X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/2d21ac55c334faf3a56e5634905ed6987fc787d4..99c3a10404e5d1ef94397ab4df5a8b74711fc4d3:/bsd/netinet/tcp_output.c diff --git a/bsd/netinet/tcp_output.c b/bsd/netinet/tcp_output.c index 250d4a2d6..83a205319 100644 --- a/bsd/netinet/tcp_output.c +++ b/bsd/netinet/tcp_output.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2012 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -81,7 +81,10 @@ #include #include +#include #include +#include +#include #include #include @@ -89,6 +92,7 @@ #include #include #include +#include #if INET6 #include #include @@ -101,10 +105,12 @@ #include #include #include +#include #if TCPDEBUG #include #endif #include +#include #if IPSEC #include @@ -114,51 +120,80 @@ #include #endif /* MAC_SOCKET */ +#include + #define DBG_LAYER_BEG NETDBG_CODE(DBG_NETTCP, 1) #define DBG_LAYER_END NETDBG_CODE(DBG_NETTCP, 3) #define DBG_FNC_TCP_OUTPUT NETDBG_CODE(DBG_NETTCP, (4 << 8) | 1) - #ifdef notyet extern struct mbuf *m_copypack(); #endif int path_mtu_discovery = 1; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW | CTLFLAG_LOCKED, &path_mtu_discovery, 1, "Enable Path MTU Discovery"); int ss_fltsz = 1; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW | CTLFLAG_LOCKED, &ss_fltsz, 1, "Slow start flight size"); int ss_fltsz_local = 8; /* starts with eight segments max */ -SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW | CTLFLAG_LOCKED, &ss_fltsz_local, 1, "Slow start flight size for local networks"); -int tcp_do_newreno = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno, CTLFLAG_RW, &tcp_do_newreno, - 0, "Enable NewReno Algorithms"); +int tcp_do_tso = 1; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_do_tso, 0, "Enable TCP Segmentation Offload"); + int tcp_ecn_outbound = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, ecn_initiate_out, CTLFLAG_RW, &tcp_ecn_outbound, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, ecn_initiate_out, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_ecn_outbound, 0, "Initiate ECN for outbound connections"); int tcp_ecn_inbound = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, ecn_negotiate_in, CTLFLAG_RW, &tcp_ecn_inbound, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, ecn_negotiate_in, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_ecn_inbound, 0, "Allow ECN negotiation for inbound connections"); int tcp_packet_chaining = 50; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, packetchain, CTLFLAG_RW, &tcp_packet_chaining, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, packetchain, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_packet_chaining, 0, "Enable TCP output packet chaining"); int tcp_output_unlocked = 1; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, socket_unlocked_on_output, CTLFLAG_RW, &tcp_output_unlocked, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, socket_unlocked_on_output, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_output_unlocked, 0, "Unlock TCP when sending packets down to IP"); -static long packchain_newlist = 0; -static long packchain_looped = 0; -static long packchain_sent = 0; +int tcp_do_rfc3390 = 1; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_do_rfc3390, 1, "Calculate intial slowstart cwnd depending on MSS"); + +int tcp_min_iaj_win = MIN_IAJ_WIN; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, min_iaj_win, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_min_iaj_win, 1, "Minimum recv win based on inter-packet arrival jitter"); + +int tcp_acc_iaj_react_limit = ACC_IAJ_REACT_LIMIT; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, acc_iaj_react_limit, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_acc_iaj_react_limit, 1, "Accumulated IAJ when receiver starts to react"); + +uint32_t tcp_do_autosendbuf = 1; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, doautosndbuf, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_do_autosendbuf, 1, "Enable send socket buffer auto-tuning"); +uint32_t tcp_autosndbuf_inc = 8 * 1024; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, autosndbufinc, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_autosndbuf_inc, 1, "Increment in send socket bufffer size"); + +uint32_t tcp_autosndbuf_max = 512 * 1024; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, autosndbufmax, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_autosndbuf_max, 1, "Maximum send socket buffer size"); + +uint32_t tcp_prioritize_acks = 1; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, ack_prioritize, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_prioritize_acks, 1, "Prioritize pure acks"); + +static int32_t packchain_newlist = 0; +static int32_t packchain_looped = 0; +static int32_t packchain_sent = 0; /* temporary: for testing */ #if IPSEC @@ -166,14 +201,25 @@ extern int ipsec_bypass; #endif extern int slowlink_wsize; /* window correction for slow links */ -extern u_long route_generation; +#if IPFIREWALL extern int fw_enable; /* firewall check for packet chaining */ extern int fw_bypass; /* firewall check: disable packet chaining if there is rules */ +#endif /* IPFIREWALL */ extern vm_size_t so_cache_zone_element_size; +#if RANDOM_IP_ID +extern int ip_use_randomid; +#endif /* RANDOM_IP_ID */ +extern u_int32_t dlil_filter_count; +extern u_int32_t kipf_count; +extern int tcp_recv_bg; +extern int maxseg_unacked; static int tcp_ip_output(struct socket *, struct tcpcb *, struct mbuf *, int, - struct mbuf *, int); + struct mbuf *, int, int, int32_t, boolean_t); + +extern uint32_t get_base_rtt(struct tcpcb *tp); +static struct mbuf* tcp_send_lroacks(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th); static __inline__ u_int16_t get_socket_id(struct socket * s) @@ -183,7 +229,7 @@ get_socket_id(struct socket * s) if (so_cache_zone_element_size == 0) { return (0); } - val = (u_int16_t)(((u_int32_t)s) / so_cache_zone_element_size); + val = (u_int16_t)(((uintptr_t)s) / so_cache_zone_element_size); if (val == 0) { val = 0xffff; } @@ -207,14 +253,21 @@ get_socket_id(struct socket * s) * ip_output_list:EMSGSIZE * ip_output_list:ENOBUFS * ip_output_list:??? [ignorable: mostly IPSEC/firewall/DLIL] - * ip6_output:??? [IPV6 only] + * ip6_output_list:EINVAL + * ip6_output_list:EOPNOTSUPP + * ip6_output_list:EHOSTUNREACH + * ip6_output_list:EADDRNOTAVAIL + * ip6_output_list:ENETUNREACH + * ip6_output_list:EMSGSIZE + * ip6_output_list:ENOBUFS + * ip6_output_list:??? [ignorable: mostly IPSEC/firewall/DLIL] */ int tcp_output(struct tcpcb *tp) { struct socket *so = tp->t_inpcb->inp_socket; - long len, recwin, sendwin; - int off, flags, error; + int32_t len, recwin, sendwin, off; + int flags, error; register struct mbuf *m; struct ip *ip = NULL; register struct ipovly *ipov = NULL; @@ -226,24 +279,30 @@ tcp_output(struct tcpcb *tp) unsigned ipoptlen, optlen, hdrlen; int idle, sendalot, lost = 0; int i, sack_rxmit; + int tso = 0; int sack_bytes_rxmt; struct sackhole *p; - - int maxburst = TCP_MAXBURST; +#ifdef IPSEC + unsigned ipsec_optlen = 0; +#endif int last_off = 0; int m_off; - struct mbuf *m_last = NULL; + int idle_time = 0; + struct mbuf *m_lastm = NULL; struct mbuf *m_head = NULL; struct mbuf *packetlist = NULL; struct mbuf *tp_inp_options = tp->t_inpcb->inp_depend4.inp4_options; #if INET6 int isipv6 = tp->t_inpcb->inp_vflag & INP_IPV6 ; - struct ip6_pktopts *inp6_pktopts = tp->t_inpcb->inp_depend6.inp6_outputopts; #endif short packchain_listadd = 0; u_int16_t socket_id = get_socket_id(so); int so_options = so->so_options; struct rtentry *rt; + u_int32_t basertt, svc_flags = 0, allocated_len; + u_int32_t lro_ackmore = (tp->t_lropktlen != 0) ? 