X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/593a1d5fd87cdf5b46dd5fcb84467b432cea0f91..4b17d6b6e417f714551ec129064745ea9919780e:/bsd/netinet/tcp_subr.c?ds=sidebyside diff --git a/bsd/netinet/tcp_subr.c b/bsd/netinet/tcp_subr.c index a94f8ad2a..531b37fa9 100644 --- a/bsd/netinet/tcp_subr.c +++ b/bsd/netinet/tcp_subr.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2012 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -82,16 +82,19 @@ #include #include #include +#include #include #include #include #include +#define tcp_minmssoverload fring #define _IP_VHL #include #include #include +#include #if INET6 #include #endif @@ -101,6 +104,7 @@ #endif #include #include +#include #if INET6 #include #endif @@ -109,6 +113,9 @@ #include #include #include +#include +#include + #if INET6 #include #endif @@ -125,12 +132,17 @@ #endif #endif /*IPSEC*/ +#undef tcp_minmssoverload + #if CONFIG_MACF_NET #include #endif /* MAC_NET */ #include #include +#include + +#include #define DBG_FNC_TCP_CLOSE NETDBG_CODE(DBG_NETTCP, ((5 << 8) | 2)) @@ -142,16 +154,18 @@ extern int ipsec_bypass; #endif 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 @@ -161,7 +175,7 @@ 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"); /* @@ -177,59 +191,87 @@ __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, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW | CTLFLAG_LOCKED, &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 tcp_tcbhashsize = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, - &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); - 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"); 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 | 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; + +/* 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. @@ -241,6 +283,10 @@ extern int path_mtu_discovery; #define TCBHASHSIZE CONFIG_TCBHASHSIZE #endif +__private_extern__ int tcp_tcbhashsize = TCBHASHSIZE; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD | CTLFLAG_LOCKED, + &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); + /* * This is the actual shape of what we allocate using the zone * allocator. Doing it this way allows us to protect both structures @@ -250,28 +296,25 @@ extern int path_mtu_discovery; * 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 - -static struct tcpcb dummy_tcb; - extern struct inpcbhead time_wait_slots[]; -extern int cur_tw_slot; -extern u_long *delack_bitmask; -extern u_long route_generation; +extern struct tcptimerlist tcp_timer_list; 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) { @@ -286,6 +329,17 @@ 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 @@ -293,7 +347,6 @@ int tcp_freeq(struct tcpcb *tp); void tcp_init() { - int hashsize = TCBHASHSIZE; vm_size_t str_size; int i; struct inpcbinfo *pcbinfo; @@ -306,45 +359,69 @@ tcp_init() tcp_keepintvl = TCPTV_KEEPINTVL; 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; pcbinfo = &tcbinfo; - if (!powerof2(hashsize)) { + if (!powerof2(tcp_tcbhashsize)) { printf("WARNING: TCB hash size not a power of 2\n"); - hashsize = 512; /* safe default */ + tcp_tcbhashsize = 512; /* safe default */ } - tcp_tcbhashsize = hashsize; - tcbinfo.hashsize = hashsize; - tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); - tcbinfo.porthashbase = hashinit(hashsize, M_PCB, + 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); + str_size = P2ROUNDUP(sizeof(struct inp_tp), sizeof(u_int64_t)); tcbinfo.ipi_zone = (void *) zinit(str_size, 120000*str_size, 8192, "tcpcb"); + zone_change(tcbinfo.ipi_zone, Z_CALLERACCT, FALSE); + zone_change(tcbinfo.ipi_zone, Z_EXPAND, TRUE); + + 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 - dummy_tcb.t_state = TCP_NSTATES; - dummy_tcb.t_flags = 0; - tcbinfo.dummy_cb = (caddr_t) &dummy_tcb; - /* + /* * allocate lock group attribute and group for tcp pcb mutexes */ - pcbinfo->mtx_grp_attr = lck_grp_attr_alloc_init(); + pcbinfo->mtx_grp_attr = lck_grp_attr_alloc_init(); pcbinfo->mtx_grp = lck_grp_alloc_init("tcppcb", pcbinfo->mtx_grp_attr); /* @@ -357,21 +434,40 @@ tcp_init() return; /* pretty much dead if this fails... */ } - - in_pcb_nat_init(&tcbinfo, AF_INET, IPPROTO_TCP, SOCK_STREAM); - - delack_bitmask = _MALLOC((4 * hashsize)/32, M_PCB, M_WAITOK); - if (delack_bitmask == 0) - panic("Delack