The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/netinet/tcp_subr.c

Version: -  FREEBSD  -  FREEBSD11  -  FREEBSD10  -  FREEBSD9  -  FREEBSD92  -  FREEBSD91  -  FREEBSD90  -  FREEBSD8  -  FREEBSD82  -  FREEBSD81  -  FREEBSD80  -  FREEBSD7  -  FREEBSD74  -  FREEBSD73  -  FREEBSD72  -  FREEBSD71  -  FREEBSD70  -  FREEBSD6  -  FREEBSD64  -  FREEBSD63  -  FREEBSD62  -  FREEBSD61  -  FREEBSD60  -  FREEBSD5  -  FREEBSD55  -  FREEBSD54  -  FREEBSD53  -  FREEBSD52  -  FREEBSD51  -  FREEBSD50  -  FREEBSD4  -  FREEBSD3  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * Redistribution and use in source and binary forms, with or without
    6  * modification, are permitted provided that the following conditions
    7  * are met:
    8  * 1. Redistributions of source code must retain the above copyright
    9  *    notice, this list of conditions and the following disclaimer.
   10  * 2. Redistributions in binary form must reproduce the above copyright
   11  *    notice, this list of conditions and the following disclaimer in the
   12  *    documentation and/or other materials provided with the distribution.
   13  * 4. Neither the name of the University nor the names of its contributors
   14  *    may be used to endorse or promote products derived from this software
   15  *    without specific prior written permission.
   16  *
   17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   27  * SUCH DAMAGE.
   28  *
   29  *      @(#)tcp_subr.c  8.2 (Berkeley) 5/24/95
   30  */
   31 
   32 #include <sys/cdefs.h>
   33 __FBSDID("$FreeBSD: releng/8.2/sys/netinet/tcp_subr.c 214860 2010-11-06 09:42:41Z lstewart $");
   34 
   35 #include "opt_compat.h"
   36 #include "opt_inet.h"
   37 #include "opt_inet6.h"
   38 #include "opt_ipsec.h"
   39 #include "opt_tcpdebug.h"
   40 
   41 #include <sys/param.h>
   42 #include <sys/systm.h>
   43 #include <sys/callout.h>
   44 #include <sys/kernel.h>
   45 #include <sys/sysctl.h>
   46 #include <sys/jail.h>
   47 #include <sys/malloc.h>
   48 #include <sys/mbuf.h>
   49 #ifdef INET6
   50 #include <sys/domain.h>
   51 #endif
   52 #include <sys/priv.h>
   53 #include <sys/proc.h>
   54 #include <sys/socket.h>
   55 #include <sys/socketvar.h>
   56 #include <sys/protosw.h>
   57 #include <sys/random.h>
   58 
   59 #include <vm/uma.h>
   60 
   61 #include <net/route.h>
   62 #include <net/if.h>
   63 #include <net/vnet.h>
   64 
   65 #include <netinet/in.h>
   66 #include <netinet/in_systm.h>
   67 #include <netinet/ip.h>
   68 #ifdef INET6
   69 #include <netinet/ip6.h>
   70 #endif
   71 #include <netinet/in_pcb.h>
   72 #ifdef INET6
   73 #include <netinet6/in6_pcb.h>
   74 #endif
   75 #include <netinet/in_var.h>
   76 #include <netinet/ip_var.h>
   77 #ifdef INET6
   78 #include <netinet6/ip6_var.h>
   79 #include <netinet6/scope6_var.h>
   80 #include <netinet6/nd6.h>
   81 #endif
   82 #include <netinet/ip_icmp.h>
   83 #include <netinet/tcp.h>
   84 #include <netinet/tcp_fsm.h>
   85 #include <netinet/tcp_seq.h>
   86 #include <netinet/tcp_timer.h>
   87 #include <netinet/tcp_var.h>
   88 #include <netinet/tcp_syncache.h>
   89 #include <netinet/tcp_offload.h>
   90 #ifdef INET6
   91 #include <netinet6/tcp6_var.h>
   92 #endif
   93 #include <netinet/tcpip.h>
   94 #ifdef TCPDEBUG
   95 #include <netinet/tcp_debug.h>
   96 #endif
   97 #include <netinet6/ip6protosw.h>
   98 
   99 #ifdef IPSEC
  100 #include <netipsec/ipsec.h>
  101 #include <netipsec/xform.h>
  102 #ifdef INET6
  103 #include <netipsec/ipsec6.h>
  104 #endif
  105 #include <netipsec/key.h>
  106 #include <sys/syslog.h>
  107 #endif /*IPSEC*/
  108 
  109 #include <machine/in_cksum.h>
  110 #include <sys/md5.h>
  111 
  112 #include <security/mac/mac_framework.h>
  113 
  114 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
  115 #ifdef INET6
  116 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
  117 #endif
  118 
  119 static int
  120 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
  121 {
  122         int error, new;
  123 
  124         new = V_tcp_mssdflt;
  125         error = sysctl_handle_int(oidp, &new, 0, req);
  126         if (error == 0 && req->newptr) {
  127                 if (new < TCP_MINMSS)
  128                         error = EINVAL;
  129                 else
  130                         V_tcp_mssdflt = new;
  131         }
  132         return (error);
  133 }
  134 
  135 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
  136     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
  137     &sysctl_net_inet_tcp_mss_check, "I",
  138     "Default TCP Maximum Segment Size");
  139 
  140 #ifdef INET6
  141 static int
  142 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
  143 {
  144         int error, new;
  145 
  146         new = V_tcp_v6mssdflt;
  147         error = sysctl_handle_int(oidp, &new, 0, req);
  148         if (error == 0 && req->newptr) {
  149                 if (new < TCP_MINMSS)
  150                         error = EINVAL;
  151                 else
  152                         V_tcp_v6mssdflt = new;
  153         }
  154         return (error);
  155 }
  156 
  157 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
  158     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
  159     &sysctl_net_inet_tcp_mss_v6_check, "I",
  160    "Default TCP Maximum Segment Size for IPv6");
  161 #endif
  162 
  163 static int
  164 vnet_sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
  165 {
  166 
  167         VNET_SYSCTL_ARG(req, arg1);
  168         return (sysctl_msec_to_ticks(oidp, arg1, arg2, req));
  169 }
  170 
  171 /*
  172  * Minimum MSS we accept and use. This prevents DoS attacks where
  173  * we are forced to a ridiculous low MSS like 20 and send hundreds
  174  * of packets instead of one. The effect scales with the available
  175  * bandwidth and quickly saturates the CPU and network interface
  176  * with packet generation and sending. Set to zero to disable MINMSS
  177  * checking. This setting prevents us from sending too small packets.
  178  */
  179 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
  180 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
  181      &VNET_NAME(tcp_minmss), 0,
  182     "Minmum TCP Maximum Segment Size");
  183 
  184 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
  185 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
  186     &VNET_NAME(tcp_do_rfc1323), 0,
  187     "Enable rfc1323 (high performance TCP) extensions");
  188 
  189 static int      tcp_log_debug = 0;
  190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
  191     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
  192 
  193 static int      tcp_tcbhashsize = 0;
  194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
  195     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
  196 
  197 static int      do_tcpdrain = 1;
  198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
  199     "Enable tcp_drain routine for extra help when low on mbufs");
  200 
  201 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
  202     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
  203 
  204 static VNET_DEFINE(int, icmp_may_rst) = 1;
  205 #define V_icmp_may_rst                  VNET(icmp_may_rst)
  206 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
  207     &VNET_NAME(icmp_may_rst), 0,
  208     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
  209 
  210 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
  211 #define V_tcp_isn_reseed_interval       VNET(tcp_isn_reseed_interval)
  212 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
  213     &VNET_NAME(tcp_isn_reseed_interval), 0,
  214     "Seconds between reseeding of ISN secret");
  215 
  216 /*
  217  * TCP bandwidth limiting sysctls.  Note that the default lower bound of
  218  * 1024 exists only for debugging.  A good production default would be
  219  * something like 6100.