1 : 0; + struct mbuf *mnext = NULL; + int sackoptlen = 0; /* * Determine length of data that should be transmitted, @@ -252,28 +311,17 @@ tcp_output(struct tcpcb *tp) * to send, then transmit; otherwise, investigate further. */ idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una); - if (idle && tp->t_rcvtime >= tp->t_rxtcur) { - /* - * We have been idle for "a while" and no acks are - * expected to clock out any data we send -- - * slow start to get ack "clock" running again. - * - * Set the slow-start flight size depending on whether - * this is a local network or not. - */ - if ( -#if INET6 - (isipv6 && in6_localaddr(&tp->t_inpcb->in6p_faddr)) || - (!isipv6 && -#endif - in_localaddr(tp->t_inpcb->inp_faddr) -#if INET6 - ) -#endif - ) - tp->snd_cwnd = tp->t_maxseg * ss_fltsz_local; - else - tp->snd_cwnd = tp->t_maxseg * ss_fltsz; + + /* Since idle_time is signed integer, the following integer subtraction + * will take care of wrap around of tcp_now + */ + idle_time = tcp_now - tp->t_rcvtime; + if (idle && idle_time >= TCP_IDLETIMEOUT(tp)) { + if (CC_ALGO(tp)->after_idle != NULL) + CC_ALGO(tp)->after_idle(tp); + DTRACE_TCP5(cc, void, NULL, struct inpcb *, tp->t_inpcb, + struct tcpcb *, tp, struct tcphdr *, NULL, + int32_t, TCP_CC_IDLE_TIMEOUT); } tp->t_flags &= ~TF_LASTIDLE; if (idle) { @@ -287,14 +335,12 @@ again: #if INET6 if (isipv6) { - KERNEL_DEBUG(DBG_LAYER_BEG, ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport), (((tp->t_inpcb->in6p_laddr.s6_addr16[0] & 0xffff) << 16) | (tp->t_inpcb->in6p_faddr.s6_addr16[0] & 0xffff)), sendalot,0,0); - } - else + } else #endif { @@ -303,6 +349,7 @@ again: (((tp->t_inpcb->inp_laddr.s_addr & 0xffff) << 16) | (tp->t_inpcb->inp_faddr.s_addr & 0xffff)), sendalot,0,0); + } /* * If the route generation id changed, we need to check that our * local (source) IP address is still valid. If it isn't either @@ -310,14 +357,34 @@ again: * come back before the TCP connection times out). */ rt = tp->t_inpcb->inp_route.ro_rt; - if (rt != NULL && rt->generation_id != route_generation) { + if (rt != NULL && (!(rt->rt_flags & RTF_UP) || + rt->generation_id != route_generation)) { struct ifnet *ifp; + struct in_ifaddr *ia = NULL; + struct in6_ifaddr *ia6 = NULL; + int found_srcaddr = 0; /* disable multipages at the socket */ somultipages(so, FALSE); + /* Disable TSO for the socket until we know more */ + tp->t_flags &= ~TF_TSO; + + if (isipv6) { + ia6 = ifa_foraddr6(&tp->t_inpcb->in6p_laddr); + if (ia6 != NULL) + found_srcaddr = 1; + } else { + ia = ifa_foraddr(tp->t_inpcb->inp_laddr.s_addr); + if (ia != NULL) + found_srcaddr = 1; + } + /* check that the source address is still valid */ - if (ifa_foraddr(tp->t_inpcb->inp_laddr.s_addr) == 0) { + if (found_srcaddr == 0) { + + soevent(so, + (SO_FILT_HINT_LOCKED | SO_FILT_HINT_NOSRCADDR)); if (tp->t_state >= TCPS_CLOSE_WAIT) { tcp_drop(tp, EADDRNOTAVAIL); @@ -326,14 +393,16 @@ again: /* set Retransmit timer if it wasn't set * reset Persist timer and shift register as the - * adversed peer window may not be valid anymore + * advertised peer window may not be valid anymore */ if (!tp->t_timer[TCPT_REXMT]) { - tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; + tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp, tp->t_rxtcur); if (tp->t_timer[TCPT_PERSIST]) { tp->t_timer[TCPT_PERSIST] = 0; tp->t_rxtshift = 0; + tp->t_persist_stop = 0; + tp->rxt_start = 0; } } @@ -346,19 +415,27 @@ again: tcp_drop(tp, EADDRNOTAVAIL); return(EADDRNOTAVAIL); } - else + else { + tcp_check_timer_state(tp); return(0); /* silently ignore, keep data in socket: address may be back */ + } } + if (ia != NULL) + IFA_REMREF(&ia->ia_ifa); + + if (ia6 != NULL) + IFA_REMREF(&ia6->ia_ifa); /* * Address is still valid; check for multipages capability * again in case the outgoing interface has changed. */ - lck_mtx_lock(rt_mtx); - rt = tp->t_inpcb->inp_route.ro_rt; - if (rt != NULL && (ifp = rt->rt_ifp) != NULL) + RT_LOCK(rt); + if ((ifp = rt->rt_ifp) != NULL) { somultipages(so, (ifp->if_hwassist & IFNET_MULTIPAGES)); - if (rt != NULL && rt->generation_id != route_generation) + tcp_set_tso(tp, ifp); + } + if (rt->rt_flags & RTF_UP) rt->generation_id = route_generation; /* * See if we should do MTU discovery. Don't do it if: @@ -374,8 +451,7 @@ again: else tp->t_flags |= TF_PMTUD; - lck_mtx_unlock(rt_mtx); - } + RT_UNLOCK(rt); } /* @@ -409,7 +485,7 @@ again: p = NULL; if (tp->sack_enable && IN_FASTRECOVERY(tp) && (p = tcp_sack_output(tp, &sack_bytes_rxmt))) { - long cwin; + int32_t cwin; cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt; if (cwin < 0) @@ -432,17 +508,28 @@ again: goto after_sack_rexmit; } else /* Can rexmit part of the current hole */ - len = ((long)ulmin(cwin, + len = ((int32_t)min(cwin, tp->snd_recover - p->rxmit)); - } else - len = ((long)ulmin(cwin, p->end - p->rxmit)); - off = p->rxmit - tp->snd_una; + } else { + len = ((int32_t)min(cwin, p->end - p->rxmit)); + } if (len > 0) { + off = p->rxmit - tp->snd_una; /* update off only if we really transmit SACK data */ sack_rxmit = 1; sendalot = 1; tcpstat.tcps_sack_rexmits++; tcpstat.tcps_sack_rexmit_bytes += min(len, tp->t_maxseg); + if (nstat_collect) { + nstat_route_tx(tp->t_inpcb->inp_route.ro_rt, 1, + min(len, tp->t_maxseg), NSTAT_TX_FLAG_RETRANSMIT); + locked_add_64(&tp->t_inpcb->inp_stat->txpackets, 1); + locked_add_64(&tp->t_inpcb->inp_stat->txbytes, + min(len, tp->t_maxseg)); + tp->t_stat.txretransmitbytes += min(len, tp->t_maxseg); + } + } else { + len = 0; } } after_sack_rexmit: @@ -485,6 +572,8 @@ after_sack_rexmit: } else { tp->t_timer[TCPT_PERSIST] = 0; tp->t_rxtshift = 0; + tp->rxt_start = 0; + tp->t_persist_stop = 0; } } @@ -505,17 +594,17 @@ after_sack_rexmit: */ if (sack_rxmit == 0) { if (sack_bytes_rxmt == 0) - len = ((long)ulmin(so->so_snd.sb_cc, sendwin) - off); + len = min(so->so_snd.sb_cc, sendwin) - off; else { - long cwin; + int32_t cwin; /* * We are inside of a SACK recovery episode and are * sending new data, having retransmitted all the * data possible in the scoreboard. */ - len = ((long)ulmin(so->so_snd.sb_cc, tp->snd_wnd) - - off); + len = min(so->so_snd.sb_cc, tp->snd_wnd) + - off; /* * Don't remove this (len > 0) check ! * We explicitly check for len > 0 here (although it @@ -530,8 +619,10 @@ after_sack_rexmit: sack_bytes_rxmt; if (cwin < 0) cwin = 0; - len = lmin(len, cwin); + len = imin(len, cwin); } + else + len = 0; } } @@ -544,29 +635,39 @@ after_sack_rexmit: flags &= ~TH_SYN; off--, len++; if (len > 0 && tp->t_state == TCPS_SYN_SENT) { - while (!