X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/490019cf9519204c5fb36b2fba54ceb983bb6b72..c3c9b80d004dbbfdf763edeb97968c6997e3b45b:/bsd/netinet/tcp_subr.c diff --git a/bsd/netinet/tcp_subr.c b/bsd/netinet/tcp_subr.c index 88d18875a..0b2a55138 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) 2000-2020 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 @@ -69,7 +68,6 @@ #include #include -#include #include #include #include @@ -91,6 +89,8 @@ #include #include #include +#include +#include #define tcp_minmssoverload fring #define _IP_VHL @@ -98,19 +98,15 @@ #include #include #include -#if INET6 #include -#endif +#include #include -#if INET6 #include -#endif #include #include #include -#if INET6 #include -#endif +#include #include #include #include @@ -120,21 +116,19 @@ #include #include -#if INET6 #include -#endif #include #if TCPDEBUG #include #endif +#include + #include #if IPSEC #include -#if INET6 #include -#endif -#endif /*IPSEC*/ +#endif /* IPSEC */ #if NECP #include @@ -142,59 +136,50 @@ #undef tcp_minmssoverload -#if CONFIG_MACF_NET -#include -#endif /* MAC_NET */ - #include #include #include #include #include +#include +#include -#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; +static tcp_cc tcp_ccgen; extern struct tcptimerlist tcp_timer_list; extern struct tcptailq tcp_tw_tailq; -int tcp_mssdflt = TCP_MSS; -SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW | CTLFLAG_LOCKED, - &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); +SYSCTL_SKMEM_TCP_INT(TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW | CTLFLAG_LOCKED, + int, tcp_mssdflt, TCP_MSS, "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, - "Default TCP Maximum Segment Size for IPv6"); -#endif - -extern int tcp_do_autorcvbuf; +SYSCTL_SKMEM_TCP_INT(TCPCTL_V6MSSDFLT, v6mssdflt, + CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_v6mssdflt, TCP6_MSS, + "Default TCP Maximum Segment Size for IPv6"); -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; +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_SKMEM_TCP_INT(OID_AUTO, fastopen_backlog, + CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_tfo_backlog, 10, + "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)"); +SYSCTL_SKMEM_TCP_INT(OID_AUTO, fastopen, CTLFLAG_RW | CTLFLAG_LOCKED, + int, tcp_fastopen, TCP_FASTOPEN_CLIENT | TCP_FASTOPEN_SERVER, + "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_SKMEM_TCP_INT(OID_AUTO, now_init, CTLFLAG_RD | CTLFLAG_LOCKED, + uint32_t, tcp_now_init, 0, "Initial tcp now value"); + +SYSCTL_SKMEM_TCP_INT(OID_AUTO, microuptime_init, CTLFLAG_RD | CTLFLAG_LOCKED, + uint32_t, tcp_microuptime_init, 0, "Initial tcp uptime value in micro seconds"); /* * Minimum MSS we accept and use. This prevents DoS attacks where @@ -204,41 +189,23 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen_fallback_min, CTLFLAG_RW | CTLFLAG_ * with packet generation and sending. Set to zero to disable MINMSS * 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 | CTLFLAG_LOCKED, - &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"); -#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_SKMEM_TCP_INT(OID_AUTO, minmss, CTLFLAG_RW | CTLFLAG_LOCKED, + int, tcp_minmss, TCP_MINMSS, "Minmum TCP Maximum Segment Size"); -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, pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED, +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, +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, +SYSCTL_SKMEM_TCP_INT(OID_AUTO, icmp_may_rst, CTLFLAG_RW | CTLFLAG_LOCKED, + static int, icmp_may_rst, 1, "Certain ICMP unreachable messages may abort connections in SYN_SENT"); -static int tcp_strict_rfc1948 = 0; -static int tcp_isn_reseed_interval = 0; +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, +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, @@ -246,43 +213,48 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, &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"); +SYSCTL_SKMEM_TCP_INT(OID_AUTO, rtt_min, CTLFLAG_RW | CTLFLAG_LOCKED, + int, tcp_TCPTV_MIN, 100, "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"); +SYSCTL_SKMEM_TCP_INT(OID_AUTO, rexmt_slop, CTLFLAG_RW, + int, tcp_rexmt_slop, TCPTV_REXMTSLOP, "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_SKMEM_TCP_INT(OID_AUTO, randomize_ports, CTLFLAG_RW | CTLFLAG_LOCKED, + __private_extern__ int, 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"); +SYSCTL_SKMEM_TCP_INT(OID_AUTO, win_scale_factor, CTLFLAG_RW | CTLFLAG_LOCKED, + __private_extern__ int, tcp_win_scale, 3, "Window scaling factor"); + +#if (DEVELOPMENT || DEBUG) +SYSCTL_SKMEM_TCP_INT(OID_AUTO, init_rtt_from_cache, + CTLFLAG_RW | CTLFLAG_LOCKED, static int, tcp_init_rtt_from_cache, 1, + "Initalize RTT from route cache"); +#else +SYSCTL_SKMEM_TCP_INT(OID_AUTO, init_rtt_from_cache, + CTLFLAG_RD | CTLFLAG_LOCKED, static int, tcp_init_rtt_from_cache, 1, + "Initalize RTT from route cache"); +#endif /* (DEVELOPMENT || DEBUG) */ + +static int tso_debug = 0; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso_debug, CTLFLAG_RW | CTLFLAG_LOCKED, + &tso_debug, 0, "TSO verbosity"); -static void tcp_cleartaocache(void); -static void tcp_notify(struct inpcb *, int); +static void tcp_notify(struct inpcb *, int); -struct zone *sack_hole_zone; -struct zone *tcp_reass_zone; -struct zone *tcp_bwmeas_zone; -struct zone *tcp_rxt_seg_zone; +struct zone *sack_hole_zone; +struct zone *tcp_reass_zone; +struct zone *tcp_bwmeas_zone; +struct zone *tcp_rxt_seg_zone; -extern int slowlink_wsize; /* window correction for slow links */ +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. * @@ -290,12 +262,12 @@ 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; +__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 @@ -305,22 +277,28 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD | CTLFLAG_LOCKED, * we avoid changing most of the rest of the code (although it needs * to be changed, eventually, for greater efficiency). */ -#define ALIGNMENT 32 -struct inp_tp { - struct inpcb inp; - struct tcpcb tcb __attribute__((aligned(ALIGNMENT))); +#define ALIGNMENT 32 +struct inp_tp { + struct inpcb inp; + struct tcpcb tcb __attribute__((aligned(ALIGNMENT))); }; #undef ALIGNMENT int get_inpcb_str_size(void); int get_tcp_str_size(void); +os_log_t tcp_mpkl_log_object = NULL; + 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 void tcp_flow_lim_stats(struct ifnet_stats_per_flow *ifs, + struct if_lim_perf_stat *stat); +static void tcp_flow_ecn_perf_stats(struct ifnet_stats_per_flow *ifs, + struct if_tcp_ecn_perf_stat *stat); static aes_encrypt_ctx tfo_ctx; /* Crypto-context for TFO */ @@ -328,21 +306,18 @@ void tcp_tfo_gen_cookie(struct inpcb *inp, u_char *out, size_t blk_size) { u_char in[CCAES_BLOCK_SIZE]; -#if INET6 int isipv6 = inp->inp_vflag & INP_IPV6; -#endif VERIFY(blk_size == CCAES_BLOCK_SIZE); bzero(&in[0], CCAES_BLOCK_SIZE); bzero(&out[0], CCAES_BLOCK_SIZE); -#if INET6 - if (isipv6) + if (isipv6) { memcpy(in, &inp->in6p_faddr, sizeof(struct in6_addr)); - else -#endif /* INET6 */ + } else { memcpy(in, &inp->inp_faddr, sizeof(struct in_addr)); + } aes_encrypt_cbc(in, NULL, 1, out, &tfo_ctx); } @@ -352,7 +327,10 @@ 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; @@ -363,18 +341,23 @@ 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. */ bzero(keystring, sizeof(keystring)); error = sysctl_io_string(req, keystring, sizeof(keystring), 0, NULL); - if (error) + if (error) { 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; @@ -384,46 +367,48 @@ tcp_sysctl_fastopenkey(__unused struct sysctl_oid *oidp, __unused void *arg1, aes_encrypt_key128((u_char *)key, &tfo_ctx); exit: - return (error); + return error; } -int get_inpcb_str_size(void) +int +get_inpcb_str_size(void) { return sizeof(struct inpcb); } -int get_tcp_str_size(void) +int +get_tcp_str_size(void) { return sizeof(struct tcpcb); } -int tcp_freeq(struct tcpcb *tp); - 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); + */ + size = size & (size - 1); } /* Check for overflow when rounding up */ @@ -438,11 +423,11 @@ static int scale_to_powerof2(int size) { } static void -tcp_tfo_init() +tcp_tfo_init(void) { u_char key[TCP_FASTOPEN_KEYLEN]; - read_random(key, sizeof(key)); + read_frandom(key, sizeof(key)); aes_encrypt_key128(key, &tfo_ctx); } @@ -454,18 +439,18 @@ 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; + uint32_t logging_config; - VERIFY((pp->pr_flags & (PR_INITIALIZED|PR_ATTACHED)) == PR_ATTACHED); + VERIFY((pp->pr_flags & (PR_INITIALIZED | PR_ATTACHED)) == PR_ATTACHED); - if (tcp_initialized) + if (tcp_initialized) { return; + } tcp_initialized = 1; tcp_ccgen = 1; - tcp_cleartaocache(); - tcp_keepinit = TCPTV_KEEP_INIT; tcp_keepidle = TCPTV_KEEP_IDLE; tcp_keepintvl = TCPTV_KEEPINTVL; @@ -474,8 +459,17 @@ tcp_init(struct protosw *pp, struct domain *dp) tcp_msl = TCPTV_MSL; microuptime(&tcp_uptime); - read_random(&tcp_now, sizeof(tcp_now)); - tcp_now = tcp_now & 0x3fffffff; /* Starts tcp internal clock at a random value */ + read_frandom(&tcp_now, sizeof(tcp_now)); + + /* Starts tcp internal clock at a random value */ + tcp_now = tcp_now & 0x3fffffff; + + /* expose initial uptime/now via systcl for utcp to keep time sync */ + tcp_now_init = tcp_now; + tcp_microuptime_init = + (uint32_t)(tcp_uptime.tv_usec + (tcp_uptime.tv_sec * USEC_PER_SEC)); + SYSCTL_SKMEM_UPDATE_FIELD(tcp.microuptime_init, tcp_microuptime_init); + SYSCTL_SKMEM_UPDATE_FIELD(tcp.now_init, tcp_now_init); tcp_tfo_init(); @@ -487,7 +481,8 @@ 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 @@ -513,69 +508,45 @@ tcp_init(struct protosw *pp, struct domain *dp) tcp_tcbhashsize = 16; } printf("WARNING: TCB hash size not a power of 2, " - "scaled from %d to %d.\n", - old_hash_size, - tcp_tcbhashsize); + "scaled from %d to %d.