X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/d41d1dae2cd00cc08c7982087d1c445180cad9f5..22ba694c5857e62b5a553b1505dcf2e509177f28:/bsd/netinet/tcp_subr.c diff --git a/bsd/netinet/tcp_subr.c b/bsd/netinet/tcp_subr.c index f1b220bc2..c9a7a6bb7 100644 --- a/bsd/netinet/tcp_subr.c +++ b/bsd/netinet/tcp_subr.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2010 Apple Inc. All rights reserved. + * Copyright (c) 2000-2013 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -82,9 +82,12 @@ #include #include #include +#include #include #include +#include + #include #include @@ -112,6 +115,9 @@ #include #include #include +#include +#include + #if INET6 #include #endif @@ -136,6 +142,9 @@ #include #include +#include + +#include #define DBG_FNC_TCP_CLOSE NETDBG_CODE(DBG_NETTCP, ((5 << 8) | 2)) @@ -145,18 +154,22 @@ extern int tcp_lq_overflow; #if IPSEC extern int ipsec_bypass; #endif +extern struct tcptimerlist tcp_timer_list; +extern struct tcptailq tcp_tw_tailq; int tcp_mssdflt = TCP_MSS; -SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); #if INET6 int tcp_v6mssdflt = TCP6_MSS; SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, - CTLFLAG_RW, &tcp_v6mssdflt , 0, + CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_v6mssdflt , 0, "Default TCP Maximum Segment Size for IPv6"); #endif +extern int tcp_do_autorcvbuf; + /* * Minimum MSS we accept and use. This prevents DoS attacks where * we are forced to a ridiculous low MSS like 20 and send hundreds @@ -166,75 +179,93 @@ SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, * checking. This setting prevents us from sending too small packets. */ int tcp_minmss = TCP_MINMSS; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); -/* - * Number of TCP segments per second we accept from remote host - * before we start to calculate average segment size. If average - * segment size drops below the minimum TCP MSS we assume a DoS - * attack and reset+drop the connection. Care has to be taken not to - * set this value too small to not kill interactive type connections - * (telnet, SSH) which send many small packets. - */ -#ifdef FIX_WORKAROUND_FOR_3894301 -__private_extern__ int tcp_minmssoverload = TCP_MINMSSOVERLOAD; -#else -__private_extern__ int tcp_minmssoverload = 0; -#endif -SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, - &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" - "be under the MINMSS Size"); - static int tcp_do_rfc1323 = 1; -SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); +// Not used static int tcp_do_rfc1644 = 0; -SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); static int do_tcpdrain = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW | CTLFLAG_LOCKED, &do_tcpdrain, 0, "Enable tcp_drain routine for extra help when low on mbufs"); -SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED, &tcbinfo.ipi_count, 0, "Number of active PCBs"); +SYSCTL_INT(_net_inet_tcp, OID_AUTO, tw_pcbcount, + CTLFLAG_RD | CTLFLAG_LOCKED, + &tcbinfo.ipi_twcount, 0, "Number of pcbs in time-wait state"); + static int icmp_may_rst = 1; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW | CTLFLAG_LOCKED, &icmp_may_rst, 0, "Certain ICMP unreachable messages may abort connections in SYN_SENT"); static int tcp_strict_rfc1948 = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly"); static int tcp_isn_reseed_interval = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); static int tcp_background_io_enabled = 1; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, background_io_enabled, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, background_io_enabled, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_background_io_enabled, 0, "Background IO Enabled"); -int tcp_TCPTV_MIN = 1; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, rtt_min, CTLFLAG_RW, +int tcp_TCPTV_MIN = 100; /* 100ms minimum RTT */ +SYSCTL_INT(_net_inet_tcp, OID_AUTO, rtt_min, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_TCPTV_MIN, 0, "min rtt value allowed"); +int tcp_rexmt_slop = TCPTV_REXMTSLOP; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmt_slop, CTLFLAG_RW, + &tcp_rexmt_slop, 0, "Slop added to retransmit timeout"); + __private_extern__ int tcp_use_randomport = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, randomize_ports, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, randomize_ports, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_use_randomport, 0, "Randomize TCP port numbers"); +extern struct tcp_cc_algo tcp_cc_newreno; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno_sockets, CTLFLAG_RD | CTLFLAG_LOCKED, + &tcp_cc_newreno.num_sockets, 0, "Number of sockets using newreno"); + +extern struct tcp_cc_algo tcp_cc_ledbat; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, background_sockets, CTLFLAG_RD | CTLFLAG_LOCKED, + &tcp_cc_ledbat.num_sockets, 0, "Number of sockets using background transport"); + +__private_extern__ int tcp_win_scale = 3; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, win_scale_factor, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_win_scale, 0, "Window scaling factor"); + static void tcp_cleartaocache(void); static void tcp_notify(struct inpcb *, int); +static void tcp_cc_init(void); + struct zone *sack_hole_zone; +struct zone *tcp_reass_zone; +struct zone *tcp_bwmeas_zone; +#if 0 +static unsigned int tcp_mptcp_dsnm_sz; +struct zone *tcp_mptcp_dsnm_zone; +#endif +/* The array containing pointers to currently implemented TCP CC algorithms */ +struct tcp_cc_algo* tcp_cc_algo_list[TCP_CC_ALGO_COUNT]; -extern unsigned int total_mb_cnt; -extern unsigned int total_cl_cnt; -extern int sbspace_factor; -extern int tcp_sockthreshold; extern int slowlink_wsize; /* window correction for slow links */ extern int path_mtu_discovery; +extern u_int32_t tcp_autorcvbuf_max; +extern u_int32_t tcp_autorcvbuf_inc_shift; +static void tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb); + +#define TCP_BWMEAS_BURST_MINSIZE 6 +#define TCP_BWMEAS_BURST_MAXSIZE 25 + +static uint32_t bwmeas_elm_size; /* * Target size of TCP PCB hash tables. Must be a power of two. @@ -247,7 +278,7 @@ extern int path_mtu_discovery; #endif __private_extern__ int tcp_tcbhashsize = TCBHASHSIZE; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD | CTLFLAG_LOCKED, &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); /* @@ -259,25 +290,23 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, * to be changed, eventually, for greater efficiency). */ #define ALIGNMENT 32 -#define ALIGNM1 (ALIGNMENT - 1) struct inp_tp { - union { - struct inpcb