(tp->t_flags & TF_SENDINPROG) && - tp->t_pktlist_head != NULL) { + while (tp->t_inpcb->inp_sndinprog_cnt == 0 && + tp->t_pktlist_head != NULL) { packetlist = tp->t_pktlist_head; packchain_listadd = tp->t_lastchain; packchain_sent++; TCP_PKTLIST_CLEAR(tp); - tp->t_flags |= TF_SENDINPROG; error = tcp_ip_output(so, tp, packetlist, packchain_listadd, tp_inp_options, - (so_options & SO_DONTROUTE)); + (so_options & SO_DONTROUTE), + (sack_rxmit | (sack_bytes_rxmt != 0)), 0, +#ifdef INET6 + isipv6); +#else + 0); +#endif + - tp->t_flags &= ~TF_SENDINPROG; } - /* tcp was closed while we were in ip; resume close */ - if ((tp->t_flags & - (TF_CLOSING|TF_SENDINPROG)) == TF_CLOSING) { + /* + * tcp was closed while we were in ip, + * resume close + */ + if (tp->t_inpcb->inp_sndinprog_cnt == 0 && + (tp->t_flags & TF_CLOSING)) { tp->t_flags &= ~TF_CLOSING; (void) tcp_close(tp); + } else { + tcp_check_timer_state(tp); } KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); - return 0; + return(0); } } @@ -580,7 +681,12 @@ after_sack_rexmit: flags &= ~TH_FIN; } - if (len < 0) { + /* The check here used to be (len < 0). Some times len is zero when + * the congestion window is closed and we need to check if persist timer + * has to be set in that case. But don't set persist until connection + * is established. + */ + if (len <= 0 && !(flags & TH_SYN)) { /* * If FIN has been sent but not acked, * but we haven't been called to retransmit, @@ -595,20 +701,101 @@ after_sack_rexmit: if (sendwin == 0) { tp->t_timer[TCPT_REXMT] = 0; tp->t_rxtshift = 0; + tp->rxt_start = 0; tp->snd_nxt = tp->snd_una; if (tp->t_timer[TCPT_PERSIST] == 0) tcp_setpersist(tp); } } + /* Automatic sizing of send socket buffer. Increase the send socket buffer + * size if all of the following criteria are met + * 1. the receiver has enough buffer space for this data + * 2. send buffer is filled to 7/8th with data (so we actually + * have data to make use of it); + * 3. our send window (slow start and congestion controlled) is + * larger than sent but unacknowledged data in send buffer. + */ + basertt = get_base_rtt(tp); + if (tcp_do_autosendbuf == 1 && + !INP_WAIT_FOR_IF_FEEDBACK(tp->t_inpcb) && !IN_FASTRECOVERY(tp) && + (so->so_snd.sb_flags & (SB_AUTOSIZE | SB_TRIM)) == SB_AUTOSIZE && + tcp_cansbgrow(&so->so_snd)) { + if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat && + so->so_snd.sb_cc >= (so->so_snd.sb_hiwat / 8 * 7) && + sendwin >= (so->so_snd.sb_cc - + (tp->snd_nxt - tp->snd_una))) { + /* Also increase the send buffer only if the + * round-trip time is not increasing because we do + * not want to contribute to latency by filling buffers. + * We also do not want to hold onto application's + * old data for too long. Interactive applications would + * rather discard old data. + */ + if (tp->t_rttcur <= + (basertt + 25)) { + if (sbreserve(&so->so_snd, + min(so->so_snd.sb_hiwat + tcp_autosndbuf_inc, + tcp_autosndbuf_max)) == 1) { + so->so_snd.sb_idealsize = so->so_snd.sb_hiwat; + } + } else { + so->so_snd.sb_idealsize = + max(tcp_sendspace, so->so_snd.sb_hiwat - + (2 * tcp_autosndbuf_inc)); + so->so_snd.sb_flags |= SB_TRIM; + } + } + } + + /* + * Truncate to the maximum segment length or enable TCP Segmentation + * Offloading (if supported by hardware) and ensure that FIN is removed + * if the length no longer contains the last data byte. + * + * TSO may only be used if we are in a pure bulk sending state. The + * presence of TCP-MD5, SACK retransmits, SACK advertizements, ipfw rules + * and IP options prevent using TSO. With TSO the TCP header is the same + * (except for the sequence number) for all generated packets. This + * makes it impossible to transmit any options which vary per generated + * segment or packet. + * + * The length of TSO bursts is limited to TCP_MAXWIN. That limit and + * removal of FIN (if not already catched here) are handled later after + * the exact length of the TCP options are known. + */ +#if IPSEC /* - * len will be >= 0 after this point. Truncate to the maximum - * segment length and ensure that FIN is removed if the length - * no longer contains the last data byte. + * Pre-calculate here as we save another lookup into the darknesses + * of IPsec that way and can actually decide if TSO is ok. */ + if (ipsec_bypass == 0) + ipsec_optlen = ipsec_hdrsiz_tcp(tp); +#endif + if (len > tp->t_maxseg) { - len = tp->t_maxseg; - sendalot = 1; + if ((tp->t_flags & TF_TSO) && tcp_do_tso && +#if RANDOM_IP_ID + ip_use_randomid && +#endif /* RANDOM_IP_ID */ + kipf_count == 0 && dlil_filter_count == 0 && + tp->rcv_numsacks == 0 && sack_rxmit == 0 && sack_bytes_rxmt == 0 && + tp->t_inpcb->inp_options == NULL && + tp->t_inpcb->in6p_options == NULL +#if IPSEC + && ipsec_optlen == 0 +#endif +#if IPFIREWALL + && (fw_enable == 0 || fw_bypass) +#endif + ) { + tso = 1; + sendalot = 0; + } else { + len = tp->t_maxseg; + sendalot = 1; + tso = 0; + } } if (sack_rxmit) { if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc)) @@ -624,47 +811,60 @@ after_sack_rexmit: * Sender silly window avoidance. We transmit under the following * conditions when len is non-zero: * - * - We have a full segment + * - we've timed out (e.g. persist timer) + * - we need to retransmit + * - We have a full segment (or more with TSO) * - This is the last buffer in a write()/send() and we are * either idle or running NODELAY - * - we've timed out (e.g. persist timer) * - we have more then 1/2 the maximum send window's worth of * data (receiver may be limited the window size) - * - we need to retransmit */ if (len) { - if (len == tp->t_maxseg) { - tp->t_flags |= TF_MAXSEGSNT; - goto send; - } - if (!(tp->t_flags & TF_MORETOCOME) && - (idle || tp->t_flags & TF_NODELAY || tp->t_flags & TF_MAXSEGSNT) && - (tp->t_flags & TF_NOPUSH) == 0 && - len + off >= so->so_snd.sb_cc) { - tp->t_flags &= ~TF_MAXSEGSNT; - goto send; - } if (tp->t_force) { tp->t_flags &= ~TF_MAXSEGSNT; goto send; } - if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) { - tp->t_flags &= ~TF_MAXSEGSNT; - goto send; - } if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { - tp->t_flags &= ~TF_MAXSEGSNT; + if (len >= tp->t_maxseg) + tp->t_flags |= TF_MAXSEGSNT; + else + tp->t_flags &= ~TF_MAXSEGSNT; goto send; } if (sack_rxmit) goto send; + + /* + * Send new data on the connection only if it is + * not flow controlled + */ + if (!INP_WAIT_FOR_IF_FEEDBACK(tp->t_inpcb) || + tp->t_state != TCPS_ESTABLISHED) { + if (len >= tp->t_maxseg) { + tp->t_flags |= TF_MAXSEGSNT; + goto send; + } + if (!