X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/8f6c56a50524aa785f7e596d52dddfb331e18961..b0d623f7f2ae71ed96e60569f61f9a9a27016e80:/bsd/netinet/tcp_output.c diff --git a/bsd/netinet/tcp_output.c b/bsd/netinet/tcp_output.c index b81dbd53a..9d37f63fd 100644 --- a/bsd/netinet/tcp_output.c +++ b/bsd/netinet/tcp_output.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2008 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -60,6 +60,12 @@ * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95 * $FreeBSD: src/sys/netinet/tcp_output.c,v 1.39.2.10 2001/07/07 04:30:38 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. + */ #define _IP_VHL @@ -75,9 +81,11 @@ #include #include +#include #include #include +#include #include #include #include @@ -102,6 +110,10 @@ #include #endif /*IPSEC*/ +#if CONFIG_MACF_NET +#include +#endif /* MAC_SOCKET */ + #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) @@ -111,7 +123,7 @@ extern struct mbuf *m_copypack(); #endif -static int path_mtu_discovery = 1; +int path_mtu_discovery = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW, &path_mtu_discovery, 1, "Enable Path MTU Discovery"); @@ -119,7 +131,7 @@ 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 = 4; /* starts with four segments max */ +int ss_fltsz_local = 8; /* starts with eight 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"); @@ -127,15 +139,30 @@ 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, + &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, + 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, + 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, 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; +int tcp_output_unlocked = 1; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, socket_unlocked_on_output, CTLFLAG_RW, &tcp_output_unlocked, + 0, "Unlock TCP when sending packets down to IP"); +static int32_t packchain_newlist = 0; +static int32_t packchain_looped = 0; +static int32_t packchain_sent = 0; /* temporary: for testing */ #if IPSEC @@ -143,11 +170,20 @@ 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; +#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; + +static int tcp_ip_output(struct socket *, struct tcpcb *, struct mbuf *, int, + struct mbuf *, int, int); static __inline__ u_int16_t get_socket_id(struct socket * s) @@ -157,7 +193,7 @@ 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; } @@ -166,13 +202,29 @@ get_socket_id(struct socket * s) /* * Tcp output routine: figure out what should be sent and send it. + * + * Returns: 0 Success + * EADDRNOTAVAIL + * ENOBUFS + * EMSGSIZE + * EHOSTUNREACH + * ENETDOWN + * ip_output_list:ENOMEM + * ip_output_list:EADDRNOTAVAIL + * ip_output_list:ENETUNREACH + * ip_output_list:EHOSTUNREACH + * ip_output_list:EACCES + * ip_output_list:EMSGSIZE + * ip_output_list:ENOBUFS + * ip_output_list:??? [ignorable: mostly IPSEC/firewall/DLIL] + * ip6_output:??? [IPV6 only] */ 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; @@ -182,26 +234,29 @@ tcp_output(struct tcpcb *tp) register struct tcphdr *th; u_char opt[TCP_MAXOLEN]; unsigned ipoptlen, optlen, hdrlen; - int idle, sendalot, howmuchsent = 0; + int idle, sendalot, lost = 0; int i, sack_rxmit; + int tso = 0; int sack_bytes_rxmt; struct sackhole *p; - +#ifdef IPSEC + unsigned ipsec_optlen = 0; +#endif int maxburst = TCP_MAXBURST; - struct rmxp_tao *taop; - struct rmxp_tao tao_noncached; int last_off = 0; int m_off; - struct mbuf *m_last = 0; - struct mbuf *m_head = 0; - struct mbuf *packetlist = 0; - struct mbuf *lastpacket = 0; + struct mbuf *m_last = 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; /* * Determine length of data that should be transmitted, @@ -267,13 +322,23 @@ again: * return error or silently do nothing (assuming the address will * come back before the TCP connection times out). */ + rt = tp->t_inpcb->inp_route.ro_rt; + if (rt != NULL && (!