X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/3e170ce000f1506b7b5d2c5c7faec85ceabb573d..4d15aeb193b2c68f1d38666c317f8d3734f5f083:/bsd/netinet/tcp_subr.c diff --git a/bsd/netinet/tcp_subr.c b/bsd/netinet/tcp_subr.c index 6fafa0f5f..b2e18eb05 100644 --- a/bsd/netinet/tcp_subr.c +++ b/bsd/netinet/tcp_subr.c @@ -1,8 +1,8 @@ /* - * Copyright (c) 2000-2015 Apple Inc. All rights reserved. + * Copyright (c) 2000-2016 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ - * + * * This file contains Original Code and/or Modifications of Original Code * as defined in and that are subject to the Apple Public Source License * Version 2.0 (the 'License'). You may not use this file except in @@ -11,10 +11,10 @@ * unlawful or unlicensed copies of an Apple operating system, or to * circumvent, violate, or enable the circumvention or violation of, any * terms of an Apple operating system software license agreement. - * + * * Please obtain a copy of the License at * http://www.opensource.apple.com/apsl/ and read it before using this file. - * + * * The Original Code and all software distributed under the License are * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, @@ -22,7 +22,7 @@ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. * Please see the License for the specific language governing rights and * limitations under the License. - * + * * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* @@ -58,7 +58,6 @@ * SUCH DAMAGE. * * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 - * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.22 2001/08/22 00:59:12 silby Exp $ */ /* * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce @@ -92,8 +91,8 @@ #include #include -#define tcp_minmssoverload fring -#define _IP_VHL +#define tcp_minmssoverload fring +#define _IP_VHL #include #include #include @@ -134,7 +133,7 @@ #if INET6 #include #endif -#endif /*IPSEC*/ +#endif /* IPSEC */ #if NECP #include @@ -154,7 +153,7 @@ #include -#define DBG_FNC_TCP_CLOSE NETDBG_CODE(DBG_NETTCP, ((5 << 8) | 2)) +#define DBG_FNC_TCP_CLOSE NETDBG_CODE(DBG_NETTCP, ((5 << 8) | 2)) extern int tcp_lq_overflow; @@ -163,38 +162,39 @@ extern struct tcptailq tcp_tw_tailq; int tcp_mssdflt = TCP_MSS; SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW | CTLFLAG_LOCKED, - &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); + &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 | CTLFLAG_LOCKED, &tcp_v6mssdflt , 0, + CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_v6mssdflt, 0, "Default TCP Maximum Segment Size for IPv6"); #endif extern int tcp_do_autorcvbuf; -int tcp_sysctl_fastopenkey(struct sysctl_oid *, void *, int , +int tcp_sysctl_fastopenkey(struct sysctl_oid *, void *, int, struct sysctl_req *); -SYSCTL_PROC(_net_inet_tcp, OID_AUTO, fastopen_key, - CTLTYPE_STRING | CTLFLAG_WR, - 0 , 0, tcp_sysctl_fastopenkey, "S", "TCP Fastopen key"); +SYSCTL_PROC(_net_inet_tcp, OID_AUTO, fastopen_key, CTLTYPE_STRING | CTLFLAG_WR, + 0, 0, tcp_sysctl_fastopenkey, "S", "TCP Fastopen key"); /* Current count of half-open TFO connections */ int tcp_tfo_halfcnt = 0; /* Maximum of half-open TFO connection backlog */ int tcp_tfo_backlog = 10; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen_backlog, CTLFLAG_RW | CTLFLAG_LOCKED, - &tcp_tfo_backlog, 0, "Backlog queue for half-open TFO connections"); +SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen_backlog, + CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_tfo_backlog, 0, + "Backlog queue for half-open TFO connections"); int tcp_fastopen = TCP_FASTOPEN_CLIENT | TCP_FASTOPEN_SERVER; SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen, CTLFLAG_RW | CTLFLAG_LOCKED, - &tcp_fastopen, 0, "Enable TCP Fastopen (RFC 7413)"); + &tcp_fastopen, 0, "Enable TCP Fastopen (RFC 7413)"); int tcp_tfo_fallback_min = 10; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen_fallback_min, CTLFLAG_RW | CTLFLAG_LOCKED, - &tcp_tfo_fallback_min, 0, "Mininum number of trials without TFO when in fallback mode"); +SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen_fallback_min, + CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_tfo_fallback_min, 0, + "Mininum number of trials without TFO when in fallback mode"); /* * Minimum MSS we accept and use. This prevents DoS attacks where @@ -206,62 +206,64 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen_fallback_min, CTLFLAG_RW | CTLFLAG_ */ int tcp_minmss = TCP_MINMSS; SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW | CTLFLAG_LOCKED, - &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); + &tcp_minmss, 0, "Minmum TCP Maximum Segment Size"); int tcp_do_rfc1323 = 1; #if (DEVELOPMENT || DEBUG) SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, - CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_do_rfc1323 , 0, - "Enable rfc1323 (high performance TCP) extensions"); + CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_do_rfc1323, 0, + "Enable rfc1323 (high performance TCP) extensions"); #endif /* (DEVELOPMENT || DEBUG) */ // Not used static int tcp_do_rfc1644 = 0; -SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW | CTLFLAG_LOCKED, - &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); +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 | CTLFLAG_LOCKED, &do_tcpdrain, 0, - "Enable tcp_drain routine for extra help when low on mbufs"); +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 | CTLFLAG_LOCKED, - &tcbinfo.ipi_count, 0, "Number of active PCBs"); +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"); +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 | CTLFLAG_LOCKED, &icmp_may_rst, 0, - "Certain ICMP unreachable messages may abort connections in SYN_SENT"); +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; static int tcp_isn_reseed_interval = 0; #if (DEVELOPMENT || DEBUG) -SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948, - CTLFLAG_RW | CTLFLAG_LOCKED, - &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly"); +SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly"); 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"); + CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); #endif /* (DEVELOPMENT || DEBUG) */ 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"); + &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"); +SYSCTL_INT(_net_inet_tcp, OID_AUTO, randomize_ports, + CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_use_randomport, 0, + "Randomize TCP port numbers"); __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"); + CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_win_scale, 0, + "Window scaling factor"); static void tcp_cleartaocache(void); static void tcp_notify(struct inpcb *, int); @@ -274,12 +276,10 @@ struct zone *tcp_rxt_seg_zone; 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 +#define TCP_BWMEAS_BURST_MINSIZE 6 +#define TCP_BWMEAS_BURST_MAXSIZE 25 static uint32_t bwmeas_elm_size; @@ -290,12 +290,12 @@ static uint32_t bwmeas_elm_size; * variable net.inet.tcp.tcbhashsize */ #ifndef TCBHASHSIZE -#define TCBHASHSIZE CONFIG_TCBHASHSIZE +#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"); + &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); /* * This is the actual shape of what we allocate using the zone @@ -317,9 +317,9 @@ 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 */ +static lck_attr_t *tcp_uptime_mtx_attr = NULL; +static lck_grp_t *tcp_uptime_mtx_grp = NULL; +static lck_grp_attr_t *tcp_uptime_mtx_grp_attr = NULL; int tcp_notsent_lowat_check(struct socket *so); static aes_encrypt_ctx tfo_ctx; /* Crypto-context for TFO */ @@ -352,7 +352,10 @@ tcp_sysctl_fastopenkey(__unused struct sysctl_oid *oidp, __unused void *arg1, __unused int arg2, struct sysctl_req *req) { int error = 0; - /* TFO-key is expressed as a string in hex format (+1 to account for \0 char) */ + /* + * TFO-key is expressed as a string in hex format + * (+1 to account for \0 char) + */ char keystring[TCP_FASTOPEN_KEYLEN * 2 + 1]; u_int32_t key[TCP_FASTOPEN_KEYLEN / sizeof(u_int32_t)]; int i; @@ -363,7 +366,8 @@ tcp_sysctl_fastopenkey(__unused struct sysctl_oid *oidp, __unused void *arg1, goto exit; } - /* sysctl_io_string copies keystring into the oldptr of the sysctl_req. + /* + * sysctl_io_string copies keystring into the oldptr of the sysctl_req. * Make sure everything is zero, to avoid putting garbage in there or * leaking the stack. */ @@ -374,7 +378,10 @@ tcp_sysctl_fastopenkey(__unused struct sysctl_oid *oidp, __unused void *arg1, goto exit; for (i = 0; i < (TCP_FASTOPEN_KEYLEN / sizeof(u_int32_t)); i++) { - /* We jump over the keystring in 8-character (4 byte in