(tp->t_flags & TF_MORETOCOME) && + (idle || tp->t_flags & TF_NODELAY || tp->t_flags & TF_MAXSEGSNT) && + (tp->t_flags & TF_NOPUSH) == 0 && + len + off >= so->so_snd.sb_cc) { + tp->t_flags &= ~TF_MAXSEGSNT; + goto send; + } + if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) { + tp->t_flags &= ~TF_MAXSEGSNT; + goto send; + } + } else { + tcpstat.tcps_fcholdpacket++; + } } /* * Compare available window to amount of window * known to peer (as advertised window less * next expected input). If the difference is at least two - * max size segments, or at least 50% of the maximum possible + * max size segments, or at least 25% of the maximum possible * window, then want to send a window update to peer. * Skip this if the connection is in T/TCP half-open state. */ @@ -674,13 +874,36 @@ after_sack_rexmit: * taking into account that we are limited by * TCP_MAXWIN << tp->rcv_scale. */ - long adv = lmin(recwin, (long)TCP_MAXWIN << tp->rcv_scale) - + int32_t adv = imin(recwin, (int)TCP_MAXWIN << tp->rcv_scale) - (tp->rcv_adv - tp->rcv_nxt); - if (adv >= (long) (2 * tp->t_maxseg)) - goto send; - if (2 * adv >= (long) so->so_rcv.sb_hiwat) - goto send; + if (adv >= (int32_t) (2 * tp->t_maxseg)) { + /* Update only if the resulting scaled value of the window changed, or + * if there is a change in the sequence since the last ack. + * This avoids what appears as dupe ACKS (see rdar://5640997) + * + * If streaming is detected avoid sending too many window updates. + * We will depend on the delack timer to send a window update + * when needed. + */ + if ((tp->t_flags & TF_STRETCHACK) == 0 && + (tp->last_ack_sent != tp->rcv_nxt || + ((recwin + adv) >> tp->rcv_scale) > recwin)) { + goto send; + } + + /* Make sure that the delayed ack timer is set if we + * delayed sending a window update because of streaming + * detection. + */ + if ((tp->t_flags & TF_STRETCHACK) != 0 && + (tp->t_flags & TF_DELACK) == 0) { + tp->t_flags |= TF_DELACK; + tp->t_timer[TCPT_DELACK] = OFFSET_FROM_START(tp, tcp_delack); + } + } + if (4 * adv >= (int32_t) so->so_rcv.sb_hiwat) + goto send; } /* @@ -706,10 +929,11 @@ after_sack_rexmit: * after the retransmission timer has been turned off. Make sure * that the retransmission timer is set. */ - if (tp->sack_enable && SEQ_GT(tp->snd_max, tp->snd_una) && - tp->t_timer[TCPT_REXMT] == 0 && + if (tp->sack_enable && (tp->t_state >= TCPS_ESTABLISHED) && + SEQ_GT(tp->snd_max, tp->snd_una) && + tp->t_timer[TCPT_REXMT] == 0 && 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); goto just_return; } /* @@ -737,6 +961,7 @@ after_sack_rexmit: if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 && tp->t_timer[TCPT_PERSIST] == 0) { tp->t_rxtshift = 0; + tp->rxt_start = 0; tcp_setpersist(tp); } just_return: @@ -744,22 +969,29 @@ just_return: * If there is no reason to send a segment, just return. * but if there is some packets left in the packet list, send them now. */ - while (!(tp->t_flags & TF_SENDINPROG) && tp->t_pktlist_head != NULL) { + while (tp->t_inpcb->inp_sndinprog_cnt == 0 && + tp->t_pktlist_head != NULL) { packetlist = tp->t_pktlist_head; packchain_listadd = tp->t_lastchain; packchain_sent++; TCP_PKTLIST_CLEAR(tp); - tp->t_flags |= TF_SENDINPROG; error = tcp_ip_output(so, tp, packetlist, packchain_listadd, - tp_inp_options, (so_options & SO_DONTROUTE)); - - tp->t_flags &= ~TF_SENDINPROG; + tp_inp_options, (so_options & SO_DONTROUTE), + (sack_rxmit | (sack_bytes_rxmt != 0)), recwin, +#ifdef INET6 + isipv6); +#else + 0); +#endif } /* tcp was closed while we were in ip; resume close */ - if ((tp->t_flags & (TF_CLOSING|TF_SENDINPROG)) == TF_CLOSING) { + if (tp->t_inpcb->inp_sndinprog_cnt == 0 && + (tp->t_flags & TF_CLOSING)) { tp->t_flags &= ~TF_CLOSING; (void) tcp_close(tp); + } else { + tcp_check_timer_state(tp); } KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return (0); @@ -779,7 +1011,7 @@ send: hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); else #endif - hdrlen = sizeof (struct tcpiphdr); + hdrlen = sizeof (struct tcpiphdr); if (flags & TH_SYN) { tp->snd_nxt = tp->iss; if ((tp->t_flags & TF_NOOPT) == 0) { @@ -794,7 +1026,7 @@ send: if ((tp->t_flags & TF_REQ_SCALE) && ((flags & TH_ACK) == 0 || (tp->t_flags & TF_RCVD_SCALE))) { - *((u_int32_t *)(opt + optlen)) = htonl( + *((u_int32_t *)(void *)(opt + optlen)) = htonl( TCPOPT_NOP << 24 | TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | @@ -875,7 +1107,7 @@ send: * the CWR flag on data packets. Pure acks don't have this set. */ if ((tp->ecn_flags & TE_SENDCWR) != 0 && len != 0 && - !SEQ_LT(tp->snd_nxt, tp->snd_max)) { + !SEQ_LT(tp->snd_nxt, tp->snd_max) && !sack_rxmit) { flags |= TH_CWR; tp->ecn_flags &= ~TE_SENDCWR; } @@ -896,7 +1128,7 @@ send: (flags & TH_RST) == 0 && ((flags & TH_ACK) == 0 || (tp->t_flags & TF_RCVD_TSTMP))) { - u_int32_t *lp = (u_int32_t *)(opt + optlen); + u_int32_t *lp = (u_int32_t *)(void *)(opt + optlen); /* Form timestamp option as shown in appendix A of RFC 1323. */ *lp++ = htonl(TCPOPT_TSTAMP_HDR); @@ -905,6 +1137,10 @@ send: optlen += TCPOLEN_TSTAMP_APPA; } + /* Note the timestamp for receive buffer autosizing */ + if (tp->rfbuf_ts == 0 && (so->so_rcv.sb_flags & SB_AUTOSIZE)) + tp->rfbuf_ts = tcp_now; + if (tp->sack_enable && ((tp->t_flags & TF_NOOPT) == 0)) { /* * Tack on the SACK permitted option *last*. @@ -946,7 +1182,7 @@ send: if (TCPS_HAVEESTABLISHED(tp->t_state) && (tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0 && MAX_TCPOPTLEN - optlen - 2 >= TCPOLEN_SACK) { - int nsack, sackoptlen, padlen; + int nsack, padlen; u_char *bp = (u_char *)opt + optlen; u_int32_t *lp; @@ -967,7 +1203,7 @@ send: tcpstat.tcps_sack_send_blocks++; *bp++ = TCPOPT_SACK; *bp++ = sackoptlen; - lp = (u_int32_t *)bp; + lp = (u_int32_t *)(void *)bp; for (i = 0; i < nsack; i++) { struct sackblk sack = tp->sackblks[i]; *lp++ = htonl(sack.start); @@ -1004,8 +1240,7 @@ send: ipoptlen = 0; } #if IPSEC - if (ipsec_bypass == 0) - ipoptlen += ipsec_hdrsiz_tcp(tp); + ipoptlen += ipsec_optlen; #endif /* @@ -1013,26 +1248,56 @@ send: * bump the packet length beyond the t_maxopd length. * Clear the FIN bit because we cut off the tail of * the segment. + * + * When doing TSO limit a burst to TCP_MAXWIN minus the + * IP, TCP and Options length to keep ip->ip_len from + * overflowing. Prevent the last segment from being + * fractional thus making them all equal sized and set + * the flag to continue sending. TSO is disabled