(rt->rt_flags & RTF_UP) || + rt->generation_id != route_generation)) { + struct ifnet *ifp; + struct in_ifaddr *ia; + + /* disable multipages at the socket */ + somultipages(so, FALSE); + + /* Disable TSO for the socket until we know more */ + tp->t_flags &= ~TF_TSO; - 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 ((ia = ifa_foraddr(tp->t_inpcb->inp_laddr.s_addr)) == NULL) { + if (tp->t_state >= TCPS_CLOSE_WAIT) { - tcp_close(tp); + tcp_drop(tp, EADDRNOTAVAIL); return(EADDRNOTAVAIL); } @@ -290,16 +355,46 @@ again: } } - 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) + if (tp->t_pktlist_head != NULL) + m_freem_list(tp->t_pktlist_head); + TCP_PKTLIST_CLEAR(tp); + + /* drop connection if source address isn't available */ + if (so->so_flags & SOF_NOADDRAVAIL) { + tcp_drop(tp, EADDRNOTAVAIL); return(EADDRNOTAVAIL); + } else - return(0); /* silently ignore and keep data in socket */ + return(0); /* silently ignore, keep data in socket: address may be back */ } + ifafree(&ia->ia_ifa); + + /* + * Address is still valid; check for multipages capability + * again in case the outgoing interface has changed. + */ + RT_LOCK(rt); + if ((ifp = rt->rt_ifp) != NULL) { + somultipages(so, (ifp->if_hwassist & IFNET_MULTIPAGES)); + 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: + * 1) it is disabled via the sysctl + * 2) the route isn't up + * 3) the MTU is locked (if it is, then discovery has been + * disabled) + */ + + if (!path_mtu_discovery || ((rt != NULL) && + (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) + tp->t_flags &= ~TF_PMTUD; + else + tp->t_flags |= TF_PMTUD; + + RT_UNLOCK(rt); } } @@ -334,7 +429,7 @@ 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) @@ -357,18 +452,20 @@ 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; + 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); } + else + len = 0; } after_sack_rexmit: /* @@ -430,17 +527,17 @@ 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 @@ -455,8 +552,10 @@ after_sack_rexmit: sack_bytes_rxmt; if (cwin < 0) cwin = 0; - len = lmin(len, cwin); + len = imin(len, cwin); } + else + len = 0; } } @@ -469,13 +568,29 @@ after_sack_rexmit: flags &= ~TH_SYN; off--, len++; 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; + while (!(tp->t_flags & TF_SENDINPROG) && + 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), (sack_rxmit | (sack_bytes_rxmt != 0))); + + 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) { + tp->t_flags &= ~TF_CLOSING; + (void) tcp_close(tp); } - KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); - return 0; + KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, + 0,0,0,0,0); + return 0; } } @@ -511,14 +626,53 @@ after_sack_rexmit: } /* - * 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. + * 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 + /* + * 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; - howmuchsent += len; - 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)) @@ -528,16 +682,13 @@ after_sack_rexmit: flags &= ~TH_FIN; } - 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); + recwin = tcp_sbspace(tp); /* * Sender silly window avoidance. We transmit under the following * conditions when len is non-zero: * - * - We have a full segment + * - 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) @@ -546,19 +697,29 @@ after_sack_rexmit: * - we need to retransmit */ if (len) { - if (len == tp->t_maxseg) + if (len >= tp->t_maxseg) { + tp->t_flags |= TF_MAXSEGSNT; goto send; + } if (!(tp->t_flags & TF_MORETOCOME) && - (idle || tp->t_flags & TF_NODELAY) && + (idle || tp->t_flags & TF_NODELAY || tp->t_flags & TF_MAXSEGSNT) && (tp->t_flags & TF_NOPUSH) == 0 && - len + off >= so->so_snd.sb_cc) + len + off >= so->so_snd.sb_cc) { + tp->t_flags &= ~TF_MAXSEGSNT; goto send; - if (tp->t_force) + } + if (tp->t_force) { + tp->t_flags &= ~TF_MAXSEGSNT; goto send; - if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) + } + 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)) /* retransmit case */ + } + if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { + tp->t_flags &= ~TF_MAXSEGSNT; goto send; + } if (sack_rxmit) goto send; } @@ -577,13 +738,22 @@ after_sack_rexmit: * taking into account that we are limited by * TCP_MAXWIN << tp->rcv_scale. */ - long adv = min(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 ((tp->last_ack_sent != tp->rcv_nxt) || (((recwin + adv) >> tp->rcv_scale) > recwin)) + goto send; + } + if (2 * adv >= (int32_t) so->so_rcv.sb_hiwat) + goto send; } /* @@ -609,7 +779,7 @@ 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) && + 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; @@ -647,9 +817,22 @@ 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); + while (!