > CCrecv then is new SYN, and can implicitly - * ack the FIN (and data) in retransmission queue. - * Complete close and delete TCPCB. Then reprocess - * segment, hoping to find new TCPCB in LISTEN state; - * - * else must be old SYN; drop it. - * else do normal processing. + * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. */ case TCPS_LAST_ACK: case TCPS_CLOSING: case TCPS_TIME_WAIT: - if ((thflags & TH_SYN) && - (to.to_flag & TOF_CC) && tp->cc_recv != 0) { - if (tp->t_state == TCPS_TIME_WAIT && - tp->t_starttime > tcp_msl) { - rstreason = BANDLIM_UNLIMITED; - goto dropwithreset; - } - if (CC_GT(to.to_cc, tp->cc_recv)) { - tp = tcp_close(tp); - tcp_unlock(so, 1, 50); - goto findpcb; - } - else - goto drop; - } break; /* continue normal processing */ /* Received a SYN while connection is already established. @@ -1858,6 +2334,12 @@ trimthenstep6: * echo of our outgoing acknowlegement numbers, but some hosts * send a reset with the sequence number at the rightmost edge * of our receive window, and we have to handle this case. + * Note 2: Paul Watson's paper "Slipping in the Window" has shown + * that brute force RST attacks are possible. To combat this, + * we use a much stricter check while in the ESTABLISHED state, + * 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. * 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. @@ -1878,17 +2360,25 @@ trimthenstep6: * SYN_RECEIVED STATE: * If passive open, return to LISTEN state. * If active open, inform user that connection was refused. - * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: + * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: * Inform user that connection was reset, and close tcb. * CLOSING, LAST_ACK STATES: * Close the tcb. * TIME_WAIT STATE: * Drop the segment - see Stevens, vol. 2, p. 964 and * RFC 1337. + * + * Radar 4803931: Allows for the case where we ACKed the FIN but + * 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)) { + 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) { case TCPS_SYN_RECEIVED: @@ -1896,6 +2386,10 @@ trimthenstep6: goto close; case TCPS_ESTABLISHED: + if (tp->last_ack_sent != th->th_seq) { + tcpstat.tcps_badrst++; + goto drop; + } case TCPS_FIN_WAIT_1: case TCPS_CLOSE_WAIT: /* @@ -1905,6 +2399,8 @@ trimthenstep6: so->so_error = ECONNRESET; close: postevent(so, 0, EV_RESET); + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_CLOSED); tp->t_state = TCPS_CLOSED; tcpstat.tcps_drops++; tp = tcp_close(tp); @@ -1922,14 +2418,11 @@ trimthenstep6: goto drop; } -#if 1 - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); -#endif /* * RFC 1323 PAWS: If we have a timestamp reply on this segment * and it's less than ts_recent, drop it. */ - if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent && + 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. */ @@ -1950,20 +2443,18 @@ trimthenstep6: tcpstat.tcps_rcvduppack++; tcpstat.tcps_rcvdupbyte += tlen; tcpstat.tcps_pawsdrop++; - goto dropafterack; + if (nstat_collect) { + nstat_route_rx(tp->t_inpcb->inp_route.ro_rt, 1, tlen, NSTAT_RX_FLAG_DUPLICATE); + locked_add_64(&inp->inp_stat->rxpackets, 1); + locked_add_64(&inp->inp_stat->rxbytes, tlen); + tp->t_stat.rxduplicatebytes += tlen; + } + if (tlen) + goto dropafterack; + goto drop; } } - /* - * T/TCP mechanism - * If T/TCP was negotiated and the segment doesn't have CC, - * or if its CC is wrong then drop the segment. - * RST segments do not have to comply with this. - */ - if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && - ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) - goto dropafterack; - /* * In the SYN-RECEIVED state, validate that the packet belongs to * this connection before trimming the data to fit the receive @@ -2011,6 +2502,12 @@ trimthenstep6: tcpstat.tcps_rcvpartduppack++; tcpstat.tcps_rcvpartdupbyte += todrop; } + if (nstat_collect) { + nstat_route_rx(tp->t_inpcb->inp_route.ro_rt, 1, todrop, NSTAT_RX_FLAG_DUPLICATE); + locked_add_64(&inp->inp_stat->rxpackets, 1); + locked_add_64(&inp->inp_stat->rxbytes, todrop); + tp->t_stat.rxduplicatebytes += todrop; + } drop_hdrlen += todrop; /* drop from the top afterwards */ th->th_seq += todrop; tlen -= todrop; @@ -2079,11 +2576,25 @@ trimthenstep6: /* * If last ACK falls within this segment's sequence numbers, * record its timestamp. - * NOTE that the test is modified according to the latest - * proposal of the tcplw@cray.com list (Braden 1993/04/26). + * NOTE: + * 1) That the test incorporates suggestions from the latest + * proposal of the tcplw@cray.com list (Braden 1993/04/26). + * 2) That updating only on newer timestamps interferes with + * our earlier PAWS tests, so this check should be solely + * predicated on the sequence space of this segment. + * 3) That we modify the segment boundary check to be + * Last.ACK.Sent <= SEG.SEQ + SEG.Len + * instead of RFC1323's + * Last.ACK.Sent < SEG.SEQ + SEG.Len, + * This modified check allows us to overcome RFC1323's + * limitations as described in Stevens TCP/IP Illustrated + * Vol. 2 p.869. In such cases, we can still calculate the + * RTT correctly when RCV.NXT == Last.ACK.Sent. */ - if ((to.to_flag & TOF_TS) != 0 && - SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { + 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))) { tp->ts_recent_age = tcp_now; tp->ts_recent = to.to_tsval; } @@ -2108,6 +2619,8 @@ trimthenstep6: if (tp->t_state == TCPS_SYN_RECEIVED || (tp->t_flags & TF_NEEDSYN)) goto step6; + else if (tp->t_flags & TF_ACKNOW) + goto dropafterack; else goto drop; } @@ -2125,44 +2638,46 @@ trimthenstep6: case