when + * IP options or IPSEC are present. */ if (len + optlen + ipoptlen > tp->t_maxopd) { /* * If there is still more to send, don't close the connection. */ flags &= ~TH_FIN; - len = tp->t_maxopd - optlen - ipoptlen; - sendalot = 1; + if (tso) { + int32_t tso_maxlen; + + tso_maxlen = tp->tso_max_segment_size ? tp->tso_max_segment_size : TCP_MAXWIN; + + if (len > tso_maxlen - hdrlen - optlen) { + len = tso_maxlen - hdrlen - optlen; + len = len - (len % (tp->t_maxopd - optlen)); + sendalot = 1; + } else if (tp->t_flags & TF_NEEDFIN) + sendalot = 1; + } else { + len = tp->t_maxopd - optlen - ipoptlen; + sendalot = 1; + } } -/*#ifdef DIAGNOSTIC*/ -#if INET6 if (max_linkhdr + hdrlen > MCLBYTES) panic("tcphdr too big"); -#else - if (max_linkhdr + hdrlen > MHLEN) - panic("tcphdr too big"); -#endif -/*#endif*/ + /* Check if there is enough data in the send socket + * buffer to start measuring bw + */ + if ((tp->t_flagsext & TF_MEASURESNDBW) != 0 && + (tp->t_bwmeas != NULL) && + (tp->t_flagsext & TF_BWMEAS_INPROGRESS) == 0 && + (so->so_snd.sb_cc - (tp->snd_max - tp->snd_una)) >= + tp->t_bwmeas->bw_minsize) { + tp->t_bwmeas->bw_size = min((so->so_snd.sb_cc - (tp->snd_max - tp->snd_una)), + tp->t_bwmeas->bw_maxsize); + tp->t_flagsext |= TF_BWMEAS_INPROGRESS; + tp->t_bwmeas->bw_start = tp->snd_max; + tp->t_bwmeas->bw_ts = tcp_now; + } + + VERIFY(tp->t_inpcb->inp_flowhash != 0); + /* * Grab a header mbuf, attaching a copy of data to * be transmitted, and initialize the header from @@ -1041,25 +1306,24 @@ send: if (len) { if (tp->t_force && len == 1) tcpstat.tcps_sndprobe++; - else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { + else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) { tcpstat.tcps_sndrexmitpack++; tcpstat.tcps_sndrexmitbyte += len; + if (nstat_collect) { + nstat_route_tx(tp->t_inpcb->inp_route.ro_rt, 1, + len, NSTAT_TX_FLAG_RETRANSMIT); + locked_add_64(&tp->t_inpcb->inp_stat->txpackets, 1); + locked_add_64(&tp->t_inpcb->inp_stat->txbytes, len); + tp->t_stat.txretransmitbytes += len; + } } else { tcpstat.tcps_sndpack++; tcpstat.tcps_sndbyte += len; - } -#ifdef notyet - if ((m = m_copypack(so->so_snd.sb_mb, off, - (int)len, max_linkhdr + hdrlen)) == 0) { - error = ENOBUFS; - goto out; + if (nstat_collect) { + locked_add_64(&tp->t_inpcb->inp_stat->txpackets, 1); + locked_add_64(&tp->t_inpcb->inp_stat->txbytes, len); + } } - /* - * m_copypack left space for our hdr; use it. - */ - m->m_len += hdrlen; - m->m_data -= hdrlen; -#else /* * try to use the new interface that allocates all * the necessary mbuf hdrs under 1 mbuf lock and @@ -1073,11 +1337,13 @@ send: * data area (no cluster attached) */ m = NULL; -#if INET6 + + /* minimum length we are going to allocate */ + allocated_len = MHLEN; if (MHLEN < hdrlen + max_linkhdr) { - MGETHDR(m, M_DONTWAIT, MT_HEADER); /* MAC-OK */ + MGETHDR(m, M_DONTWAIT, MT_HEADER); if (m == NULL) { - error = ENOBUFS; + error = ENOBUFS; goto out; } MCLGET(m, M_DONTWAIT); @@ -1088,20 +1354,21 @@ send: } m->m_data += max_linkhdr; m->m_len = hdrlen; + allocated_len = MCLBYTES; } -#endif - if (len <= MHLEN - hdrlen - max_linkhdr) { + if (len <= allocated_len - hdrlen - max_linkhdr) { if (m == NULL) { - MGETHDR(m, M_DONTWAIT, MT_HEADER); /* MAC-OK */ + VERIFY(allocated_len <= MHLEN); + MGETHDR(m, M_DONTWAIT, MT_HEADER); if (m == NULL) { - error = ENOBUFS; + error = ENOBUFS; goto out; } m->m_data += max_linkhdr; m->m_len = hdrlen; } /* makes sure we still have data left to be sent at this point */ - if (so->so_snd.sb_mb == NULL || off == -1) { + if (so->so_snd.sb_mb == NULL || off < 0) { if (m != NULL) m_freem(m); error = 0; /* should we return an error? */ goto out; @@ -1128,8 +1395,8 @@ send: * m_copym_with_hdrs to avoid rescanning from the beginning of the socket buffer mbuf list. * setting the mbuf pointer to NULL is sufficient to disable the hint mechanism. */ - if (m_head != so->so_snd.sb_mb || last_off != off) - m_last = NULL; + if (m_head != so->so_snd.sb_mb || sack_rxmit || last_off != off) + m_lastm = NULL; last_off = off + len; m_head = so->so_snd.sb_mb; @@ -1143,7 +1410,7 @@ send: * m_copym_with_hdrs will always return the last mbuf pointer and the offset into it that * it acted on to fullfill the current request, whether a valid 'hint' was passed in or not */ - if ((m = m_copym_with_hdrs(so->so_snd.sb_mb, off, (int) len, M_DONTWAIT, &m_last, &m_off)) == NULL) { + if ((m = m_copym_with_hdrs(so->so_snd.sb_mb, off, len, M_DONTWAIT, &m_lastm, &m_off)) == NULL) { error = ENOBUFS; goto out; } @@ -1151,7 +1418,6 @@ send: m->m_len = hdrlen; } } -#endif /* * If we're sending everything we've got, set PUSH. * (This will keep happy those implementations which only @@ -1175,12 +1441,14 @@ send: error = ENOBUFS; goto out; } -#if INET6 - if (isipv6 && (MHLEN < hdrlen + max_linkhdr) && - MHLEN >= hdrlen) { - MH_ALIGN(m, hdrlen); - } else -#endif + if (MHLEN < (hdrlen + max_linkhdr)) { + MCLGET(m, M_DONTWAIT); + if ((m->m_flags & M_EXT) == 0) { + m_freem(m); + error = ENOBUFS; + goto out; + } + } m->m_data += max_linkhdr; m->m_len = hdrlen; } @@ -1191,18 +1459,23 @@ send: #if INET6 if (isipv6) { ip6 = mtod(m, struct ip6_hdr *); - th = (struct tcphdr *)(ip6 + 1); + th = (struct tcphdr *)(void *)(ip6 + 1); tcp_fillheaders(tp, ip6, th); + if ((tp->ecn_flags & TE_SENDIPECT) != 0 && len && + !SEQ_LT(tp->snd_nxt, tp->snd_max) && !sack_rxmit) { + ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); + } + svc_flags |= PKT_SCF_IPV6; } else #endif /* INET6 */ { ip = mtod(m, struct ip *); ipov = (struct ipovly *)ip; - th = (struct tcphdr *)(ip + 1); + th = (struct tcphdr *)(void *)(ip + 1); /* this picks up the pseudo header (w/o the length) */ tcp_fillheaders(tp, ip, th); if ((tp->ecn_flags & TE_SENDIPECT) != 0 && len && - !SEQ_LT(tp->snd_nxt, tp->snd_max)) { + !SEQ_LT(tp->snd_nxt, tp->snd_max) && !sack_rxmit) { ip->ip_tos = IPTOS_ECN_ECT0; } } @@ -1212,7 +1485,7 @@ send: * window for use in delaying messages about window sizes. * If resending a FIN, be sure not to use a new sequence number. */ - if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && + if (flags & TH_FIN && (tp->t_flags & TF_SENTFIN) && tp->snd_nxt == tp->snd_max) tp->snd_nxt--; /* @@ -1239,6 +1512,8 @@ send: tp->sackhint.sack_bytes_rexmit += len; } th->th_ack = htonl(tp->rcv_nxt); + tp->last_ack_sent = tp->rcv_nxt; + if (optlen) { bcopy(opt, th + 1, optlen); th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; @@ -1248,20 +1523,32 @@ send: * Calculate receive window. Don't shrink window, * but avoid silly window syndrome. */ - if (recwin < (long)(so->so_rcv.sb_hiwat / 4) && recwin < (long)tp->t_maxseg) + if (recwin < (int32_t)(so->so_rcv.sb_hiwat / 4) && recwin < (int)tp->t_maxseg) recwin = 0; - if (recwin < (long)(tp->rcv_adv - tp->rcv_nxt)) - recwin = (long)(tp->rcv_adv - tp->rcv_nxt); + if (recwin < (int32_t)(tp->rcv_adv - tp->rcv_nxt)) + recwin = (int32_t)(tp->rcv_adv - tp->rcv_nxt); if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0) { - if (recwin > (long)slowlink_wsize) + if (recwin > (int32_t)slowlink_wsize) recwin = slowlink_wsize; - th->th_win = htons((u_short) (recwin>>tp->rcv_scale)); } - else { - if (recwin > (long)(TCP_MAXWIN << tp->rcv_scale)) - recwin = (long)(TCP_MAXWIN << tp->rcv_scale); - th->th_win = htons((u_short) (recwin>>tp->rcv_scale)); + +#if TRAFFIC_MGT + if (tcp_recv_bg == 1 || IS_TCP_RECV_BG(so)) { + if (tp->acc_iaj > tcp_acc_iaj_react_limit) { + uint32_t min_iaj_win = tcp_min_iaj_win * tp->t_maxseg; + if (tp->iaj_rwintop == 0 || + SEQ_LT(tp->iaj_rwintop, tp->rcv_adv)) + tp->iaj_rwintop = tp->rcv_adv; + if (SEQ_LT(tp->iaj_rwintop, tp->rcv_nxt + min_iaj_win)) + tp->iaj_rwintop = tp->rcv_nxt + min_iaj_win; + recwin = min(tp->iaj_rwintop - tp->rcv_nxt, recwin); + } } +#endif /* TRAFFIC_MGT */ + + if (recwin > (int32_t)(TCP_MAXWIN << tp->rcv_scale)) + recwin = (int32_t)(TCP_MAXWIN << tp->rcv_scale); + th->th_win = htons((u_short) (recwin>>tp->rcv_scale)); /* * Adjust the RXWIN0SENT flag - indicate that we have advertised @@ -1271,7 +1558,7 @@ send: * to read more data then can be buffered prior to transmitting on * the connection. */ - if (recwin == 0) + if (th->th_win == 0) tp->t_flags |= TF_RXWIN0SENT; else tp->t_flags &= ~TF_RXWIN0SENT; @@ -1293,13 +1580,17 @@ send: */ m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ #if INET6 - if (isipv6) + if (isipv6) { /* * ip6_plen is not need to be filled now, and will be filled * in ip6_output. */ - th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), - sizeof(struct tcphdr) + optlen + len); + m->m_pkthdr.csum_flags = CSUM_TCPIPV6; + m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); + if (len + optlen) + th->th_sum = in_addword(th->th_sum, + htons((u_short)(optlen + len))); + } else #endif /* INET6 */ { @@ -1310,6 +1601,23 @@ send: htons((u_short)(optlen + len))); } + /* + * Enable TSO and specify the size of the segments. + * The TCP pseudo header checksum is always provided. + */ + if (tso) { +#if INET6 + if (isipv6) + m->m_pkthdr.csum_flags = CSUM_TSO_IPV6; + else +#endif /* INET6 */ + m->m_pkthdr.csum_flags = CSUM_TSO_IPV4; + + m->m_pkthdr.tso_segsz = tp->t_maxopd - optlen; + } + else + m->m_pkthdr.tso_segsz = 0; + /* * In transmit state, time the transmission and arrange for * the retransmit. In persist state, just set snd_max. @@ -1338,7 +1646,7 @@ send: * not currently timing anything. */ if (tp->t_rtttime == 0) { - tp->t_rtttime = 1; + tp->t_rtttime = tcp_now; tp->t_rtseq = startseq; tcpstat.tcps_segstimed++; } @@ -1359,8 +1667,10 @@ timer: if (tp->t_timer[TCPT_PERSIST]) { tp->t_timer[TCPT_PERSIST] = 0; tp->t_rxtshift = 0; + tp->rxt_start = 0; + tp->t_persist_stop = 0; } - tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; + tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp, tp->t_rxtcur); } } else { /* @@ -1392,11 +1702,11 @@ timer: * to handle ttl and tos; we could keep them in * the template, but need a way to checksum without them. */ +#ifdef INET6 /* * m->m_pkthdr.len should have been set before cksum calcuration, * because in6_cksum() need it. */ -#if INET6 if (isipv6) { /* * we separately set hoplimit for every segment, since the @@ -1410,51 +1720,23 @@ timer: : NULL); /* TODO: IPv6 IP6TOS_ECT bit on */ -#if IPSEC - if (ipsec_bypass == 0 && ipsec_setsocket(m, so) != 0) { - m_freem(m); - error = ENOBUFS; - goto out; - } -#endif /*IPSEC*/ - m->m_pkthdr.socket_id = socket_id; - error = ip6_output(m, - inp6_pktopts, - &tp->t_inpcb->in6p_route, - (so_options & SO_DONTROUTE), NULL, NULL, 0); + KERNEL_DEBUG(DBG_LAYER_BEG, + ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport), + (((tp->t_inpcb->in6p_laddr.s6_addr16[0] & 0xffff) << 16) | + (tp->t_inpcb->in6p_faddr.s6_addr16[0] & 0xffff)), + sendalot,0,0); } else #endif /* INET6 */ - { - ip->ip_len = m->m_pkthdr.len; -#if INET6 - if (isipv6) - ip->ip_ttl = in6_selecthlim(tp->t_inpcb, - tp->t_inpcb->in6p_route.ro_rt ? - tp->t_inpcb->in6p_route.ro_rt->rt_ifp - : NULL); - else -#endif /* INET6 */ - ip->ip_ttl = tp->t_inpcb->inp_ip_ttl; /* XXX */ - ip->ip_tos |= (tp->t_inpcb->inp_ip_tos & ~IPTOS_ECN_MASK); /* XXX */ - - -#if INET6 - if (isipv6) { - KERNEL_DEBUG(DBG_LAYER_BEG, - ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport), - (((tp->t_inpcb->in6p_laddr.s6_addr16[0] & 0xffff) << 16) | - (tp->t_inpcb->in6p_faddr.s6_addr16[0] & 0xffff)), - 0,0,0); - } - else -#endif { - KERNEL_DEBUG(DBG_LAYER_BEG, - ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport), - (((tp->t_inpcb->inp_laddr.s_addr & 0xffff) << 16) | - (tp->t_inpcb->inp_faddr.s_addr & 0xffff)), - 0,0,0); - } + ip->ip_len = m->m_pkthdr.len; + ip->ip_ttl = tp->t_inpcb->inp_ip_ttl; /* XXX */ + ip->ip_tos |= (tp->t_inpcb->inp_ip_tos & ~IPTOS_ECN_MASK);/* XXX */ + KERNEL_DEBUG(DBG_LAYER_BEG, + ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport), + (((tp->t_inpcb->inp_laddr.s_addr & 0xffff) << 16) | + (tp->t_inpcb->inp_faddr.s_addr & 0xffff)), + 0,0,0); + } /* * See if we should do MTU discovery. @@ -1465,9 +1747,11 @@ timer: * 4) the MTU is not locked (if it is, then discovery has been * disabled for that route) */ - - if (path_mtu_discovery && (tp->t_flags & TF_PMTUD)) - ip->ip_off |= IP_DF; +#ifdef INET6 + if (!isipv6) +#endif + if (path_mtu_discovery && (tp->t_flags & TF_PMTUD)) + ip->ip_off |= IP_DF; #if IPSEC if (ipsec_bypass == 0) @@ -1479,9 +1763,51 @@ timer: */ lost = 0; m->m_pkthdr.socket_id = socket_id; + + /* + * Embed the flow hash in pkt hdr and mark the packet as + * capable of flow controlling + */ + m->m_pkthdr.m_flowhash = tp->t_inpcb->inp_flowhash; + m->m_pkthdr.m_fhflags |= + (PF_TAG_TCP | PF_TAG_FLOWHASH | PF_TAG_FLOWADV); + m->m_nextpkt = NULL; + + if (tp->t_inpcb->inp_last_outifp != NULL && + tp->t_inpcb->inp_last_outifp != lo_ifp) { + /* Hint to prioritize this packet if + * 1. if the packet has no data + * 2. the interface supports transmit-start model and did + * not disable ACK prioritization. + * 3. Only ACK flag is set. + * 4. there is no outstanding data on this connection. + */ + if (tcp_prioritize_acks != 0 && len == 0 && + (tp->t_inpcb->inp_last_outifp->if_eflags & + (IFEF_TXSTART | IFEF_NOACKPRI)) == IFEF_TXSTART && + th->th_flags == TH_ACK && tp->snd_una == tp->snd_max && + tp->t_timer[TCPT_REXMT] == 0) { + svc_flags |= PKT_SCF_TCP_ACK; + } + set_packet_service_class(m, so, MBUF_SC_UNSPEC, svc_flags); + } + tp->t_pktlist_sentlen += len; tp->t_lastchain++; + +#ifdef INET6 + if (isipv6) { + DTRACE_TCP5(send, struct mbuf *, m, struct inpcb *, tp->t_inpcb, + struct ip6 *, ip6, struct tcpcb *, tp, struct tcphdr *, + th); + } else +#endif + { + DTRACE_TCP5(send, struct mbuf *, m, struct inpcb *, tp->t_inpcb, + struct ip *, ip, struct tcpcb *, tp, struct tcphdr *, th); + } + if (tp->t_pktlist_head != NULL) { tp->t_pktlist_tail->m_nextpkt = m; tp->t_pktlist_tail = m; @@ -1490,25 +1816,53 @@ timer: tp->t_pktlist_head = tp->t_pktlist_tail = m; } + if ((lro_ackmore) && (!sackoptlen) && (!tp->t_timer[TCPT_PERSIST]) && + ((th->th_flags & TH_ACK) == TH_ACK) && (!len) && + (tp->t_state == TCPS_ESTABLISHED)) { + /* For a pure ACK, see if you need to send more of them */ + mnext = tcp_send_lroacks(tp, m, th); + if (mnext) { + tp->t_pktlist_tail->m_nextpkt = mnext; + if (mnext->m_nextpkt == NULL) { + tp->t_pktlist_tail = mnext; + tp->t_lastchain++; + } else { + struct mbuf *tail, *next; + next = mnext->m_nextpkt; + tail = next->m_nextpkt; + while (tail) { + next = tail; + tail = tail->m_nextpkt; + tp->t_lastchain++; + } + tp->t_pktlist_tail = next; + } + } + } + if (sendalot == 0 || (tp->t_state != TCPS_ESTABLISHED) || (tp->snd_cwnd <= (tp->snd_wnd / 8)) || (tp->t_flags & (TH_PUSH | TF_ACKNOW)) || tp->t_force != 0 || tp->t_lastchain >= tcp_packet_chaining) { error = 0; - while (!(tp->t_flags & TF_SENDINPROG) && - tp->t_pktlist_head != NULL) { + while (tp->t_inpcb->inp_sndinprog_cnt == 0 && + tp->t_pktlist_head != NULL) { packetlist = tp->t_pktlist_head; packchain_listadd = tp->t_lastchain; packchain_sent++; lost = tp->t_pktlist_sentlen; TCP_PKTLIST_CLEAR(tp); - tp->t_flags |= TF_SENDINPROG; error = tcp_ip_output(so, tp, packetlist, packchain_listadd, tp_inp_options, - (so_options & SO_DONTROUTE)); + (so_options & SO_DONTROUTE), + (sack_rxmit | (sack_bytes_rxmt != 0)), recwin, +#ifdef INET6 + isipv6); +#else + 0); +#endif - tp->t_flags &= ~TF_SENDINPROG; if (error) { /* * Take into account the rest of unsent @@ -1523,24 +1877,19 @@ timer: } } /* tcp was closed while we were in ip; resume close */ - if ((tp->t_flags & (TF_CLOSING|TF_SENDINPROG)) == TF_CLOSING) { + if (tp->t_inpcb->inp_sndinprog_cnt == 0 && + (tp->t_flags & TF_CLOSING)) { tp->t_flags &= ~TF_CLOSING; (void) tcp_close(tp); return (0); } - } - else { + } else { error = 0; packchain_looped++; tcpstat.tcps_sndtotal++; - if (recwin > 0 && SEQ_GT(tp->rcv_nxt+recwin, tp->rcv_adv)) - tp->rcv_adv = tp->rcv_nxt + recwin; - tp->last_ack_sent = tp->rcv_nxt; - tp->t_flags &= ~(TF_ACKNOW|TF_DELACK); goto again; } - } if (error) { /* * Assume that the packets were lost, so back out the @@ -1568,11 +1917,20 @@ out: TCP_PKTLIST_CLEAR(tp); if (error == ENOBUFS) { - if (!tp->t_timer[TCPT_REXMT] && - !tp->t_timer[TCPT_PERSIST]) - tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; - tcp_quench(tp->t_inpcb, 0); + if (!tp->t_timer[TCPT_REXMT] && + !tp->t_timer[TCPT_PERSIST]) + tp->t_timer[TCPT_REXMT] = + OFFSET_FROM_START(tp, tp->t_rxtcur); + + tp->snd_cwnd = tp->t_maxseg; + tp->t_bytes_acked = 0; + + tcp_check_timer_state(tp); KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); + + DTRACE_TCP5(cc, void, NULL, struct inpcb *, tp->t_inpcb, + struct tcpcb *, tp, struct tcphdr *, NULL, + int32_t, TCP_CC_OUTPUT_ERROR); return (0); } if (error == EMSGSIZE) { @@ -1581,60 +1939,138 @@ out: * for us. tcp_mtudisc() will, as its last action, * initiate retransmission, so it is important to * not do so here. + * + * If TSO was active we either got an interface + * without TSO capabilits or TSO was turned off. + * Disable it for this connection as too and + * immediatly retry with MSS sized segments generated + * by this function. */ + if (tso) + tp->t_flags &= ~TF_TSO; + tcp_mtudisc(tp->t_inpcb, 0); + tcp_check_timer_state(tp); + KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return 0; } if ((error == EHOSTUNREACH || error == ENETDOWN) && TCPS_HAVERCVDSYN(tp->t_state)) { tp->t_softerror = error; + tcp_check_timer_state(tp); KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return (0); } + tcp_check_timer_state(tp); KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return (error); } tcpstat.tcps_sndtotal++; - /* - * Data sent (as far as we can tell). - * If this advertises a larger window than any other segment, - * then remember the size of the advertised window. - * Any pending ACK has now been sent. - */ - if (recwin > 0 && SEQ_GT(tp->rcv_nxt+recwin, tp->rcv_adv)) - tp->rcv_adv = tp->rcv_nxt + recwin; - tp->last_ack_sent = tp->rcv_nxt; - tp->t_flags &= ~(TF_ACKNOW|TF_DELACK); - KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END,0,0,0,0,0); - if (sendalot && (!tcp_do_newreno || --maxburst)) + if (sendalot) goto again; + + tcp_check_timer_state(tp); return (0); } static int tcp_ip_output(struct socket *so, struct tcpcb *tp, struct mbuf *pkt, - int cnt, struct mbuf *opt, int flags) + int cnt, struct mbuf *opt, int flags, int sack_in_progress, int recwin, + boolean_t isipv6) { int error = 0; boolean_t chain; boolean_t unlocked = FALSE; + struct inpcb *inp = tp->t_inpcb; + struct ip_out_args ipoa = + { IFSCOPE_NONE, { 0 }, IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR }; + struct route ro; + struct ifnet *outif = NULL; +#ifdef INET6 + struct ip6_out_args ip6oa = + { IFSCOPE_NONE, { 0 }, IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR }; + struct route_in6 ro6; + struct flowadv *adv = + (isipv6 ? &ip6oa.ip6oa_flowadv : &ipoa.ipoa_flowadv); +#else + struct flowadv *adv = &ipoa.ipoa_flowadv; +#endif /* !INET6 */ + + /* If socket was bound to an ifindex, tell ip_output about it */ + if (inp->inp_flags & INP_BOUND_IF) { +#ifdef INET6 + if (isipv6) { + ip6oa.ip6oa_boundif = inp->inp_boundifp->if_index; + ip6oa.ip6oa_flags |= IP6OAF_BOUND_IF; + } else +#endif + { + ipoa.ipoa_boundif = inp->inp_boundifp->if_index; + ipoa.ipoa_flags |= IPOAF_BOUND_IF; + } + } + + if (inp->inp_flags & INP_NO_IFT_CELLULAR) { +#ifdef INET6 + if (isipv6) + ip6oa.ip6oa_flags |= IP6OAF_NO_CELLULAR; + else +#endif + ipoa.ipoa_flags |= IPOAF_NO_CELLULAR; + } +#ifdef INET6 + if (isipv6) + flags |= IPV6_OUTARGS; + else +#endif + flags |= IP_OUTARGS; + + /* Copy the cached route and take an extra reference */ +#ifdef INET6 + if (isipv6) + in6p_route_copyout(inp, &ro6); + else +#endif + inp_route_copyout(inp, &ro); + + /* + * Data sent (as far as we can tell). + * If this advertises a larger window than any other segment, + * then remember the size of the advertised window. + * Make sure ACK/DELACK conditions are cleared before + * we unlock the socket. + */ + if (recwin > 0 && SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv)) + tp->rcv_adv = tp->rcv_nxt + recwin; + tp->last_ack_sent = tp->rcv_nxt; + tp->t_flags &= ~(TF_ACKNOW | TF_DELACK); + tp->t_timer[TCPT_DELACK] = 0; + tp->t_unacksegs = 0; + + /* Increment the count of outstanding send operations */ + inp->inp_sndinprog_cnt++; /* - * If allowed, unlock TCP socket while in IP + * If allowed, unlock TCP socket