\n", + old_hash_size, + 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)); - tcbinfo.ipi_zone = zinit(str_size, 120000*str_size, 8192, "tcpcb"); - zone_change(tcbinfo.ipi_zone, Z_CALLERACCT, FALSE); - zone_change(tcbinfo.ipi_zone, Z_EXPAND, TRUE); + &tcbinfo.ipi_porthashmask); + str_size = (vm_size_t)P2ROUNDUP(sizeof(struct inp_tp), sizeof(u_int64_t)); + tcbinfo.ipi_zone = zone_create("tcpcb", str_size, ZC_NONE); tcbinfo.ipi_gc = tcp_gc; tcbinfo.ipi_timer = tcp_itimer; in_pcbinfo_attach(&tcbinfo); - str_size = P2ROUNDUP(sizeof(struct sackhole), sizeof(u_int64_t)); - sack_hole_zone = zinit(str_size, 120000*str_size, 8192, "sack_hole zone"); - zone_change(sack_hole_zone, Z_CALLERACCT, FALSE); - zone_change(sack_hole_zone, Z_EXPAND, TRUE); + str_size = (vm_size_t)P2ROUNDUP(sizeof(struct sackhole), sizeof(u_int64_t)); + sack_hole_zone = zone_create("sack_hole zone", str_size, ZC_NONE); - str_size = P2ROUNDUP(sizeof(struct tseg_qent), sizeof(u_int64_t)); - tcp_reass_zone = zinit(str_size, (nmbclusters >> 4) * 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); + str_size = (vm_size_t)P2ROUNDUP(sizeof(struct tseg_qent), sizeof(u_int64_t)); + tcp_reass_zone = zone_create("tcp_reass_zone", str_size, ZC_NONE); - 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); + str_size = (vm_size_t)P2ROUNDUP(sizeof(struct bwmeas), sizeof(u_int64_t)); + tcp_bwmeas_zone = zone_create("tcp_bwmeas_zone", str_size, ZC_ZFREE_CLEARMEM); - str_size = P2ROUNDUP(sizeof(struct tcp_ccstate), sizeof(u_int64_t)); - tcp_cc_zone = zinit(str_size, 20000 * str_size, 0, "tcp_cc_zone"); - zone_change(tcp_cc_zone, Z_CALLERACCT, FALSE); - zone_change(tcp_cc_zone, Z_EXPAND, TRUE); + str_size = (vm_size_t)P2ROUNDUP(sizeof(struct tcp_ccstate), sizeof(u_int64_t)); + tcp_cc_zone = zone_create("tcp_cc_zone", str_size, ZC_NONE); - str_size = P2ROUNDUP(sizeof(struct tcp_rxt_seg), sizeof(u_int64_t)); - tcp_rxt_seg_zone = zinit(str_size, 10000 * str_size, 0, - "tcp_rxt_seg_zone"); - zone_change(tcp_rxt_seg_zone, Z_CALLERACCT, FALSE); - zone_change(tcp_rxt_seg_zone, Z_EXPAND, TRUE); + str_size = (vm_size_t)P2ROUNDUP(sizeof(struct tcp_rxt_seg), sizeof(u_int64_t)); + tcp_rxt_seg_zone = zone_create("tcp_rxt_seg_zone", str_size, ZC_NONE); -#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; - _max_protohdr = max_protohdr; /* round it up */ + _max_protohdr = (int)max_protohdr; /* round it up */ } - if (max_linkhdr + max_protohdr > MCLBYTES) + if (max_linkhdr + max_protohdr > MCLBYTES) { panic("tcp_init"); + } #undef TCP_MINPROTOHDR /* Initialize time wait and timer lists */ @@ -584,39 +555,59 @@ 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); - - /* Initialize TCP LRO data structures */ - tcp_lro_init(); + tcp_uptime_lock = lck_spin_alloc_init(tcp_uptime_mtx_grp, + tcp_uptime_mtx_attr); /* Initialize TCP Cache */ tcp_cache_init(); + tcp_mpkl_log_object = MPKL_CREATE_LOGOBJECT("com.apple.xnu.tcp"); + if (tcp_mpkl_log_object == NULL) { + panic("MPKL_CREATE_LOGOBJECT failed"); + } + + logging_config = atm_get_diagnostic_config(); + if (logging_config & 0x80000000) { + tcp_log_privacy = 1; + } + + PE_parse_boot_argn("tcp_log", &tcp_log_enable_flags, sizeof(tcp_log_enable_flags)); + /* - * If more than 60 MB of mbuf pool is available, increase the + * If more than 4GB of actual memory is available, increase the * maximum allowed receive and send socket buffer size. */ - if (nmbclusters > 30720) { - tcp_autorcvbuf_max = 1024 * 1024; - tcp_autosndbuf_max = 1024 * 1024; + if (mem_actual >= (1ULL << (GBSHIFT + 2))) { + tcp_autorcvbuf_max = 4 * 1024 * 1024; + tcp_autosndbuf_max = 4 * 1024 * 1024; + + SYSCTL_SKMEM_UPDATE_FIELD(tcp.autorcvbufmax, tcp_autorcvbuf_max); + SYSCTL_SKMEM_UPDATE_FIELD(tcp.autosndbufmax, tcp_autosndbuf_max); } } @@ -626,46 +617,42 @@ 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; -#if INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { struct ip6_hdr *ip6; ip6 = (struct ip6_hdr *)ip_ptr; ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | - (inp->inp_flow & IPV6_FLOWINFO_MASK); + (inp->inp_flow & IPV6_FLOWINFO_MASK); ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | - (IPV6_VERSION & 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, - htonl(sizeof (struct tcphdr) + IPPROTO_TCP)); - } 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)); + htonl(sizeof(struct tcphdr) + IPPROTO_TCP)); + } else { + 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; @@ -685,20 +672,20 @@ 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; m = m_get(M_DONTWAIT, MT_HEADER); - if (m == NULL) - return (0); + if (m == NULL) { + return NULL; + } m->m_len = sizeof(struct tcptemp); n = mtod(m, struct tcptemp *); tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t); - return (n); + return n; } /* @@ -717,80 +704,74 @@ tcp_maketemplate(tp) */ void tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, - tcp_seq ack, tcp_seq seq, int flags, struct tcp_respond_args *tra) + tcp_seq ack, tcp_seq seq, uint8_t flags, struct tcp_respond_args *tra) { - int tlen; + uint16_t tlen; int win = 0; struct route *ro = 0; struct route sro; struct ip *ip; struct tcphdr *nth; -#if INET6 struct route_in6 *ro6 = 0; struct route_in6 sro6; struct ip6_hdr *ip6; int isipv6; -#endif /* INET6 */ struct ifnet *outif; + int sotc = SO_TC_UNSPEC; + bool check_qos_marking_again = FALSE; -#if INET6 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6; ip6 = ipgen; -#endif /* INET6 */ ip = ipgen; if (tp) { + check_qos_marking_again = tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_POLICY_OVERRIDE ? FALSE : TRUE; if (!(flags & TH_RST)) { win = tcp_sbspace(tp); - if (win > (int32_t)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) + if (isipv6) { ro6 = &tp->t_inpcb->in6p_route; - else -#endif /* INET6 */ - ro = &tp->t_inpcb->inp_route; + } else { + ro = &tp->t_inpcb->inp_route; + } } else { -#if INET6 if (isipv6) { ro6 = &sro6; - bzero(ro6, sizeof *ro6); - } else -#endif /* INET6 */ - { + bzero(ro6, sizeof(*ro6)); + } else { ro = &sro; - bzero(ro, sizeof *ro); + bzero(ro, sizeof(*ro)); } } if (m == 0) { - m = m_gethdr(M_DONTWAIT, MT_HEADER); /* MAC-OK */ - if (m == NULL) + m = m_gethdr(M_DONTWAIT, MT_HEADER); /* MAC-OK */ + if (m == NULL) { return; + } tlen = 0; m->m_data += max_linkhdr; -#if INET6 if (isipv6) { VERIFY((MHLEN - max_linkhdr) >= - (sizeof (*ip6) + sizeof (*nth))); - bcopy((caddr_t)ip6, mtod(m, caddr_t), - sizeof(struct ip6_hdr)); + (sizeof(*ip6) + sizeof(*nth))); + bcopy((caddr_t)ip6, mtod(m, caddr_t), + sizeof(struct ip6_hdr)); ip6 = mtod(m, struct ip6_hdr *); nth = (struct tcphdr *)(void *)(ip6 + 1); - } else -#endif /* INET6 */ - { + } else { VERIFY((MHLEN - max_linkhdr) >= - (sizeof (*ip) + sizeof (*nth))); + (sizeof(*ip) + sizeof(*nth))); bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); ip = mtod(m, struct ip *); nth = (struct tcphdr *)(void *)(ip + 1); } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); #if MPTCP - if ((tp) && (tp->t_mpflags & TMPF_RESET)) + if ((tp) && (tp->t_mpflags & TMPF_RESET)) { flags = (TH_RST | TH_ACK); - else + } else #endif flags = TH_ACK; } else { @@ -799,21 +780,18 @@ 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; } -#if INET6 +#define xchg(a, b, type) { type t; t = a; a = b; b = t; } 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 *)(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); - } + } else { + /* 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 @@ -826,78 +804,60 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, xchg(nth->th_dport, nth->th_sport, n_short); #undef xchg } -#if INET6 if (isipv6) { - ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + - tlen)); - tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); - } else -#endif - { - tlen += sizeof (struct tcpiphdr); - ip->ip_len = tlen; - ip->ip_ttl = ip_defttl; - } + ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) + + tlen)); + tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr); + } else { + tlen += sizeof(struct tcpiphdr); + ip->ip_len = tlen; + ip->ip_ttl = (uint8_t)ip_defttl; + } m->m_len = tlen; m->m_pkthdr.len = tlen; m->m_pkthdr.rcvif = 0; -#if CONFIG_MACF_NET - if (tp != NULL && tp->t_inpcb != NULL) { - /* - * Packet is associated with a socket, so allow the - * label of the response to reflect the socket label. - */ - mac_mbuf_label_associate_inpcb(tp->t_inpcb, m); - } else { - /* - * Packet is not associated with a socket, so possibly - * update the label in place. - */ - mac_netinet_tcp_reply(m); + if (tra->keep_alive) { + m->m_pkthdr.pkt_flags |= PKTF_KEEPALIVE; } -#endif nth->th_seq = htonl(seq); nth->th_ack = htonl(ack); nth->th_x2 = 0; - nth->th_off = sizeof (struct tcphdr) >> 2; + nth->th_off = sizeof(struct tcphdr) >> 2; nth->th_flags = flags; - if (tp) + if (tp) { nth->th_win = htons((u_short) (win >> tp->rcv_scale)); - else + } else { nth->th_win = htons((u_short)win); + } nth->th_urp = 0; -#if INET6 if (isipv6) { nth->th_sum = 0; nth->th_sum = in6_pseudo(&ip6->ip6_src, &ip6->ip6_dst, - htonl((tlen - sizeof (struct ip6_hdr)) + IPPROTO_TCP)); + htonl((tlen - sizeof(struct ip6_hdr)) + IPPROTO_TCP)); m->m_pkthdr.csum_flags = CSUM_TCPIPV6; m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, - ro6 && ro6->ro_rt ? - ro6->ro_rt->rt_ifp : - NULL); - } else -#endif /* INET6 */ - { + ro6 && ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); + } else { nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, - htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); + htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); m->m_pkthdr.csum_flags = CSUM_TCP; m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); } #if TCPDEBUG - if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) + if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) { tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); + } #endif #if NECP - necp_mark_packet_from_socket(m, tp ? tp->t_inpcb : NULL, 0, 0); + necp_mark_packet_from_socket(m, tp ? tp->t_inpcb : NULL, 0, 0, 0, 0); #endif /* NECP */ #if IPSEC if (tp != NULL && tp->t_inpcb->inp_sp != NULL && - ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) { + ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) { m_freem(m); return; } @@ -908,60 +868,125 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, if (isipv6) { svc_flags |= PKT_SCF_IPV6; } - set_packet_service_class(m, tp->t_inpcb->inp_socket, - MBUF_SC_UNSPEC, svc_flags); + sotc = tp->t_inpcb->inp_socket->so_traffic_class; + if ((flags & TH_RST) == 0) { + set_packet_service_class(m, tp->t_inpcb->inp_socket, + sotc, svc_flags); + } else { + m_set_service_class(m, MBUF_SC_BK_SYS); + } /* Embed flowhash and flow control flags */ m->m_pkthdr.pkt_flowsrc = FLOWSRC_INPCB; m->m_pkthdr.pkt_flowid = tp->t_inpcb->inp_flowhash; - m->m_pkthdr.pkt_flags |= PKTF_FLOW_ID | PKTF_FLOW_LOCALSRC; -#if MPTCP - /* Disable flow advisory when using MPTCP. */ - if (!