X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/8f6c56a50524aa785f7e596d52dddfb331e18961..490019cf9519204c5fb36b2fba54ceb983bb6b72:/bsd/netinet/tcp_subr.c diff --git a/bsd/netinet/tcp_subr.c b/bsd/netinet/tcp_subr.c index 2f61e0e1e..88d18875a 100644 --- a/bsd/netinet/tcp_subr.c +++ b/bsd/netinet/tcp_subr.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2015 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -60,7 +60,12 @@ * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.22 2001/08/22 00:59:12 silby Exp $ */ - +/* + * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce + * support for mandatory and extensible security protections. This notice + * is included in support of clause 2.2 (b) of the Apple Public License, + * Version 2.0. + */ #include #include @@ -69,9 +74,7 @@ #include #include #include -#if INET6 #include -#endif #include #include #include @@ -79,17 +82,22 @@ #include #include #include +#include #include +#include - +#include #include #include +#include +#define tcp_minmssoverload fring #define _IP_VHL #include #include #include +#include #if INET6 #include #endif @@ -99,6 +107,7 @@ #endif #include #include +#include #if INET6 #include #endif @@ -107,6 +116,10 @@ #include #include #include +#include +#include +#include + #if INET6 #include #endif @@ -123,30 +136,66 @@ #endif #endif /*IPSEC*/ -#include +#if NECP +#include +#endif /* NECP */ + +#undef tcp_minmssoverload + +#if CONFIG_MACF_NET +#include +#endif /* MAC_NET */ + +#include +#include +#include #include +#include + +#include #define DBG_FNC_TCP_CLOSE NETDBG_CODE(DBG_NETTCP, ((5 << 8) | 2)) extern int tcp_lq_overflow; -/* temporary: for testing */ -#if IPSEC -extern int ipsec_bypass; -extern lck_mtx_t *sadb_mutex; -#endif +extern struct tcptimerlist tcp_timer_list; +extern struct tcptailq tcp_tw_tailq; int tcp_mssdflt = TCP_MSS; -SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); #if INET6 int tcp_v6mssdflt = TCP6_MSS; SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, - CTLFLAG_RW, &tcp_v6mssdflt , 0, + CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_v6mssdflt , 0, "Default TCP Maximum Segment Size for IPv6"); #endif +extern int tcp_do_autorcvbuf; + +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"); + +/* Current count of half-open TFO connections */ +int tcp_tfo_halfcnt = 0; + +/* Maximum of half-open TFO connection backlog */ +int tcp_tfo_backlog = 10; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen_backlog, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_tfo_backlog, 0, "Backlog queue for half-open TFO connections"); + +int tcp_fastopen = TCP_FASTOPEN_CLIENT | TCP_FASTOPEN_SERVER; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_fastopen, 0, "Enable TCP Fastopen (RFC 7413)"); + +int tcp_tfo_fallback_min = 10; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen_fallback_min, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_tfo_fallback_min, 0, "Mininum number of trials without TFO when in fallback mode"); + /* * Minimum MSS we accept and use. This prevents DoS attacks where * we are forced to a ridiculous low MSS like 20 and send hundreds @@ -156,60 +205,83 @@ SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, * checking. This setting prevents us from sending too small packets. */ int tcp_minmss = TCP_MINMSS; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); - -/* - * Number of TCP segments per second we accept from remote host - * before we start to calculate average segment size. If average - * segment size drops below the minimum TCP MSS we assume a DoS - * attack and reset+drop the connection. Care has to be taken not to - * set this value too small to not kill interactive type connections - * (telnet, SSH) which send many small packets. - */ -#ifdef FIX_WORKAROUND_FOR_3894301 -__private_extern__ int tcp_minmssoverload = TCP_MINMSSOVERLOAD; -#else -__private_extern__ int tcp_minmssoverload = 0; -#endif -SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, - &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" - "be under the MINMSS Size"); - -static int tcp_do_rfc1323 = 1; -SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, - &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); - +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, +SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); -static int tcp_tcbhashsize = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, - &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); - static int do_tcpdrain = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW | CTLFLAG_LOCKED, &do_tcpdrain, 0, "Enable tcp_drain routine for extra help when low on mbufs"); -SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED, &tcbinfo.ipi_count, 0, "Number of active PCBs"); +SYSCTL_INT(_net_inet_tcp, OID_AUTO, tw_pcbcount, + CTLFLAG_RD | CTLFLAG_LOCKED, + &tcbinfo.ipi_twcount, 0, "Number of pcbs in time-wait state"); + static int icmp_may_rst = 1; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW | CTLFLAG_LOCKED, &icmp_may_rst, 0, "Certain ICMP unreachable messages may abort connections in SYN_SENT"); static int tcp_strict_rfc1948 = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948, CTLFLAG_RW, +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"); -static int tcp_isn_reseed_interval = 0; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, +SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, + CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); +#endif /* (DEVELOPMENT || DEBUG) */ + +int tcp_TCPTV_MIN = 100; /* 100ms minimum RTT */ +SYSCTL_INT(_net_inet_tcp, OID_AUTO, rtt_min, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_TCPTV_MIN, 0, "min rtt value allowed"); + +int tcp_rexmt_slop = TCPTV_REXMTSLOP; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmt_slop, CTLFLAG_RW, + &tcp_rexmt_slop, 0, "Slop added to retransmit timeout"); + +__private_extern__ int tcp_use_randomport = 0; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, randomize_ports, CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_use_randomport, 0, "Randomize TCP port numbers"); + +__private_extern__ int tcp_win_scale = 3; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, win_scale_factor, + CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_win_scale, 0, "Window scaling factor"); static void tcp_cleartaocache(void); static void tcp_notify(struct inpcb *, int); + 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; + +extern u_int32_t tcp_autorcvbuf_max; +extern u_int32_t tcp_autorcvbuf_inc_shift; +static void tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb); + +#define TCP_BWMEAS_BURST_MINSIZE 6 +#define TCP_BWMEAS_BURST_MAXSIZE 25 + +static uint32_t bwmeas_elm_size; /* * Target size of TCP PCB hash tables. Must be a power of two. @@ -218,9 +290,13 @@ struct zone *sack_hole_zone; * variable net.inet.tcp.tcbhashsize */ #ifndef TCBHASHSIZE -#define TCBHASHSIZE 4096 +#define TCBHASHSIZE CONFIG_TCBHASHSIZE #endif +__private_extern__ int tcp_tcbhashsize = TCBHASHSIZE; +SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD | CTLFLAG_LOCKED, + &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); + /* * This is the actual shape of what we allocate using the zone * allocator. Doing it this way allows us to protect both structures @@ -230,125 +306,317 @@ struct zone *sack_hole_zone; * to be changed, eventually, for greater efficiency). */ #define ALIGNMENT 32 -#define ALIGNM1 (ALIGNMENT - 1) struct inp_tp { - union { - struct inpcb inp; - char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; - } inp_tp_u; - struct tcpcb tcb; + struct inpcb inp; + struct tcpcb tcb __attribute__((aligned(ALIGNMENT))); }; #undef ALIGNMENT -#undef ALIGNM1 -static struct tcpcb dummy_tcb; +int get_inpcb_str_size(void); +int get_tcp_str_size(void); +static void tcpcb_to_otcpcb(struct tcpcb *, struct otcpcb *); -extern struct inpcbhead time_wait_slots[]; -extern int cur_tw_slot; -extern u_long *delack_bitmask; -extern u_long route_generation; +static lck_attr_t *tcp_uptime_mtx_attr = NULL; /* mutex attributes */ +static lck_grp_t *tcp_uptime_mtx_grp = NULL; /* mutex group definition */ +static lck_grp_attr_t *tcp_uptime_mtx_grp_attr = NULL; /* mutex group attributes */ +int tcp_notsent_lowat_check(struct socket *so); +static aes_encrypt_ctx tfo_ctx; /* Crypto-context for TFO */ -int get_inpcb_str_size() +void +tcp_tfo_gen_cookie(struct inpcb *inp, u_char *out, size_t blk_size) { - return sizeof(struct inpcb); + 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) + memcpy(in, &inp->in6p_faddr, sizeof(struct in6_addr)); + else +#endif /* INET6 */ + memcpy(in, &inp->inp_faddr, sizeof(struct in_addr)); + + aes_encrypt_cbc(in, NULL, 1, out, &tfo_ctx); +} + +__private_extern__ int +tcp_sysctl_fastopenkey(__unused struct sysctl_oid *oidp, __unused void *arg1, + __unused int arg2, struct sysctl_req *req) +{ + int error = 0; + /* TFO-key is expressed as a string in hex format (+1 to account for \0 char) */ + char keystring[TCP_FASTOPEN_KEYLEN * 2 + 1]; + u_int32_t key[TCP_FASTOPEN_KEYLEN / sizeof(u_int32_t)]; + int i; + + /* -1, because newlen is len without the terminating \0 character */ + if (req->newlen != (sizeof(keystring) - 1)) { + error = EINVAL; + goto exit; + } + + /* sysctl_io_string copies keystring into the oldptr of the sysctl_req. + * Make sure everything is zero, to avoid putting garbage in there or + * leaking the stack. + */ + bzero(keystring, sizeof(keystring)); + + error = sysctl_io_string(req, keystring, sizeof(keystring), 0, NULL); + if (error) + goto exit; + + for (i = 0; i < (TCP_FASTOPEN_KEYLEN / sizeof(u_int32_t)); i++) { + /* We jump over the keystring in 8-character (4 byte in hex) steps */ + if (sscanf(&keystring[i * 8], "%8x", &key[i]) != 1) { + error = EINVAL; + goto exit; + } + } + + aes_encrypt_key128((u_char *)key, &tfo_ctx); + +exit: + return (error); } +int get_inpcb_str_size(void) +{ + return sizeof(struct inpcb); +} -int get_tcp_str_size() +int get_tcp_str_size(void) { return sizeof(struct tcpcb); } int tcp_freeq(struct tcpcb *tp); +static int scale_to_powerof2(int size); + +/* + * This helper routine returns one of the following scaled value of size: + * 1. Rounded down power of two value of size if the size value passed as + * argument is not a power of two and the rounded up value overflows. + * OR + * 2. Rounded up power of two value of size if the size value passed as + * argument is not a power of two and the rounded up value does not overflow + * OR + * 3. Same value as argument size if it is already a power of two. + */ +static int scale_to_powerof2(int size) { + /* Handle special case of size = 0 */ + int ret = size ? size : 1; + + if (!powerof2(ret)) { + while(!powerof2(size)) { + /* + * Clear out least significant + * set bit till size is left with + * its highest set bit at which point + * it is rounded down power of two. + */ + size = size & (size -1); + } + + /* Check for overflow when rounding up */ + if (0 == (size << 1)) { + ret = size; + } else { + ret = size << 1; + } + } + + return ret; +} + +static void +tcp_tfo_init() +{ + u_char key[TCP_FASTOPEN_KEYLEN]; + + read_random(key, sizeof(key)); + aes_encrypt_key128(key, &tfo_ctx); +} /* * Tcp initialization */ void -tcp_init() +tcp_init(struct protosw *pp, struct domain *dp) { - int hashsize = TCBHASHSIZE; +#pragma unused(dp) + static int tcp_initialized = 0; vm_size_t str_size; - int i; - struct inpcbinfo *pcbinfo; - + struct inpcbinfo *pcbinfo; + + VERIFY((pp->pr_flags & (PR_INITIALIZED|PR_ATTACHED)) == PR_ATTACHED); + + if (tcp_initialized) + return; + tcp_initialized = 1; + tcp_ccgen = 1; tcp_cleartaocache(); - tcp_delacktime = TCPTV_DELACK; tcp_keepinit = TCPTV_KEEP_INIT; tcp_keepidle = TCPTV_KEEP_IDLE; tcp_keepintvl = TCPTV_KEEPINTVL; + tcp_keepcnt = TCPTV_KEEPCNT; tcp_maxpersistidle = TCPTV_KEEP_IDLE; tcp_msl = TCPTV_MSL; + + microuptime(&tcp_uptime); read_random(&tcp_now, sizeof(tcp_now)); - tcp_now = tcp_now & 0x7fffffff; /* Starts tcp internal 500ms clock at a random value */ + tcp_now = tcp_now & 0x3fffffff; /* Starts tcp internal clock at a random value */ + tcp_tfo_init(); LIST_INIT(&tcb); - tcbinfo.listhead = &tcb; + tcbinfo.ipi_listhead = &tcb; + pcbinfo = &tcbinfo; - if (!powerof2(hashsize)) { - printf("WARNING: TCB hash size not a power of 2\n"); - hashsize = 512; /* safe default */ - } - tcp_tcbhashsize = hashsize; - tcbinfo.hashsize = hashsize; - tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); - tcbinfo.porthashbase = hashinit(hashsize, M_PCB, - &tcbinfo.porthashmask); - str_size = (vm_size_t) sizeof(struct inp_tp); - tcbinfo.ipi_zone = (void *) zinit(str_size, 120000*str_size, 8192, "tcpcb"); + /* + * allocate lock group attribute and group for tcp pcb mutexes + */ + pcbinfo->ipi_lock_grp_attr = lck_grp_attr_alloc_init(); + pcbinfo->ipi_lock_grp = lck_grp_alloc_init("tcppcb", pcbinfo->ipi_lock_grp_attr); + + /* + * allocate the lock attribute for tcp pcb mutexes + */ + pcbinfo->ipi_lock_attr = lck_attr_alloc_init(); + + if ((pcbinfo->ipi_lock = lck_rw_alloc_init(pcbinfo->ipi_lock_grp, + pcbinfo->ipi_lock_attr)) == NULL) { + panic("%s: unable to allocate PCB lock\n", __func__); + /* NOTREACHED */ + } + + if (tcp_tcbhashsize == 0) { + /* Set to default */ + tcp_tcbhashsize = 512; + } + + if (!powerof2(tcp_tcbhashsize)) { + int old_hash_size = tcp_tcbhashsize; + tcp_tcbhashsize = scale_to_powerof2(tcp_tcbhashsize); + /* Lower limit of 16 */ + if (tcp_tcbhashsize < 16) { + tcp_tcbhashsize = 16; + } + printf("WARNING: TCB hash size not a power of 2, " + "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_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"); - tcp_reass_maxseg = nmbclusters / 16; + 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); + + 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 */ - if (max_protohdr < TCP_MINPROTOHDR) - max_protohdr = TCP_MINPROTOHDR; - if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) + if (max_protohdr < TCP_MINPROTOHDR) { + _max_protohdr = TCP_MINPROTOHDR; + _max_protohdr = max_protohdr; /* round it up */ + } + if (max_linkhdr + max_protohdr > MCLBYTES) panic("tcp_init"); #undef TCP_MINPROTOHDR - dummy_tcb.t_state = TCP_NSTATES; - dummy_tcb.t_flags = 0; - tcbinfo.dummy_cb = (caddr_t) &dummy_tcb; - /* - * allocate lock group attribute and group for tcp pcb mutexes - */ - pcbinfo->mtx_grp_attr = lck_grp_attr_alloc_init(); - lck_grp_attr_setdefault(pcbinfo->mtx_grp_attr); - pcbinfo->mtx_grp = lck_grp_alloc_init("tcppcb", pcbinfo->mtx_grp_attr); - + /* Initialize time wait and timer lists */ + TAILQ_INIT(&tcp_tw_tailq); + + bzero(&tcp_timer_list, sizeof(tcp_timer_list)); + LIST_INIT(&tcp_timer_list.lhead); /* - * allocate the lock attribute for tcp pcb mutexes + * allocate lock group attribute, group and attribute for the tcp timer list */ - pcbinfo->mtx_attr = lck_attr_alloc_init(); - lck_attr_setdefault(pcbinfo->mtx_attr); - - if ((pcbinfo->mtx = lck_rw_alloc_init(pcbinfo->mtx_grp, pcbinfo->mtx_attr)) == NULL) { - printf("tcp_init: mutex not alloced!\n"); - return; /* pretty much dead if this fails... */ + tcp_timer_list.mtx_grp_attr = lck_grp_attr_alloc_init(); + tcp_timer_list.mtx_grp = lck_grp_alloc_init("tcptimerlist", tcp_timer_list.mtx_grp_attr); + tcp_timer_list.mtx_attr = lck_attr_alloc_init(); + if ((tcp_timer_list.mtx = lck_mtx_alloc_init(tcp_timer_list.mtx_grp, tcp_timer_list.mtx_attr)) == NULL) { + panic("failed to allocate memory for tcp_timer_list.mtx\n"); + }; + if ((tcp_timer_list.call = thread_call_allocate(tcp_run_timerlist, NULL)) == NULL) { + panic("failed to allocate call entry 1 in tcp_init\n"); } + /* + * allocate lock group attribute, group and attribute for tcp_uptime_lock + */ + tcp_uptime_mtx_grp_attr = lck_grp_attr_alloc_init(); + tcp_uptime_mtx_grp = lck_grp_alloc_init("tcpuptime", tcp_uptime_mtx_grp_attr); + tcp_uptime_mtx_attr = lck_attr_alloc_init(); + tcp_uptime_lock = lck_spin_alloc_init(tcp_uptime_mtx_grp, tcp_uptime_mtx_attr); - in_pcb_nat_init(&tcbinfo, AF_INET, IPPROTO_TCP, SOCK_STREAM); - - delack_bitmask = _MALLOC((4 * hashsize)/32, M_PCB, M_WAITOK); - if (delack_bitmask == 0) - panic("Delack Memory"); + /* Initialize TCP LRO data structures */ + tcp_lro_init(); - for (i=0; i < (tcbinfo.hashsize / 32); i++) - delack_bitmask[i] = 0; + /* Initialize TCP Cache */ + tcp_cache_init(); - for (i=0; i < N_TIME_WAIT_SLOTS; i++) { - LIST_INIT(&time_wait_slots[i]); + /* + * If more than 60 MB of mbuf pool is available, increase the + * maximum allowed receive and send socket buffer size. + */ + if (nmbclusters > 30720) { + tcp_autorcvbuf_max = 1024 * 1024; + tcp_autosndbuf_max = 1024 * 1024; } } @@ -372,14 +640,15 @@ tcp_fillheaders(tp, ip_ptr, tcp_ptr) ip6 = (struct ip6_hdr *)ip_ptr; ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | - (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); + (inp->inp_flow & IPV6_FLOWINFO_MASK); ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | (IPV6_VERSION & IPV6_VERSION_MASK); ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_plen = sizeof(struct tcphdr); ip6->ip6_src = inp->in6p_laddr; ip6->ip6_dst = inp->in6p_faddr; - tcp_hdr->th_sum = 0; + tcp_hdr->th_sum = in6_pseudo(&inp->in6p_laddr, &inp->in6p_faddr, + htonl(sizeof (struct tcphdr) + IPPROTO_TCP)); } else #endif { @@ -447,15 +716,10 @@ tcp_maketemplate(tp) * NOTE: If m != NULL, then ti must point to *inside* the mbuf. */ void -tcp_respond(tp, ipgen, th, m, ack, seq, flags) - struct tcpcb *tp; - void *ipgen; - register struct tcphdr *th; - register struct mbuf *m; - tcp_seq ack, seq; - int flags; +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) { - register int tlen; + int tlen; int win = 0; struct route *ro = 0; struct route sro; @@ -467,7 +731,7 @@ tcp_respond(tp, ipgen, th, m, ack, seq, flags) struct ip6_hdr *ip6; int isipv6; #endif /* INET6 */ - int ipflags = 0; + struct ifnet *outif; #if INET6 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6; @@ -477,9 +741,9 @@ tcp_respond(tp, ipgen, th, m, ack, seq, flags) if (tp) { if (!(flags & TH_RST)) { - win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); - if (win > (long)TCP_MAXWIN << tp->rcv_scale) - win = (long)TCP_MAXWIN << tp->rcv_scale; + win = tcp_sbspace(tp); + if (win > (int32_t)TCP_MAXWIN << tp->rcv_scale) + win = (int32_t)TCP_MAXWIN << tp->rcv_scale; } #if INET6 if (isipv6) @@ -500,25 +764,34 @@ tcp_respond(tp, ipgen, th, m, ack, seq, flags) } } if (m == 0) { - m = m_gethdr(M_DONTWAIT, MT_HEADER); + m = m_gethdr(M_DONTWAIT, MT_HEADER); /* MAC-OK */ if (m == NULL) return; tlen = 0; m->m_data += max_linkhdr; #if INET6 if (isipv6) { + VERIFY((MHLEN - max_linkhdr) >= + (sizeof (*ip6) + sizeof (*nth))); bcopy((caddr_t)ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(m, struct ip6_hdr *); - nth = (struct tcphdr *)(ip6 + 1); + nth = (struct tcphdr *)(void *)(ip6 + 1); } else #endif /* INET6 */ { + VERIFY((MHLEN - max_linkhdr) >= + (sizeof (*ip) + sizeof (*nth))); bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); ip = mtod(m, struct ip *); - nth = (struct tcphdr *)(ip + 1); + nth = (struct tcphdr *)(void *)(ip + 1); } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); +#if MPTCP + if ((tp) && (tp->t_mpflags & TMPF_RESET)) + flags = (TH_RST | TH_ACK); + else +#endif flags = TH_ACK; } else { m_freem(m->m_next); @@ -529,13 +802,17 @@ tcp_respond(tp, ipgen, th, m, ack, seq, flags) #define xchg(a,b,type) { type t; t=a; a=b; b=t; } #if INET6 if (isipv6) { + /* Expect 32-bit aligned IP on strict-align platforms */ + IP6_HDR_STRICT_ALIGNMENT_CHECK(ip6); xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); - nth = (struct tcphdr *)(ip6 + 1); + nth = (struct tcphdr *)(void *)(ip6 + 1); } else #endif /* INET6 */ { + /* Expect 32-bit aligned IP on strict-align platforms */ + IP_HDR_STRICT_ALIGNMENT_CHECK(ip); xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); - nth = (struct tcphdr *)(ip + 1); + nth = (struct tcphdr *)(void *)(ip + 1); } if (th != nth) { /* @@ -564,6 +841,22 @@ tcp_respond(tp, ipgen, th, m, ack, seq, flags) 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); + } +#endif + nth->th_seq = htonl(seq); nth->th_ack = htonl(ack); nth->th_x2 = 0; @@ -577,9 +870,10 @@ tcp_respond(tp, ipgen, th, m, ack, seq, flags) #if INET6 if (isipv6) { nth->th_sum = 0; - nth->th_sum = in6_cksum(m, IPPROTO_TCP, - sizeof(struct ip6_hdr), - tlen - sizeof(struct ip6_hdr)); + nth->th_sum = in6_pseudo(&ip6->ip6_src, &ip6->ip6_dst, + htonl((tlen - sizeof (struct ip6_hdr)) + IPPROTO_TCP)); + m->m_pkthdr.csum_flags = CSUM_TCPIPV6; + m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, ro6 && ro6->ro_rt ? ro6->ro_rt->rt_ifp : @@ -596,26 +890,97 @@ tcp_respond(tp, ipgen, th, m, ack, seq, flags) 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); +#endif /* NECP */ + #if IPSEC - if (ipsec_bypass == 0 && ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) { + if (tp != NULL && tp->t_inpcb->inp_sp != NULL && + ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) { m_freem(m); return; } #endif + + if (tp != NULL) { + u_int32_t svc_flags = 0; + if (isipv6) { + svc_flags |= PKT_SCF_IPV6; + } + set_packet_service_class(m, tp->t_inpcb->inp_socket, + MBUF_SC_UNSPEC, svc_flags); + + /* Embed flowhash and flow control flags */ + m->m_pkthdr.pkt_flowsrc = FLOWSRC_INPCB; + m->m_pkthdr.pkt_flowid = tp->t_inpcb->inp_flowhash; + m->m_pkthdr.pkt_flags |= PKTF_FLOW_ID | PKTF_FLOW_LOCALSRC; +#if MPTCP + /* Disable flow advisory when using MPTCP. */ + if (!(tp->t_mpflags & TMPF_MPTCP_TRUE)) +#endif /* MPTCP */ + m->m_pkthdr.pkt_flags |= PKTF_FLOW_ADV; + m->m_pkthdr.pkt_proto = IPPROTO_TCP; + } + #if INET6 if (isipv6) { - (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL, 0); - if (ro6 == &sro6 && ro6->ro_rt) { - rtfree(ro6->ro_rt); - ro6->ro_rt = NULL; - } + struct ip6_out_args ip6oa = { tra->ifscope, { 0 }, + IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR, 0 }; + + if (tra->ifscope != IFSCOPE_NONE) + ip6oa.ip6oa_flags |= IP6OAF_BOUND_IF; + if (tra->nocell) + ip6oa.ip6oa_flags |= IP6OAF_NO_CELLULAR; + if (tra->noexpensive) + ip6oa.ip6oa_flags |= IP6OAF_NO_EXPENSIVE; + if (tra->awdl_unrestricted) + ip6oa.ip6oa_flags |= IP6OAF_AWDL_UNRESTRICTED; + + (void) ip6_output(m, NULL, ro6, IPV6_OUTARGS, NULL, + NULL, &ip6oa); + + if (tp != NULL && ro6 != NULL && ro6->ro_rt != NULL && + (outif = ro6->ro_rt->rt_ifp) != + tp->t_inpcb->in6p_last_outifp) + tp->t_inpcb->in6p_last_outifp = outif; + + if (ro6 == &sro6) + ROUTE_RELEASE(ro6); } else #endif /* INET6 */ { - (void) ip_output_list(m, 0, NULL, ro, ipflags, NULL); - if (ro == &sro && ro->ro_rt) { - rtfree(ro->ro_rt); - ro->ro_rt = NULL; + struct ip_out_args ipoa = { tra->ifscope, { 0 }, + IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR, 0 }; + + if (tra->ifscope != IFSCOPE_NONE) + ipoa.ipoa_flags |= IPOAF_BOUND_IF; + if (tra->nocell) + ipoa.ipoa_flags |= IPOAF_NO_CELLULAR; + if (tra->noexpensive) + ipoa.ipoa_flags |= IPOAF_NO_EXPENSIVE; + if (tra->awdl_unrestricted) + ipoa.ipoa_flags |= IPOAF_AWDL_UNRESTRICTED; + + if (ro != &sro) { + /* Copy the cached route and take an extra reference */ + inp_route_copyout(tp->t_inpcb, &sro); + } + /* + * For consistency, pass a local route copy. + */ + (void) ip_output(m, NULL, &sro, IP_OUTARGS, NULL, &ipoa); + + if (tp != NULL && sro.ro_rt != NULL && + (outif = sro.ro_rt->rt_ifp) != + tp->t_inpcb->inp_last_outifp) + tp->t_inpcb->inp_last_outifp = outif; + + if (ro != &sro) { + /* Synchronize cached PCB route */ + inp_route_copyin(tp->t_inpcb, &sro); + } else { + ROUTE_RELEASE(&sro); } } } @@ -637,13 +1002,15 @@ tcp_newtcpcb(inp) int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ - if (so->cached_in_sock_layer == 0) { - it = (struct inp_tp *)inp; + calculate_tcp_clock(); + + if ((so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) == 0) { + it = (struct inp_tp *)(void *)inp; tp = &it->tcb; + } else { + tp = (struct tcpcb *)(void *)inp->inp_saved_ppcb; } - else - tp = (struct tcpcb *) inp->inp_saved_ppcb; - + bzero((char *) tp, sizeof(struct tcpcb)); LIST_INIT(&tp->t_segq); tp->t_maxseg = tp->t_maxopd = @@ -654,8 +1021,11 @@ tcp_newtcpcb(inp) if (tcp_do_rfc1323) tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); - tp->sack_enable = tcp_do_sack; + if (tcp_do_sack) + tp->t_flagsext |= TF_SACK_ENABLE; + TAILQ_INIT(&tp->snd_holes); + SLIST_INIT(&tp->t_rxt_segments); tp->t_inpcb = inp; /* XXX */ /* * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no @@ -664,11 +1034,34 @@ tcp_newtcpcb(inp) */ tp->t_srtt = TCPTV_SRTTBASE; tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; - tp->t_rttmin = TCPTV_MIN; + tp->t_rttmin = tcp_TCPTV_MIN; tp->t_rxtcur = TCPTV_RTOBASE; - tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; + + if (tcp_use_newreno) + /* use newreno by default */ + tp->tcp_cc_index = TCP_CC_ALGO_NEWRENO_INDEX; + else + tp->tcp_cc_index = TCP_CC_ALGO_CUBIC_INDEX; + + tcp_cc_allocate_state(tp); + + if (CC_ALGO(tp)->init != NULL) + CC_ALGO(tp)->init(tp); + + tp->snd_cwnd = TCP_CC_CWND_INIT_BYTES; tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; - tp->t_rcvtime = 0; + tp->snd_ssthresh_prev = TCP_MAXWIN << TCP_MAX_WINSHIFT; + tp->t_rcvtime = tcp_now; + tp->tentry.timer_start = tcp_now; + tp->t_persist_timeout = tcp_max_persist_timeout; + tp->t_persist_stop = 0; + tp->t_flagsext |= TF_RCVUNACK_WAITSS; + tp->t_rexmtthresh = tcprexmtthresh; + + /* Clear time wait tailq entry */ + tp->t_twentry.tqe_next = NULL; + tp->t_twentry.tqe_prev = NULL; + /* * IPv4 TTL initialization is necessary for an IPv6 socket as well, * because the socket may be bound to an IPv6 wildcard address, @@ -690,8 +1083,13 @@ tcp_drop(tp, errno) int errno; { struct socket *so = tp->t_inpcb->inp_socket; - +#if CONFIG_DTRACE + struct inpcb *inp = tp->t_inpcb; +#endif + if (TCPS_HAVERCVDSYN(tp->t_state)) { + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_CLOSED); tp->t_state = TCPS_CLOSED; (void) tcp_output(tp); tcpstat.tcps_drops++; @@ -703,6 +1101,84 @@ tcp_drop(tp, errno) return (tcp_close(tp)); } +void +tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt) +{ + u_int32_t rtt = rt->rt_rmx.rmx_rtt; + int isnetlocal = (tp->t_flags & TF_LOCAL); + + if (rtt != 0) { + /* + * XXX the lock bit for RTT indicates that the value + * is also a minimum value; this is subject to time. + */ + if (rt->rt_rmx.rmx_locks & RTV_RTT) + tp->t_rttmin = rtt / (RTM_RTTUNIT / TCP_RETRANSHZ); + else + tp->t_rttmin = isnetlocal ? tcp_TCPTV_MIN : TCPTV_REXMTMIN; + tp->t_srtt = rtt / (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE)); + tcpstat.tcps_usedrtt++; + if (rt->rt_rmx.rmx_rttvar) { + tp->t_rttvar = rt->rt_rmx.rmx_rttvar / + (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE)); + tcpstat.tcps_usedrttvar++; + } else { + /* default variation is +- 1 rtt */ + tp->t_rttvar = + tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; + } + TCPT_RANGESET(tp->t_rxtcur, + ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, + tp->t_rttmin, TCPTV_REXMTMAX, + TCP_ADD_REXMTSLOP(tp)); + } +} + +static inline void +tcp_update_ecn_perf_stats(struct tcpcb *tp, + struct if_tcp_ecn_perf_stat *stat) +{ + u_int64_t curval, oldval; + struct inpcb *inp = tp->t_inpcb; + stat->total_txpkts += inp->inp_stat->txpackets; + stat->total_rxpkts += inp->inp_stat->rxpackets; + stat->total_rxmitpkts += tp->t_stat.rxmitpkts; + stat->total_oopkts += tp->t_rcvoopack; + stat->total_reorderpkts += (tp->t_reordered_pkts + tp->t_pawsdrop + + tp->t_dsack_sent + tp->t_dsack_recvd); + + /* Average RTT */ + curval = (tp->t_srtt >> TCP_RTT_SHIFT); + if (curval > 0 && tp->t_rttupdated >= 16) { + if (stat->rtt_avg == 0) { + stat->rtt_avg = curval; + } else { + oldval = stat->rtt_avg; + stat->rtt_avg = + ((oldval << 4) - oldval + curval) >> 4; + } + } + + /* RTT variance */ + curval = tp->t_rttvar >> TCP_RTTVAR_SHIFT; + if (curval > 0 && tp->t_rttupdated >= 16) { + if (stat->rtt_var == 0) { + stat->rtt_var = curval; + } else { + oldval = stat->rtt_var; + stat->rtt_var = + ((oldval << 4) - oldval + curval) >> 4; + } + } + + /* Total number of SACK recovery episodes */ + stat->sack_episodes += tp->t_sack_recovery_episode; + + if (inp->inp_socket->so_error == ECONNRESET) + stat->rst_drop++; + return; +} + /* * Close a TCP control block: * discard all space held by the tcp @@ -718,31 +1194,46 @@ tcp_close(tp) #if INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ - register struct rtentry *rt; + struct route *ro; + struct rtentry *rt; int dosavessthresh; - if ( inp->inp_ppcb == NULL) /* tcp_close was called previously, bail */ - return NULL; - - /* Clear the timers before we delete the PCB. */ - { - int i; - for (i = 0; i < TCPT_NTIMERS; i++) { - tp->t_timer[i] = 0; - } - } + /* tcp_close was called previously, bail */ + if (inp->inp_ppcb == NULL) + return(NULL); + tcp_canceltimers(tp); KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_START, tp,0,0,0,0); - switch (tp->t_state) - { - case TCPS_ESTABLISHED: - case TCPS_FIN_WAIT_1: - case TCPS_CLOSING: - case TCPS_CLOSE_WAIT: - case TCPS_LAST_ACK: - break; + + /* + * If another thread for this tcp is currently in ip (indicated by + * the TF_SENDINPROG flag), defer the cleanup until after it returns + * back to tcp. This is done to serialize the close until after all + * pending output is finished, in order to avoid having the PCB be + * detached and the cached route cleaned, only for ip to cache the + * route back into the PCB again. Note that we've cleared all the + * timers at this point. Set TF_CLOSING to indicate to tcp_output() + * that is should call us again once it returns from ip; at that + * point both flags should be cleared and we can proceed further + * with the cleanup. + */ + if ((tp->t_flags & TF_CLOSING) || + inp->inp_sndinprog_cnt > 0) { + tp->t_flags |= TF_CLOSING; + return (NULL); } + DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp, + struct tcpcb *, tp, int32_t, TCPS_CLOSED); + +#if INET6 + ro = (isipv6 ? (struct route *)&inp->in6p_route : &inp->inp_route); +#else + ro = &inp->inp_route; +#endif + rt = ro->ro_rt; + if (rt != NULL) + RT_LOCK_SPIN(rt); /* * If we got enough samples through the srtt filter, @@ -756,32 +1247,33 @@ tcp_close(tp) * update anything that the user "locked". */ if (tp->t_rttupdated >= 16) { - register u_long i = 0; + register u_int32_t i = 0; + #if INET6 if (isipv6) { struct sockaddr_in6 *sin6; - if ((rt = inp->in6p_route.ro_rt) == NULL) + if (rt == NULL) goto no_valid_rt; - sin6 = (struct sockaddr_in6 *)rt_key(rt); + sin6 = (struct sockaddr_in6 *)(void *)rt_key(rt); if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) goto no_valid_rt; } else #endif /* INET6 */ - rt = inp->inp_route.ro_rt; - if (rt == NULL || - ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr - == INADDR_ANY || rt->generation_id != route_generation) { - if (tp->t_state >= TCPS_CLOSE_WAIT) - tp->t_state = TCPS_CLOSING; - + 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, + int32_t, TCPS_CLOSED); + tp->t_state = TCPS_CLOSED; goto no_valid_rt; } + RT_LOCK_ASSERT_HELD(rt); if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { i = tp->t_srtt * - (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); + (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE)); if (rt->rt_rmx.rmx_rtt && i) /* * filter this update to half the old & half @@ -797,7 +1289,7 @@ tcp_close(tp) } if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { i = tp->t_rttvar * - (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); + (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE)); if (rt->rt_rmx.rmx_rttvar && i) rt->rt_rmx.rmx_rttvar = (rt->rt_rmx.rmx_rttvar + i) / 2; @@ -834,7 +1326,7 @@ tcp_close(tp) i = (i + tp->t_maxseg / 2) / tp->t_maxseg; if (i < 2) i = 2; - i *= (u_long)(tp->t_maxseg + + i *= (u_int32_t)(tp->t_maxseg + #if INET6 (isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : @@ -852,39 +1344,191 @@ tcp_close(tp) tcpstat.tcps_cachedssthresh++; } } - rt = inp->inp_route.ro_rt; - if (rt) { - /* - * mark route for deletion if no information is - * cached. - */ - if ((tp->t_flags & TF_LQ_OVERFLOW) && tcp_lq_overflow && - ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0)){ - if (rt->rt_rmx.rmx_rtt == 0) - rt->rt_flags |= RTF_DELCLONE; + + /* + * Mark route for deletion if no information is cached. + */ + if (rt != NULL && (so->so_flags & SOF_OVERFLOW) && tcp_lq_overflow) { + if (!(rt->rt_rmx.rmx_locks & RTV_RTT) && + rt->rt_rmx.rmx_rtt == 0) { + rt->rt_flags |= RTF_DELCLONE; } } - no_valid_rt: + +no_valid_rt: + if (rt != NULL) + RT_UNLOCK(rt); + /* free the reassembly queue, if any */ (void) tcp_freeq(tp); + /* Collect ECN related statistics */ + if (tp->ecn_flags & TE_SETUPSENT) { + if (tp->ecn_flags & TE_CLIENT_SETUP) { + INP_INC_IFNET_STAT(inp, ecn_client_setup); + if (TCP_ECN_ENABLED(tp)) { + INP_INC_IFNET_STAT(inp, + ecn_client_success); + } else if (tp->ecn_flags & TE_LOST_SYN) { + INP_INC_IFNET_STAT(inp, ecn_syn_lost); + } else { + INP_INC_IFNET_STAT(inp, + ecn_peer_nosupport); + } + } else { + INP_INC_IFNET_STAT(inp, ecn_server_setup); + if (TCP_ECN_ENABLED(tp)) { + INP_INC_IFNET_STAT(inp, + ecn_server_success); + } else if (tp->ecn_flags & TE_LOST_SYNACK) { + INP_INC_IFNET_STAT(inp, + ecn_synack_lost); + } else { + INP_INC_IFNET_STAT(inp, + ecn_peer_nosupport); + } + } + } else { + INP_INC_IFNET_STAT(inp, ecn_off_conn); + } + if (TCP_ECN_ENABLED(tp)) { + if (tp->ecn_flags & TE_RECV_ECN_CE) { + tcpstat.tcps_ecn_conn_recv_ce++; + INP_INC_IFNET_STAT(inp, ecn_conn_recv_ce); + } + if (tp->ecn_flags & TE_RECV_ECN_ECE) { + tcpstat.tcps_ecn_conn_recv_ece++; + INP_INC_IFNET_STAT(inp, ecn_conn_recv_ece); + } + if (tp->ecn_flags & (TE_RECV_ECN_CE | TE_RECV_ECN_ECE)) { + if (tp->t_stat.txretransmitbytes > 0 || + tp->t_stat.rxoutoforderbytes > 0) { + tcpstat.tcps_ecn_conn_pl_ce++; + INP_INC_IFNET_STAT(inp, ecn_conn_plce); + } else { + tcpstat.tcps_ecn_conn_nopl_ce++; + INP_INC_IFNET_STAT(inp, ecn_conn_noplce); + } + } else { + if (tp->t_stat.txretransmitbytes > 0 || + tp->t_stat.rxoutoforderbytes > 0) { + tcpstat.tcps_ecn_conn_plnoce++; + INP_INC_IFNET_STAT(inp, ecn_conn_plnoce); + } + } + } + + /* Aggregate performance stats */ + if (inp->inp_last_outifp != NULL && !(tp->t_flags & TF_LOCAL)) { + struct ifnet *ifp = inp->inp_last_outifp; + ifnet_lock_shared(ifp); + if ((ifp->if_refflags & (IFRF_ATTACHED | IFRF_DETACHING)) == + IFRF_ATTACHED) { + if (inp->inp_vflag & INP_IPV6) { + ifp->if_ipv6_stat->timestamp = net_uptime(); + if (TCP_ECN_ENABLED(tp)) { + tcp_update_ecn_perf_stats(tp, + &ifp->if_ipv6_stat->ecn_on); + } else { + tcp_update_ecn_perf_stats(tp, + &ifp->if_ipv6_stat->ecn_off); + } + } else { + ifp->if_ipv4_stat->timestamp = net_uptime(); + if (TCP_ECN_ENABLED(tp)) { + tcp_update_ecn_perf_stats(tp, + &ifp->if_ipv4_stat->ecn_on); + } else { + tcp_update_ecn_perf_stats(tp, + &ifp->if_ipv4_stat->ecn_off); + } + } + } + ifnet_lock_done(ifp); + } + tcp_free_sackholes(tp); + if (tp->t_bwmeas != NULL) { + tcp_bwmeas_free(tp); + } + tcp_rxtseg_clean(tp); + /* Free the packet list */ + if (tp->t_pktlist_head != NULL) + m_freem_list(tp->t_pktlist_head); + TCP_PKTLIST_CLEAR(tp); + +#if MPTCP + /* Clear MPTCP state */ + if ((so->so_flags & SOF_MPTCP_TRUE) || + (so->so_flags & SOF_MP_SUBFLOW)) { + soevent(so, (SO_FILT_HINT_LOCKED | SO_FILT_HINT_DELETEOK)); + } + tp->t_mpflags = 0; + tp->t_mptcb = NULL; +#endif /* MPTCP */ -#ifdef __APPLE__ - if (so->cached_in_sock_layer) + if (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) inp->inp_saved_ppcb = (caddr_t) tp; -#endif - soisdisconnected(so); + tp->t_state = TCPS_CLOSED; + + /* Issue a wakeup before detach so that we don't miss + * a wakeup + */ + sodisconnectwakeup(so); + + /* + * Clean up any LRO state + */ + if (tp->t_flagsext & TF_LRO_OFFLOADED) { + tcp_lro_remove_state(inp->inp_laddr, inp->inp_faddr, + inp->inp_lport, inp->inp_fport); + tp->t_flagsext &= ~TF_LRO_OFFLOADED; + } + + /* + * If this is a socket that does not want to wakeup the device + * for it's traffic, the application might need to know that the + * socket is closed, send a notification. + */ + if ((so->so_options & SO_NOWAKEFROMSLEEP) && + inp->inp_state != INPCB_STATE_DEAD && + !(inp->inp_flags2 & INP2_TIMEWAIT)) + socket_post_kev_msg_closed(so); + + if (CC_ALGO(tp)->cleanup != NULL) { + CC_ALGO(tp)->cleanup(tp); + } + + if (tp->t_ccstate != NULL) { + zfree(tcp_cc_zone, tp->t_ccstate); + tp->t_ccstate = NULL; + } + tp->tcp_cc_index = TCP_CC_ALGO_NONE; + + /* Can happen if we close the socket before receiving the third ACK */ + if ((tp->t_tfo_flags & TFO_F_COOKIE_VALID)) { + OSDecrementAtomic(&tcp_tfo_halfcnt); + + /* Panic if something has gone terribly wrong. */ + VERIFY(tcp_tfo_halfcnt >= 0); + + tp->t_tfo_flags &= ~TFO_F_COOKIE_VALID; + } + #if INET6 - if (INP_CHECK_SOCKAF(so, AF_INET6)) + if (SOCK_CHECK_DOM(so, PF_INET6)) in6_pcbdetach(inp); else #endif /* INET6 */ in_pcbdetach(inp); + + /* Call soisdisconnected after detach because it might unlock the socket */ + soisdisconnected(so); tcpstat.tcps_closed++; - KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_END, tcpstat.tcps_closed,0,0,0,0); - return ((struct tcpcb *)0); + KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_END, + tcpstat.tcps_closed, 0, 0, 0, 0); + return(NULL); } int @@ -898,49 +1542,51 @@ tcp_freeq(tp) while((q = LIST_FIRST(&tp->t_segq)) != NULL) { LIST_REMOVE(q, tqe_q); m_freem(q->tqe_m); - FREE(q, M_TSEGQ); - tcp_reass_qsize--; + zfree(tcp_reass_zone, q); rv = 1; } + tp->t_reassqlen = 0; return (rv); } -void -tcp_drain() -{ + /* - * ###LD 05/19/04 locking issue, tcpdrain is disabled, deadlock situation with tcbinfo.mtx + * 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 */ - if (do_tcpdrain) - { - struct inpcb *inpb; - struct tcpcb *tcpb; - struct tseg_qent *te; +void +tcp_drain() +{ + struct inpcb *inp; + struct tcpcb *tp; - /* - * Walk the tcpbs, if existing, and flush the reassembly queue, - * if there is one... - * XXX: The "Net/3" implementation doesn't imply that the TCP - * reassembly queue should be flushed, but in a situation - * where we're really low on mbufs, this is potentially - * usefull. - */ - lck_rw_lock_exclusive(tcbinfo.mtx); - for (inpb = LIST_FIRST(tcbinfo.listhead); inpb; - inpb = LIST_NEXT(inpb, inp_list)) { - if ((tcpb = intotcpcb(inpb))) { - while ((te = LIST_FIRST(&tcpb->t_segq)) - != NULL) { - LIST_REMOVE(te, tqe_q); - m_freem(te->tqe_m); - FREE(te, M_TSEGQ); - tcp_reass_qsize--; - } - } - } - lck_rw_done(tcbinfo.mtx); + if (!lck_rw_try_lock_exclusive(tcbinfo.ipi_lock)) + return; + + LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) { + if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) != + WNT_STOPUSING) { + tcp_lock(inp->inp_socket, 1, 0); + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) + == WNT_STOPUSING) { + /* lost a race, try the next one */ + tcp_unlock(inp->inp_socket, 1, 0); + continue; + } + tp = intotcpcb(inp); + if (do_tcpdrain) + tcp_freeq(tp); + + so_drain_extended_bk_idle(inp->inp_socket); + + tcp_unlock(inp->inp_socket, 1, 0); + } } + lck_rw_done(tcbinfo.ipi_lock); + } /* @@ -986,11 +1632,100 @@ tcp_notify(inp, error) #endif } +struct bwmeas* +tcp_bwmeas_alloc(struct tcpcb *tp) +{ + struct bwmeas *elm; + elm = zalloc(tcp_bwmeas_zone); + if (elm == NULL) + return(elm); + + bzero(elm, bwmeas_elm_size); + elm->bw_minsizepkts = TCP_BWMEAS_BURST_MINSIZE; + elm->bw_maxsizepkts = TCP_BWMEAS_BURST_MAXSIZE; + elm->bw_minsize = elm->bw_minsizepkts * tp->t_maxseg; + elm->bw_maxsize = elm->bw_maxsizepkts * tp->t_maxseg; + return(elm); +} + +void +tcp_bwmeas_free(struct tcpcb* tp) +{ + zfree(tcp_bwmeas_zone, tp->t_bwmeas); + tp->t_bwmeas = NULL; + tp->t_flagsext &= ~(TF_MEASURESNDBW); +} + +/* + * tcpcb_to_otcpcb