inp; - char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; - } inp_tp_u; - struct tcpcb tcb; + struct inpcb inp; + struct tcpcb tcb __attribute__((aligned(ALIGNMENT))); }; #undef ALIGNMENT -#undef ALIGNM1 - -extern struct inpcbhead time_wait_slots[]; -extern u_int32_t *delack_bitmask; int get_inpcb_str_size(void); int get_tcp_str_size(void); static void tcpcb_to_otcpcb(struct tcpcb *, struct otcpcb *); +static lck_attr_t *tcp_uptime_mtx_attr = NULL; /* mutex attributes */ +static lck_grp_t *tcp_uptime_mtx_grp = NULL; /* mutex group definition */ +static lck_grp_attr_t *tcp_uptime_mtx_grp_attr = NULL; /* mutex group attributes */ +int tcp_notsent_lowat_check(struct socket *so); + + int get_inpcb_str_size(void) { return sizeof(struct inpcb); @@ -291,84 +320,156 @@ int get_tcp_str_size(void) int tcp_freeq(struct tcpcb *tp); +/* + * Initialize TCP congestion control algorithms. + */ + +void +tcp_cc_init(void) +{ + bzero(&tcp_cc_algo_list, sizeof(tcp_cc_algo_list)); + tcp_cc_algo_list[TCP_CC_ALGO_NEWRENO_INDEX] = &tcp_cc_newreno; + tcp_cc_algo_list[TCP_CC_ALGO_BACKGROUND_INDEX] = &tcp_cc_ledbat; +} /* * Tcp initialization */ void -tcp_init() +tcp_init(struct protosw *pp, struct domain *dp) { +#pragma unused(dp) + static int tcp_initialized = 0; vm_size_t str_size; - int i; - struct inpcbinfo *pcbinfo; - + struct inpcbinfo *pcbinfo; + + VERIFY((pp->pr_flags & (PR_INITIALIZED|PR_ATTACHED)) == PR_ATTACHED); + + if (tcp_initialized) + return; + tcp_initialized = 1; + tcp_ccgen = 1; tcp_cleartaocache(); tcp_keepinit = TCPTV_KEEP_INIT; tcp_keepidle = TCPTV_KEEP_IDLE; tcp_keepintvl = TCPTV_KEEPINTVL; + tcp_keepcnt = TCPTV_KEEPCNT; tcp_maxpersistidle = TCPTV_KEEP_IDLE; tcp_msl = TCPTV_MSL; - read_random(&tcp_now, sizeof(tcp_now)); - tcp_now = tcp_now & 0x3fffffff; /* Starts tcp internal 100ms clock at a random value */ + microuptime(&tcp_uptime); + read_random(&tcp_now, sizeof(tcp_now)); + tcp_now = tcp_now & 0x3fffffff; /* Starts tcp internal clock at a random value */ LIST_INIT(&tcb); - tcbinfo.listhead = &tcb; + tcbinfo.ipi_listhead = &tcb; + pcbinfo = &tcbinfo; + /* + * allocate lock group attribute and group for tcp pcb mutexes + */ + pcbinfo->ipi_lock_grp_attr = lck_grp_attr_alloc_init(); + pcbinfo->ipi_lock_grp = lck_grp_alloc_init("tcppcb", pcbinfo->ipi_lock_grp_attr); + + /* + * allocate the lock attribute for tcp pcb mutexes + */ + pcbinfo->ipi_lock_attr = lck_attr_alloc_init(); + + if ((pcbinfo->ipi_lock = lck_rw_alloc_init(pcbinfo->ipi_lock_grp, + pcbinfo->ipi_lock_attr)) == NULL) { + panic("%s: unable to allocate PCB lock\n", __func__); + /* NOTREACHED */ + } + if (!powerof2(tcp_tcbhashsize)) { printf("WARNING: TCB hash size not a power of 2\n"); tcp_tcbhashsize = 512; /* safe default */ } - tcbinfo.hashsize = tcp_tcbhashsize; - tcbinfo.hashbase = hashinit(tcp_tcbhashsize, M_PCB, &tcbinfo.hashmask); - tcbinfo.porthashbase = hashinit(tcp_tcbhashsize, M_PCB, - &tcbinfo.porthashmask); - str_size = (vm_size_t) sizeof(struct inp_tp); - tcbinfo.ipi_zone = (void *) zinit(str_size, 120000*str_size, 8192, "tcpcb"); + tcbinfo.ipi_hashbase = hashinit(tcp_tcbhashsize, M_PCB, &tcbinfo.ipi_hashmask); + tcbinfo.ipi_porthashbase = hashinit(tcp_tcbhashsize, M_PCB, + &tcbinfo.ipi_porthashmask); + str_size = P2ROUNDUP(sizeof(struct inp_tp), sizeof(u_int64_t)); + tcbinfo.ipi_zone = zinit(str_size, 120000*str_size, 8192, "tcpcb"); + zone_change(tcbinfo.ipi_zone, Z_CALLERACCT, FALSE); + zone_change(tcbinfo.ipi_zone, Z_EXPAND, TRUE); + + tcbinfo.ipi_gc = tcp_gc; + in_pcbinfo_attach(&tcbinfo); + + str_size = P2ROUNDUP(sizeof(struct sackhole), sizeof(u_int64_t)); sack_hole_zone = zinit(str_size, 120000*str_size, 8192, "sack_hole zone"); + zone_change(sack_hole_zone, Z_CALLERACCT, FALSE); + zone_change(sack_hole_zone, Z_EXPAND, TRUE); + tcp_reass_maxseg = nmbclusters / 16; + str_size = P2ROUNDUP(sizeof(struct tseg_qent), sizeof(u_int64_t)); + tcp_reass_zone = zinit(str_size, (tcp_reass_maxseg + 1) * str_size, + 0, "tcp_reass_zone"); + if (tcp_reass_zone == NULL) { + panic("%s: failed allocating tcp_reass_zone", __func__); + /* NOTREACHED */ + } + zone_change(tcp_reass_zone, Z_CALLERACCT, FALSE); + zone_change(tcp_reass_zone, Z_EXPAND, TRUE); + + bwmeas_elm_size = P2ROUNDUP(sizeof(struct bwmeas), sizeof(u_int64_t)); + tcp_bwmeas_zone = zinit(bwmeas_elm_size, (100 * bwmeas_elm_size), 0, "tcp_bwmeas_zone"); + if (tcp_bwmeas_zone == NULL) { + panic("%s: failed allocating tcp_bwmeas_zone", __func__); + /* NOTREACHED */ + } + zone_change(tcp_bwmeas_zone, Z_CALLERACCT, FALSE); + zone_change(tcp_bwmeas_zone, Z_EXPAND, TRUE); #if INET6 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) #else /* INET6 */ #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) #endif /* INET6 */ - if (max_protohdr < TCP_MINPROTOHDR) - max_protohdr = TCP_MINPROTOHDR; - if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) + if (max_protohdr < TCP_MINPROTOHDR) { + _max_protohdr = TCP_MINPROTOHDR; + _max_protohdr = max_protohdr; /* round it up */ + } + if (max_linkhdr + max_protohdr > MCLBYTES) panic("tcp_init"); #undef TCP_MINPROTOHDR - /* - * allocate lock group attribute and group for tcp pcb mutexes - */ - pcbinfo->mtx_grp_attr = lck_grp_attr_alloc_init(); - pcbinfo->mtx_grp = lck_grp_alloc_init("tcppcb", pcbinfo->mtx_grp_attr); - + /* Initialize time wait and timer lists */ + TAILQ_INIT(&tcp_tw_tailq); + + bzero(&tcp_timer_list, sizeof(tcp_timer_list)); + LIST_INIT(&tcp_timer_list.lhead); /* - * allocate the lock attribute for tcp pcb mutexes + * allocate lock group attribute, group and attribute for the tcp timer list */ - pcbinfo->mtx_attr = lck_attr_alloc_init(); - - if ((pcbinfo->mtx = lck_rw_alloc_init(pcbinfo->mtx_grp, pcbinfo->mtx_attr)) == NULL) { - printf("tcp_init: mutex not alloced!