hex) steps */ + /* + * We jump over the keystring in 8-character (4 byte in hex) + * steps + */ if (sscanf(&keystring[i * 8], "%8x", &key[i]) != 1) { error = EINVAL; goto exit; @@ -387,14 +394,16 @@ exit: return (error); } -int get_inpcb_str_size(void) +int +get_inpcb_str_size(void) { - return sizeof(struct inpcb); + return (sizeof(struct inpcb)); } -int get_tcp_str_size(void) +int +get_tcp_str_size(void) { - return sizeof(struct tcpcb); + return (sizeof(struct tcpcb)); } int tcp_freeq(struct tcpcb *tp); @@ -403,26 +412,27 @@ static int scale_to_powerof2(int size); /* * This helper routine returns one of the following scaled value of size: - * 1. Rounded down power of two value of size if the size value passed as + * 1. Rounded down power of two value of size if the size value passed as * argument is not a power of two and the rounded up value overflows. * OR - * 2. Rounded up power of two value of size if the size value passed as - * argument is not a power of two and the rounded up value does not overflow + * 2. Rounded up power of two value of size if the size value passed as + * argument is not a power of two and the rounded up value does not overflow * OR * 3. Same value as argument size if it is already a power of two. - */ -static int scale_to_powerof2(int size) { + */ +static int +scale_to_powerof2(int size) { /* Handle special case of size = 0 */ int ret = size ? size : 1; if (!powerof2(ret)) { - while(!powerof2(size)) { - /* + while (!powerof2(size)) { + /* * Clear out least significant * set bit till size is left with * its highest set bit at which point * it is rounded down power of two. - */ + */ size = size & (size -1); } @@ -434,11 +444,11 @@ static int scale_to_powerof2(int size) { } } - return ret; + return (ret); } static void -tcp_tfo_init() +tcp_tfo_init(void) { u_char key[TCP_FASTOPEN_KEYLEN]; @@ -454,7 +464,7 @@ tcp_init(struct protosw *pp, struct domain *dp) { #pragma unused(dp) static int tcp_initialized = 0; - vm_size_t str_size; + vm_size_t str_size; struct inpcbinfo *pcbinfo; VERIFY((pp->pr_flags & (PR_INITIALIZED|PR_ATTACHED)) == PR_ATTACHED); @@ -475,7 +485,9 @@ tcp_init(struct protosw *pp, struct domain *dp) microuptime(&tcp_uptime); read_random(&tcp_now, sizeof(tcp_now)); - tcp_now = tcp_now & 0x3fffffff; /* Starts tcp internal clock at a random value */ + + /* Starts tcp internal clock at a random value */ + tcp_now = tcp_now & 0x3fffffff; tcp_tfo_init(); @@ -487,7 +499,8 @@ tcp_init(struct protosw *pp, struct domain *dp) * 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); + pcbinfo->ipi_lock_grp = lck_grp_alloc_init("tcppcb", + pcbinfo->ipi_lock_grp_attr); /* * allocate the lock attribute for tcp pcb mutexes @@ -518,7 +531,8 @@ tcp_init(struct protosw *pp, struct domain *dp) tcp_tcbhashsize); } - tcbinfo.ipi_hashbase = hashinit(tcp_tcbhashsize, M_PCB, &tcbinfo.ipi_hashmask); + 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)); @@ -531,7 +545,8 @@ tcp_init(struct protosw *pp, struct domain *dp) 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"); + 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); @@ -546,7 +561,8 @@ tcp_init(struct protosw *pp, struct domain *dp) 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"); + 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 */ @@ -566,9 +582,9 @@ tcp_init(struct protosw *pp, struct domain *dp) zone_change(tcp_rxt_seg_zone, Z_EXPAND, TRUE); #if INET6 -#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) +#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) #else /* INET6 */ -#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) +#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) #endif /* INET6 */ if (max_protohdr < TCP_MINPROTOHDR) { _max_protohdr = TCP_MINPROTOHDR; @@ -584,25 +600,32 @@ tcp_init(struct protosw *pp, struct domain *dp) 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 + * 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_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) { + 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"); }; - if ((tcp_timer_list.call = thread_call_allocate(tcp_run_timerlist, NULL)) == NULL) { + tcp_timer_list.call = thread_call_allocate(tcp_run_timerlist, NULL); + if (tcp_timer_list.call == NULL) { panic("failed to allocate call entry 1 in tcp_init\n"); } /* - * allocate lock group attribute, group and attribute for tcp_uptime_lock + * 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_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); + tcp_uptime_lock = lck_spin_alloc_init(tcp_uptime_mtx_grp, + tcp_uptime_mtx_attr); /* Initialize TCP LRO data structures */ tcp_lro_init(); @@ -617,6 +640,12 @@ tcp_init(struct protosw *pp, struct domain *dp) if (nmbclusters > 30720) { tcp_autorcvbuf_max = 1024 * 1024; tcp_autosndbuf_max = 1024 * 1024; + + /* + * Receive buffer max for cellular interfaces supporting + * Carrier Aggregation is higher + */ + tcp_autorcvbuf_max_ca = 2 * 1024 * 1024; } } @@ -626,10 +655,7 @@ tcp_init(struct protosw *pp, struct domain *dp) * of the tcpcb each time to conserve mbufs. */ void -tcp_fillheaders(tp, ip_ptr, tcp_ptr) - struct tcpcb *tp; - void *ip_ptr; - void *tcp_ptr; +tcp_fillheaders(struct tcpcb *tp, void *ip_ptr, void *tcp_ptr) { struct inpcb *inp = tp->t_inpcb; struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; @@ -643,8 +669,9 @@ tcp_fillheaders(tp, ip_ptr, tcp_ptr) (inp->inp_flow & IPV6_FLOWINFO_MASK); ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | (IPV6_VERSION & IPV6_VERSION_MASK); + ip6->ip6_plen = htons(sizeof(struct tcphdr)); ip6->ip6_nxt = IPPROTO_TCP; - ip6->ip6_plen = sizeof(struct tcphdr); + ip6->ip6_hlim = 0; ip6->ip6_src = inp->in6p_laddr; ip6->ip6_dst = inp->in6p_faddr; tcp_hdr->th_sum = in6_pseudo(&inp->in6p_laddr, &inp->in6p_faddr, @@ -652,20 +679,21 @@ tcp_fillheaders(tp, ip_ptr, tcp_ptr) } else #endif { - struct ip *ip = (struct ip *) ip_ptr; - - ip->ip_vhl = IP_VHL_BORING; - ip->ip_tos = 0; - ip->ip_len = 0; - ip->ip_id = 0; - ip->ip_off = 0; - ip->ip_ttl = 0; - ip->ip_sum = 0; - ip->ip_p = IPPROTO_TCP; - ip->ip_src = inp->inp_laddr; - ip->ip_dst = inp->inp_faddr; - tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, - htons(sizeof(struct tcphdr) + IPPROTO_TCP)); + struct ip *ip = (struct ip *) ip_ptr; + + ip->ip_vhl = IP_VHL_BORING; + ip->ip_tos = 0; + ip->ip_len = 0; + ip->ip_id = 0; + ip->ip_off = 0; + ip->ip_ttl = 0; + ip->ip_sum = 0; + ip->ip_p = IPPROTO_TCP; + ip->ip_src = inp->inp_laddr; + ip->ip_dst = inp->inp_faddr; + tcp_hdr->th_sum = + in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, + htons(sizeof(struct tcphdr) + IPPROTO_TCP)); } tcp_hdr->th_sport = inp->inp_lport; @@ -685,8 +713,7 @@ tcp_fillheaders(tp, ip_ptr, tcp_ptr) * use for this function is in keepalives, which use tcp_respond. */ struct tcptemp * -tcp_maketemplate(tp) - struct tcpcb *tp; +tcp_maketemplate(struct tcpcb *tp) { struct mbuf *m; struct tcptemp *n; @@ -732,6 +759,7 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, int isipv6; #endif /* INET6 */ struct ifnet *outif; + int sotc = SO_TC_UNSPEC; #if INET6 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6; @@ -755,12 +783,12 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, #if INET6 if (isipv6) { ro6 = &sro6; - bzero(ro6, sizeof *ro6); + bzero(ro6, sizeof(*ro6)); } else #endif /* INET6 */ { ro = &sro; - bzero(ro, sizeof *ro); + bzero(ro, sizeof(*ro)); } } if (m == 0) { @@ -773,8 +801,8 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, if (isipv6) { VERIFY((MHLEN - max_linkhdr) >= (sizeof (*ip6) + sizeof (*nth))); - bcopy((caddr_t)ip6, mtod(m, caddr_t), - sizeof(struct ip6_hdr)); + bcopy((caddr_t)ip6, mtod(m, caddr_t), + sizeof(struct ip6_hdr)); ip6 = mtod(m, struct ip6_hdr *); nth = (struct tcphdr *)(void *)(ip6 + 1); } else @@ -799,7 +827,7 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, m->m_data = (caddr_t)ipgen; /* m_len is set later */ tlen = 0; -#define xchg(a,b,type) { type t; t=a; a=b; b=t; } +#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 */ @@ -808,12 +836,12 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 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 *)(void *)(ip + 1); - } + { + /* 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 *)(void *)(ip + 1); + } if (th != nth) { /* * this is usually a case when an extension header @@ -833,11 +861,11 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); } else #endif - { - tlen += sizeof (struct tcpiphdr); - ip->ip_len = tlen; - ip->ip_ttl = ip_defttl; - } + { + tlen += sizeof (struct tcpiphdr); + ip->ip_len = tlen; + ip->ip_ttl = ip_defttl; + } m->m_len = tlen; m->m_pkthdr.len = tlen; m->m_pkthdr.rcvif = 0; @@ -875,9 +903,7 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 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 : - NULL); + ro6 && ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); } else #endif /* INET6 */ { @@ -908,8 +934,9 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, if (isipv6) { svc_flags |= PKT_SCF_IPV6; } + sotc = tp->t_inpcb->inp_socket->so_traffic_class; set_packet_service_class(m, tp->t_inpcb->inp_socket, - MBUF_SC_UNSPEC, svc_flags); + sotc, svc_flags); /* Embed flowhash and flow control flags */ m->m_pkthdr.pkt_flowsrc = FLOWSRC_INPCB; @@ -926,7 +953,8 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, #if INET6 if (isipv6) { struct ip6_out_args ip6oa = { tra->ifscope, { 0 }, - IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR, 0 }; + IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR, 0, + SO_TC_UNSPEC, _NET_SERVICE_TYPE_UNSPEC}; if (tra->ifscope != IFSCOPE_NONE) ip6oa.ip6oa_flags |= IP6OAF_BOUND_IF; @@ -936,7 +964,14 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, ip6oa.ip6oa_flags |= IP6OAF_NO_EXPENSIVE; if (tra->awdl_unrestricted) ip6oa.ip6oa_flags |= IP6OAF_AWDL_UNRESTRICTED; - + if (tra->intcoproc_allowed) + ip6oa.ip6oa_flags |= IP6OAF_INTCOPROC_ALLOWED; + ip6oa.ip6oa_sotc = sotc; + if (tp != NULL) { + if ((tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_ALLOWED)) + ip6oa.ip6oa_flags |= IP6OAF_QOSMARKING_ALLOWED; + ip6oa.ip6oa_netsvctype = tp->t_inpcb->inp_socket->so_netsvctype; + } (void) ip6_output(m, NULL, ro6, IPV6_OUTARGS, NULL, NULL, &ip6oa); @@ -951,7 +986,8 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, #endif /* INET6 */ { struct ip_out_args ipoa = { tra->ifscope, { 0 }, - IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR, 0 }; + IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR, 0, + SO_TC_UNSPEC, _NET_SERVICE_TYPE_UNSPEC }; if (tra->ifscope != IFSCOPE_NONE) ipoa.ipoa_flags |= IPOAF_BOUND_IF; @@ -961,7 +997,12 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, ipoa.ipoa_flags |= IPOAF_NO_EXPENSIVE; if (tra->awdl_unrestricted) ipoa.ipoa_flags |= IPOAF_AWDL_UNRESTRICTED; - + ipoa.ipoa_sotc = sotc; + if (tp != NULL) { + if ((tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_ALLOWED)) + ipoa.ipoa_flags |= IPOAF_QOSMARKING_ALLOWED; + ipoa.ipoa_netsvctype = tp->t_inpcb->inp_socket->so_netsvctype; + } if (ro != &sro) { /* Copy the cached route and take an extra reference */ inp_route_copyout(tp->t_inpcb, &sro); @@ -992,12 +1033,11 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, * come from the zone allocator set up in tcp_init(). */ struct tcpcb * -tcp_newtcpcb(inp) - struct inpcb *inp; +tcp_newtcpcb(struct inpcb *inp) { struct inp_tp *it; - register struct tcpcb *tp; - register struct socket *so = inp->inp_socket; + struct tcpcb *tp; + struct socket *so = inp->inp_socket; #if INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ @@ -1005,12 +1045,12 @@ tcp_newtcpcb(inp) calculate_tcp_clock(); if ((so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) == 0) { - it = (struct inp_tp *)(void *)inp; - tp = &it->tcb; + it = (struct inp_tp *)(void *)inp; + tp = &it->tcb; } else { - tp = (struct tcpcb *)(void *)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 = @@ -1026,14 +1066,16 @@ tcp_newtcpcb(inp) TAILQ_INIT(&tp->snd_holes); SLIST_INIT(&tp->t_rxt_segments); - tp->t_inpcb = inp; /* XXX */ + SLIST_INIT(&tp->t_notify_ack); + tp->t_inpcb = inp; /* * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives * reasonable initial retransmit time. */ tp->t_srtt = TCPTV_SRTTBASE; - tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; + tp->t_rttvar = + ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; tp->t_rttmin = tcp_TCPTV_MIN; tp->t_rxtcur = TCPTV_RTOBASE; @@ -1078,9 +1120,7 @@ tcp_newtcpcb(inp) * then send a RST to peer. */ struct tcpcb * -tcp_drop(tp, errno) - register struct tcpcb *tp; - int errno; +tcp_drop(struct tcpcb *tp, int errno) { struct socket *so = tp->t_inpcb->inp_socket; #if CONFIG_DTRACE @@ -1102,7 +1142,7 @@ tcp_drop(tp, errno) } void -tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt) +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); @@ -1115,17 +1155,19 @@ tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt) 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)); + 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)); + (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; + tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; } TCPT_RANGESET(tp->t_rxtcur, ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, @@ -1134,6 +1176,50 @@ tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt) } } +static inline void +tcp_update_ecn_perf_stats(struct tcpcb *tp, + struct if_tcp_ecn_perf_stat *stat) +{ + u_int64_t curval, oldval; + struct inpcb *inp = tp->t_inpcb; + stat->total_txpkts += inp->inp_stat->txpackets; + stat->total_rxpkts += inp->inp_stat->rxpackets; + stat->total_rxmitpkts += tp->t_stat.rxmitpkts; + stat->total_oopkts += tp->t_rcvoopack; + stat->total_reorderpkts += (tp->t_reordered_pkts + tp->t_pawsdrop + + tp->t_dsack_sent + tp->t_dsack_recvd); + + /* Average RTT */ + curval = (tp->t_srtt >> TCP_RTT_SHIFT); + if (curval > 0 && tp->t_rttupdated >= 16) { + if (stat->rtt_avg == 0) { + stat->rtt_avg = curval; + } else { + oldval = stat->rtt_avg; + stat->rtt_avg = + ((oldval << 4) - oldval + curval) >> 4; + } + } + + /* RTT variance */ + curval = tp->t_rttvar >> TCP_RTTVAR_SHIFT; + if (curval > 0 && tp->t_rttupdated >= 16) { + if (stat->rtt_var == 0) { + stat->rtt_var = curval; + } else { + oldval = stat->rtt_var; + stat->rtt_var = + ((oldval << 4) - oldval + curval) >> 4; + } + } + + /* Total number of SACK recovery episodes */ + stat->sack_episodes += tp->t_sack_recovery_episode; + + if (inp->inp_socket->so_error == ECONNRESET) + stat->rst_drop++; +} + /* * Close a TCP control block: * discard all space held by the tcp @@ -1141,8 +1227,7 @@ tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt) * wake up any sleepers */ struct tcpcb * -tcp_close(tp) - register struct tcpcb *tp; +tcp_close(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so = inp->inp_socket; @@ -1154,11 +1239,11 @@ tcp_close(tp) int dosavessthresh; /* tcp_close was called previously, bail */ - if (inp->inp_ppcb == NULL) - return(NULL); + if (inp->inp_ppcb == NULL) + return (NULL); tcp_canceltimers(tp); - KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_START, tp,0,0,0,0); + KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_START, tp, 0, 0, 0, 0); /* * If another thread for this tcp is currently in ip (indicated by @@ -1202,7 +1287,7 @@ tcp_close(tp) * update anything that the user "locked". */ if (tp->t_rttupdated >= 16) { - register u_int32_t i = 0; + u_int32_t i = 0; #if INET6 if (isipv6) { @@ -1272,8 +1357,8 @@ tcp_close(tp) else dosavessthresh = (i < so->so_snd.sb_hiwat / 2); if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && - i != 0 && rt->rt_rmx.rmx_ssthresh != 0) - || dosavessthresh) { + i != 0 && rt->rt_rmx.rmx_ssthresh != 0) || + dosavessthresh) { /* * convert the limit from user data bytes to * packets then to packet data bytes. @@ -1283,14 +1368,10 @@ tcp_close(tp) i = 2; i *= (u_int32_t)(tp->t_maxseg + #if INET6 - (isipv6 ? sizeof (struct ip6_hdr) + - sizeof (struct tcphdr) : -#endif - sizeof (struct tcpiphdr) -#if INET6 - ) -#endif - ); + isipv6 ? sizeof (struct ip6_hdr) + + sizeof (struct tcphdr) : +#endif /* INET6 */ + sizeof (struct tcpiphdr)); if (rt->rt_rmx.rmx_ssthresh) rt->rt_rmx.rmx_ssthresh = (rt->rt_rmx.rmx_ssthresh + i) / 2; @@ -1316,25 +1397,97 @@ no_valid_rt: /* free the reassembly queue, if any */ (void) tcp_freeq(tp); + + /* Collect ECN related statistics */ + if (tp->ecn_flags & TE_SETUPSENT) { + if (tp->ecn_flags & TE_CLIENT_SETUP) { + INP_INC_IFNET_STAT(inp, ecn_client_setup); + if (TCP_ECN_ENABLED(tp)) { + INP_INC_IFNET_STAT(inp, + ecn_client_success); + } else if (tp->ecn_flags & TE_LOST_SYN) { + INP_INC_IFNET_STAT(inp, ecn_syn_lost); + } else { + INP_INC_IFNET_STAT(inp, + ecn_peer_nosupport); + } + } else { + INP_INC_IFNET_STAT(inp, ecn_server_setup); + if (TCP_ECN_ENABLED(tp)) { + INP_INC_IFNET_STAT(inp, + ecn_server_success); + } else if (tp->ecn_flags & TE_LOST_SYNACK) { + INP_INC_IFNET_STAT(inp, + ecn_synack_lost); + } else { + INP_INC_IFNET_STAT(inp, + ecn_peer_nosupport); + } + } + } else { + INP_INC_IFNET_STAT(inp, ecn_off_conn); + } if (TCP_ECN_ENABLED(tp)) { - if (tp->ecn_flags & TE_RECV_ECN_CE) + if (tp->ecn_flags & TE_RECV_ECN_CE) { tcpstat.tcps_ecn_conn_recv_ce++; - if (tp->ecn_flags & TE_RECV_ECN_ECE) + INP_INC_IFNET_STAT(inp, ecn_conn_recv_ce); + } + if (tp->ecn_flags & TE_RECV_ECN_ECE) { tcpstat.tcps_ecn_conn_recv_ece++; + INP_INC_IFNET_STAT(inp, ecn_conn_recv_ece); + } if (tp->ecn_flags & (TE_RECV_ECN_CE | TE_RECV_ECN_ECE)) { if (tp->t_stat.txretransmitbytes > 0 || - tp->t_stat.rxoutoforderbytes > 0) + tp->t_stat.rxoutoforderbytes > 0) { tcpstat.tcps_ecn_conn_pl_ce++; - else + INP_INC_IFNET_STAT(inp, ecn_conn_plce); + } else { tcpstat.tcps_ecn_conn_nopl_ce++; + INP_INC_IFNET_STAT(inp, ecn_conn_noplce); + } } else { if (tp->t_stat.txretransmitbytes > 0 || - tp->t_stat.rxoutoforderbytes > 0) + tp->t_stat.rxoutoforderbytes > 0) { tcpstat.tcps_ecn_conn_plnoce++; + INP_INC_IFNET_STAT(inp, ecn_conn_plnoce); + } + } + } + + /* Aggregate performance stats */ + if (inp->inp_last_outifp != NULL && !