(tp->t_flags & TF_SENDINPROG) && 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), (sack_rxmit | (sack_bytes_rxmt != 0))); + + 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) { + tp->t_flags &= ~TF_CLOSING; + (void) tcp_close(tp); } KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return (0); @@ -692,7 +875,90 @@ send: optlen += 4; } } + } + + /* + RFC 3168 states that: + - If you ever sent an ECN-setup SYN/SYN-ACK you must be prepared + to handle the TCP ECE flag, even if you also later send a + non-ECN-setup SYN/SYN-ACK. + - If you ever send a non-ECN-setup SYN/SYN-ACK, you must not set + the ip ECT flag. + + It is not clear how the ECE flag would ever be set if you never + set the IP ECT flag on outbound packets. All the same, we use + the TE_SETUPSENT to indicate that we have committed to handling + the TCP ECE flag correctly. We use the TE_SENDIPECT to indicate + whether or not we should set the IP ECT flag on outbound packets. + */ + /* + * For a SYN-ACK, send an ECN setup SYN-ACK + */ + if (tcp_ecn_inbound && (flags & (TH_SYN | TH_ACK)) == (TH_SYN | TH_ACK)) { + if ((tp->ecn_flags & TE_SETUPRECEIVED) != 0) { + if ((tp->ecn_flags & TE_SETUPSENT) == 0) { + /* Setting TH_ECE makes this an ECN-setup SYN-ACK */ + flags |= TH_ECE; + + /* + * Record that we sent the ECN-setup and default to + * setting IP ECT. + */ + tp->ecn_flags |= (TE_SETUPSENT | TE_SENDIPECT); + } + else { + /* + * We sent an ECN-setup SYN-ACK but it was dropped. + * Fallback to non-ECN-setup SYN-ACK and clear flag + * that to indicate we should not send data with IP ECT set. + * + * Pretend we didn't receive an ECN-setup SYN. + */ + tp->ecn_flags &= ~TE_SETUPRECEIVED; + } + } + } + else if (tcp_ecn_outbound && (flags & (TH_SYN | TH_ACK)) == TH_SYN) { + if ((tp->ecn_flags & TE_SETUPSENT) == 0) { + /* Setting TH_ECE and TH_CWR makes this an ECN-setup SYN */ + flags |= (TH_ECE | TH_CWR); + + /* + * Record that we sent the ECN-setup and default to + * setting IP ECT. + */ + tp->ecn_flags |= (TE_SETUPSENT | TE_SENDIPECT); + } + else { + /* + * We sent an ECN-setup SYN but it was dropped. + * Fall back to no ECN and clear flag indicating + * we should send data with IP ECT set. + */ + tp->ecn_flags &= ~TE_SENDIPECT; + } + } + + /* + * Check if we should set the TCP CWR flag. + * CWR flag is sent when we reduced the congestion window because + * we received a TCP ECE or we performed a fast retransmit. We + * never set the CWR flag on retransmitted packets. We only set + * 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)) { + flags |= TH_CWR; + tp->ecn_flags &= ~TE_SENDCWR; + } + + /* + * Check if we should set the TCP ECE flag. + */ + if ((tp->ecn_flags & TE_SENDECE) != 0 && len == 0) { + flags |= TH_ECE; + } /* * Send a timestamp and echo-reply if this is a SYN and our side @@ -804,15 +1070,14 @@ send: else #endif { - if (tp->t_inpcb->inp_options) { - ipoptlen = tp->t_inpcb->inp_options->m_len - + if (tp_inp_options) { + ipoptlen = tp_inp_options->m_len - offsetof(struct ipoption, ipopt_list); } else ipoptlen = 0; } #if IPSEC - if (ipsec_bypass == 0) - ipoptlen += ipsec_hdrsiz_tcp(tp); + ipoptlen += ipsec_optlen; #endif /* @@ -820,15 +1085,34 @@ 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; - howmuchsent += len; - 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*/ @@ -849,7 +1133,7 @@ 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; } else { @@ -883,7 +1167,7 @@ send: m = NULL; #if INET6 if (MHLEN < hdrlen + max_linkhdr) { - MGETHDR(m, M_DONTWAIT, MT_HEADER); + MGETHDR(m, M_DONTWAIT, MT_HEADER); /* MAC-OK */ if (m == NULL) { error = ENOBUFS; goto out; @@ -900,7 +1184,7 @@ send: #endif if (len <= MHLEN - hdrlen - max_linkhdr) { if (m == NULL) { - MGETHDR(m, M_DONTWAIT, MT_HEADER); + MGETHDR(m, M_DONTWAIT, MT_HEADER); /* MAC-OK */ if (m == NULL) { error = ENOBUFS; goto out; @@ -909,7 +1193,7 @@ send: 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; @@ -936,7 +1220,7 @@ 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) + if (m_head != so->so_snd.sb_mb || sack_rxmit || last_off != off) m_last = NULL; last_off = off + len; m_head = so->so_snd.sb_mb; @@ -951,7 +1235,7 @@ 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_last, &m_off)) == NULL) { error = ENOBUFS; goto out; } @@ -978,7 +1262,7 @@ send: else tcpstat.tcps_sndwinup++; - MGETHDR(m, M_DONTWAIT, MT_HEADER); + MGETHDR(m, M_DONTWAIT, MT_HEADER); /* MAC-OK */ if (m == NULL) { error = ENOBUFS; goto out; @@ -993,6 +1277,9 @@ send: m->m_len = hdrlen; } m->m_pkthdr.rcvif = 0; +#if CONFIG_MACF_NET + mac_mbuf_label_associate_inpcb(tp->t_inpcb, m); +#endif #if INET6 if (isipv6) { ip6 = mtod(m, struct ip6_hdr *); @@ -1006,6 +1293,10 @@ send: th = (struct tcphdr *)(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)) { + ip->ip_tos = IPTOS_ECN_ECT0; + } } /* @@ -1040,6 +1331,8 @@ 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; @@ -1049,18 +1342,18 @@ 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)); + 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; + 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)); } @@ -1109,7 +1402,25 @@ send: if (len + optlen) th->th_sum = in_addword(th->th_sum, htons((u_short)(optlen + len))); - } + } + + /* + * Enable TSO and specify the size of the segments. + * The TCP pseudo header checksum is always provided. + * XXX: Fixme: This is currently not the case for IPv6. + */ + 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 @@ -1183,7 +1494,7 @@ timer: /* * Trace. */ - if (so->so_options & SO_DEBUG) + if (so_options & SO_DEBUG) tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0); #endif @@ -1220,13 +1531,12 @@ timer: #endif /*IPSEC*/ m->m_pkthdr.socket_id = socket_id; error = ip6_output(m, - tp->t_inpcb->in6p_outputopts, + inp6_pktopts, &tp->t_inpcb->in6p_route, - (so->so_options & SO_DONTROUTE), NULL, NULL, 0); + (so_options & SO_DONTROUTE), NULL, NULL, 0); } else #endif /* INET6 */ { - struct rtentry *rt; ip->ip_len = m->m_pkthdr.len; #if INET6 if (isipv6) @@ -1237,7 +1547,7 @@ timer: else #endif /* INET6 */ ip->ip_ttl = tp->t_inpcb->inp_ip_ttl; /* XXX */ - ip->ip_tos = tp->t_inpcb->inp_ip_tos; /* XXX */ + ip->ip_tos |= (tp->t_inpcb->inp_ip_tos & ~IPTOS_ECN_MASK); /* XXX */ #if INET6 @@ -1259,18 +1569,18 @@ timer: } /* - * See if we should do MTU discovery. We do it only if the following - * are true: - * 1) we have a valid route to the destination - * 2) the MTU is not locked (if it is, then discovery has been - * disabled) + * See if we should do MTU discovery. + * Look at the flag updated on the following criterias: + * 1) Path MTU discovery is authorized by the sysctl + * 2) The route isn't set yet (unlikely but could happen) + * 3) The route is up + * 4) the MTU is not locked (if it is, then discovery has been + * disabled for that route) */ - if (path_mtu_discovery - && (rt = tp->t_inpcb->inp_route.ro_rt) - && rt->rt_flags & RTF_UP - && !(rt->rt_rmx.rmx_locks & RTV_MTU)) { + + if (path_mtu_discovery && (tp->t_flags & TF_PMTUD)) ip->ip_off |= IP_DF; - } + #if IPSEC if (ipsec_bypass == 0) ipsec_setsocket(m, so); @@ -1279,38 +1589,63 @@ timer: /* * The socket is kept locked while sending out packets in ip_output, even if packet chaining is not active. */ - + lost = 0; 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; + m->m_nextpkt = NULL; + tp->t_pktlist_sentlen += len; + tp->t_lastchain++; + if (tp->t_pktlist_head != NULL) { + tp->t_pktlist_tail->m_nextpkt = m; + tp->t_pktlist_tail = m; + } else { packchain_newlist++; - packetlist = lastpacket = m; - packchain_listadd=0; + tp->t_pktlist_head = tp->t_pktlist_tail = m; } - 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; + 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) { + 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), (sack_rxmit | (sack_bytes_rxmt != 0))); + + tp->t_flags &= ~TF_SENDINPROG; + if (error) { + /* + * Take into account the rest of unsent + * packets in the packet list for this tcp + * into "lost", since we're about to free + * the whole list below. + */ + lost += tp->t_pktlist_sentlen; + break; + } else { + lost = 0; + } + } + /* tcp was closed while we were in ip; resume close */ + if ((tp->t_flags & (TF_CLOSING|TF_SENDINPROG)) == TF_CLOSING) { + tp->t_flags &= ~TF_CLOSING; + (void) tcp_close(tp); + return (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; @@ -1319,10 +1654,12 @@ timer: } } if (error) { - /* - * We know that the packet was lost, so back out the - * sequence number advance, if any. + * Assume that the packets were lost, so back out the + * sequence number advance, if any. Note that the "lost" + * variable represents the amount of user data sent during + * the recent call to ip_output_list() plus the amount of + * user data in the packet list for this tcp at the moment. */ if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) { /* @@ -1331,22 +1668,25 @@ timer: */ if ((flags & TH_SYN) == 0) { if (sack_rxmit) { - p->rxmit -= howmuchsent; - tp->sackhint.sack_bytes_rexmit -= howmuchsent; + p->rxmit -= lost; + tp->sackhint.sack_bytes_rexmit -= lost; } else - tp->snd_nxt -= howmuchsent; + tp->snd_nxt -= lost; } } - howmuchsent = 0; out: + if (tp->t_pktlist_head != NULL) + m_freem_list(tp->t_pktlist_head); + 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 (packetlist) - m_freem_list(packetlist); - tp->t_lastchain = 0; + if (!tp->t_timer[TCPT_REXMT] && + !tp->t_timer[TCPT_PERSIST]) + tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; + + tp->snd_cwnd = tp->t_maxseg; + tp->t_bytes_acked = 0; + KERNEL_DEBUG(DBG_FNC_TCP_OUTPUT | DBG_FUNC_END, 0,0,0,0,0); return (0); } @@ -1356,30 +1696,30 @@ 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); - 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++; /* @@ -1388,7 +1728,7 @@ sentit: * 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)) + 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); @@ -1399,6 +1739,100 @@ sentit: return (0); } +static int +tcp_ip_output(struct socket *so, struct tcpcb *tp, struct mbuf *pkt, + int cnt, struct mbuf *opt, int flags, int sack_in_progress) +{ + int error = 0; + boolean_t chain; + boolean_t unlocked = FALSE; + struct inpcb *inp = tp->t_inpcb; + struct ip_out_args ipoa; + struct route ro; + + /* If socket was bound to an ifindex, tell ip_output about it */ + ipoa.ipoa_ifscope = (inp->inp_flags & INP_BOUND_IF) ? + inp->inp_boundif : IFSCOPE_NONE; + flags |= IP_OUTARGS; + + /* Copy the cached route and take an extra reference */ + inp_route_copyout(inp, &ro); + + /* + * Make sure ACK/DELACK conditions are cleared before + * we unlock the socket. + */ + tp->t_flags &= ~(TF_ACKNOW | TF_DELACK); + + /* + * If allowed, unlock TCP socket while in IP + * but only if the connection is established and + * if we're not sending from an upcall. + */ + if (tcp_output_unlocked && ((so->so_flags & SOF_UPCALLINUSE) == 0) && + (tp->t_state == TCPS_ESTABLISHED) && (sack_in_progress == 0)) { + unlocked = TRUE; + socket_unlock(so, 0); + } + + /* + * Don't send down a chain of packets when: + * - TCP chaining is disabled + * - there is an IPsec rule set + * - there is a non default rule set for the firewall + */ + + chain = tcp_packet_chaining > 1 +#if IPSEC + && ipsec_bypass +#endif +#if IPFIREWALL + && (fw_enable == 0 || fw_bypass) +#endif + ; // I'm important, not extraneous + + + while (pkt != NULL) { + struct mbuf *npkt = pkt->m_nextpkt; + + if (!chain) { + pkt->m_nextpkt = NULL; + /* + * If we are not chaining, make sure to set the packet + * list count to 0 so that IP takes the right path; + * this is important for cases such as IPSec where a + * single mbuf might result in multiple mbufs as part + * of the encapsulation. If a non-zero count is passed + * down to IP, the head of the chain might change and + * we could end up skipping it (thus generating bogus + * packets). Fixing it in IP would be desirable, but + * for now this would do it. + */ + cnt = 0; + } + error = ip_output_list(pkt, cnt, opt, &ro, flags, 0, &ipoa); + if (chain || error) { + /* + * If we sent down a chain then we are done since + * the callee had taken care of everything; else + * we need to free the rest of the chain ourselves. + */ + if (!chain) + m_freem_list(npkt); + break; + } + pkt = npkt; + } + + if (unlocked) + socket_lock(so, 0); + + /* Synchronize cached PCB route */ + inp_route_copyin(inp, &ro); + + return (error); +} + void tcp_setpersist(tp) register struct tcpcb *tp;