  220  */
  221 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
  222     "TCP inflight data limiting");
  223 
  224 static VNET_DEFINE(int, tcp_inflight_enable) = 0;
  225 #define V_tcp_inflight_enable           VNET(tcp_inflight_enable)
  226 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
  227     &VNET_NAME(tcp_inflight_enable), 0,
  228     "Enable automatic TCP inflight data limiting");
  229 
  230 static int      tcp_inflight_debug = 0;
  231 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
  232     &tcp_inflight_debug, 0,
  233     "Debug TCP inflight calculations");
  234 
  235 static VNET_DEFINE(int, tcp_inflight_rttthresh);
  236 #define V_tcp_inflight_rttthresh        VNET(tcp_inflight_rttthresh)
  237 SYSCTL_VNET_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh,
  238     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_inflight_rttthresh), 0,
  239     vnet_sysctl_msec_to_ticks, "I",
  240     "RTT threshold below which inflight will deactivate itself");
  241 
  242 static VNET_DEFINE(int, tcp_inflight_min) = 6144;
  243 #define V_tcp_inflight_min              VNET(tcp_inflight_min)
  244 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
  245     &VNET_NAME(tcp_inflight_min), 0,
  246     "Lower-bound for TCP inflight window");
  247 
  248 static VNET_DEFINE(int, tcp_inflight_max) = TCP_MAXWIN << TCP_MAX_WINSHIFT;
  249 #define V_tcp_inflight_max              VNET(tcp_inflight_max)
  250 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
  251     &VNET_NAME(tcp_inflight_max), 0,
  252     "Upper-bound for TCP inflight window");
  253 
  254 static VNET_DEFINE(int, tcp_inflight_stab) = 20;
  255 #define V_tcp_inflight_stab             VNET(tcp_inflight_stab)
  256 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
  257     &VNET_NAME(tcp_inflight_stab), 0,
  258     "Inflight Algorithm Stabilization 20 = 2 packets");
  259 
  260 VNET_DEFINE(uma_zone_t, sack_hole_zone);
  261 #define V_sack_hole_zone                VNET(sack_hole_zone)
  262 
  263 static struct inpcb *tcp_notify(struct inpcb *, int);
  264 static void     tcp_isn_tick(void *);
  265 static char *   tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
  266                     void *ip4hdr, const void *ip6hdr);
  267 
  268 /*
  269  * Target size of TCP PCB hash tables. Must be a power of two.
  270  *
  271  * Note that this can be overridden by the kernel environment
  272  * variable net.inet.tcp.tcbhashsize
  273  */
  274 #ifndef TCBHASHSIZE
  275 #define TCBHASHSIZE     512
  276 #endif
  277 
  278 /*
  279  * XXX
  280  * Callouts should be moved into struct tcp directly.  They are currently
  281  * separate because the tcpcb structure is exported to userland for sysctl
  282  * parsing purposes, which do not know about callouts.
  283  */
  284 struct tcpcb_mem {
  285         struct  tcpcb           tcb;
  286         struct  tcp_timer       tt;
  287 };
  288 
  289 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
  290 #define V_tcpcb_zone                    VNET(tcpcb_zone)
  291 
  292 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
  293 struct callout isn_callout;
  294 static struct mtx isn_mtx;
  295 
  296 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
  297 #define ISN_LOCK()      mtx_lock(&isn_mtx)
  298 #define ISN_UNLOCK()    mtx_unlock(&isn_mtx)
  299 
  300 /*
  301  * TCP initialization.
  302  */
  303 static void
  304 tcp_zone_change(void *tag)
  305 {
  306 
  307         uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
  308         uma_zone_set_max(V_tcpcb_zone, maxsockets);
  309         tcp_tw_zone_change();
  310 }
  311 
  312 static int
  313 tcp_inpcb_init(void *mem, int size, int flags)
  314 {
  315         struct inpcb *inp = mem;
  316 
  317         INP_LOCK_INIT(inp, "inp", "tcpinp");
  318         return (0);
  319 }
  320 
  321 void
  322 tcp_init(void)
  323 {
  324         int hashsize;
  325 
  326         INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
  327         LIST_INIT(&V_tcb);
  328 #ifdef VIMAGE
  329         V_tcbinfo.ipi_vnet = curvnet;
  330 #endif
  331         V_tcbinfo.ipi_listhead = &V_tcb;
  332         hashsize = TCBHASHSIZE;
  333         TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
  334         if (!powerof2(hashsize)) {
  335                 printf("WARNING: TCB hash size not a power of 2\n");
  336                 hashsize = 512; /* safe default */
  337         }
  338         V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
  339             &V_tcbinfo.ipi_hashmask);
  340         V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
  341             &V_tcbinfo.ipi_porthashmask);
  342         V_tcbinfo.ipi_zone = uma_zcreate("tcp_inpcb", sizeof(struct inpcb),
  343             NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
  344         uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
  345         V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
  346 
  347         /*
  348          * These have to be type stable for the benefit of the timers.
  349          */
  350         V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
  351             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
  352         uma_zone_set_max(V_tcpcb_zone, maxsockets);
  353 
  354         tcp_tw_init();
  355         syncache_init();
  356         tcp_hc_init();
  357         tcp_reass_init();
  358 
  359         TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
  360         V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
  361             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
  362 
  363         /* Skip initialization of globals for non-default instances. */
  364         if (!IS_DEFAULT_VNET(curvnet))
  365                 return;
  366 
  367         /* XXX virtualize those bellow? */
  368         tcp_delacktime = TCPTV_DELACK;
  369         tcp_keepinit = TCPTV_KEEP_INIT;
  370         tcp_keepidle = TCPTV_KEEP_IDLE;
  371         tcp_keepintvl = TCPTV_KEEPINTVL;
  372         tcp_maxpersistidle = TCPTV_KEEP_IDLE;
  373         tcp_msl = TCPTV_MSL;
  374         tcp_rexmit_min = TCPTV_MIN;
  375         if (tcp_rexmit_min < 1)
  376                 tcp_rexmit_min = 1;
  377         tcp_rexmit_slop = TCPTV_CPU_VAR;
  378         tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
  379         tcp_tcbhashsize = hashsize;
  380 
  381 #ifdef INET6
  382 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
  383 #else /* INET6 */
  384 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
  385 #endif /* INET6 */
  386         if (max_protohdr < TCP_MINPROTOHDR)
  387                 max_protohdr = TCP_MINPROTOHDR;
  388         if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
  389                 panic("tcp_init");
  390 #undef TCP_MINPROTOHDR
  391 
  392         ISN_LOCK_INIT();
  393         callout_init(&isn_callout, CALLOUT_MPSAFE);
  394         callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
  395         EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
  396                 SHUTDOWN_PRI_DEFAULT);
  397         EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
  398                 EVENTHANDLER_PRI_ANY);
  399 }
  400 
  401 #ifdef VIMAGE
  402 void
  403 tcp_destroy(void)
  404 {
  405 
  406         tcp_reass_destroy();
  407         tcp_hc_destroy();
  408         syncache_destroy();
  409         tcp_tw_destroy();
  410 
  411         /* XXX check that hashes are empty! */
  412         hashdestroy(V_tcbinfo.ipi_hashbase, M_PCB,
  413             V_tcbinfo.ipi_hashmask);
  414         hashdestroy(V_tcbinfo.ipi_porthashbase, M_PCB,
  415             V_tcbinfo.ipi_porthashmask);
  416 
  417         uma_zdestroy(V_sack_hole_zone);
  418         uma_zdestroy(V_tcpcb_zone);
  419         uma_zdestroy(V_tcbinfo.ipi_zone);
  420 
  421         INP_INFO_LOCK_DESTROY(&V_tcbinfo);
  422 }
  423 #endif
  424 
  425 void
  426 tcp_fini(void *xtp)
  427 {
  428 
  429         callout_stop(&isn_callout);
  430 }
  431 
  432 /*
  433  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
  434  * tcp_template used to store this data in mbufs, but we now recopy it out
  435  * of the tcpcb each time to conserve mbufs.
  436  */
  437 void
  438 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
  439 {
  440         struct tcphdr *th = (struct tcphdr *)tcp_ptr;
  441 
  442         INP_WLOCK_ASSERT(inp);
  443 
  444 #ifdef INET6
  445         if ((inp->inp_vflag & INP_IPV6) != 0) {
  446                 struct ip6_hdr *ip6;
  447 
  448                 ip6 = (struct ip6_hdr *)ip_ptr;
  449                 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
  450                         (inp->inp_flow & IPV6_FLOWINFO_MASK);
  451                 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
  452                         (IPV6_VERSION & IPV6_VERSION_MASK);
  453                 ip6->ip6_nxt = IPPROTO_TCP;
  454                 ip6->ip6_plen = htons(sizeof(struct tcphdr));
  455                 ip6->ip6_src = inp->in6p_laddr;
  456                 ip6->ip6_dst = inp->in6p_faddr;
  457         } else
  458 #endif
  459         {
  460                 struct ip *ip;
  461 
  462                 ip = (struct ip *)ip_ptr;
  463                 ip->ip_v = IPVERSION;
  464                 ip->ip_hl = 5;
  465                 ip->ip_tos = inp->inp_ip_tos;
  466                 ip->ip_len = 0;
  467                 ip->ip_id = 0;
  468                 ip->ip_off = 0;
  469                 ip->ip_ttl = inp->inp_ip_ttl;
  470                 ip->ip_sum = 0;
  471                 ip->ip_p = IPPROTO_TCP;
  472                 ip->ip_src = inp->inp_laddr;
  473                 ip->ip_dst = inp->inp_faddr;
  474         }
  475         th->th_sport = inp->inp_lport;
  476         th->th_dport = inp->inp_fport;
  477         th->th_seq = 0;
  478         th->th_ack = 0;
  479         th->th_x2 = 0;
  480         th->th_off = 5;
  481         th->th_flags = 0;
  482         th->th_win = 0;
  483         th->th_urp = 0;
  484         th->th_sum = 0;         /* in_pseudo() is called later for ipv4 */
  485 }
  486 
  487 /*
  488  * Create template to be used to send tcp packets on a connection.