(tp->t_flags & TF_LOCAL)) { + struct ifnet *ifp = inp->inp_last_outifp; + ifnet_lock_shared(ifp); + if ((ifp->if_refflags & (IFRF_ATTACHED | IFRF_DETACHING)) == + IFRF_ATTACHED) { + if (inp->inp_vflag & INP_IPV6) { + ifp->if_ipv6_stat->timestamp = net_uptime(); + if (TCP_ECN_ENABLED(tp)) { + tcp_update_ecn_perf_stats(tp, + &ifp->if_ipv6_stat->ecn_on); + } else { + tcp_update_ecn_perf_stats(tp, + &ifp->if_ipv6_stat->ecn_off); + } + } else { + ifp->if_ipv4_stat->timestamp = net_uptime(); + if (TCP_ECN_ENABLED(tp)) { + tcp_update_ecn_perf_stats(tp, + &ifp->if_ipv4_stat->ecn_on); + } else { + tcp_update_ecn_perf_stats(tp, + &ifp->if_ipv4_stat->ecn_off); + } + } } + ifnet_lock_done(ifp); } tcp_free_sackholes(tp); + tcp_notify_ack_free(tp); + + inp_decr_sndbytes_allunsent(so, tp->snd_una); + if (tp->t_bwmeas != NULL) { tcp_bwmeas_free(tp); } @@ -1359,13 +1512,14 @@ no_valid_rt: tp->t_state = TCPS_CLOSED; - /* Issue a wakeup before detach so that we don't miss + /* + * Issue a wakeup before detach so that we don't miss * a wakeup */ sodisconnectwakeup(so); - /* - * Clean up any LRO state + /* + * Clean up any LRO state */ if (tp->t_flagsext & TF_LRO_OFFLOADED) { tcp_lro_remove_state(inp->inp_laddr, inp->inp_faddr, @@ -1410,23 +1564,23 @@ no_valid_rt: #endif /* INET6 */ in_pcbdetach(inp); - /* Call soisdisconnected after detach because it might unlock the socket */ + /* + * 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(NULL); + return (NULL); } int -tcp_freeq(tp) - struct tcpcb *tp; +tcp_freeq(struct tcpcb *tp) { - - register struct tseg_qent *q; + struct tseg_qent *q; int rv = 0; - while((q = LIST_FIRST(&tp->t_segq)) != NULL) { + while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { LIST_REMOVE(q, tqe_q); m_freem(q->tqe_m); zfree(tcp_reass_zone, q); @@ -1444,12 +1598,12 @@ tcp_freeq(tp) * Do it next time if the pcbinfo lock is in use */ void -tcp_drain() +tcp_drain(void) { struct inpcb *inp; struct tcpcb *tp; - if (!lck_rw_try_lock_exclusive(tcbinfo.ipi_lock)) + if (!lck_rw_try_lock_exclusive(tcbinfo.ipi_lock)) return; LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) { @@ -1461,10 +1615,10 @@ tcp_drain() /* lost a race, try the next one */ tcp_unlock(inp->inp_socket, 1, 0); continue; - } + } tp = intotcpcb(inp); - if (do_tcpdrain) + if (do_tcpdrain) tcp_freeq(tp); so_drain_extended_bk_idle(inp->inp_socket); @@ -1485,13 +1639,11 @@ tcp_drain() * reporting soft errors (yet - a kqueue filter may be added). */ static void -tcp_notify(inp, error) - struct inpcb *inp; - int error; +tcp_notify(struct inpcb *inp, int error) { struct tcpcb *tp; - if (inp == NULL || (inp->inp_state == INPCB_STATE_DEAD)) + if (inp == NULL || (inp->inp_state == INPCB_STATE_DEAD)) return; /* pcb is gone already */ tp = (struct tcpcb *)inp->inp_ppcb; @@ -1504,8 +1656,8 @@ tcp_notify(inp, error) * can never complete. */ if (tp->t_state == TCPS_ESTABLISHED && - (error == EHOSTUNREACH || error == ENETUNREACH || - error == EHOSTDOWN)) { + (error == EHOSTUNREACH || error == ENETUNREACH || + error == EHOSTDOWN)) { return; } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && tp->t_softerror) @@ -1519,30 +1671,56 @@ tcp_notify(inp, error) #endif } -struct bwmeas* +struct bwmeas * tcp_bwmeas_alloc(struct tcpcb *tp) { struct bwmeas *elm; elm = zalloc(tcp_bwmeas_zone); if (elm == NULL) - return(elm); + 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); + return (elm); } void -tcp_bwmeas_free(struct tcpcb* tp) +tcp_bwmeas_free(struct tcpcb *tp) { zfree(tcp_bwmeas_zone, tp->t_bwmeas); tp->t_bwmeas = NULL; tp->t_flagsext &= ~(TF_MEASURESNDBW); } +int +get_tcp_inp_list(struct inpcb **inp_list, int n, inp_gen_t gencnt) +{ + struct tcpcb *tp; + struct inpcb *inp; + int i = 0; + + 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; + } + + 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; + } + return (i); +} + /* * 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. @@ -1556,7 +1734,8 @@ tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp) otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST]; otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP]; otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL]; - otp->t_inpcb = (_TCPCB_PTR(struct inpcb *))VM_KERNEL_ADDRPERM(tp->t_inpcb); + otp->t_inpcb = + (_TCPCB_PTR(struct inpcb *))VM_KERNEL_ADDRPERM(tp->t_inpcb); otp->t_state = tp->t_state; otp->t_flags = tp->t_flags; otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0; @@ -1598,8 +1777,8 @@ tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp) 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->cc_send = 0; + otp->cc_recv = 0; otp->snd_recover = tp->snd_recover; otp->snd_cwnd_prev = tp->snd_cwnd_prev; otp->snd_ssthresh_prev = tp->snd_ssthresh_prev; @@ -1611,8 +1790,7 @@ tcp_pcblist SYSCTL_HANDLER_ARGS { #pragma unused(oidp, arg1, arg2) int error, i = 0, n; - struct inpcb *inp, **inp_list; - struct tcpcb *tp; + struct inpcb **inp_list; inp_gen_t gencnt; struct xinpgen xig; @@ -1623,15 +1801,15 @@ tcp_pcblist SYSCTL_HANDLER_ARGS lck_rw_lock_shared(tcbinfo.ipi_lock); if (req->oldptr == USER_ADDR_NULL) { n = tcbinfo.ipi_count; - req->oldidx = 2 * (sizeof xig) + req->oldidx = 2 * (sizeof(xig)) + (n + n/8) * sizeof(struct xtcpcb); lck_rw_done(tcbinfo.ipi_lock); - return 0; + return (0); } if (req->newptr != USER_ADDR_NULL) { lck_rw_done(tcbinfo.ipi_lock); - return EPERM; + return (EPERM); } /* @@ -1641,70 +1819,68 @@ tcp_pcblist SYSCTL_HANDLER_ARGS n = tcbinfo.ipi_count; bzero(&xig, sizeof(xig)); - xig.xig_len = 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); + error = SYSCTL_OUT(req, &xig, sizeof(xig)); if (error) { lck_rw_done(tcbinfo.ipi_lock); - return error; + return (error); } /* * We are done if there is no pcb */ if (n == 0) { lck_rw_done(tcbinfo.ipi_lock); - return 0; + return (0); } - inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK); + inp_list = _MALLOC(n * sizeof (*inp_list), M_TEMP, M_WAITOK); if (inp_list == 0) { lck_rw_done(tcbinfo.ipi_lock); - return ENOMEM; - } - - 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; + return (ENOMEM); } - 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; + n = get_tcp_inp_list(inp_list, n, gencnt); error = 0; for (i = 0; i < n; i++) { + struct xtcpcb xt; + caddr_t inp_ppcb; + struct inpcb *inp; + inp = inp_list[i]; - if (inp->inp_gencnt <= gencnt && - inp->inp_state != INPCB_STATE_DEAD) { - struct xtcpcb xt; - caddr_t inp_ppcb; - - bzero(&xt, sizeof(xt)); - xt.xt_len = sizeof xt; - /* XXX should avoid extra copy */ - inpcb_to_compat(inp, &xt.xt_inp); - inp_ppcb = inp->inp_ppcb; - if (inp_ppcb != NULL) { - 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); + + if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) + continue; + tcp_lock(inp->inp_socket, 1, NULL); + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { + tcp_unlock(inp->inp_socket, 1, NULL); + continue; + } + if (inp->inp_gencnt > gencnt) { + tcp_unlock(inp->inp_socket, 1, NULL); + continue; + } + + bzero(&xt, sizeof(xt)); + xt.xt_len = sizeof(xt); + /* XXX should avoid extra copy */ + inpcb_to_compat(inp, &xt.xt_inp); + inp_ppcb = inp->inp_ppcb; + if (inp_ppcb != NULL) { + 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); + + tcp_unlock(inp->inp_socket, 1, NULL); + + error = SYSCTL_OUT(req, &xt, sizeof(xt)); } if (!error) { /* @@ -1715,15 +1891,15 @@ tcp_pcblist SYSCTL_HANDLER_ARGS * might be necessary to retry. */ bzero(&xig, sizeof(xig)); - xig.xig_len = 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); + error = SYSCTL_OUT(req, &xig, sizeof(xig)); } FREE(inp_list, M_TEMP); lck_rw_done(tcbinfo.ipi_lock); - return error; + return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, @@ -1734,59 +1910,59 @@ SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, static void tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp) { - otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first); - otp->t_dupacks = tp->t_dupacks; + otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first); + otp->t_dupacks = tp->t_dupacks; otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT]; otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST]; otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP]; otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL]; - otp->t_state = tp->t_state; - otp->t_flags = tp->t_flags; - otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0; - 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 = 0; + otp->t_state = tp->t_state; + otp->t_flags = tp->t_flags; + otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0; + 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 = 0; + otp->cc_recv = 0; + 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 = 0; } @@ -1794,95 +1970,97 @@ static int tcp_pcblist64 SYSCTL_HANDLER_ARGS { #pragma unused(oidp, arg1, arg2) - int error, i = 0, n; - struct inpcb *inp, **inp_list; - struct tcpcb *tp; - inp_gen_t gencnt; - struct xinpgen xig; - - /* - * 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.