X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/a39ff7e25e19b3a8c3020042a3872ca9ec9659f1..c3c9b80d004dbbfdf763edeb97968c6997e3b45b:/bsd/netinet/tcp_subr.c diff --git a/bsd/netinet/tcp_subr.c b/bsd/netinet/tcp_subr.c index 363dea99f..0b2a55138 100644 --- a/bsd/netinet/tcp_subr.c +++ b/bsd/netinet/tcp_subr.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2018 Apple Inc. All rights reserved. + * Copyright (c) 2000-2020 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -68,7 +68,6 @@ #include #include -#include #include #include #include @@ -90,27 +89,23 @@ #include #include #include +#include +#include -#define tcp_minmssoverload fring -#define _IP_VHL +#define tcp_minmssoverload fring +#define _IP_VHL #include #include #include #include -#if INET6 #include #include -#endif #include -#if INET6 #include -#endif #include #include #include -#if INET6 #include -#endif #include #include #include @@ -121,20 +116,18 @@ #include #include -#if INET6 #include -#endif #include #if TCPDEBUG #include #endif +#include + #include #if IPSEC #include -#if INET6 #include -#endif #endif /* IPSEC */ #if NECP @@ -143,57 +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)) static tcp_cc tcp_ccgen; -extern int tcp_lq_overflow; extern struct tcptimerlist tcp_timer_list; extern struct tcptailq tcp_tw_tailq; SYSCTL_SKMEM_TCP_INT(TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW | CTLFLAG_LOCKED, - int, tcp_mssdflt, TCP_MSS, "Default TCP Maximum Segment Size"); + int, tcp_mssdflt, TCP_MSS, "Default TCP Maximum Segment Size"); -#if INET6 SYSCTL_SKMEM_TCP_INT(TCPCTL_V6MSSDFLT, v6mssdflt, - CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_v6mssdflt, TCP6_MSS, - "Default TCP Maximum Segment Size for IPv6"); -#endif + 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, 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"); + 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 */ SYSCTL_SKMEM_TCP_INT(OID_AUTO, fastopen_backlog, - CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_tfo_backlog, 10, - "Backlog queue for half-open TFO connections"); + CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_tfo_backlog, 10, + "Backlog queue for half-open TFO connections"); 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_fastopen, TCP_FASTOPEN_CLIENT | TCP_FASTOPEN_SERVER, + "Enable TCP Fastopen (RFC 7413)"); SYSCTL_SKMEM_TCP_INT(OID_AUTO, now_init, CTLFLAG_RD | CTLFLAG_LOCKED, - uint32_t, tcp_now_init, 0, "Initial tcp now value"); + 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"); + uint32_t, tcp_microuptime_init, 0, "Initial tcp uptime value in micro seconds"); /* * Minimum MSS we accept and use. This prevents DoS attacks where @@ -204,75 +190,70 @@ SYSCTL_SKMEM_TCP_INT(OID_AUTO, microuptime_init, CTLFLAG_RD | CTLFLAG_LOCKED, * checking. This setting prevents us from sending too small packets. */ SYSCTL_SKMEM_TCP_INT(OID_AUTO, minmss, CTLFLAG_RW | CTLFLAG_LOCKED, - int, tcp_minmss, TCP_MINMSS, "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, do_tcpdrain, CTLFLAG_RW | CTLFLAG_LOCKED, - static int, do_tcpdrain, 0, - "Enable tcp_drain routine for extra help when low on mbufs"); + int, tcp_minmss, TCP_MINMSS, "Minmum TCP Maximum Segment Size"); SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED, - &tcbinfo.ipi_count, 0, "Number of active PCBs"); + &tcbinfo.ipi_count, 0, "Number of active PCBs"); SYSCTL_INT(_net_inet_tcp, OID_AUTO, tw_pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED, - &tcbinfo.ipi_twcount, 0, "Number of pcbs in time-wait state"); + &tcbinfo.ipi_twcount, 0, "Number of pcbs in time-wait state"); 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, 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, - &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly"); + &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly"); SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, - CTLFLAG_RW | CTLFLAG_LOCKED, - &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); + CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); #endif /* (DEVELOPMENT || DEBUG) */ SYSCTL_SKMEM_TCP_INT(OID_AUTO, rtt_min, CTLFLAG_RW | CTLFLAG_LOCKED, - int, tcp_TCPTV_MIN, 100, "min rtt value allowed"); + int, tcp_TCPTV_MIN, 100, "min rtt value allowed"); SYSCTL_SKMEM_TCP_INT(OID_AUTO, rexmt_slop, CTLFLAG_RW, - int, tcp_rexmt_slop, TCPTV_REXMTSLOP, "Slop added to retransmit timeout"); + int, tcp_rexmt_slop, TCPTV_REXMTSLOP, "Slop added to retransmit timeout"); 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_use_randomport, 0, + "Randomize TCP port numbers"); SYSCTL_SKMEM_TCP_INT(OID_AUTO, win_scale_factor, CTLFLAG_RW | CTLFLAG_LOCKED, - __private_extern__ int, tcp_win_scale, 3, "Window scaling factor"); + __private_extern__ int, tcp_win_scale, 3, "Window scaling factor"); -static void tcp_cleartaocache(void); -static void tcp_notify(struct inpcb *, int); +#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) */ -struct zone *sack_hole_zone; -struct zone *tcp_reass_zone; -struct zone *tcp_bwmeas_zone; -struct zone *tcp_rxt_seg_zone; +static int tso_debug = 0; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso_debug, CTLFLAG_RW | CTLFLAG_LOCKED, + &tso_debug, 0, "TSO verbosity"); -extern int slowlink_wsize; /* window correction for slow links */ +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; + +extern int slowlink_wsize; /* window correction for slow links */ extern int path_mtu_discovery; 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; +#define TCP_BWMEAS_BURST_MINSIZE 6 +#define TCP_BWMEAS_BURST_MAXSIZE 25 /* * Target size of TCP PCB hash tables. Must