\n"); - return; /* pretty much dead if this fails... */ + tcp_timer_list.mtx_grp_attr = lck_grp_attr_alloc_init(); + tcp_timer_list.mtx_grp = lck_grp_alloc_init("tcptimerlist", tcp_timer_list.mtx_grp_attr); + tcp_timer_list.mtx_attr = lck_attr_alloc_init(); + if ((tcp_timer_list.mtx = lck_mtx_alloc_init(tcp_timer_list.mtx_grp, tcp_timer_list.mtx_attr)) == NULL) { + panic("failed to allocate memory for tcp_timer_list.mtx\n"); + }; + tcp_timer_list.fast_quantum = TCP_FASTTIMER_QUANTUM; + tcp_timer_list.slow_quantum = TCP_SLOWTIMER_QUANTUM; + if ((tcp_timer_list.call = thread_call_allocate(tcp_run_timerlist, NULL)) == NULL) { + panic("failed to allocate call entry 1 in tcp_init\n"); } - delack_bitmask = _MALLOC((4 * tcp_tcbhashsize)/32, M_PCB, M_WAITOK); - if (delack_bitmask == 0) - panic("Delack Memory"); - - for (i=0; i < (tcbinfo.hashsize / 32); i++) - delack_bitmask[i] = 0; + /* + * allocate lock group attribute, group and attribute for tcp_uptime_lock + */ + tcp_uptime_mtx_grp_attr = lck_grp_attr_alloc_init(); + tcp_uptime_mtx_grp = lck_grp_alloc_init("tcpuptime", tcp_uptime_mtx_grp_attr); + tcp_uptime_mtx_attr = lck_attr_alloc_init(); + tcp_uptime_lock = lck_spin_alloc_init(tcp_uptime_mtx_grp, tcp_uptime_mtx_attr); - for (i=0; i < N_TIME_WAIT_SLOTS; i++) { - LIST_INIT(&time_wait_slots[i]); - } + /* Initialize TCP congestion control algorithms list */ + tcp_cc_init(); - timeout(tcp_fasttimo, NULL, hz/TCP_RETRANSHZ); + /* Initialize TCP LRO data structures */ + tcp_lro_init(); } /* @@ -391,14 +492,15 @@ tcp_fillheaders(tp, ip_ptr, tcp_ptr) ip6 = (struct ip6_hdr *)ip_ptr; ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | - (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); + (inp->inp_flow & IPV6_FLOWINFO_MASK); ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | (IPV6_VERSION & IPV6_VERSION_MASK); ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_plen = sizeof(struct tcphdr); ip6->ip6_src = inp->in6p_laddr; ip6->ip6_dst = inp->in6p_faddr; - tcp_hdr->th_sum = 0; + tcp_hdr->th_sum = in6_pseudo(&inp->in6p_laddr, &inp->in6p_faddr, + htonl(sizeof (struct tcphdr) + IPPROTO_TCP)); } else #endif { @@ -474,7 +576,8 @@ tcp_respond( tcp_seq ack, tcp_seq seq, int flags, - unsigned int ifscope + unsigned int ifscope, + unsigned int nocell ) { register int tlen; @@ -489,6 +592,7 @@ tcp_respond( struct ip6_hdr *ip6; int isipv6; #endif /* INET6 */ + struct ifnet *outif; #if INET6 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6; @@ -528,18 +632,27 @@ tcp_respond( m->m_data += max_linkhdr; #if INET6 if (isipv6) { + VERIFY((MHLEN - max_linkhdr) >= + (sizeof (*ip6) + sizeof (*nth))); bcopy((caddr_t)ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(m, struct ip6_hdr *); - nth = (struct tcphdr *)(ip6 + 1); + nth = (struct tcphdr *)(void *)(ip6 + 1); } else #endif /* INET6 */ { + VERIFY((MHLEN - max_linkhdr) >= + (sizeof (*ip) + sizeof (*nth))); bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); ip = mtod(m, struct ip *); - nth = (struct tcphdr *)(ip + 1); + nth = (struct tcphdr *)(void *)(ip + 1); } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); +#if MPTCP + if ((tp) && (tp->t_mpflags & TMPF_RESET)) + flags = (TH_RST | TH_ACK); + else +#endif flags = TH_ACK; } else { m_freem(m->m_next); @@ -550,13 +663,17 @@ tcp_respond( #define xchg(a,b,type) { type t; t=a; a=b; b=t; } #if INET6 if (isipv6) { + /* Expect 32-bit aligned IP on strict-align platforms */ + IP6_HDR_STRICT_ALIGNMENT_CHECK(ip6); xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); - nth = (struct tcphdr *)(ip6 + 1); + nth = (struct tcphdr *)(void *)(ip6 + 1); } else #endif /* INET6 */ { + /* Expect 32-bit aligned IP on strict-align platforms */ + IP_HDR_STRICT_ALIGNMENT_CHECK(ip); xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); - nth = (struct tcphdr *)(ip + 1); + nth = (struct tcphdr *)(void *)(ip + 1); } if (th != nth) { /* @@ -614,9 +731,10 @@ tcp_respond( #if INET6 if (isipv6) { nth->th_sum = 0; - nth->th_sum = in6_cksum(m, IPPROTO_TCP, - sizeof(struct ip6_hdr), - tlen - sizeof(struct ip6_hdr)); + nth->th_sum = in6_pseudo(&ip6->ip6_src, &ip6->ip6_dst, + htonl((tlen - sizeof (struct ip6_hdr)) + IPPROTO_TCP)); + m->m_pkthdr.csum_flags = CSUM_TCPIPV6; + m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, ro6 && ro6->ro_rt ? ro6->ro_rt->rt_ifp : @@ -639,21 +757,57 @@ tcp_respond( return; } #endif -#if PKT_PRIORITY - if (tp != NULL) - set_traffic_class(m, tp->t_inpcb->inp_socket, MBUF_TC_NONE); -#endif /* PKT_PRIORITY */ + + if (tp != NULL) { + u_int32_t svc_flags = 0; + if (isipv6) { + svc_flags |= PKT_SCF_IPV6; + } + set_packet_service_class(m, tp->t_inpcb->inp_socket, + MBUF_SC_UNSPEC, svc_flags); + + /* Embed flowhash and flow control flags */ + m->m_pkthdr.pkt_flowsrc = FLOWSRC_INPCB; + m->m_pkthdr.pkt_flowid = tp->t_inpcb->inp_flowhash; + m->m_pkthdr.pkt_flags |= PKTF_FLOW_ID | PKTF_FLOW_LOCALSRC; +#if MPTCP + /* Disable flow advisory when using MPTCP. */ + if (!(tp->t_mpflags & TMPF_MPTCP_TRUE)) +#endif /* MPTCP */ + m->m_pkthdr.pkt_flags |= PKTF_FLOW_ADV; + m->m_pkthdr.pkt_proto = IPPROTO_TCP; + } + #if INET6 if (isipv6) { - (void)ip6_output(m, NULL, ro6, 0, NULL, NULL, 0); - if (ro6 == &sro6 && ro6->ro_rt) { - rtfree(ro6->ro_rt); - ro6->ro_rt = NULL; - } + struct ip6_out_args ip6oa = { ifscope, { 0 }, + IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR, 0 }; + + if (ifscope != IFSCOPE_NONE) + ip6oa.ip6oa_flags |= IP6OAF_BOUND_IF; + if (nocell) + ip6oa.ip6oa_flags |= IP6OAF_NO_CELLULAR; + + (void) ip6_output(m, NULL, ro6, IPV6_OUTARGS, NULL, + NULL, &ip6oa); + + if (tp != NULL && ro6 != NULL && ro6->ro_rt != NULL && + (outif = ro6->ro_rt->rt_ifp) != + tp->t_inpcb->in6p_last_outifp) + tp->t_inpcb->in6p_last_outifp = outif; + + if (ro6 == &sro6) + ROUTE_RELEASE(ro6); } else #endif /* INET6 */ { - struct ip_out_args ipoa = { ifscope }; + struct ip_out_args ipoa = { ifscope, { 0 }, + IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR, 0 }; + + if (ifscope != IFSCOPE_NONE) + ipoa.ipoa_flags |= IPOAF_BOUND_IF; + if (nocell) + ipoa.ipoa_flags |= IPOAF_NO_CELLULAR; if (ro != &sro) { /* Copy the cached route and take an extra reference */ @@ -664,11 +818,16 @@ tcp_respond( */ (void) ip_output(m, NULL, &sro, IP_OUTARGS, NULL, &ipoa); + if (tp != NULL && sro.ro_rt != NULL && + (outif = sro.ro_rt->rt_ifp) != + tp->t_inpcb->inp_last_outifp) + tp->t_inpcb->inp_last_outifp = outif; + if (ro != &sro) { /* Synchronize cached PCB route */ inp_route_copyin(tp->t_inpcb, &sro); - } else if (sro.ro_rt != NULL) { - rtfree(sro.ro_rt); + } else { + ROUTE_RELEASE(&sro); } } } @@ -690,13 +849,15 @@ tcp_newtcpcb(inp) int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ - if (so->cached_in_sock_layer == 0) { - it = (struct inp_tp *)inp; + calculate_tcp_clock(); + + if (!so->cached_in_sock_layer) { + it = (struct inp_tp *)(void *)inp; tp = &it->tcb; + } else { + tp = (struct tcpcb *)(void *)inp->inp_saved_ppcb; } - else - tp = (struct tcpcb *) inp->inp_saved_ppcb; - + bzero((char *) tp, sizeof(struct tcpcb)); LIST_INIT(&tp->t_segq); tp->t_maxseg = tp->t_maxopd = @@ -707,7 +868,9 @@ tcp_newtcpcb(inp) if (tcp_do_rfc1323) tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); - tp->sack_enable = tcp_do_sack; + if (tcp_do_sack) + tp->t_flagsext |= TF_SACK_ENABLE; + TAILQ_INIT(&tp->snd_holes); tp->t_inpcb = inp; /* XXX */ /* @@ -719,12 +882,29 @@ tcp_newtcpcb(inp) tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; tp->t_rttmin = tcp_TCPTV_MIN; tp->t_rxtcur = TCPTV_RTOBASE; + + /* Initialize