X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/593a1d5fd87cdf5b46dd5fcb84467b432cea0f91..440d4c6cfad24426bfddca7518f16c17f4e443f7:/bsd/netinet/tcp_usrreq.c?ds=inline diff --git a/bsd/netinet/tcp_usrreq.c b/bsd/netinet/tcp_usrreq.c index d194a867f..19405c584 100644 --- a/bsd/netinet/tcp_usrreq.c +++ b/bsd/netinet/tcp_usrreq.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2012 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -76,6 +76,7 @@ #include #include +#include #include #include @@ -105,6 +106,11 @@ #include #endif /*IPSEC*/ +void tcp_fill_info(struct tcpcb *, struct tcp_info *); +errno_t tcp_fill_info_for_info_tuple(struct info_tuple *, struct tcp_info *); + +int tcp_sysctl_info(struct sysctl_oid *, void *, int , struct sysctl_req *); + /* * TCP protocol interface to socket abstraction. */ @@ -120,30 +126,30 @@ static struct tcpcb * static struct tcpcb * tcp_usrclosed(struct tcpcb *); -__private_extern__ int tcp_win_scale = 3; -SYSCTL_INT(_net_inet_tcp, OID_AUTO, win_scale_factor, CTLFLAG_RW, - &tcp_win_scale, 0, "Window scaling factor"); - static u_int32_t tcps_in_sw_cksum; -SYSCTL_UINT(_net_inet_tcp, OID_AUTO, in_sw_cksum, CTLFLAG_RD, +SYSCTL_UINT(_net_inet_tcp, OID_AUTO, in_sw_cksum, CTLFLAG_RD | CTLFLAG_LOCKED, &tcps_in_sw_cksum, 0, "Number of received packets checksummed in software"); static u_int64_t tcps_in_sw_cksum_bytes; -SYSCTL_QUAD(_net_inet_tcp, OID_AUTO, in_sw_cksum_bytes, CTLFLAG_RD, +SYSCTL_QUAD(_net_inet_tcp, OID_AUTO, in_sw_cksum_bytes, CTLFLAG_RD | CTLFLAG_LOCKED, &tcps_in_sw_cksum_bytes, "Amount of received data checksummed in software"); static u_int32_t tcps_out_sw_cksum; -SYSCTL_UINT(_net_inet_tcp, OID_AUTO, out_sw_cksum, CTLFLAG_RD, +SYSCTL_UINT(_net_inet_tcp, OID_AUTO, out_sw_cksum, CTLFLAG_RD | CTLFLAG_LOCKED, &tcps_out_sw_cksum, 0, "Number of transmitted packets checksummed in software"); static u_int64_t tcps_out_sw_cksum_bytes; -SYSCTL_QUAD(_net_inet_tcp, OID_AUTO, out_sw_cksum_bytes, CTLFLAG_RD, +SYSCTL_QUAD(_net_inet_tcp, OID_AUTO, out_sw_cksum_bytes, CTLFLAG_RD | CTLFLAG_LOCKED, &tcps_out_sw_cksum_bytes, "Amount of transmitted data checksummed in software"); +extern uint32_t tcp_autorcvbuf_max; + +extern void tcp_sbrcv_trim(struct tcpcb *tp, struct sockbuf *sb); + #if TCPDEBUG #define TCPDEBUG0 int ostate = 0 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0 @@ -155,13 +161,8 @@ SYSCTL_QUAD(_net_inet_tcp, OID_AUTO, out_sw_cksum_bytes, CTLFLAG_RD, #define TCPDEBUG2(req) #endif -#if CONFIG_USESOCKTHRESHOLD -__private_extern__ unsigned int tcp_sockthreshold = 64; -#else -__private_extern__ unsigned int tcp_sockthreshold = 0; -#endif -SYSCTL_INT(_net_inet_tcp, OID_AUTO, sockthreshold, CTLFLAG_RW, - &tcp_sockthreshold , 0, "TCP Socket size increased if less than threshold"); +SYSCTL_PROC(_net_inet_tcp, OID_AUTO, info, CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, + 0 , 0, tcp_sysctl_info, "S", "TCP info per tuple"); /* * TCP attaches to socket via pru_attach(), reserving space, @@ -186,7 +187,7 @@ tcp_usr_attach(struct socket *so, __unused int proto, struct proc *p) error = EISCONN; goto out; } - + error = tcp_attach(so, p); if (error) goto out; @@ -217,14 +218,15 @@ tcp_usr_detach(struct socket *so) if (inp == 0 || (inp->inp_state == INPCB_STATE_DEAD)) { return EINVAL; /* XXX */ } -#if 1 - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); -#endif + lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); tp = intotcpcb(inp); /* In case we got disconnected from the peer */ if (tp == 0) goto out; TCPDEBUG1(); + + calculate_tcp_clock(); + tp = tcp_disconnect(tp); out: TCPDEBUG2(PRU_DETACH); @@ -238,6 +240,7 @@ out: } \ tp = intotcpcb(inp); \ TCPDEBUG1(); \ + calculate_tcp_clock(); \ } while(0) #define COMMON_END(req) out: TCPDEBUG2(req); return error; goto out @@ -276,7 +279,7 @@ tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct proc *p) * Must check for multicast addresses and disallow binding * to them. */ - sinp = (struct sockaddr_in *)nam; + sinp = (struct sockaddr_in *)(void *)nam; if (sinp->sin_family == AF_INET && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { error = EAFNOSUPPORT; @@ -309,7 +312,7 @@ tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct proc *p) * Must check for multicast addresses and disallow binding * to them. */ - sin6p = (struct sockaddr_in6 *)nam; + sin6p = (struct sockaddr_in6 *)(void *)nam; if (sin6p->sin6_family == AF_INET6 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) { error = EAFNOSUPPORT; @@ -415,6 +418,8 @@ tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct proc *p) tp = intotcpcb(inp); TCPDEBUG1(); + calculate_tcp_clock(); + if (nam->sa_family != 0 && nam->sa_family != AF_INET) { error = EAFNOSUPPORT; goto out; @@ -422,7 +427,7 @@ tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct proc *p) /* * Must disallow TCP ``connections'' to multicast addresses. */ - sinp = (struct sockaddr_in *)nam; + sinp = (struct sockaddr_in *)(void *)nam; if (sinp->sin_family == AF_INET && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) { error = EAFNOSUPPORT; @@ -455,7 +460,7 @@ tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct proc *p) /* * Must disallow TCP ``connections'' to multicast addresses. */ - sin6p = (struct sockaddr_in6 *)nam; + sin6p = (struct sockaddr_in6 *)(void *)nam; if (sin6p->sin6_family == AF_INET6 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) { error = EAFNOSUPPORT; @@ -505,9 +510,7 @@ tcp_usr_disconnect(struct socket *so) struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp; -#if 1 - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); -#endif + lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); COMMON_START(); /* In case we got disconnected from the peer */ if (tp == 0) @@ -529,6 +532,8 @@ tcp_usr_accept(struct socket *so, struct sockaddr **nam) struct tcpcb *tp = NULL; TCPDEBUG0; + in_setpeeraddr(so, nam); + if (so->so_state & SS_ISDISCONNECTED) { error = ECONNABORTED; goto out; @@ -538,7 +543,9 @@ tcp_usr_accept(struct socket *so, struct sockaddr **nam) } tp = intotcpcb(inp); TCPDEBUG1(); - in_setpeeraddr(so, nam); + + calculate_tcp_clock(); + COMMON_END(PRU_ACCEPT); } @@ -560,6 +567,9 @@ tcp6_usr_accept(struct socket *so, struct sockaddr **nam) } tp = intotcpcb(inp); TCPDEBUG1(); + + calculate_tcp_clock(); + in6_mapped_peeraddr(so, nam); COMMON_END(PRU_ACCEPT); } @@ -614,6 +624,8 @@ tcp_usr_rcvd(struct socket *so, __unused int flags) /* In case we got disconnected from the peer */ if (tp == 0) goto out; + tcp_sbrcv_trim(tp, &so->so_rcv); + tcp_output(tp); COMMON_END(PRU_RCVD); } @@ -681,6 +693,9 @@ tcp_usr_send(struct socket *so, int flags, struct mbuf *m, #endif /* INET6 */ tp = intotcpcb(inp); TCPDEBUG1(); + + calculate_tcp_clock(); + if (control) { /* TCP doesn't do control messages (rights, creds, etc) */ if (control->m_len) { @@ -706,11 +721,11 @@ tcp_usr_send(struct socket *so, int flags, struct mbuf *m, error = tcp6_connect(tp, nam, p); else #endif /* INET6 */ - error = tcp_connect(tp, nam, p); + error = tcp_connect(tp, nam, p); if (error) goto out; tp->snd_wnd = TTCP_CLIENT_SND_WND; - tcp_mss(tp, -1); + tcp_mss(tp, -1, IFSCOPE_NONE); } if (flags & PRUS_EOF) { @@ -729,7 +744,9 @@ tcp_usr_send(struct socket *so, int flags, struct mbuf *m, tp->t_flags &= ~TF_MORETOCOME; } } else { - if (sbspace(&so->so_snd) < -512) { + if (sbspace(&so->so_snd) == 0) { + /* if no space is left in sockbuf, + * do not try to squeeze in OOB traffic */ m_freem(m); error = ENOBUFS; goto out; @@ -759,7 +776,7 @@ tcp_usr_send(struct socket *so, int flags, struct mbuf *m, if (error) goto out; tp->snd_wnd = TTCP_CLIENT_SND_WND; - tcp_mss(tp, -1); + tcp_mss(tp, -1, IFSCOPE_NONE); } tp->snd_up = tp->snd_una + so->so_snd.sb_cc; tp->t_force = 1; @@ -875,11 +892,12 @@ tcp_connect(tp, nam, p) struct inpcb *inp = tp->t_inpcb, *oinp; struct socket *so = inp->inp_socket; struct tcpcb *otp; - struct sockaddr_in *sin = (struct sockaddr_in *)nam; - struct sockaddr_in *ifaddr; + struct sockaddr_in *sin = (struct sockaddr_in *)(void *)nam; + struct sockaddr_in ifaddr; struct rmxp_tao *taop; struct rmxp_tao tao_noncached; int error; + struct ifnet *outif = NULL; if (inp->inp_lport == 0) { error = in_pcbbind(inp, (struct sockaddr *)0, p); @@ -892,7 +910,7 @@ tcp_connect(tp, nam, p) * earlier incarnation of this same connection still in * TIME_WAIT state, creating an ADDRINUSE error. */ - error = in_pcbladdr(inp, nam, &ifaddr); + error = in_pcbladdr(inp, nam, &ifaddr, &outif); if (error) return error; @@ -900,7 +918,7 @@ tcp_connect(tp, nam, p) oinp = in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port, inp->inp_laddr.s_addr != INADDR_ANY ? inp->inp_laddr - : ifaddr->sin_addr, + : ifaddr.sin_addr, inp->inp_lport, 0, NULL); tcp_lock(inp->inp_socket, 0, 0); @@ -915,7 +933,7 @@ tcp_connect(tp, nam, p) if (oinp != inp && (otp = intotcpcb(oinp)) != NULL && otp->t_state == TCPS_TIME_WAIT && - otp->t_starttime < (u_long)tcp_msl && + ((int)(tcp_now - otp->t_starttime)) < tcp_msl && (otp->t_flags & TF_RCVD_CC)) otp = tcp_close(otp); else { @@ -928,7 +946,7 @@ tcp_connect(tp, nam, p) tcp_unlock(oinp->inp_socket, 1, 0); } skip_oinp: - if ((inp->inp_laddr.s_addr == INADDR_ANY ? ifaddr->sin_addr.s_addr : + if ((inp->inp_laddr.s_addr == INADDR_ANY ? ifaddr.sin_addr.s_addr : inp->inp_laddr.s_addr) == sin->sin_addr.s_addr && inp->inp_lport == sin->sin_port) return EINVAL; @@ -938,37 +956,29 @@ skip_oinp: lck_rw_lock_exclusive(inp->inp_pcbinfo->mtx); socket_lock(inp->inp_socket, 0); } - if (inp->inp_laddr.s_addr == INADDR_ANY) - inp->inp_laddr = ifaddr->sin_addr; + if (inp->inp_laddr.s_addr == INADDR_ANY) { + inp->inp_laddr = ifaddr.sin_addr; + inp->inp_last_outifp = outif; + } inp->inp_faddr = sin->sin_addr; inp->inp_fport = sin->sin_port; in_pcbrehash(inp); lck_rw_done(inp->inp_pcbinfo->mtx); - /* Compute window scaling to requesti according to sb_hiwat - * or leave us some room to increase potentially increase the window size depending - * on the default win scale - */ - while (tp->request_r_scale < TCP_MAX_WINSHIFT && - (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) - tp->request_r_scale++; + if (inp->inp_flowhash == 0) + inp->inp_flowhash = inp_calc_flowhash(inp); - /* - * Inflate window size only if no setsockopt was performed on the recv sockbuf and - * if we're not over our number of active pcbs. - */ - - if (((so->so_rcv.sb_flags & SB_USRSIZE) == 0) && (inp->inp_pcbinfo->ipi_count < tcp_sockthreshold)) { - tp->request_r_scale = max(tcp_win_scale, tp->request_r_scale); - so->so_rcv.sb_hiwat = min(TCP_MAXWIN << tp->request_r_scale, (sb_max / (MSIZE+MCLBYTES)) * MCLBYTES); - } + tcp_set_max_rwinscale(tp, so); soisconnecting(so); tcpstat.tcps_connattempt++; tp->t_state = TCPS_SYN_SENT; - tp->t_timer[TCPT_KEEP] = tcp_keepinit; + tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, + tp->t_keepinit ? tp->t_keepinit : tcp_keepinit); tp->iss = tcp_new_isn(tp); tcp_sendseqinit(tp); + if (nstat_collect) + nstat_route_connect_attempt(inp->inp_route.ro_rt); /* * Generate a CC value for this connection and @@ -1001,26 +1011,31 @@ tcp6_connect(tp, nam, p) struct inpcb *inp = tp->t_inpcb, *oinp; struct socket *so = inp->inp_socket; struct tcpcb *otp; - struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; + struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)(void *)nam; struct in6_addr addr6; struct rmxp_tao *taop; struct rmxp_tao tao_noncached; - int error; + int error = 0; + struct ifnet *outif = NULL; if (inp->inp_lport == 0) { error = in6_pcbbind(inp, (struct sockaddr *)0, p); if (error) - return error; + goto done; } /* * Cannot simply call in_pcbconnect, because there might be an * earlier incarnation of this same connection still in * TIME_WAIT state, creating an ADDRINUSE error. + * + * in6_pcbladdr() might return an ifp with its reference held + * even in the error case, so make sure that it's released + * whenever it's non-NULL. */ - error = in6_pcbladdr(inp, nam, &addr6); + error = in6_pcbladdr(inp, nam, &addr6, &outif); if (error) - return error; + goto done; tcp_unlock(inp->inp_socket, 0, 0); oinp = in6_pcblookup_hash(inp->inp_pcbinfo, &sin6->sin6_addr, sin6->sin6_port, @@ -1032,11 +1047,13 @@ tcp6_connect(tp, nam, p) if (oinp) { if (oinp != inp && (otp = intotcpcb(oinp)) != NULL && otp->t_state == TCPS_TIME_WAIT && - otp->t_starttime < (u_long)tcp_msl && - (otp->t_flags & TF_RCVD_CC)) + ((int)(tcp_now - otp->t_starttime)) < tcp_msl && + (otp->t_flags & TF_RCVD_CC)) { otp = tcp_close(otp); - else - return EADDRINUSE; + } else { + error = EADDRINUSE; + goto done; + } } if (!lck_rw_try_lock_exclusive(inp->inp_pcbinfo->mtx)) { /*lock inversion issue, mostly with udp multicast packets */ @@ -1044,8 +1061,10 @@ tcp6_connect(tp, nam, p) lck_rw_lock_exclusive(inp->inp_pcbinfo->mtx); socket_lock(inp->inp_socket, 0); } - if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) + if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { inp->in6p_laddr = addr6; + inp->in6p_last_outifp = outif; /* no reference needed */ + } inp->in6p_faddr = sin6->sin6_addr; inp->inp_fport = sin6->sin6_port; if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0) @@ -1053,17 +1072,20 @@ tcp6_connect(tp, nam, p) in_pcbrehash(inp); lck_rw_done(inp->inp_pcbinfo->mtx); - /* Compute window scaling to request. */ - while (tp->request_r_scale < TCP_MAX_WINSHIFT && - (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat) - tp->request_r_scale++; + if (inp->inp_flowhash == 0) + inp->inp_flowhash = inp_calc_flowhash(inp); + + tcp_set_max_rwinscale(tp, so); soisconnecting(so); tcpstat.tcps_connattempt++; tp->t_state = TCPS_SYN_SENT; - tp->t_timer[TCPT_KEEP] = tcp_keepinit; + tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, + tp->t_keepinit ? tp->t_keepinit : tcp_keepinit); tp->iss = tcp_new_isn(tp); tcp_sendseqinit(tp); + if (nstat_collect) + nstat_route_connect_attempt(inp->inp_route.ro_rt); /* * Generate a CC value for this connection and @@ -1083,10 +1105,204 @@ tcp6_connect(tp, nam, p) tp->t_flags |= TF_SENDCCNEW; } - return 0; +done: + if (outif != NULL) + ifnet_release(outif); + + return (error); } #endif /* INET6 */ +/* + * Export TCP internal state information via a struct tcp_info + */ +__private_extern__ void +tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti) +{ + struct inpcb *inp = tp->t_inpcb; + + bzero(ti, sizeof(*ti)); + + ti->tcpi_state = tp->t_state; + + if (tp->t_state > TCPS_LISTEN) { + if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP)) + ti->tcpi_options |= TCPI_OPT_TIMESTAMPS; + if (tp->t_flags & TF_SACK_PERMIT) + ti->tcpi_options |= TCPI_OPT_SACK; + if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) { + ti->tcpi_options |= TCPI_OPT_WSCALE; + ti->tcpi_snd_wscale = tp->snd_scale; + ti->tcpi_rcv_wscale = tp->rcv_scale; + } + + /* Are we in retranmission episode */ + if (tp->snd_max != tp->snd_nxt) + ti->tcpi_flags |= TCPI_FLAG_LOSSRECOVERY; + else + ti->tcpi_flags &= ~TCPI_FLAG_LOSSRECOVERY; + + ti->tcpi_rto = tp->t_timer[TCPT_REXMT] ? tp->t_rxtcur : 0; + ti->tcpi_snd_mss = tp->t_maxseg; + ti->tcpi_rcv_mss = tp->t_maxseg; + + ti->tcpi_rttcur = tp->t_rttcur; + ti->tcpi_srtt = tp->t_srtt >> TCP_RTT_SHIFT; + ti->tcpi_rttvar = tp->t_rttvar >> TCP_RTTVAR_SHIFT; + + ti->tcpi_snd_ssthresh = tp->snd_ssthresh; + ti->tcpi_snd_cwnd = tp->snd_cwnd; + ti->tcpi_snd_sbbytes = tp->t_inpcb->inp_socket->so_snd.sb_cc; + + ti->tcpi_rcv_space = tp->rcv_wnd; + + ti->tcpi_snd_wnd = tp->snd_wnd; + ti->tcpi_snd_nxt = tp->snd_nxt; + ti->tcpi_rcv_nxt = tp->rcv_nxt; + + /* convert bytes/msec to bits/sec */ + if ((tp->t_flagsext & TF_MEASURESNDBW) != 0 && + tp->t_bwmeas != NULL) { + ti->tcpi_snd_bw = (tp->t_bwmeas->bw_sndbw * 8000); + } + + ti->tcpi_last_outif = (tp->t_inpcb->inp_last_outifp == NULL) ? 0 : + tp->t_inpcb->inp_last_outifp->if_index; + + //atomic_get_64(ti->tcpi_txbytes, &inp->inp_stat->txbytes); + ti->tcpi_txbytes = inp->inp_stat->txbytes; + ti->tcpi_txretransmitbytes = tp->t_stat.txretransmitbytes; + ti->tcpi_txunacked = tp->snd_max - tp->snd_una; + + //atomic_get_64(ti->tcpi_rxbytes, &inp->inp_stat->rxbytes); + ti->tcpi_rxbytes = inp->inp_stat->rxbytes; + ti->tcpi_rxduplicatebytes = tp->t_stat.rxduplicatebytes; + } +} + +__private_extern__ errno_t +tcp_fill_info_for_info_tuple(struct info_tuple *itpl, struct tcp_info *ti) +{ + struct inpcbinfo *pcbinfo = NULL; + struct inpcb *inp = NULL; + struct socket *so; + struct tcpcb *tp; + + if (itpl->itpl_proto == IPPROTO_TCP) + pcbinfo = &tcbinfo; + else + return EINVAL; + + if (itpl->itpl_local_sa.sa_family == AF_INET && + itpl->itpl_remote_sa.sa_family == AF_INET) { + inp = in_pcblookup_hash(pcbinfo, + itpl->itpl_remote_sin.sin_addr, + itpl->itpl_remote_sin.sin_port, + itpl->itpl_local_sin.sin_addr, + itpl->itpl_local_sin.sin_port, + 0, NULL); + } else if (itpl->itpl_local_sa.sa_family == AF_INET6 && + itpl->itpl_remote_sa.sa_family == AF_INET6) { + struct in6_addr ina6_local; + struct in6_addr ina6_remote; + + ina6_local = itpl->itpl_local_sin6.sin6_addr; + if (IN6_IS_SCOPE_LINKLOCAL(&ina6_local) && itpl->itpl_local_sin6.sin6_scope_id) + ina6_local.s6_addr16[1] = htons(itpl->itpl_local_sin6.sin6_scope_id); + + ina6_remote = itpl->itpl_remote_sin6.sin6_addr; + if (IN6_IS_SCOPE_LINKLOCAL(&ina6_remote) && itpl->itpl_remote_sin6.sin6_scope_id) + ina6_remote.s6_addr16[1] = htons(itpl->itpl_remote_sin6.sin6_scope_id); + + inp = in6_pcblookup_hash(pcbinfo, + &ina6_remote, + itpl->itpl_remote_sin6.sin6_port, + &ina6_local, + itpl->itpl_local_sin6.sin6_port, + 0, NULL); + } else + return EINVAL; + if (inp == NULL || (so = inp->inp_socket) == NULL) + return ENOENT; + + socket_lock(so, 0); + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { + socket_unlock(so, 0); + return ENOENT; + } + tp = intotcpcb(inp); + + tcp_fill_info(tp, ti); + socket_unlock(so, 0); + + return 0; +} + + +__private_extern__ int +tcp_sysctl_info(__unused struct sysctl_oid *oidp, __unused void *arg1, __unused int arg2, struct sysctl_req *req) +{ + int error; + struct tcp_info ti; + struct info_tuple itpl; + + if (req->newptr == USER_ADDR_NULL) { + return EINVAL; + } + if (req->newlen < sizeof(struct info_tuple)) { + return EINVAL; + } + error = SYSCTL_IN(req, &itpl, sizeof(struct info_tuple)); + if (error != 0) { + return error; + } + error = tcp_fill_info_for_info_tuple(&itpl, &ti); + if (error != 0) { + return error; + } + error = SYSCTL_OUT(req, &ti, sizeof(struct tcp_info)); + if (error != 0) { + return error; + } + + return 0; +} + +static int +tcp_lookup_peer_pid_locked(struct socket *so, pid_t *out_pid) +{ + int error = EHOSTUNREACH; + *out_pid = -1; + if ((so->so_state & SS_ISCONNECTED) == 0) return ENOTCONN; + + struct inpcb *inp = (struct inpcb*)so->so_pcb; + uint16_t lport = inp->inp_lport; + uint16_t fport = inp->inp_fport; + struct inpcb *finp = NULL; + + if (inp->inp_vflag & INP_IPV6) { + struct in6_addr laddr6 = inp->in6p_laddr; + struct in6_addr faddr6 = inp->in6p_faddr; + socket_unlock(so, 0); + finp = in6_pcblookup_hash(&tcbinfo, &laddr6, lport, &faddr6, fport, 0, NULL); + socket_lock(so, 0); + } else if (inp->inp_vflag & INP_IPV4) { + struct in_addr laddr4 = inp->inp_laddr; + struct in_addr faddr4 = inp->inp_faddr; + socket_unlock(so, 0); + finp = in_pcblookup_hash(&tcbinfo, laddr4, lport, faddr4, fport, 0, NULL); + socket_lock(so, 0); + } + + if (finp) { + *out_pid = finp->inp_socket->last_pid; + error = 0; + in_pcb_checkstate(finp, WNT_RELEASE, 0); + } + + return error; +} + /* * The new sockopt interface makes it possible for us to block in the * copyin/out step (if we take a page fault). Taking a page fault at @@ -1108,7 +1324,9 @@ tcp_ctloutput(so, sopt) if (inp == NULL) { return (ECONNRESET); } - if (sopt->sopt_level != IPPROTO_TCP) { + /* Allow at this level */ + if (sopt->sopt_level != IPPROTO_TCP && + !(sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_FLUSH)) { #if INET6 if (INP_CHECK_SOCKAF(so, AF_INET6)) error = ip6_ctloutput(so, sopt); @@ -1122,6 +1340,8 @@ tcp_ctloutput(so, sopt) return (ECONNRESET); } + calculate_tcp_clock(); + switch (sopt->sopt_dir) { case SOPT_SET: switch (sopt->sopt_name) { @@ -1153,7 +1373,69 @@ tcp_ctloutput(so, sopt) else tp->t_flags &= ~opt; break; + case TCP_RXT_FINDROP: + error = sooptcopyin(sopt, &optval, sizeof optval, + sizeof optval); + if (error) + break; + opt = TF_RXTFINDROP; + if (optval) + tp->t_flagsext |= opt; + else + tp->t_flagsext &= ~opt; + break; + case TCP_MEASURE_SND_BW: + error = sooptcopyin(sopt, &optval, sizeof optval, + sizeof optval); + if (error) + break; + opt = TF_MEASURESNDBW; + if (optval) { + if (tp->t_bwmeas == NULL) { + tp->t_bwmeas = tcp_bwmeas_alloc(tp); + if (tp->t_bwmeas == NULL) { + error = ENOMEM; + break; + } + } + tp->t_flagsext |= opt; + } else { + tp->t_flagsext &= ~opt; + /* Reset snd bw measurement state */ + tp->t_flagsext &= ~(TF_BWMEAS_INPROGRESS); + if (tp->t_bwmeas != NULL) { + tcp_bwmeas_free(tp); + } + } + break; + case TCP_MEASURE_BW_BURST: { + struct tcp_measure_bw_burst in; + uint32_t minpkts, maxpkts; + bzero(&in, sizeof(in)); + error = sooptcopyin(sopt, &in, sizeof(in), + sizeof(in)); + if (error) + break; + if ((tp->t_flagsext & TF_MEASURESNDBW) == 0 || + tp->t_bwmeas == NULL) { + error = EINVAL; + break; + } + minpkts = (in.min_burst_size != 0) ? in.min_burst_size : + tp->t_bwmeas->bw_minsizepkts; + maxpkts = (in.max_burst_size != 0) ? in.max_burst_size : + tp->t_bwmeas->bw_maxsizepkts; + if (minpkts > maxpkts) { + error = EINVAL; + break; + } + tp->t_bwmeas->bw_minsizepkts = minpkts; + tp->t_bwmeas->bw_maxsizepkts = maxpkts; + tp->t_bwmeas->bw_minsize = (minpkts * tp->t_maxseg); + tp->t_bwmeas->bw_maxsize = (maxpkts * tp->t_maxseg); + break; + } case TCP_MAXSEG: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); @@ -1176,10 +1458,70 @@ tcp_ctloutput(so, sopt) error = EINVAL; else { tp->t_keepidle = optval * TCP_RETRANSHZ; - tp->t_timer[TCPT_KEEP] = TCP_KEEPIDLE(tp); /* reset the timer to new value */ + tp->t_timer[TCPT_KEEP] = OFFSET_FROM_START(tp, + TCP_KEEPIDLE(tp)); /* reset the timer to new value */ + tcp_check_timer_state(tp); } break; - + + case TCP_CONNECTIONTIMEOUT: + error = sooptcopyin(sopt, &optval, sizeof optval, + sizeof optval); + if (error) + break; + if (optval < 0) + error = EINVAL; + else + tp->t_keepinit = optval * TCP_RETRANSHZ; + break; + + case PERSIST_TIMEOUT: + error = sooptcopyin(sopt, &optval, sizeof optval, + sizeof optval); + if (error) + break; + if (optval < 0) + error = EINVAL; + else + tp->t_persist_timeout = optval * TCP_RETRANSHZ; + break; + case TCP_RXT_CONNDROPTIME: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error) + break; + if (optval < 0) + error = EINVAL; + else + tp->rxt_conndroptime = optval * TCP_RETRANSHZ; + break; + case TCP_NOTSENT_LOWAT: + error = sooptcopyin(sopt, &optval, sizeof(optval), + sizeof(optval)); + if (error) + break; + if (optval < 0) { + error = EINVAL; + break; + } else { + if (optval == 0) { + so->so_flags &= ~(SOF_NOTSENT_LOWAT); + tp->t_notsent_lowat = 0; + } else { + so->so_flags |= SOF_NOTSENT_LOWAT; + tp->t_notsent_lowat = optval; + } + } + break; + + case SO_FLUSH: + if ((error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval))) != 0) + break; + + error = inp_flush(inp, optval); + break; + default: error = ENOPROTOOPT; break; @@ -1203,6 +1545,55 @@ tcp_ctloutput(so, sopt) case TCP_NOPUSH: optval = tp->t_flags & TF_NOPUSH; break; + case TCP_CONNECTIONTIMEOUT: + optval = tp->t_keepinit / TCP_RETRANSHZ; + break; + case PERSIST_TIMEOUT: + optval = tp->t_persist_timeout / TCP_RETRANSHZ; + break; + case TCP_RXT_CONNDROPTIME: + optval = tp->rxt_conndroptime / TCP_RETRANSHZ; + break; + case TCP_RXT_FINDROP: + optval = tp->t_flagsext & TF_RXTFINDROP; + break; + case TCP_MEASURE_SND_BW: + optval = tp->t_flagsext & TF_MEASURESNDBW; + break; + case TCP_INFO: { + struct tcp_info ti; + + tcp_fill_info(tp, &ti); + error = sooptcopyout(sopt, &ti, sizeof(struct tcp_info)); + goto done; + /* NOT REACHED */ + } + case TCP_MEASURE_BW_BURST: { + struct tcp_measure_bw_burst out; + if ((tp->t_flagsext & TF_MEASURESNDBW) == 0 || + tp->t_bwmeas == NULL) { + error = EINVAL; + break; + } + out.min_burst_size = tp->t_bwmeas->bw_minsizepkts; + out.max_burst_size = tp->t_bwmeas->bw_maxsizepkts; + error = sooptcopyout(sopt, &out, sizeof(out)); + goto done; + } + case TCP_NOTSENT_LOWAT: + if ((so->so_flags & SOF_NOTSENT_LOWAT) != 0) { + optval = tp->t_notsent_lowat; + } else { + optval = 0; + } + break; + case TCP_PEER_PID: { + pid_t pid; + error = tcp_lookup_peer_pid_locked(so, &pid); + if (error == 0) + error = sooptcopyout(sopt, &pid, sizeof(pid)); + goto done; + } default: error = ENOPROTOOPT; break; @@ -1211,6 +1602,7 @@ tcp_ctloutput(so, sopt) error = sooptcopyout(sopt, &optval, sizeof optval); break; } +done: return (error); } @@ -1219,12 +1611,47 @@ tcp_ctloutput(so, sopt) * sizes, respectively. These are obsolescent (this information should * be set by the route). */ -u_long tcp_sendspace = 1448*256; -SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW, - &tcp_sendspace , 0, "Maximum outgoing TCP datagram size"); -u_long tcp_recvspace = 