be a power of two. @@ -281,12 +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 @@ -296,16 +277,18 @@ 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; @@ -323,21 +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); } @@ -369,8 +349,9 @@ tcp_sysctl_fastopenkey(__unused struct sysctl_oid *oidp, __unused void *arg1, 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++) { /* @@ -386,19 +367,19 @@ 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) { - return (sizeof(struct inpcb)); + return sizeof(struct inpcb); } int get_tcp_str_size(void) { - return (sizeof(struct tcpcb)); + return sizeof(struct tcpcb); } static int scale_to_powerof2(int size); @@ -414,7 +395,8 @@ static int scale_to_powerof2(int size); * 3. Same value as argument size if it is already a power of two. */ static int -scale_to_powerof2(int size) { +scale_to_powerof2(int size) +{ /* Handle special case of size = 0 */ int ret = size ? size : 1; @@ -426,7 +408,7 @@ scale_to_powerof2(int size) { * 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 */ @@ -437,7 +419,7 @@ scale_to_powerof2(int size) { } } - return (ret); + return ret; } static void @@ -459,16 +441,16 @@ tcp_init(struct protosw *pp, struct domain *dp) static int tcp_initialized = 0; 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; @@ -485,7 +467,7 @@ tcp_init(struct protosw *pp, struct domain *dp) /* expose initial uptime/now via systcl for utcp to keep time sync */ tcp_now_init = tcp_now; tcp_microuptime_init = - tcp_uptime.tv_usec + (tcp_uptime.tv_sec * USEC_PER_SEC); + (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); @@ -526,72 +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_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 = 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 sackhole), sizeof(u_int64_t)); + sack_hole_zone = zone_create("sack_hole 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 = 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 = 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); - -#if INET6 -#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) -#else /* INET6 */ -#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) -#endif /* INET6 */ + 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); + + 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 = (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 = (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); + +#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 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 */ @@ -610,7 +565,8 @@ tcp_init(struct protosw *pp, struct domain *dp) if ((tcp_timer_list.mtx = lck_mtx_alloc_init(tcp_timer_list.mtx_grp, tcp_timer_list.mtx_attr)) == NULL) { panic("failed to allocate memory for tcp_timer_list.mtx\n"); - }; + } + ; tcp_timer_list.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"); @@ -627,32 +583,31 @@ tcp_init(struct protosw *pp, struct domain *dp) tcp_uptime_lock = lck_spin_alloc_init(tcp_uptime_mtx_grp, tcp_uptime_mtx_attr); - /* Initialize TCP LRO data structures */ - tcp_lro_init(); - /* 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) { - #if CONFIG_EMBEDDED - tcp_autorcvbuf_max = 2 * 1024 * 1024; - tcp_autosndbuf_max = 2 * 1024 * 1024; - #else - tcp_autorcvbuf_max = 1024 * 1024; - tcp_autosndbuf_max = 1024 * 1024; - #endif /* CONFIG_EMBEDDED */ + 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); - - /* - * Receive buffer max for cellular interfaces supporting - * Carrier Aggregation is higher - */ - tcp_autorcvbuf_max_ca = 2 * 1024 * 1024; } } @@ -667,25 +622,22 @@ 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_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 - { + htonl(sizeof(struct tcphdr) + IPPROTO_TCP)); + } else { struct ip *ip = (struct ip *) ip_ptr; ip->ip_vhl = IP_VHL_BORING; @@ -726,13 +678,14 @@ tcp_maketemplate(struct tcpcb *tp) 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; } /* @@ -751,81 +704,74 @@ tcp_maketemplate(struct tcpcb *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 */ - { + } else { ro = &sro; 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))); + (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 { @@ -834,16 +780,13 @@ 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 */ - { + } 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); @@ -861,76 +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); + 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 = ip_defttl; + 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 */ - { + } 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; } @@ -942,17 +869,31 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, svc_flags |= PKT_SCF_IPV6; } sotc = tp->t_inpcb->inp_socket->so_traffic_class; - set_packet_service_class(m, tp->t_inpcb->inp_socket, - sotc, svc_flags); + 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 | 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; bzero(&ip6oa, sizeof(ip6oa)); @@ -961,36 +902,58 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, ip6oa.ip6oa_sotc = SO_TC_UNSPEC; ip6oa.ip6oa_netsvctype = _NET_SERVICE_TYPE_UNSPEC; - if (tra->ifscope != IFSCOPE_NONE) + 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) + } + 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)) + 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 = outif; } - if (ro6 == &sro6) + if (ro6 == &sro6) { ROUTE_RELEASE(ro6); - } else -#endif /* INET6 */ - { + } + } else { struct ip_out_args ipoa; bzero(&ipoa, sizeof(ipoa)); ipoa.ipoa_boundif = tra->ifscope; @@ -998,18 +961,30 @@ tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, ipoa.ipoa_sotc = SO_TC_UNSPEC; ipoa.ipoa_netsvctype = _NET_SERVICE_TYPE_UNSPEC; - if (tra->ifscope != IFSCOPE_NONE) + 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)) + 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) { @@ -1021,11 +996,19 @@ 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 = outif; - } if (ro != &sro) { /* Synchronize cached PCB route */ @@ -1048,31 +1031,24 @@ tcp_newtcpcb(struct inpcb *inp) struct inp_tp *it; struct tcpcb *tp; struct socket *so = inp->inp_socket; -#if INET6 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); @@ -1089,47 +1065,67 @@ tcp_newtcpcb(struct inpcb *inp) 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) + 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 */ } /* @@ -1147,16 +1143,21 @@ tcp_drop(struct tcpcb *tp, int 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 @@ -1165,19 +1166,24 @@ 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 + } 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)); @@ -1187,11 +1193,19 @@ tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt) tp->t_rttvar = 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 @@ -1200,8 +1214,9 @@ tcp_create_ifnet_stats_per_flow(struct tcpcb *tp, { struct inpcb *inp; struct socket *so; - if (tp == NULL || ifs == NULL) + if (tp == NULL || ifs == NULL) { return; + } bzero(ifs, sizeof(*ifs)); inp = tp->t_inpcb; @@ -1230,7 +1245,7 @@ tcp_create_ifnet_stats_per_flow(struct tcpcb *tp, } else { ifs->bw_sndbw_max = 0; } - if (tp->t_bwmeas!= NULL && tp->t_bwmeas->bw_rcvbw_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; @@ -1277,8 +1292,9 @@ tcp_flow_ecn_perf_stats(struct ifnet_stats_per_flow *ifs, /* SACK episodes */ stat->sack_episodes += ifs->sack_recovery_episodes; - if (ifs->connreset) + if (ifs->connreset) { stat->rst_drop++; + } } static inline void @@ -1295,14 +1311,14 @@ tcp_flow_lim_stats(struct ifnet_stats_per_flow *ifs, 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, + 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, + stat->lim_dl_max_bandwidth = MAX(stat->lim_dl_max_bandwidth, ifs->bw_rcvbw_max); } @@ -1333,17 +1349,17 @@ tcp_flow_lim_stats(struct ifnet_stats_per_flow *ifs, if (stat->lim_rtt_min == 0) { stat->lim_rtt_min = ifs->rttmin; } else { - stat->lim_rtt_min = min(stat->lim_rtt_min, ifs->rttmin); + stat->lim_rtt_min = MIN(stat->lim_rtt_min, ifs->rttmin); } /* connection timeouts */ stat->lim_conn_attempts++; - if (ifs->conntimeout) + if (ifs->conntimeout) { stat->lim_conn_timeouts++; + } /* bytes sent using background delay-based algorithms */ stat->lim_bk_txpkts += ifs->bk_txpackets; - } /* @@ -1357,17 +1373,18 @@ 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); @@ -1385,22 +1402,21 @@ tcp_close(struct tcpcb *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, @@ -1415,20 +1431,19 @@ tcp_close(struct tcpcb *tp) */ if (tp->t_rttupdated >= 16) { 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, @@ -1441,7 +1456,7 @@ tcp_close(struct tcpcb *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. @@ -1450,19 +1465,27 @@ tcp_close(struct tcpcb *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: @@ -1479,10 +1502,11 @@ tcp_close(struct tcpcb *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) { @@ -1491,19 +1515,19 @@ tcp_close(struct tcpcb *tp) * 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 /* INET6 */ - sizeof (struct tcpiphdr)); - 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++; } } @@ -1511,7 +1535,7 @@ tcp_close(struct tcpcb *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; @@ -1519,8 +1543,9 @@ tcp_close(struct tcpcb *tp) } no_valid_rt: - if (rt != NULL) + if (rt != NULL) { RT_UNLOCK(rt); + } /* free the reassembly queue, if any */ (void) tcp_freeq(tp); @@ -1539,12 +1564,14 @@ no_valid_rt: } 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 (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) - inp->inp_saved_ppcb = (caddr_t) tp; + if (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) { + inp->inp_saved_ppcb = (caddr_t) tp; + } tp->t_state = TCPS_CLOSED; @@ -1555,13 +1582,10 @@ no_valid_rt: 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 @@ -1570,8 +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); @@ -1593,12 +1618,11 @@ 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 @@ -1607,7 +1631,7 @@ no_valid_rt: tcpstat.tcps_closed++; KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_END, tcpstat.tcps_closed, 0, 0, 0, 0); - return (NULL); + return NULL; } int @@ -1623,47 +1647,38 @@ tcp_freeq(struct tcpcb *tp) 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(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) { + 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 */ socket_unlock(inp->inp_socket, 1); continue; } tp = intotcpcb(inp); - if (do_tcpdrain) - tcp_freeq(tp); - so_drain_extended_bk_idle(inp->inp_socket); socket_unlock(inp->inp_socket, 1); } } lck_rw_done(tcbinfo.ipi_lock); - } /* @@ -1679,9 +1694,9 @@ 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); @@ -1700,29 +1715,21 @@ tcp_notify(struct inpcb *inp, int error) 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 * 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_minsize = elm->bw_minsizepkts * tp->t_maxseg; - return (elm); + return elm; } void @@ -1742,21 +1749,25 @@ get_tcp_inp_list(struct inpcb **inp_list, int n, inp_gen_t gencnt) LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) { if (inp->inp_gencnt <= gencnt && - inp->inp_state != INPCB_STATE_DEAD) + inp->inp_state != INPCB_STATE_DEAD) { inp_list[i++] = inp; - if (i >= n) + } + 