while in IP * but only if the connection is established and - * if we're not sending from an upcall. - */ + * in a normal mode where reentrancy on the tcpcb won't be + * an issue: + * - there is no SACK episode + * - we're not in Fast Recovery mode + * - if we're not sending from an upcall. + */ + if (tcp_output_unlocked && !so->so_upcallusecount && + (tp->t_state == TCPS_ESTABLISHED) && (sack_in_progress == 0) && + ((tp->t_flags & TF_FASTRECOVERY) == 0)) { - if (tcp_output_unlocked && ((so->so_flags & SOF_UPCALLINUSE) == 0) && - (tp->t_state == TCPS_ESTABLISHED)) { - unlocked = TRUE; - socket_unlock(so, 0); + unlocked = TRUE; + socket_unlock(so, 0); } - + /* * Don't send down a chain of packets when: * - TCP chaining is disabled @@ -1642,11 +2078,15 @@ tcp_ip_output(struct socket *so, struct tcpcb *tp, struct mbuf *pkt, * - there is a non default rule set for the firewall */ - chain = tcp_packet_chaining > 1 && + chain = tcp_packet_chaining > 1 #if IPSEC - ipsec_bypass && + && ipsec_bypass +#endif +#if IPFIREWALL + && (fw_enable == 0 || fw_bypass) #endif - (fw_enable == 0 || fw_bypass); + ; // I'm important, not extraneous + while (pkt != NULL) { struct mbuf *npkt = pkt->m_nextpkt; @@ -1666,13 +2106,16 @@ tcp_ip_output(struct socket *so, struct tcpcb *tp, struct mbuf *pkt, */ cnt = 0; } -#if CONFIG_FORCE_OUT_IFP - error = ip_output_list(pkt, cnt, opt, &tp->t_inpcb->inp_route, - flags, 0, tp->t_inpcb->pdp_ifp); -#else - error = ip_output_list(pkt, cnt, opt, &tp->t_inpcb->inp_route, - flags, 0, NULL); +#ifdef INET6 + if (isipv6) + error = ip6_output_list(pkt, cnt, + inp->in6p_outputopts, &ro6, flags, NULL, NULL, + &ip6oa); + else #endif + error = ip_output_list(pkt, cnt, opt, &ro, flags, NULL, + &ipoa); + if (chain || error) { /* * If we sent down a chain then we are done since @@ -1689,6 +2132,71 @@ tcp_ip_output(struct socket *so, struct tcpcb *tp, struct mbuf *pkt, if (unlocked) socket_lock(so, 0); + /* + * Enter flow controlled state if the connection is established + * and is not in recovery. + * + * A connection will enter suspended state even if it is in + * recovery. + */ + if (((adv->code == FADV_FLOW_CONTROLLED && !IN_FASTRECOVERY(tp)) || + adv->code == FADV_SUSPENDED) && + !(tp->t_flags & TF_CLOSING) && + tp->t_state == TCPS_ESTABLISHED) { + int rc; + rc = inp_set_fc_state(inp, adv->code); + + if (rc == 1) + DTRACE_TCP5(cc, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, struct tcphdr *, NULL, + int32_t, ((adv->code == FADV_FLOW_CONTROLLED) ? + TCP_CC_FLOW_CONTROL : TCP_CC_SUSPEND)); + } + + /* + * When an interface queue gets suspended, some of the + * packets are dropped. Return ENOBUFS, to update the + * pcb state. + */ + if (adv->code == FADV_SUSPENDED) + error = ENOBUFS; + + VERIFY(inp->inp_sndinprog_cnt > 0); + if ( --inp->inp_sndinprog_cnt == 0) + inp->inp_flags &= ~(INP_FC_FEEDBACK); + +#ifdef INET6 + if (isipv6) { + if (ro6.ro_rt != NULL && (outif = ro6.ro_rt->rt_ifp) != + inp->in6p_last_outifp) + inp->in6p_last_outifp = outif; + } else +#endif + if (ro.ro_rt != NULL && (outif = ro.ro_rt->rt_ifp) != + inp->inp_last_outifp) + inp->inp_last_outifp = outif; + + if ((inp->inp_flags & INP_NO_IFT_CELLULAR) && outif != NULL && + outif->if_type == IFT_CELLULAR) + soevent(inp->inp_socket, + (SO_FILT_HINT_LOCKED|SO_FILT_HINT_IFDENIED)); + + /* Synchronize cached PCB route & options */ +#ifdef INET6 + if (isipv6) + in6p_route_copyin(inp, &ro6); + else +#endif + inp_route_copyin(inp, &ro); + + if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift == 0 && + tp->t_inpcb->inp_route.ro_rt != NULL) { + /* If we found the route and there is an rtt on it + * reset the retransmit timer + */ + tcp_getrt_rtt(tp, tp->t_inpcb->in6p_route.ro_rt); + tp->t_timer[TCPT_REXMT] = OFFSET_FROM_START(tp, tp->t_rxtcur); + } return (error); } @@ -1698,14 +2206,104 @@ tcp_setpersist(tp) { int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; - if (tp->t_timer[TCPT_REXMT]) - panic("tcp_setpersist: retransmit pending"); + /* If a PERSIST_TIMER option was set we will limit the + * time the persist timer will be active for that connection + * in order to avoid DOS by using zero window probes. + * see rdar://5805356 + */ + + if ((tp->t_persist_timeout != 0) && + (tp->t_timer[TCPT_PERSIST] == 0) && + (tp->t_persist_stop == 0)) { + tp->t_persist_stop = tcp_now + tp->t_persist_timeout; + } + /* * Start/restart persistance timer. */ TCPT_RANGESET(tp->t_timer[TCPT_PERSIST], t * tcp_backoff[tp->t_rxtshift], - TCPTV_PERSMIN, TCPTV_PERSMAX); + TCPTV_PERSMIN, TCPTV_PERSMAX, 0); + tp->t_timer[TCPT_PERSIST] = OFFSET_FROM_START(tp, tp->t_timer[TCPT_PERSIST]); + if (tp->t_rxtshift < TCP_MAXRXTSHIFT) tp->t_rxtshift++; } + +/* + * Send as many acks as data coalesced. Every other packet when stretch + * ACK is not enabled. Every 8 packets, if stretch ACK is enabled. + */ +static struct mbuf* +tcp_send_lroacks(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th) +{ + struct mbuf *mnext = NULL, *ack_chain = NULL, *tail = NULL; + int count = 0; + tcp_seq org_ack = ntohl(th->th_ack); + tcp_seq prev_ack = 0; + int tack_offset = 28; /* XXX IPv6 not supported */ + int ack_size = (tp->t_flags & TF_STRETCHACK) ? + (maxseg_unacked * tp->t_maxseg) : (tp->t_maxseg << 1); + int segs_acked = (tp->t_flags & TF_STRETCHACK) ? maxseg_unacked : 2; + struct mbuf *prev_ack_pkt = NULL; + struct socket *so = tp->t_inpcb->inp_socket; + + count = tp->t_lropktlen/tp->t_maxseg; + + prev_ack = (org_ack - tp->t_lropktlen) + ack_size; + if (prev_ack < org_ack) { + ack_chain = m_dup(m, M_DONTWAIT); + if (ack_chain) { + th->th_ack = htonl(prev_ack); + tail = ack_chain; + count -= segs_acked; /* accounts for prev_ack packet */ + count = (count <= segs_acked) ? 0 : count - segs_acked; + tcpstat.tcps_sndacks++; + so_tc_update_stats(m, so, m_get_service_class(m)); + } else { + return NULL; + } + } + else { + tp->t_lropktlen = 0; + return NULL; + } + + prev_ack_pkt = ack_chain; + + while (count > 0) { + if ((prev_ack + ack_size) < org_ack) { + prev_ack += ack_size; + } else { + /* + * The last ACK sent must have the ACK number that TCP + * thinks is the last sent ACK number. + */ + prev_ack = org_ack; + } + mnext = m_dup(prev_ack_pkt, M_DONTWAIT); + if (mnext) { + HTONL(prev_ack); + bcopy(&prev_ack, mtod(prev_ack_pkt, caddr_t) + tack_offset, 4); + NTOHL(prev_ack); + tail->m_nextpkt = mnext; + tail = mnext; + count -= segs_acked; + tcpstat.tcps_sndacks++; + so_tc_update_stats(m, so, m_get_service_class(m)); + if (lrodebug == 5) { + printf("%s: lropktlen = %d count = %d, th_ack = %x \n", + __func__, tp->t_lropktlen, count, + th->th_ack); + } + } else { + if (lrodebug == 5) { + printf("%s: failed to alloc mbuf.\n", __func__); + } + break; + } + prev_ack_pkt = mnext; + } + tp->t_lropktlen = 0; + return ack_chain; +}