X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/de355530ae67247cbd0da700edb3a2a1dae884c2..0c530ab8987f0ae6a1a3d9284f40182b88852816:/bsd/netinet/tcp_output.c diff --git a/bsd/netinet/tcp_output.c b/bsd/netinet/tcp_output.c index d3e5558d3..b336fa6d8 100644 --- a/bsd/netinet/tcp_output.c +++ b/bsd/netinet/tcp_output.c @@ -113,7 +113,7 @@ int ss_fltsz = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW, &ss_fltsz, 1, "Slow start flight size"); -int ss_fltsz_local = TCP_MAXWIN; /* something large */ +int ss_fltsz_local = 4; /* starts with four segments max */ SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW, &ss_fltsz_local, 1, "Slow start flight size for local networks"); @@ -121,7 +121,14 @@ int tcp_do_newreno = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno, CTLFLAG_RW, &tcp_do_newreno, 0, "Enable NewReno Algorithms"); -struct mbuf *m_copym_with_hdrs __P((struct mbuf*, int, int, int, struct mbuf**, int*)); +int tcp_packet_chaining = 50; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, packetchain, CTLFLAG_RW, &tcp_packet_chaining, + 0, "Enable TCP output packet chaining"); + +struct mbuf *m_copym_with_hdrs(struct mbuf*, int, int, int, struct mbuf**, int*); +static long packchain_newlist = 0; +static long packchain_looped = 0; +static long packchain_sent = 0; /* temporary: for testing */ @@ -129,15 +136,36 @@ struct mbuf *m_copym_with_hdrs __P((struct mbuf*, int, int, int, struct mbuf**, extern int ipsec_bypass; #endif +extern int slowlink_wsize; /* window correction for slow links */ +extern u_long route_generation; +extern int fw_enable; /* firewall is on: disable packet chaining */ +extern int ipsec_bypass; + +extern vm_size_t so_cache_zone_element_size; + +static __inline__ u_int16_t +get_socket_id(struct socket * s) +{ + u_int16_t val; + + if (so_cache_zone_element_size == 0) { + return (0); + } + val = (u_int16_t)(((u_int32_t)s) / so_cache_zone_element_size); + if (val == 0) { + val = 0xffff; + } + return (val); +} + /* * Tcp output routine: figure out what should be sent and send it. */ int -tcp_output(tp) - register struct tcpcb *tp; +tcp_output(struct tcpcb *tp) { - register struct socket *so = tp->t_inpcb->inp_socket; - register long len, win; + struct socket *so = tp->t_inpcb->inp_socket; + long len, recwin, sendwin; int off, flags, error; register struct mbuf *m; struct ip *ip = NULL; @@ -148,59 +176,43 @@ tcp_output(tp) register struct tcphdr *th; u_char opt[TCP_MAXOLEN]; unsigned ipoptlen, optlen, hdrlen; - int idle, sendalot; + int idle, sendalot, howmuchsent = 0; + int i, sack_rxmit; + int sack_bytes_rxmt; + struct sackhole *p; + int maxburst = TCP_MAXBURST; struct rmxp_tao *taop; struct rmxp_tao tao_noncached; -#if INET6 - int isipv6; -#endif - int last_off; + int last_off = 0; int m_off; struct mbuf *m_last = 0; struct mbuf *m_head = 0; - - - KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_START, 0,0,0,0,0); + struct mbuf *packetlist = 0; + struct mbuf *lastpacket = 0; #if INET6 - if (isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 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)), - 0,0,0); - } - else + int isipv6 = tp->t_inpcb->inp_vflag & INP_IPV6 ; #endif + short packchain_listadd = 0; + u_int16_t socket_id = get_socket_id(so); + - { - 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); - } /* * Determine length of data that should be transmitted, * and flags that will be used. * If there is some data or critical controls (SYN, RST) * to send, then transmit; otherwise, investigate further. */ - idle = (tp->snd_max == tp->snd_una); -#ifdef __APPLE__ + idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una); if (idle && tp->t_rcvtime >= tp->t_rxtcur) { -#else - if (idle && (ticks - tp->t_rcvtime) >= tp->t_rxtcur) { -#endif /* * 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)) || @@ -215,12 +227,144 @@ tcp_output(tp) else tp->snd_cwnd = tp->t_maxseg * ss_fltsz; } + tp->t_flags &= ~TF_LASTIDLE; + if (idle) { + if (tp->t_flags & TF_MORETOCOME) { + tp->t_flags |= TF_LASTIDLE; + idle = 0; + } + } again: + KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_START, 0,0,0,0,0); + +#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 +#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)), + 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 + * return error or silently do nothing (assuming the address will + * come back before the TCP connection times out). + */ + + if ((tp->t_inpcb->inp_route.ro_rt != NULL && + (tp->t_inpcb->inp_route.ro_rt->generation_id != route_generation)) || (tp->t_inpcb->inp_route.ro_rt == NULL)) { + /* check that the source address is still valid */ + if (ifa_foraddr(tp->t_inpcb->inp_laddr.s_addr) == 0) { + if (tp->t_state >= TCPS_CLOSE_WAIT) { + tcp_close(tp); + return(EADDRNOTAVAIL); + } + + /* set Retransmit timer if it wasn't set + * reset Persist timer and shift register as the + * adversed peer window may not be valid anymore + */ + + if (!tp->t_timer[TCPT_REXMT]) { + tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; + if (tp->t_timer[TCPT_PERSIST]) { + tp->t_timer[TCPT_PERSIST] = 0; + tp->t_rxtshift = 0; + } + } + + if (packetlist) { + error = ip_output_list(packetlist, packchain_listadd, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, + (so->so_options & SO_DONTROUTE), 0); + tp->t_lastchain = 0; + } + if (so->so_flags & SOF_NOADDRAVAIL) + return(EADDRNOTAVAIL); + else + return(0); /* silently ignore and keep data in socket */ + } + } + } + + /* + * If we've recently taken a timeout, snd_max will be greater than + * snd_nxt. There may be SACK information that allows us to avoid + * resending already delivered data. Adjust snd_nxt accordingly. + */ + if (tp->sack_enable && SEQ_LT(tp->snd_nxt, tp->snd_max)) + tcp_sack_adjust(tp); sendalot = 0; off = tp->snd_nxt - tp->snd_una; - win = min(tp->snd_wnd, tp->snd_cwnd); + sendwin = min(tp->snd_wnd, tp->snd_cwnd); + + if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0) + sendwin = min(sendwin, slowlink_wsize); flags = tcp_outflags[tp->t_state]; + /* + * Send any SACK-generated retransmissions. If we're explicitly trying + * to send out new data (when sendalot is 1), bypass this function. + * If we retransmit in fast recovery mode, decrement snd_cwnd, since + * we're replacing a (future) new transmission with a retransmission + * now, and we previously incremented snd_cwnd in tcp_input(). + */ + /* + * Still in sack recovery , reset rxmit flag to zero. + */ + sack_rxmit = 0; + sack_bytes_rxmt = 0; + len = 0; + p = NULL; + if (tp->sack_enable && IN_FASTRECOVERY(tp) && + (p = tcp_sack_output(tp, &sack_bytes_rxmt))) { + long cwin; + + cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt; + if (cwin < 0) + cwin = 0; + /* Do not retransmit SACK segments beyond snd_recover */ + if (SEQ_GT(p->end, tp->snd_recover)) { + /* + * (At least) part of sack hole extends beyond + * snd_recover. Check to see if we can rexmit data + * for this hole. + */ + if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { + /* + * Can't rexmit any more data for this hole. + * That data will be rexmitted in the next + * sack recovery episode, when snd_recover + * moves past p->rxmit. + */ + p = NULL; + goto after_sack_rexmit; + } else + /* Can rexmit part of the current hole */ + len = ((long)ulmin(cwin, + tp->snd_recover - p->rxmit)); + } else + len = ((long)ulmin(cwin, p->end - p->rxmit)); + off = p->rxmit - tp->snd_una; + if (len > 0) { + sack_rxmit = 1; + sendalot = 1; + tcpstat.tcps_sack_rexmits++; + tcpstat.tcps_sack_rexmit_bytes += + min(len, tp->t_maxseg); + } + } +after_sack_rexmit: /* * Get standard flags, and add SYN or FIN if requested by 'hidden' * state flags. @@ -237,7 +381,7 @@ again: * and go to transmit state. */ if (tp->t_force) { - if (win == 0) { + if (sendwin == 0) { /* * If we still have some data to send, then * clear the FIN bit. Usually this would @@ -256,18 +400,58 @@ again: */ if (off < so->so_snd.sb_cc) flags &= ~TH_FIN; - win = 1; + sendwin = 1; } else { tp->t_timer[TCPT_PERSIST] = 0; tp->t_rxtshift = 0; } } - len = (long)ulmin(so->so_snd.sb_cc, win) - off; - - if ((taop = tcp_gettaocache(tp->t_inpcb)) == NULL) { - taop = &tao_noncached; - bzero(taop, sizeof(*taop)); + /* + * If snd_nxt == snd_max and we have transmitted a FIN, the + * offset will be > 0 even if so_snd.sb_cc is 0, resulting in + * a negative length. This can also occur when TCP opens up + * its congestion window while receiving additional duplicate + * acks after fast-retransmit because TCP will reset snd_nxt + * to snd_max after the fast-retransmit. + * + * In the normal retransmit-FIN-only case, however, snd_nxt will + * be set to snd_una, the offset will be 0, and the length may + * wind up 0. + * + * If sack_rxmit is true we are retransmitting from the scoreboard + * in which case len is already set. + */ + if (sack_rxmit == 0) { + if (sack_bytes_rxmt == 0) + len = ((long)ulmin(so->so_snd.sb_cc, sendwin) - off); + else { + long 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); + /* + * Don't remove this (len > 0) check ! + * We explicitly check for len > 0 here (although it + * isn't really necessary), to work around a gcc + * optimization issue - to force gcc to compute + * len above. Without this check, the computation + * of len is bungled by the optimizer. + */ + if (len > 0) { + cwin = tp->snd_cwnd - + (tp->snd_nxt - tp->sack_newdata) - + sack_bytes_rxmt; + if (cwin < 0) + cwin = 0; + len = lmin(len, cwin); + } + } } /* @@ -278,22 +462,23 @@ again: if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { flags &= ~TH_SYN; off--, len++; - if (len > 0 && tp->t_state == TCPS_SYN_SENT && - taop->tao_ccsent == 0) { + if (len > 0 && tp->t_state == TCPS_SYN_SENT) { + if (packetlist) { + error = ip_output_list(packetlist, packchain_listadd, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, + (so->so_options & SO_DONTROUTE), 0); + tp->t_lastchain = 0; + } KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return 0; } } /* - * Be careful not to send data and/or FIN on SYN segments - * in cases when no CC option will be sent. + * Be careful not to send data and/or FIN on SYN segments. * This measure is needed to prevent interoperability problems * with not fully conformant TCP implementations. */ - if ((flags & TH_SYN) && - ((tp->t_flags & TF_NOOPT) || !(tp->t_flags & TF_REQ_CC) || - ((flags & TH_ACK) && !(tp->t_flags & TF_RCVD_CC)))) { + if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) { len = 0; flags &= ~TH_FIN; } @@ -302,7 +487,7 @@ again: /* * If FIN has been sent but not acked, * but we haven't been called to retransmit, - * len will be -1. Otherwise, window shrank + * len will be < 0. Otherwise, window shrank * after we sent into it. If window shrank to 0, * cancel pending retransmit, pull snd_nxt back * to (closed) window, and set the persist timer @@ -310,7 +495,7 @@ again: * close completely, just wait for an ACK. */ len = 0; - if (win == 0) { + if (sendwin == 0) { tp->t_timer[TCPT_REXMT] = 0; tp->t_rxtshift = 0; tp->snd_nxt = tp->snd_una; @@ -318,24 +503,41 @@ again: tcp_setpersist(tp); } } + + /* + * 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. + */ if (len > tp->t_maxseg) { len = tp->t_maxseg; + howmuchsent += len; sendalot = 1; } - if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) - flags &= ~TH_FIN; + if (sack_rxmit) { + if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc)) + flags &= ~TH_FIN; + } else { + if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc)) + flags &= ~TH_FIN; + } - win = sbspace(&so->so_rcv); + if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0 ) /* Clips window size for slow links */ + recwin = min(sbspace(&so->so_rcv), slowlink_wsize); + else + recwin = sbspace(&so->so_rcv); /* - * Sender silly window avoidance. If connection is idle - * and can send all data, a maximum segment, - * at least a maximum default-size segment do it, - * or are forced, do it; otherwise don't bother. - * If peer's buffer is tiny, then send - * when window is at least half open. - * If retransmitting (possibly after persist timer forced us - * to send into a small window), then must resend. + * Sender silly window avoidance. We transmit under the following + * conditions when len is non-zero: + * + * - We have a full segment + * - 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) @@ -349,7 +551,9 @@ again: goto send; if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) goto send; - if (SEQ_LT(tp->snd_nxt, tp->snd_max)) + if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */ + goto send; + if (sack_rxmit) goto send; } @@ -359,14 +563,15 @@ again: * next expected input). If the difference is at least two * max size segments, or at least 50% 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. */ - if (win > 0) { + if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN)) { /* * "adv" is the amount we can increase the window, * taking into account that we are limited by * TCP_MAXWIN << tp->rcv_scale. */ - long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) - + long adv = min(recwin, (long)TCP_MAXWIN << tp->rcv_scale) - (tp->rcv_adv - tp->rcv_nxt); if (adv >= (long) (2 * tp->t_maxseg)) @@ -376,7 +581,8 @@ again: } /* - * Send if we owe peer an ACK. + * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW + * is also a catch-all for the retransmit timer timeout case. */ if (tp->t_flags & TF_ACKNOW) goto send; @@ -387,13 +593,22 @@ again: goto send; /* * If our state indicates that FIN should be sent - * and we have not yet done so, or we're retransmitting the FIN, - * then we need to send. + * and we have not yet done so, then we need to send. */ if (flags & TH_FIN && ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) goto send; - + /* + * In SACK, it is possible for tcp_output to fail to send a segment + * 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 && + tp->t_timer[TCPT_PERSIST] == 0) { + tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; + goto just_return; + } /* * TCP window updates are not reliable, rather a polling protocol * using ``persist'' packets is used to insure receipt of window @@ -421,10 +636,15 @@ again: tp->t_rxtshift = 0; tcp_setpersist(tp); } - +just_return: /* - * No reason to send a segment, 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. */ + if (packetlist) { + error = ip_output_list(packetlist, packchain_listadd, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, + (so->so_options & SO_DONTROUTE), 0); + } KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return (0); @@ -486,78 +706,92 @@ send: optlen += TCPOLEN_TSTAMP_APPA; } - /* - * Send `CC-family' options if our side wants to use them (TF_REQ_CC), - * options are allowed (!TF_NOOPT) and it's not a RST. - */ - if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && - (flags & TH_RST) == 0) { - switch (flags & (TH_SYN|TH_ACK)) { - /* - * This is a normal ACK, send CC if we received CC before - * from our peer. + if (tp->sack_enable && ((tp->t_flags & TF_NOOPT) == 0)) { + /* + * Tack on the SACK permitted option *last*. + * And do padding of options after tacking this on. + * This is because of MSS, TS, WinScale and Signatures are + * all present, we have just 2 bytes left for the SACK + * permitted option, which is just enough. */ - case TH_ACK: - if (!