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->inp_state != INPCB_STATE_DEAD) { inp_list[i++] = inp; - if (i >= n) + } + if (i >= n) { break; + } } - return (i); + return i; } /* @@ -1840,14 +1851,14 @@ tcp_pcblist SYSCTL_HANDLER_ARGS if (req->oldptr == USER_ADDR_NULL) { n = tcbinfo.ipi_count; req->oldidx = 2 * (sizeof(xig)) - + (n + n/8) * sizeof(struct xtcpcb); + + (n + n / 8) * sizeof(struct xtcpcb); lck_rw_done(tcbinfo.ipi_lock); - return (0); + return 0; } if (req->newptr != USER_ADDR_NULL) { lck_rw_done(tcbinfo.ipi_lock); - return (EPERM); + return EPERM; } /* @@ -1864,20 +1875,20 @@ tcp_pcblist SYSCTL_HANDLER_ARGS error = SYSCTL_OUT(req, &xig, sizeof(xig)); if (error) { lck_rw_done(tcbinfo.ipi_lock); - return (error); + return error; } /* * We are done if there is no pcb */ if (n == 0) { lck_rw_done(tcbinfo.ipi_lock); - return (0); + return 0; } - inp_list = _MALLOC(n * sizeof (*inp_list), M_TEMP, M_WAITOK); + inp_list = _MALLOC(n * sizeof(*inp_list), M_TEMP, M_WAITOK); if (inp_list == 0) { lck_rw_done(tcbinfo.ipi_lock); - return (ENOMEM); + return ENOMEM; } n = get_tcp_inp_list(inp_list, n, gencnt); @@ -1890,8 +1901,9 @@ tcp_pcblist SYSCTL_HANDLER_ARGS inp = inp_list[i]; - if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) + 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); @@ -1913,8 +1925,9 @@ tcp_pcblist SYSCTL_HANDLER_ARGS } else { bzero((char *) &xt.xt_tp, sizeof(xt.xt_tp)); } - if (inp->inp_socket) + if (inp->inp_socket) { sotoxsocket(inp->inp_socket, &xt.xt_socket); + } socket_unlock(inp->inp_socket, 1); @@ -1937,14 +1950,14 @@ tcp_pcblist SYSCTL_HANDLER_ARGS } FREE(inp_list, M_TEMP); lck_rw_done(tcbinfo.ipi_lock); - return (error); + return error; } 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 !CONFIG_EMBEDDED +#if XNU_TARGET_OS_OSX static void tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp) @@ -2022,14 +2035,14 @@ tcp_pcblist64 SYSCTL_HANDLER_ARGS if (req->oldptr == USER_ADDR_NULL) { n = tcbinfo.ipi_count; req->oldidx = 2 * (sizeof(xig)) - + (n + n/8) * sizeof(struct xtcpcb64); + + (n + n / 8) * sizeof(struct xtcpcb64); lck_rw_done(tcbinfo.ipi_lock); - return (0); + return 0; } if (req->newptr != USER_ADDR_NULL) { lck_rw_done(tcbinfo.ipi_lock); - return (EPERM); + return EPERM; } /* @@ -2046,20 +2059,20 @@ tcp_pcblist64 SYSCTL_HANDLER_ARGS error = SYSCTL_OUT(req, &xig, sizeof(xig)); if (error) { lck_rw_done(tcbinfo.ipi_lock); - return (error); + return error; } /* * We are done if there is no pcb */ if (n == 0) { lck_rw_done(tcbinfo.ipi_lock); - return (0); + return 0; } - inp_list = _MALLOC(n * sizeof (*inp_list), M_TEMP, M_WAITOK); + inp_list = _MALLOC(n * sizeof(*inp_list), M_TEMP, M_WAITOK); if (inp_list == 0) { lck_rw_done(tcbinfo.ipi_lock); - return (ENOMEM); + return ENOMEM; } n = get_tcp_inp_list(inp_list, n, gencnt); @@ -2071,8 +2084,9 @@ tcp_pcblist64 SYSCTL_HANDLER_ARGS inp = inp_list[i]; - if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) + 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); @@ -2088,42 +2102,44 @@ tcp_pcblist64 SYSCTL_HANDLER_ARGS 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) + if (inp->inp_ppcb != NULL) { tcpcb_to_xtcpcb64((struct tcpcb *)inp->inp_ppcb, &xt); - if (inp->inp_socket) + } + 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)); + /* + * 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); + 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 /* !CONFIG_EMBEDDED */ +#endif /* XNU_TARGET_OS_OSX */ static int tcp_pcblist_n SYSCTL_HANDLER_ARGS @@ -2133,33 +2149,45 @@ tcp_pcblist_n SYSCTL_HANDLER_ARGS error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo); - return (error); + return error; } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist_n, - CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, - tcp_pcblist_n, "S,xtcpcb_n", "List of active TCP connections"); + 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 tcp_count_opportunistic(unsigned int ifindex, u_int32_t flags) { - return (inpcb_count_opportunistic(ifindex, &tcbinfo, flags)); + return inpcb_count_opportunistic(ifindex, &tcbinfo, flags); } __private_extern__ uint32_t tcp_find_anypcb_byaddr(struct ifaddr *ifa) { - return (inpcb_find_anypcb_byaddr(ifa, &tcbinfo)); + return inpcb_find_anypcb_byaddr(ifa, &tcbinfo); } static void @@ -2169,9 +2197,10 @@ tcp_handle_msgsize(struct ip *ip, struct inpcb *inp) u_short ifscope = IFSCOPE_NONE; int mtu; struct sockaddr_in icmpsrc = { - sizeof (struct sockaddr_in), - AF_INET, 0, { 0 }, - { 0, 0, 0, 0, 0, 0, 0, 0 } }; + .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; icp = (struct icmp *)(void *) @@ -2193,8 +2222,9 @@ tcp_handle_msgsize(struct ip *ip, struct inpcb *inp) * lock bit, indicating that we are no longer doing MTU * discovery. */ - if (ROUTE_UNUSABLE(&(inp->inp_route)) == false) + if (ROUTE_UNUSABLE(&(inp->inp_route)) == false) { rt = inp->inp_route.ro_rt; + } /* * icmp6_mtudisc_update scopes the routing lookup @@ -2207,12 +2237,13 @@ tcp_handle_msgsize(struct ip *ip, struct inpcb *inp) * Take the interface scope from cached route or * the last outgoing interface from inp */ - if (rt != NULL) + if (rt != NULL) { ifscope = (rt->rt_ifp != NULL) ? rt->rt_ifp->if_index : IFSCOPE_NONE; - else + } else { ifscope = (inp->inp_last_outifp != NULL) ? inp->inp_last_outifp->if_index : IFSCOPE_NONE; + } if ((rt == NULL) || !