TCPS_SYN_RECEIVED: tcpstat.tcps_connects++; - soisconnected(so); /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->snd_scale = tp->requested_s_scale; tp->rcv_scale = tp->request_r_scale; + tp->snd_wnd = th->th_win << tp->snd_scale; + tiwin = tp->snd_wnd; } - /* - * Upon successful completion of 3-way handshake, - * update cache.CC if it was undefined, pass any queued - * data to the user, and advance state appropriately. - */ - if ((taop = tcp_gettaocache(inp)) != NULL && - taop->tao_cc == 0) - taop->tao_cc = tp->cc_recv; - /* * Make transitions: * SYN-RECEIVED -> ESTABLISHED * SYN-RECEIVED* -> FIN-WAIT-1 */ + tp->t_starttime = tcp_now; if (tp->t_flags & TF_NEEDFIN) { + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_FIN_WAIT_1); tp->t_state = TCPS_FIN_WAIT_1; tp->t_flags &= ~TF_NEEDFIN; } 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] = TCP_KEEPIDLE(tp); + tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, TCP_KEEPIDLE(tp)); + if (nstat_collect) + nstat_route_connect_success(tp->t_inpcb->inp_route.ro_rt); } /* * 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) - (void) tcp_reass(tp, (struct tcphdr *)0, 0, + (void) tcp_reass(tp, (struct tcphdr *)0, &tlen, (struct mbuf *)0); tp->snd_wl1 = th->th_seq - 1; - /* fall into ... */ + + /* FALLTHROUGH */ + + isconnected = TRUE; /* * In ESTABLISHED state: drop duplicate ACKs; ACK out of range @@ -2179,7 +2694,13 @@ trimthenstep6: case TCPS_CLOSING: case TCPS_LAST_ACK: case TCPS_TIME_WAIT: - + if (SEQ_GT(th->th_ack, tp->snd_max)) { + tcpstat.tcps_rcvacktoomuch++; + goto dropafterack; + } + if (tp->sack_enable && + (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes))) + tcp_sack_doack(tp, &to, th->th_ack); if (SEQ_LEQ(th->th_ack, tp->snd_una)) { if (tlen == 0 && tiwin == tp->snd_wnd) { tcpstat.tcps_rcvdupack++; @@ -2210,38 +2731,90 @@ trimthenstep6: if (tp->t_timer[TCPT_REXMT] == 0 || th->th_ack != tp->snd_una) tp->t_dupacks = 0; - else if (++tp->t_dupacks == tcprexmtthresh) { - tcp_seq onxt = tp->snd_nxt; - u_int win = - min(tp->snd_wnd, tp->snd_cwnd) / 2 / - tp->t_maxseg; - if (tcp_do_newreno && SEQ_LT(th->th_ack, - tp->snd_recover)) { - /* False retransmit, should not - * cut window + else if (++tp->t_dupacks > tcprexmtthresh || + IN_FASTRECOVERY(tp)) { + if (tp->sack_enable && 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. */ + 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) + tp->snd_cwnd = tp->snd_ssthresh; + } + } else tp->snd_cwnd += tp->t_maxseg; - tp->t_dupacks = 0; - (void) tcp_output(tp); - goto drop; + + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, th, + int32_t, TCP_CC_IN_FASTRECOVERY); + + (void) tcp_output(tp); + goto drop; + } else if (tp->t_dupacks == tcprexmtthresh) { + tcp_seq onxt = tp->snd_nxt; + + /* + * If we're doing sack, check to + * see if we're already in sack + * recovery. If we're not doing sack, + * check to see if we're in newreno + * recovery. + */ + if (tp->sack_enable) { + if (IN_FASTRECOVERY(tp)) { + tp->t_dupacks = 0; + break; + } + } else { + if (SEQ_LEQ(th->th_ack, + tp->snd_recover)) { + tp->t_dupacks = 0; + break; + } } - if (win < 2) - win = 2; - tp->snd_ssthresh = win * tp->t_maxseg; + + /* + * 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, th); + ENTER_FASTRECOVERY(tp); tp->snd_recover = tp->snd_max; tp->t_timer[TCPT_REXMT] = 0; tp->t_rtttime = 0; + tp->ecn_flags |= TE_SENDCWR; + if (tp->sack_enable) { + tcpstat.tcps_sack_recovery_episode++; + tp->sack_newdata = tp->snd_nxt; + tp->snd_cwnd = tp->t_maxseg; + + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, th, + int32_t, TCP_CC_ENTER_FASTRECOVERY); + + (void) tcp_output(tp); + goto drop; + } tp->snd_nxt = th->th_ack; tp->snd_cwnd = tp->t_maxseg; (void) tcp_output(tp); tp->snd_cwnd = tp->snd_ssthresh + - tp->t_maxseg * tp->t_dupacks; + tp->t_maxseg * tp->t_dupacks; if (SEQ_GT(onxt, tp->snd_nxt)) tp->snd_nxt = onxt; - goto drop; - } else if (tp->t_dupacks > tcprexmtthresh) { - tp->snd_cwnd += tp->t_maxseg; - (void) tcp_output(tp); + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, th, + int32_t, TCP_CC_ENTER_FASTRECOVERY); goto drop; } } else @@ -2252,36 +2825,37 @@ trimthenstep6: * If the congestion window was inflated to account * for the other side's cached packets, retract it. */ - if (tcp_do_newreno == 0) { - if (tp->t_dupacks >= tcprexmtthresh && - tp->snd_cwnd > tp->snd_ssthresh) - tp->snd_cwnd = tp->snd_ssthresh; - tp->t_dupacks = 0; - } else if (tp->t_dupacks >= tcprexmtthresh && - !tcp_newreno(tp, th)) { - /* - * Window inflation should have left us with approx. - * snd_ssthresh outstanding data. But in case we - * would be inclined to send a burst, better to do - * it via the slow start mechanism. - */ - if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max)) - tp->snd_cwnd = - tp->snd_max - th->th_ack + tp->t_maxseg; - else - tp->snd_cwnd = tp->snd_ssthresh; - tp->t_dupacks = 0; - } + if (IN_FASTRECOVERY(tp)) { + if (SEQ_LT(th->th_ack, tp->snd_recover)) { + if (tp->sack_enable) + tcp_sack_partialack(tp, th); + else + tcp_newreno_partial_ack(tp, th); + + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, th, + int32_t, TCP_CC_PARTIAL_ACK); + } else { + EXIT_FASTRECOVERY(tp); + if (CC_ALGO(tp)->post_fr != NULL) + CC_ALGO(tp)->post_fr(tp, th); + tp->t_dupacks = 0; - if (tp->t_dupacks < tcprexmtthresh) + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, th, + int32_t, TCP_CC_EXIT_FASTRECOVERY); + } + } else { + /* + * We were not in fast recovery. Reset the duplicate ack + * counter. + */ tp->t_dupacks = 0; - - if (SEQ_GT(th->th_ack, tp->snd_max)) { - tcpstat.tcps_rcvacktoomuch++; - goto dropafterack; } + + /* - * If we reach this point, ACK is not a duplicate, + * If we reach this point, ACK is not a duplicate, * i.e., it ACKs something we sent. */ if (tp->t_flags & TF_NEEDSYN) { @@ -2314,11 +2888,22 @@ process_ACK: * original cwnd and ssthresh, and proceed to transmit where * we left off. */ - if (tp->t_rxtshift == 1 && tcp_now < tp->t_badrxtwin) { + if (tp->t_rxtshift == 1 && + TSTMP_LT(tcp_now, tp->t_badrxtwin)) { + ++tcpstat.tcps_sndrexmitbad; tp->snd_cwnd = tp->snd_cwnd_prev; tp->snd_ssthresh = tp->snd_ssthresh_prev; + tp->snd_recover = tp->snd_recover_prev; + if (tp->t_flags & TF_WASFRECOVERY) + ENTER_FASTRECOVERY(tp); tp->snd_nxt = tp->snd_max; tp->t_badrxtwin = 0; /* XXX probably not required */ + tp->t_rxtshift = 0; + tp->rxt_start = 0; + + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, th, + int32_t, TCP_CC_BAD_REXMT_RECOVERY); } /* @@ -2330,11 +2915,18 @@ process_ACK: * timer backoff (cf., Phil Karn's retransmit alg.). * Recompute the initial retransmit timer. * Also makes sure we have a valid time stamp in hand + * + * Some boxes send broken timestamp replies + * during the SYN+ACK phase, ignore + * timestamps of 0 or we could calculate a + * huge RTT and blow up the retransmit timer. */ - if (((to.to_flag & TOF_TS) != 0) && (to.to_tsecr != 0)) - tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); - else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) - tcp_xmit_timer(tp, tp->t_rtttime); + if (((to.to_flags & TOF_TS) != 0) && (to.to_tsecr != 0) && + TSTMP_GEQ(tcp_now, to.to_tsecr)) { + tcp_xmit_timer(tp, tcp_now - to.to_tsecr); + } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { + tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); + } /* * If all outstanding data is acked, stop retransmit @@ -2346,7 +2938,7 @@ process_ACK: tp->t_timer[TCPT_REXMT] = 0; needoutput = 1; } else if (tp->t_timer[TCPT_PERSIST] == 0) - tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; + tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp, tp->t_rxtcur); /* * If no data (only SYN) was ACK'd, @@ -2355,26 +2947,35 @@ process_ACK: if (acked == 0) goto step6; - /* - * When new data is acked, open the congestion window. - * If the window gives us less than ssthresh packets - * in flight, open exponentially (maxseg per packet). - * Otherwise open linearly: maxseg per window - * (maxseg^2 / cwnd per packet). - */ - { - register u_int cw = tp->snd_cwnd; - register u_int incr = tp->t_maxseg; + if ((thflags & TH_ECE) != 0 && + (tp->ecn_flags & TE_SETUPSENT) != 0) { + /* + * Reduce the congestion window if we haven't done so. + */ + if (!tp->sack_enable && !IN_FASTRECOVERY(tp) && + SEQ_GEQ(th->th_ack, tp->snd_recover)) { + tcp_reduce_congestion_window(tp, th); + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, th, + int32_t, TCP_CC_ECN_RCVD); + } + } - if (cw > tp->snd_ssthresh) - incr = incr * incr / cw; /* - * If t_dupacks != 0 here, it indicates that we are still - * in NewReno fast recovery mode, so we leave the congestion - * window alone. + * When new data is acked, open the congestion window. + * The specifics of how this is achieved are up to the + * congestion control algorithm in use for this connection. + * + * The calculations in this function assume that snd_una is + * not updated yet. */ - if (tcp_do_newreno == 0 || tp->t_dupacks == 0) - tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<snd_scale); + if (!IN_FASTRECOVERY(tp)) { + if (CC_ALGO(tp)->ack_rcvd != NULL) + CC_ALGO(tp)->ack_rcvd(tp, th); + + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, th, + int32_t, TCP_CC_ACK_RCVD); } if (acked > so->so_snd.sb_cc) { tp->snd_wnd -= so->so_snd.sb_cc; @@ -2385,17 +2986,29 @@ process_ACK: tp->snd_wnd -= acked; ourfinisacked = 0; } - sowwakeup(so); /* detect una wraparound */ - if (SEQ_GEQ(tp->snd_una, tp->snd_recover) && - SEQ_LT(th->th_ack, tp->snd_recover)) - tp->snd_recover = th->th_ack; - if (SEQ_GT(tp->snd_una, tp->snd_high) && - SEQ_LEQ(th->th_ack, tp->snd_high)) - tp->snd_high = th->th_ack - 1; + if ( !IN_FASTRECOVERY(tp) && + SEQ_GT(tp->snd_una, 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)) + EXIT_FASTRECOVERY(tp); + tp->snd_una = th->th_ack; + if (tp->sack_enable) { + if (SEQ_GT(tp->snd_una, tp->snd_recover)) + tp->snd_recover = tp->snd_una; + } if (SEQ_LT(tp->snd_nxt, tp->snd_una)) tp->snd_nxt = tp->snd_una; + + /* + * sowwakeup must happen after snd_una, et al. are updated so that + * the sequence numbers are in sync with so_snd + */ + sowwakeup(so); switch (tp->t_state) { @@ -2414,13 +3027,16 @@ process_ACK: * we'll hang forever. */ if (so->so_state & SS_CANTRCVMORE) { - soisdisconnected(so); - tp->t_timer[TCPT_2MSL] = tcp_maxidle; + add_to_time_wait(tp, tcp_maxidle); + isconnected = FALSE; + isdisconnected = TRUE; } - add_to_time_wait(tp); + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_FIN_WAIT_2); tp->t_state = TCPS_FIN_WAIT_2; - goto drop; + /* fall through and make sure we also recognize data ACKed with the FIN */ } + tp->t_flags |= TF_ACKNOW; break; /* @@ -2431,18 +3047,20 @@ process_ACK: */ case TCPS_CLOSING: if (ourfinisacked) { + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_TIME_WAIT); tp->t_state = TCPS_TIME_WAIT; tcp_canceltimers(tp); /* Shorten TIME_WAIT [RFC-1644, p.28] */ if (tp->cc_recv != 0 && - tp->t_starttime < tcp_msl) - tp->t_timer[TCPT_2MSL] = - tp->t_rxtcur * TCPTV_TWTRUNC; + ((int)(tcp_now - tp->t_starttime)) < tcp_msl) + add_to_time_wait(tp, tp->t_rxtcur * TCPTV_TWTRUNC); else - tp->t_timer[TCPT_2MSL] = 2 * tcp_msl; - add_to_time_wait(tp); - soisdisconnected(so); + add_to_time_wait(tp, 2 * tcp_msl); + isconnected = FALSE; + isdisconnected = TRUE; } + tp->t_flags |= TF_ACKNOW; break; /* @@ -2464,8 +3082,7 @@ process_ACK: * it and restart the finack timer. */ case TCPS_TIME_WAIT: - tp->t_timer[TCPT_2MSL] = 2 * tcp_msl; - add_to_time_wait(tp); + add_to_time_wait(tp, 2 * tcp_msl); goto dropafterack; } } @@ -2538,14 +3155,14 @@ step6: * but if two URG's are pending at once, some out-of-band * data may creep in... ick. */ - if (th->th_urp <= (u_long)tlen + if (th->th_urp <= (u_int32_t)tlen #if SO_OOBINLINE && (so->so_options & SO_OOBINLINE) == 0 #endif ) tcp_pulloutofband(so, th, m, drop_hdrlen); /* hdr drop is delayed */ - } else + } else { /* * If no out of band data is expected, * pull receive urgent pointer along @@ -2553,8 +3170,27 @@ step6: */ if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) tp->rcv_up = tp->rcv_nxt; -dodata: /* XXX */ + } +dodata: + + /* Set socket's connect or disconnect state correcly before doing data. + * The following might unlock the socket if there is an upcall or a socket + * filter. + */ + if (isconnected) { + soisconnected(so); + } else if (isdisconnected) { + soisdisconnected(so); + } + /* Let's check the state of pcb just to make sure that it did not get closed + * when we unlocked above + */ + if (inp->inp_state == INPCB_STATE_DEAD) { + /* Just drop the packet that we are processing and return */ + goto drop; + } + /* * Process the segment text, merging it into the TCP sequencing queue, * and arranging for acknowledgment of receipt if necessary. @@ -2563,24 +3199,32 @@ dodata: /* XXX */ * case PRU_RCVD). If a FIN has already been received on this * connection then we just ignore the text. */ - if ((tlen || (thflags&TH_FIN)) && + if ((tlen || (thflags & TH_FIN)) && TCPS_HAVERCVDFIN(tp->t_state) == 0) { + tcp_seq save_start = th->th_seq; + tcp_seq save_end = th->th_seq + tlen; m_adj(m, drop_hdrlen); /* delayed header drop */ /* - * Insert segment which inludes th into reassembly queue of tcp with - * control block tp. Return TH_FIN if reassembly now includes - * a segment with FIN. This handle the common case inline (segment - * is the next to be received on an established connection, and the - * queue is empty), avoiding linkage into and removal from the queue - * and repetition of various conversions. - * Set DELACK for segments received in order, but ack immediately - * when segments are out of order (so fast retransmit can work). + * Insert segment which includes th into TCP reassembly queue + * with control block tp. Set thflags to whether reassembly now + * includes a segment with FIN. This handles the common case + * inline (segment is the next to be received on an established + * connection, and the queue is empty), avoiding linkage into + * and removal from the queue and repetition of various + * conversions. + * Set DELACK for segments received in order, but ack + * immediately when segments are out of order (so + * fast retransmit can work). */ if (th->th_seq == tp->rcv_nxt && LIST_EMPTY(&tp->t_segq) && TCPS_HAVEESTABLISHED(tp->t_state)) { - if (DELAY_ACK(tp) && ((tp->t_flags & TF_ACKNOW) == 0)) { - tp->t_flags |= TF_DELACK; + if (DELAY_ACK(tp, th) && ((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); + } + tp->t_unacksegs++; } else { tp->t_flags |= TF_ACKNOW; @@ -2589,14 +3233,22 @@ dodata: /* XXX */ thflags = th->th_flags & TH_FIN; tcpstat.tcps_rcvpack++; tcpstat.tcps_rcvbyte += tlen; + if (nstat_collect) { + locked_add_64(&inp->inp_stat->rxpackets, 1); + locked_add_64(&inp->inp_stat->rxbytes, tlen); + } ND6_HINT(tp); - if (sbappend(&so->so_rcv, m)) + so_recv_data_stat(so, m, drop_hdrlen); + if (sbappendstream(&so->so_rcv, m)) sorwakeup(so); } else { thflags = tcp_reass(tp, th, &tlen, m); tp->t_flags |= TF_ACKNOW; } + if (tlen > 0 && tp->sack_enable) + tcp_update_sack_list(tp, save_start, save_end); + if (tp->t_flags & TF_DELACK) { #if INET6 @@ -2619,7 +3271,9 @@ dodata: /* XXX */ * our window, in order to estimate the sender's * buffer size. */ - len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); + len = (u_int)(so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt)); + if (len > so->so_rcv.sb_maxused) + so->so_rcv.sb_maxused = len; } else { m_freem(m); thflags &= ~TH_FIN; @@ -2636,12 +3290,18 @@ dodata: /* XXX */ /* * If connection is half-synchronized * (ie NEEDSYN flag on) then delay ACK, - * so it may be piggybacked when SYN is sent. - * Otherwise, since we received a FIN then no - * more input can be expected, send ACK now. + * If connection is half-synchronized + * (ie NEEDSYN flag on) then delay ACK, + * so it may be piggybacked when SYN is sent. + * Otherwise, since we received a FIN then no + * more input can be expected, send ACK now. */ - if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) { - tp->t_flags |= TF_DELACK; + 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); + } + tp->t_unacksegs++; } else { tp->t_flags |= TF_ACKNOW; @@ -2655,8 +3315,10 @@ dodata: /* XXX */ * enter the CLOSE_WAIT state. */ case TCPS_SYN_RECEIVED: - /*FALLTHROUGH*/ + tp->t_starttime = tcp_now; case TCPS_ESTABLISHED: + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_CLOSE_WAIT); tp->t_state = TCPS_CLOSE_WAIT; break; @@ -2665,6 +3327,8 @@ dodata: /* XXX */ * 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); tp->t_state = TCPS_CLOSING; break; @@ -2674,20 +3338,20 @@ dodata: /* XXX */ * standard timers. */ case TCPS_FIN_WAIT_2: + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_TIME_WAIT); tp->t_state = TCPS_TIME_WAIT; tcp_canceltimers(tp); /* Shorten TIME_WAIT [RFC-1644, p.28] */ if (tp->cc_recv != 0 && - tp->t_starttime < tcp_msl) { - tp->t_timer[TCPT_2MSL] = - tp->t_rxtcur * TCPTV_TWTRUNC; + ((int)(tcp_now - tp->t_starttime)) < tcp_msl) { + add_to_time_wait(tp, tp->t_rxtcur * TCPTV_TWTRUNC); /* For transaction client, force ACK now. */ tp->t_flags |= TF_ACKNOW; + tp->t_unacksegs = 0; } else - tp->t_timer[TCPT_2MSL] = 2 * tcp_msl; - - add_to_time_wait(tp); + add_to_time_wait(tp, 2 * tcp_msl); soisdisconnected(so); break; @@ -2695,8 +3359,7 @@ dodata: /* XXX */ * In TIME_WAIT state restart the 2 MSL time_wait timer. */ case TCPS_TIME_WAIT: - tp->t_timer[TCPT_2MSL] = 2 * tcp_msl; - add_to_time_wait(tp); + add_to_time_wait(tp, 2 * tcp_msl); break; } } @@ -2709,8 +3372,13 @@ dodata: /* XXX */ /* * Return any desired output. */ - if (needoutput || (tp->t_flags & TF_ACKNOW)) + if (needoutput || (tp->t_flags & TF_ACKNOW)) { (void) tcp_output(tp); + } + + tcp_check_timer_state(tp); + + tcp_unlock(so, 1, 0); KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); return; @@ -2745,6 +3413,8 @@ dropafterack: m_freem(m); tp->t_flags |= TF_ACKNOW; (void) tcp_output(tp); + + /* Don't need to check timer state as we should have done it during tcp_output */ tcp_unlock(so, 1, 0); KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); return; @@ -2788,22 +3458,22 @@ dropwithreset: 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); + TH_RST, ifscope, nocell); 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); + (tcp_seq)0, TH_RST|TH_ACK, ifscope, nocell); } /* destroy temporarily created socket */ if (dropsocket) { (void) soabort(so); tcp_unlock(so, 1, 0); } - else - if ((inp != NULL) && (nosock == 0)) - tcp_unlock(so, 1, 0); + else if ((inp != NULL) && (nosock == 0)) { + tcp_unlock(so, 1, 0); + } KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); return; dropnosock: @@ -2823,20 +3493,24 @@ drop: (void) soabort(so); tcp_unlock(so, 1, 0); } - else - if (nosock == 0) - tcp_unlock(so, 1, 0); + else if (nosock == 0) { + tcp_unlock(so, 1, 0); + } KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0); return; } static void -tcp_dooptions(tp, cp, cnt, th, to) +tcp_dooptions(tp, cp, cnt, th, to, input_ifscope) +/* + * Parse TCP options and place in tcpopt. + */ struct tcpcb *tp; u_char *cp; int cnt; struct tcphdr *th; struct tcpopt *to; + unsigned int input_ifscope; { u_short mss = 0; int opt, optlen; @@ -2865,7 +3539,11 @@ tcp_dooptions(tp, cp, cnt, th, to) if (!(th->th_flags & TH_SYN)) continue; bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); + +#if BYTE_ORDER != BIG_ENDIAN NTOHS(mss); +#endif + break; case TCPOPT_WINDOW: @@ -2880,13 +3558,20 @@ tcp_dooptions(tp, cp, cnt, th, to) case TCPOPT_TIMESTAMP: if (optlen != TCPOLEN_TIMESTAMP) continue; - to->to_flag |= TOF_TS; + to->to_flags |= TOF_TS; bcopy((char *)cp + 2, (char *)&to->to_tsval, sizeof(to->to_tsval)); + +#if BYTE_ORDER != BIG_ENDIAN NTOHL(to->to_tsval); +#endif + bcopy((char *)cp + 6, (char *)&to->to_tsecr, sizeof(to->to_tsecr)); + +#if BYTE_ORDER != BIG_ENDIAN NTOHL(to->to_tsecr); +#endif /* * A timestamp received in a SYN makes @@ -2898,49 +3583,25 @@ tcp_dooptions(tp, cp, cnt, th, to) tp->ts_recent_age = tcp_now; } break; - case TCPOPT_CC: - if (optlen != TCPOLEN_CC) + case TCPOPT_SACK_PERMITTED: + if (!tcp_do_sack || + optlen != TCPOLEN_SACK_PERMITTED) continue; - to->to_flag |= TOF_CC; - bcopy((char *)cp + 2, - (char *)&to->to_cc, sizeof(to->to_cc)); - NTOHL(to->to_cc); - /* - * A CC or CC.new option received in a SYN makes - * it ok to send CC in subsequent segments. - */ if (th->th_flags & TH_SYN) - tp->t_flags |= TF_RCVD_CC; + to->to_flags |= TOF_SACK; break; - case TCPOPT_CCNEW: - if (optlen != TCPOLEN_CC) - continue; - if (!(th->th_flags & TH_SYN)) + case TCPOPT_SACK: + if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) continue; - to->to_flag |= TOF_CCNEW; - bcopy((char *)cp + 2, - (char *)&to->to_cc, sizeof(to->to_cc)); - NTOHL(to->to_cc); - /* - * A CC or CC.new option received in a SYN makes - * it ok to send CC in subsequent segments. - */ - tp->t_flags |= TF_RCVD_CC; - break; - case TCPOPT_CCECHO: - if (optlen != TCPOLEN_CC) - continue; - if (!(th->th_flags & TH_SYN)) - continue; - to->to_flag |= TOF_CCECHO; - bcopy((char *)cp + 2, - (char *)&to->to_ccecho, sizeof(to->to_ccecho)); - NTOHL(to->to_ccecho); + to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; + to->to_sacks = cp + 2; + tcpstat.tcps_sack_rcv_blocks++; + break; } } if (th->th_flags & TH_SYN) - tcp_mss(tp, mss); /* sets t_maxseg */ + tcp_mss(tp, mss, input_ifscope); /* sets t_maxseg */ } /* @@ -2979,6 +3640,36 @@ tcp_pulloutofband(so, th, m, off) panic("tcp_pulloutofband"); } +uint32_t +get_base_rtt(struct tcpcb *tp) +{ + uint32_t base_rtt = 0, i; + for (i = 0; i < N_RTT_BASE; ++i) { + if (tp->rtt_hist[i] != 0 && + (base_rtt == 0 || tp->rtt_hist[i] < base_rtt)) + base_rtt = tp->rtt_hist[i]; + } + return base_rtt; +} + +/* Each value of RTT base represents the minimum RTT seen in a minute. + * We keep upto N_RTT_BASE minutes worth of history. + */ +void +update_base_rtt(struct tcpcb *tp, uint32_t rtt) +{ + if (++tp->rtt_count >= rtt_samples_per_slot) { + int i=0; + for (i = (N_RTT_BASE-1); i > 0; --i) { + tp->rtt_hist[i] = tp->rtt_hist[i-1]; + } + tp->rtt_hist[0] = rtt; + tp->rtt_count = 0; + } else { + tp->rtt_hist[0] = min(tp->rtt_hist[0], rtt); + } +} + /* * Collect new round-trip time estimate * and update averages and current timeout. @@ -2992,15 +3683,26 @@ tcp_xmit_timer(tp, rtt) tcpstat.tcps_rttupdated++; tp->t_rttupdated++; + + if (rtt > 0) { + tp->t_rttcur = rtt; + update_base_rtt(tp, rtt); + } + if (tp->t_srtt != 0) { /* * srtt is stored as fixed point with 5 bits after the - * binary point (i.e., scaled by 8). The following magic + * binary point (i.e., scaled by 32). The following magic * is equivalent to the smoothing algorithm in rfc793 with * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed - * point). Adjust rtt to origin 0. + * point). + * + * Freebsd adjusts rtt to origin 0 by subtracting 1 from the provided + * rtt value. This was required because of the way t_rtttime was + * initiailised to 1 before. Since we changed t_rtttime to be based on + * tcp_now, this extra adjustment is not needed. */ - delta = ((rtt - 1) << TCP_DELTA_SHIFT) + delta = (rtt << TCP_DELTA_SHIFT) - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); if ((tp->t_srtt += delta) <= 0) @@ -3021,6 +3723,8 @@ tcp_xmit_timer(tp, rtt) delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); if ((tp->t_rttvar += delta) <= 0) tp->t_rttvar = 1; + if (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. @@ -3029,9 +3733,12 @@ tcp_xmit_timer(tp, 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; } + nstat_route_rtt(tp->t_inpcb->inp_route.ro_rt, tp->t_srtt, tp->t_rttvar); tp->t_rtttime = 0; tp->t_rxtshift = 0; + tp->rxt_start = 0; /* * the retransmit should happen at rtt + 4 * rttvar. @@ -3045,7 +3752,8 @@ tcp_xmit_timer(tp, rtt) * the minimum feasible timer (which is 2 ticks). */ TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), - max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); + max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX, + TCP_ADD_REXMTSLOP(tp)); /* * We received an ack for a packet that wasn't retransmitted; @@ -3057,6 +3765,38 @@ tcp_xmit_timer(tp, rtt) tp->t_softerror = 0; } +static inline unsigned int +tcp_maxmtu(struct rtentry *rt) +{ + unsigned int maxmtu; + + 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); + + return (maxmtu); +} + +#if INET6 +static inline unsigned int +tcp_maxmtu6(struct rtentry *rt) +{ + unsigned int maxmtu; + + RT_LOCK_ASSERT_HELD(rt); + lck_rw_lock_shared(nd_if_rwlock); + if (rt->rt_rmx.rmx_mtu == 0) + maxmtu = IN6_LINKMTU(rt->rt_ifp); + else + maxmtu = MIN(rt->rt_rmx.rmx_mtu, IN6_LINKMTU(rt->rt_ifp)); + lck_rw_done(nd_if_rwlock); + + return (maxmtu); +} +#endif + /* * Determine a reasonable value for maxseg size. * If the route is known, check route for mtu. @@ -3081,23 +3821,23 @@ tcp_xmit_timer(tp, rtt) * NOTE that this routine is only called when we process an incoming * segment, for outgoing segments only tcp_mssopt is called. * - * In case of T/TCP, we call this routine during implicit connection - * setup as well (offer = -1), to initialize maxseg from the cached - * MSS of our peer. */ void -tcp_mss(tp, offer) +tcp_mss(tp, offer, input_ifscope) struct tcpcb *tp; int offer; + unsigned int input_ifscope; { register struct rtentry *rt; struct ifnet *ifp; register int rtt, mss; - u_long bufsize; + 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; @@ -3111,12 +3851,31 @@ tcp_mss(tp, offer) #else #define min_protoh (sizeof (struct tcpiphdr)) #endif + #if INET6 - if (isipv6) - rt = tcp_rtlookup6(inp); + if (isipv6) { + rt = tcp_rtlookup6(inp, input_ifscope); + if (rt != NULL && + (IN6_IS_ADDR_LOOPBACK(&inp->in6p_faddr) || + IN6_IS_ADDR_LINKLOCAL(&inp->in6p_faddr) || + rt->rt_gateway->sa_family == AF_LINK || + in6_localaddr(&inp->in6p_faddr))) { + tp->t_flags |= TF_LOCAL; + } + } else #endif /* INET6 */ - rt = tcp_rtlookup(inp); + { + rt = tcp_rtlookup(inp, input_ifscope); + if (rt != NULL && + (rt->rt_gateway->sa_family == AF_LINK || + rt->rt_ifp->if_flags & IFF_LOOPBACK || + in_localaddr(inp->inp_faddr))) { + tp->t_flags |= TF_LOCAL; + } + } + isnetlocal = (tp->t_flags & TF_LOCAL); + if (rt == NULL) { tp->t_maxopd = tp->t_maxseg = #if INET6 @@ -3182,12 +3941,14 @@ tcp_mss(tp, offer) * is also a minimum value; this is subject to time. */ if (rt->rt_rmx.rmx_locks & RTV_RTT) - tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); - tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); + tp->t_rttmin = rtt / (RTM_RTTUNIT / TCP_RETRANSHZ); + else + tp->t_rttmin = isnetlocal ? tcp_TCPTV_MIN : TCPTV_REXMTMIN; + tp->t_srtt = rtt / (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE)); tcpstat.tcps_usedrtt++; if (rt->rt_rmx.rmx_rttvar) { tp->t_rttvar = rt->rt_rmx.rmx_rttvar / - (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); + (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE)); tcpstat.tcps_usedrttvar++; } else { /* default variation is +- 1 rtt */ @@ -3196,34 +3957,30 @@ tcp_mss(tp, offer) } TCPT_RANGESET(tp->t_rxtcur, ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, - tp->t_rttmin, TCPTV_REXMTMAX); + tp->t_rttmin, TCPTV_REXMTMAX, + TCP_ADD_REXMTSLOP(tp)); } - /* - * if there's an mtu associated with the route, use it - * else, use the link mtu. - */ - if (rt->rt_rmx.rmx_mtu) - mss = rt->rt_rmx.rmx_mtu - min_protoh; else - { - mss = -#if INET6 - (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : -#endif - ifp->if_mtu + 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 - - min_protoh; + mss -= min_protoh; + + if (rt->rt_rmx.rmx_mtu == 0) { #if INET6 if (isipv6) { - if (!in6_localaddr(&inp->in6p_faddr)) + if (!isnetlocal) mss = min(mss, tcp_v6mssdflt); } else #endif /* INET6 */ - if (!in_localaddr(inp->inp_faddr)) + if (!isnetlocal) mss = min(mss, tcp_mssdflt); } + mss = min(mss, offer); /* * maxopd stores the maximum length of data AND options @@ -3235,18 +3992,20 @@ tcp_mss(tp, offer) tp->t_maxopd = mss; /* - * In case of T/TCP, origoffer==-1 indicates, that no segments - * were received yet. In this case we just guess, otherwise - * we do the same as before T/TCP. + * 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)) mss -= TCPOLEN_TSTAMP_APPA; - if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && - (origoffer == -1 || - (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) - mss -= TCPOLEN_CC_APPA; + tp->t_maxseg = mss; + + /* + * Calculate corrected value for sb_max; ensure to upgrade the + * numerator for large sb_max values else it will overflow. + */ + sb_max_corrected = (sb_max * (u_int64_t)MCLBYTES) / (MSIZE + MCLBYTES); /* * If there's a pipesize (ie loopback), change the socket @@ -3263,9 +4022,9 @@ tcp_mss(tp, offer) if (bufsize < mss) mss = bufsize; else { - bufsize = roundup(bufsize, mss); - if (bufsize > sb_max) - bufsize = sb_max; + bufsize = (((bufsize + (u_int64_t)mss - 1) / (u_int64_t)mss) * (u_int64_t)mss); + if (bufsize > sb_max_corrected) + bufsize = sb_max_corrected; (void)sbreserve(&so->so_snd, bufsize); } tp->t_maxseg = mss; @@ -3276,29 +4035,13 @@ tcp_mss(tp, offer) #endif bufsize = so->so_rcv.sb_hiwat; if (bufsize > mss) { - bufsize = roundup(bufsize, mss); - if (bufsize > sb_max) - bufsize = sb_max; + bufsize = (((bufsize + (u_int64_t)mss - 1) / (u_int64_t)mss) * (u_int64_t)mss); + if (bufsize > sb_max_corrected) + bufsize = sb_max_corrected; (void)sbreserve(&so->so_rcv, bufsize); } - /* - * Set the slow-start flight size depending on whether this - * is a local network or not. - */ - if ( -#if INET6 - (isipv6 && in6_localaddr(&inp->in6p_faddr)) || - (!isipv6 && -#endif - in_localaddr(inp->inp_faddr) -#if INET6 - ) -#endif - ) - tp->snd_cwnd = mss * ss_fltsz_local; - else - tp->snd_cwnd = mss * ss_fltsz; + set_tcp_stream_priority(so); if (rt->rt_rmx.rmx_ssthresh) { /* @@ -3309,7 +4052,23 @@ tcp_mss(tp, offer) */ tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); tcpstat.tcps_usedssthresh++; + } else { + tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; } + + + /* + * Set the slow-start flight size depending on whether this + * is a local network or not. + */ + if (CC_ALGO(tp)->cwnd_init != NULL) + CC_ALGO(tp)->cwnd_init(tp); + + DTRACE_TCP5(cc, void, NULL, struct inpcb *, tp->t_inpcb, struct tcpcb *, tp, + struct tcphdr *, NULL, int32_t, TCP_CC_CWND_INIT); + + /* Route locked during lookup above */ + RT_UNLOCK(rt); } /* @@ -3320,6 +4079,7 @@ tcp_mssopt(tp) struct tcpcb *tp; { struct rtentry *rt; + int mss; #if INET6 int isipv6; int min_protoh; @@ -3332,18 +4092,20 @@ tcp_mssopt(tp) #else #define min_protoh (sizeof (struct tcpiphdr)) #endif + #if INET6 if (isipv6) - rt = tcp_rtlookup6(tp->t_inpcb); + rt = tcp_rtlookup6(tp->t_inpcb, IFSCOPE_NONE); else #endif /* INET6 */ - rt = tcp_rtlookup(tp->t_inpcb); - if (rt == NULL) - return + rt = tcp_rtlookup(tp->t_inpcb, IFSCOPE_NONE); + if (rt == NULL) { + return ( #if INET6 isipv6 ? tcp_v6mssdflt : #endif /* INET6 */ - tcp_mssdflt; + tcp_mssdflt); + } /* * Slower link window correction: * If a value is specificied for slowlink_wsize use it for PPP links @@ -3355,29 +4117,30 @@ tcp_mssopt(tp) tp->t_flags |= TF_SLOWLINK; } - return rt->rt_ifp->if_mtu - min_protoh; +#if INET6 + mss = (isipv6 ? tcp_maxmtu6(rt) : tcp_maxmtu(rt)); +#else + mss = tcp_maxmtu(rt); +#endif + /* Route locked during lookup above */ + RT_UNLOCK(rt); + return (mss - min_protoh); } - /* - * Checks for partial ack. If partial ack arrives, force the retransmission - * of the next unacknowledged segment, do not clear tp->t_dupacks, and return - * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to - * be started again. If the ack advances at least to tp->snd_recover, return 0. + * On a partial ack arrives, force the retransmission of the + * next unacknowledged segment. Do not clear tp->t_dupacks. + * By setting snd_nxt to th_ack, this forces retransmission timer to + * be started again. */ -static int -tcp_newreno(tp, th) +static void +tcp_newreno_partial_ack(tp, th) struct tcpcb *tp; struct tcphdr *th; { - if (SEQ_LT(th->th_ack, tp->snd_recover)) { tcp_seq onxt = tp->snd_nxt; - u_long ocwnd = tp->snd_cwnd; -#ifdef __APPLE__ + u_int32_t ocwnd = tp->snd_cwnd; tp->t_timer[TCPT_REXMT] = 0; -#else - callout_stop(tp->tt_rexmt); -#endif tp->t_rtttime = 0; tp->snd_nxt = th->th_ack; /* @@ -3395,10 +4158,12 @@ tcp_newreno(tp, th) * Partial window deflation. Relies on fact that tp->snd_una * not updated yet. */ - tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); - return (1); - } - return (0); + if (tp->snd_cwnd > th->th_ack - tp->snd_una) + tp->snd_cwnd -= th->th_ack - tp->snd_una; + else + tp->snd_cwnd = 0; + tp->snd_cwnd += tp->t_maxseg; + } /* @@ -3411,9 +4176,9 @@ tcp_newreno(tp, th) * The listening TCP socket "head" must be locked */ static int -tcpdropdropablreq(struct socket *head) +tcp_dropdropablreq(struct socket *head) { - struct socket *so; + struct socket *so, *sonext; unsigned int i, j, qlen; static int rnd; static struct timeval old_runtime; @@ -3421,7 +4186,14 @@ tcpdropdropablreq(struct socket *head) struct timeval tv; struct inpcb *inp = NULL; struct tcpcb *tp; - + + if ((head->so_options & SO_ACCEPTCONN) == 0) + return 0; + + so = TAILQ_FIRST(&head->so_incomp); + if (!so) + return 0; + microtime(&tv); if ((i = (tv.tv_sec - old_runtime.tv_sec)) != 0) { old_runtime = tv; @@ -3429,9 +4201,6 @@ tcpdropdropablreq(struct socket *head) cur_cnt = 0; } - so = TAILQ_FIRST(&head->so_incomp); - if (!so) - return 0; qlen = head->so_incqlen; if (++cur_cnt > qlen || old_cnt > qlen) { @@ -3441,51 +4210,202 @@ tcpdropdropablreq(struct socket *head) while (j-- && so) so = TAILQ_NEXT(so, so_list); } - /* Find a connection that is not already closing */ + /* Find a connection that is not already closing (or being served) */ while (so) { inp = (struct inpcb *)so->so_pcb; - if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) != WNT_STOPUSING) - break; + sonext = TAILQ_NEXT(so, so_list); + + if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) != WNT_STOPUSING) { + /* Avoid the issue of a socket being accepted by one input thread + * and being dropped by another input thread. + * If we can't get a hold on this mutex, then grab the next socket in line. + */ + if (lck_mtx_try_lock(&inp->inpcb_mtx)) { + so->so_usecount++; + if ((so->so_usecount == 2) && + (so->so_state & SS_INCOMP) != 0 && + (so->so_flags & SOF_INCOMP_INPROGRESS) == 0) + break; + else {/* don't use if being accepted or used in any other way */ + in_pcb_checkstate(inp, WNT_RELEASE, 1); + tcp_unlock(so, 1, 0); + } + } + else { + /* do not try to lock the inp in in_pcb_checkstate + * because the lock is already held in some other thread. + * Only drop the inp_wntcnt reference. + */ + in_pcb_checkstate(inp, WNT_RELEASE, 1); + } + } + so = sonext; - so = TAILQ_NEXT(so, so_list); } if (!so) return 0; - - /* Let's remove this connection from the incomplete list */ - tcp_lock(so, 1, 0); - + + /* Makes sure socket is still in the right state to be discarded */ + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { tcp_unlock(so, 1, 0); return 0; } - head->so_incqlen--; - head->so_qlen--; + + if (so->so_usecount != 2 || !(so->so_state & SS_INCOMP)) { + /* do not discard: that socket is being accepted */ + tcp_unlock(so, 1, 0); + return 0; + } + TAILQ_REMOVE(&head->so_incomp, so, so_list); tcp_unlock(head, 0, 0); - - so->so_head = NULL; - so->so_usecount--; /* No more held by so_head */ - /* - * We do not want to lose track of the PCB right away in case we receive - * more segments from the peer - */ + lck_mtx_assert(&inp->inpcb_mtx, LCK_MTX_ASSERT_OWNED); tp = sototcpcb(so); - tp->t_flags |= TF_LQ_OVERFLOW; - tp->t_state = TCPS_CLOSED; - (void) tcp_output(tp); + so->so_flags |= SOF_OVERFLOW; + so->so_head = NULL; + + tcp_close(tp); + tp->t_unacksegs = 0; + + if (inp->inp_wantcnt > 0 && inp->inp_wantcnt != WNT_STOPUSING) { + /* Some one has a wantcnt on this pcb. Since WNT_ACQUIRE + * doesn't require a lock, it could have happened while + * we are holding the lock. This pcb will have to + * be garbage collected later. + * Release the reference held for so_incomp queue + */ + so->so_usecount--; + + tcp_unlock(so, 1, 0); + } else { + /* Unlock this socket and leave the reference on. We need to + * acquire the pcbinfo lock in order to fully dispose it off + */ + tcp_unlock(so, 0, 0); + + lck_rw_lock_exclusive(tcbinfo.mtx); + + tcp_lock(so, 0, 0); + + /* Release the reference held for so_incomp queue */ + so->so_usecount--; + + if (so->so_usecount != 1 || + (inp->inp_wantcnt > 0 && inp->inp_wantcnt != WNT_STOPUSING)) { + /* There is an extra wantcount or usecount that must + * have been added when the socket was unlocked. This + * socket will have to be garbage collected later + */ + tcp_unlock(so, 1, 0); + } else { + + /* Drop the reference held for this function */ + so->so_usecount--; + + in_pcbdispose(inp); + } + lck_rw_done(tcbinfo.mtx); + } tcpstat.tcps_drops++; - soisdisconnected(so); - tcp_canceltimers(tp); - add_to_time_wait(tp); - - tcp_unlock(so, 1, 0); + tcp_lock(head, 0, 0); - - return 1; - + head->so_incqlen--; + head->so_qlen--; + return(1); +} + +/* Set background congestion control on a socket */ +void +tcp_set_background_cc(struct socket *so) +{ + tcp_set_new_cc(so, TCP_CC_ALGO_BACKGROUND_INDEX); +} + +/* Set foreground congestion control on a socket */ +void +tcp_set_foreground_cc(struct socket *so) +{ + tcp_set_new_cc(so, TCP_CC_ALGO_NEWRENO_INDEX); +} + +static void +tcp_set_new_cc(struct socket *so, uint16_t cc_index) +{ + struct inpcb *inp = sotoinpcb(so); + struct tcpcb *tp = intotcpcb(inp); + uint16_t old_cc_index = 0; + if (tp->tcp_cc_index != cc_index) { + + old_cc_index = tp->tcp_cc_index; + + if (CC_ALGO(tp)->cleanup != NULL) + CC_ALGO(tp)->cleanup(tp); + tp->tcp_cc_index = cc_index; + + /* Decide if the connection is just starting or if + * we have sent some packets on it. + */ + if (tp->snd_nxt > tp->iss) { + /* Already sent some packets */ + if (CC_ALGO(tp)->switch_to != NULL) + CC_ALGO(tp)->switch_to(tp, old_cc_index); + } else { + if (CC_ALGO(tp)->init != NULL) + CC_ALGO(tp)->init(tp); + } + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, NULL, + int32_t, TCP_CC_CHANGE_ALGO); + } } +static int +tcp_getstat SYSCTL_HANDLER_ARGS +{ +#pragma unused(oidp, arg1, arg2) + + int error; + + if (req->oldptr == 0) { + req->oldlen= (size_t)sizeof(struct tcpstat); + } + + error = SYSCTL_OUT(req, &tcpstat, MIN(sizeof (tcpstat), req->oldlen)); + + return (error); + +} + +SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, + tcp_getstat, "S,tcpstat", "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); + +static int +sysctl_rexmtthresh SYSCTL_HANDLER_ARGS +{ +#pragma unused(arg1, arg2) + + int error, val = tcprexmtthresh; + + error = sysctl_handle_int(oidp, &val, 0, req); + if (error || !req->newptr) + return (error); + + /* + * Constrain the number of duplicate ACKs + * to consider for TCP fast retransmit + * to either 2 or 3 + */ + + if (val < 2 || val > 3) + return (EINVAL); + + tcprexmtthresh = val; + + 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");