congestion control algorithm for this connection + * to newreno by default + */ + tp->tcp_cc_index = TCP_CC_ALGO_NEWRENO_INDEX; + if (CC_ALGO(tp)->init != NULL) { + CC_ALGO(tp)->init(tp); + } + tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; - tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->snd_ssthresh_prev = TCP_MAXWIN << TCP_MAX_WINSHIFT; - tp->t_rcvtime = 0; - tp->t_bw_rtttime = 0; + tp->t_rcvtime = tcp_now; + tp->tentry.timer_start = tcp_now; + tp->t_persist_timeout = tcp_max_persist_timeout; + tp->t_persist_stop = 0; + tp->t_flagsext |= TF_RCVUNACK_WAITSS; + tp->t_rexmtthresh = tcprexmtthresh; + + /* Clear time wait tailq entry */ + tp->t_twentry.tqe_next = NULL; + tp->t_twentry.tqe_prev = NULL; + /* * IPv4 TTL initialization is necessary for an IPv6 socket as well, * because the socket may be bound to an IPv6 wildcard address, @@ -746,8 +926,13 @@ tcp_drop(tp, errno) int errno; { struct socket *so = tp->t_inpcb->inp_socket; - +#if CONFIG_DTRACE + struct inpcb *inp = tp->t_inpcb; +#endif + if (TCPS_HAVERCVDSYN(tp->t_state)) { + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_CLOSED); tp->t_state = TCPS_CLOSED; (void) tcp_output(tp); tcpstat.tcps_drops++; @@ -759,6 +944,39 @@ tcp_drop(tp, errno) return (tcp_close(tp)); } +void +tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt) +{ + u_int32_t rtt = rt->rt_rmx.rmx_rtt; + int isnetlocal = (tp->t_flags & TF_LOCAL); + + if (rtt != 0) { + /* + * XXX the lock bit for RTT indicates that the value + * is also a minimum value; this is subject to time. + */ + if (rt->rt_rmx.rmx_locks & RTV_RTT) + tp->t_rttmin = rtt / (RTM_RTTUNIT / TCP_RETRANSHZ); + else + tp->t_rttmin = isnetlocal ? tcp_TCPTV_MIN : TCPTV_REXMTMIN; + tp->t_srtt = rtt / (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE)); + tcpstat.tcps_usedrtt++; + if (rt->rt_rmx.rmx_rttvar) { + tp->t_rttvar = rt->rt_rmx.rmx_rttvar / + (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE)); + tcpstat.tcps_usedrttvar++; + } else { + /* default variation is +- 1 rtt */ + tp->t_rttvar = + tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; + } + TCPT_RANGESET(tp->t_rxtcur, + ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, + tp->t_rttmin, TCPTV_REXMTMAX, + TCP_ADD_REXMTSLOP(tp)); + } +} + /* * Close a TCP control block: * discard all space held by the tcp @@ -774,30 +992,16 @@ tcp_close(tp) #if INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ + struct route *ro; struct rtentry *rt; int dosavessthresh; - if ( inp->inp_ppcb == NULL) /* tcp_close was called previously, bail */ - return NULL; - - /* Clear the timers before we delete the PCB. */ - { - int i; - for (i = 0; i < TCPT_NTIMERS; i++) { - tp->t_timer[i] = 0; - } - } + /* tcp_close was called previously, bail */ + if ( inp->inp_ppcb == NULL) + return(NULL); + tcp_canceltimers(tp); KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_START, tp,0,0,0,0); - switch (tp->t_state) - { - case TCPS_ESTABLISHED: - case TCPS_FIN_WAIT_1: - case TCPS_CLOSING: - case TCPS_CLOSE_WAIT: - case TCPS_LAST_ACK: - break; - } /* * If another thread for this tcp is currently in ip (indicated by @@ -811,16 +1015,25 @@ tcp_close(tp) * point both flags should be cleared and we can proceed further * with the cleanup. */ - if (tp->t_flags & (TF_CLOSING|TF_SENDINPROG)) { + if ((tp->t_flags & TF_CLOSING) || + inp->inp_sndinprog_cnt > 0) { tp->t_flags |= TF_CLOSING; return (NULL); } + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_CLOSED); + + if (CC_ALGO(tp)->cleanup != NULL) { + CC_ALGO(tp)->cleanup(tp); + } + #if INET6 - rt = isipv6 ? inp->in6p_route.ro_rt : inp->inp_route.ro_rt; + ro = (isipv6 ? (struct route *)&inp->in6p_route : &inp->inp_route); #else - rt = inp->inp_route.ro_rt; + ro = &inp->inp_route; #endif + rt = ro->ro_rt; if (rt != NULL) RT_LOCK_SPIN(rt); @@ -844,17 +1057,21 @@ tcp_close(tp) if (rt == NULL) goto no_valid_rt; - sin6 = (struct sockaddr_in6 *)rt_key(rt); + sin6 = (struct sockaddr_in6 *)(void *)rt_key(rt); if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) goto no_valid_rt; } else #endif /* INET6 */ - if (rt == NULL || !(rt->rt_flags & RTF_UP) || - ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr == - INADDR_ANY || rt->generation_id != route_generation) { - if (tp->t_state >= TCPS_CLOSE_WAIT) + if (ROUTE_UNUSABLE(ro) || + SIN(rt_key(rt))->sin_addr.s_addr == INADDR_ANY) { + if (tp->t_state >= TCPS_CLOSE_WAIT) { + DTRACE_TCP4(state__change, + void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, + TCPS_CLOSING); tp->t_state = TCPS_CLOSING; + } goto no_valid_rt; } @@ -951,27 +1168,50 @@ no_valid_rt: (void) tcp_freeq(tp); tcp_free_sackholes(tp); + if (tp->t_bwmeas != NULL) { + tcp_bwmeas_free(tp); + } /* Free the packet list */ if (tp->t_pktlist_head != NULL) m_freem_list(tp->t_pktlist_head); TCP_PKTLIST_CLEAR(tp); -#ifdef __APPLE__ +#if MPTCP + /* Clear MPTCP state */ + tp->t_mpflags = 0; +#endif /* MPTCP */ + if (so->cached_in_sock_layer) inp->inp_saved_ppcb = (caddr_t) tp; -#endif - soisdisconnected(so); + /* Issue a wakeup before detach so that we don't miss + * a wakeup + */ + sodisconnectwakeup(so); + + /* + * Clean up any LRO state + */ + if (tp->t_flagsext & TF_LRO_OFFLOADED) { + tcp_lro_remove_state(inp->inp_laddr, inp->inp_faddr, + inp->inp_lport, + inp->inp_fport); + tp->t_flagsext &= ~TF_LRO_OFFLOADED; + } + tp->t_state = TCPS_CLOSED; #if INET6 - if (INP_CHECK_SOCKAF(so, AF_INET6)) + if (SOCK_CHECK_DOM(so, PF_INET6)) in6_pcbdetach(inp); else #endif /* INET6 */ in_pcbdetach(inp); + + /* Call soisdisconnected after detach because it might unlock the socket */ + soisdisconnected(so); tcpstat.tcps_closed++; KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_END, tcpstat.tcps_closed,0,0,0,0); - return ((struct tcpcb *)0); + return(NULL); } int @@ -985,7 +1225,7 @@ tcp_freeq(tp) while((q = LIST_FIRST(&tp->t_segq)) != NULL) { LIST_REMOVE(q, tqe_q); m_freem(q->tqe_m); - FREE(q, M_TSEGQ); + zfree(tcp_reass_zone, q); tcp_reass_qsize--; rv = 1; } @@ -997,34 +1237,32 @@ tcp_drain() { if (do_tcpdrain) { - struct inpcb *inpb; - struct tcpcb *tcpb; - struct tseg_qent *te; - + struct inpcb *inp; + struct tcpcb *tp; /* * Walk the tcpbs, if existing, and flush the reassembly queue, * if there is one... - * XXX: The "Net/3" implementation doesn't imply that the TCP - * reassembly queue should be flushed, but in a situation - * where we're really low on mbufs, this is potentially - * usefull. + * Do it next time if the pcbinfo lock is in use */ - if (!lck_rw_try_lock_exclusive(tcbinfo.mtx)) /* do it next time if the lock is in use */ + if (!lck_rw_try_lock_exclusive(tcbinfo.ipi_lock)) return; - for (inpb = LIST_FIRST(tcbinfo.listhead); inpb; - inpb = LIST_NEXT(inpb, inp_list)) { - if ((tcpb = intotcpcb(inpb))) { - while ((te = LIST_FIRST(&tcpb->t_segq)) - != NULL) { - LIST_REMOVE(te, tqe_q); - m_freem(te->tqe_m); - FREE(te, M_TSEGQ); - tcp_reass_qsize--; - } + LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) { + if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) != + WNT_STOPUSING) { + tcp_lock(inp->inp_socket, 1, 0); + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) + == WNT_STOPUSING) { + /* lost a race, try the next one */ + tcp_unlock(inp->inp_socket, 1, 0); + continue; + } + tp = intotcpcb(inp); + tcp_freeq(tp); + tcp_unlock(inp->inp_socket, 1, 0); } } - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); } } @@ -1072,6 +1310,30 @@ tcp_notify(inp, error) #endif } +struct bwmeas* +tcp_bwmeas_alloc(struct tcpcb *tp) +{ + struct bwmeas *elm; + elm = zalloc(tcp_bwmeas_zone); + if (elm == NULL) + return(elm); + + bzero(elm, bwmeas_elm_size); + elm->bw_minsizepkts = TCP_BWMEAS_BURST_MINSIZE; + elm->bw_maxsizepkts = TCP_BWMEAS_BURST_MAXSIZE; + elm->bw_minsize = elm->bw_minsizepkts * tp->t_maxseg; + elm->bw_maxsize = elm->bw_maxsizepkts * tp->t_maxseg; + return(elm); +} + +void +tcp_bwmeas_free(struct tcpcb* tp) +{ + zfree(tcp_bwmeas_zone, tp->t_bwmeas); + tp->t_bwmeas = NULL; + tp->t_flagsext &= ~(TF_MEASURESNDBW); +} + /* * tcpcb_to_otcpcb copies specific bits of a tcpcb to a otcpcb format. * The otcpcb data structure is passed to user space and must not change. @@ -1083,7 +1345,7 @@ tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp) otp->t_segq = (u_int32_t)(uintptr_t)tp->t_segq.lh_first; otp->t_dupacks = tp->t_dupacks; - for (i = 0; i < TCPT_NTIMERS; i++) + for (i = 0; i < TCPT_NTIMERS_EXT; i++) otp->t_timer[i] = tp->t_timer[i]; otp->t_inpcb = (_TCPCB_PTR(struct inpcb *))(uintptr_t)tp->t_inpcb; otp->t_state = tp->t_state; @@ -1132,34 +1394,34 @@ tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp) otp->snd_recover = tp->snd_recover; otp->snd_cwnd_prev = tp->snd_cwnd_prev; otp->snd_ssthresh_prev = tp->snd_ssthresh_prev; - otp->t_badrxtwin = tp->t_badrxtwin; + otp->t_badrxtwin = 0; } static int tcp_pcblist SYSCTL_HANDLER_ARGS { #pragma unused(oidp, arg1, arg2) - int error, i, n; + int error, i = 0, n; struct inpcb *inp, **inp_list; + struct tcpcb *tp; inp_gen_t gencnt; struct xinpgen xig; - int slot; /* * The process of preparing the TCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ - lck_rw_lock_shared(tcbinfo.mtx); + lck_rw_lock_shared(tcbinfo.ipi_lock); if (req->oldptr == USER_ADDR_NULL) { n = tcbinfo.ipi_count; req->oldidx = 2 * (sizeof xig) + (n + n/8) * sizeof(struct xtcpcb); - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return 0; } if (req->newptr != USER_ADDR_NULL) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return EPERM; } @@ -1176,41 +1438,36 @@ tcp_pcblist SYSCTL_HANDLER_ARGS xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return error; } /* * We are done if there is no pcb */ if (n == 0) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return 0; } inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK); if (inp_list == 0) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return ENOMEM; } - for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n; - inp = LIST_NEXT(inp, inp_list)) { -#ifdef __APPLE__ - if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) -#else - if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp)) -#endif + LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) { + if (inp->inp_gencnt <= gencnt && + inp->inp_state != INPCB_STATE_DEAD) inp_list[i++] = inp; + if (i >= n) break; } - for (slot = 0; slot < N_TIME_WAIT_SLOTS; slot++) { - struct inpcb *inpnxt; - - for (inp = time_wait_slots[slot].lh_first; inp && i < n; inp = inpnxt) { - inpnxt = inp->inp_list.le_next; - if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) - inp_list[i++] = inp; - } + TAILQ_FOREACH(tp, &tcp_tw_tailq, t_twentry) { + inp = tp->t_inpcb; + if (inp->inp_gencnt <= gencnt && + inp->inp_state != INPCB_STATE_DEAD) + inp_list[i++] = inp; + if (i >= n) break; } n = i; @@ -1218,7 +1475,8 @@ tcp_pcblist SYSCTL_HANDLER_ARGS error = 0; for (i = 0; i < n; i++) { inp = inp_list[i]; - if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) { + if (inp->inp_gencnt <= gencnt && + inp->inp_state != INPCB_STATE_DEAD) { struct xtcpcb xt; caddr_t inp_ppcb; @@ -1228,7 +1486,8 @@ tcp_pcblist SYSCTL_HANDLER_ARGS inpcb_to_compat(inp, &xt.xt_inp); inp_ppcb = inp->inp_ppcb; if (inp_ppcb != NULL) { - tcpcb_to_otcpcb((struct tcpcb *)inp_ppcb, + tcpcb_to_otcpcb( + (struct tcpcb *)(void *)inp_ppcb, &xt.xt_tp); } else { bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); @@ -1254,14 +1513,13 @@ tcp_pcblist SYSCTL_HANDLER_ARGS error = SYSCTL_OUT(req, &xig, sizeof xig); } FREE(inp_list, M_TEMP); - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return error; } -SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, +SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); -#if !CONFIG_EMBEDDED static void tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp) @@ -1270,7 +1528,7 @@ tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp) otp->t_segq = (u_int32_t)(uintptr_t)tp->t_segq.lh_first; otp->t_dupacks = tp->t_dupacks; - for (i = 0; i < TCPT_NTIMERS; i++) + for (i = 0; i < TCPT_NTIMERS_EXT; i++) otp->t_timer[i] = tp->t_timer[i]; otp->t_state = tp->t_state; otp->t_flags = tp->t_flags; @@ -1318,7 +1576,7 @@ tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp) otp->snd_recover = tp->snd_recover; otp->snd_cwnd_prev = tp->snd_cwnd_prev; otp->snd_ssthresh_prev = tp->snd_ssthresh_prev; - otp->t_badrxtwin = tp->t_badrxtwin; + otp->t_badrxtwin = 0; } @@ -1326,27 +1584,27 @@ static int tcp_pcblist64 SYSCTL_HANDLER_ARGS { #pragma unused(oidp, arg1, arg2) - int error, i, n; + int error, i = 0, n; struct inpcb *inp, **inp_list; + struct tcpcb *tp; inp_gen_t gencnt; struct xinpgen xig; - int slot; /* * The process of preparing the TCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ - lck_rw_lock_shared(tcbinfo.mtx); + lck_rw_lock_shared(tcbinfo.ipi_lock); if (req->oldptr == USER_ADDR_NULL) { n = tcbinfo.ipi_count; req->oldidx = 2 * (sizeof xig) + (n + n/8) * sizeof(struct xtcpcb64); - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return 0; } if (req->newptr != USER_ADDR_NULL) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return EPERM; } @@ -1363,41 +1621,36 @@ tcp_pcblist64 SYSCTL_HANDLER_ARGS xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return error; } /* * We are done if there is