  489  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
  490  * use for this function is in keepalives, which use tcp_respond.
  491  */
  492 struct tcptemp *
  493 tcpip_maketemplate(struct inpcb *inp)
  494 {
  495         struct tcptemp *t;
  496 
  497         t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
  498         if (t == NULL)
  499                 return (NULL);
  500         tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
  501         return (t);
  502 }
  503 
  504 /*
  505  * Send a single message to the TCP at address specified by
  506  * the given TCP/IP header.  If m == NULL, then we make a copy
  507  * of the tcpiphdr at ti and send directly to the addressed host.
  508  * This is used to force keep alive messages out using the TCP
  509  * template for a connection.  If flags are given then we send
  510  * a message back to the TCP which originated the * segment ti,
  511  * and discard the mbuf containing it and any other attached mbufs.
  512  *
  513  * In any case the ack and sequence number of the transmitted
  514  * segment are as specified by the parameters.
  515  *
  516  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
  517  */
  518 void
  519 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
  520     tcp_seq ack, tcp_seq seq, int flags)
  521 {
  522         int tlen;
  523         int win = 0;
  524         struct ip *ip;
  525         struct tcphdr *nth;
  526 #ifdef INET6
  527         struct ip6_hdr *ip6;
  528         int isipv6;
  529 #endif /* INET6 */
  530         int ipflags = 0;
  531         struct inpcb *inp;
  532 
  533         KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
  534 
  535 #ifdef INET6
  536         isipv6 = ((struct ip *)ipgen)->ip_v == 6;
  537         ip6 = ipgen;
  538 #endif /* INET6 */
  539         ip = ipgen;
  540 
  541         if (tp != NULL) {
  542                 inp = tp->t_inpcb;
  543                 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
  544                 INP_WLOCK_ASSERT(inp);
  545         } else
  546                 inp = NULL;
  547 
  548         if (tp != NULL) {
  549                 if (!(flags & TH_RST)) {
  550                         win = sbspace(&inp->inp_socket->so_rcv);
  551                         if (win > (long)TCP_MAXWIN << tp->rcv_scale)
  552                                 win = (long)TCP_MAXWIN << tp->rcv_scale;
  553                 }
  554         }
  555         if (m == NULL) {
  556                 m = m_gethdr(M_DONTWAIT, MT_DATA);
  557                 if (m == NULL)
  558                         return;
  559                 tlen = 0;
  560                 m->m_data += max_linkhdr;
  561 #ifdef INET6
  562                 if (isipv6) {
  563                         bcopy((caddr_t)ip6, mtod(m, caddr_t),
  564                               sizeof(struct ip6_hdr));
  565                         ip6 = mtod(m, struct ip6_hdr *);
  566                         nth = (struct tcphdr *)(ip6 + 1);
  567                 } else
  568 #endif /* INET6 */
  569               {
  570                 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
  571                 ip = mtod(m, struct ip *);
  572                 nth = (struct tcphdr *)(ip + 1);
  573               }
  574                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
  575                 flags = TH_ACK;
  576         } else {
  577                 /*
  578                  *  reuse the mbuf. 
  579                  * XXX MRT We inherrit the FIB, which is lucky.
  580                  */
  581                 m_freem(m->m_next);
  582                 m->m_next = NULL;
  583                 m->m_data = (caddr_t)ipgen;
  584                 /* m_len is set later */
  585                 tlen = 0;
  586 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
  587 #ifdef INET6
  588                 if (isipv6) {
  589                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
  590                         nth = (struct tcphdr *)(ip6 + 1);
  591                 } else
  592 #endif /* INET6 */
  593               {
  594                 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
  595                 nth = (struct tcphdr *)(ip + 1);
  596               }
  597                 if (th != nth) {
  598                         /*
  599                          * this is usually a case when an extension header
  600                          * exists between the IPv6 header and the
  601                          * TCP header.
  602                          */
  603                         nth->th_sport = th->th_sport;
  604                         nth->th_dport = th->th_dport;
  605                 }
  606                 xchg(nth->th_dport, nth->th_sport, uint16_t);
  607 #undef xchg
  608         }
  609 #ifdef INET6
  610         if (isipv6) {
  611                 ip6->ip6_flow = 0;
  612                 ip6->ip6_vfc = IPV6_VERSION;
  613                 ip6->ip6_nxt = IPPROTO_TCP;
  614                 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
  615                                                 tlen));
  616                 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
  617         } else
  618 #endif
  619         {
  620                 tlen += sizeof (struct tcpiphdr);
  621                 ip->ip_len = tlen;
  622                 ip->ip_ttl = V_ip_defttl;
  623                 if (V_path_mtu_discovery)
  624                         ip->ip_off |= IP_DF;
  625         }
  626         m->m_len = tlen;
  627         m->m_pkthdr.len = tlen;
  628         m->m_pkthdr.rcvif = NULL;
  629 #ifdef MAC
  630         if (inp != NULL) {
  631                 /*
  632                  * Packet is associated with a socket, so allow the
  633                  * label of the response to reflect the socket label.
  634                  */
  635                 INP_WLOCK_ASSERT(inp);
  636                 mac_inpcb_create_mbuf(inp, m);
  637         } else {
  638                 /*
  639                  * Packet is not associated with a socket, so possibly
  640                  * update the label in place.
  641                  */
  642                 mac_netinet_tcp_reply(m);
  643         }
  644 #endif
  645         nth->th_seq = htonl(seq);
  646         nth->th_ack = htonl(ack);
  647         nth->th_x2 = 0;
  648         nth->th_off = sizeof (struct tcphdr) >> 2;
  649         nth->th_flags = flags;
  650         if (tp != NULL)
  651                 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
  652         else
  653                 nth->th_win = htons((u_short)win);
  654         nth->th_urp = 0;
  655 #ifdef INET6
  656         if (isipv6) {
  657                 nth->th_sum = 0;
  658                 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
  659                                         sizeof(struct ip6_hdr),
  660                                         tlen - sizeof(struct ip6_hdr));
  661                 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
  662                     NULL, NULL);
  663         } else
  664 #endif /* INET6 */
  665         {
  666                 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
  667                     htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
  668                 m->m_pkthdr.csum_flags = CSUM_TCP;
  669                 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
  670         }
  671 #ifdef TCPDEBUG
  672         if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
  673                 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
  674 #endif
  675 #ifdef INET6
  676         if (isipv6)
  677                 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
  678         else
  679 #endif /* INET6 */
  680         (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
  681 }
  682 
  683 /*
  684  * Create a new TCP control block, making an
  685  * empty reassembly queue and hooking it to the argument
  686  * protocol control block.  The `inp' parameter must have
  687  * come from the zone allocator set up in tcp_init().
  688  */
  689 struct tcpcb *
  690 tcp_newtcpcb(struct inpcb *inp)
  691 {
  692         struct tcpcb_mem *tm;
  693         struct tcpcb *tp;
  694 #ifdef INET6
  695         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
  696 #endif /* INET6 */
  697 
  698         tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
  699         if (tm == NULL)
  700                 return (NULL);
  701         tp = &tm->tcb;
  702 #ifdef VIMAGE
  703         tp->t_vnet = inp->inp_vnet;
  704 #endif
  705         tp->t_timers = &tm->tt;
  706         /*      LIST_INIT(&tp->t_segq); */      /* XXX covered by M_ZERO */
  707         tp->t_maxseg = tp->t_maxopd =
  708 #ifdef INET6
  709                 isipv6 ? V_tcp_v6mssdflt :
  710 #endif /* INET6 */
  711                 V_tcp_mssdflt;
  712 
  713         /* Set up our timeouts. */
  714         callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
  715         callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
  716         callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
  717         callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
  718         callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
  719 
  720         if (V_tcp_do_rfc1323)
  721                 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
  722         if (V_tcp_do_sack)
  723                 tp->t_flags |= TF_SACK_PERMIT;
  724         TAILQ_INIT(&tp->snd_holes);
  725         tp->t_inpcb = inp;      /* XXX */
  726         /*
  727          * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
  728          * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
  729          * reasonable initial retransmit time.
