xnu-517.9.5.tar.gz

[apple/xnu.git] / bsd / netinet / tcp_usrreq.c

diff --git a/bsd/netinet/tcp_usrreq.c b/bsd/netinet/tcp_usrreq.c
index d68e6f45d5b21048fb9a9a2a6b2b91866e5bdf6b..246e031c06c53a2a76b2fc98e309ab37b619ab0a 100644 (file)
--- a/bsd/netinet/tcp_usrreq.c
+++ b/bsd/netinet/tcp_usrreq.c
@@ -259,7 +259,7 @@ tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct proc *p)
        }
        inp->inp_vflag &= ~INP_IPV4;
        inp->inp_vflag |= INP_IPV6;
        }
        inp->inp_vflag &= ~INP_IPV4;
        inp->inp_vflag |= INP_IPV6;

-       if (ip6_mapped_addr_on && (inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
+       if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {

                if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
                        inp->inp_vflag |= INP_IPV4;
                else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
                if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
                        inp->inp_vflag |= INP_IPV4;
                else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {


@@ -310,8 +310,7 @@ tcp6_usr_listen(struct socket *so, struct proc *p)
        COMMON_START();
        if (inp->inp_lport == 0) {
                inp->inp_vflag &= ~INP_IPV4;
        COMMON_START();
        if (inp->inp_lport == 0) {
                inp->inp_vflag &= ~INP_IPV4;

-               if (ip6_mapped_addr_on &&
-                   (inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
+               if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)

                        inp->inp_vflag |= INP_IPV4;
                error = in6_pcbbind(inp, (struct sockaddr *)0, p);
        }
                        inp->inp_vflag |= INP_IPV4;
                error = in6_pcbbind(inp, (struct sockaddr *)0, p);
        }


@@ -384,9 +383,8 @@ tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct proc *p)
        if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
                struct sockaddr_in sin;
 
        if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
                struct sockaddr_in sin;
 

-               if (!ip6_mapped_addr_on ||
-                   (inp->inp_flags & IN6P_IPV6_V6ONLY))
-                       return(EINVAL);
+               if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)
+                       return (EINVAL);

 
                in6_sin6_2_sin(&sin, sin6p);
                inp->inp_vflag |= INP_IPV4;
 
                in6_sin6_2_sin(&sin, sin6p);
                inp->inp_vflag |= INP_IPV4;


@@ -984,12 +982,24 @@ tcp_ctloutput(so, sopt)
                        if (error)
                                break;
 
                        if (error)
                                break;
 

-                       if (optval > 0 && optval <= tp->t_maxseg)
+                       if (optval > 0 && optval <= tp->t_maxseg &&
+                           optval + 40 >= tcp_minmss)

                                tp->t_maxseg = optval;
                        else
                                error = EINVAL;
                        break;
 
                                tp->t_maxseg = optval;
                        else
                                error = EINVAL;
                        break;
 

+                case TCP_KEEPALIVE:
+                        error = sooptcopyin(sopt, &optval, sizeof optval,
+                                            sizeof optval);
+                        if (error)
+                                break;
+                        if (optval < 0)
+                                error = EINVAL;
+                       else
+                               tp->t_keepidle = optval * PR_SLOWHZ;
+                        break;
+               

                default:
                        error = ENOPROTOOPT;
                        break;
                default:
                        error = ENOPROTOOPT;
                        break;


@@ -1004,6 +1014,9 @@ tcp_ctloutput(so, sopt)
                case TCP_MAXSEG:
                        optval = tp->t_maxseg;
                        break;
                case TCP_MAXSEG:
                        optval = tp->t_maxseg;
                        break;

+               case TCP_KEEPALIVE:
+                       optval = tp->t_keepidle / PR_SLOWHZ;
+                       break;

                case TCP_NOOPT:
                        optval = tp->t_flags & TF_NOOPT;
                        break;
                case TCP_NOOPT:
                        optval = tp->t_flags & TF_NOOPT;
                        break;


@@ -1034,6 +1047,11 @@ u_long   tcp_recvspace = 1024*16;
 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 
     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
 
 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 
     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
 

+__private_extern__ int tcp_sockthreshold = 256;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, sockthreshold, CTLFLAG_RW, 
+    &tcp_sockthreshold , 0, "TCP Socket size increased if less than threshold");
+
+#define TCP_INCREASED_SPACE    65535   /* Automatically increase tcp send/rcv space to this value */

 /*
  * Attach TCP protocol to socket, allocating
  * internet protocol control block, tcp control block,
 /*
  * Attach TCP protocol to socket, allocating
  * internet protocol control block, tcp control block,


@@ -1051,15 +1069,28 @@ tcp_attach(so, p)
        int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != NULL;
 #endif
 
        int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != NULL;
 #endif
 

-       if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
-               error = soreserve(so, tcp_sendspace, tcp_recvspace);
-               if (error)
-                       return (error);
-       }

        error = in_pcballoc(so, &tcbinfo, p);
        if (error)
                return (error);
        error = in_pcballoc(so, &tcbinfo, p);
        if (error)
                return (error);

+

        inp = sotoinpcb(so);
        inp = sotoinpcb(so);

+
+       if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
+               /*
+                * The goal is to let clients have large send/rcv default windows (TCP_INCREASED_SPACE)
+                * while not hogging mbuf space for servers. This is done by watching a threshold
+                * of tcpcbs in use and bumping the default send and rcvspace only if under that threshold.
+                * The theory being that busy servers have a lot more active tcpcbs and don't want the potential
+                * memory penalty of having much larger sockbuffs. The sysctl allows to fine tune that threshold value.          */
+
+               if (inp->inp_pcbinfo->ipi_count < tcp_sockthreshold)
+                       error = soreserve(so, MAX(TCP_INCREASED_SPACE, tcp_sendspace), MAX(TCP_INCREASED_SPACE,tcp_recvspace));
+               else    
+                       error = soreserve(so, tcp_sendspace, tcp_recvspace);
+               if (error)
+                       return (error);
+       }
+

 #if INET6
        if (isipv6) {
                inp->inp_vflag |= INP_IPV6;
 #if INET6
        if (isipv6) {
                inp->inp_vflag |= INP_IPV6;