no pcb */ if (n == 0) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return 0; } inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK); if (inp_list == 0) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return ENOMEM; } - for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n; - inp = LIST_NEXT(inp, inp_list)) { -#ifdef __APPLE__ - if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) -#else - if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp)) -#endif + LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) { + if (inp->inp_gencnt <= gencnt && + inp->inp_state != INPCB_STATE_DEAD) inp_list[i++] = inp; + if (i >= n) break; } - for (slot = 0; slot < N_TIME_WAIT_SLOTS; slot++) { - struct inpcb *inpnxt; - - for (inp = time_wait_slots[slot].lh_first; inp && i < n; inp = inpnxt) { - inpnxt = inp->inp_list.le_next; - if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) - inp_list[i++] = inp; - } + TAILQ_FOREACH(tp, &tcp_tw_tailq, t_twentry) { + inp = tp->t_inpcb; + if (inp->inp_gencnt <= gencnt && + inp->inp_state != INPCB_STATE_DEAD) + inp_list[i++] = inp; + if (i >= n) break; } n = i; @@ -1406,43 +1659,77 @@ tcp_pcblist64 SYSCTL_HANDLER_ARGS for (i = 0; i < n; i++) { inp = inp_list[i]; if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) { - struct xtcpcb64 xt; - - bzero(&xt, sizeof(xt)); - xt.xt_len = sizeof xt; - inpcb_to_xinpcb64(inp, &xt.xt_inpcb); - xt.xt_inpcb.inp_ppcb = (u_int64_t)(uintptr_t)inp->inp_ppcb; - if (inp->inp_ppcb != NULL) - tcpcb_to_xtcpcb64((struct tcpcb *)inp->inp_ppcb, &xt); - if (inp->inp_socket) - sotoxsocket64(inp->inp_socket, &xt.xt_inpcb.xi_socket); - error = SYSCTL_OUT(req, &xt, sizeof xt); + struct xtcpcb64 xt; + + bzero(&xt, sizeof(xt)); + xt.xt_len = sizeof xt; + inpcb_to_xinpcb64(inp, &xt.xt_inpcb); + xt.xt_inpcb.inp_ppcb = (u_int64_t)(uintptr_t)inp->inp_ppcb; + if (inp->inp_ppcb != NULL) + tcpcb_to_xtcpcb64((struct tcpcb *)inp->inp_ppcb, &xt); + if (inp->inp_socket) + sotoxsocket64(inp->inp_socket, &xt.xt_inpcb.xi_socket); + error = SYSCTL_OUT(req, &xt, sizeof xt); } } if (!error) { - /* - * Give the user an updated idea of our state. - * If the generation differs from what we told - * her before, she knows that something happened - * while we were processing this request, and it - * might be necessary to retry. - */ - bzero(&xig, sizeof(xig)); - xig.xig_len = sizeof xig; - xig.xig_gen = tcbinfo.ipi_gencnt; - xig.xig_sogen = so_gencnt; - xig.xig_count = tcbinfo.ipi_count; - error = SYSCTL_OUT(req, &xig, sizeof xig); + /* + * Give the user an updated idea of our state. + * If the generation differs from what we told + * her before, she knows that something happened + * while we were processing this request, and it + * might be necessary to retry. + */ + bzero(&xig, sizeof(xig)); + xig.xig_len = sizeof xig; + xig.xig_gen = tcbinfo.ipi_gencnt; + xig.xig_sogen = so_gencnt; + xig.xig_count = tcbinfo.ipi_count; + error = SYSCTL_OUT(req, &xig, sizeof xig); } FREE(inp_list, M_TEMP); - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return error; } -SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist64, CTLFLAG_RD, 0, 0, +SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist64, CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, tcp_pcblist64, "S,xtcpcb64", "List of active TCP connections"); -#endif /* !CONFIG_EMBEDDED */ + +static int +tcp_pcblist_n SYSCTL_HANDLER_ARGS +{ +#pragma unused(oidp, arg1, arg2) + int error = 0; + + error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo); + + return error; +} + + +SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist_n, CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, + tcp_pcblist_n, "S,xtcpcb_n", "List of active TCP connections"); + + +__private_extern__ void +tcp_get_ports_used(uint32_t ifindex, int protocol, uint32_t wildcardok, + bitstr_t *bitfield) +{ + inpcb_get_ports_used(ifindex, protocol, wildcardok, bitfield, &tcbinfo); +} + +__private_extern__ uint32_t +tcp_count_opportunistic(unsigned int ifindex, u_int32_t flags) +{ + return inpcb_count_opportunistic(ifindex, &tcbinfo, flags); +} + +__private_extern__ uint32_t +tcp_find_anypcb_byaddr(struct ifaddr *ifa) +{ + return inpcb_find_anypcb_byaddr(ifa, &tcbinfo); +} void tcp_ctlinput(cmd, sa, vip) @@ -1452,16 +1739,13 @@ tcp_ctlinput(cmd, sa, vip) { tcp_seq icmp_tcp_seq; struct ip *ip = vip; - struct tcphdr *th; struct in_addr faddr; struct inpcb *inp; struct tcpcb *tp; - - void (*notify)(struct inpcb *, int) = tcp_notify; - struct icmp *icp; + void (*notify)(struct inpcb *, int) = tcp_notify; - faddr = ((struct sockaddr_in *)sa)->sin_addr; + faddr = ((struct sockaddr_in *)(void *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) return; @@ -1481,19 +1765,22 @@ tcp_ctlinput(cmd, sa, vip) else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) return; if (ip) { - icp = (struct icmp *)((caddr_t)ip - - offsetof(struct icmp, icmp_ip)); - th = (struct tcphdr *)((caddr_t)ip - + (IP_VHL_HL(ip->ip_vhl) << 2)); - inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, - ip->ip_src, th->th_sport, 0, NULL); + struct tcphdr th; + struct icmp *icp; + + icp = (struct icmp *)(void *) + ((caddr_t)ip - offsetof(struct icmp, icmp_ip)); + bcopy(((caddr_t)ip + (IP_VHL_HL(ip->ip_vhl) << 2)), + &th, sizeof (th)); + inp = in_pcblookup_hash(&tcbinfo, faddr, th.th_dport, + ip->ip_src, th.th_sport, 0, NULL); if (inp != NULL && inp->inp_socket != NULL) { tcp_lock(inp->inp_socket, 1, 0); if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { tcp_unlock(inp->inp_socket, 1, 0); return; } - icmp_tcp_seq = htonl(th->th_seq); + icmp_tcp_seq = htonl(th.th_seq); tp = intotcpcb(inp); if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && SEQ_LT(icmp_tcp_seq, tp->snd_max)) { @@ -1618,10 +1905,10 @@ tcp6_ctlinput(cmd, sa, d) in6_pcbnotify(&tcbinfo, sa, th.th_dport, (struct sockaddr *)ip6cp->ip6c_src, - th.th_sport, cmd, notify); + th.th_sport, cmd, NULL, notify); } else { in6_pcbnotify(&tcbinfo, sa, 0, - (struct sockaddr *)(size_t)sa6_src, 0, cmd, notify); + (struct sockaddr *)(size_t)sa6_src, 0, cmd, NULL, notify); } } #endif /* INET6 */ @@ -1687,7 +1974,7 @@ tcp_new_isn(tp) if (((tp->t_state == TCPS_LISTEN) || (tp->t_state == TCPS_TIME_WAIT)) && tcp_strict_rfc1948 == 0) #ifdef __APPLE__ - return random(); + return RandomULong(); #else return arc4random(); #endif @@ -1773,7 +2060,7 @@ tcp_mtudisc( if (tp) { #if INET6 if (isipv6) - rt = tcp_rtlookup6(inp); + rt = tcp_rtlookup6(inp, IFSCOPE_NONE); else #endif /* INET6 */ rt = tcp_rtlookup(inp, IFSCOPE_NONE); @@ -1832,11 +2119,19 @@ tcp_mtudisc( (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) mss -= TCPOLEN_TSTAMP_APPA; +#if MPTCP + mss -= mptcp_adj_mss(tp, TRUE); +#endif if (so->so_snd.sb_hiwat < mss) mss = so->so_snd.sb_hiwat; tp->t_maxseg = mss; + /* + * Reset the slow-start flight size as it may depends on the new MSS + */ + if (CC_ALGO(tp)->cwnd_init != NULL) + CC_ALGO(tp)->cwnd_init(tp); tcpstat.tcps_mturesent++; tp->t_rtttime = 0; tp->snd_nxt = tp->snd_una; @@ -1866,15 +2161,19 @@ tcp_rtlookup(inp, input_ifscope) if ((rt = ro->ro_rt) != NULL) RT_LOCK(rt); - if (rt == NULL || !