1448*384; -SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, - &tcp_recvspace , 0, "Maximum incoming TCP datagram size"); +u_int32_t tcp_sendspace = 1448*256; +u_int32_t tcp_recvspace = 1448*384; + +/* During attach, the size of socket buffer allocated is limited to + * sb_max in sbreserve. Disallow setting the tcp send and recv space + * to be more than sb_max because that will cause tcp_attach to fail + * (see radar 5713060) + */ +static int +sysctl_tcp_sospace(struct sysctl_oid *oidp, __unused void *arg1, + __unused int arg2, struct sysctl_req *req) { + u_int32_t new_value = 0, *space_p = NULL; + int changed = 0, error = 0; + u_quad_t sb_effective_max = (sb_max / (MSIZE+MCLBYTES)) * MCLBYTES; + + switch (oidp->oid_number) { + case TCPCTL_SENDSPACE: + space_p = &tcp_sendspace; + break; + case TCPCTL_RECVSPACE: + space_p = &tcp_recvspace; + break; + default: + return EINVAL; + } + error = sysctl_io_number(req, *space_p, sizeof(u_int32_t), + &new_value, &changed); + if (changed) { + if (new_value > 0 && new_value <= sb_effective_max) { + *space_p = new_value; + } else { + error = ERANGE; + } + } + return error; +} + +SYSCTL_PROC(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_sendspace , 0, &sysctl_tcp_sospace, "IU", "Maximum outgoing TCP datagram size"); +SYSCTL_PROC(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, + &tcp_recvspace , 0, &sysctl_tcp_sospace, "IU", "Maximum incoming TCP datagram size"); /* @@ -1246,7 +1673,6 @@ tcp_attach(so, p) register struct tcpcb *tp; struct inpcb *inp; int error; - u_long sb_effective_max; #if INET6 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0; #endif @@ -1258,27 +1684,14 @@ tcp_attach(so, p) inp = sotoinpcb(so); if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { - /* - * The goal is to let clients machines use large send/rcv default windows to compensate for link - * latency and make sure the receiver is not constraining the sender window. - * But we doon't want to have a few connections use all our mbuf space for servers. - * This is done by watching a threshold of tcpcbs in use and bumping the default send and rcvspace - * only if that threshold isn't reached. - * We're also advertising a much bigger window size (tuneable by sysctl) in correlation with * the max socket buffer size if - * we consider that we have enough ressources for it. This window will be adjusted depending on the - * global socket layer buffer use with the use of tcp_sbpace - */ - - if (inp->inp_pcbinfo->ipi_count < tcp_sockthreshold) { - sb_effective_max = (sb_max / (MSIZE+MCLBYTES)) * MCLBYTES; - error = soreserve(so, max(min((TCP_MAXWIN << tcp_win_scale)/4, sb_effective_max), tcp_sendspace), - max(min((TCP_MAXWIN << tcp_win_scale)/2, sb_effective_max), tcp_recvspace)); - } - else - error = soreserve(so, tcp_sendspace, tcp_recvspace); + error = soreserve(so, tcp_sendspace, tcp_recvspace); if (error) return (error); } + if ((so->so_rcv.sb_flags & SB_USRSIZE) == 0) + so->so_rcv.sb_flags |= SB_AUTOSIZE; + if ((so->so_snd.sb_flags & SB_USRSIZE) == 0) + so->so_snd.sb_flags |= SB_AUTOSIZE; #if INET6 if (isipv6) { @@ -1302,6 +1715,9 @@ tcp_attach(so, p) so->so_state |= nofd; return (ENOBUFS); } + if (nstat_collect) { + nstat_tcp_new_pcb(inp); + } tp->t_state = TCPS_CLOSED; return (0); } @@ -1374,7 +1790,7 @@ tcp_usrclosed(tp) soisdisconnected(tp->t_inpcb->inp_socket); /* To prevent the connection hanging in FIN_WAIT_2 forever. */ if (tp->t_state == TCPS_FIN_WAIT_2) - tp->t_timer[TCPT_2MSL] = tcp_maxidle; + tp->t_timer[TCPT_2MSL] = OFFSET_FROM_START(tp, tcp_maxidle); } return (tp); }