copies specific bits of a tcpcb to a otcpcb format. + * The otcpcb data structure is passed to user space and must not change. + */ +static void +tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp) +{ + otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first); + otp->t_dupacks = tp->t_dupacks; + otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT]; + otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST]; + otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP]; + otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL]; + otp->t_inpcb = (_TCPCB_PTR(struct inpcb *))VM_KERNEL_ADDRPERM(tp->t_inpcb); + otp->t_state = tp->t_state; + otp->t_flags = tp->t_flags; + otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0; + otp->snd_una = tp->snd_una; + otp->snd_max = tp->snd_max; + otp->snd_nxt = tp->snd_nxt; + otp->snd_up = tp->snd_up; + otp->snd_wl1 = tp->snd_wl1; + otp->snd_wl2 = tp->snd_wl2; + otp->iss = tp->iss; + otp->irs = tp->irs; + otp->rcv_nxt = tp->rcv_nxt; + otp->rcv_adv = tp->rcv_adv; + otp->rcv_wnd = tp->rcv_wnd; + otp->rcv_up = tp->rcv_up; + otp->snd_wnd = tp->snd_wnd; + otp->snd_cwnd = tp->snd_cwnd; + otp->snd_ssthresh = tp->snd_ssthresh; + otp->t_maxopd = tp->t_maxopd; + otp->t_rcvtime = tp->t_rcvtime; + otp->t_starttime = tp->t_starttime; + otp->t_rtttime = tp->t_rtttime; + otp->t_rtseq = tp->t_rtseq; + otp->t_rxtcur = tp->t_rxtcur; + otp->t_maxseg = tp->t_maxseg; + otp->t_srtt = tp->t_srtt; + otp->t_rttvar = tp->t_rttvar; + otp->t_rxtshift = tp->t_rxtshift; + otp->t_rttmin = tp->t_rttmin; + otp->t_rttupdated = tp->t_rttupdated; + otp->max_sndwnd = tp->max_sndwnd; + otp->t_softerror = tp->t_softerror; + otp->t_oobflags = tp->t_oobflags; + otp->t_iobc = tp->t_iobc; + otp->snd_scale = tp->snd_scale; + otp->rcv_scale = tp->rcv_scale; + otp->request_r_scale = tp->request_r_scale; + otp->requested_s_scale = tp->requested_s_scale; + otp->ts_recent = tp->ts_recent; + otp->ts_recent_age = tp->ts_recent_age; + otp->last_ack_sent = tp->last_ack_sent; + otp->cc_send = 0; + otp->cc_recv = 0; + otp->snd_recover = tp->snd_recover; + otp->snd_cwnd_prev = tp->snd_cwnd_prev; + otp->snd_ssthresh_prev = tp->snd_ssthresh_prev; + otp->t_badrxtwin = 0; +} + static int tcp_pcblist SYSCTL_HANDLER_ARGS { - int error, i, n; +#pragma unused(oidp, arg1, arg2) + int error, i = 0, n; struct inpcb *inp, **inp_list; + struct tcpcb *tp; inp_gen_t gencnt; struct xinpgen xig; @@ -998,17 +1733,17 @@ tcp_pcblist SYSCTL_HANDLER_ARGS * The process of preparing the TCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ - lck_rw_lock_shared(tcbinfo.mtx); + lck_rw_lock_shared(tcbinfo.ipi_lock); if (req->oldptr == USER_ADDR_NULL) { n = tcbinfo.ipi_count; req->oldidx = 2 * (sizeof xig) + (n + n/8) * sizeof(struct xtcpcb); - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return 0; } if (req->newptr != USER_ADDR_NULL) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return EPERM; } @@ -1025,38 +1760,45 @@ tcp_pcblist SYSCTL_HANDLER_ARGS xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return error; } - /* - * We are done if there is no pcb - */ - if (n == 0) { - lck_rw_done(tcbinfo.mtx); - return 0; + /* + * We are done if there is no pcb + */ + if (n == 0) { + lck_rw_done(tcbinfo.ipi_lock); + return 0; } inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK); if (inp_list == 0) { - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return ENOMEM; } - for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n; - inp = LIST_NEXT(inp, inp_list)) { -#ifdef __APPLE__ - if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) -#else - if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp)) -#endif + LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) { + if (inp->inp_gencnt <= gencnt && + inp->inp_state != INPCB_STATE_DEAD) inp_list[i++] = inp; + if (i >= n) break; } + + TAILQ_FOREACH(tp, &tcp_tw_tailq, t_twentry) { + inp = tp->t_inpcb; + if (inp->inp_gencnt <= gencnt && + inp->inp_state != INPCB_STATE_DEAD) + inp_list[i++] = inp; + if (i >= n) break; + } + n = i; error = 0; for (i = 0; i < n; i++) { inp = inp_list[i]; - if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) { + if (inp->inp_gencnt <= gencnt && + inp->inp_state != INPCB_STATE_DEAD) { struct xtcpcb xt; caddr_t inp_ppcb; @@ -1066,10 +1808,12 @@ tcp_pcblist SYSCTL_HANDLER_ARGS inpcb_to_compat(inp, &xt.xt_inp); inp_ppcb = inp->inp_ppcb; if (inp_ppcb != NULL) { - bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); - } - else + tcpcb_to_otcpcb( + (struct tcpcb *)(void *)inp_ppcb, + &xt.xt_tp); + } else { bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); + } if (inp->inp_socket) sotoxsocket(inp->inp_socket, &xt.xt_socket); error = SYSCTL_OUT(req, &xt, sizeof xt); @@ -1091,92 +1835,227 @@ tcp_pcblist SYSCTL_HANDLER_ARGS error = SYSCTL_OUT(req, &xig, sizeof xig); } FREE(inp_list, M_TEMP); - lck_rw_done(tcbinfo.mtx); + lck_rw_done(tcbinfo.ipi_lock); return error; } -SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, +SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, + CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); -#ifndef __APPLE__ + +static void +tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp) +{ + otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first); + otp->t_dupacks = tp->t_dupacks; + otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT]; + otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST]; + otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP]; + otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL]; + otp->t_state = tp->t_state; + otp->t_flags = tp->t_flags; + otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0; + otp->snd_una = tp->snd_una; + otp->snd_max = tp->snd_max; + otp->snd_nxt = tp->snd_nxt; + otp->snd_up = tp->snd_up; + otp->snd_wl1 = tp->snd_wl1; + otp->snd_wl2 = tp->snd_wl2; + otp->iss = tp->iss; + otp->irs = tp->irs; + otp->rcv_nxt = tp->rcv_nxt; + otp->rcv_adv = tp->rcv_adv; + otp->rcv_wnd = tp->rcv_wnd; + otp->rcv_up = tp->rcv_up; + otp->snd_wnd = tp->snd_wnd; + otp->snd_cwnd = tp->snd_cwnd; + otp->snd_ssthresh = tp->snd_ssthresh; + otp->t_maxopd = tp->t_maxopd; + otp->t_rcvtime = tp->t_rcvtime; + otp->t_starttime = tp->t_starttime; + otp->t_rtttime = tp->t_rtttime; + otp->t_rtseq = tp->t_rtseq; + otp->t_rxtcur = tp->t_rxtcur; + otp->t_maxseg = tp->t_maxseg; + otp->t_srtt = tp->t_srtt; + otp->t_rttvar = tp->t_rttvar; + otp->t_rxtshift = tp->t_rxtshift; + otp->t_rttmin = tp->t_rttmin; + otp->t_rttupdated = tp->t_rttupdated; + otp->max_sndwnd = tp->max_sndwnd; + otp->t_softerror = tp->t_softerror; + otp->t_oobflags = tp->t_oobflags; + otp->t_iobc = tp->t_iobc; + otp->snd_scale = tp->snd_scale; + otp->rcv_scale = tp->rcv_scale; + otp->request_r_scale = tp->request_r_scale; + otp->requested_s_scale = tp->requested_s_scale; + otp->ts_recent = tp->ts_recent; + otp->ts_recent_age = tp->ts_recent_age; + otp->last_ack_sent = tp->last_ack_sent; + otp->cc_send = 0; + otp->cc_recv = 0; + otp->snd_recover = tp->snd_recover; + otp->snd_cwnd_prev = tp->snd_cwnd_prev; + otp->snd_ssthresh_prev = tp->snd_ssthresh_prev; + otp->t_badrxtwin = 0; +} + + static int -tcp_getcred(SYSCTL_HANDLER_ARGS) +tcp_pcblist64 SYSCTL_HANDLER_ARGS { - struct sockaddr_in addrs[2]; - struct inpcb *inp; - int error, s; +#pragma unused(oidp, arg1, arg2) + int error, i = 0, n; + struct inpcb *inp, **inp_list; + struct tcpcb *tp; + inp_gen_t gencnt; + struct xinpgen xig; - error = suser(req->p); - if (error) - return (error); - error = SYSCTL_IN(req, addrs, sizeof(addrs)); - if (error) - return (error); - s = splnet(); - inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, - addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); - if (inp == NULL || inp->inp_socket == NULL) { - error = ENOENT; - goto out; - } - error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(*(kauth_cred_t)0); -out: - splx(s); - return (error); + /* + * The process of preparing the TCB list is too time-consuming and + * resource-intensive to repeat twice on every request. + */ + lck_rw_lock_shared(tcbinfo.ipi_lock); + if (req->oldptr == USER_ADDR_NULL) { + n = tcbinfo.ipi_count; + req->oldidx = 2 * (sizeof xig) + + (n + n/8) * sizeof(struct xtcpcb64); + lck_rw_done(tcbinfo.ipi_lock); + return 0; + } + + if (req->newptr != USER_ADDR_NULL) { + lck_rw_done(tcbinfo.ipi_lock); + return EPERM; + } + + /* + * OK, now we're committed to doing something. + */ + gencnt = tcbinfo.ipi_gencnt; + n = tcbinfo.ipi_count; + + bzero(&xig, sizeof(xig)); + xig.xig_len = sizeof xig; + xig.xig_count = n; + xig.xig_gen = gencnt; + xig.xig_sogen = so_gencnt; + error = SYSCTL_OUT(req, &xig, sizeof xig); + if (error) { + lck_rw_done(tcbinfo.ipi_lock); + return error; + } + /* + * We are done if there is no pcb + */ + if (n == 0) { + lck_rw_done(tcbinfo.ipi_lock); + return 0; + } + + inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK); + if (inp_list == 0) { + lck_rw_done(tcbinfo.ipi_lock); + return ENOMEM; + } + + LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) { + if (inp->inp_gencnt <= gencnt && + inp->inp_state != INPCB_STATE_DEAD) + inp_list[i++] = inp; + if (i >= n) break; + } + + TAILQ_FOREACH(tp, &tcp_tw_tailq, t_twentry) { + inp = tp->t_inpcb; + if (inp->inp_gencnt <= gencnt && + inp->inp_state != INPCB_STATE_DEAD) + inp_list[i++] = inp; + if (i >= n) break; + } + + n = i; + + error = 0; + for (i = 0; i < n; i++) { + inp = inp_list[i]; + if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) { + struct xtcpcb64 xt; + + bzero(&xt, sizeof(xt)); + xt.xt_len = sizeof xt; + inpcb_to_xinpcb64(inp, &xt.xt_inpcb); + xt.xt_inpcb.inp_ppcb = (uint64_t)VM_KERNEL_ADDRPERM(inp->inp_ppcb); + if (inp->inp_ppcb != NULL) + tcpcb_to_xtcpcb64((struct tcpcb *)inp->inp_ppcb, &xt); + if (inp->inp_socket) + sotoxsocket64(inp->inp_socket, &xt.xt_inpcb.xi_socket); + error = SYSCTL_OUT(req, &xt, sizeof xt); + } + } + if (!error) { + /* + * Give the user an updated idea of our state. + * If the generation differs from what we told + * her before, she knows that something happened + * while we were processing this request, and it + * might be necessary to retry. + */ + bzero(&xig, sizeof(xig)); + xig.xig_len = sizeof xig; + xig.xig_gen = tcbinfo.ipi_gencnt; + xig.xig_sogen = so_gencnt; + xig.xig_count = tcbinfo.ipi_count; + error = SYSCTL_OUT(req, &xig, sizeof xig); + } + FREE(inp_list, M_TEMP); + lck_rw_done(tcbinfo.ipi_lock); + return error; } -SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, - 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection"); +SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist64, + CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, + tcp_pcblist64, "S,xtcpcb64", "List of active TCP connections"); + -#if INET6 static int -tcp6_getcred(SYSCTL_HANDLER_ARGS) +tcp_pcblist_n SYSCTL_HANDLER_ARGS { - struct sockaddr_in6 addrs[2]; - struct inpcb *inp; - int error, s, mapped = 0; +#pragma unused(oidp, arg1, arg2) + int error = 0; + + error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo); + + return error; +} - error = suser(req->p); - if (error) - return (error); - error = SYSCTL_IN(req, addrs, sizeof(addrs)); - if (error) - return (error); - if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { - if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) - mapped = 1; - else - return (EINVAL); - } - s = splnet(); - if (mapped == 1) - inp = in_pcblookup_hash(&tcbinfo, - *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], - addrs[1].sin6_port, - *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], - addrs[0].sin6_port, - 0, NULL); - else - inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, - addrs[1].sin6_port, - &addrs[0].sin6_addr, addrs[0].sin6_port, - 0, NULL); - if (inp == NULL || inp->inp_socket == NULL) { - error = ENOENT; - goto out; - } - error = SYSCTL_OUT(req, inp->inp_socket->so_cred, - sizeof(*(kauth_cred_t)0); -out: - splx(s); - return (error); + +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"); + + +__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); } -SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, - 0, 0, - tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection"); -#endif -#endif /* __APPLE__*/ +__private_extern__ uint32_t +tcp_count_opportunistic(unsigned int ifindex, u_int32_t flags) +{ + return inpcb_count_opportunistic(ifindex, &tcbinfo, flags); +} + +__private_extern__ uint32_t +tcp_find_anypcb_byaddr(struct ifaddr *ifa) +{ + return inpcb_find_anypcb_byaddr(ifa, &tcbinfo); +} void tcp_ctlinput(cmd, sa, vip) @@ -1184,48 +2063,110 @@ tcp_ctlinput(cmd, sa, vip) struct sockaddr *sa; void *vip; { + tcp_seq icmp_tcp_seq; struct ip *ip = vip; - struct tcphdr *th; struct in_addr faddr; struct inpcb *inp; struct tcpcb *tp; + void (*notify)(struct inpcb *, int) = tcp_notify; - tcp_seq icmp_seq; - faddr = ((struct sockaddr_in *)sa)->sin_addr; + faddr = ((struct sockaddr_in *)(void *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) return; - if (cmd == PRC_QUENCH) - notify = tcp_quench; + if ((unsigned)cmd >= PRC_NCMDS) + return; + + if (cmd == PRC_MSGSIZE) + notify = tcp_mtudisc; else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT) && ip) notify = tcp_drop_syn_sent; - else if (cmd == PRC_MSGSIZE) - notify = tcp_mtudisc; else if (PRC_IS_REDIRECT(cmd)) { ip = 0; notify = in_rtchange; } else if (cmd == PRC_HOSTDEAD) ip = 0; - else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) + /* Source quench is deprecated */ + else if (cmd == PRC_QUENCH) + return; + else if (inetctlerrmap[cmd] == 0) return; if (ip) { - th = (struct tcphdr *)((caddr_t)ip - + (IP_VHL_HL(ip->ip_vhl) << 2)); - inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, - ip->ip_src, th->th_sport, 0, NULL); + struct tcphdr th; + struct icmp *icp; + + icp = (struct icmp *)(void *) + ((caddr_t)ip - offsetof(struct icmp, icmp_ip)); + bcopy(((caddr_t)ip + (IP_VHL_HL(ip->ip_vhl) << 2)), + &th, sizeof (th)); + inp = in_pcblookup_hash(&tcbinfo, faddr, th.th_dport, + ip->ip_src, th.th_sport, 0, NULL); if (inp != NULL && inp->inp_socket != NULL) { tcp_lock(inp->inp_socket, 1, 0); if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { tcp_unlock(inp->inp_socket, 1, 0); return; } - icmp_seq = htonl(th->th_seq); + icmp_tcp_seq = htonl(th.th_seq); tp = intotcpcb(inp); - if (SEQ_GEQ(icmp_seq, tp->snd_una) && - SEQ_LT(icmp_seq, tp->snd_max)) + if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && + SEQ_LT(icmp_tcp_seq, tp->snd_max)) { + if (cmd == PRC_MSGSIZE) { + + /* + * MTU discovery: + * If we got a needfrag and there is a host route to the + * original destination, and the MTU is not locked, then + * set the MTU in the route to the suggested new value + * (if given) and then notify as usual. The ULPs will + * notice that the MTU has changed and adapt accordingly. + * If no new MTU was suggested, then we guess a new one + * less than the current value. If the new MTU is + * unreasonably small (defined by sysctl tcp_minmss), then + * we reset the MTU to the interface value and enable the + * lock bit, indicating that we are no longer doing MTU + * discovery. + */ + struct rtentry *rt; + int mtu; + struct sockaddr_in icmpsrc = { sizeof (struct sockaddr_in), AF_INET, + 0 , { 0 }, { 0,0,0,0,0,0,0,0 } }; + icmpsrc.sin_addr = icp->icmp_ip.ip_dst; + + rt = rtalloc1((struct sockaddr *)&icmpsrc, 0, + RTF_CLONING | RTF_PRCLONING); + if (rt != NULL) { + RT_LOCK(rt); + if ((rt->rt_flags & RTF_HOST) && + !(rt->rt_rmx.rmx_locks & RTV_MTU)) { + mtu = ntohs(icp->icmp_nextmtu); + if (!mtu) + mtu = ip_next_mtu(rt->rt_rmx. + rmx_mtu, 1); +#if DEBUG_MTUDISC + printf("MTU for %s reduced to %d\n", + inet_ntop(AF_INET, + &icmpsrc.sin_addr, ipv4str, + sizeof (ipv4str)), mtu); +#endif + if (mtu < max(296, (tcp_minmss + + sizeof (struct tcpiphdr)))) { + /* rt->rt_rmx.rmx_mtu = + rt->rt_ifp->if_mtu; */ + rt->rt_rmx.rmx_locks |= RTV_MTU; + } else if (rt->rt_rmx.rmx_mtu > mtu) { + rt->rt_rmx.rmx_mtu = mtu; + } + } + RT_UNLOCK(rt); + rtfree(rt); + } + } + (*notify)(inp, inetctlerrmap[cmd]); + } tcp_unlock(inp->inp_socket, 1, 0); } } else @@ -1255,12 +2196,15 @@ tcp6_ctlinput(cmd, sa, d) sa->sa_len != sizeof(struct sockaddr_in6)) return; - if (cmd == PRC_QUENCH) - notify = tcp_quench; - else if (cmd == PRC_MSGSIZE) + if ((unsigned)cmd >= PRC_NCMDS) + return; + + if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc; - else if (!PRC_IS_REDIRECT(cmd) && - ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) + else if (!PRC_IS_REDIRECT(cmd) && (inet6ctlerrmap[cmd] == 0)) + return; + /* Source quench is deprecated */ + else if (cmd == PRC_QUENCH) return; /* if the parameter is from icmp6, decode it. */ @@ -1292,10 +2236,11 @@ tcp6_ctlinput(cmd, sa, d) in6_pcbnotify(&tcbinfo, sa, th.th_dport, (struct sockaddr *)ip6cp->ip6c_src, - th.th_sport, cmd, notify); - } else - in6_pcbnotify(&tcbinfo, sa, 0, (struct sockaddr *)sa6_src, - 0, cmd, notify); + th.th_sport, cmd, NULL, notify); + } else { + in6_pcbnotify(&tcbinfo, sa, 0, + (struct sockaddr *)(size_t)sa6_src, 0, cmd, NULL, notify); + } } #endif /* INET6 */ @@ -1345,10 +2290,6 @@ tcp6_ctlinput(cmd, sa, d) #define ISN_BYTES_PER_SECOND 1048576 -u_char isn_secret[32]; -int isn_last_reseed; -MD5_CTX isn_ctx; - tcp_seq tcp_new_isn(tp) struct tcpcb *tp; @@ -1356,12 +2297,15 @@ tcp_new_isn(tp) u_int32_t md5_buffer[4]; tcp_seq new_isn; struct timeval timenow; + u_char isn_secret[32]; + int 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 random(); + return RandomULong(); #else return arc4random(); #endif @@ -1405,21 +2349,6 @@ tcp_new_isn(tp) return new_isn; } -/* - * When a source quench is received, close congestion window - * to one segment. We will gradually open it again as we proceed. - */ -void -tcp_quench( - struct inpcb *inp, - __unused int errno -) -{ - struct tcpcb *tp = intotcpcb(inp); - - if (tp) - tp->snd_cwnd = tp->t_maxseg; -} /* * When a specific ICMP unreachable message is received and the @@ -1455,6 +2384,7 @@ tcp_mtudisc( struct socket *so = inp->inp_socket; int offered; int mss; + u_int32_t mtu; #if INET6 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ @@ -1462,21 +2392,35 @@ tcp_mtudisc( if (tp) { #if INET6 if (isipv6) - rt = tcp_rtlookup6(inp); + rt = tcp_rtlookup6(inp, IFSCOPE_NONE); else #endif /* INET6 */ - rt = tcp_rtlookup(inp); + 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; + + /* Route locked during lookup above */ + if (rt != NULL) + RT_UNLOCK(rt); return; } taop = rmx_taop(rt->rt_rmx); offered = taop->tao_mssopt; - mss = rt->rt_rmx.rmx_mtu - + mtu = rt->rt_rmx.rmx_mtu; + + /* Route locked during lookup above */ + RT_UNLOCK(rt); + +#if NECP + // Adjust MTU if necessary. + mtu = necp_socket_get_effective_mtu(inp, mtu); +#endif /* NECP */ + + mss = mtu - #if INET6 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : @@ -1514,11 +2458,19 @@ tcp_mtudisc( (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) mss -= TCPOLEN_TSTAMP_APPA; +#if MPTCP + mss -= mptcp_adj_mss(tp, TRUE); +#endif if (so->so_snd.sb_hiwat < mss) mss = so->so_snd.sb_hiwat; tp->t_maxseg = mss; + /* + * Reset the slow-start flight size as it may depends on the new MSS + */ + if (CC_ALGO(tp)->cwnd_init != NULL) + CC_ALGO(tp)->cwnd_init(tp); tcpstat.tcps_mturesent++; tp->t_rtttime = 0; tp->snd_nxt = tp->snd_una; @@ -1530,56 +2482,210 @@ tcp_mtudisc( * Look-up the routing entry to the peer of this inpcb. If no route * is found and it cannot be allocated the return NULL. This routine * is called by TCP routines that access the rmx structure and by tcp_mss - * to get the interface MTU. + * to get the interface MTU. If a route is found, this routine will + * hold the rtentry lock; the caller is responsible for unlocking. */ struct rtentry * -tcp_rtlookup(inp) +tcp_rtlookup(inp, input_ifscope) struct inpcb *inp; + unsigned int input_ifscope; { struct route *ro; struct rtentry *rt; + struct tcpcb *tp; + + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); ro = &inp->inp_route; - if (ro == NULL) - return (NULL); - rt = ro->ro_rt; - if (rt == NULL || !