(rt->rt_flags & RTF_HOST) || @@ -2237,17 +2268,18 @@ tcp_handle_msgsize(struct ip *ip, struct inpcb *inp) * < route's MTU. We may want to adopt * that change. */ - if (mtu == 0) + 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); + sizeof(ipv4str)), mtu); #endif if (mtu < max(296, (tcp_minmss + - sizeof (struct tcpiphdr)))) { + sizeof(struct tcpiphdr)))) { rt->rt_rmx.rmx_locks |= RTV_MTU; } else if (rt->rt_rmx.rmx_mtu > mtu) { rt->rt_rmx.rmx_mtu = mtu; @@ -2271,31 +2303,36 @@ tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip, __unused struct ifnet *ifp 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) + if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) { return; + } - if ((unsigned)cmd >= PRC_NCMDS) + if ((unsigned)cmd >= PRC_NCMDS) { return; + } /* Source quench is deprecated */ - if (cmd == PRC_QUENCH) - return; + if (cmd == PRC_QUENCH) { + return; + } - if (cmd == PRC_MSGSIZE) + 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) + } 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) + else if (cmd == PRC_HOSTDEAD) { ip = NULL; - else if (inetctlerrmap[cmd] == 0 && !PRC_IS_REDIRECT(cmd)) + } else if (inetctlerrmap[cmd] == 0 && !PRC_IS_REDIRECT(cmd)) { return; + } if (ip == NULL) { @@ -2330,8 +2367,9 @@ tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip, __unused struct ifnet *ifp tp = intotcpcb(inp); if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && SEQ_LT(icmp_tcp_seq, tp->snd_max)) { - if (cmd == PRC_MSGSIZE) + if (cmd == PRC_MSGSIZE) { tcp_handle_msgsize(ip, inp); + } (*notify)(inp, inetctlerrmap[cmd]); } @@ -2339,13 +2377,11 @@ tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip, __unused struct ifnet *ifp socket_unlock(inp->inp_socket, 1); } -#if INET6 void tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d, __unused struct ifnet *ifp) { tcp_seq icmp_tcp_seq; struct in6_addr *dst; - struct tcphdr *th; void (*notify)(struct inpcb *, int) = tcp_notify; struct ip6_hdr *ip6; struct mbuf *m; @@ -2357,16 +2393,24 @@ tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d, __unused struct ifnet *ifp) 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; + } /* Source quench is deprecated */ - if (cmd == PRC_QUENCH) + if (cmd == PRC_QUENCH) { return; + } - if ((unsigned)cmd >= PRC_NCMDS) + if ((unsigned)cmd >= PRC_NCMDS) { return; + } /* if the parameter is from icmp6, decode it. */ if (d != NULL) { @@ -2380,26 +2424,28 @@ tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d, __unused struct ifnet *ifp) } else { m = NULL; ip6 = NULL; - off = 0; /* fool gcc */ + off = 0; /* fool gcc */ sa6_src = &sa6_any; dst = NULL; } - if (cmd == PRC_MSGSIZE) + if (cmd == PRC_MSGSIZE) { notify = tcp_mtudisc; - else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || + } else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && - ip6 != NULL) + 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) + else if (cmd == PRC_HOSTDEAD) { ip6 = NULL; - else if (inet6ctlerrmap[cmd] == 0 && !PRC_IS_REDIRECT(cmd)) + } else if (inet6ctlerrmap[cmd] == 0 && !PRC_IS_REDIRECT(cmd)) { return; + } if (ip6 == NULL) { @@ -2408,12 +2454,21 @@ tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d, __unused struct ifnet *ifp) return; } + /* Check if we can safely get the ports from the tcp hdr */ if (m == NULL || - (m->m_pkthdr.len < (int32_t) (off + offsetof(struct tcphdr, th_seq)))) + (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); - th = (struct tcphdr *)(void *)mtodo(m, off); - icmp_tcp_seq = ntohl(th->th_seq); + 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 (cmd == PRC_MSGSIZE) { mtu = ntohl(icmp6->icmp6_mtu); @@ -2421,12 +2476,13 @@ tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d, __unused struct ifnet *ifp) * If no alternative MTU was proposed, or the proposed * MTU was too small, set to the min. */ - if (mtu < IPV6_MMTU) + if (mtu < IPV6_MMTU) { mtu = IPV6_MMTU - 8; + } } - inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_dst, th->th_dport, - &ip6->ip6_src, th->th_sport, 0, NULL); + 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) { @@ -2453,15 +2509,16 @@ tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d, __unused struct ifnet *ifp) * is smaller than the current one. */ if (mtu < tp->t_maxseg + - (sizeof (*th) + sizeof (*ip6))) + (sizeof(struct tcphdr) + sizeof(struct ip6_hdr))) { (*notify)(inp, inetctlerrmap[cmd]); - } else + } + } else { (*notify)(inp, inetctlerrmap[cmd]); + } } } socket_unlock(inp->inp_socket, 1); } -#endif /* INET6 */ /* @@ -2507,7 +2564,7 @@ tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d, __unused struct ifnet *ifp) * */ -#define ISN_BYTES_PER_SECOND 1048576 +#define ISN_BYTES_PER_SECOND 1048576 tcp_seq tcp_new_isn(struct tcpcb *tp) @@ -2516,24 +2573,24 @@ tcp_new_isn(struct tcpcb *tp) 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) #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_frandom(&isn_secret, sizeof(isn_secret)); #else @@ -2548,15 +2605,12 @@ tcp_new_isn(struct tcpcb *tp) sizeof(u_short)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); -#if INET6 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, sizeof(struct in6_addr)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, sizeof(struct in6_addr)); - } else -#endif - { + } else { MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, sizeof(struct in_addr)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, @@ -2566,7 +2620,7 @@ tcp_new_isn(struct tcpcb *tp) MD5Final((u_char *) &md5_buffer, &isn_ctx); new_isn = (tcp_seq) md5_buffer[0]; new_isn += timenow.tv_sec * (ISN_BYTES_PER_SECOND / hz); - return (new_isn); + return new_isn; } @@ -2580,8 +2634,9 @@ 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); + } } /* @@ -2591,48 +2646,45 @@ tcp_drop_syn_sent(struct inpcb *inp, int 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; - u_int32_t