  730          */
  731         tp->t_srtt = TCPTV_SRTTBASE;
  732         tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
  733         tp->t_rttmin = tcp_rexmit_min;
  734         tp->t_rxtcur = TCPTV_RTOBASE;
  735         tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
  736         tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
  737         tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
  738         tp->t_rcvtime = ticks;
  739         tp->t_bw_rtttime = ticks;
  740         /*
  741          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
  742          * because the socket may be bound to an IPv6 wildcard address,
  743          * which may match an IPv4-mapped IPv6 address.
  744          */
  745         inp->inp_ip_ttl = V_ip_defttl;
  746         inp->inp_ppcb = tp;
  747         return (tp);            /* XXX */
  748 }
  749 
  750 /*
  751  * Drop a TCP connection, reporting
  752  * the specified error.  If connection is synchronized,
  753  * then send a RST to peer.
  754  */
  755 struct tcpcb *
  756 tcp_drop(struct tcpcb *tp, int errno)
  757 {
  758         struct socket *so = tp->t_inpcb->inp_socket;
  759 
  760         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
  761         INP_WLOCK_ASSERT(tp->t_inpcb);
  762 
  763         if (TCPS_HAVERCVDSYN(tp->t_state)) {
  764                 tp->t_state = TCPS_CLOSED;
  765                 (void) tcp_output_reset(tp);
  766                 TCPSTAT_INC(tcps_drops);
  767         } else
  768                 TCPSTAT_INC(tcps_conndrops);
  769         if (errno == ETIMEDOUT && tp->t_softerror)
  770                 errno = tp->t_softerror;
  771         so->so_error = errno;
  772         return (tcp_close(tp));
  773 }
  774 
  775 void
  776 tcp_discardcb(struct tcpcb *tp)
  777 {
  778         struct inpcb *inp = tp->t_inpcb;
  779         struct socket *so = inp->inp_socket;
  780 #ifdef INET6
  781         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
  782 #endif /* INET6 */
  783 
  784         INP_WLOCK_ASSERT(inp);
  785 
  786         /*
  787          * Make sure that all of our timers are stopped before we
  788          * delete the PCB.
  789          */
  790         callout_stop(&tp->t_timers->tt_rexmt);
  791         callout_stop(&tp->t_timers->tt_persist);
  792         callout_stop(&tp->t_timers->tt_keep);
  793         callout_stop(&tp->t_timers->tt_2msl);
  794         callout_stop(&tp->t_timers->tt_delack);
  795 
  796         /*
  797          * If we got enough samples through the srtt filter,
  798          * save the rtt and rttvar in the routing entry.
  799          * 'Enough' is arbitrarily defined as 4 rtt samples.
  800          * 4 samples is enough for the srtt filter to converge
  801          * to within enough % of the correct value; fewer samples
  802          * and we could save a bogus rtt. The danger is not high
  803          * as tcp quickly recovers from everything.
  804          * XXX: Works very well but needs some more statistics!
  805          */
  806         if (tp->t_rttupdated >= 4) {
  807                 struct hc_metrics_lite metrics;
  808                 u_long ssthresh;
  809 
  810                 bzero(&metrics, sizeof(metrics));
  811                 /*
  812                  * Update the ssthresh always when the conditions below
  813                  * are satisfied. This gives us better new start value
  814                  * for the congestion avoidance for new connections.
  815                  * ssthresh is only set if packet loss occured on a session.
  816                  *
  817                  * XXXRW: 'so' may be NULL here, and/or socket buffer may be
  818                  * being torn down.  Ideally this code would not use 'so'.
  819                  */
  820                 ssthresh = tp->snd_ssthresh;
  821                 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
  822                         /*
  823                          * convert the limit from user data bytes to
  824                          * packets then to packet data bytes.
  825                          */
  826                         ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
  827                         if (ssthresh < 2)
  828                                 ssthresh = 2;
  829                         ssthresh *= (u_long)(tp->t_maxseg +
  830 #ifdef INET6
  831                                       (isipv6 ? sizeof (struct ip6_hdr) +
  832                                                sizeof (struct tcphdr) :
  833 #endif
  834                                        sizeof (struct tcpiphdr)
  835 #ifdef INET6
  836                                        )
  837 #endif
  838                                       );
  839                 } else
  840                         ssthresh = 0;
  841                 metrics.rmx_ssthresh = ssthresh;
  842 
  843                 metrics.rmx_rtt = tp->t_srtt;
  844                 metrics.rmx_rttvar = tp->t_rttvar;
  845                 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
  846                 metrics.rmx_bandwidth = tp->snd_bandwidth;
  847                 metrics.rmx_cwnd = tp->snd_cwnd;
  848                 metrics.rmx_sendpipe = 0;
  849                 metrics.rmx_recvpipe = 0;
  850 
  851                 tcp_hc_update(&inp->inp_inc, &metrics);
  852         }
  853 
  854         /* free the reassembly queue, if any */
  855         tcp_reass_flush(tp);
  856         /* Disconnect offload device, if any. */
  857         tcp_offload_detach(tp);
  858                 
  859         tcp_free_sackholes(tp);
  860         inp->inp_ppcb = NULL;
  861         tp->t_inpcb = NULL;
  862         uma_zfree(V_tcpcb_zone, tp);
  863 }
  864 
  865 /*
  866  * Attempt to close a TCP control block, marking it as dropped, and freeing
  867  * the socket if we hold the only reference.
  868  */
  869 struct tcpcb *
  870 tcp_close(struct tcpcb *tp)
  871 {
  872         struct inpcb *inp = tp->t_inpcb;
  873         struct socket *so;
  874 
  875         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
  876         INP_WLOCK_ASSERT(inp);
  877 
  878         /* Notify any offload devices of listener close */
  879         if (tp->t_state == TCPS_LISTEN)
  880                 tcp_offload_listen_close(tp);
  881         in_pcbdrop(inp);
  882         TCPSTAT_INC(tcps_closed);
  883         KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
  884         so = inp->inp_socket;
  885         soisdisconnected(so);
  886         if (inp->inp_flags & INP_SOCKREF) {
  887                 KASSERT(so->so_state & SS_PROTOREF,
  888                     ("tcp_close: !SS_PROTOREF"));
  889                 inp->inp_flags &= ~INP_SOCKREF;
  890                 INP_WUNLOCK(inp);
  891                 ACCEPT_LOCK();
  892                 SOCK_LOCK(so);
  893                 so->so_state &= ~SS_PROTOREF;
  894                 sofree(so);
  895                 return (NULL);
  896         }
  897         return (tp);
  898 }
  899 
  900 void
  901 tcp_drain(void)
  902 {
  903         VNET_ITERATOR_DECL(vnet_iter);
  904 
  905         if (!do_tcpdrain)
  906                 return;
  907 
  908         VNET_LIST_RLOCK_NOSLEEP();
  909         VNET_FOREACH(vnet_iter) {
  910                 CURVNET_SET(vnet_iter);
  911                 struct inpcb *inpb;
  912                 struct tcpcb *tcpb;
  913 
  914         /*
  915          * Walk the tcpbs, if existing, and flush the reassembly queue,
  916          * if there is one...
  917          * XXX: The "Net/3" implementation doesn't imply that the TCP
  918          *      reassembly queue should be flushed, but in a situation
  919          *      where we're really low on mbufs, this is potentially
  920          *      usefull.
  921          */
  922                 INP_INFO_RLOCK(&V_tcbinfo);
  923                 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
  924                         if (inpb->inp_flags & INP_TIMEWAIT)
  925                                 continue;
  926                         INP_WLOCK(inpb);
  927                         if ((tcpb = intotcpcb(inpb)) != NULL) {
  928                                 tcp_reass_flush(tcpb);
  929                                 tcp_clean_sackreport(tcpb);
  930                         }
  931                         INP_WUNLOCK(inpb);
  932                 }
  933                 INP_INFO_RUNLOCK(&V_tcbinfo);
  934                 CURVNET_RESTORE();
  935         }
  936         VNET_LIST_RUNLOCK_NOSLEEP();
  937 }
  938 
  939 /*
  940  * Notify a tcp user of an asynchronous error;
  941  * store error as soft error, but wake up user
  942  * (for now, won't do anything until can select for soft error).
  943  *
  944  * Do not wake up user since there currently is no mechanism for
  945  * reporting soft errors (yet - a kqueue filter may be added).
  946  */
  947 static struct inpcb *
  948 tcp_notify(struct inpcb *inp, int error)
  949 {
  950         struct tcpcb *tp;
  951 
  952         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
  953         INP_WLOCK_ASSERT(inp);
  954 
  955         if ((inp->inp_flags & INP_TIMEWAIT) ||
  956             (inp->inp_flags & INP_DROPPED))
  957                 return (inp);
  958 
  959         tp = intotcpcb(inp);
  960         KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
  961 
  962         /*
  963          * Ignore some errors if we are hooked up.