(tp->t_flags & TF_RCVD_CC)) - break; - /*FALLTHROUGH*/ - /* - * We can only get here in T/TCP's SYN_SENT* state, when - * we're a sending a non-SYN segment without waiting for - * the ACK of our SYN. A check above assures that we only - * do this if our peer understands T/TCP. + * If this is the first SYN of connection (not a SYN + * ACK), include SACK permitted option. If this is a + * SYN ACK, include SACK permitted option if peer has + * already done so. This is only for active connect, + * since the syncache takes care of the passive connect. */ - case 0: - opt[optlen++] = TCPOPT_NOP; - opt[optlen++] = TCPOPT_NOP; - opt[optlen++] = TCPOPT_CC; - opt[optlen++] = TCPOLEN_CC; - *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); - - optlen += 4; - break; + if ((flags & TH_SYN) && + (!(flags & TH_ACK) || (tp->t_flags & TF_SACK_PERMIT))) { + u_char *bp; + bp = (u_char *)opt + optlen; + + *bp++ = TCPOPT_SACK_PERMITTED; + *bp++ = TCPOLEN_SACK_PERMITTED; + optlen += TCPOLEN_SACK_PERMITTED; + } /* - * This is our initial SYN, check whether we have to use - * CC or CC.new. + * Send SACKs if necessary. This should be the last + * option processed. Only as many SACKs are sent as + * are permitted by the maximum options size. + * + * In general, SACK blocks consume 8*n+2 bytes. + * So a full size SACK blocks option is 34 bytes + * (to generate 4 SACK blocks). At a minimum, + * we need 10 bytes (to generate 1 SACK block). + * If TCP Timestamps (12 bytes) and TCP Signatures + * (18 bytes) are both present, we'll just have + * 10 bytes for SACK options 40 - (12 + 18). */ - case TH_SYN: - opt[optlen++] = TCPOPT_NOP; - opt[optlen++] = TCPOPT_NOP; - opt[optlen++] = tp->t_flags & TF_SENDCCNEW ? - TCPOPT_CCNEW : TCPOPT_CC; - opt[optlen++] = TCPOLEN_CC; - *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send); - optlen += 4; - break; + 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; + u_char *bp = (u_char *)opt + optlen; + u_int32_t *lp; - /* - * This is a SYN,ACK; send CC and CC.echo if we received - * CC from our peer. - */ - case (TH_SYN|TH_ACK): - if (tp->t_flags & TF_RCVD_CC) { - opt[optlen++] = TCPOPT_NOP; - opt[optlen++] = TCPOPT_NOP; - opt[optlen++] = TCPOPT_CC; - opt[optlen++] = TCPOLEN_CC; - *(u_int32_t *)&opt[optlen] = - htonl(tp->cc_send); - optlen += 4; - opt[optlen++] = TCPOPT_NOP; - opt[optlen++] = TCPOPT_NOP; - opt[optlen++] = TCPOPT_CCECHO; - opt[optlen++] = TCPOLEN_CC; - *(u_int32_t *)&opt[optlen] = - htonl(tp->cc_recv); - optlen += 4; + nsack = (MAX_TCPOPTLEN - optlen - 2) / TCPOLEN_SACK; + nsack = min(nsack, tp->rcv_numsacks); + sackoptlen = (2 + nsack * TCPOLEN_SACK); + + /* + * First we need to pad options so that the + * SACK blocks can start at a 4-byte boundary + * (sack option and length are at a 2 byte offset). + */ + padlen = (MAX_TCPOPTLEN - optlen - sackoptlen) % 4; + optlen += padlen; + while (padlen-- > 0) + *bp++ = TCPOPT_NOP; + + tcpstat.tcps_sack_send_blocks++; + *bp++ = TCPOPT_SACK; + *bp++ = sackoptlen; + lp = (u_int32_t *)bp; + for (i = 0; i < nsack; i++) { + struct sackblk sack = tp->sackblks[i]; + *lp++ = htonl(sack.start); + *lp++ = htonl(sack.end); } - break; + optlen += sackoptlen; } - } + } + + /* Pad TCP options to a 4 byte boundary */ + if (optlen < MAX_TCPOPTLEN && (optlen % sizeof(u_int32_t))) { + int pad = sizeof(u_int32_t) - (optlen % sizeof(u_int32_t)); + u_char *bp = (u_char *)opt + optlen; + + optlen += pad; + while (pad) { + *bp++ = TCPOPT_EOL; + pad--; + } + } + + hdrlen += optlen; - hdrlen += optlen; #if INET6 if (isipv6) ipoptlen = ip6_optlen(tp->t_inpcb); @@ -567,9 +801,8 @@ send: if (tp->t_inpcb->inp_options) { ipoptlen = tp->t_inpcb->inp_options->m_len - offsetof(struct ipoption, ipopt_list); - } else { + } else ipoptlen = 0; - } } #if IPSEC if (ipsec_bypass == 0) @@ -588,6 +821,7 @@ send: */ flags &= ~TH_FIN; len = tp->t_maxopd - optlen - ipoptlen; + howmuchsent += len; sendalot = 1; } @@ -668,6 +902,12 @@ send: 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 (m != NULL) m_freem(m); + error = 0; /* should we return an error? */ + goto out; + } m_copydata(so->so_snd.sb_mb, off, (int) len, mtod(m, caddr_t) + hdrlen); m->m_len += len; @@ -694,7 +934,13 @@ send: m_last = NULL; last_off = off + len; m_head = so->so_snd.sb_mb; - + + /* makes sure we still have data left to be sent at this point */ + if (m_head == NULL) { + error = 0; /* should we return an error? */ + goto out; + } + /* * 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 @@ -740,7 +986,7 @@ send: m->m_data += max_linkhdr; m->m_len = hdrlen; } - m->m_pkthdr.rcvif = (struct ifnet *)0; + m->m_pkthdr.rcvif = 0; #if INET6 if (isipv6) { ip6 = mtod(m, struct ip6_hdr *); @@ -777,10 +1023,16 @@ send: * case, since we know we aren't doing a retransmission. * (retransmit and persist are mutually exclusive...) */ - if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST]) - th->th_seq = htonl(tp->snd_nxt); - else - th->th_seq = htonl(tp->snd_max); + if (sack_rxmit == 0) { + if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST]) + th->th_seq = htonl(tp->snd_nxt); + else + th->th_seq = htonl(tp->snd_max); + } else { + th->th_seq = htonl(p->rxmit); + p->rxmit += len; + tp->sackhint.sack_bytes_rexmit += len; + } th->th_ack = htonl(tp->rcv_nxt); if (optlen) { bcopy(opt, th + 1, optlen); @@ -791,13 +1043,33 @@ send: * Calculate receive window. Don't shrink window, * but avoid silly window syndrome. */ - if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg) - win = 0; - if (win < (long)(tp->rcv_adv - tp->rcv_nxt)) - win = (long)(tp->rcv_adv - tp->rcv_nxt); - if (win > (long)TCP_MAXWIN << tp->rcv_scale) - win = (long)TCP_MAXWIN << tp->rcv_scale; - th->th_win = htons((u_short) (win>>tp->rcv_scale)); + if (recwin < (long)(so->so_rcv.sb_hiwat / 4) && recwin < (long)tp->t_maxseg) + recwin = 0; + if (recwin < (long)(tp->rcv_adv - tp->rcv_nxt)) + recwin = (long)(tp->rcv_adv - tp->rcv_nxt); + if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0) { + if (recwin > (long)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)); + } + + /* + * Adjust the RXWIN0SENT flag - indicate that we have advertised + * a 0 window. This may cause the remote transmitter to stall. This + * flag tells soreceive() to disable delayed acknowledgements when + * draining the buffer. This can occur if the receiver is attempting + * to read more data then can be buffered prior to transmitting on + * the connection. + */ + if (recwin == 0) + tp->t_flags |= TF_RXWIN0SENT; + else + tp->t_flags &= ~TF_RXWIN0SENT; if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); th->th_flags |= TH_URG; @@ -831,10 +1103,6 @@ send: if (len + optlen) th->th_sum = in_addword(th->th_sum, htons((u_short)(optlen + len))); - - /* IP version must be set here for ipv4/ipv6 checking later */ - KASSERT(ip->ip_v == IPVERSION, - ("%s: IP version incorrect: %d", __FUNCTION__, ip->ip_v)); } /* @@ -855,6 +1123,8 @@ send: tp->t_flags |= TF_SENTFIN; } } + if (sack_rxmit) + goto timer; tp->snd_nxt += len; if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { tp->snd_max = tp->snd_nxt; @@ -877,17 +1147,31 @@ send: * Initialize shift counter which is used for backoff * of retransmit time. */ +timer: if (tp->t_timer[TCPT_REXMT] == 0 && - tp->snd_nxt != tp->snd_una) { - tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; + ((sack_rxmit && tp->snd_nxt != tp->snd_max) || + tp->snd_nxt != tp->snd_una)) { if (tp->t_timer[TCPT_PERSIST]) { tp->t_timer[TCPT_PERSIST] = 0; tp->t_rxtshift = 0; } + tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; } - } else - if (SEQ_GT(tp->snd_nxt + len, tp->snd_max)) + } else { + /* + * Persist case, update snd_max but since we are in + * persist mode (no window) we do not update