(rt->rt_flags & RTF_UP) || - rt->generation_id != route_generation) { + if (ROUTE_UNUSABLE(ro)) { + if (rt != NULL) { + RT_UNLOCK(rt); + rt = NULL; + } + ROUTE_RELEASE(ro); /* No route yet, so try to acquire one */ if (inp->inp_faddr.s_addr != INADDR_ANY) { unsigned int ifscope; ro->ro_dst.sa_family = AF_INET; ro->ro_dst.sa_len = sizeof(struct sockaddr_in); - ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = + ((struct sockaddr_in *)(void *)&ro->ro_dst)->sin_addr = inp->inp_faddr; /* @@ -1885,15 +2184,15 @@ tcp_rtlookup(inp, input_ifscope) * input_ifscope is IFSCOPE_NONE). */ ifscope = (inp->inp_flags & INP_BOUND_IF) ? - inp->inp_boundif : input_ifscope; + inp->inp_boundifp->if_index : input_ifscope; - if (rt != NULL) - RT_UNLOCK(rt); - rtalloc_scoped_ign(ro, 0, ifscope); + rtalloc_scoped(ro, ifscope); if ((rt = ro->ro_rt) != NULL) RT_LOCK(rt); } } + if (rt != NULL) + RT_LOCK_ASSERT_HELD(rt); /* * Update MTU discovery determination. Don't do it if: @@ -1926,6 +2225,14 @@ tcp_rtlookup(inp, input_ifscope) tcp_set_tso(tp, rt->rt_ifp); } + /* Note if the peer is local */ + if (rt != NULL && + (rt->rt_gateway->sa_family == AF_LINK || + rt->rt_ifp->if_flags & IFF_LOOPBACK || + in_localaddr(inp->inp_faddr))) { + tp->t_flags |= TF_LOCAL; + } + /* * Caller needs to call RT_UNLOCK(rt). */ @@ -1934,8 +2241,9 @@ tcp_rtlookup(inp, input_ifscope) #if INET6 struct rtentry * -tcp_rtlookup6(inp) +tcp_rtlookup6(inp, input_ifscope) struct inpcb *inp; + unsigned int input_ifscope; { struct route_in6 *ro6; struct rtentry *rt; @@ -1947,23 +2255,40 @@ tcp_rtlookup6(inp) if ((rt = ro6->ro_rt) != NULL) RT_LOCK(rt); - if (rt == NULL || !(rt->rt_flags & RTF_UP) || - rt->generation_id != route_generation) { + if (ROUTE_UNUSABLE(ro6)) { + if (rt != NULL) { + RT_UNLOCK(rt); + rt = NULL; + } + ROUTE_RELEASE(ro6); /* No route yet, so try to acquire one */ if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) { struct sockaddr_in6 *dst6; + unsigned int ifscope; dst6 = (struct sockaddr_in6 *)&ro6->ro_dst; dst6->sin6_family = AF_INET6; dst6->sin6_len = sizeof(*dst6); dst6->sin6_addr = inp->in6p_faddr; - if (rt != NULL) - RT_UNLOCK(rt); - rtalloc_ign((struct route *)ro6, 0); + + /* + * If the socket was bound to an interface, then + * the bound-to-interface takes precedence over + * the inbound interface passed in by the caller + * (if we get here as part of the output path then + * input_ifscope is IFSCOPE_NONE). + */ + ifscope = (inp->inp_flags & INP_BOUND_IF) ? + inp->inp_boundifp->if_index : input_ifscope; + + rtalloc_scoped((struct route *)ro6, ifscope); if ((rt = ro6->ro_rt) != NULL) RT_LOCK(rt); } } + if (rt != NULL) + RT_LOCK_ASSERT_HELD(rt); + /* * Update path MTU Discovery determination * while looking up the route: @@ -2004,6 +2329,15 @@ tcp_rtlookup6(inp) tcp_set_tso(tp, rt->rt_ifp); } + /* Note if the peer is local */ + if (rt != NULL && + (IN6_IS_ADDR_LOOPBACK(&inp->in6p_faddr) || + IN6_IS_ADDR_LINKLOCAL(&inp->in6p_faddr) || + rt->rt_gateway->sa_family == AF_LINK || + in6_localaddr(&inp->in6p_faddr))) { + tp->t_flags |= TF_LOCAL; + } + /* * Caller needs to call RT_UNLOCK(rt). */ @@ -2035,7 +2369,7 @@ ipsec_hdrsiz_tcp(tp) #if INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { ip6 = mtod(m, struct ip6_hdr *); - th = (struct tcphdr *)(ip6 + 1); + th = (struct tcphdr *)(void *)(ip6 + 1); m->m_pkthdr.len = m->m_len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); tcp_fillheaders(tp, ip6, th); @@ -2068,7 +2402,7 @@ tcp_gettaocache(inp) #if INET6 if ((inp->inp_vflag & INP_IPV6) != 0) - rt = tcp_rtlookup6(inp); + rt = tcp_rtlookup6(inp, IFSCOPE_NONE); else #endif /* INET6 */ rt = tcp_rtlookup(inp, IFSCOPE_NONE); @@ -2112,7 +2446,7 @@ tcp_lock(struct socket *so, int refcount, void *lr) lr_saved = lr; if (so->so_pcb != NULL) { - lck_mtx_lock(((struct inpcb *)so->so_pcb)->inpcb_mtx); + lck_mtx_lock(&((struct inpcb *)so->so_pcb)->inpcb_mtx); } else { panic("tcp_lock: so=%p NO PCB! lr=%p lrh= %s\n", so, lr_saved, solockhistory_nr(so)); @@ -2142,9 +2476,11 @@ tcp_unlock(struct socket *so, int refcount, void *lr) lr_saved = lr; #ifdef MORE_TCPLOCK_DEBUG - printf("tcp_unlock: so=%p sopcb=%p lock=%p ref=%x lr=%p\n", - so, so->so_pcb, ((struct inpcb *)so->so_pcb)->inpcb_mtx, - so->so_usecount, lr_saved); + printf("tcp_unlock: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x " + "lr=0x%llx\n", (uint64_t)VM_KERNEL_ADDRPERM(so), + (uint64_t)VM_KERNEL_ADDRPERM(so->so_pcb), + (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)), + so->so_usecount, (uint64_t)VM_KERNEL_ADDRPERM(lr_saved)); #endif if (refcount) so->so_usecount--; @@ -2159,11 +2495,11 @@ tcp_unlock(struct socket *so, int refcount, void *lr) so, so->so_usecount, lr_saved, solockhistory_nr(so)); /* NOTREACHED */ } else { - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, + lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); so->unlock_lr[so->next_unlock_lr] = lr_saved; so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; - lck_mtx_unlock(((struct inpcb *)so->so_pcb)->inpcb_mtx); + lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx); } return (0); } @@ -2179,7 +2515,7 @@ tcp_getlock( if (so->so_usecount < 0) panic("tcp_getlock: so=%p usecount=%x lrh= %s\n", so, so->so_usecount, solockhistory_nr(so)); - return(inp->inpcb_mtx); + return(&inp->inpcb_mtx); } else { panic("tcp_getlock: so=%p NULL so_pcb %s\n", @@ -2188,28 +2524,60 @@ tcp_getlock( } } +/* Determine if we can grow the recieve socket buffer to avoid sending + * a zero window update to the peer. We allow even socket buffers that + * have fixed size (set by the application) to grow if the resource + * constraints are met. They will also be trimmed after the application + * reads data. + */ +static void +tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb) { + u_int32_t rcvbufinc = tp->t_maxseg << tcp_autorcvbuf_inc_shift; + u_int32_t rcvbuf = sb->sb_hiwat; + struct socket *so = tp->t_inpcb->inp_socket; + + /* + * If message delivery is enabled, do not count + * unordered bytes in receive buffer towards hiwat + */ + if (so->so_flags & SOF_ENABLE_MSGS) + rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes; + + if (tcp_do_autorcvbuf == 1 && + tcp_cansbgrow(sb) && + (tp->t_flags & TF_SLOWLINK) == 0 && + (rcvbuf - sb->sb_cc) < rcvbufinc && + (rcvbuf < tcp_autorcvbuf_max)) { + sbreserve(sb, (sb->sb_hiwat + rcvbufinc)); + } +} + int32_t tcp_sbspace(struct tcpcb *tp) { struct sockbuf *sb = &tp->t_inpcb->inp_socket->so_rcv; - int32_t space, newspace; + u_int32_t rcvbuf = sb->sb_hiwat; + int32_t space; + struct socket *so = tp->t_inpcb->inp_socket; + + /* + * If message delivery is enabled, do not count + * unordered bytes in receive buffer towards hiwat mark. + * This value is used to return correct rwnd that does + * not reflect the extra unordered bytes added to the + * receive socket buffer. + */ + if (so->so_flags & SOF_ENABLE_MSGS) + rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes; + + tcp_sbrcv_grow_rwin(tp, sb); - space = ((int32_t) imin((sb->sb_hiwat - sb->sb_cc), + space = ((int32_t) imin((rcvbuf - sb->sb_cc), (sb->sb_mbmax - sb->sb_mbcnt))); if (space < 0) space = 0; -#if TRAFFIC_MGT - if (tp->t_inpcb->inp_socket->so_traffic_mgt_flags & TRAFFIC_MGT_SO_BG_REGULATE) { - if (tcp_background_io_enabled && - tp->t_inpcb->inp_socket->so_traffic_mgt_flags & TRAFFIC_MGT_SO_BG_SUPPRESSED) { - tp->t_flags |= TF_RXWIN0SENT; - return 0; /* Triggers TCP window closing by responding there is no space */ - } - } -#endif /* TRAFFIC_MGT */ - - /* Avoid inscreasing window size if the current window + /* Avoid increasing window size if the current window * is already very low, we could be in "persist" mode and * we could break some apps (see rdar://5409343) */ @@ -2222,42 +2590,29 @@ tcp_sbspace(struct tcpcb *tp) if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0 ) return imin(space, slowlink_wsize); - /* - * Check for ressources constraints before over-ajusting the amount of space we can - * advertise in the TCP window size updates. - */ - - if (sbspace_factor && (tp->t_inpcb->inp_pcbinfo->ipi_count < tcp_sockthreshold) && - (total_mb_cnt / 8) < (mbstat.m_clusters / sbspace_factor)) { - if (space < (int32_t)(sb->sb_maxused - sb->sb_cc)) {/* make sure we don't constrain the window if we have enough ressources */ - space = (int32_t) imax((sb->sb_maxused - sb->sb_cc), tp->rcv_maxbyps); - } - newspace = (int32_t) imax(((int32_t)sb->sb_maxused - sb->sb_cc), (int32_t)tp->rcv_maxbyps); - - if (newspace > space) - space = newspace; - } return space; } /* * Checks TCP Segment Offloading capability for a given connection and interface pair. */ void -tcp_set_tso(tp, ifp) - struct tcpcb *tp; - struct ifnet *ifp; +tcp_set_tso(struct tcpcb *tp, struct ifnet *ifp) { +#if MPTCP + /* + * We can't use TSO if this tcpcb belongs to an MPTCP session. + */ + if (tp->t_mpflags & TMPF_MPTCP_TRUE) { + tp->t_flags &= ~TF_TSO; + return; + } +#endif #if INET6 struct inpcb *inp = tp->t_inpcb; int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; if (isipv6) { - /* - * Radar 6921834: Disable TSO IPv6 because there is no support - * for TSO & HW checksum in ip6_output yet - */ -#if 0 - if (ifp && ifp->if_hwassist & IFNET_TSO_IPV6) { + if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV6)) { tp->t_flags |= TF_TSO; if (ifp->if_tso_v6_mtu != 0) tp->tso_max_segment_size = ifp->if_tso_v6_mtu; @@ -2266,12 +2621,11 @@ tcp_set_tso(tp, ifp) } else tp->t_flags &= ~TF_TSO; -#endif } else #endif /* INET6 */ { - if (ifp && ifp->if_hwassist & IFNET_TSO_IPV4) { + if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV4)) { tp->t_flags |= TF_TSO; if (ifp->if_tso_v4_mtu != 0) tp->tso_max_segment_size = ifp->if_tso_v4_mtu; @@ -2281,4 +2635,111 @@ tcp_set_tso(tp, ifp) tp->t_flags &= ~TF_TSO; } } + +#define TIMEVAL_TO_TCPHZ(_tv_) ((_tv_).tv_sec * TCP_RETRANSHZ + (_tv_).tv_usec / TCP_RETRANSHZ_TO_USEC) + +/* Function to calculate the tcp clock. The tcp clock will get updated + * at the boundaries of the tcp layer. This is done at 3 places: + * 1. Right before processing an input tcp packet + * 2. Whenever a connection wants to access the network using tcp_usrreqs + * 3. When a tcp timer fires or before tcp slow timeout + * + */ + +void +calculate_tcp_clock() +{ + struct timeval tv = tcp_uptime; + struct timeval interval = {0, TCP_RETRANSHZ_TO_USEC}; + struct timeval now, hold_now; + uint32_t incr = 0; + + microuptime(&now); + + /* + * Update coarse-grained networking timestamp (in sec.); the idea + * is to update the counter returnable via net_uptime() when + * we read time. + */ + net_update_uptime_secs(now.tv_sec); + + timevaladd(&tv, &interval); + if (timevalcmp(&now, &tv, >)) { + /* time to update the clock */ + lck_spin_lock(tcp_uptime_lock); + if (timevalcmp(&tcp_uptime, &now, >=)) { + /* clock got updated while we were waiting for the lock */ + lck_spin_unlock(tcp_uptime_lock); + return; + } + + microuptime(&now); + hold_now = now; + tv = tcp_uptime; + timevalsub(&now, &tv); + + incr = TIMEVAL_TO_TCPHZ(now); + if (incr > 0) { + tcp_uptime = hold_now; + tcp_now += incr; + } + + lck_spin_unlock(tcp_uptime_lock); + } + return; +} + +/* Compute receive window scaling that we are going to request + * for this connection based on sb_hiwat. Try to leave some + * room to potentially increase the window size upto a maximum + * defined by the constant tcp_autorcvbuf_max. + */ +void +tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so) { + u_int32_t maxsockbufsize; + + tp->request_r_scale = max(tcp_win_scale, tp->request_r_scale); + maxsockbufsize = ((so->so_rcv.sb_flags & SB_USRSIZE) != 0) ? + so->so_rcv.sb_hiwat : tcp_autorcvbuf_max; + + while (tp->request_r_scale < TCP_MAX_WINSHIFT && + (TCP_MAXWIN << tp->request_r_scale) < maxsockbufsize) + tp->request_r_scale++; + tp->request_r_scale = min(tp->request_r_scale, TCP_MAX_WINSHIFT); + +} + +int +tcp_notsent_lowat_check(struct socket *so) { + struct inpcb *inp = sotoinpcb(so); + struct tcpcb *tp = NULL; + int notsent = 0; + if (inp != NULL) { + tp = intotcpcb(inp); + } + + notsent = so->so_snd.sb_cc - + (tp->snd_nxt - tp->snd_una); + + /* When we send a FIN or SYN, not_sent can be negative. + * In that case also we need to send a write event to the + * process if it is waiting. In the FIN case, it will + * get an error from send because cantsendmore will be set. + */ + if (notsent <= tp->t_notsent_lowat) { + return(1); + } + + /* When Nagle's algorithm is not disabled, it is better + * to wakeup the client until there is atleast one + * maxseg of data to write. + */ + if ((tp->t_flags & TF_NODELAY) == 0 && + notsent > 0 && notsent < tp->t_maxseg) { + return(1); + } + return(0); +} + + /* DSEP Review Done pl-20051213-v02 @3253,@3391,@3400 */