  964          * If connection hasn't completed, has retransmitted several times,
  965          * and receives a second error, give up now.  This is better
  966          * than waiting a long time to establish a connection that
  967          * can never complete.
  968          */
  969         if (tp->t_state == TCPS_ESTABLISHED &&
  970             (error == EHOSTUNREACH || error == ENETUNREACH ||
  971              error == EHOSTDOWN)) {
  972                 return (inp);
  973         } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
  974             tp->t_softerror) {
  975                 tp = tcp_drop(tp, error);
  976                 if (tp != NULL)
  977                         return (inp);
  978                 else
  979                         return (NULL);
  980         } else {
  981                 tp->t_softerror = error;
  982                 return (inp);
  983         }
  984 #if 0
  985         wakeup( &so->so_timeo);
  986         sorwakeup(so);
  987         sowwakeup(so);
  988 #endif
  989 }
  990 
  991 static int
  992 tcp_pcblist(SYSCTL_HANDLER_ARGS)
  993 {
  994         int error, i, m, n, pcb_count;
  995         struct inpcb *inp, **inp_list;
  996         inp_gen_t gencnt;
  997         struct xinpgen xig;
  998 
  999         /*
 1000          * The process of preparing the TCB list is too time-consuming and
 1001          * resource-intensive to repeat twice on every request.
 1002          */
 1003         if (req->oldptr == NULL) {
 1004                 n = V_tcbinfo.ipi_count + syncache_pcbcount();
 1005                 n += imax(n / 8, 10);
 1006                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
 1007                 return (0);
 1008         }
 1009 
 1010         if (req->newptr != NULL)
 1011                 return (EPERM);
 1012 
 1013         /*
 1014          * OK, now we're committed to doing something.
 1015          */
 1016         INP_INFO_RLOCK(&V_tcbinfo);
 1017         gencnt = V_tcbinfo.ipi_gencnt;
 1018         n = V_tcbinfo.ipi_count;
 1019         INP_INFO_RUNLOCK(&V_tcbinfo);
 1020 
 1021         m = syncache_pcbcount();
 1022 
 1023         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
 1024                 + (n + m) * sizeof(struct xtcpcb));
 1025         if (error != 0)
 1026                 return (error);
 1027 
 1028         xig.xig_len = sizeof xig;
 1029         xig.xig_count = n + m;
 1030         xig.xig_gen = gencnt;
 1031         xig.xig_sogen = so_gencnt;
 1032         error = SYSCTL_OUT(req, &xig, sizeof xig);
 1033         if (error)
 1034                 return (error);
 1035 
 1036         error = syncache_pcblist(req, m, &pcb_count);
 1037         if (error)
 1038                 return (error);
 1039 
 1040         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
 1041         if (inp_list == NULL)
 1042                 return (ENOMEM);
 1043 
 1044         INP_INFO_RLOCK(&V_tcbinfo);
 1045         for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
 1046             inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
 1047                 INP_WLOCK(inp);
 1048                 if (inp->inp_gencnt <= gencnt) {
 1049                         /*
 1050                          * XXX: This use of cr_cansee(), introduced with
 1051                          * TCP state changes, is not quite right, but for
 1052                          * now, better than nothing.
 1053                          */
 1054                         if (inp->inp_flags & INP_TIMEWAIT) {
 1055                                 if (intotw(inp) != NULL)
 1056                                         error = cr_cansee(req->td->td_ucred,
 1057                                             intotw(inp)->tw_cred);
 1058                                 else
 1059                                         error = EINVAL; /* Skip this inp. */
 1060                         } else
 1061                                 error = cr_canseeinpcb(req->td->td_ucred, inp);
 1062                         if (error == 0) {
 1063                                 in_pcbref(inp);
 1064                                 inp_list[i++] = inp;
 1065                         }
 1066                 }
 1067                 INP_WUNLOCK(inp);
 1068         }
 1069         INP_INFO_RUNLOCK(&V_tcbinfo);
 1070         n = i;
 1071 
 1072         error = 0;
 1073         for (i = 0; i < n; i++) {
 1074                 inp = inp_list[i];
 1075                 INP_RLOCK(inp);
 1076                 if (inp->inp_gencnt <= gencnt) {
 1077                         struct xtcpcb xt;
 1078                         void *inp_ppcb;
 1079 
 1080                         bzero(&xt, sizeof(xt));
 1081                         xt.xt_len = sizeof xt;
 1082                         /* XXX should avoid extra copy */
 1083                         bcopy(inp, &xt.xt_inp, sizeof *inp);
 1084                         inp_ppcb = inp->inp_ppcb;
 1085                         if (inp_ppcb == NULL)
 1086                                 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
 1087                         else if (inp->inp_flags & INP_TIMEWAIT) {
 1088                                 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
 1089                                 xt.xt_tp.t_state = TCPS_TIME_WAIT;
 1090                         } else
 1091                                 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
 1092                         if (inp->inp_socket != NULL)
 1093                                 sotoxsocket(inp->inp_socket, &xt.xt_socket);
 1094                         else {
 1095                                 bzero(&xt.xt_socket, sizeof xt.xt_socket);
 1096                                 xt.xt_socket.xso_protocol = IPPROTO_TCP;
 1097                         }
 1098                         xt.xt_inp.inp_gencnt = inp->inp_gencnt;
 1099                         INP_RUNLOCK(inp);
 1100                         error = SYSCTL_OUT(req, &xt, sizeof xt);
 1101                 } else
 1102                         INP_RUNLOCK(inp);
 1103         }
 1104         INP_INFO_WLOCK(&V_tcbinfo);
 1105         for (i = 0; i < n; i++) {
 1106                 inp = inp_list[i];
 1107                 INP_WLOCK(inp);
 1108                 if (!in_pcbrele(inp))
 1109                         INP_WUNLOCK(inp);
 1110         }
 1111         INP_INFO_WUNLOCK(&V_tcbinfo);
 1112 
 1113         if (!error) {
 1114                 /*
 1115                  * Give the user an updated idea of our state.
 1116                  * If the generation differs from what we told
 1117                  * her before, she knows that something happened
 1118                  * while we were processing this request, and it
 1119                  * might be necessary to retry.
 1120                  */
 1121                 INP_INFO_RLOCK(&V_tcbinfo);
 1122                 xig.xig_gen = V_tcbinfo.ipi_gencnt;
 1123                 xig.xig_sogen = so_gencnt;
 1124                 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
 1125                 INP_INFO_RUNLOCK(&V_tcbinfo);
 1126                 error = SYSCTL_OUT(req, &xig, sizeof xig);
 1127         }
 1128         free(inp_list, M_TEMP);
 1129         return (error);
 1130 }
 1131 
 1132 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
 1133     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
 1134 
 1135 static int
 1136 tcp_getcred(SYSCTL_HANDLER_ARGS)
 1137 {
 1138         struct xucred xuc;
 1139         struct sockaddr_in addrs[2];
 1140         struct inpcb *inp;
 1141         int error;
 1142 
 1143         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 1144         if (error)
 1145                 return (error);
 1146         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 1147         if (error)
 1148                 return (error);
 1149         INP_INFO_RLOCK(&V_tcbinfo);
 1150         inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
 1151             addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
 1152         if (inp != NULL) {
 1153                 INP_RLOCK(inp);
 1154                 INP_INFO_RUNLOCK(&V_tcbinfo);
 1155                 if (inp->inp_socket == NULL)
 1156                         error = ENOENT;
 1157                 if (error == 0)
 1158                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 1159                 if (error == 0)
 1160                         cru2x(inp->inp_cred, &xuc);
 1161                 INP_RUNLOCK(inp);
 1162         } else {
 1163                 INP_INFO_RUNLOCK(&V_tcbinfo);
 1164                 error = ENOENT;
 1165         }
 1166         if (error == 0)
 1167                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 1168         return (error);
 1169 }
 1170 
 1171 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
 1172     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
 1173     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
 1174 
 1175 #ifdef INET6
 1176 static int
 1177 tcp6_getcred(SYSCTL_HANDLER_ARGS)
 1178 {
 1179         struct xucred xuc;
 1180         struct sockaddr_in6 addrs[2];
 1181         struct inpcb *inp;
 1182         int error, mapped = 0;
 1183 
 1184         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 1185         if (error)
 1186                 return (error);
 1187         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 1188         if (error)
 1189                 return (error);
 1190         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
 1191             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
 1192                 return (error);
 1193         }
 1194         if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
 1195                 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
 1196                         mapped = 1;
 1197                 else
 1198                         return (EINVAL);
 1199         }
 1200 
 1201         INP_INFO_RLOCK(&V_tcbinfo);
 1202         if (mapped == 1)
 1203                 inp = in_pcblookup_hash(&V_tcbinfo,
 1204                         *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
 1205                         addrs[1].sin6_port,
 1206                         *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
 1207                         addrs[0].sin6_port,
 1208                         0, NULL);
 1209         else
 1210                 inp = in6_pcblookup_hash(&V_tcbinfo,
 1211                         &addrs[1].sin6_addr, addrs[1].sin6_port,
 1212                         &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
 1213         if (inp != NULL) {
 1214                 INP_RLOCK(inp);
 1215                 INP_INFO_RUNLOCK(&V_tcbinfo);
 1216                 if (inp->inp_socket == NULL)
 1217                         error = ENOENT;
 1218                 if (error == 0)
 1219                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 1220                 if (error == 0)
 1221                         cru2x(inp->inp_cred, &xuc);
 1222                 INP_RUNLOCK(inp);
 1223         } else {
 1224                 INP_INFO_RUNLOCK(&V_tcbinfo);
 1225                 error = ENOENT;
 1226         }
 1227         if (error == 0)
 1228                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 1229         return (error);
 1230 }
 1231 
 1232 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
 1233     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
 1234     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
 1235 #endif
 1236 
 1237 
 1238 void
 1239 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 1240 {
 1241         struct ip *ip = vip;
 1242         struct tcphdr *th;
 1243         struct in_addr faddr;
 1244         struct inpcb *inp;
 1245         struct tcpcb *tp;
 1246         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
 1247         struct icmp *icp;
 1248         struct in_conninfo inc;
 1249         tcp_seq icmp_tcp_seq;
 1250         int mtu;
 1251 
 1252         faddr = ((struct sockaddr_in *)sa)->sin_addr;
 1253         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
 1254                 return;
 1255 
 1256         if (cmd == PRC_MSGSIZE)
 1257                 notify = tcp_mtudisc;
 1258         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
 1259                 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
 1260                 notify = tcp_drop_syn_sent;
 1261         /*
 1262          * Redirects don't need to be handled up here.