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.ipi_lock); - return 0; - } - - if (req->newptr != USER_ADDR_NULL) { - lck_rw_done(tcbinfo.ipi_lock); - 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.ipi_lock); - return error; - } - /* - * We are done if there is no pcb - */ - if (n == 0) { - 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.ipi_lock); - return ENOMEM; - } + int error, i = 0, n; + struct inpcb **inp_list; + inp_gen_t gencnt; + struct xinpgen xig; - 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; - } + /* + * 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.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.ipi_lock); + return (0); + } - 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; - - 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 = (uint64_t)VM_KERNEL_ADDRPERM(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) { + if (req->newptr != USER_ADDR_NULL) { + lck_rw_done(tcbinfo.ipi_lock); + 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.ipi_lock); + return (error); + } + /* + * We are done if there is no pcb + */ + if (n == 0) { + 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.ipi_lock); + return (ENOMEM); + } + + n = get_tcp_inp_list(inp_list, n, gencnt); + + error = 0; + for (i = 0; i < n; i++) { + struct xtcpcb64 xt; + struct inpcb *inp; + + inp = inp_list[i]; + + if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) + continue; + tcp_lock(inp->inp_socket, 1, NULL); + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { + tcp_unlock(inp->inp_socket, 1, NULL); + continue; + } + if (inp->inp_gencnt > gencnt) { + tcp_unlock(inp->inp_socket, 1, NULL); + continue; + } + + bzero(&xt, sizeof(xt)); + xt.xt_len = sizeof(xt); + inpcb_to_xinpcb64(inp, &xt.xt_inpcb); + xt.xt_inpcb.inp_ppcb = + (uint64_t)VM_KERNEL_ADDRPERM(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); + + tcp_unlock(inp->inp_socket, 1, NULL); + + 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 @@ -1891,20 +2069,20 @@ tcp_pcblist64 SYSCTL_HANDLER_ARGS * might be necessary to retry. */ bzero(&xig, sizeof(xig)); - xig.xig_len = 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.ipi_lock); - return error; + error = SYSCTL_OUT(req, &xig, sizeof(xig)); + } + FREE(inp_list, M_TEMP); + lck_rw_done(tcbinfo.ipi_lock); + return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist64, CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, - tcp_pcblist64, "S,xtcpcb64", "List of active TCP connections"); + tcp_pcblist64, "S,xtcpcb64", "List of active TCP connections"); static int @@ -1912,43 +2090,136 @@ tcp_pcblist_n SYSCTL_HANDLER_ARGS { #pragma unused(oidp, arg1, arg2) int error = 0; - + error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo); - - return error; + + return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist_n, CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, - tcp_pcblist_n, "S,xtcpcb_n", "List of active TCP connections"); + 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 flags, bitstr_t *bitfield) { - inpcb_get_ports_used(ifindex, protocol, flags, + inpcb_get_ports_used(ifindex, protocol, flags, bitfield, &tcbinfo); } __private_extern__ uint32_t tcp_count_opportunistic(unsigned int ifindex, u_int32_t flags) { - return inpcb_count_opportunistic(ifindex, &tcbinfo, 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); + return (inpcb_find_anypcb_byaddr(ifa, &tcbinfo)); +} + +static void +tcp_handle_msgsize(struct ip *ip, struct inpcb *inp) +{ + struct rtentry *rt = NULL; + u_short ifscope = IFSCOPE_NONE; + int mtu; + struct sockaddr_in icmpsrc = { + sizeof (struct sockaddr_in), + AF_INET, 0, { 0 }, + { 0, 0, 0, 0, 0, 0, 0, 0 } }; + struct icmp *icp = NULL; + + icp = (struct icmp *)(void *) + ((caddr_t)ip - offsetof(struct icmp, icmp_ip)); + + icmpsrc.sin_addr = icp->icmp_ip.ip_dst; + + /* + * 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. + */ + if (ROUTE_UNUSABLE(&(inp->inp_route)) == false) + rt = inp->inp_route.ro_rt; + + /* + * icmp6_mtudisc_update scopes the routing lookup + * to the incoming interface (delivered from mbuf + * packet header. + * That is mostly ok but for asymmetric networks + * that may be an issue. + * Frag needed OR Packet too big really communicates + * MTU for the out data path. + * Take the interface scope from cached route or + * the last outgoing interface from inp + */ + if (rt != NULL) + ifscope = (rt->rt_ifp != NULL) ? + rt->rt_ifp->if_index : IFSCOPE_NONE; + else + ifscope = (inp->inp_last_outifp != NULL) ? + inp->inp_last_outifp->if_index : IFSCOPE_NONE; + + if ((rt == NULL) || + !(rt->rt_flags & RTF_HOST) || + (rt->rt_flags & (RTF_CLONING | RTF_PRCLONING))) { + rt = rtalloc1_scoped((struct sockaddr *)&icmpsrc, 0, + RTF_CLONING | RTF_PRCLONING, ifscope); + } else if (rt) { + RT_LOCK(rt); + rtref(rt); + RT_UNLOCK(rt); + } + + if (rt != NULL) { + RT_LOCK(rt); + if ((rt->rt_flags & RTF_HOST) && + !(rt->rt_rmx.rmx_locks & RTV_MTU)) { + mtu = ntohs(icp->icmp_nextmtu); + /* + * XXX Stock BSD has changed the following + * to compare with icp->icmp_ip.ip_len + * to converge faster when sent packet + * < route's MTU. We may want to adopt + * that change. + */ + if (mtu == 0) + 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_locks |= RTV_MTU; + } else if (rt->rt_rmx.rmx_mtu > mtu) { + rt->rt_rmx.rmx_mtu = mtu; + } + } + RT_UNLOCK(rt); + rtfree(rt); + } } void -tcp_ctlinput(cmd, sa, vip) - int cmd; - struct sockaddr *sa; - void *vip; +tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) { tcp_seq icmp_tcp_seq; struct ip *ip = vip; @@ -1976,7 +2247,7 @@ tcp_ctlinput(cmd, sa, vip) } else if (cmd == PRC_HOSTDEAD) ip = 0; /* Source quench is deprecated */ - else if (cmd == PRC_QUENCH) + else if (cmd == PRC_QUENCH) return; else if (inetctlerrmap[cmd] == 0) return; @@ -1986,13 +2257,17 @@ tcp_ctlinput(cmd, sa, vip) 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)); + /* + * Only the first 8 bytes of TCP header will be returned. + */ + bzero(&th, sizeof(th)); + bcopy(((caddr_t)ip + (IP_VHL_HL(ip->ip_vhl) << 2)), &th, 8); 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) { + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == + WNT_STOPUSING) { tcp_unlock(inp->inp_socket, 1, 0); return; } @@ -2000,57 +2275,8 @@ tcp_ctlinput(cmd, sa, vip) tp = intotcpcb(inp); 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); - } - } + if (cmd == PRC_MSGSIZE) + tcp_handle_msgsize(ip, inp); (*notify)(inp, inetctlerrmap[cmd]); } @@ -2062,10 +2288,7 @@ tcp_ctlinput(cmd, sa, vip) #if INET6 void -tcp6_ctlinput(cmd, sa, d) - int cmd; - struct sockaddr *sa; - void *d; +tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) { struct tcphdr th; void (*notify)(struct inpcb *, int) = tcp_notify; @@ -2091,7 +2314,7 @@ tcp6_ctlinput(cmd, sa, d) else if (!PRC_IS_REDIRECT(cmd) && (inet6ctlerrmap[cmd] == 0)) return; /* Source quench is deprecated */ - else if (cmd == PRC_QUENCH) + else if (cmd == PRC_QUENCH) return; /* if the parameter is from icmp6, decode it. */ @@ -2139,7 +2362,7 @@ tcp6_ctlinput(cmd, sa, d) * 1. In SYN-ACK packets. * 2. In SYN packets. * - * The ISNs in SYN-ACK packets have no monotonicity requirement, + * The ISNs in SYN-ACK packets have no monotonicity requirement, * and should be as unpredictable as possible to avoid the possibility * of spoofing and/or connection hijacking. To satisfy this * requirement, SYN-ACK ISNs are generated via the arc4random() @@ -2175,11 +2398,10 @@ tcp6_ctlinput(cmd, sa, d) * */ -#define ISN_BYTES_PER_SECOND 1048576 +#define ISN_BYTES_PER_SECOND 1048576 tcp_seq -tcp_new_isn(tp) - struct tcpcb *tp; +tcp_new_isn(struct tcpcb *tp) { u_int32_t md5_buffer[4]; tcp_seq new_isn; @@ -2189,19 +2411,19 @@ tcp_new_isn(tp) MD5_CTX isn_ctx; /* Use arc4random for SYN-ACKs when not in exact RFC1948 mode. */ - if (((tp->t_state == TCPS_LISTEN) || (tp->t_state == TCPS_TIME_WAIT)) - && tcp_strict_rfc1948 == 0) + if (((tp->t_state == TCPS_LISTEN) || (tp->t_state == TCPS_TIME_WAIT)) && + tcp_strict_rfc1948 == 0) #ifdef __APPLE__ - return RandomULong(); + return (RandomULong()); #else - return arc4random(); + return (arc4random()); #endif getmicrotime(&timenow); /* Seed if this is the first use, reseed if requested. */ if ((isn_last_reseed == 0) || ((tcp_strict_rfc1948 == 0) && (tcp_isn_reseed_interval > 0) && - (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) + (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) < (u_int)timenow.tv_sec))) { #ifdef __APPLE__ read_random(&isn_secret, sizeof(isn_secret)); @@ -2210,30 +2432,32 @@ tcp_new_isn(tp) #endif isn_last_reseed = timenow.tv_sec; } - + /* Compute the md5 hash and return the ISN. */ MD5Init(&isn_ctx); - MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); - MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); + MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, + sizeof(u_short)); + MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, + sizeof(u_short)); #if INET6 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, - sizeof(struct in6_addr)); + sizeof(struct in6_addr)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, - sizeof(struct in6_addr)); + sizeof(struct in6_addr)); } else #endif { MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, - sizeof(struct in_addr)); + sizeof(struct in_addr)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, - sizeof(struct in_addr)); + sizeof(struct in_addr)); } MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); MD5Final((u_char *) &md5_buffer, &isn_ctx); new_isn = (tcp_seq) md5_buffer[0]; new_isn += timenow.tv_sec * (ISN_BYTES_PER_SECOND / hz); - return new_isn; + return (new_isn); } @@ -2243,9 +2467,7 @@ tcp_new_isn(tp) * is controlled by the icmp_may_rst sysctl. */ void -tcp_drop_syn_sent(inp, errno) - struct inpcb *inp; - int errno; +tcp_drop_syn_sent(struct inpcb *inp, int errno) { struct tcpcb *tp = intotcpcb(inp); @@ -2272,8 +2494,13 @@ tcp_mtudisc( int offered; int mss; u_int32_t mtu; + u_int32_t protoHdrOverhead = sizeof (struct tcpiphdr); #if INET6 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; + + if (isipv6) + protoHdrOverhead = sizeof(struct ip6_hdr) + + sizeof(struct tcphdr); #endif /* INET6 */ if (tp) { @@ -2282,7 +2509,7 @@ tcp_mtudisc( rt = tcp_rtlookup6(inp, IFSCOPE_NONE); else #endif /* INET6 */ - rt = tcp_rtlookup(inp, IFSCOPE_NONE); + rt = tcp_rtlookup(inp, IFSCOPE_NONE); if (!rt || !rt->rt_rmx.rmx_mtu) { tp->t_maxopd = tp->t_maxseg = #if INET6 @@ -2306,17 +2533,7 @@ tcp_mtudisc( // Adjust MTU if necessary. mtu = necp_socket_get_effective_mtu(inp, mtu); #endif /* NECP */ - - mss = mtu - -#if INET6 - (isipv6 ? - sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : -#endif /* INET6 */ - sizeof(struct tcpiphdr) -#if INET6 - ) -#endif /* INET6 */ - ; + mss = mtu - protoHdrOverhead; if (offered) mss = min(mss, offered); @@ -2354,7 +2571,8 @@ tcp_mtudisc( tp->t_maxseg = mss; /* - * Reset the slow-start flight size as it may depends on the new 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); @@ -2373,9 +2591,7 @@ tcp_mtudisc( * hold the rtentry lock; the caller is responsible for unlocking. */ struct rtentry * -tcp_rtlookup(inp, input_ifscope) - struct inpcb *inp; - unsigned int input_ifscope; +tcp_rtlookup(struct inpcb *inp, unsigned int input_ifscope) { struct route *ro; struct rtentry *rt; @@ -2428,29 +2644,23 @@ tcp_rtlookup(inp, input_ifscope) * disabled) */ - tp = intotcpcb(inp); + tp = intotcpcb(inp); - if (!path_mtu_discovery || ((rt != NULL) && - (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) + if (!path_mtu_discovery || ((rt != NULL) && + (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) tp->t_flags &= ~TF_PMTUD; else tp->t_flags |= TF_PMTUD; -#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); soif2kcl(inp->inp_socket, (rt->rt_ifp->if_eflags & IFEF_2KCL)); + tcp_set_ecn(tp, rt->rt_ifp); + if (inp->inp_last_outifp == NULL) + inp->inp_last_outifp = rt->rt_ifp; } /* Note if the peer is local */ @@ -2460,18 +2670,16 @@ tcp_rtlookup(inp, input_ifscope) in_localaddr(inp->inp_faddr))) { tp->t_flags |= TF_LOCAL; } - + /* * Caller needs to call RT_UNLOCK(rt). */ - return rt; + return (rt); } #if INET6 struct rtentry * -tcp_rtlookup6(inp, input_ifscope) - struct inpcb *inp; - unsigned int input_ifscope; +tcp_rtlookup6(struct inpcb *inp, unsigned int input_ifscope) { struct route_in6 *ro6; struct rtentry *rt; @@ -2526,7 +2734,7 @@ tcp_rtlookup6(inp, input_ifscope) */ - tp = intotcpcb(inp); + tp = intotcpcb(inp); /* * Update MTU discovery determination. Don't do it if: @@ -2536,27 +2744,21 @@ tcp_rtlookup6(inp, input_ifscope) * disabled) */ - if (!path_mtu_discovery || ((rt != NULL) && - (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) + if (!path_mtu_discovery || ((rt != NULL) && + (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) tp->t_flags &= ~TF_PMTUD; else tp->t_flags |= TF_PMTUD; -#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); soif2kcl(inp->inp_socket, (rt->rt_ifp->if_eflags & IFEF_2KCL)); + tcp_set_ecn(tp, rt->rt_ifp); + if (inp->inp_last_outifp == NULL) + inp->inp_last_outifp = rt->rt_ifp; } /* Note if the peer is local */ @@ -2571,15 +2773,14 @@ tcp_rtlookup6(inp, input_ifscope) /* * Caller needs to call RT_UNLOCK(rt). */ - return rt; + return (rt); } #endif /* INET6 */ #if IPSEC /* compute ESP/AH header size for TCP, including outer IP header. */ size_t -ipsec_hdrsiz_tcp(tp) - struct tcpcb *tp; +ipsec_hdrsiz_tcp(struct tcpcb *tp) { struct inpcb *inp; struct mbuf *m; @@ -2591,10 +2792,10 @@ ipsec_hdrsiz_tcp(tp) struct tcphdr *th; if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) - return 0; + return (0); MGETHDR(m, M_DONTWAIT, MT_DATA); /* MAC-OK */ if (!m) - return 0; + return (0); #if INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { @@ -2606,17 +2807,17 @@ ipsec_hdrsiz_tcp(tp) hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); } else #endif /* INET6 */ - { - ip = mtod(m, struct ip *); - th = (struct tcphdr *)(ip + 1); - m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); - tcp_fillheaders(tp, ip, th); - hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); - } + { + ip = mtod(m, struct ip *); + th = (struct tcphdr *)(ip + 1); + m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); + tcp_fillheaders(tp, ip, th); + hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); + } m_free(m); - return hdrsiz; + return (hdrsiz); } -#endif /*IPSEC*/ +#endif /* IPSEC */ /* * Return a pointer to the cached information about the remote host. @@ -2624,8 +2825,7 @@ ipsec_hdrsiz_tcp(tp) * the route metrics. */ struct rmxp_tao * -tcp_gettaocache(inp) - struct inpcb *inp; +tcp_gettaocache(struct inpcb *inp) { struct rtentry *rt; struct rmxp_tao *taop; @@ -2643,9 +2843,9 @@ tcp_gettaocache(inp) /* Route locked during lookup above */ if (rt != NULL) RT_UNLOCK(rt); - return NULL; + return (NULL); } - + taop = rmx_taop(rt->rt_rmx); /* Route locked during lookup above */ RT_UNLOCK(rt); @@ -2661,7 +2861,7 @@ tcp_gettaocache(inp) * nothing in the cache left over. */ static void -tcp_cleartaocache() +tcp_cleartaocache(void) { } @@ -2678,14 +2878,15 @@ tcp_lock(struct socket *so, int refcount, void *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", + 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=%p ref=%x lrh= %s\n", - so, so->so_pcb, lr_saved, so->so_usecount, solockhistory_nr(so)); + so, so->so_pcb, lr_saved, so->so_usecount, + solockhistory_nr(so)); /* NOTREACHED */ } if (refcount) @@ -2716,12 +2917,12 @@ tcp_unlock(struct socket *so, int refcount, void *lr) so->so_usecount--; if (so->so_usecount < 0) { - panic("tcp_unlock: so=%p usecount=%x lrh= %s\n", + 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", + 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 { @@ -2735,28 +2936,25 @@ tcp_unlock(struct socket *so, int refcount, void *lr) } lck_mtx_t * -tcp_getlock( - struct socket *so, - __unused int locktype) +tcp_getlock(struct socket *so, __unused int locktype) { struct inpcb *inp = sotoinpcb(so); if (so->so_pcb) { 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); - } - else { - panic("tcp_getlock: so=%p NULL so_pcb %s\n", + 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 %s\n", so, solockhistory_nr(so)); return (so->so_proto->pr_domain->dom_mtx); } } -/* +/* * 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 + * 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. @@ -2768,6 +2966,8 @@ tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb) u_int32_t rcvbuf = sb->sb_hiwat; struct socket *so = tp->t_inpcb->inp_socket; + if (tcp_recv_bg == 1 || IS_TCP_RECV_BG(so)) + return; /* * If message delivery is enabled, do not count * unordered bytes in receive buffer towards hiwat @@ -2791,12 +2991,17 @@ tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb) int32_t tcp_sbspace(struct tcpcb *tp) { - struct sockbuf *sb = &tp->t_inpcb->inp_socket->so_rcv; - u_int32_t rcvbuf = sb->sb_hiwat; - int32_t space; struct socket *so = tp->t_inpcb->inp_socket; + struct sockbuf *sb = &so->so_rcv; + u_int32_t rcvbuf; + int32_t space; int32_t pending = 0; + tcp_sbrcv_grow_rwin(tp, sb); + + /* hiwat might have changed */ + rcvbuf = sb->sb_hiwat; + /* * If message delivery is enabled, do not count * unordered bytes in receive buffer towards hiwat mark. @@ -2807,11 +3012,9 @@ tcp_sbspace(struct tcpcb *tp) 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((rcvbuf - sb->sb_cc), (sb->sb_mbmax - sb->sb_mbcnt))); - if (space < 0) + if (space < 0) space = 0; #if CONTENT_FILTER @@ -2823,20 +3026,21 @@ tcp_sbspace(struct tcpcb *tp) else space -= pending; - /* Avoid increasing 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) */ - if (space < tp->t_maxseg) - return space; + if (space < tp->t_maxseg) + return (space); - /* Clip window size for slower link */ + /* Clip window size for slower link */ - if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0 ) - return imin(space, slowlink_wsize); + if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0) + return (imin(space, slowlink_wsize)); - return space; + return (space); } /* * Checks TCP Segment Offloading capability for a given connection @@ -2865,20 +3069,20 @@ tcp_set_tso(struct tcpcb *tp, struct ifnet *ifp) if (isipv6) { if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV6)) { tp->t_flags |= TF_TSO; - if (ifp->if_tso_v6_mtu != 0) + 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; - } else + } else #endif /* INET6 */ { if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV4)) { tp->t_flags |= TF_TSO; - if (ifp->if_tso_v4_mtu != 0) + 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; @@ -2887,18 +3091,20 @@ tcp_set_tso(struct tcpcb *tp, struct ifnet *ifp) } } -#define TIMEVAL_TO_TCPHZ(_tv_) ((_tv_).tv_sec * TCP_RETRANSHZ + (_tv_).tv_usec / TCP_RETRANSHZ_TO_USEC) +#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 +/* + * 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 + * 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() +calculate_tcp_clock(void) { struct timeval tv = tcp_uptime; struct timeval interval = {0, TCP_RETRANSHZ_TO_USEC}; @@ -2935,18 +3141,20 @@ calculate_tcp_clock() tcp_now += incr; } - lck_spin_unlock(tcp_uptime_lock); - } - return; + lck_spin_unlock(tcp_uptime_lock); + } } -/* 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 +/* + * 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) { +tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so, + u_int32_t rcvbuf_max) +{ u_int32_t maxsockbufsize; if (!tcp_do_rfc1323) { tp->request_r_scale = 0; @@ -2955,7 +3163,7 @@ tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so) { 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; + so->so_rcv.sb_hiwat : rcvbuf_max; while (tp->request_r_scale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << tp->request_r_scale) < maxsockbufsize) @@ -2973,31 +3181,34 @@ tcp_notsent_lowat_check(struct socket *so) { tp = intotcpcb(inp); } - notsent = so->so_snd.sb_cc - + 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 + /* + * 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); + return (1); } - /* When Nagle's algorithm is not disabled, it is better + /* + * 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 && + if ((tp->t_flags & TF_NODELAY) == 0 && notsent > 0 && notsent < tp->t_maxseg) { - return(1); + return (1); } - return(0); + return (0); } void -tcp_rxtseg_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end) { +tcp_rxtseg_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end) +{ struct tcp_rxt_seg *rxseg = NULL, *prev = NULL, *next = NULL; u_int32_t rxcount = 0; @@ -3065,7 +3276,6 @@ tcp_rxtseg_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end) { } else { SLIST_INSERT_HEAD(&tp->t_rxt_segments, rxseg, rx_link); } - return; } struct tcp_rxt_seg * @@ -3136,7 +3346,7 @@ tcp_rxtseg_dsack_for_tlp(struct tcpcb *tp) SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) { if (rxseg->rx_count == 1 && - SLIST_NEXT(rxseg,rx_link) == NULL && + SLIST_NEXT(rxseg, rx_link) == NULL && (rxseg->rx_flags & TCP_RXT_DSACK_FOR_TLP)) { dsack_for_tlp = TRUE; break; @@ -3146,7 +3356,8 @@ tcp_rxtseg_dsack_for_tlp(struct tcpcb *tp) } u_int32_t -tcp_rxtseg_total_size(struct tcpcb *tp) { +tcp_rxtseg_total_size(struct tcpcb *tp) +{ struct tcp_rxt_seg *rxseg; u_int32_t total_size = 0; @@ -3172,16 +3383,15 @@ tcp_get_connectivity_status(struct tcpcb *tp, } if (tp->t_rtimo_probes >= TCP_CONNECTIVITY_PROBES_MAX) connstatus->read_probe_failed = 1; - if (tp->t_inpcb != NULL && tp->t_inpcb->inp_last_outifp != NULL - && (tp->t_inpcb->inp_last_outifp->if_eflags & IFEF_PROBE_CONNECTIVITY)) + if (tp->t_inpcb != NULL && tp->t_inpcb->inp_last_outifp != NULL && + (tp->t_inpcb->inp_last_outifp->if_eflags & IFEF_PROBE_CONNECTIVITY)) connstatus->probe_activated = 1; - return; } boolean_t tfo_enabled(const struct tcpcb *tp) { - return !!(tp->t_flagsext & TF_FASTOPEN); + return ((tp->t_flagsext & TF_FASTOPEN)? TRUE : FALSE); } void @@ -3190,3 +3400,423 @@ tcp_disable_tfo(struct tcpcb *tp) tp->t_flagsext &= ~TF_FASTOPEN; } +static struct mbuf * +tcp_make_keepalive_frame(struct tcpcb *tp, struct ifnet *ifp, + boolean_t is_probe) +{ + struct inpcb *inp = tp->t_inpcb; + struct tcphdr *th; + u_int8_t *data; + int win = 0; + struct mbuf *m; + + /* + * The code assumes the IP + TCP headers fit in an mbuf packet header + */ + _CASSERT(sizeof(struct ip) + sizeof(struct tcphdr) <= _MHLEN); + _CASSERT(sizeof(struct ip6_hdr) + sizeof(struct tcphdr) <= _MHLEN); + + MGETHDR(m, M_WAIT, MT_HEADER); + if (m == NULL) { + return (NULL); + } + m->m_pkthdr.pkt_proto = IPPROTO_TCP; + + data = mbuf_datastart(m); + + if (inp->inp_vflag & INP_IPV4) { + bzero(data, sizeof(struct ip) + sizeof(struct tcphdr)); + th = (struct tcphdr *)(void *) (data + sizeof(struct ip)); + m->m_len = sizeof(struct ip) + sizeof(struct tcphdr); + m->m_pkthdr.len = m->m_len; + } else { + VERIFY(inp->inp_vflag & INP_IPV6); + + bzero(data, sizeof(struct ip6_hdr) + + sizeof(struct tcphdr)); + th = (struct tcphdr *)(void *)(data + sizeof(struct ip6_hdr)); + m->m_len = sizeof(struct ip6_hdr) + + sizeof(struct tcphdr); + m->m_pkthdr.len = m->m_len; + } + + tcp_fillheaders(tp, data, th); + + if (inp->inp_vflag & INP_IPV4) { + struct ip *ip; + + ip = (__typeof__(ip))(void *)data; + + ip->ip_id = ip_randomid(); + ip->ip_len = htons(sizeof(struct ip) + sizeof(struct tcphdr)); + ip->ip_ttl = inp->inp_ip_ttl; + ip->ip_tos |= (inp->inp_ip_tos & ~IPTOS_ECN_MASK); + ip->ip_sum = in_cksum_hdr(ip); + } else { + struct ip6_hdr *ip6; + + ip6 = (__typeof__(ip6))(void *)data; + + ip6->ip6_plen = htons(sizeof(struct tcphdr)); + ip6->ip6_hlim = in6_selecthlim(inp, ifp); + ip6->ip6_flow = ip6->ip6_flow & ~IPV6_FLOW_ECN_MASK; + + if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) + ip6->ip6_src.s6_addr16[1] = 0; + if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) + ip6->ip6_dst.s6_addr16[1] = 0; + } + th->th_flags = TH_ACK; + + win = tcp_sbspace(tp); + if (win > ((int32_t)TCP_MAXWIN << tp->rcv_scale)) + win = (int32_t)TCP_MAXWIN << tp->rcv_scale; + th->th_win = htons((u_short) (win >> tp->rcv_scale)); + + if (is_probe) { + th->th_seq = htonl(tp->snd_una - 1); + } else { + th->th_seq = htonl(tp->snd_una); + } + th->th_ack = htonl(tp->rcv_nxt); + + /* Force recompute TCP checksum to be the final value */ + th->th_sum = 0; + if (inp->inp_vflag & INP_IPV4) { + th->th_sum = inet_cksum(m, IPPROTO_TCP, + sizeof(struct ip), sizeof(struct tcphdr)); + } else { + th->th_sum = inet6_cksum(m, IPPROTO_TCP, + sizeof(struct ip6_hdr), sizeof(struct tcphdr)); + } + + return (m); +} + +void +tcp_fill_keepalive_offload_frames(ifnet_t ifp, + struct ifnet_keepalive_offload_frame *frames_array, + u_int32_t frames_array_count, size_t frame_data_offset, + u_int32_t *used_frames_count) +{ + struct inpcb *inp; + inp_gen_t gencnt; + u_int32_t frame_index = *used_frames_count; + + if (ifp == NULL || frames_array == NULL || + frames_array_count == 0 || + frame_index >= frames_array_count || + frame_data_offset >= IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) + return; + + /* + * This function is called outside the regular TCP processing + * so we need to update the TCP clock. + */ + calculate_tcp_clock(); + + lck_rw_lock_shared(tcbinfo.ipi_lock); + gencnt = tcbinfo.ipi_gencnt; + LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) { + struct socket *so; + struct ifnet_keepalive_offload_frame *frame; + struct mbuf *m = NULL; + struct tcpcb *tp = intotcpcb(inp); + + if (frame_index >= frames_array_count) + break; + + if (inp->inp_gencnt > gencnt || + inp->inp_state == INPCB_STATE_DEAD) + continue; + + if ((so = inp->inp_socket) == NULL || + (so->so_state & SS_DEFUNCT)) + continue; + /* + * check for keepalive offload flag without socket + * lock to avoid a deadlock + */ + if (!