Memory"); - - for (i=0; i < (tcbinfo.hashsize / 32); i++) - delack_bitmask[i] = 0; - for (i=0; i < N_TIME_WAIT_SLOTS; i++) { LIST_INIT(&time_wait_slots[i]); } - timeout(tcp_fasttimo, NULL, hz/TCP_RETRANSHZ); + bzero(&tcp_timer_list, sizeof(tcp_timer_list)); + LIST_INIT(&tcp_timer_list.lhead); + /* + * allocate lock group attribute, group and attribute for the tcp timer list + */ + 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"); + } + + /* + * 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); + + /* Initialize TCP congestion control algorithms list */ + tcp_cc_init(); + + /* Initialize TCP LRO data structures */ + tcp_lro_init(); } /* @@ -401,7 +497,9 @@ tcp_fillheaders(tp, ip_ptr, tcp_ptr) 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_cksum_phdr(&inp->in6p_laddr, + &inp->in6p_faddr, htonl(sizeof(struct tcphdr)), + htonl(IPPROTO_TCP)); } else #endif { @@ -477,11 +575,8 @@ tcp_respond( tcp_seq ack, tcp_seq seq, int flags, -#if CONFIG_FORCE_OUT_IFP - ifnet_t ifp -#else - __unused ifnet_t ifp -#endif + unsigned int ifscope, + unsigned int nocell ) { register int tlen; @@ -496,7 +591,7 @@ tcp_respond( struct ip6_hdr *ip6; int isipv6; #endif /* INET6 */ - int ipflags = 0; + struct ifnet *outif; #if INET6 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6; @@ -507,8 +602,8 @@ tcp_respond( if (tp) { if (!(flags & TH_RST)) { win = tcp_sbspace(tp); - if (win > (long)TCP_MAXWIN << tp->rcv_scale) - win = (long)TCP_MAXWIN << tp->rcv_scale; + if (win > (int32_t)TCP_MAXWIN << tp->rcv_scale) + win = (int32_t)TCP_MAXWIN << tp->rcv_scale; } #if INET6 if (isipv6) @@ -539,13 +634,13 @@ tcp_respond( 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 */ { 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)); flags = TH_ACK; @@ -558,13 +653,17 @@ 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) { /* @@ -608,12 +707,7 @@ tcp_respond( mac_netinet_tcp_reply(m); } #endif - -#if CONFIG_IP_EDGEHOLE - if (tp && tp->t_inpcb) - ip_edgehole_mbuf_tag(tp->t_inpcb, m); -#endif - + nth->th_seq = htonl(seq); nth->th_ack = htonl(ack); nth->th_x2 = 0; @@ -627,9 +721,11 @@ 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_cksum_phdr(&ip6->ip6_src, + &ip6->ip6_dst, htons((u_short)(tlen - sizeof(struct ip6_hdr))), + htonl(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 : @@ -652,24 +748,71 @@ tcp_respond( return; } #endif + + 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.m_flowhash = tp->t_inpcb->inp_flowhash; + m->m_pkthdr.m_fhflags |= + (PF_TAG_TCP | PF_TAG_FLOWHASH | PF_TAG_FLOWADV); + } + #if INET6 if (isipv6) { - (void)ip6_output(m, NULL, ro6, ipflags, 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 }; + + 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 (ro6->ro_rt != NULL) { + if (ro6 == &sro6) { + rtfree(ro6->ro_rt); + ro6->ro_rt = NULL; + } else if ((outif = ro6->ro_rt->rt_ifp) != + tp->t_inpcb->in6p_last_outifp) { + tp->t_inpcb->in6p_last_outifp = outif; + } } } else #endif /* INET6 */ { -#if CONFIG_FORCE_OUT_IFP - ifp = (tp && tp->t_inpcb) ? tp->t_inpcb->pdp_ifp : - (ifp && (ifp->if_flags & IFF_POINTOPOINT) != 0) ? ifp : NULL; -#endif - (void) ip_output_list(m, 0, NULL, ro, ipflags, NULL, ifp); - if (ro == &sro && ro->ro_rt) { - rtfree(ro->ro_rt); - ro->ro_rt = NULL; + struct ip_out_args ipoa = { ifscope, { 0 }, + IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR }; + + 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 */ + inp_route_copyout(tp->t_inpcb, &sro); + } + /* + * For consistency, pass a local route copy. + */ + (void) ip_output(m, NULL, &sro, IP_OUTARGS, NULL, &ipoa); + + if (ro != &sro) { + if (sro.ro_rt != NULL && + (outif = sro.ro_rt->rt_ifp) != + tp->t_inpcb->inp_last_outifp) + tp->t_inpcb->inp_last_outifp = outif; + /* Synchronize cached PCB route */ + inp_route_copyin(tp->t_inpcb, &sro); + } else if (sro.ro_rt != NULL) { + rtfree(sro.ro_rt); } } } @@ -691,13 +834,15 @@ tcp_newtcpcb(inp) int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ + calculate_tcp_clock(); + if (so->cached_in_sock_layer == 0) { - it = (struct inp_tp *)inp; + it = (struct inp_tp *)(void *)inp; tp = &it->tcb; } else - tp = (struct tcpcb *) inp->inp_saved_ppcb; - + tp = (struct tcpcb *)(void *)inp->inp_saved_ppcb; + bzero((char *) tp, sizeof(struct tcpcb)); LIST_INIT(&tp->t_segq); tp->t_maxseg = tp->t_maxopd = @@ -720,12 +865,23 @@ 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; /* * IPv4 TTL initialization is necessary for an IPv6 socket as well, * because the socket may be bound to an IPv6 wildcard address, @@ -747,8 +903,13 @@ 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++; @@ -760,6 +921,39 @@ 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 @@ -775,30 +969,14 @@ tcp_close(tp) #if INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ - register struct rtentry *rt; + 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; - } - } + 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 @@ -812,12 +990,24 @@ 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); } - lck_mtx_lock(rt_mtx); + 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; +#else + rt = inp->inp_route.ro_rt; +#endif + if (rt != NULL) + RT_LOCK_SPIN(rt); + /* * If we got enough samples through the srtt filter, * save the rtt and rttvar in the routing entry. @@ -830,30 +1020,32 @@ tcp_close(tp) * update anything that the user "locked". */ if (tp->t_rttupdated >= 16) { - register u_long i = 0; + register u_int32_t i = 0; #if INET6 if (isipv6) { struct sockaddr_in6 *sin6; - if ((rt = inp->in6p_route.ro_rt) == NULL) + 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 */ - rt = inp->inp_route.ro_rt; - if (rt == NULL || - ((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 (rt == NULL || !(rt->rt_flags & RTF_UP) || + ((struct sockaddr_in *)(void *)rt_key(rt))->sin_addr.s_addr == + INADDR_ANY || rt->generation_id != route_generation) { + 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; } + RT_LOCK_ASSERT_HELD(rt); if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { i = tp->t_srtt * (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE)); @@ -909,7 +1101,7 @@ tcp_close(tp) i = (i + tp->t_maxseg / 2) / tp->t_maxseg; if (i < 2) i = 2; - i *= (u_long)(tp->t_maxseg + + i *= (u_int32_t)(tp->t_maxseg + #if INET6 (isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : @@ -927,25 +1119,28 @@ tcp_close(tp) tcpstat.tcps_cachedssthresh++; } } - rt = inp->inp_route.ro_rt; - if (rt) { - /* - * mark route for deletion if no information is - * cached. - */ - if ((so->so_flags & SOF_OVERFLOW) && tcp_lq_overflow && - ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0)){ - if (rt->rt_rmx.rmx_rtt == 0) - rt->rt_flags |= RTF_DELCLONE; + + /* + * Mark route for deletion if no information is cached. + */ + if (rt != NULL && (so->so_flags & SOF_OVERFLOW) && tcp_lq_overflow) { + if (!(rt->rt_rmx.rmx_locks & RTV_RTT) && + rt->rt_rmx.rmx_rtt == 0) { + rt->rt_flags |= RTF_DELCLONE; } } - no_valid_rt: - /* free the reassembly queue, if any */ - lck_mtx_unlock(rt_mtx); +no_valid_rt: + if (rt != NULL) + RT_UNLOCK(rt); + + /* free the reassembly queue, if any */ (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) @@ -956,17 +1151,33 @@ tcp_close(tp) if (so->cached_in_sock_layer) inp->inp_saved_ppcb = (caddr_t) tp; #endif + /* 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; + } - soisdisconnected(so); #if INET6 if (INP_CHECK_SOCKAF(so, AF_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 @@ -980,7 +1191,7 @@ 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; } @@ -1014,7 +1225,7 @@ tcp_drain() != NULL) { LIST_REMOVE(te, tqe_q); m_freem(te->tqe_m); - FREE(te, M_TSEGQ); + zfree(tcp_reass_zone, te); tcp_reass_qsize--; } } @@ -1067,6 +1278,93 @@ 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. + */ +static void +tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp) +{ + int i; + + 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_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; + otp->t_flags = tp->t_flags; + otp->t_force = tp->t_force; + otp->snd_una = tp->snd_una; + otp->snd_max = tp->snd_max; + otp->snd_nxt = tp->snd_nxt; + otp->snd_up = tp->snd_up; + otp->snd_wl1 = tp->snd_wl1; + otp->snd_wl2 = tp->snd_wl2; + otp->iss = tp->iss; + otp->irs = tp->irs; + otp->rcv_nxt = tp->rcv_nxt; + otp->rcv_adv = tp->rcv_adv; + otp->rcv_wnd = tp->rcv_wnd; + otp->rcv_up = tp->rcv_up; + otp->snd_wnd = tp->snd_wnd; + otp->snd_cwnd = tp->snd_cwnd; + otp->snd_ssthresh = tp->snd_ssthresh; + otp->t_maxopd = tp->t_maxopd; + otp->t_rcvtime = tp->t_rcvtime; + otp->t_starttime = tp->t_starttime; + otp->t_rtttime = tp->t_rtttime; + otp->t_rtseq = tp->t_rtseq; + otp->t_rxtcur = tp->t_rxtcur; + otp->t_maxseg = tp->t_maxseg; + otp->t_srtt = tp->t_srtt; + otp->t_rttvar = tp->t_rttvar; + otp->t_rxtshift = tp->t_rxtshift; + otp->t_rttmin = tp->t_rttmin; + otp->t_rttupdated = tp->t_rttupdated; + otp->max_sndwnd = tp->max_sndwnd; + otp->t_softerror = tp->t_softerror; + otp->t_oobflags = tp->t_oobflags; + otp->t_iobc = tp->t_iobc; + otp->snd_scale = tp->snd_scale; + otp->rcv_scale = tp->rcv_scale; + otp->request_r_scale = tp->request_r_scale; + otp->requested_s_scale = tp->requested_s_scale; + otp->ts_recent = tp->ts_recent; + otp->ts_recent_age = tp->ts_recent_age; + otp->last_ack_sent = tp->last_ack_sent; + otp->cc_send = tp->cc_send; + otp->cc_recv = tp->cc_recv; + 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; +} + static int tcp_pcblist SYSCTL_HANDLER_ARGS { @@ -1160,10 +1458,11 @@ tcp_pcblist SYSCTL_HANDLER_ARGS inpcb_to_compat(inp, &xt.xt_inp); inp_ppcb = inp->inp_ppcb; if (inp_ppcb != NULL) { - bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); - } - else + tcpcb_to_otcpcb((struct tcpcb *)(void *)inp_ppcb, + &xt.xt_tp); + } else { bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); + } if (inp->inp_socket) sotoxsocket(inp->inp_socket, &xt.xt_socket); error = SYSCTL_OUT(req, &xt, sizeof xt); @@ -1189,88 +1488,219 @@ tcp_pcblist SYSCTL_HANDLER_ARGS 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"); -#ifndef __APPLE__ +#if !CONFIG_EMBEDDED + +static void +tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp) +{ + int i; + + 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_EXT; i++) + otp->t_timer[i] = tp->t_timer[i]; + otp->t_state = tp->t_state; + otp->t_flags = tp->t_flags; + otp->t_force = tp->t_force; + otp->snd_una = tp->snd_una; + otp->snd_max = tp->snd_max; + otp->snd_nxt = tp->snd_nxt; + otp->snd_up = tp->snd_up; + otp->snd_wl1 = tp->snd_wl1; + otp->snd_wl2 = tp->snd_wl2; + otp->iss = tp->iss; + otp->irs = tp->irs; + otp->rcv_nxt = tp->rcv_nxt; + otp->rcv_adv = tp->rcv_adv; + otp->rcv_wnd = tp->rcv_wnd; + otp->rcv_up = tp->rcv_up; + otp->snd_wnd = tp->snd_wnd; + otp->snd_cwnd = tp->snd_cwnd; + otp->snd_ssthresh = tp->snd_ssthresh; + otp->t_maxopd = tp->t_maxopd; + otp->t_rcvtime = tp->t_rcvtime; + otp->t_starttime = tp->t_starttime; + otp->t_rtttime = tp->t_rtttime; + otp->t_rtseq = tp->t_rtseq; + otp->t_rxtcur = tp->t_rxtcur; + otp->t_maxseg = tp->t_maxseg; + otp->t_srtt = tp->t_srtt; + otp->t_rttvar = tp->t_rttvar; + otp->t_rxtshift = tp->t_rxtshift; + otp->t_rttmin = tp->t_rttmin; + otp->t_rttupdated = tp->t_rttupdated; + otp->max_sndwnd = tp->max_sndwnd; + otp->t_softerror = tp->t_softerror; + otp->t_oobflags = tp->t_oobflags; + otp->t_iobc = tp->t_iobc; + otp->snd_scale = tp->snd_scale; + otp->rcv_scale = tp->rcv_scale; + otp->request_r_scale = tp->request_r_scale; + otp->requested_s_scale = tp->requested_s_scale; + otp->ts_recent = tp->ts_recent; + otp->ts_recent_age = tp->ts_recent_age; + otp->last_ack_sent = tp->last_ack_sent; + otp->cc_send = tp->cc_send; + otp->cc_recv = tp->cc_recv; + 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; +} + + static int -tcp_getcred(SYSCTL_HANDLER_ARGS) +tcp_pcblist64 SYSCTL_HANDLER_ARGS { - struct sockaddr_in addrs[2]; - struct inpcb *inp; - int error, s; - - error = suser(req->p); - if (error) - return (error); - error = SYSCTL_IN(req, addrs, sizeof(addrs)); - if (error) - return (error); - s = splnet(); - inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, - addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); - if (inp == NULL || inp->inp_socket == NULL) { - error = ENOENT; - goto out; - } - error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(*(kauth_cred_t)0); -out: - splx(s); - return (error); +#pragma unused(oidp, arg1, arg2) + int error, i, n; + struct inpcb *inp, **inp_list; + 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); + 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); + return 0; + } + + if (req->newptr != USER_ADDR_NULL) { + lck_rw_done(tcbinfo.mtx); + return EPERM; + } + + /* + * OK, now we're committed to doing something. + */ + gencnt = tcbinfo.ipi_gencnt; + n = tcbinfo.ipi_count; + + bzero(&xig, sizeof(xig)); + xig.xig_len = sizeof xig; + xig.xig_count = n; + xig.xig_gen = gencnt; + xig.xig_sogen = so_gencnt; + error = SYSCTL_OUT(req, &xig, sizeof