 1263          */
 1264         else if (PRC_IS_REDIRECT(cmd))
 1265                 return;
 1266         /*
 1267          * Source quench is depreciated.
 1268          */
 1269         else if (cmd == PRC_QUENCH)
 1270                 return;
 1271         /*
 1272          * Hostdead is ugly because it goes linearly through all PCBs.
 1273          * XXX: We never get this from ICMP, otherwise it makes an
 1274          * excellent DoS attack on machines with many connections.
 1275          */
 1276         else if (cmd == PRC_HOSTDEAD)
 1277                 ip = NULL;
 1278         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
 1279                 return;
 1280         if (ip != NULL) {
 1281                 icp = (struct icmp *)((caddr_t)ip
 1282                                       - offsetof(struct icmp, icmp_ip));
 1283                 th = (struct tcphdr *)((caddr_t)ip
 1284                                        + (ip->ip_hl << 2));
 1285                 INP_INFO_WLOCK(&V_tcbinfo);
 1286                 inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
 1287                     ip->ip_src, th->th_sport, 0, NULL);
 1288                 if (inp != NULL)  {
 1289                         INP_WLOCK(inp);
 1290                         if (!(inp->inp_flags & INP_TIMEWAIT) &&
 1291                             !(inp->inp_flags & INP_DROPPED) &&
 1292                             !(inp->inp_socket == NULL)) {
 1293                                 icmp_tcp_seq = htonl(th->th_seq);
 1294                                 tp = intotcpcb(inp);
 1295                                 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
 1296                                     SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
 1297                                         if (cmd == PRC_MSGSIZE) {
 1298                                             /*
 1299                                              * MTU discovery:
 1300                                              * If we got a needfrag set the MTU
 1301                                              * in the route to the suggested new
 1302                                              * value (if given) and then notify.
 1303                                              */
 1304                                             bzero(&inc, sizeof(inc));
 1305                                             inc.inc_faddr = faddr;
 1306                                             inc.inc_fibnum =
 1307                                                 inp->inp_inc.inc_fibnum;
 1308 
 1309                                             mtu = ntohs(icp->icmp_nextmtu);
 1310                                             /*
 1311                                              * If no alternative MTU was
 1312                                              * proposed, try the next smaller
 1313                                              * one.  ip->ip_len has already
 1314                                              * been swapped in icmp_input().
 1315                                              */
 1316                                             if (!mtu)
 1317                                                 mtu = ip_next_mtu(ip->ip_len,
 1318                                                  1);
 1319                                             if (mtu < V_tcp_minmss
 1320                                                  + sizeof(struct tcpiphdr))
 1321                                                 mtu = V_tcp_minmss
 1322                                                  + sizeof(struct tcpiphdr);
 1323                                             /*
 1324                                              * Only cache the the MTU if it
 1325                                              * is smaller than the interface
 1326                                              * or route MTU.  tcp_mtudisc()
 1327                                              * will do right thing by itself.
 1328                                              */
 1329                                             if (mtu <= tcp_maxmtu(&inc, NULL))
 1330                                                 tcp_hc_updatemtu(&inc, mtu);
 1331                                         }
 1332 
 1333                                         inp = (*notify)(inp, inetctlerrmap[cmd]);
 1334                                 }
 1335                         }
 1336                         if (inp != NULL)
 1337                                 INP_WUNLOCK(inp);
 1338                 } else {
 1339                         bzero(&inc, sizeof(inc));
 1340                         inc.inc_fport = th->th_dport;
 1341                         inc.inc_lport = th->th_sport;
 1342                         inc.inc_faddr = faddr;
 1343                         inc.inc_laddr = ip->ip_src;
 1344                         syncache_unreach(&inc, th);
 1345                 }
 1346                 INP_INFO_WUNLOCK(&V_tcbinfo);
 1347         } else
 1348                 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
 1349 }
 1350 
 1351 #ifdef INET6
 1352 void
 1353 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 1354 {
 1355         struct tcphdr th;
 1356         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
 1357         struct ip6_hdr *ip6;
 1358         struct mbuf *m;
 1359         struct ip6ctlparam *ip6cp = NULL;
 1360         const struct sockaddr_in6 *sa6_src = NULL;
 1361         int off;
 1362         struct tcp_portonly {
 1363                 u_int16_t th_sport;
 1364                 u_int16_t th_dport;
 1365         } *thp;
 1366 
 1367         if (sa->sa_family != AF_INET6 ||
 1368             sa->sa_len != sizeof(struct sockaddr_in6))
 1369                 return;
 1370 
 1371         if (cmd == PRC_MSGSIZE)
 1372                 notify = tcp_mtudisc;
 1373         else if (!PRC_IS_REDIRECT(cmd) &&
 1374                  ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
 1375                 return;
 1376         /* Source quench is depreciated. */
 1377         else if (cmd == PRC_QUENCH)
 1378                 return;
 1379 
 1380         /* if the parameter is from icmp6, decode it. */
 1381         if (d != NULL) {
 1382                 ip6cp = (struct ip6ctlparam *)d;
 1383                 m = ip6cp->ip6c_m;
 1384                 ip6 = ip6cp->ip6c_ip6;
 1385                 off = ip6cp->ip6c_off;
 1386                 sa6_src = ip6cp->ip6c_src;
 1387         } else {
 1388                 m = NULL;
 1389                 ip6 = NULL;
 1390                 off = 0;        /* fool gcc */
 1391                 sa6_src = &sa6_any;
 1392         }
 1393 
 1394         if (ip6 != NULL) {
 1395                 struct in_conninfo inc;
 1396                 /*
 1397                  * XXX: We assume that when IPV6 is non NULL,
 1398                  * M and OFF are valid.
 1399                  */
 1400 
 1401                 /* check if we can safely examine src and dst ports */
 1402                 if (m->m_pkthdr.len < off + sizeof(*thp))
 1403                         return;
 1404 
 1405                 bzero(&th, sizeof(th));
 1406                 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
 1407 
 1408                 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
 1409                     (struct sockaddr *)ip6cp->ip6c_src,
 1410                     th.th_sport, cmd, NULL, notify);
 1411 
 1412                 bzero(&inc, sizeof(inc));
 1413                 inc.inc_fport = th.th_dport;
 1414                 inc.inc_lport = th.th_sport;
 1415                 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
 1416                 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
 1417                 inc.inc_flags |= INC_ISIPV6;
 1418                 INP_INFO_WLOCK(&V_tcbinfo);
 1419                 syncache_unreach(&inc, &th);
 1420                 INP_INFO_WUNLOCK(&V_tcbinfo);
 1421         } else
 1422                 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
 1423                               0, cmd, NULL, notify);
 1424 }
 1425 #endif /* INET6 */
 1426 
 1427 
 1428 /*
 1429  * Following is where TCP initial sequence number generation occurs.