(tp->t_mpflags & TMPF_MPTCP_TRUE)) -#endif /* MPTCP */ - m->m_pkthdr.pkt_flags |= PKTF_FLOW_ADV; + m->m_pkthdr.pkt_flags |= (PKTF_FLOW_ID | PKTF_FLOW_LOCALSRC | PKTF_FLOW_ADV); m->m_pkthdr.pkt_proto = IPPROTO_TCP; + m->m_pkthdr.tx_tcp_pid = tp->t_inpcb->inp_socket->last_pid; + m->m_pkthdr.tx_tcp_e_pid = tp->t_inpcb->inp_socket->e_pid; + + if (flags & TH_RST) { + m->m_pkthdr.comp_gencnt = tp->t_comp_gencnt; + } + } else { + if (flags & TH_RST) { + m->m_pkthdr.comp_gencnt = TCP_ACK_COMPRESSION_DUMMY; + m_set_service_class(m, MBUF_SC_BK_SYS); + } } -#if INET6 if (isipv6) { - struct ip6_out_args ip6oa = { tra->ifscope, { 0 }, - IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR, 0 }; - - if (tra->ifscope != IFSCOPE_NONE) + struct ip6_out_args ip6oa; + bzero(&ip6oa, sizeof(ip6oa)); + ip6oa.ip6oa_boundif = tra->ifscope; + ip6oa.ip6oa_flags = IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR; + ip6oa.ip6oa_sotc = SO_TC_UNSPEC; + ip6oa.ip6oa_netsvctype = _NET_SERVICE_TYPE_UNSPEC; + + if (tra->ifscope != IFSCOPE_NONE) { ip6oa.ip6oa_flags |= IP6OAF_BOUND_IF; - if (tra->nocell) + } + if (tra->nocell) { ip6oa.ip6oa_flags |= IP6OAF_NO_CELLULAR; - if (tra->noexpensive) + } + if (tra->noexpensive) { ip6oa.ip6oa_flags |= IP6OAF_NO_EXPENSIVE; - if (tra->awdl_unrestricted) + } + if (tra->noconstrained) { + ip6oa.ip6oa_flags |= IP6OAF_NO_CONSTRAINED; + } + 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.qos_marking_gencount = tp->t_inpcb->inp_policyresult.results.qos_marking_gencount; + if (check_qos_marking_again) { + ip6oa.ip6oa_flags |= IP6OAF_REDO_QOSMARKING_POLICY; + } + ip6oa.ip6oa_netsvctype = tp->t_inpcb->inp_socket->so_netsvctype; + } (void) ip6_output(m, NULL, ro6, IPV6_OUTARGS, NULL, NULL, &ip6oa); + if (check_qos_marking_again) { + struct inpcb *inp = tp->t_inpcb; + inp->inp_policyresult.results.qos_marking_gencount = ip6oa.qos_marking_gencount; + if (ip6oa.ip6oa_flags & IP6OAF_QOSMARKING_ALLOWED) { + inp->inp_socket->so_flags1 |= SOF1_QOSMARKING_ALLOWED; + } else { + inp->inp_socket->so_flags1 &= ~SOF1_QOSMARKING_ALLOWED; + } + } + if (tp != NULL && ro6 != NULL && ro6->ro_rt != NULL && (outif = ro6->ro_rt->rt_ifp) != - tp->t_inpcb->in6p_last_outifp) + tp->t_inpcb->in6p_last_outifp) { tp->t_inpcb->in6p_last_outifp = outif; + } - if (ro6 == &sro6) + if (ro6 == &sro6) { ROUTE_RELEASE(ro6); - } else -#endif /* INET6 */ - { - struct ip_out_args ipoa = { tra->ifscope, { 0 }, - IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR, 0 }; - - if (tra->ifscope != IFSCOPE_NONE) + } + } else { + struct ip_out_args ipoa; + bzero(&ipoa, sizeof(ipoa)); + ipoa.ipoa_boundif = tra->ifscope; + ipoa.ipoa_flags = IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR; + ipoa.ipoa_sotc = SO_TC_UNSPEC; + ipoa.ipoa_netsvctype = _NET_SERVICE_TYPE_UNSPEC; + + if (tra->ifscope != IFSCOPE_NONE) { ipoa.ipoa_flags |= IPOAF_BOUND_IF; - if (tra->nocell) + } + if (tra->nocell) { ipoa.ipoa_flags |= IPOAF_NO_CELLULAR; - if (tra->noexpensive) + } + if (tra->noexpensive) { ipoa.ipoa_flags |= IPOAF_NO_EXPENSIVE; - if (tra->awdl_unrestricted) + } + if (tra->noconstrained) { + ipoa.ipoa_flags |= IPOAF_NO_CONSTRAINED; + } + 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; + } + if (!(tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_POLICY_OVERRIDE)) { + ipoa.ipoa_flags |= IPOAF_REDO_QOSMARKING_POLICY; + } + ipoa.qos_marking_gencount = tp->t_inpcb->inp_policyresult.results.qos_marking_gencount; + 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); @@ -971,11 +996,20 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, */ (void) ip_output(m, NULL, &sro, IP_OUTARGS, NULL, &ipoa); + if (check_qos_marking_again) { + struct inpcb *inp = tp->t_inpcb; + inp->inp_policyresult.results.qos_marking_gencount = ipoa.qos_marking_gencount; + if (ipoa.ipoa_flags & IPOAF_QOSMARKING_ALLOWED) { + inp->inp_socket->so_flags1 |= SOF1_QOSMARKING_ALLOWED; + } else { + inp->inp_socket->so_flags1 &= ~SOF1_QOSMARKING_ALLOWED; + } + } if (tp != NULL && sro.ro_rt != NULL && (outif = sro.ro_rt->rt_ifp) != - tp->t_inpcb->inp_last_outifp) + tp->t_inpcb->inp_last_outifp) { tp->t_inpcb->inp_last_outifp = outif; - + } if (ro != &sro) { /* Synchronize cached PCB route */ inp_route_copyin(tp->t_inpcb, &sro); @@ -992,84 +1026,106 @@ 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; -#if INET6 + struct tcpcb *tp; + struct socket *so = inp->inp_socket; int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; -#endif /* INET6 */ + uint32_t random_32; 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 = -#if INET6 - isipv6 ? tcp_v6mssdflt : -#endif /* INET6 */ - tcp_mssdflt; + tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; - if (tcp_do_rfc1323) - tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); - if (tcp_do_sack) - tp->t_flagsext |= TF_SACK_ENABLE; + tp->t_flags = (TF_REQ_SCALE | TF_REQ_TSTMP); + tp->t_flagsext |= TF_SACK_ENABLE; 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; - if (tcp_use_newreno) + if (tcp_use_newreno) { /* use newreno by default */ tp->tcp_cc_index = TCP_CC_ALGO_NEWRENO_INDEX; - else + } else { tp->tcp_cc_index = TCP_CC_ALGO_CUBIC_INDEX; + } tcp_cc_allocate_state(tp); - if (CC_ALGO(tp)->init != NULL) + if (CC_ALGO(tp)->init != NULL) { CC_ALGO(tp)->init(tp); + } - tp->snd_cwnd = TCP_CC_CWND_INIT_BYTES; + tp->snd_cwnd = tcp_initial_cwnd(tp); tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->snd_ssthresh_prev = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->t_rcvtime = tcp_now; tp->tentry.timer_start = tcp_now; + tp->rcv_unackwin = tcp_now; tp->t_persist_timeout = tcp_max_persist_timeout; tp->t_persist_stop = 0; tp->t_flagsext |= TF_RCVUNACK_WAITSS; - tp->t_rexmtthresh = tcprexmtthresh; + tp->t_rexmtthresh = (uint8_t)tcprexmtthresh; + tp->rfbuf_ts = tcp_now; + tp->rfbuf_space = tcp_initial_cwnd(tp); + tp->t_forced_acks = TCP_FORCED_ACKS_COUNT; + + /* Enable bandwidth measurement on this connection */ + tp->t_flagsext |= TF_MEASURESNDBW; + if (tp->t_bwmeas == NULL) { + tp->t_bwmeas = tcp_bwmeas_alloc(tp); + if (tp->t_bwmeas == NULL) { + tp->t_flagsext &= ~TF_MEASURESNDBW; + } + } /* Clear time wait tailq entry */ tp->t_twentry.tqe_next = NULL; tp->t_twentry.tqe_prev = NULL; + read_frandom(&random_32, sizeof(random_32)); + if (__probable(tcp_do_ack_compression)) { + tp->t_comp_gencnt = random_32; + if (tp->t_comp_gencnt <= TCP_ACK_COMPRESSION_DUMMY) { + tp->t_comp_gencnt = TCP_ACK_COMPRESSION_DUMMY + 1; + } + tp->t_comp_lastinc = tcp_now; + } + + if (__probable(tcp_randomize_timestamps)) { + tp->t_ts_offset = random_32; + } + /* * IPv4 TTL initialization is necessary for an IPv6 socket as well, * because the socket may be bound to an IPv6 wildcard address, * which may match an IPv4-mapped IPv6 address. */ - inp->inp_ip_ttl = ip_defttl; + inp->inp_ip_ttl = (uint8_t)ip_defttl; inp->inp_ppcb = (caddr_t)tp; - return (tp); /* XXX */ + return tp; /* XXX */ } /* @@ -1078,9 +1134,7 @@ 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 @@ -1089,79 +1143,144 @@ tcp_drop(tp, errno) if (TCPS_HAVERCVDSYN(tp->t_state)) { DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, - struct tcpcb *, tp, int32_t, TCPS_CLOSED); + struct tcpcb *, tp, int32_t, TCPS_CLOSED); tp->t_state = TCPS_CLOSED; (void) tcp_output(tp); tcpstat.tcps_drops++; - } else + } else { tcpstat.tcps_conndrops++; - if (errno == ETIMEDOUT && tp->t_softerror) + } + if (errno == ETIMEDOUT && tp->t_softerror) { errno = tp->t_softerror; - so->so_error = errno; - return (tcp_close(tp)); + } + so->so_error = (u_short)errno; + + TCP_LOG_CONNECTION_SUMMARY(tp); + + return tcp_close(tp); } 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); - if (rtt != 0) { + TCP_LOG_RTM_RTT(tp, rt); + + if (rtt != 0 && tcp_init_rtt_from_cache != 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) + 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)); + } 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)); + (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; } + + /* + * The RTO formula in the route metric case is based on: + * 4 * srtt + 8 * rttvar + * modulo the min, max and slop + */ TCPT_RANGESET(tp->t_rxtcur, - ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, - tp->t_rttmin, TCPTV_REXMTMAX, - TCP_ADD_REXMTSLOP(tp)); + ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, + tp->t_rttmin, TCPTV_REXMTMAX, + TCP_ADD_REXMTSLOP(tp)); + } + + TCP_LOG_RTT_INFO(tp); +} + +static inline void +tcp_create_ifnet_stats_per_flow(struct tcpcb *tp, + struct ifnet_stats_per_flow *ifs) +{ + struct inpcb *inp; + struct socket *so; + if (tp == NULL || ifs == NULL) { + return; + } + + bzero(ifs, sizeof(*ifs)); + inp = tp->t_inpcb; + so = inp->inp_socket; + + ifs->ipv4 = (inp->inp_vflag & INP_IPV6) ? 0 : 1; + ifs->local = (tp->t_flags & TF_LOCAL) ? 1 : 0; + ifs->connreset = (so->so_error == ECONNRESET) ? 1 : 0; + ifs->conntimeout = (so->so_error == ETIMEDOUT) ? 1 : 0; + ifs->ecn_flags = tp->ecn_flags; + ifs->txretransmitbytes = tp->t_stat.txretransmitbytes; + ifs->rxoutoforderbytes = tp->t_stat.rxoutoforderbytes; + ifs->rxmitpkts = tp->t_stat.rxmitpkts; + ifs->rcvoopack = tp->t_rcvoopack; + ifs->pawsdrop = tp->t_pawsdrop; + ifs->sack_recovery_episodes = tp->t_sack_recovery_episode; + ifs->reordered_pkts = tp->t_reordered_pkts; + ifs->dsack_sent = tp->t_dsack_sent; + ifs->dsack_recvd = tp->t_dsack_recvd; + ifs->srtt = tp->t_srtt; + ifs->rttupdated = tp->t_rttupdated; + ifs->rttvar = tp->t_rttvar; + ifs->rttmin = get_base_rtt(tp); + if (tp->t_bwmeas != NULL && tp->t_bwmeas->bw_sndbw_max > 0) { + ifs->bw_sndbw_max = tp->t_bwmeas->bw_sndbw_max; + } else { + ifs->bw_sndbw_max = 0; + } + if (tp->t_bwmeas != NULL && tp->t_bwmeas->bw_rcvbw_max > 0) { + ifs->bw_rcvbw_max = tp->t_bwmeas->bw_rcvbw_max; + } else { + ifs->bw_rcvbw_max = 0; } + ifs->bk_txpackets = so->so_tc_stats[MBUF_TC_BK].txpackets; + ifs->txpackets = inp->inp_stat->txpackets; + ifs->rxpackets = inp->inp_stat->rxpackets; } static inline void -tcp_update_ecn_perf_stats(struct tcpcb *tp, +tcp_flow_ecn_perf_stats(struct ifnet_stats_per_flow *ifs, 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); + stat->total_txpkts += ifs->txpackets; + stat->total_rxpkts += ifs->rxpackets; + stat->total_rxmitpkts += ifs->rxmitpkts; + stat->total_oopkts += ifs->rcvoopack; + stat->total_reorderpkts += (ifs->reordered_pkts + + ifs->pawsdrop + ifs->dsack_sent + ifs->dsack_recvd); /* Average RTT */ - curval = (tp->t_srtt >> TCP_RTT_SHIFT); - if (curval > 0 && tp->t_rttupdated >= 16) { + curval = ifs->srtt >> TCP_RTT_SHIFT; + if (curval > 0 && ifs->rttupdated >= 16) { if (stat->rtt_avg == 0) { stat->rtt_avg = curval; } else { oldval = stat->rtt_avg; - stat->rtt_avg = - ((oldval << 4) - oldval + curval) >> 4; + stat->rtt_avg = ((oldval << 4) - oldval + curval) >> 4; } } /* RTT variance */ - curval = tp->t_rttvar >> TCP_RTTVAR_SHIFT; - if (curval > 0 && tp->t_rttupdated >= 16) { + curval = ifs->rttvar >> TCP_RTTVAR_SHIFT; + if (curval > 0 && ifs->rttupdated >= 16) { if (stat->rtt_var == 0) { stat->rtt_var = curval; } else { @@ -1171,12 +1290,76 @@ tcp_update_ecn_perf_stats(struct tcpcb *tp, } } - /* Total number of SACK recovery episodes */ - stat->sack_episodes += tp->t_sack_recovery_episode; - - if (inp->inp_socket->so_error == ECONNRESET) + /* SACK episodes */ + stat->sack_episodes += ifs->sack_recovery_episodes; + if (ifs->connreset) { stat->rst_drop++; - return; + } +} + +static inline void +tcp_flow_lim_stats(struct