(inp->inp_flags2 & INP2_KEEPALIVE_OFFLOAD)) { + continue; + } + + if (!(inp->inp_vflag & (INP_IPV4 | INP_IPV6))) { + continue; + } + if (inp->inp_ppcb == NULL || + in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) + continue; + tcp_lock(so, 1, 0); + /* Release the want count */ + if (inp->inp_ppcb == NULL || + (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING)) { + tcp_unlock(so, 1, 0); + continue; + } + if ((inp->inp_vflag & INP_IPV4) && + (inp->inp_laddr.s_addr == INADDR_ANY || + inp->inp_faddr.s_addr == INADDR_ANY)) { + tcp_unlock(so, 1, 0); + continue; + } + if ((inp->inp_vflag & INP_IPV6) && + (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) || + IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))) { + tcp_unlock(so, 1, 0); + continue; + } + if (inp->inp_lport == 0 || inp->inp_fport == 0) { + tcp_unlock(so, 1, 0); + continue; + } + if (inp->inp_last_outifp == NULL || + inp->inp_last_outifp->if_index != ifp->if_index) { + tcp_unlock(so, 1, 0); + continue; + } + if ((inp->inp_vflag & INP_IPV4) && frame_data_offset + + sizeof(struct ip) + sizeof(struct tcphdr) > + IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) { + tcp_unlock(so, 1, 0); + continue; + } else if (!(inp->inp_vflag & INP_IPV4) && frame_data_offset + + sizeof(struct ip6_hdr) + sizeof(struct tcphdr) > + IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) { + tcp_unlock(so, 1, 0); + continue; + } + /* + * There is no point in waking up the device for connections + * that are not established. Long lived connection are meant + * for processes that will sent and receive data + */ + if (tp->t_state != TCPS_ESTABLISHED) { + tcp_unlock(so, 1, 0); + continue; + } + /* + * This inp has all the information that is needed to + * generate an offload frame. + */ + frame = &frames_array[frame_index]; + frame->type = IFNET_KEEPALIVE_OFFLOAD_FRAME_TCP; + frame->ether_type = (inp->inp_vflag & INP_IPV4) ? + IFNET_KEEPALIVE_OFFLOAD_FRAME_ETHERTYPE_IPV4 : + IFNET_KEEPALIVE_OFFLOAD_FRAME_ETHERTYPE_IPV6; + frame->interval = tp->t_keepidle > 0 ? tp->t_keepidle : + tcp_keepidle; + frame->keep_cnt = TCP_CONN_KEEPCNT(tp); + frame->keep_retry = TCP_CONN_KEEPINTVL(tp); + frame->local_port = ntohs(inp->inp_lport); + frame->remote_port = ntohs(inp->inp_fport); + frame->local_seq = tp->snd_nxt; + frame->remote_seq = tp->rcv_nxt; + if (inp->inp_vflag & INP_IPV4) { + frame->length = frame_data_offset + + sizeof(struct ip) + sizeof(struct tcphdr); + frame->reply_length = frame->length; + + frame->addr_length = sizeof(struct in_addr); + bcopy(&inp->inp_laddr, frame->local_addr, + sizeof(struct in_addr)); + bcopy(&inp->inp_faddr, frame->remote_addr, + sizeof(struct in_addr)); + } else { + struct in6_addr *ip6; + + frame->length = frame_data_offset + + sizeof(struct ip6_hdr) + sizeof(struct tcphdr); + frame->reply_length = frame->length; + + frame->addr_length = sizeof(struct in6_addr); + ip6 = (struct in6_addr *)(void *)frame->local_addr; + bcopy(&inp->in6p_laddr, ip6, sizeof(struct in6_addr)); + if (IN6_IS_SCOPE_EMBED(ip6)) + ip6->s6_addr16[1] = 0; + + ip6 = (struct in6_addr *)(void *)frame->remote_addr; + bcopy(&inp->in6p_faddr, ip6, sizeof(struct in6_addr)); + if (IN6_IS_SCOPE_EMBED(ip6)) + ip6->s6_addr16[1] = 0; + } + + /* + * First the probe + */ + m = tcp_make_keepalive_frame(tp, ifp, TRUE); + if (m == NULL) { + tcp_unlock(so, 1, 0); + continue; + } + bcopy(m->m_data, frame->data + frame_data_offset, + m->m_len); + m_freem(m); + + /* + * Now the response packet to incoming probes + */ + m = tcp_make_keepalive_frame(tp, ifp, FALSE); + if (m == NULL) { + tcp_unlock(so, 1, 0); + continue; + } + bcopy(m->m_data, frame->reply_data + frame_data_offset, + m->m_len); + m_freem(m); + + frame_index++; + tcp_unlock(so, 1, 0); + } + lck_rw_done(tcbinfo.ipi_lock); + *used_frames_count = frame_index; +} + +errno_t +tcp_notify_ack_id_valid(struct tcpcb *tp, struct socket *so, + u_int32_t notify_id) +{ + struct tcp_notify_ack_marker *elm; + + if (so->so_snd.sb_cc == 0) + return (ENOBUFS); + + SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) { + /* Duplicate id is not allowed */ + if (elm->notify_id == notify_id) + return (EINVAL); + /* Duplicate position is not allowed */ + if (elm->notify_snd_una == tp->snd_una + so->so_snd.sb_cc) + return (EINVAL); + } + return (0); +} + +errno_t +tcp_add_notify_ack_marker(struct tcpcb *tp, u_int32_t notify_id) +{ + struct tcp_notify_ack_marker *nm, *elm = NULL; + struct socket *so = tp->t_inpcb->inp_socket; + + MALLOC(nm, struct tcp_notify_ack_marker *, sizeof (*nm), + M_TEMP, M_WAIT | M_ZERO); + if (nm == NULL) + return (ENOMEM); + nm->notify_id = notify_id; + nm->notify_snd_una = tp->snd_una + so->so_snd.sb_cc; + + SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) { + if (SEQ_GT(nm->notify_snd_una, elm->notify_snd_una)) + break; + } + + if (elm == NULL) { + VERIFY(SLIST_EMPTY(&tp->t_notify_ack)); + SLIST_INSERT_HEAD(&tp->t_notify_ack, nm, notify_next); + } else { + SLIST_INSERT_AFTER(elm, nm, notify_next); + } + tp->t_notify_ack_count++; + return (0); +} + +void +tcp_notify_ack_free(struct tcpcb *tp) +{ + struct tcp_notify_ack_marker *elm, *next; + if (SLIST_EMPTY(&tp->t_notify_ack)) + return; + + SLIST_FOREACH_SAFE(elm, &tp->t_notify_ack, notify_next, next) { + SLIST_REMOVE(&tp->t_notify_ack, elm, tcp_notify_ack_marker, + notify_next); + FREE(elm, M_TEMP); + } + SLIST_INIT(&tp->t_notify_ack); + tp->t_notify_ack_count = 0; +} + +inline void +tcp_notify_acknowledgement(struct tcpcb *tp, struct socket *so) +{ + struct tcp_notify_ack_marker *elm; + + elm = SLIST_FIRST(&tp->t_notify_ack); + if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) { + soevent(so, SO_FILT_HINT_LOCKED | SO_FILT_HINT_NOTIFY_ACK); + } +} + +void +tcp_get_notify_ack_count(struct tcpcb *tp, + struct tcp_notify_ack_complete *retid) +{ + struct tcp_notify_ack_marker *elm; + size_t complete = 0; + + SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) { + if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) + complete++; + else + break; + } + retid->notify_pending = tp->t_notify_ack_count - complete; + retid->notify_complete_count = min(TCP_MAX_NOTIFY_ACK, complete); +} + +void +tcp_get_notify_ack_ids(struct tcpcb *tp, + struct tcp_notify_ack_complete *retid) +{ + size_t i = 0; + struct tcp_notify_ack_marker *elm, *next; + + SLIST_FOREACH_SAFE(elm, &tp->t_notify_ack, notify_next, next) { + if (i >= retid->notify_complete_count) + break; + if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) { + retid->notify_complete_id[i++] = elm->notify_id; + SLIST_REMOVE(&tp->t_notify_ack, elm, + tcp_notify_ack_marker, notify_next); + FREE(elm, M_TEMP); + tp->t_notify_ack_count--; + } else { + break; + } + } +} + +bool +tcp_notify_ack_active(struct socket *so) +{ + if ((SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) && + SOCK_TYPE(so) == SOCK_STREAM) { + struct tcpcb *tp = intotcpcb(sotoinpcb(so)); + + if (!SLIST_EMPTY(&tp->t_notify_ack)) { + struct tcp_notify_ack_marker *elm; + elm = SLIST_FIRST(&tp->t_notify_ack); + if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) + return (true); + } + } + return (false); +} + +inline int32_t +inp_get_sndbytes_allunsent(struct socket *so, u_int32_t th_ack) +{ + struct inpcb *inp = sotoinpcb(so); + struct tcpcb *tp = intotcpcb(inp); + + if ((so->so_snd.sb_flags & SB_SNDBYTE_CNT) && + so->so_snd.sb_cc > 0) { + int32_t unsent, sent; + sent = tp->snd_max - th_ack; + if (tp->t_flags & TF_SENTFIN) + sent--; + unsent = so->so_snd.sb_cc - sent; + return (unsent); + } + return (0); +}