 1430  *
 1431  * There are two places where we must use initial sequence numbers:
 1432  * 1.  In SYN-ACK packets.
 1433  * 2.  In SYN packets.
 1434  *
 1435  * All ISNs for SYN-ACK packets are generated by the syncache.  See
 1436  * tcp_syncache.c for details.
 1437  *
 1438  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
 1439  * depends on this property.  In addition, these ISNs should be
 1440  * unguessable so as to prevent connection hijacking.  To satisfy
 1441  * the requirements of this situation, the algorithm outlined in
 1442  * RFC 1948 is used, with only small modifications.
 1443  *
 1444  * Implementation details:
 1445  *
 1446  * Time is based off the system timer, and is corrected so that it
 1447  * increases by one megabyte per second.  This allows for proper
 1448  * recycling on high speed LANs while still leaving over an hour
 1449  * before rollover.
 1450  *
 1451  * As reading the *exact* system time is too expensive to be done
 1452  * whenever setting up a TCP connection, we increment the time
 1453  * offset in two ways.  First, a small random positive increment
 1454  * is added to isn_offset for each connection that is set up.
 1455  * Second, the function tcp_isn_tick fires once per clock tick
 1456  * and increments isn_offset as necessary so that sequence numbers
 1457  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
 1458  * random positive increments serve only to ensure that the same
 1459  * exact sequence number is never sent out twice (as could otherwise
 1460  * happen when a port is recycled in less than the system tick
 1461  * interval.)
 1462  *
 1463  * net.inet.tcp.isn_reseed_interval controls the number of seconds
 1464  * between seeding of isn_secret.  This is normally set to zero,
 1465  * as reseeding should not be necessary.
 1466  *
 1467  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
 1468  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
 1469  * general, this means holding an exclusive (write) lock.
 1470  */
 1471 
 1472 #define ISN_BYTES_PER_SECOND 1048576
 1473 #define ISN_STATIC_INCREMENT 4096
 1474 #define ISN_RANDOM_INCREMENT (4096 - 1)
 1475 
 1476 static VNET_DEFINE(u_char, isn_secret[32]);
 1477 static VNET_DEFINE(int, isn_last_reseed);
 1478 static VNET_DEFINE(u_int32_t, isn_offset);
 1479 static VNET_DEFINE(u_int32_t, isn_offset_old);
 1480 
 1481 #define V_isn_secret                    VNET(isn_secret)
 1482 #define V_isn_last_reseed               VNET(isn_last_reseed)
 1483 #define V_isn_offset                    VNET(isn_offset)
 1484 #define V_isn_offset_old                VNET(isn_offset_old)
 1485 
 1486 tcp_seq
 1487 tcp_new_isn(struct tcpcb *tp)
 1488 {
 1489         MD5_CTX isn_ctx;
 1490         u_int32_t md5_buffer[4];
 1491         tcp_seq new_isn;
 1492 
 1493         INP_WLOCK_ASSERT(tp->t_inpcb);
 1494 
 1495         ISN_LOCK();
 1496         /* Seed if this is the first use, reseed if requested. */
 1497         if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
 1498              (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
 1499                 < (u_int)ticks))) {
 1500                 read_random(&V_isn_secret, sizeof(V_isn_secret));
 1501                 V_isn_last_reseed = ticks;
 1502         }
 1503 
 1504         /* Compute the md5 hash and return the ISN. */
 1505         MD5Init(&isn_ctx);
 1506         MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
 1507         MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
 1508 #ifdef INET6
 1509         if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
 1510                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
 1511                           sizeof(struct in6_addr));
 1512                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
 1513                           sizeof(struct in6_addr));
 1514         } else
 1515 #endif
 1516         {
 1517                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
 1518                           sizeof(struct in_addr));
 1519                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
 1520                           sizeof(struct in_addr));
 1521         }
 1522         MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
 1523         MD5Final((u_char *) &md5_buffer, &isn_ctx);
 1524         new_isn = (tcp_seq) md5_buffer[0];
 1525         V_isn_offset += ISN_STATIC_INCREMENT +
 1526                 (arc4random() & ISN_RANDOM_INCREMENT);
 1527         new_isn += V_isn_offset;
 1528         ISN_UNLOCK();
 1529         return (new_isn);
 1530 }
 1531 
 1532 /*
 1533  * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
 1534  * to keep time flowing at a relatively constant rate.  If the random
 1535  * increments have already pushed us past the projected offset, do nothing.
 1536  */
 1537 static void
 1538 tcp_isn_tick(void *xtp)
 1539 {
 1540         VNET_ITERATOR_DECL(vnet_iter);
 1541         u_int32_t projected_offset;
 1542 
 1543         VNET_LIST_RLOCK_NOSLEEP();
 1544         ISN_LOCK();
 1545         VNET_FOREACH(vnet_iter) {
 1546                 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
 1547                 projected_offset =
 1548                     V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
 1549 
 1550                 if (SEQ_GT(projected_offset, V_isn_offset))
 1551                         V_isn_offset = projected_offset;
 1552 
 1553                 V_isn_offset_old = V_isn_offset;
 1554                 CURVNET_RESTORE();
 1555         }
 1556         ISN_UNLOCK();
 1557         VNET_LIST_RUNLOCK_NOSLEEP();
 1558         callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
 1559 }
 1560 
 1561 /*
 1562  * When a specific ICMP unreachable message is received and the
 1563  * connection state is SYN-SENT, drop the connection.  This behavior
 1564  * is controlled by the icmp_may_rst sysctl.
 1565  */
 1566 struct inpcb *
 1567 tcp_drop_syn_sent(struct inpcb *inp, int errno)
 1568 {
 1569         struct tcpcb *tp;
 1570 
 1571         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 1572         INP_WLOCK_ASSERT(inp);
 1573 
 1574         if ((inp->inp_flags & INP_TIMEWAIT) ||
 1575             (inp->inp_flags & INP_DROPPED))
 1576                 return (inp);
 1577 
 1578         tp = intotcpcb(inp);
 1579         if (tp->t_state != TCPS_SYN_SENT)
 1580                 return (inp);
 1581 
 1582         tp = tcp_drop(tp, errno);
 1583         if (tp != NULL)
 1584                 return (inp);
 1585         else
 1586                 return (NULL);
 1587 }
 1588 
 1589 /*
 1590  * When `need fragmentation' ICMP is received, update our idea of the MSS
 1591  * based on the new value in the route.  Also nudge TCP to send something,
 1592  * since we know the packet we just sent was dropped.
 1593  * This duplicates some code in the tcp_mss() function in tcp_input.c.
 1594  */
 1595 struct inpcb *
 1596 tcp_mtudisc(struct inpcb *inp, int errno)
 1597 {
 1598         struct tcpcb *tp;
 1599         struct socket *so;
 1600 
 1601         INP_WLOCK_ASSERT(inp);
 1602         if ((inp->inp_flags & INP_TIMEWAIT) ||
 1603             (inp->inp_flags & INP_DROPPED))
 1604                 return (inp);
 1605 
 1606         tp = intotcpcb(inp);
 1607         KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
 1608 
 1609         tcp_mss_update(tp, -1, NULL, NULL);
 1610   
 1611         so = inp->inp_socket;
 1612         SOCKBUF_LOCK(&so->so_snd);
 1613         /* If the mss is larger than the socket buffer, decrease the mss. */
 1614         if (so->so_snd.sb_hiwat < tp->t_maxseg)
 1615                 tp->t_maxseg = so->so_snd.sb_hiwat;
 1616         SOCKBUF_UNLOCK(&so->so_snd);
 1617 
 1618         TCPSTAT_INC(tcps_mturesent);
 1619         tp->t_rtttime = 0;
 1620         tp->snd_nxt = tp->snd_una;
 1621         tcp_free_sackholes(tp);
 1622         tp->snd_recover = tp->snd_max;
 1623         if (tp->t_flags & TF_SACK_PERMIT)
 1624                 EXIT_FASTRECOVERY(tp);
 1625         tcp_output_send(tp);
 1626         return (inp);
 1627 }
 1628 
 1629 /*
 1630  * Look-up the routing entry to the peer of this inpcb.  If no route
 1631  * is found and it cannot be allocated, then return 0.  This routine
 1632  * is called by TCP routines that access the rmx structure and by
 1633  * tcp_mss_update to get the peer/interface MTU.