(rt->rt_flags & RTF_UP) || rt->generation_id != route_generation) { + if ((rt = ro->ro_rt) != NULL) + RT_LOCK(rt); + + if (ROUTE_UNUSABLE(ro)) { + if (rt != NULL) { + RT_UNLOCK(rt); + rt = NULL; + } + ROUTE_RELEASE(ro); /* No route yet, so try to acquire one */ if (inp->inp_faddr.s_addr != INADDR_ANY) { + unsigned int ifscope; + ro->ro_dst.sa_family = AF_INET; ro->ro_dst.sa_len = sizeof(struct sockaddr_in); - ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = + ((struct sockaddr_in *)(void *)&ro->ro_dst)->sin_addr = inp->inp_faddr; - rtalloc(ro); - rt = ro->ro_rt; + + /* + * If the socket was bound to an interface, then + * the bound-to-interface takes precedence over + * the inbound interface passed in by the caller + * (if we get here as part of the output path then + * input_ifscope is IFSCOPE_NONE). + */ + ifscope = (inp->inp_flags & INP_BOUND_IF) ? + inp->inp_boundifp->if_index : input_ifscope; + + rtalloc_scoped(ro, ifscope); + if ((rt = ro->ro_rt) != NULL) + RT_LOCK(rt); } } + if (rt != NULL) + RT_LOCK_ASSERT_HELD(rt); + + /* + * Update MTU discovery determination. Don't do it if: + * 1) it is disabled via the sysctl + * 2) the route isn't up + * 3) the MTU is locked (if it is, then discovery has been + * disabled) + */ + + tp = intotcpcb(inp); + + if (!path_mtu_discovery || ((rt != NULL) && + (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) + tp->t_flags &= ~TF_PMTUD; + else + tp->t_flags |= TF_PMTUD; + +#if CONFIG_IFEF_NOWINDOWSCALE + if (tcp_obey_ifef_nowindowscale && + tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL && + (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE)) { + /* Window scaling is enabled on this interface */ + tp->t_flags &= ~TF_REQ_SCALE; + } +#endif + + if (rt != NULL && rt->rt_ifp != NULL) { + somultipages(inp->inp_socket, + (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES)); + tcp_set_tso(tp, rt->rt_ifp); + soif2kcl(inp->inp_socket, + (rt->rt_ifp->if_eflags & IFEF_2KCL)); + tcp_set_ecn(tp, rt->rt_ifp); + } + + /* 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))) { + tp->t_flags |= TF_LOCAL; + } + + /* + * Caller needs to call RT_UNLOCK(rt). + */ return rt; } #if INET6 struct rtentry * -tcp_rtlookup6(inp) +tcp_rtlookup6(inp, input_ifscope) struct inpcb *inp; + unsigned int input_ifscope; { struct route_in6 *ro6; struct rtentry *rt; + struct tcpcb *tp; + + lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED); ro6 = &inp->in6p_route; - rt = ro6->ro_rt; - if (rt == NULL || !(rt->rt_flags & RTF_UP)) { + if ((rt = ro6->ro_rt) != NULL) + RT_LOCK(rt); + + if (ROUTE_UNUSABLE(ro6)) { + if (rt != NULL) { + RT_UNLOCK(rt); + rt = NULL; + } + ROUTE_RELEASE(ro6); /* No route yet, so try to acquire one */ if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) { struct sockaddr_in6 *dst6; + unsigned int ifscope; dst6 = (struct sockaddr_in6 *)&ro6->ro_dst; dst6->sin6_family = AF_INET6; dst6->sin6_len = sizeof(*dst6); dst6->sin6_addr = inp->in6p_faddr; - rtalloc((struct route *)ro6); - rt = ro6->ro_rt; + + /* + * If the socket was bound to an interface, then + * the bound-to-interface takes precedence over + * the inbound interface passed in by the caller + * (if we get here as part of the output path then + * input_ifscope is IFSCOPE_NONE). + */ + ifscope = (inp->inp_flags & INP_BOUND_IF) ? + inp->inp_boundifp->if_index : input_ifscope; + + rtalloc_scoped((struct route *)ro6, ifscope); + if ((rt = ro6->ro_rt) != NULL) + RT_LOCK(rt); } } + if (rt != NULL) + RT_LOCK_ASSERT_HELD(rt); + + /* + * Update path MTU Discovery determination + * while looking up the route: + * 1) we have a valid route to the destination + * 2) the MTU is not locked (if it is, then discovery has been + * disabled) + */ + + + tp = intotcpcb(inp); + + /* + * Update MTU discovery determination. Don't do it if: + * 1) it is disabled via the sysctl + * 2) the route isn't up + * 3) the MTU is locked (if it is, then discovery has been + * disabled) + */ + + if (!path_mtu_discovery || ((rt != NULL) && + (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) + tp->t_flags &= ~TF_PMTUD; + else + tp->t_flags |= TF_PMTUD; + +#if CONFIG_IFEF_NOWINDOWSCALE + if (tcp_obey_ifef_nowindowscale && + tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL && + (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE)) { + /* Window scaling is not enabled on this interface */ + tp->t_flags &= ~TF_REQ_SCALE; + } +#endif + + if (rt != NULL && rt->rt_ifp != NULL) { + somultipages(inp->inp_socket, + (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES)); + tcp_set_tso(tp, rt->rt_ifp); + soif2kcl(inp->inp_socket, + (rt->rt_ifp->if_eflags & IFEF_2KCL)); + tcp_set_ecn(tp, rt->rt_ifp); + } + + /* 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; + } + + /* + * Caller needs to call RT_UNLOCK(rt). + */ return rt; } #endif /* INET6 */ @@ -1601,15 +2707,14 @@ ipsec_hdrsiz_tcp(tp) if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) return 0; - MGETHDR(m, M_DONTWAIT, MT_DATA); + MGETHDR(m, M_DONTWAIT, MT_DATA); /* MAC-OK */ if (!m) return 0; - lck_mtx_lock(sadb_mutex); #if INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { ip6 = mtod(m, struct ip6_hdr *); - th = (struct tcphdr *)(ip6 + 1); + th = (struct tcphdr *)(void *)(ip6 + 1); m->m_pkthdr.len = m->m_len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); tcp_fillheaders(tp, ip6, th); @@ -1623,7 +2728,6 @@ ipsec_hdrsiz_tcp(tp) tcp_fillheaders(tp, ip, th); hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); } - lck_mtx_unlock(sadb_mutex); m_free(m); return hdrsiz; } @@ -1639,20 +2743,28 @@ tcp_gettaocache(inp) struct inpcb *inp; { struct rtentry *rt; + struct rmxp_tao *taop; #if INET6 if ((inp->inp_vflag & INP_IPV6) != 0) - rt = tcp_rtlookup6(inp); + rt = tcp_rtlookup6(inp, IFSCOPE_NONE); else #endif /* INET6 */ - rt = tcp_rtlookup(inp); + rt = tcp_rtlookup(inp, IFSCOPE_NONE); /* Make sure this is a host route and is up. */ if (rt == NULL || - (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) + (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) { + /* Route locked during lookup above */ + if (rt != NULL) + RT_UNLOCK(rt); return NULL; - - return rmx_taop(rt->rt_rmx); + } + + taop = rmx_taop(rt->rt_rmx); + /* Route locked during lookup above */ + RT_UNLOCK(rt); + return (taop); } /* @@ -1669,86 +2781,527 @@ tcp_cleartaocache() } int -tcp_lock(so, refcount, lr) - struct socket *so; - int refcount; - int lr; +tcp_lock(struct socket *so, int refcount, void *lr) { - int lr_saved; -#ifdef __ppc__ - if (lr == 0) { - __asm__ volatile("mflr %0" : "=r" (lr_saved)); - } - else lr_saved = lr; -#endif + void *lr_saved; - if (so->so_pcb) { - lck_mtx_lock(((struct inpcb *)so->so_pcb)->inpcb_mtx); - } - else { - panic("tcp_lock: so=%x NO PCB! lr=%x\n", so, lr_saved); - lck_mtx_lock(so->so_proto->pr_domain->dom_mtx); + if (lr == NULL) + lr_saved = __builtin_return_address(0); + else + lr_saved = lr; + + if (so->so_pcb != NULL) { + lck_mtx_lock(&((struct inpcb *)so->so_pcb)->inpcb_mtx); + } else { + panic("tcp_lock: so=%p NO PCB! lr=%p lrh= %s\n", + so, lr_saved, solockhistory_nr(so)); + /* NOTREACHED */ } - if (so->so_usecount < 0) - panic("tcp_lock: so=%x so_pcb=%x lr=%x ref=%x\n", - so, so->so_pcb, lr_saved, so->so_usecount); - + if (so->so_usecount < 0) { + panic("tcp_lock: so=%p so_pcb=%p lr=%p ref=%x lrh= %s\n", + so, so->so_pcb, lr_saved, so->so_usecount, solockhistory_nr(so)); + /* NOTREACHED */ + } if (refcount) so->so_usecount++; - so->reserved3 = (void *)lr_saved; + so->lock_lr[so->next_lock_lr] = lr_saved; + so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; return (0); } int -tcp_unlock(so, refcount, lr) - struct socket *so; - int refcount; - int lr; +tcp_unlock(struct socket *so, int refcount, void *lr) { - int lr_saved; -#ifdef __ppc__ - if (lr == 0) { - __asm__ volatile("mflr %0" : "=r" (lr_saved)); - } - else lr_saved = lr; -#endif + void *lr_saved; + + if (lr == NULL) + lr_saved = __builtin_return_address(0); + else + lr_saved = lr; #ifdef MORE_TCPLOCK_DEBUG - printf("tcp_unlock: so=%x sopcb=%x lock=%x ref=%x lr=%x\n", - so, so->so_pcb, ((struct inpcb *)so->so_pcb)->inpcb_mtx, so->so_usecount, lr_saved); + printf("tcp_unlock: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x " + "lr=0x%llx\n", (uint64_t)VM_KERNEL_ADDRPERM(so), + (uint64_t)VM_KERNEL_ADDRPERM(so->so_pcb), + (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)), + so->so_usecount, (uint64_t)VM_KERNEL_ADDRPERM(lr_saved)); #endif if (refcount) so->so_usecount--; - if (so->so_usecount < 0) - panic("tcp_unlock: so=%x usecount=%x\n", so, so->so_usecount); - if (so->so_pcb == NULL) { - panic("tcp_unlock: so=%x NO PCB usecount=%x lr=%x\n", so, so->so_usecount, lr_saved); - lck_mtx_unlock(so->so_proto->pr_domain->dom_mtx); + if (so->so_usecount < 0) { + panic("tcp_unlock: so=%p usecount=%x lrh= %s\n", + so, so->so_usecount, solockhistory_nr(so)); + /* NOTREACHED */ } - 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); + if (so->so_pcb == NULL) { + panic("tcp_unlock: so=%p NO PCB usecount=%x lr=%p lrh= %s\n", + so, so->so_usecount, lr_saved, solockhistory_nr(so)); + /* NOTREACHED */ + } else { + lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx, + LCK_MTX_ASSERT_OWNED); + so->unlock_lr[so->next_unlock_lr] = lr_saved; + so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; + lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx); } - so->reserved4 = (void *)lr_saved; return (0); } lck_mtx_t * -tcp_getlock(so, locktype) - struct socket *so; - int locktype; +tcp_getlock( + struct socket *so, + __unused int locktype) { struct inpcb *inp = sotoinpcb(so); if (so->so_pcb) { if (so->so_usecount < 0) - panic("tcp_getlock: so=%x