ifnet_stats_per_flow *ifs, + struct if_lim_perf_stat *stat) +{ + u_int64_t curval, oldval; + + stat->lim_total_txpkts += ifs->txpackets; + stat->lim_total_rxpkts += ifs->rxpackets; + stat->lim_total_retxpkts += ifs->rxmitpkts; + stat->lim_total_oopkts += ifs->rcvoopack; + + if (ifs->bw_sndbw_max > 0) { + /* convert from bytes per ms to bits per second */ + ifs->bw_sndbw_max *= 8000; + stat->lim_ul_max_bandwidth = MAX(stat->lim_ul_max_bandwidth, + ifs->bw_sndbw_max); + } + + if (ifs->bw_rcvbw_max > 0) { + /* convert from bytes per ms to bits per second */ + ifs->bw_rcvbw_max *= 8000; + stat->lim_dl_max_bandwidth = MAX(stat->lim_dl_max_bandwidth, + ifs->bw_rcvbw_max); + } + + /* Average RTT */ + curval = ifs->srtt >> TCP_RTT_SHIFT; + if (curval > 0 && ifs->rttupdated >= 16) { + if (stat->lim_rtt_average == 0) { + stat->lim_rtt_average = curval; + } else { + oldval = stat->lim_rtt_average; + stat->lim_rtt_average = + ((oldval << 4) - oldval + curval) >> 4; + } + } + + /* RTT variance */ + curval = ifs->rttvar >> TCP_RTTVAR_SHIFT; + if (curval > 0 && ifs->rttupdated >= 16) { + if (stat->lim_rtt_variance == 0) { + stat->lim_rtt_variance = curval; + } else { + oldval = stat->lim_rtt_variance; + stat->lim_rtt_variance = + ((oldval << 4) - oldval + curval) >> 4; + } + } + + if (stat->lim_rtt_min == 0) { + stat->lim_rtt_min = ifs->rttmin; + } else { + stat->lim_rtt_min = MIN(stat->lim_rtt_min, ifs->rttmin); + } + + /* connection timeouts */ + stat->lim_conn_attempts++; + if (ifs->conntimeout) { + stat->lim_conn_timeouts++; + } + + /* bytes sent using background delay-based algorithms */ + stat->lim_bk_txpkts += ifs->bk_txpackets; } /* @@ -1186,24 +1369,25 @@ tcp_update_ecn_perf_stats(struct tcpcb *tp, * 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; -#if INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; -#endif /* INET6 */ struct route *ro; struct rtentry *rt; int dosavessthresh; + struct ifnet_stats_per_flow ifs; /* tcp_close was called previously, bail */ - if (inp->inp_ppcb == NULL) - return(NULL); + if (inp->inp_ppcb == NULL) { + return NULL; + } + + tcp_del_fsw_flow(tp); 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 @@ -1218,22 +1402,21 @@ tcp_close(tp) * with the cleanup. */ if ((tp->t_flags & TF_CLOSING) || - inp->inp_sndinprog_cnt > 0) { + inp->inp_sndinprog_cnt > 0) { tp->t_flags |= TF_CLOSING; - return (NULL); + return NULL; } + TCP_LOG_CONNECTION_SUMMARY(tp); + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, - struct tcpcb *, tp, int32_t, TCPS_CLOSED); + struct tcpcb *, tp, int32_t, TCPS_CLOSED); -#if INET6 ro = (isipv6 ? (struct route *)&inp->in6p_route : &inp->inp_route); -#else - ro = &inp->inp_route; -#endif rt = ro->ro_rt; - if (rt != NULL) + if (rt != NULL) { RT_LOCK_SPIN(rt); + } /* * If we got enough samples through the srtt filter, @@ -1247,21 +1430,20 @@ 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; + bool log_rtt = false; -#if INET6 if (isipv6) { struct sockaddr_in6 *sin6; - if (rt == NULL) + if (rt == NULL) { goto no_valid_rt; + } sin6 = (struct sockaddr_in6 *)(void *)rt_key(rt); - if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) + if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { goto no_valid_rt; - } - else -#endif /* INET6 */ - if (ROUTE_UNUSABLE(ro) || + } + } else if (ROUTE_UNUSABLE(ro) || SIN(rt_key(rt))->sin_addr.s_addr == INADDR_ANY) { DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, struct tcpcb *, tp, @@ -1274,7 +1456,7 @@ tcp_close(tp) if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { i = tp->t_srtt * (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE)); - if (rt->rt_rmx.rmx_rtt && i) + if (rt->rt_rmx.rmx_rtt && i) { /* * filter this update to half the old & half * the new values, converting scale. @@ -1283,19 +1465,27 @@ tcp_close(tp) */ rt->rt_rmx.rmx_rtt = (rt->rt_rmx.rmx_rtt + i) / 2; - else + } else { rt->rt_rmx.rmx_rtt = i; + } tcpstat.tcps_cachedrtt++; + log_rtt = true; } if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { i = tp->t_rttvar * (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE)); - if (rt->rt_rmx.rmx_rttvar && i) + if (rt->rt_rmx.rmx_rttvar && i) { rt->rt_rmx.rmx_rttvar = (rt->rt_rmx.rmx_rttvar + i) / 2; - else + } else { rt->rt_rmx.rmx_rttvar = i; + } tcpstat.tcps_cachedrttvar++; + log_rtt = true; + } + if (log_rtt) { + TCP_LOG_RTM_RTT(tp, rt); + TCP_LOG_RTT_INFO(tp); } /* * The old comment here said: @@ -1312,35 +1502,32 @@ tcp_close(tp) * way to calculate the pipesize, it will have to do. */ i = tp->snd_ssthresh; - if (rt->rt_rmx.rmx_sendpipe != 0) + if (rt->rt_rmx.rmx_sendpipe != 0) { dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2); - else + } 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. */ i = (i + tp->t_maxseg / 2) / tp->t_maxseg; - if (i < 2) + if (i < 2) { 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 - ); - if (rt->rt_rmx.rmx_ssthresh) + isipv6 ? sizeof(struct ip6_hdr) + + sizeof(struct tcphdr) : + sizeof(struct tcpiphdr)); + if (rt->rt_rmx.rmx_ssthresh) { rt->rt_rmx.rmx_ssthresh = (rt->rt_rmx.rmx_ssthresh + i) / 2; - else + } else { rt->rt_rmx.rmx_ssthresh = i; + } tcpstat.tcps_cachedssthresh++; } } @@ -1348,7 +1535,7 @@ tcp_close(tp) /* * Mark route for deletion if no information is cached. */ - if (rt != NULL && (so->so_flags & SOF_OVERFLOW) && tcp_lq_overflow) { + if (rt != NULL && (so->so_flags & SOF_OVERFLOW)) { if (!(rt->rt_rmx.rmx_locks & RTV_RTT) && rt->rt_rmx.rmx_rtt == 0) { rt->rt_flags |= RTF_DELCLONE; @@ -1356,135 +1543,49 @@ tcp_close(tp) } no_valid_rt: - if (rt != NULL) + if (rt != NULL) { RT_UNLOCK(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) { - tcpstat.tcps_ecn_conn_recv_ce++; - 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) { - tcpstat.tcps_ecn_conn_pl_ce++; - 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) { - 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); - } + /* performance stats per interface */ + tcp_create_ifnet_stats_per_flow(tp, &ifs); + tcp_update_stats_per_flow(&ifs, inp->inp_last_outifp); 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); } tcp_rxtseg_clean(tp); /* Free the packet list */ - if (tp->t_pktlist_head != NULL) + if (tp->t_pktlist_head != NULL) { m_freem_list(tp->t_pktlist_head); + } TCP_PKTLIST_CLEAR(tp); -#if MPTCP - /* Clear MPTCP state */ - if ((so->so_flags & SOF_MPTCP_TRUE) || - (so->so_flags & SOF_MP_SUBFLOW)) { - soevent(so, (SO_FILT_HINT_LOCKED | SO_FILT_HINT_DELETEOK)); + if (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) { + inp->inp_saved_ppcb = (caddr_t) tp; } - tp->t_mpflags = 0; - tp->t_mptcb = NULL; -#endif /* MPTCP */ - - if (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) - inp->inp_saved_ppcb = (caddr_t) tp; 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 + /* + * Make sure to clear the TCP Keep Alive Offload as it is + * ref counted on the interface */ - 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; - } + tcp_clear_keep_alive_offload(so); /* * If this is a socket that does not want to wakeup the device @@ -1493,8 +1594,9 @@ no_valid_rt: */ if ((so->so_options & SO_NOWAKEFROMSLEEP) && inp->inp_state != INPCB_STATE_DEAD && - !(inp->inp_flags2 & INP2_TIMEWAIT)) + !(inp->inp_flags2 & INP2_TIMEWAIT)) { socket_post_kev_msg_closed(so); + } if (CC_ALGO(tp)->cleanup != NULL) { CC_ALGO(tp)->cleanup(tp); @@ -1516,77 +1618,67 @@ no_valid_rt: tp->t_tfo_flags &= ~TFO_F_COOKIE_VALID; } -#if INET6 - if (SOCK_CHECK_DOM(so, PF_INET6)) + if (SOCK_CHECK_DOM(so, PF_INET6)) { in6_pcbdetach(inp); - else -#endif /* INET6 */ - in_pcbdetach(inp); + } else { + 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); rv = 1; } tp->t_reassqlen = 0; - return (rv); + return rv; } -/* - * Walk the tcpbs, if existing, and flush the reassembly queue, - * if there is one when do_tcpdrain is enabled - * Also defunct the extended background idle socket - * 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) { if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) != - WNT_STOPUSING) { - tcp_lock(inp->inp_socket, 1, 0); + WNT_STOPUSING) { + socket_lock(inp->inp_socket, 1); if (in_pcb_checkstate(inp, WNT_RELEASE, 1) - == WNT_STOPUSING) { + == WNT_STOPUSING) { /* lost a race, try the next one */ - tcp_unlock(inp->inp_socket, 1, 0); + socket_unlock(inp->inp_socket, 1); continue; - } + } tp = intotcpcb(inp); - if (do_tcpdrain) - tcp_freeq(tp); - so_drain_extended_bk_idle(inp->inp_socket); - tcp_unlock(inp->inp_socket, 1, 0); + socket_unlock(inp->inp_socket, 1); } } lck_rw_done(tcbinfo.ipi_lock); - } /* @@ -1598,17 +1690,16 @@ 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; + VERIFY(tp != NULL); /* * Ignore some errors if we are hooked up. * If connection hasn't completed, has retransmitted several times, @@ -1617,45 +1708,68 @@ tcp_notify(inp, error) * can never complete. */ if (tp->t_state == TCPS_ESTABLISHED && - (error == EHOSTUNREACH || error == ENETUNREACH || - error == EHOSTDOWN)) { - return; + (error == EHOSTUNREACH || error == ENETUNREACH || + error == EHOSTDOWN)) { + if (inp->inp_route.ro_rt) { + rtfree(inp->inp_route.ro_rt); + inp->inp_route.ro_rt = (struct rtentry *)NULL; + } } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && - tp->t_softerror) + tp->t_softerror) { tcp_drop(tp, error); - else + } else { tp->t_softerror = error; -#if 0 - wakeup((caddr_t) &so->so_timeo); - sorwakeup(so); - sowwakeup(so); -#endif + } } -struct bwmeas* +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 = zalloc_flags(tcp_bwmeas_zone, Z_ZERO | Z_WAITOK); 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. @@ -1669,7 +1783,8 @@ 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; @@ -1724,8 +1839,7 @@ 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; @@ -1736,8 +1850,8 @@ 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) - + (n + n/8) * sizeof(struct xtcpcb); + req->oldidx = 2 * (sizeof(xig)) + + (n + n / 8) * sizeof(struct xtcpcb); lck_rw_done(tcbinfo.ipi_lock); return 0; } @@ -1754,11 +1868,11 @@ 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; @@ -1768,56 +1882,56 @@ tcp_pcblist SYSCTL_HANDLER_ARGS */ 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; - } - - 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; + } + socket_lock(inp->inp_socket, 1); + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { + socket_unlock(inp->inp_socket, 1); + continue; + } + if (inp->inp_gencnt > gencnt) { + socket_unlock(inp->inp_socket, 1); + 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); } + + socket_unlock(inp->inp_socket, 1); + + error = SYSCTL_OUT(req, &xt, sizeof(xt)); } if (!error) { /* @@ -1828,11 +1942,11 @@ 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); @@ -1840,66 +1954,67 @@ tcp_pcblist SYSCTL_HANDLER_ARGS } SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, - CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, - tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); + CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, + tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); +#if XNU_TARGET_OS_OSX 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 = 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; + 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; } @@ -1907,142 +2022,160 @@ 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) { - /* - * 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.ipi_lock); - return 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; + } + socket_lock(inp->inp_socket, 1); + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { + socket_unlock(inp->inp_socket, 1); + continue; + } + if (inp->inp_gencnt > gencnt) { + socket_unlock(inp->inp_socket, 1); + 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); + } + + socket_unlock(inp->inp_socket, 1); + + 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.