xig); + if (error) { + lck_rw_done(tcbinfo.mtx); + return error; + } + /* + * We are done if there is no pcb + */ + if (n == 0) { + lck_rw_done(tcbinfo.mtx); + return 0; + } + + inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK); + if (inp_list == 0) { + lck_rw_done(tcbinfo.mtx); + 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 + inp_list[i++] = inp; + } + + 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; + } + } + + n = i; + + error = 0; + 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); + } + } + 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); + } + FREE(inp_list, M_TEMP); + lck_rw_done(tcbinfo.mtx); + return error; } -SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, - 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection"); +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 */ -#if INET6 static int -tcp6_getcred(SYSCTL_HANDLER_ARGS) +tcp_pcblist_n SYSCTL_HANDLER_ARGS { - struct sockaddr_in6 addrs[2]; - struct inpcb *inp; - int error, s, mapped = 0; - - error = suser(req->p); - if (error) - return (error); - error = SYSCTL_IN(req, addrs, sizeof(addrs)); - if (error) - return (error); - if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { - if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) - mapped = 1; - else - return (EINVAL); - } - s = splnet(); - if (mapped == 1) - inp = in_pcblookup_hash(&tcbinfo, - *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], - addrs[1].sin6_port, - *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], - addrs[0].sin6_port, - 0, NULL); - else - inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, - addrs[1].sin6_port, - &addrs[0].sin6_addr, addrs[0].sin6_port, - 0, NULL); - if (inp == NULL || inp->inp_socket == NULL) { - error = ENOENT; - goto out; - } - error = SYSCTL_OUT(req, inp->inp_socket->so_cred, - sizeof(*(kauth_cred_t)0); -out: - splx(s); - return (error); +#pragma unused(oidp, arg1, arg2) + int error = 0; + + error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo); + + return error; } -SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, - 0, 0, - tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection"); -#endif -#endif /* __APPLE__*/ + +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(unsigned int ifindex, uint8_t *bitfield) +{ + inpcb_get_ports_used(ifindex, bitfield, &tcbinfo); +} + +__private_extern__ uint32_t +tcp_count_opportunistic(unsigned int ifindex, u_int32_t flags) +{ + return inpcb_count_opportunistic(ifindex, &tcbinfo, flags); +} void tcp_ctlinput(cmd, sa, vip) @@ -1278,48 +1708,107 @@ tcp_ctlinput(cmd, sa, vip) struct sockaddr *sa; void *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; - tcp_seq icmp_seq; - 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; - if (cmd == PRC_QUENCH) - notify = tcp_quench; + if (cmd == PRC_MSGSIZE) + notify = tcp_mtudisc; else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT) && ip) notify = tcp_drop_syn_sent; - else if (cmd == PRC_MSGSIZE) - notify = tcp_mtudisc; else if (PRC_IS_REDIRECT(cmd)) { ip = 0; notify = in_rtchange; } else if (cmd == PRC_HOSTDEAD) ip = 0; + /* Source quench is deprecated */ + else if (cmd == PRC_QUENCH) + return; else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) return; if (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_seq = htonl(th->th_seq); + icmp_tcp_seq = htonl(th.th_seq); tp = intotcpcb(inp); - if (SEQ_GEQ(icmp_seq, tp->snd_una) && - SEQ_LT(icmp_seq, tp->snd_max)) + if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && + SEQ_LT(icmp_tcp_seq, tp->snd_max)) { + if (cmd == PRC_MSGSIZE) { + + /* + * MTU discovery: + * If we got a needfrag and there is a host route to the + * original destination, and the MTU is not locked, then + * set the MTU in the route to the suggested new value + * (if given) and then notify as usual. The ULPs will + * notice that the MTU has changed and adapt accordingly. + * If no new MTU was suggested, then we guess a new one + * less than the current value. If the new MTU is + * unreasonably small (defined by sysctl tcp_minmss), then + * we reset the MTU to the interface value and enable the + * lock bit, indicating that we are no longer doing MTU + * discovery. + */ + struct rtentry *rt; + int mtu; + struct sockaddr_in icmpsrc = { sizeof (struct sockaddr_in), AF_INET, + 0 , { 0 }, { 0,0,0,0,0,0,0,0 } }; + icmpsrc.sin_addr = icp->icmp_ip.ip_dst; + + rt = rtalloc1((struct sockaddr *)&icmpsrc, 0, + RTF_CLONING | RTF_PRCLONING); + if (rt != NULL) { + RT_LOCK(rt); + if ((rt->rt_flags & RTF_HOST) && + !