protoHdrOverhead = sizeof (struct tcpiphdr); -#if INET6 + u_int32_t protoHdrOverhead = sizeof(struct tcpiphdr); int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; - 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); -#endif /* INET6 */ + } - if (tp) { -#if INET6 - if (isipv6) + if (tp != NULL) { + if (isipv6) { rt = tcp_rtlookup6(inp, IFSCOPE_NONE); - else -#endif /* INET6 */ + } 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 */ @@ -2644,8 +2696,9 @@ tcp_mtudisc( #endif /* NECP */ mss = mtu - protoHdrOverhead; - 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 @@ -2663,19 +2716,22 @@ 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; @@ -2685,8 +2741,9 @@ tcp_mtudisc( * 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; @@ -2711,8 +2768,9 @@ tcp_rtlookup(struct inpcb *inp, unsigned int input_ifscope) 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) { @@ -2727,7 +2785,7 @@ tcp_rtlookup(struct inpcb *inp, unsigned int 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 @@ -2740,12 +2798,14 @@ tcp_rtlookup(struct inpcb *inp, unsigned int 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: @@ -2758,10 +2818,11 @@ tcp_rtlookup(struct inpcb *inp, unsigned int input_ifscope) tp = intotcpcb(inp); if (!path_mtu_discovery || ((rt != NULL) && - (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) + (!(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 (rt != NULL && rt->rt_ifp != NULL) { somultipages(inp->inp_socket, @@ -2772,25 +2833,23 @@ tcp_rtlookup(struct inpcb *inp, unsigned int input_ifscope) 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); + return rt; } -#if INET6 struct rtentry * tcp_rtlookup6(struct inpcb *inp, unsigned int input_ifscope) { @@ -2801,8 +2860,9 @@ tcp_rtlookup6(struct inpcb *inp, unsigned int input_ifscope) 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) { @@ -2831,12 +2891,14 @@ tcp_rtlookup6(struct inpcb *inp, unsigned int 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 @@ -2858,10 +2920,11 @@ tcp_rtlookup6(struct inpcb *inp, unsigned int input_ifscope) */ if (!path_mtu_discovery || ((rt != NULL) && - (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) + (!(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 (rt != NULL && rt->rt_ifp != NULL) { somultipages(inp->inp_socket, @@ -2873,23 +2936,22 @@ tcp_rtlookup6(struct inpcb *inp, unsigned int input_ifscope) 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; + } } /* * Caller needs to call RT_UNLOCK(rt). */ - return (rt); + return rt; } -#endif /* INET6 */ #if IPSEC /* compute ESP/AH header size for TCP, including outer IP header. */ @@ -2900,28 +2962,25 @@ ipsec_hdrsiz_tcp(struct tcpcb *tp) 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)) - return (0); - MGETHDR(m, M_DONTWAIT, MT_DATA); /* MAC-OK */ - if (!m) - return (0); + if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) { + return 0; + } + 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 */ - { + } else { ip = mtod(m, struct ip *); th = (struct tcphdr *)(ip + 1); m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); @@ -2929,82 +2988,35 @@ ipsec_hdrsiz_tcp(struct tcpcb *tp) hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); } m_free(m); - return (hdrsiz); + 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(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(void) -{ -} - 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) { if (so->so_flags & SOF_MP_SUBFLOW) { struct mptcb *mp_tp = tptomptp(sototcpcb(so)); - VERIFY(mp_tp); - - mpte_lock_assert_notheld(mp_tp->mpt_mpte); + struct socket *mp_so = mptetoso(mp_tp->mpt_mpte); - mpte_lock(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)) { - mpte_unlock(mp_tp->mpt_mpte); + socket_unlock(mp_so, refcount); goto retry; } } else { @@ -3019,7 +3031,7 @@ retry: goto retry; } } - } else { + } else { panic("tcp_lock: so=%p NO PCB! lr=%p lrh= %s\n", so, lr_saved, solockhistory_nr(so)); /* NOTREACHED */ @@ -3031,11 +3043,12 @@ retry: 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 @@ -3043,10 +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 " @@ -3055,8 +3069,9 @@ 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", @@ -3069,22 +3084,22 @@ tcp_unlock(struct socket *so, int refcount, void *lr) /* NOTREACHED */ } else { so->unlock_lr[so->next_unlock_lr] = lr_saved; - so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; + 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); - VERIFY(mp_tp); - mpte_lock_assert_held(mp_tp->mpt_mpte); + socket_lock_assert_owned(mp_so); - mpte_unlock(mp_tp->mpt_mpte); + 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 * @@ -3092,22 +3107,24 @@ tcp_getlock(struct socket *so, int flags) { struct inpcb *inp = sotoinpcb(so); - if (so->so_pcb) { - if (so->so_usecount < 0) + 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 (mpte_getlock(mp_tp->mpt_mpte, flags)); + return mp_so->so_proto->pr_getlock(mp_so, flags); } else { - return (&inp->inpcb_mtx); + 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; } } @@ -3125,23 +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 (tcp_recv_bg == 1 || IS_TCP_RECV_BG(so)) + if (tcp_recv_bg == 1 || IS_TCP_RECV_BG(so)) { return; - /* - * If message delivery is enabled, do not count - * unordered bytes in receive buffer towards hiwat - */ - if (so->so_flags & SOF_ENABLE_MSGS) - rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes; + } if (tcp_do_autorcvbuf == 1 && - tcp_cansbgrow(sb) && - (tp->t_flags & TF_SLOWLINK) == 0 && - (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)); } @@ -3156,34 +3168,33 @@ tcp_sbspace(struct