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"); + CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, + tcp_pcblist64, "S,xtcpcb64", "List of active TCP connections"); +#endif /* XNU_TARGET_OS_OSX */ static int tcp_pcblist_n SYSCTL_HANDLER_ARGS { #pragma unused(oidp, arg1, arg2) int error = 0; - + error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo); - + return error; } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist_n, - CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, - tcp_pcblist_n, "S,xtcpcb_n", "List of active TCP connections"); + CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, + tcp_pcblist_n, "S,xtcpcb_n", "List of active TCP connections"); + +static int +tcp_progress_indicators SYSCTL_HANDLER_ARGS +{ +#pragma unused(oidp, arg1, arg2) + + return ntstat_tcp_progress_indicators(req); +} + +SYSCTL_PROC(_net_inet_tcp, OID_AUTO, progress, + CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, 0, 0, + tcp_progress_indicators, "S", "Various items that indicate the current state of progress on the link"); __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, - bitfield, &tcbinfo); + inpcb_get_ports_used(ifindex, protocol, flags, bitfield, + &tcbinfo); } __private_extern__ uint32_t @@ -2057,192 +2190,335 @@ tcp_find_anypcb_byaddr(struct ifaddr *ifa) return inpcb_find_anypcb_byaddr(ifa, &tcbinfo); } -void -tcp_ctlinput(cmd, sa, vip) - int cmd; - struct sockaddr *sa; - void *vip; +static void +tcp_handle_msgsize(struct ip *ip, struct inpcb *inp) { - tcp_seq icmp_tcp_seq; - struct ip *ip = vip; - struct in_addr faddr; - struct inpcb *inp; - struct tcpcb *tp; + struct rtentry *rt = NULL; + u_short ifscope = IFSCOPE_NONE; + int mtu; + struct sockaddr_in icmpsrc = { + .sin_len = sizeof(struct sockaddr_in), + .sin_family = AF_INET, .sin_port = 0, .sin_addr = { .s_addr = 0 }, + .sin_zero = { 0, 0, 0, 0, 0, 0, 0, 0 } + }; + struct icmp *icp = NULL; - void (*notify)(struct inpcb *, int) = tcp_notify; + icp = (struct icmp *)(void *) + ((caddr_t)ip - offsetof(struct icmp, icmp_ip)); - faddr = ((struct sockaddr_in *)(void *)sa)->sin_addr; - if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) - return; + icmpsrc.sin_addr = icp->icmp_ip.ip_dst; - if ((unsigned)cmd >= PRC_NCMDS) - return; + /* + * 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; + } - 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 (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 (inetctlerrmap[cmd] == 0) - return; - if (ip) { - struct tcphdr th; - struct icmp *icp; - - icp = (struct icmp *)(void *) - ((caddr_t)ip - offsetof(struct icmp, icmp_ip)); - bcopy(((caddr_t)ip + (IP_VHL_HL(ip->ip_vhl) << 2)), - &th, sizeof (th)); - inp = in_pcblookup_hash(&tcbinfo, faddr, th.th_dport, - ip->ip_src, th.th_sport, 0, NULL); - if (inp != NULL && inp->inp_socket != NULL) { - tcp_lock(inp->inp_socket, 1, 0); - if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { - tcp_unlock(inp->inp_socket, 1, 0); - return; - } - icmp_tcp_seq = htonl(th.th_seq); - 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); - } - } + /* + * 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; + } - (*notify)(inp, inetctlerrmap[cmd]); + 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; } - tcp_unlock(inp->inp_socket, 1, 0); } - } else + RT_UNLOCK(rt); + rtfree(rt); + } +} + +void +tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip, __unused struct ifnet *ifp) +{ + tcp_seq icmp_tcp_seq; + struct ip *ip = vip; + struct in_addr faddr; + struct inpcb *inp; + struct tcpcb *tp; + struct tcphdr *th; + struct icmp *icp; + void (*notify)(struct inpcb *, int) = tcp_notify; + + faddr = ((struct sockaddr_in *)(void *)sa)->sin_addr; + if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) { + return; + } + + if ((unsigned)cmd >= PRC_NCMDS) { + return; + } + + /* Source quench is deprecated */ + if (cmd == PRC_QUENCH) { + return; + } + + if (cmd == PRC_MSGSIZE) { + notify = tcp_mtudisc; + } else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || + cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || + cmd == PRC_TIMXCEED_INTRANS) && ip) { + notify = tcp_drop_syn_sent; + } + /* + * Hostdead is ugly because it goes linearly through all PCBs. + * XXX: We never get this from ICMP, otherwise it makes an + * excellent DoS attack on machines with many connections. + */ + else if (cmd == PRC_HOSTDEAD) { + ip = NULL; + } else if (inetctlerrmap[cmd] == 0 && !PRC_IS_REDIRECT(cmd)) { + return; + } + + + if (ip == NULL) { in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); + return; + } + + icp = (struct icmp *)(void *) + ((caddr_t)ip - offsetof(struct icmp, icmp_ip)); + th = (struct tcphdr *)(void *)((caddr_t)ip + (IP_VHL_HL(ip->ip_vhl) << 2)); + icmp_tcp_seq = ntohl(th->th_seq); + + inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, + ip->ip_src, th->th_sport, 0, NULL); + + if (inp == NULL || + inp->inp_socket == NULL) { + return; + } + + socket_lock(inp->inp_socket, 1); + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == + WNT_STOPUSING) { + socket_unlock(inp->inp_socket, 1); + return; + } + + if (PRC_IS_REDIRECT(cmd)) { + /* signal EHOSTDOWN, as it flushes the cached route */ + (*notify)(inp, EHOSTDOWN); + } else { + tp = intotcpcb(inp); + if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && + SEQ_LT(icmp_tcp_seq, tp->snd_max)) { + if (cmd == PRC_MSGSIZE) { + tcp_handle_msgsize(ip, inp); + } + + (*notify)(inp, inetctlerrmap[cmd]); + } + } + socket_unlock(inp->inp_socket, 1); } -#if INET6 void -tcp6_ctlinput(cmd, sa, d) - int cmd; - struct sockaddr *sa; - void *d; +tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d, __unused struct ifnet *ifp) { - struct tcphdr th; + tcp_seq icmp_tcp_seq; + struct in6_addr *dst; void (*notify)(struct inpcb *, int) = tcp_notify; struct ip6_hdr *ip6; struct mbuf *m; + struct inpcb *inp; + struct tcpcb *tp; + struct icmp6_hdr *icmp6; struct ip6ctlparam *ip6cp = NULL; const struct sockaddr_in6 *sa6_src = NULL; - int off; - struct tcp_portonly { - u_int16_t th_sport; - u_int16_t th_dport; - } *thp; + unsigned int mtu; + unsigned int off; + + struct tcp_ports { + uint16_t th_sport; + uint16_t th_dport; + } t_ports; if (sa->sa_family != AF_INET6 || - sa->sa_len != sizeof(struct sockaddr_in6)) + sa->sa_len != sizeof(struct sockaddr_in6)) { return; + } - if ((unsigned)cmd >= PRC_NCMDS) + /* Source quench is deprecated */ + if (cmd == PRC_QUENCH) { return; + } - if (cmd == PRC_MSGSIZE) - notify = tcp_mtudisc; - else if (!PRC_IS_REDIRECT(cmd) && (inet6ctlerrmap[cmd] == 0)) - return; - /* Source quench is deprecated */ - else if (cmd == PRC_QUENCH) + if ((unsigned)cmd >= PRC_NCMDS) { return; + } /* if the parameter is from icmp6, decode it. */ if (d != NULL) { ip6cp = (struct ip6ctlparam *)d; + icmp6 = ip6cp->ip6c_icmp6; m = ip6cp->ip6c_m; ip6 = ip6cp->ip6c_ip6; off = ip6cp->ip6c_off; sa6_src = ip6cp->ip6c_src; + dst = ip6cp->ip6c_finaldst; } else { m = NULL; ip6 = NULL; - off = 0; /* fool gcc */ + off = 0; /* fool gcc */ sa6_src = &sa6_any; + dst = NULL; + } + + if (cmd == PRC_MSGSIZE) { + notify = tcp_mtudisc; + } else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || + cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && + ip6 != NULL) { + notify = tcp_drop_syn_sent; + } + /* + * Hostdead is ugly because it goes linearly through all PCBs. + * XXX: We never get this from ICMP, otherwise it makes an + * excellent DoS attack on machines with many connections. + */ + else if (cmd == PRC_HOSTDEAD) { + ip6 = NULL; + } else if (inet6ctlerrmap[cmd] == 0 && !PRC_IS_REDIRECT(cmd)) { + return; + } + + + if (ip6 == NULL) { + in6_pcbnotify(&tcbinfo, sa, 0, (struct sockaddr *)(size_t)sa6_src, + 0, cmd, NULL, notify); + return; + } + + /* Check if we can safely get the ports from the tcp hdr */ + if (m == NULL || + (m->m_pkthdr.len < + (int32_t) (off + sizeof(struct tcp_ports)))) { + return; + } + bzero(&t_ports, sizeof(struct tcp_ports)); + m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); + + off += sizeof(struct tcp_ports); + if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { + return; } + m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); + icmp_tcp_seq = ntohl(icmp_tcp_seq); - if (ip6) { + if (cmd == PRC_MSGSIZE) { + mtu = ntohl(icmp6->icmp6_mtu); /* - * XXX: We assume that when IPV6 is non NULL, - * M and OFF are valid. + * If no alternative MTU was proposed, or the proposed + * MTU was too small, set to the min. */ + if (mtu < IPV6_MMTU) { + mtu = IPV6_MMTU - 8; + } + } - /* check if we can safely examine src and dst ports */ - if (m->m_pkthdr.len < off + sizeof(*thp)) - return; + inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_dst, t_ports.th_dport, + &ip6->ip6_src, t_ports.th_sport, 0, NULL); + + if (inp == NULL || + inp->inp_socket == NULL) { + return; + } - bzero(&th, sizeof(th)); - m_copydata(m, off, sizeof(*thp), (caddr_t)&th); + socket_lock(inp->inp_socket, 1); + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == + WNT_STOPUSING) { + socket_unlock(inp->inp_socket, 1); + return; + } - in6_pcbnotify(&tcbinfo, sa, th.th_dport, - (struct sockaddr *)ip6cp->ip6c_src, - th.th_sport, cmd, NULL, notify); + if (PRC_IS_REDIRECT(cmd)) { + /* signal EHOSTDOWN, as it flushes the cached route */ + (*notify)(inp, EHOSTDOWN); } else { - in6_pcbnotify(&tcbinfo, sa, 0, - (struct sockaddr *)(size_t)sa6_src, 0, cmd, NULL, notify); + tp = intotcpcb(inp); + if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && + SEQ_LT(icmp_tcp_seq, tp->snd_max)) { + if (cmd == PRC_MSGSIZE) { + /* + * Only process the offered MTU if it + * is smaller than the current one. + */ + if (mtu < tp->t_maxseg + + (sizeof(struct tcphdr) + sizeof(struct ip6_hdr))) { + (*notify)(inp, inetctlerrmap[cmd]); + } + } else { + (*notify)(inp, inetctlerrmap[cmd]); + } + } } + socket_unlock(inp->inp_socket, 1); } -#endif /* INET6 */ /* @@ -2252,7 +2528,7 @@ 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() @@ -2291,56 +2567,54 @@ tcp6_ctlinput(cmd, sa, d) #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; struct timeval timenow; u_char isn_secret[32]; - int isn_last_reseed = 0; + long isn_last_reseed = 0; 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)timenow.tv_sec))) { + (((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)); + read_frandom(&isn_secret, sizeof(isn_secret)); #else read_random_unlimited(&isn_secret, sizeof(isn_secret)); #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)); -#if INET6 + 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 ((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)); - } else -#endif - { + sizeof(struct in6_addr)); + } else { 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); @@ -2356,14 +2630,13 @@ 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); - if (tp && tp->t_state == TCPS_SYN_SENT) + if (tp && tp->t_state == TCPS_SYN_SENT) { tcp_drop(tp, errno); + } } /* @@ -2373,43 +2646,45 @@ tcp_drop_syn_sent(inp, errno) * This duplicates some code in the tcp_mss() function in tcp_input.c. */ void -tcp_mtudisc( - struct inpcb *inp, - __unused int errno -) +tcp_mtudisc(struct inpcb *inp, __unused int errno) { struct tcpcb *tp = intotcpcb(inp); struct rtentry *rt; - struct rmxp_tao *taop; struct socket *so = inp->inp_socket; - int offered; int mss; u_int32_t mtu; -#if INET6 + u_int32_t protoHdrOverhead = sizeof(struct tcpiphdr); int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; -#endif /* INET6 */ - if (tp) { -#if INET6 - if (isipv6) + /* + * Nothing left to send after the socket is defunct or TCP is in the closed state + */ + if ((so->so_state & SS_DEFUNCT) || (tp != NULL && tp->t_state == TCPS_CLOSED)) { + return; + } + + if (isipv6) { + protoHdrOverhead = sizeof(struct ip6_hdr) + + sizeof(struct tcphdr); + } + + if (tp != NULL) { + if (isipv6) { rt = tcp_rtlookup6(inp, IFSCOPE_NONE); - else -#endif /* INET6 */ - rt = tcp_rtlookup(inp, IFSCOPE_NONE); + } else { + 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; + isipv6 ? tcp_v6mssdflt : + tcp_mssdflt; /* Route locked during lookup above */ - if (rt != NULL) + if (rt != NULL) { RT_UNLOCK(rt); + } return; } - taop = rmx_taop(rt->rt_rmx); - offered = taop->tao_mssopt; mtu = rt->rt_rmx.rmx_mtu; /* Route locked during lookup above */ @@ -2419,20 +2694,11 @@ tcp_mtudisc( // Adjust MTU if necessary. mtu = necp_socket_get_effective_mtu(inp, mtu); #endif /* NECP */ + mss = mtu - protoHdrOverhead; - mss = mtu - -#if INET6 - (isipv6 ? - sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : -#endif /* INET6 */ - sizeof(struct tcpiphdr) -#if INET6 - ) -#endif /* INET6 */ - ; - - if (offered) - mss = min(mss, offered); + if (tp->t_maxopd) { + mss = min(mss, tp->t_maxopd); + } /* * XXX - The above conditional probably violates the TCP * spec. The problem is that, since we don't know the @@ -2450,27 +2716,34 @@ tcp_mtudisc( * will get recorded and the new parameters should get * recomputed. For Further Study. */ - if (tp->t_maxopd <= mss) + if (tp->t_maxopd <= mss) { return; + } tp->t_maxopd = mss; - if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && - (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) + if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP && + (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) { mss -= TCPOLEN_TSTAMP_APPA; + } #if MPTCP mss -= mptcp_adj_mss(tp, TRUE); #endif - if (so->so_snd.sb_hiwat < mss) + if (so->so_snd.sb_hiwat < mss) { mss = so->so_snd.sb_hiwat; + } tp->t_maxseg = mss; + ASSERT(tp->t_maxseg); + /* - * 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) + 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; @@ -2486,19 +2759,18 @@ 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; struct tcpcb *tp; - lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); + LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); ro = &inp->inp_route; - if ((rt = ro->ro_rt) != NULL) + if ((rt = ro->ro_rt) != NULL) { RT_LOCK(rt); + } if (ROUTE_UNUSABLE(ro)) { if (rt != NULL) { @@ -2513,7 +2785,7 @@ tcp_rtlookup(inp, input_ifscope) ro->ro_dst.sa_family = AF_INET; ro->ro_dst.sa_len = sizeof(struct sockaddr_in); ((struct sockaddr_in *)(void *)&ro->ro_dst)->sin_addr = - inp->inp_faddr; + inp->inp_faddr; /* * If the socket was bound to an interface, then @@ -2526,12 +2798,14 @@ tcp_rtlookup(inp, input_ifscope) inp->inp_boundifp->if_index : input_ifscope; rtalloc_scoped(ro, ifscope); - if ((rt = ro->ro_rt) != NULL) + if ((rt = ro->ro_rt) != NULL) { RT_LOCK(rt); + } } } - if (rt != NULL) + if (rt != NULL) { RT_LOCK_ASSERT_HELD(rt); + } /* * Update MTU discovery determination. Don't do it if: @@ -2541,22 +2815,14 @@ 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 + } 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, @@ -2565,37 +2831,38 @@ tcp_rtlookup(inp, input_ifscope) 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 */ if (rt != NULL && !(rt->rt_ifp->if_flags & IFF_POINTOPOINT) && - (rt->rt_gateway->sa_family == AF_LINK || - rt->rt_ifp->if_flags & IFF_LOOPBACK || - in_localaddr(inp->inp_faddr))) { + (rt->rt_gateway->sa_family == AF_LINK || + rt->rt_ifp->if_flags & IFF_LOOPBACK || + in_localaddr(inp->inp_faddr))) { tp->t_flags |= TF_LOCAL; } - + /* * Caller needs to call RT_UNLOCK(rt). */ 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; struct tcpcb *tp; - lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); + LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); ro6 = &inp->in6p_route; - if ((rt = ro6->ro_rt) != NULL) + if ((rt = ro6->ro_rt) != NULL) { RT_LOCK(rt); + } if (ROUTE_UNUSABLE(ro6)) { if (rt != NULL) { @@ -2624,12 +2891,14 @@ tcp_rtlookup6(inp, input_ifscope) inp->inp_boundifp->if_index : input_ifscope; rtalloc_scoped((struct route *)ro6, ifscope); - if ((rt = ro6->ro_rt) != NULL) + if ((rt = ro6->ro_rt) != NULL) { RT_LOCK(rt); + } } } - if (rt != NULL) + if (rt != NULL) { RT_LOCK_ASSERT_HELD(rt); + } /* * Update path MTU Discovery determination @@ -2640,7 +2909,7 @@ tcp_rtlookup6(inp, input_ifscope) */ - tp = intotcpcb(inp); + tp = intotcpcb(inp); /* * Update MTU discovery determination. Don't do it if: @@ -2650,20 +2919,12 @@ 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 + } 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, @@ -2672,15 +2933,18 @@ tcp_rtlookup6(inp, input_ifscope) 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 */ - if (rt != NULL && !(rt->rt_ifp->if_flags & IFF_POINTOPOINT) && - (IN6_IS_ADDR_LOOPBACK(&inp->in6p_faddr) || - IN6_IS_ADDR_LINKLOCAL(&inp->in6p_faddr) || - rt->rt_gateway->sa_family == AF_LINK || - in6_localaddr(&inp->in6p_faddr))) { - tp->t_flags |= TF_LOCAL; + /* Note if the peer is local */ + if (!(rt->rt_ifp->if_flags & IFF_POINTOPOINT) && + (IN6_IS_ADDR_LOOPBACK(&inp->in6p_faddr) || + IN6_IS_ADDR_LINKLOCAL(&inp->in6p_faddr) || + rt->rt_gateway->sa_family == AF_LINK || + in6_localaddr(&inp->in6p_faddr))) { + tp->t_flags |= TF_LOCAL; + } } /* @@ -2688,126 +2952,103 @@ tcp_rtlookup6(inp, input_ifscope) */ 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; size_t hdrsiz; struct ip *ip; -#if INET6 struct ip6_hdr *ip6 = NULL; -#endif /* INET6 */ struct tcphdr *th; - if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) + if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) { return 0; - MGETHDR(m, M_DONTWAIT, MT_DATA); /* MAC-OK */ - if (!m) + } + MGETHDR(m, M_DONTWAIT, MT_DATA); /* MAC-OK */ + if (!m) { return 0; + } -#if INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)(void *)(ip6 + 1); m->m_pkthdr.len = m->m_len = - sizeof(struct ip6_hdr) + sizeof(struct tcphdr); + sizeof(struct ip6_hdr) + sizeof(struct tcphdr); tcp_fillheaders(tp, ip6, th); 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); - } + } else { + 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; } -#endif /*IPSEC*/ - -/* - * Return a pointer to the cached information about the remote host. - * The cached information is stored in the protocol specific part of - * the route metrics. - */ -struct rmxp_tao * -tcp_gettaocache(inp) - struct inpcb *inp; -{ - struct rtentry *rt; - struct rmxp_tao *taop; - -#if INET6 - if ((inp->inp_vflag & INP_IPV6) != 0) - rt = tcp_rtlookup6(inp, IFSCOPE_NONE); - else -#endif /* INET6 */ - 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)) { - /* Route locked during lookup above */ - if (rt != NULL) - RT_UNLOCK(rt); - return NULL; - } - - taop = rmx_taop(rt->rt_rmx); - /* Route locked during lookup above */ - RT_UNLOCK(rt); - return (taop); -} - -/* - * Clear all the TAO cache entries, called from tcp_init. - * - * XXX - * This routine is just an empty one, because we assume that the routing - * routing tables are initialized at the same time when TCP, so there is - * nothing in the cache left over. - */ -static void -tcp_cleartaocache() -{ -} +#endif /* IPSEC */ int tcp_lock(struct socket *so, int refcount, void *lr) { void *lr_saved; - if (lr == NULL) + if (lr == NULL) { lr_saved = __builtin_return_address(0); - else + } else { lr_saved = lr; + } +retry: 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", + if (so->so_flags & SOF_MP_SUBFLOW) { + struct mptcb *mp_tp = tptomptp(sototcpcb(so)); + struct socket *mp_so = mptetoso(mp_tp->mpt_mpte); + + socket_lock(mp_so, refcount); + + /* + * Check if we became non-MPTCP while waiting for the lock. + * If yes, we have to retry to grab the right lock. + */ + if (!