xnu-517.9.5.tar.gz

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

diff --git a/bsd/netinet/tcp_subr.c b/bsd/netinet/tcp_subr.c
index 1fccf314b9809362e3d403287db62a9a904e64f6..cb7bda6538acd562dccfb2219e89779aa3036d9d 100644 (file)
--- a/bsd/netinet/tcp_subr.c
+++ b/bsd/netinet/tcp_subr.c
@@ -137,6 +137,18 @@ SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
        "Default TCP Maximum Segment Size for IPv6");
 #endif
 
+/*
+ * Minimum MSS we accept and use. This prevents DoS attacks where
+ * we are forced to a ridiculous low MSS like 20 and send hundreds
+ * of packets instead of one. The effect scales with the available
+ * bandwidth and quickly saturates the CPU and network interface
+ * with packet generation and sending. Set to zero to disable MINMSS
+ * checking. This setting prevents us from sending too small packets.
+ */
+int    tcp_minmss = TCP_MINMSS;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
+    &tcp_minmss , 0, "Minmum TCP Maximum Segment 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");
@@ -211,6 +223,7 @@ static struct tcpcb dummy_tcb;
 extern struct  inpcbhead       time_wait_slots[];
 extern int             cur_tw_slot;
 extern u_long          *delack_bitmask;
+extern u_long  route_generation;
 
 
 int  get_inpcb_str_size()
@@ -246,6 +259,9 @@ tcp_init()
        tcp_keepintvl = TCPTV_KEEPINTVL;
        tcp_maxpersistidle = TCPTV_KEEP_IDLE;
        tcp_msl = TCPTV_MSL;
+       read_random(&tcp_now, sizeof(tcp_now));
+       tcp_now  = tcp_now & 0x7fffffffffffffff; /* Starts tcp internal 500ms clock at a random value */
+
 
        LIST_INIT(&tcb);
        tcbinfo.listhead = &tcb;
@@ -268,6 +284,13 @@ tcp_init()
        tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
                                 ZONE_INTERRUPT, 0);
 #endif
+
+       tcp_reass_maxseg = nmbclusters / 16;
+#ifndef __APPLE__
+       TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
+           &tcp_reass_maxseg);
+#endif
+
 #if INET6
 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
 #else /* INET6 */
@@ -686,6 +709,9 @@ tcp_close(tp)
        register struct rtentry *rt;
        int dosavessthresh;
 
+       if ( inp->inp_ppcb == NULL) /* tcp_close was called previously, bail */
+               return;
+
 #ifndef __APPLE__
        /*
         * Make sure that all of our timers are stopped before we
@@ -696,7 +722,16 @@ tcp_close(tp)
        callout_stop(tp->tt_keep);
        callout_stop(tp->tt_2msl);
        callout_stop(tp->tt_delack);
+#else
+       /* Clear the timers before we delete the PCB. */
+       {
+               int i;
+               for (i = 0; i < TCPT_NTIMERS; i++) {
+                       tp->t_timer[i] = 0;
+               }
+       }
 #endif
+       
 
        KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_START, tp,0,0,0,0);
        switch (tp->t_state) 
@@ -734,11 +769,16 @@ tcp_close(tp)
                                goto no_valid_rt;
                }
                else
-#endif /* INET6 */             
-               if ((rt = inp->inp_route.ro_rt) == NULL ||
+#endif /* INET6 */
+               rt = inp->inp_route.ro_rt;      
+               if (rt == NULL ||
                    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr
-                   == INADDR_ANY)
+                   == INADDR_ANY || rt->generation_id != route_generation) {
+                       if (tp->t_state >= TCPS_CLOSE_WAIT)
+                               tp->t_state = TCPS_CLOSING;
+
                        goto no_valid_rt;
+               }
 
                if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
                        i = tp->t_srtt *
@@ -859,6 +899,7 @@ tcp_freeq(tp)
                LIST_REMOVE(q, tqe_q);
                m_freem(q->tqe_m);
                FREE(q, M_TSEGQ);
+               tcp_reass_qsize--;
                rv = 1;
        }
        return (rv);
@@ -889,6 +930,7 @@ tcp_drain()
                                        LIST_REMOVE(te, tqe_q);
                                        m_freem(te->tqe_m);
                                        FREE(te, M_TSEGQ);
+                                       tcp_reass_qsize--;
                                }
                        }
                }
@@ -909,7 +951,12 @@ tcp_notify(inp, error)
        struct inpcb *inp;
        int error;
 {
-       struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
+       struct tcpcb *tp;
+
+       if (inp == NULL)
+               return; /* pcb is gone already */
+
+       tp = (struct tcpcb *)inp->inp_ppcb;
 
        /*
         * Ignore some errors if we are hooked up.
@@ -1447,13 +1494,7 @@ tcp_mtudisc(inp, errno)
                if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
                    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
                        mss -= TCPOLEN_CC_APPA;
-#if    (MCLBYTES & (MCLBYTES - 1)) == 0
-               if (mss > MCLBYTES)
-                       mss &= ~(MCLBYTES-1);
-#else
-               if (mss > MCLBYTES)
-                       mss = mss / MCLBYTES * MCLBYTES;
-#endif
+
                if (so->so_snd.sb_hiwat < mss)
                        mss = so->so_snd.sb_hiwat;
 
@@ -1483,7 +1524,7 @@ tcp_rtlookup(inp)
        if (ro == NULL)
                return (NULL);
        rt = ro->ro_rt;
-       if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
+       if (rt == NULL || !(rt->rt_flags & RTF_UP) || rt->generation_id != route_generation) {
                /* No route yet, so try to acquire one */
                if (inp->inp_faddr.s_addr != INADDR_ANY) {
                        ro->ro_dst.sa_family = AF_INET;