tcpcb *tp) int32_t space; int32_t pending = 0; + 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; - /* - * 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; - 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 @@ -3191,15 +3202,17 @@ tcp_sbspace(struct tcpcb *tp) * we could break some apps (see rdar://5409343) */ - if (space < tp->t_maxseg) - return (space); + if (space < tp->t_maxseg) { + return space; + } /* Clip window size for slower link */ - if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0) - return (imin(space, slowlink_wsize)); + if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0) { + return imin(space, slowlink_wsize); + } - return (space); + return space; } /* * Checks TCP Segment Offloading capability for a given connection @@ -3208,50 +3221,115 @@ 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; + /* + * 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 (tunnel_if_index != 0) { + ifnet_head_lock_shared(); + tunnel_ifp = ifindex2ifnet[tunnel_if_index]; + ifnet_head_done(); + } + + if (tunnel_ifp == NULL) { + return; + } + + 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 && (ifp->if_hwassist & IFNET_TSO_IPV6)) { + if (ifp->if_hwassist & IFNET_TSO_IPV6) { tp->t_flags |= TF_TSO; - if (ifp->if_tso_v6_mtu != 0) + if (ifp->if_tso_v6_mtu != 0) { tp->tso_max_segment_size = ifp->if_tso_v6_mtu; - else + } else { tp->tso_max_segment_size = TCP_MAXWIN; - } else - tp->t_flags &= ~TF_TSO; - - } else -#endif /* INET6 */ - - { - if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV4)) { + } + } + } else { + if (ifp->if_hwassist & IFNET_TSO_IPV4) { tp->t_flags |= TF_TSO; - if (ifp->if_tso_v4_mtu != 0) + if (ifp->if_tso_v4_mtu != 0) { tp->tso_max_segment_size = ifp->if_tso_v4_mtu; - else + } else { tp->tso_max_segment_size = TCP_MAXWIN; - } else - tp->t_flags &= ~TF_TSO; + } + 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_) ((_tv_).tv_sec * TCP_RETRANSHZ + \ - (_tv_).tv_usec / TCP_RETRANSHZ_TO_USEC) +#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 @@ -3266,7 +3344,7 @@ void 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; @@ -3311,37 +3389,43 @@ calculate_tcp_clock(void) * defined by the constant tcp_autorcvbuf_max. */ void -tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so, - u_int32_t rcvbuf_max) +tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so) { - u_int32_t maxsockbufsize; - if (!tcp_do_rfc1323) { - tp->request_r_scale = 0; - return; - } + 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 : rcvbuf_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); } + if (tp == NULL) { + return 0; + } + notsent = so->so_snd.sb_cc - - (tp->snd_nxt - tp->snd_una); + (tp->snd_nxt - tp->snd_una); /* * When we send a FIN or SYN, not_sent can be negative. @@ -3350,7 +3434,7 @@ tcp_notsent_lowat_check(struct socket *so) { * get an error from send because cantsendmore will be set. */ if (notsent <= tp->t_notsent_lowat) { - return (1); + return 1; } /* @@ -3359,28 +3443,30 @@ tcp_notsent_lowat_check(struct socket *so) { * maxseg of data to write. */ if ((tp->t_flags & TF_NODELAY) == 0 && - notsent > 0 && notsent < tp->t_maxseg) { - return (1); + 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) { 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; @@ -3418,8 +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) { @@ -3441,17 +3528,47 @@ 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 @@ -3473,26 +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 @@ -3500,8 +3618,9 @@ 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 && @@ -3511,7 +3630,7 @@ tcp_rxtseg_dsack_for_tlp(struct tcpcb *tp) break; } } - return (dsack_for_tlp); + return dsack_for_tlp; } u_int32_t @@ -3523,15 +3642,16 @@ tcp_rxtseg_total_size(struct tcpcb *tp) 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)) { @@ -3540,17 +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)) + (tp->t_inpcb->inp_last_outifp->if_eflags & IFEF_PROBE_CONNECTIVITY)) { connstatus->probe_activated = 1; + } } boolean_t tfo_enabled(const struct tcpcb *tp) { - return ((tp->t_flagsext & TF_FASTOPEN)? TRUE : FALSE); + return (tp->t_flagsext & TF_FASTOPEN)? TRUE : FALSE; } void @@ -3577,7 +3699,7 @@ tcp_make_keepalive_frame(struct tcpcb *tp, struct ifnet *ifp, MGETHDR(m, M_WAIT, MT_HEADER); if (m == NULL) { - return (NULL); + return NULL; } m->m_pkthdr.pkt_proto = IPPROTO_TCP; @@ -3621,16 +3743,19 @@ tcp_make_keepalive_frame(struct tcpcb *tp, struct ifnet *ifp, 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)) + if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) { ip6->ip6_src.s6_addr16[1] = 0; - if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) + } + 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; + 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) { @@ -3650,7 +3775,7 @@ tcp_make_keepalive_frame(struct tcpcb *tp, struct ifnet *ifp, sizeof(struct ip6_hdr), sizeof(struct tcphdr)); } - return (m); + return m; } void @@ -3666,8 +3791,9 @@ tcp_fill_keepalive_offload_frames(ifnet_t ifp, if (ifp == NULL || frames_array == NULL || frames_array_count == 0 || frame_index >= frames_array_count || - frame_data_offset >= IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) + frame_data_offset >= IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) { return; + } /* * This function is called outside the regular TCP processing @@ -3683,16 +3809,19 @@ tcp_fill_keepalive_offload_frames(ifnet_t ifp, struct mbuf *m = NULL; struct tcpcb *tp = intotcpcb(inp); - if (frame_index >= frames_array_count) + if (frame_index >= frames_array_count) { break; + } if (inp->inp_gencnt > gencnt || - inp->inp_state == INPCB_STATE_DEAD) + inp->inp_state == INPCB_STATE_DEAD) { continue; + } if ((so = inp->inp_socket) == NULL || - (so->so_state & SS_DEFUNCT)) + (so->so_state & SS_DEFUNCT)) { continue; + } /* * check for keepalive offload flag without socket * lock to avoid a deadlock @@ -3705,8 +3834,9 @@ tcp_fill_keepalive_offload_frames(ifnet_t ifp, continue; } if (inp->inp_ppcb == NULL || - in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) + in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) { continue; + } socket_lock(so, 1); /* Release the want count */ if (inp->inp_ppcb == NULL || @@ -3764,17 +3894,22 @@ tcp_fill_keepalive_offload_frames(ifnet_t ifp, frame->ether_type = (inp->inp_vflag & INP_IPV4) ? IFNET_KEEPALIVE_OFFLOAD_FRAME_ETHERTYPE_IPV4 : IFNET_KEEPALIVE_OFFLOAD_FRAME_ETHERTYPE_IPV6; - frame->interval = tp->t_keepidle > 0 ? tp->t_keepidle : - tcp_keepidle; - frame->keep_cnt = TCP_CONN_KEEPCNT(tp); - frame->keep_retry = TCP_CONN_KEEPINTVL(tp); + frame->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) { - frame->length = frame_data_offset + - sizeof(struct ip) + sizeof(struct tcphdr); + 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); @@ -3785,20 +3920,23 @@ tcp_fill_keepalive_offload_frames(ifnet_t ifp, } else { struct in6_addr *ip6; - frame->length = frame_data_offset + - sizeof(struct ip6_hdr) + sizeof(struct tcphdr); + 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)) + 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)) + if (IN6_IS_SCOPE_EMBED(ip6)) { ip6->s6_addr16[1] = 0; + } } /* @@ -3832,24 +3970,130 @@ tcp_fill_keepalive_offload_frames(ifnet_t ifp, *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); + 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); + 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); + if (elm->notify_snd_una == tp->snd_una + so->so_snd.sb_cc) { + return EINVAL; + } } - return (0); + return 0; } errno_t @@ -3858,16 +4102,18 @@ 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), + MALLOC(nm, struct tcp_notify_ack_marker *, sizeof(*nm), M_TEMP, M_WAIT | M_ZERO); - if (nm == NULL) - return (ENOMEM); + 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)) + if (SEQ_GT(nm->notify_snd_una, elm->notify_snd_una)) { break; + } } if (elm == NULL) { @@ -3877,15 +4123,16 @@ tcp_add_notify_ack_marker(struct tcpcb *tp, u_int32_t notify_id) SLIST_INSERT_AFTER(elm, nm, notify_next); } tp->t_notify_ack_count++; - return (0); + return 0; } void tcp_notify_ack_free(struct tcpcb *tp) { struct tcp_notify_ack_marker *elm, *next; - if (SLIST_EMPTY(&tp->t_notify_ack)) + 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, @@ -3912,13 +4159,15 @@ tcp_get_notify_ack_count(struct tcpcb *tp, struct tcp_notify_ack_complete *retid) { struct tcp_notify_ack_marker *elm; - size_t complete = 0; + uint32_t complete = 0; SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) { - if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) + if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) { + ASSERT(complete < UINT32_MAX); complete++; - else + } else { break; + } } retid->notify_pending = tp->t_notify_ack_count - complete; retid->notify_complete_count = min(TCP_MAX_NOTIFY_ACK, complete); @@ -3932,8 +4181,9 @@ tcp_get_notify_ack_ids(struct tcpcb *tp, struct tcp_notify_ack_marker *elm, *next; SLIST_FOREACH_SAFE(elm, &tp->t_notify_ack, notify_next, next) { - if (i >= retid->notify_complete_count) + 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, @@ -3956,11 +4206,12 @@ tcp_notify_ack_active(struct socket *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); + if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) { + return true; + } } } - return (false); + return false; } inline int32_t @@ -3973,30 +4224,33 @@ inp_get_sndbytes_allunsent(struct socket *so, u_int32_t th_ack) so->so_snd.sb_cc > 0) { int32_t unsent, sent; sent = tp->snd_max - th_ack; - if (tp->t_flags & TF_SENTFIN) + if (tp->t_flags & TF_SENTFIN) { sent--; + } unsent = so->so_snd.sb_cc - sent; - return (unsent); + return unsent; } - return (0); + return 0; } #define IFP_PER_FLOW_STAT(_ipv4_, _stat_) { \ if (_ipv4_) { \ - ifp->if_ipv4_stat->_stat_++; \ + ifp->if_ipv4_stat->_stat_++; \ } else { \ - ifp->if_ipv6_stat->_stat_++; \ + 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, +void +tcp_update_stats_per_flow(struct ifnet_stats_per_flow *ifs, struct ifnet *ifp) { - if (ifp == NULL || !IF_FULLY_ATTACHED(ifp)) + if (ifp == NULL || !IF_FULLY_ATTACHED(ifp)) { return; + } ifnet_lock_shared(ifp); if (ifs->ecn_flags & TE_SETUPSENT) { @@ -4084,21 +4338,29 @@ void tcp_update_stats_per_flow(struct ifnet_stats_per_flow *ifs, IFP_PER_FLOW_STAT(ifs->ipv4, ecn_off.rxmit_drop); } } - if (ifs->ecn_fallback_synloss) + if (ifs->ecn_fallback_synloss) { IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_synloss); - if (ifs->ecn_fallback_droprst) + } + if (ifs->ecn_fallback_droprst) { IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_droprst); - if (ifs->ecn_fallback_droprxmt) + } + if (ifs->ecn_fallback_droprxmt) { IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_droprxmt); - if (ifs->ecn_fallback_ce) + } + if (ifs->ecn_fallback_ce) { IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_ce); - if (ifs->ecn_fallback_reorder) + } + if (ifs->ecn_fallback_reorder) { IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_reorder); - if (ifs->ecn_recv_ce > 0) + } + if (ifs->ecn_recv_ce > 0) { IFP_PER_FLOW_STAT(ifs->ipv4, ecn_recv_ce); - if (ifs->ecn_recv_ece > 0) + } + 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); } +