(so->so_flags & SOF_MP_SUBFLOW)) { + socket_unlock(mp_so, refcount); + goto retry; + } + } else { + lck_mtx_lock(&((struct inpcb *)so->so_pcb)->inpcb_mtx); + + if (so->so_flags & SOF_MP_SUBFLOW) { + /* + * While waiting for the lock, we might have + * become MPTCP-enabled (see mptcp_subflow_socreate). + */ + lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx); + goto retry; + } + } + } 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=%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) + if (refcount) { so->so_usecount++; + } so->lock_lr[so->next_lock_lr] = lr_saved; - so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; - return (0); + so->next_lock_lr = (so->next_lock_lr + 1) % SO_LCKDBG_MAX; + return 0; } int @@ -2815,10 +3056,11 @@ tcp_unlock(struct socket *so, int refcount, void *lr) { void *lr_saved; - if (lr == NULL) + if (lr == NULL) { lr_saved = __builtin_return_address(0); - else + } else { lr_saved = lr; + } #ifdef MORE_TCPLOCK_DEBUG printf("tcp_unlock: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x " @@ -2827,51 +3069,68 @@ tcp_unlock(struct socket *so, int refcount, void *lr) (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)), so->so_usecount, (uint64_t)VM_KERNEL_ADDRPERM(lr_saved)); #endif - if (refcount) + if (refcount) { 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 { - 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); + so->next_unlock_lr = (so->next_unlock_lr + 1) % SO_LCKDBG_MAX; + + if (so->so_flags & SOF_MP_SUBFLOW) { + struct mptcb *mp_tp = tptomptp(sototcpcb(so)); + struct socket *mp_so = mptetoso(mp_tp->mpt_mpte); + + socket_lock_assert_owned(mp_so); + + socket_unlock(mp_so, refcount); + } else { + LCK_MTX_ASSERT(&((struct inpcb *)so->so_pcb)->inpcb_mtx, + LCK_MTX_ASSERT_OWNED); + lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx); + } } - return (0); + return 0; } lck_mtx_t * -tcp_getlock( - struct socket *so, - __unused int locktype) +tcp_getlock(struct socket *so, int flags) { 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", + 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)); + } + + if (so->so_flags & SOF_MP_SUBFLOW) { + struct mptcb *mp_tp = tptomptp(sototcpcb(so)); + struct socket *mp_so = mptetoso(mp_tp->mpt_mpte); + + return mp_so->so_proto->pr_getlock(mp_so, flags); + } else { + 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); + 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. @@ -2883,21 +3142,18 @@ 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 message delivery is enabled, do not count - * unordered bytes in receive buffer towards hiwat - */ - if (so->so_flags & SOF_ENABLE_MSGS) - rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes; + if (tcp_recv_bg == 1 || IS_TCP_RECV_BG(so)) { + return; + } if (tcp_do_autorcvbuf == 1 && - tcp_cansbgrow(sb) && - (tp->t_flags & TF_SLOWLINK) == 0 && - (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) == 0 && - (rcvbuf - sb->sb_cc) < rcvbufinc && - rcvbuf < tcp_autorcvbuf_max && - (sb->sb_idealsize > 0 && - sb->sb_hiwat <= (sb->sb_idealsize + rcvbufinc))) { + tcp_cansbgrow(sb) && + (tp->t_flags & TF_SLOWLINK) == 0 && + (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) == 0 && + (rcvbuf - sb->sb_cc) < rcvbufinc && + rcvbuf < tcp_autorcvbuf_max && + (sb->sb_idealsize > 0 && + sb->sb_hiwat <= (sb->sb_idealsize + rcvbufinc))) { sbreserve(sb, min((sb->sb_hiwat + rcvbufinc), tcp_autorcvbuf_max)); } @@ -2906,50 +3162,55 @@ 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; - /* - * If message delivery is enabled, do not count - * unordered bytes in receive buffer towards hiwat mark. - * This value is used to return correct rwnd that does - * not reflect the extra unordered bytes added to the - * receive socket buffer. - */ - if (so->so_flags & SOF_ENABLE_MSGS) - rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes; + if (so->so_flags & SOF_MP_SUBFLOW) { + /* We still need to grow TCP's buffer to have a BDP-estimate */ + tcp_sbrcv_grow_rwin(tp, sb); + + return mptcp_sbspace(tptomptp(tp)); + } tcp_sbrcv_grow_rwin(tp, sb); + /* hiwat might have changed */ + rcvbuf = sb->sb_hiwat; + space = ((int32_t) imin((rcvbuf - sb->sb_cc), - (sb->sb_mbmax - sb->sb_mbcnt))); - if (space < 0) + (sb->sb_mbmax - sb->sb_mbcnt))); + if (space < 0) { space = 0; + } #if CONTENT_FILTER /* Compensate for data being processed by content filters */ pending = cfil_sock_data_space(sb); #endif /* CONTENT_FILTER */ - if (pending > space) + if (pending > space) { space = 0; - else + } 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) + 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 ) + if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0) { return imin(space, slowlink_wsize); + } return space; } @@ -2960,63 +3221,130 @@ tcp_sbspace(struct tcpcb *tp) void tcp_set_tso(struct tcpcb *tp, struct ifnet *ifp) { -#if INET6 struct inpcb *inp; int isipv6; -#endif /* INET6 */ + struct ifnet *tunnel_ifp = NULL; +#define IFNET_TSO_MASK (IFNET_TSO_IPV6 | IFNET_TSO_IPV4) + + tp->t_flags &= ~TF_TSO; + + if (ifp == NULL) { + return; + } + #if MPTCP /* * We can't use TSO if this tcpcb belongs to an MPTCP session. */ if (tp->t_mpflags & TMPF_MPTCP_TRUE) { - tp->t_flags &= ~TF_TSO; return; } #endif -#if INET6 inp = tp->t_inpcb; 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; - - } else -#endif /* INET6 */ + /* + * We can't use TSO if the TSO capability of the tunnel interface does + * not match the capability of another interface known by TCP + */ + if (inp->inp_policyresult.results.result == NECP_KERNEL_POLICY_RESULT_IP_TUNNEL) { + u_int tunnel_if_index = inp->inp_policyresult.results.result_parameter.tunnel_interface_index; - { - 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; - } -} + if (tunnel_if_index != 0) { + ifnet_head_lock_shared(); + tunnel_ifp = ifindex2ifnet[tunnel_if_index]; + ifnet_head_done(); + } -#define TIMEVAL_TO_TCPHZ(_tv_) ((_tv_).tv_sec * TCP_RETRANSHZ + (_tv_).tv_usec / TCP_RETRANSHZ_TO_USEC) + if (tunnel_ifp == NULL) { + return; + } -/* Function to calculate the tcp clock. The tcp clock will get updated + if ((ifp->if_hwassist & IFNET_TSO_MASK) != (tunnel_ifp->if_hwassist & IFNET_TSO_MASK)) { + if (tso_debug > 0) { + os_log(OS_LOG_DEFAULT, + "%s: %u > %u TSO 0 tunnel_ifp %s hwassist mismatch with ifp %s", + __func__, + ntohs(tp->t_inpcb->inp_lport), ntohs(tp->t_inpcb->inp_fport), + tunnel_ifp->if_xname, ifp->if_xname); + } + return; + } + if (inp->inp_last_outifp != NULL && + (inp->inp_last_outifp->if_hwassist & IFNET_TSO_MASK) != (tunnel_ifp->if_hwassist & IFNET_TSO_MASK)) { + if (tso_debug > 0) { + os_log(OS_LOG_DEFAULT, + "%s: %u > %u TSO 0 tunnel_ifp %s hwassist mismatch with inp_last_outifp %s", + __func__, + ntohs(tp->t_inpcb->inp_lport), ntohs(tp->t_inpcb->inp_fport), + tunnel_ifp->if_xname, inp->inp_last_outifp->if_xname); + } + return; + } + if ((inp->inp_flags & INP_BOUND_IF) && inp->inp_boundifp != NULL && + (inp->inp_boundifp->if_hwassist & IFNET_TSO_MASK) != (tunnel_ifp->if_hwassist & IFNET_TSO_MASK)) { + if (tso_debug > 0) { + os_log(OS_LOG_DEFAULT, + "%s: %u > %u TSO 0 tunnel_ifp %s hwassist mismatch with inp_boundifp %s", + __func__, + ntohs(tp->t_inpcb->inp_lport), ntohs(tp->t_inpcb->inp_fport), + tunnel_ifp->if_xname, inp->inp_boundifp->if_xname); + } + return; + } + } + + if (isipv6) { + if (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 { + if (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; + } + if (INTF_ADJUST_MTU_FOR_CLAT46(ifp)) { + tp->tso_max_segment_size -= + CLAT46_HDR_EXPANSION_OVERHD; + } + } + } + + if (tso_debug > 1) { + os_log(OS_LOG_DEFAULT, "%s: %u > %u TSO %d ifp %s", + __func__, + ntohs(tp->t_inpcb->inp_lport), + ntohs(tp->t_inpcb->inp_fport), + (tp->t_flags & TF_TSO) != 0, + ifp != NULL ? ifp->if_xname : ""); + } +} + +#define TIMEVAL_TO_TCPHZ(_tv_) ((uint32_t)((_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 + * 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}; + struct timeval interval = {.tv_sec = 0, .tv_usec = TCP_RETRANSHZ_TO_USEC}; struct timeval now, hold_now; uint32_t incr = 0; @@ -3027,7 +3355,7 @@ calculate_tcp_clock() * is to update the counter returnable via net_uptime() when * we read time. */ - net_update_uptime_secs(now.tv_sec); + net_update_uptime_with_time(&now); timevaladd(&tv, &interval); if (timevalcmp(&now, &tv, >)) { @@ -3050,82 +3378,95 @@ 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) { - u_int32_t maxsockbufsize; - if (!tcp_do_rfc1323) { - tp->request_r_scale = 0; - return; - } +tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so) +{ + uint32_t maxsockbufsize; - tp->request_r_scale = max(tcp_win_scale, tp->request_r_scale); + tp->request_r_scale = MAX((uint8_t)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 : tcp_autorcvbuf_max; + /* + * Window scale should not exceed what is needed + * to send the max receive window size; adding 1 to TCP_MAXWIN + * ensures that. + */ while (tp->request_r_scale < TCP_MAX_WINSHIFT && - (TCP_MAXWIN << tp->request_r_scale) < maxsockbufsize) + ((TCP_MAXWIN + 1) << tp->request_r_scale) < maxsockbufsize) { tp->request_r_scale++; - tp->request_r_scale = min(tp->request_r_scale, TCP_MAX_WINSHIFT); - + } + tp->request_r_scale = MIN(tp->request_r_scale, TCP_MAX_WINSHIFT); } int -tcp_notsent_lowat_check(struct socket *so) { +tcp_notsent_lowat_check(struct socket *so) +{ struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp = NULL; int notsent = 0; + if (inp != NULL) { tp = intotcpcb(inp); } - notsent = so->so_snd.sb_cc - - (tp->snd_nxt - tp->snd_una); + if (tp == NULL) { + return 0; + } + + 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 && - notsent > 0 && notsent < tp->t_maxseg) { - return(1); + if ((tp->t_flags & TF_NODELAY) == 0 && + notsent > 0 && notsent < tp->t_maxseg) { + 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; + uint16_t rxcount = 0; - if (SLIST_EMPTY(&tp->t_rxt_segments)) + if (SLIST_EMPTY(&tp->t_rxt_segments)) { tp->t_dsack_lastuna = tp->snd_una; + } /* * First check if there is a segment already existing for this * sequence space. */ SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) { - if (SEQ_GT(rxseg->rx_start, start)) + if (SEQ_GT(rxseg->rx_start, start)) { break; + } prev = rxseg; } next = rxseg; @@ -3163,8 +3504,9 @@ tcp_rxtseg_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end) { rxcount = next->rx_count; } } - if (!SEQ_LT(start, end)) + if (!SEQ_LT(start, end)) { return; + } rxseg = (struct tcp_rxt_seg *) zalloc(tcp_rxt_seg_zone); if (rxseg == NULL) { @@ -3180,24 +3522,53 @@ 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 * tcp_rxtseg_find(struct tcpcb *tp, tcp_seq start, tcp_seq end) { struct tcp_rxt_seg *rxseg; - if (SLIST_EMPTY(&tp->t_rxt_segments)) - return (NULL); + if (SLIST_EMPTY(&tp->t_rxt_segments)) { + return NULL; + } SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) { if (SEQ_LEQ(rxseg->rx_start, start) && - SEQ_GEQ(rxseg->rx_end, end)) - return (rxseg); - if (SEQ_GT(rxseg->rx_start, start)) + SEQ_GEQ(rxseg->rx_end, end)) { + return rxseg; + } + if (SEQ_GT(rxseg->rx_start, start)) { + break; + } + } + return NULL; +} + +void +tcp_rxtseg_set_spurious(struct tcpcb *tp, tcp_seq start, tcp_seq end) +{ + struct tcp_rxt_seg *rxseg; + if (SLIST_EMPTY(&tp->t_rxt_segments)) { + return; + } + + SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) { + if (SEQ_GEQ(rxseg->rx_start, start) && + SEQ_LEQ(rxseg->rx_end, end)) { + /* + * If the segment was retransmitted only once, mark it as + * spurious. + */ + if (rxseg->rx_count == 1) { + rxseg->rx_flags |= TCP_RXT_SPURIOUS; + } + } + + if (SEQ_GEQ(rxseg->rx_start, end)) { break; + } } - return (NULL); + return; } void @@ -3219,26 +3590,27 @@ tcp_rxtseg_detect_bad_rexmt(struct tcpcb *tp, tcp_seq th_ack) boolean_t bad_rexmt; struct tcp_rxt_seg *rxseg; - if (SLIST_EMPTY(&tp->t_rxt_segments)) - return (FALSE); + if (SLIST_EMPTY(&tp->t_rxt_segments)) { + return FALSE; + } /* * If all of the segments in this window are not cumulatively * acknowledged, then there can still be undetected packet loss. * Do not restore congestion window in that case. */ - if (SEQ_LT(th_ack, tp->snd_recover)) - return (FALSE); + if (SEQ_LT(th_ack, tp->snd_recover)) { + return FALSE; + } bad_rexmt = TRUE; SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) { - if (rxseg->rx_count > 1 || - !