usecount=%x\n", so, so->so_usecount); - return(inp->inpcb_mtx); + panic("tcp_getlock: so=%p usecount=%x lrh= %s\n", + so, so->so_usecount, solockhistory_nr(so)); + return(&inp->inpcb_mtx); } else { - panic("tcp_getlock: so=%x NULL so_pcb\n", so); + panic("tcp_getlock: so=%p NULL so_pcb %s\n", + so, solockhistory_nr(so)); return (so->so_proto->pr_domain->dom_mtx); } } + +/* + * Determine if we can grow the recieve socket buffer to avoid sending + * a zero window update to the peer. We allow even socket buffers that + * have fixed size (set by the application) to grow if the resource + * constraints are met. They will also be trimmed after the application + * reads data. + */ +static void +tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb) +{ + u_int32_t rcvbufinc = tp->t_maxseg << 4; + u_int32_t rcvbuf = sb->sb_hiwat; + struct socket *so = tp->t_inpcb->inp_socket; + + /* + * If message delivery is enabled, do not count + * unordered bytes in receive buffer towards hiwat + */ + if (so->so_flags & SOF_ENABLE_MSGS) + rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes; + + if (tcp_do_autorcvbuf == 1 && + tcp_cansbgrow(sb) && + (tp->t_flags & TF_SLOWLINK) == 0 && + (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)); + } +} + +int32_t +tcp_sbspace(struct tcpcb *tp) +{ + struct sockbuf *sb = &tp->t_inpcb->inp_socket->so_rcv; + u_int32_t rcvbuf = sb->sb_hiwat; + int32_t space; + struct socket *so = tp->t_inpcb->inp_socket; + int32_t pending = 0; + + /* + * If message delivery is enabled, do not count + * unordered bytes in receive buffer towards hiwat mark. + * This value is used to return correct rwnd that does + * not reflect the extra unordered bytes added to the + * receive socket buffer. + */ + if (so->so_flags & SOF_ENABLE_MSGS) + rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes; + + tcp_sbrcv_grow_rwin(tp, sb); + + space = ((int32_t) imin((rcvbuf - sb->sb_cc), + (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) + space = 0; + else + space -= pending; + + /* Avoid increasing window size if the current window + * is already very low, we could be in "persist" mode and + * we could break some apps (see rdar://5409343) + */ + + if (space < tp->t_maxseg) + return space; + + /* Clip window size for slower link */ + + if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0 ) + return imin(space, slowlink_wsize); + + return space; +} +/* + * Checks TCP Segment Offloading capability for a given connection + * and interface pair. + */ +void +tcp_set_tso(struct tcpcb *tp, struct ifnet *ifp) +{ +#if INET6 + struct inpcb *inp; + int isipv6; +#endif /* INET6 */ +#if MPTCP + /* + * We can't use TSO if this tcpcb belongs to an MPTCP session. + */ + if (tp->t_mpflags & TMPF_MPTCP_TRUE) { + tp->t_flags &= ~TF_TSO; + return; + } +#endif +#if INET6 + inp = tp->t_inpcb; + isipv6 = (inp->inp_vflag & INP_IPV6) != 0; + + if (isipv6) { + if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV6)) { + tp->t_flags |= TF_TSO; + if (ifp->if_tso_v6_mtu != 0) + tp->tso_max_segment_size = ifp->if_tso_v6_mtu; + else + tp->tso_max_segment_size = TCP_MAXWIN; + } else + tp->t_flags &= ~TF_TSO; + + } else +#endif /* INET6 */ + + { + if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV4)) { + tp->t_flags |= TF_TSO; + if (ifp->if_tso_v4_mtu != 0) + tp->tso_max_segment_size = ifp->if_tso_v4_mtu; + else + tp->tso_max_segment_size = TCP_MAXWIN; + } else + tp->t_flags &= ~TF_TSO; + } +} + +#define TIMEVAL_TO_TCPHZ(_tv_) ((_tv_).tv_sec * TCP_RETRANSHZ + (_tv_).tv_usec / TCP_RETRANSHZ_TO_USEC) + +/* Function to calculate the tcp clock. The tcp clock will get updated + * at the boundaries of the tcp layer. This is done at 3 places: + * 1. Right before processing an input tcp packet + * 2. Whenever a connection wants to access the network using tcp_usrreqs + * 3. When a tcp timer fires or before tcp slow timeout + * + */ + +void +calculate_tcp_clock() +{ + struct timeval tv = tcp_uptime; + struct timeval interval = {0, TCP_RETRANSHZ_TO_USEC}; + struct timeval now, hold_now; + uint32_t incr = 0; + + microuptime(&now); + + /* + * Update coarse-grained networking timestamp (in sec.); the idea + * is to update the counter returnable via net_uptime() when + * we read time. + */ + net_update_uptime_secs(now.tv_sec); + + timevaladd(&tv, &interval); + if (timevalcmp(&now, &tv, >)) { + /* time to update the clock */ + lck_spin_lock(tcp_uptime_lock); + if (timevalcmp(&tcp_uptime, &now, >=)) { + /* clock got updated while waiting for the lock */ + lck_spin_unlock(tcp_uptime_lock); + return; + } + + microuptime(&now); + hold_now = now; + tv = tcp_uptime; + timevalsub(&now, &tv); + + incr = TIMEVAL_TO_TCPHZ(now); + if (incr > 0) { + tcp_uptime = hold_now; + tcp_now += incr; + } + + lck_spin_unlock(tcp_uptime_lock); + } + return; +} + +/* Compute receive window scaling that we are going to request + * for this connection based on sb_hiwat. Try to leave some + * room to potentially increase the window size upto a maximum + * defined by the constant tcp_autorcvbuf_max. + */ +void +tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so) { + u_int32_t maxsockbufsize; + if (!tcp_do_rfc1323) { + tp->request_r_scale = 0; + return; + } + + tp->request_r_scale = max(tcp_win_scale, tp->request_r_scale); + maxsockbufsize = ((so->so_rcv.sb_flags & SB_USRSIZE) != 0) ? + so->so_rcv.sb_hiwat : tcp_autorcvbuf_max; + + while (tp->request_r_scale < TCP_MAX_WINSHIFT && + (TCP_MAXWIN << tp->request_r_scale) < maxsockbufsize) + tp->request_r_scale++; + tp->request_r_scale = min(tp->request_r_scale, TCP_MAX_WINSHIFT); + +} + +int +tcp_notsent_lowat_check(struct socket *so) { + struct inpcb *inp = sotoinpcb(so); + struct tcpcb *tp = NULL; + int notsent = 0; + if (inp != NULL) { + tp = intotcpcb(inp); + } + + notsent = so->so_snd.sb_cc - + (tp->snd_nxt - tp->snd_una); + + /* When we send a FIN or SYN, not_sent can be negative. + * In that case also we need to send a write event to the + * process if it is waiting. In the FIN case, it will + * get an error from send because cantsendmore will be set. + */ + if (notsent <= tp->t_notsent_lowat) { + return(1); + } + + /* When Nagle's algorithm is not disabled, it is better + * to wakeup the client until there is atleast one + * maxseg of data to write. + */ + if ((tp->t_flags & TF_NODELAY) == 0 && + notsent > 0 && notsent < tp->t_maxseg) { + return(1); + } + return(0); +} + +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; + + 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)) + break; + prev = rxseg; + } + next = rxseg; + + /* check if prev seg is for this sequence */ + if (prev != NULL && SEQ_LEQ(prev->rx_start, start) && + SEQ_GEQ(prev->rx_end, end)) { + prev->rx_count++; + return; + } + + /* + * There are a couple of possibilities at this point. + * 1. prev overlaps with the beginning of this sequence + * 2. next overlaps with the end of this sequence + * 3. there is no overlap. + */ + + if (prev != NULL && SEQ_GT(prev->rx_end, start)) { + if (prev->rx_start == start && SEQ_GT(end, prev->rx_end)) { + start = prev->rx_end + 1; + prev->rx_count++; + } else { + prev->rx_end = (start - 1); + rxcount = prev->rx_count; + } + } + + if (next != NULL && SEQ_LT(next->rx_start, end)) { + if (SEQ_LEQ(next->rx_end, end)) { + end = next->rx_start - 1; + next->rx_count++; + } else { + next->rx_start = end + 1; + rxcount = next->rx_count; + } + } + if (!SEQ_LT(start, end)) + return; + + rxseg = (struct tcp_rxt_seg *) zalloc(tcp_rxt_seg_zone); + if (rxseg == NULL) { + return; + } + bzero(rxseg, sizeof(*rxseg)); + rxseg->rx_start = start; + rxseg->rx_end = end; + rxseg->rx_count = rxcount + 1; + + if (prev != NULL) { + SLIST_INSERT_AFTER(prev, rxseg, rx_link); + } else { + SLIST_INSERT_HEAD(&tp->t_rxt_segments, rxseg, rx_link); + } + return; +} + +struct tcp_rxt_seg * +tcp_rxtseg_find(struct tcpcb *tp, tcp_seq start, tcp_seq end) +{ + struct tcp_rxt_seg *rxseg; + if (SLIST_EMPTY(&tp->t_rxt_segments)) + return (NULL); + + 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)) + break; + } + return (NULL); +} + +void +tcp_rxtseg_clean(struct tcpcb *tp) +{ + struct tcp_rxt_seg *rxseg, *next; + + SLIST_FOREACH_SAFE(rxseg, &tp->t_rxt_segments, rx_link, next) { + SLIST_REMOVE(&tp->t_rxt_segments, rxseg, + tcp_rxt_seg, rx_link); + zfree(tcp_rxt_seg_zone, rxseg); + } + tp->t_dsack_lastuna = tp->snd_max; +} + +boolean_t +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 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); + + bad_rexmt = TRUE; + SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) { + if (rxseg->rx_count > 1 || + !(rxseg->rx_flags & TCP_RXT_SPURIOUS)) { + bad_rexmt = FALSE; + break; + } + } + return (bad_rexmt); +} + +boolean_t +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); + + SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) { + if (rxseg->rx_count == 1 && + SLIST_NEXT(rxseg,rx_link) == NULL && + (rxseg->rx_flags & TCP_RXT_DSACK_FOR_TLP)) { + dsack_for_tlp = TRUE; + break; + } + } + return (dsack_for_tlp); +} + +u_int32_t +tcp_rxtseg_total_size(struct tcpcb *tp) { + struct tcp_rxt_seg *rxseg; + u_int32_t total_size = 0; + + SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) { + total_size += (rxseg->rx_end - rxseg->rx_start) + 1; + } + return (total_size); +} + +void +tcp_get_connectivity_status(struct tcpcb *tp, + struct tcp_conn_status *connstatus) +{ + if (tp == NULL || connstatus == NULL) + return; + bzero(connstatus, sizeof(*connstatus)); + if (tp->t_rxtshift >= TCP_CONNECTIVITY_PROBES_MAX) { + if (TCPS_HAVEESTABLISHED(tp->t_state)) { + connstatus->write_probe_failed = 1; + } else { + connstatus->conn_probe_failed = 1; + } + } + 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)) + connstatus->probe_activated = 1; + return; +} + +boolean_t +tfo_enabled(const struct tcpcb *tp) +{ + return !!(tp->t_flagsext & TF_FASTOPEN); +} + +void +tcp_disable_tfo(struct tcpcb *tp) +{ + tp->t_flagsext &= ~TF_FASTOPEN; +} +