snd_nxt. + */ + int xlen = len; + if (flags & TH_SYN) + ++xlen; + if (flags & TH_FIN) { + ++xlen; + tp->t_flags |= TF_SENTFIN; + } + if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max)) tp->snd_max = tp->snd_nxt + len; + } #if TCPDEBUG /* @@ -928,17 +1212,18 @@ send: goto out; } #endif /*IPSEC*/ + m->m_pkthdr.socket_id = socket_id; error = ip6_output(m, tp->t_inpcb->in6p_outputopts, &tp->t_inpcb->in6p_route, - (so->so_options & SO_DONTROUTE), NULL, NULL); + (so->so_options & SO_DONTROUTE), NULL, NULL, 0); } else #endif /* INET6 */ { struct rtentry *rt; ip->ip_len = m->m_pkthdr.len; #if INET6 - if (INP_CHECK_SOCKAF(so, AF_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 @@ -984,29 +1269,78 @@ send: if (ipsec_bypass == 0) ipsec_setsocket(m, so); #endif /*IPSEC*/ - error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, - (so->so_options & SO_DONTROUTE), 0); - } + + /* + * The socket is kept locked while sending out packets in ip_output, even if packet chaining is not active. + */ + + m->m_pkthdr.socket_id = socket_id; + if (packetlist) { + m->m_nextpkt = NULL; + lastpacket->m_nextpkt = m; + lastpacket = m; + packchain_listadd++; + } + else { + m->m_nextpkt = NULL; + packchain_newlist++; + packetlist = lastpacket = m; + packchain_listadd=0; + } + + if ((ipsec_bypass == 0) || fw_enable || sendalot == 0 || (tp->t_state != TCPS_ESTABLISHED) || + (tp->snd_cwnd <= (tp->snd_wnd / 4)) || + (tp->t_flags & (TH_PUSH | TF_ACKNOW)) || tp->t_force != 0 || + packchain_listadd >= tcp_packet_chaining) { + lastpacket->m_nextpkt = 0; + error = ip_output_list(packetlist, packchain_listadd, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, + (so->so_options & SO_DONTROUTE), 0); + tp->t_lastchain = packchain_listadd; + packchain_sent++; + packetlist = NULL; + if (error == 0) + howmuchsent = 0; + } + 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) { /* * We know that the packet was lost, so back out the * sequence number advance, if any. */ - if (tp->t_force == 0 || !tp->t_timer[TCPT_PERSIST]) { + if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) { /* * No need to check for TH_FIN here because * the TF_SENTFIN flag handles that case. */ - if ((flags & TH_SYN) == 0) - tp->snd_nxt -= len; + if ((flags & TH_SYN) == 0) { + if (sack_rxmit) { + p->rxmit -= howmuchsent; + tp->sackhint.sack_bytes_rexmit -= howmuchsent; + } else + tp->snd_nxt -= howmuchsent; + } } + howmuchsent = 0; out: 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 (packetlist) + m_freem_list(packetlist); + tp->t_lastchain = 0; KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return (0); } @@ -1018,18 +1352,28 @@ out: * not do so here. */ tcp_mtudisc(tp->t_inpcb, 0); + if (packetlist) + m_freem_list(packetlist); + tp->t_lastchain = 0; 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; + if (packetlist) + m_freem_list(packetlist); + tp->t_lastchain = 0; KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return (0); } + if (packetlist) + m_freem_list(packetlist); + tp->t_lastchain = 0; KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return (error); } +sentit: tcpstat.tcps_sndtotal++; /* @@ -1038,13 +1382,14 @@ out: * then remember the size of the advertised window. * Any pending ACK has now been sent. */ - if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) - tp->rcv_adv = tp->rcv_nxt + win; + 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); - if (sendalot) + + KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END,0,0,0,0,0); + if (sendalot && (!tcp_do_newreno || --maxburst)) goto again; - KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return (0); } @@ -1053,7 +1398,6 @@ tcp_setpersist(tp) register struct tcpcb *tp; { int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; - int tt; if (tp->t_timer[TCPT_REXMT]) panic("tcp_setpersist: retransmit pending");