(rxseg->rx_flags & TCP_RXT_SPURIOUS)) { + if (!(rxseg->rx_flags & TCP_RXT_SPURIOUS)) { bad_rexmt = FALSE; break; } } - return (bad_rexmt); + return bad_rexmt; } boolean_t @@ -3246,37 +3618,40 @@ tcp_rxtseg_dsack_for_tlp(struct tcpcb *tp) { boolean_t dsack_for_tlp = FALSE; struct tcp_rxt_seg *rxseg; - if (SLIST_EMPTY(&tp->t_rxt_segments)) - return (FALSE); + if (SLIST_EMPTY(&tp->t_rxt_segments)) { + return FALSE; + } 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; } } - return (dsack_for_tlp); + return dsack_for_tlp; } 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; SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) { total_size += (rxseg->rx_end - rxseg->rx_start) + 1; } - return (total_size); + return total_size; } void tcp_get_connectivity_status(struct tcpcb *tp, - struct tcp_conn_status *connstatus) + struct tcp_conn_status *connstatus) { - if (tp == NULL || connstatus == NULL) + if (tp == NULL || connstatus == NULL) { return; + } bzero(connstatus, sizeof(*connstatus)); if (tp->t_rxtshift >= TCP_CONNECTIVITY_PROBES_MAX) { if (TCPS_HAVEESTABLISHED(tp->t_state)) { @@ -3285,18 +3660,19 @@ tcp_get_connectivity_status(struct tcpcb *tp, connstatus->conn_probe_failed = 1; } } - if (tp->t_rtimo_probes >= TCP_CONNECTIVITY_PROBES_MAX) + 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 @@ -3305,3 +3681,686 @@ 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 = rfc6864 ? 0 : ip_randomid(); + ip->ip_off = htons(IP_DF); + 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; + } + socket_lock(so, 1); + /* Release the want count */ + if (inp->inp_ppcb == NULL || + (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING)) { + socket_unlock(so, 1); + continue; + } + if ((inp->inp_vflag & INP_IPV4) && + (inp->inp_laddr.s_addr == INADDR_ANY || + inp->inp_faddr.s_addr == INADDR_ANY)) { + socket_unlock(so, 1); + continue; + } + if ((inp->inp_vflag & INP_IPV6) && + (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) || + IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))) { + socket_unlock(so, 1); + continue; + } + if (inp->inp_lport == 0 || inp->inp_fport == 0) { + socket_unlock(so, 1); + continue; + } + if (inp->inp_last_outifp == NULL || + inp->inp_last_outifp->if_index != ifp->if_index) { + socket_unlock(so, 1); + continue; + } + if ((inp->inp_vflag & INP_IPV4) && frame_data_offset + + sizeof(struct ip) + sizeof(struct tcphdr) > + IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) { + socket_unlock(so, 1); + continue; + } else if (!(inp->inp_vflag & INP_IPV4) && frame_data_offset + + sizeof(struct ip6_hdr) + sizeof(struct tcphdr) > + IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) { + socket_unlock(so, 1); + 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) { + socket_unlock(so, 1); + 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 = (uint16_t)(tp->t_keepidle > 0 ? tp->t_keepidle : + tcp_keepidle); + frame->keep_cnt = (uint8_t)TCP_CONN_KEEPCNT(tp); + frame->keep_retry = (uint16_t)TCP_CONN_KEEPINTVL(tp); + if (so->so_options & SO_NOWAKEFROMSLEEP) { + frame->flags |= + IFNET_KEEPALIVE_OFFLOAD_FLAG_NOWAKEFROMSLEEP; + } + 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) { + ASSERT(frame_data_offset + sizeof(struct ip) + sizeof(struct tcphdr) <= UINT8_MAX); + frame->length = (uint8_t)(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; + + ASSERT(frame_data_offset + sizeof(struct ip6_hdr) + sizeof(struct tcphdr) <= UINT8_MAX); + frame->length = (uint8_t)(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) { + socket_unlock(so, 1); + 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) { + socket_unlock(so, 1); + continue; + } + bcopy(m->m_data, frame->reply_data + frame_data_offset, + m->m_len); + m_freem(m); + + frame_index++; + socket_unlock(so, 1); + } + lck_rw_done(tcbinfo.ipi_lock); + *used_frames_count = frame_index; +} + +static bool +inp_matches_kao_frame(ifnet_t ifp, struct ifnet_keepalive_offload_frame *frame, + struct inpcb *inp) +{ + if (inp->inp_ppcb == NULL) { + return false; + } + /* Release the want count */ + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { + return false; + } + if (inp->inp_last_outifp == NULL || + inp->inp_last_outifp->if_index != ifp->if_index) { + return false; + } + if (frame->local_port != ntohs(inp->inp_lport) || + frame->remote_port != ntohs(inp->inp_fport)) { + return false; + } + if (inp->inp_vflag & INP_IPV4) { + if (memcmp(&inp->inp_laddr, frame->local_addr, + sizeof(struct in_addr)) != 0 || + memcmp(&inp->inp_faddr, frame->remote_addr, + sizeof(struct in_addr)) != 0) { + return false; + } + } else if (inp->inp_vflag & INP_IPV6) { + if (memcmp(&inp->inp_laddr, frame->local_addr, + sizeof(struct in6_addr)) != 0 || + memcmp(&inp->inp_faddr, frame->remote_addr, + sizeof(struct in6_addr)) != 0) { + return false; + } + } else { + return false; + } + return true; +} + +int +tcp_notify_kao_timeout(ifnet_t ifp, + struct ifnet_keepalive_offload_frame *frame) +{ + struct inpcb *inp = NULL; + struct socket *so = NULL; + bool found = false; + + /* + * Unlock the list before posting event on the matching socket + */ + lck_rw_lock_shared(tcbinfo.ipi_lock); + + LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) { + if ((so = inp->inp_socket) == NULL || + (so->so_state & SS_DEFUNCT)) { + continue; + } + 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; + } + socket_lock(so, 1); + if (inp_matches_kao_frame(ifp, frame, inp)) { + /* + * Keep the matching socket locked + */ + found = true; + break; + } + socket_unlock(so, 1); + } + lck_rw_done(tcbinfo.ipi_lock); + + if (found) { + ASSERT(inp != NULL); + ASSERT(so != NULL); + ASSERT(so == inp->inp_socket); + /* + * Drop the TCP connection like tcptimers() does + */ + struct tcpcb *tp = inp->inp_ppcb; + + tcpstat.tcps_keepdrops++; + soevent(so, + (SO_FILT_HINT_LOCKED | SO_FILT_HINT_TIMEOUT)); + tp = tcp_drop(tp, ETIMEDOUT); + + tcpstat.tcps_ka_offload_drops++; + os_log_info(OS_LOG_DEFAULT, "%s: dropped lport %u fport %u\n", + __func__, frame->local_port, frame->remote_port); + + socket_unlock(so, 1); + } + + return 0; +} + +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; + uint32_t complete = 0; + + SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) { + if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) { + ASSERT(complete < UINT32_MAX); + 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; +} + +#define IFP_PER_FLOW_STAT(_ipv4_, _stat_) { \ + if (_ipv4_) { \ + ifp->if_ipv4_stat->_stat_++; \ + } else { \ + ifp->if_ipv6_stat->_stat_++; \ + } \ +} + +#define FLOW_ECN_ENABLED(_flags_) \ + ((_flags_ & (TE_ECN_ON)) == (TE_ECN_ON)) + +void +tcp_update_stats_per_flow(struct ifnet_stats_per_flow *ifs, + struct ifnet *ifp) +{ + if (ifp == NULL || !IF_FULLY_ATTACHED(ifp)) { + return; + } + + ifnet_lock_shared(ifp); + if (ifs->ecn_flags & TE_SETUPSENT) { + if (ifs->ecn_flags & TE_CLIENT_SETUP) { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_client_setup); + if (FLOW_ECN_ENABLED(ifs->ecn_flags)) { + IFP_PER_FLOW_STAT(ifs->ipv4, + ecn_client_success); + } else if (ifs->ecn_flags & TE_LOST_SYN) { + IFP_PER_FLOW_STAT(ifs->ipv4, + ecn_syn_lost); + } else { + IFP_PER_FLOW_STAT(ifs->ipv4, + ecn_peer_nosupport); + } + } else { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_server_setup); + if (FLOW_ECN_ENABLED(ifs->ecn_flags)) { + IFP_PER_FLOW_STAT(ifs->ipv4, + ecn_server_success); + } else if (ifs->ecn_flags & TE_LOST_SYN) { + IFP_PER_FLOW_STAT(ifs->ipv4, + ecn_synack_lost); + } else { + IFP_PER_FLOW_STAT(ifs->ipv4, + ecn_peer_nosupport); + } + } + } else { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_off_conn); + } + if (FLOW_ECN_ENABLED(ifs->ecn_flags)) { + if (ifs->ecn_flags & TE_RECV_ECN_CE) { + tcpstat.tcps_ecn_conn_recv_ce++; + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_recv_ce); + } + if (ifs->ecn_flags & TE_RECV_ECN_ECE) { + tcpstat.tcps_ecn_conn_recv_ece++; + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_recv_ece); + } + if (ifs->ecn_flags & (TE_RECV_ECN_CE | TE_RECV_ECN_ECE)) { + if (ifs->txretransmitbytes > 0 || + ifs->rxoutoforderbytes > 0) { + tcpstat.tcps_ecn_conn_pl_ce++; + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_plce); + } else { + tcpstat.tcps_ecn_conn_nopl_ce++; + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_noplce); + } + } else { + if (ifs->txretransmitbytes > 0 || + ifs->rxoutoforderbytes > 0) { + tcpstat.tcps_ecn_conn_plnoce++; + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_plnoce); + } + } + } + + /* Other stats are interesting for non-local connections only */ + if (ifs->local) { + ifnet_lock_done(ifp); + return; + } + + if (ifs->ipv4) { + ifp->if_ipv4_stat->timestamp = net_uptime(); + if (FLOW_ECN_ENABLED(ifs->ecn_flags)) { + tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv4_stat->ecn_on); + } else { + tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv4_stat->ecn_off); + } + } else { + ifp->if_ipv6_stat->timestamp = net_uptime(); + if (FLOW_ECN_ENABLED(ifs->ecn_flags)) { + tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv6_stat->ecn_on); + } else { + tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv6_stat->ecn_off); + } + } + + if (ifs->rxmit_drop) { + if (FLOW_ECN_ENABLED(ifs->ecn_flags)) { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_on.rxmit_drop); + } else { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_off.rxmit_drop); + } + } + if (ifs->ecn_fallback_synloss) { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_synloss); + } + if (ifs->ecn_fallback_droprst) { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_droprst); + } + if (ifs->ecn_fallback_droprxmt) { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_droprxmt); + } + if (ifs->ecn_fallback_ce) { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_ce); + } + if (ifs->ecn_fallback_reorder) { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_reorder); + } + if (ifs->ecn_recv_ce > 0) { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_recv_ce); + } + if (ifs->ecn_recv_ece > 0) { + IFP_PER_FLOW_STAT(ifs->ipv4, ecn_recv_ece); + } + + tcp_flow_lim_stats(ifs, &ifp->if_lim_stat); + ifnet_lock_done(ifp); +} +