(rt->rt_rmx.rmx_locks & RTV_MTU)) { + mtu = ntohs(icp->icmp_nextmtu); + if (!mtu) + mtu = ip_next_mtu(rt->rt_rmx. + rmx_mtu, 1); +#if DEBUG_MTUDISC + printf("MTU for %s reduced to %d\n", + inet_ntop(AF_INET, + &icmpsrc.sin_addr, ipv4str, + sizeof (ipv4str)), mtu); +#endif + if (mtu < max(296, (tcp_minmss + + sizeof (struct tcpiphdr)))) { + /* rt->rt_rmx.rmx_mtu = + rt->rt_ifp->if_mtu; */ + rt->rt_rmx.rmx_locks |= RTV_MTU; + } else if (rt->rt_rmx.rmx_mtu > mtu) { + rt->rt_rmx.rmx_mtu = mtu; + } + } + RT_UNLOCK(rt); + rtfree(rt); + } + } + (*notify)(inp, inetctlerrmap[cmd]); + } tcp_unlock(inp->inp_socket, 1, 0); } } else @@ -1349,13 +1838,14 @@ tcp6_ctlinput(cmd, sa, d) sa->sa_len != sizeof(struct sockaddr_in6)) return; - if (cmd == PRC_QUENCH) - notify = tcp_quench; - else if (cmd == PRC_MSGSIZE) + if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc; else if (!PRC_IS_REDIRECT(cmd) && ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) return; + /* Source quench is deprecated */ + else if (cmd == PRC_QUENCH) + return; /* if the parameter is from icmp6, decode it. */ if (d != NULL) { @@ -1386,10 +1876,11 @@ tcp6_ctlinput(cmd, sa, d) in6_pcbnotify(&tcbinfo, sa, th.th_dport, (struct sockaddr *)ip6cp->ip6c_src, - th.th_sport, cmd, notify); - } else - in6_pcbnotify(&tcbinfo, sa, 0, (struct sockaddr *)sa6_src, - 0, cmd, notify); + th.th_sport, cmd, NULL, notify); + } else { + in6_pcbnotify(&tcbinfo, sa, 0, + (struct sockaddr *)(size_t)sa6_src, 0, cmd, NULL, notify); + } } #endif /* INET6 */ @@ -1498,23 +1989,6 @@ tcp_new_isn(tp) return new_isn; } -/* - * When a source quench is received, close congestion window - * to one segment. We will gradually open it again as we proceed. - */ -void -tcp_quench( - struct inpcb *inp, - __unused int errno -) -{ - struct tcpcb *tp = intotcpcb(inp); - - if (tp) { - tp->snd_cwnd = tp->t_maxseg; - tp->t_bytes_acked = 0; - } -} /* * When a specific ICMP unreachable message is received and the @@ -1555,20 +2029,22 @@ tcp_mtudisc( #endif /* INET6 */ if (tp) { - lck_mtx_lock(rt_mtx); #if INET6 if (isipv6) - rt = tcp_rtlookup6(inp); + rt = tcp_rtlookup6(inp, IFSCOPE_NONE); else #endif /* INET6 */ - rt = tcp_rtlookup(inp); + rt = tcp_rtlookup(inp, IFSCOPE_NONE); if (!rt || !rt->rt_rmx.rmx_mtu) { tp->t_maxopd = tp->t_maxseg = #if INET6 isipv6 ? tcp_v6mssdflt : #endif /* INET6 */ tcp_mssdflt; - lck_mtx_unlock(rt_mtx); + + /* Route locked during lookup above */ + if (rt != NULL) + RT_UNLOCK(rt); return; } taop = rmx_taop(rt->rt_rmx); @@ -1584,7 +2060,9 @@ tcp_mtudisc( #endif /* INET6 */ ; - lck_mtx_unlock(rt_mtx); + /* Route locked during lookup above */ + RT_UNLOCK(rt); + if (offered) mss = min(mss, offered); /* @@ -1617,6 +2095,11 @@ tcp_mtudisc( 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; @@ -1626,39 +2109,54 @@ tcp_mtudisc( /* * Look-up the routing entry to the peer of this inpcb. If no route - * is found and it cannot be allocated then return NULL. This routine + * is found and it cannot be allocated the return NULL. This routine * is called by TCP routines that access the rmx structure and by tcp_mss - * to get the interface MTU. + * to get the interface MTU. If a route is found, this routine will + * hold the rtentry lock; the caller is responsible for unlocking. */ struct rtentry * -tcp_rtlookup(inp) +tcp_rtlookup(inp, input_ifscope) struct inpcb *inp; + unsigned int input_ifscope; { struct route *ro; struct rtentry *rt; struct tcpcb *tp; - ro = &inp->inp_route; - if (ro == NULL) - return (NULL); - rt = ro->ro_rt; + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); - lck_mtx_assert(rt_mtx, LCK_MTX_ASSERT_OWNED); + ro = &inp->inp_route; + if ((rt = ro->ro_rt) != NULL) + RT_LOCK(rt); - if (rt == NULL || !(rt->rt_flags & RTF_UP) || rt->generation_id != route_generation) { + if (rt == NULL || !(rt->rt_flags & RTF_UP) || + rt->generation_id != route_generation) { /* 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; - rtalloc_ign_locked(ro, 0UL); - rt = ro->ro_rt; + + /* + * 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; + + if (rt != NULL) + RT_UNLOCK(rt); + rtalloc_scoped(ro, ifscope); + if ((rt = ro->ro_rt) != NULL) + RT_LOCK(rt); } } - if (rt != NULL && rt->rt_ifp != NULL) - somultipages(inp->inp_socket, - (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES)); /* * Update MTU discovery determination. Don't do it if: @@ -1676,47 +2174,72 @@ tcp_rtlookup(inp) else tp->t_flags |= TF_PMTUD; -#ifdef IFEF_NOWINDOWSCALE - if (tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL && - (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE) != 0) - { - // Timestamps are not enabled on this interface - tp->t_flags &= ~(TF_REQ_SCALE); +#if CONFIG_IFEF_NOWINDOWSCALE + if (tcp_obey_ifef_nowindowscale && + tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL && + (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE)) { + /* Window scaling is enabled on this interface */ + tp->t_flags &= ~TF_REQ_SCALE; } #endif + if (rt != NULL && rt->rt_ifp != NULL) { + somultipages(inp->inp_socket, + (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES)); + tcp_set_tso(tp, rt->rt_ifp); + } + + /* + * Caller needs to call RT_UNLOCK(rt). + */ return rt; } #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; struct tcpcb *tp; - lck_mtx_assert(rt_mtx, LCK_MTX_ASSERT_OWNED); + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); ro6 = &inp->in6p_route; - rt = ro6->ro_rt; - if (rt == NULL || !(rt->rt_flags & RTF_UP)) { + if ((rt = ro6->ro_rt) != NULL) + RT_LOCK(rt); + + if (rt == NULL || !(rt->rt_flags & RTF_UP) || + rt->generation_id != route_generation) { /* 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; - rtalloc_ign_locked((struct route *)ro6, 0UL); - rt = ro6->ro_rt; + + /* + * 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; + + if (rt != NULL) + RT_UNLOCK(rt); + rtalloc_scoped((struct route *)ro6, ifscope); + if ((rt = ro6->ro_rt) != NULL) + RT_LOCK(rt); } } - if (rt != NULL && rt->rt_ifp != NULL) - somultipages(inp->inp_socket, - (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES)); /* * Update path MTU Discovery determination * while looking up the route: @@ -1742,6 +2265,24 @@ tcp_rtlookup6(inp) else tp->t_flags |= TF_PMTUD; +#if CONFIG_IFEF_NOWINDOWSCALE + if (tcp_obey_ifef_nowindowscale && + tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL && + (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE)) { + /* Window scaling is not enabled on this interface */ + tp->t_flags &= ~TF_REQ_SCALE; + } +#endif + + if (rt != NULL && rt->rt_ifp != NULL) { + somultipages(inp->inp_socket, + (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES)); + tcp_set_tso(tp, rt->rt_ifp); + } + + /* + * Caller needs to call RT_UNLOCK(rt). + */ return rt; } #endif /* INET6 */ @@ -1770,7 +2311,7 @@ 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); @@ -1801,23 +2342,25 @@ tcp_gettaocache(inp) struct rtentry *rt; struct rmxp_tao *taop; - lck_mtx_lock(rt_mtx); #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); + rt = tcp_rtlookup(inp, IFSCOPE_NONE); /* Make sure this is a host route and is up. */ if (rt == NULL || (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) { - lck_mtx_unlock(rt_mtx); + /* Route locked during lookup above */ + if (rt != NULL) + RT_UNLOCK(rt); return NULL; } taop = rmx_taop(rt->rt_rmx); - lck_mtx_unlock(rt_mtx); + /* Route locked during lookup above */ + RT_UNLOCK(rt); return (taop); } @@ -1835,62 +2378,68 @@ tcp_cleartaocache() } int -tcp_lock(so, refcount, lr) - struct socket *so; - int refcount; - int lr; +tcp_lock(struct socket *so, int refcount, void *lr) { - int lr_saved; - if (lr == 0) - lr_saved = (unsigned int) __builtin_return_address(0); - else lr_saved = lr; + void *lr_saved; - if (so->so_pcb) { - lck_mtx_lock(((struct inpcb *)so->so_pcb)->inpcb_mtx); - } - else { - panic("tcp_lock: so=%p NO PCB! lr=%x\n", so, lr_saved); - lck_mtx_lock(so->so_proto->pr_domain->dom_mtx); + if (lr == NULL) + lr_saved = __builtin_return_address(0); + else + lr_saved = lr; + + if (so->so_pcb != NULL) { + 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)); + /* NOTREACHED */ } - if (so->so_usecount < 0) - panic("tcp_lock: so=%p so_pcb=%p lr=%x ref=%x\n", - so, so->so_pcb, lr_saved, so->so_usecount); - + if (so->so_usecount < 0) { + panic("tcp_lock: so=%p so_pcb=%p lr=%p ref=%x lrh= %s\n", + so, so->so_pcb, lr_saved, so->so_usecount, solockhistory_nr(so)); + /* NOTREACHED */ + } if (refcount) so->so_usecount++; - so->lock_lr[so->next_lock_lr] = (u_int32_t)lr_saved; + so->lock_lr[so->next_lock_lr] = lr_saved; so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; return (0); } int -tcp_unlock(so, refcount, lr) - struct socket *so; - int refcount; - int lr; +tcp_unlock(struct socket *so, int refcount, void *lr) { - int lr_saved; - if (lr == 0) - lr_saved = (unsigned int) __builtin_return_address(0); - else lr_saved = lr; + void *lr_saved; + + if (lr == NULL) + lr_saved = __builtin_return_address(0); + else + lr_saved = lr; #ifdef MORE_TCPLOCK_DEBUG - printf("tcp_unlock: so=%p sopcb=%x lock=%x ref=%x lr=%x\n", - so, so->so_pcb, ((struct inpcb *)so->so_pcb)->inpcb_mtx, so->so_usecount, lr_saved); + 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); #endif if (refcount) so->so_usecount--; - if (so->so_usecount < 0) - panic("tcp_unlock: so=%p usecount=%x\n", so, so->so_usecount); - if (so->so_pcb == NULL) - panic("tcp_unlock: so=%p NO PCB usecount=%x lr=%x\n", so, so->so_usecount, lr_saved); - else { - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); - so->unlock_lr[so->next_unlock_lr] = (u_int32_t)lr_saved; + if (so->so_usecount < 0) { + panic("tcp_unlock: so=%p usecount=%x lrh= %s\n", + so, so->so_usecount, solockhistory_nr(so)); + /* NOTREACHED */ + } + if (so->so_pcb == NULL) { + panic("tcp_unlock: so=%p NO PCB usecount=%x lr=%p lrh= %s\n", + so, so->so_usecount, lr_saved, solockhistory_nr(so)); + /* NOTREACHED */ + } else { + 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); } @@ -1904,34 +2453,49 @@ tcp_getlock( if (so->so_pcb) { if (so->so_usecount < 0) - panic("tcp_getlock: so=%p usecount=%x\n", so, so->so_usecount); - return(inp->inpcb_mtx); + panic("tcp_getlock: so=%p usecount=%x lrh= %s\n", + so, so->so_usecount, solockhistory_nr(so)); + return(&inp->inpcb_mtx); } else { - panic("tcp_getlock: so=%p NULL so_pcb\n", so); + panic("tcp_getlock: so=%p NULL so_pcb %s\n", + so, solockhistory_nr(so)); return (so->so_proto->pr_domain->dom_mtx); } } -long + +/* 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; + if (tcp_do_autorcvbuf == 1 && + tcp_cansbgrow(sb) && + (tp->t_flags & TF_SLOWLINK) == 0 && + (sb->sb_hiwat - sb->sb_cc) < rcvbufinc && + (sb->sb_hiwat < 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; - long space, newspace; + int32_t space; - space = ((long) lmin((sb->sb_hiwat - sb->sb_cc), - (sb->sb_mbmax - sb->sb_mbcnt))); + tcp_sbrcv_grow_rwin(tp, sb); -#if TRAFFIC_MGT - if (tp->t_inpcb->inp_socket->so_traffic_mgt_flags & TRAFFIC_MGT_SO_BACKGROUND) { - 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 */ + space = ((int32_t) imin((sb->sb_hiwat - sb->sb_cc), + (sb->sb_mbmax - sb->sb_mbcnt))); + if (space < 0) + space = 0; - /* 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) */ @@ -1942,23 +2506,143 @@ tcp_sbspace(struct tcpcb *tp) /* Clip window size for slower link */ if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0 ) - return lmin(space, slowlink_wsize); + 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. - */ + 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; +{ +#if INET6 + struct inpcb *inp = tp->t_inpcb; + int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; + + if (isipv6) { + 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; + else + tp->tso_max_segment_size = TCP_MAXWIN; + } else + tp->t_flags &= ~TF_TSO; - if (sbspace_factor && (tp->t_inpcb->inp_pcbinfo->ipi_count < tcp_sockthreshold) && - (total_mb_cnt / 8) < (mbstat.m_clusters / sbspace_factor)) { - if (space < (long)(sb->sb_maxused - sb->sb_cc)) {/* make sure we don't constrain the window if we have enough ressources */ - space = (long) lmax((sb->sb_maxused - sb->sb_cc), tp->rcv_maxbyps); + } else +#endif /* INET6 */ + + { + 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; + else + tp->tso_max_segment_size = TCP_MAXWIN; + } else + 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; + + timevaladd(&tv, &interval); + microuptime(&now); + 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; } - newspace = (long) lmax(((long)sb->sb_maxused - sb->sb_cc), (long)tp->rcv_maxbyps); - if (newspace > space) - space = newspace; + 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); } - return space; + + 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 */