 1634  */
 1635 u_long
 1636 tcp_maxmtu(struct in_conninfo *inc, int *flags)
 1637 {
 1638         struct route sro;
 1639         struct sockaddr_in *dst;
 1640         struct ifnet *ifp;
 1641         u_long maxmtu = 0;
 1642 
 1643         KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
 1644 
 1645         bzero(&sro, sizeof(sro));
 1646         if (inc->inc_faddr.s_addr != INADDR_ANY) {
 1647                 dst = (struct sockaddr_in *)&sro.ro_dst;
 1648                 dst->sin_family = AF_INET;
 1649                 dst->sin_len = sizeof(*dst);
 1650                 dst->sin_addr = inc->inc_faddr;
 1651                 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
 1652         }
 1653         if (sro.ro_rt != NULL) {
 1654                 ifp = sro.ro_rt->rt_ifp;
 1655                 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
 1656                         maxmtu = ifp->if_mtu;
 1657                 else
 1658                         maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
 1659 
 1660                 /* Report additional interface capabilities. */
 1661                 if (flags != NULL) {
 1662                         if (ifp->if_capenable & IFCAP_TSO4 &&
 1663                             ifp->if_hwassist & CSUM_TSO)
 1664                                 *flags |= CSUM_TSO;
 1665                 }
 1666                 RTFREE(sro.ro_rt);
 1667         }
 1668         return (maxmtu);
 1669 }
 1670 
 1671 #ifdef INET6
 1672 u_long
 1673 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
 1674 {
 1675         struct route_in6 sro6;
 1676         struct ifnet *ifp;
 1677         u_long maxmtu = 0;
 1678 
 1679         KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
 1680 
 1681         bzero(&sro6, sizeof(sro6));
 1682         if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
 1683                 sro6.ro_dst.sin6_family = AF_INET6;
 1684                 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
 1685                 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
 1686                 rtalloc_ign((struct route *)&sro6, 0);
 1687         }
 1688         if (sro6.ro_rt != NULL) {
 1689                 ifp = sro6.ro_rt->rt_ifp;
 1690                 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
 1691                         maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
 1692                 else
 1693                         maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
 1694                                      IN6_LINKMTU(sro6.ro_rt->rt_ifp));
 1695 
 1696                 /* Report additional interface capabilities. */
 1697                 if (flags != NULL) {
 1698                         if (ifp->if_capenable & IFCAP_TSO6 &&
 1699                             ifp->if_hwassist & CSUM_TSO)
 1700                                 *flags |= CSUM_TSO;
 1701                 }
 1702                 RTFREE(sro6.ro_rt);
 1703         }
 1704 
 1705         return (maxmtu);
 1706 }
 1707 #endif /* INET6 */
 1708 
 1709 #ifdef IPSEC
 1710 /* compute ESP/AH header size for TCP, including outer IP header. */
 1711 size_t
 1712 ipsec_hdrsiz_tcp(struct tcpcb *tp)
 1713 {
 1714         struct inpcb *inp;
 1715         struct mbuf *m;
 1716         size_t hdrsiz;
 1717         struct ip *ip;
 1718 #ifdef INET6
 1719         struct ip6_hdr *ip6;
 1720 #endif
 1721         struct tcphdr *th;
 1722 
 1723         if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
 1724                 return (0);
 1725         MGETHDR(m, M_DONTWAIT, MT_DATA);
 1726         if (!m)
 1727                 return (0);
 1728 
 1729 #ifdef INET6
 1730         if ((inp->inp_vflag & INP_IPV6) != 0) {
 1731                 ip6 = mtod(m, struct ip6_hdr *);
 1732                 th = (struct tcphdr *)(ip6 + 1);
 1733                 m->m_pkthdr.len = m->m_len =
 1734                         sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
 1735                 tcpip_fillheaders(inp, ip6, th);
 1736                 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
 1737         } else
 1738 #endif /* INET6 */
 1739         {
 1740                 ip = mtod(m, struct ip *);
 1741                 th = (struct tcphdr *)(ip + 1);
 1742                 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
 1743                 tcpip_fillheaders(inp, ip, th);
 1744                 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
 1745         }
 1746 
 1747         m_free(m);
 1748         return (hdrsiz);
 1749 }
 1750 #endif /* IPSEC */
 1751 
 1752 /*
 1753  * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
 1754  *
 1755  * This code attempts to calculate the bandwidth-delay product as a
 1756  * means of determining the optimal window size to maximize bandwidth,
 1757  * minimize RTT, and avoid the over-allocation of buffers on interfaces and
 1758  * routers.  This code also does a fairly good job keeping RTTs in check
 1759  * across slow links like modems.  We implement an algorithm which is very
 1760  * similar (but not meant to be) TCP/Vegas.  The code operates on the
 1761  * transmitter side of a TCP connection and so only effects the transmit
 1762  * side of the connection.
 1763  *
 1764  * BACKGROUND:  TCP makes no provision for the management of buffer space
 1765  * at the end points or at the intermediate routers and switches.  A TCP
 1766  * stream, whether using NewReno or not, will eventually buffer as
 1767  * many packets as it is able and the only reason this typically works is
 1768  * due to the fairly small default buffers made available for a connection
 1769  * (typicaly 16K or 32K).  As machines use larger windows and/or window
 1770  * scaling it is now fairly easy for even a single TCP connection to blow-out
 1771  * all available buffer space not only on the local interface, but on
 1772  * intermediate routers and switches as well.  NewReno makes a misguided
 1773  * attempt to 'solve' this problem by waiting for an actual failure to occur,
 1774  * then backing off, then steadily increasing the window again until another
 1775  * failure occurs, ad-infinitum.  This results in terrible oscillation that
 1776  * is only made worse as network loads increase and the idea of intentionally
 1777  * blowing out network buffers is, frankly, a terrible way to manage network
 1778  * resources.
 1779  *
 1780  * It is far better to limit the transmit window prior to the failure
 1781  * condition being achieved.  There are two general ways to do this:  First
 1782  * you can 'scan' through different transmit window sizes and locate the
 1783  * point where the RTT stops increasing, indicating that you have filled the
 1784  * pipe, then scan backwards until you note that RTT stops decreasing, then
 1785  * repeat ad-infinitum.  This method works in principle but has severe
 1786  * implementation issues due to RTT variances, timer granularity, and
 1787  * instability in the algorithm which can lead to many false positives and
 1788  * create oscillations as well as interact badly with other TCP streams
 1789  * implementing the same algorithm.
 1790  *
 1791  * The second method is to limit the window to the bandwidth delay product
 1792  * of the link.  This is the method we implement.  RTT variances and our
 1793  * own manipulation of the congestion window, bwnd, can potentially
 1794  * destabilize the algorithm.  For this reason we have to stabilize the
 1795  * elements used to calculate the window.  We do this by using the minimum
 1796  * observed RTT, the long term average of the observed bandwidth, and
 1797  * by adding two segments worth of slop.  It isn't perfect but it is able
 1798  * to react to changing conditions and gives us a very stable basis on
 1799  * which to extend the algorithm.
 1800  */
 1801 void
 1802 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
 1803 {
 1804         u_long bw;
 1805         u_long bwnd;
 1806         int save_ticks;
 1807 
 1808         INP_WLOCK_ASSERT(tp->t_inpcb);
 1809 
 1810         /*
 1811          * If inflight_enable is disabled in the middle of a tcp connection,
 1812          * make sure snd_bwnd is effectively disabled.
 1813          */
 1814         if (V_tcp_inflight_enable == 0 ||
 1815             tp->t_rttlow < V_tcp_inflight_rttthresh) {
 1816                 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
 1817                 tp->snd_bandwidth = 0;
 1818                 return;
 1819         }
 1820 
 1821         /*
 1822          * Figure out the bandwidth.  Due to the tick granularity this
 1823          * is a very rough number and it MUST be averaged over a fairly
 1824          * long period of time.  XXX we need to take into account a link
 1825          * that is not using all available bandwidth, but for now our
 1826          * slop will ramp us up if this case occurs and the bandwidth later
 1827          * increases.
 1828          *
 1829          * Note: if ticks rollover 'bw' may wind up negative.  We must
 1830          * effectively reset t_bw_rtttime for this case.
 1831          */
 1832         save_ticks = ticks;
 1833         if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
 1834                 return;
 1835 
 1836         bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
 1837             (save_ticks - tp->t_bw_rtttime);
 1838         tp->t_bw_rtttime = save_ticks;
 1839         tp->t_bw_rtseq = ack_seq;
 1840         if (tp->t_bw_rtttime == 0 || (int)bw < 0)
 1841                 return;
 1842         bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
 1843 
 1844         tp->snd_bandwidth = bw;
 1845 
 1846         /*
 1847          * Calculate the semi-static bandwidth delay product, plus two maximal
 1848          * segments.  The additional slop puts us squarely in the sweet
 1849          * spot and also handles the bandwidth run-up case and stabilization.
 1850          * Without the slop we could be locking ourselves into a lower
 1851          * bandwidth.
 1852          *
 1853          * Situations Handled:
 1854          *      (1) Prevents over-queueing of packets on LANs, especially on
 1855          *          high speed LANs, allowing larger TCP buffers to be
 1856          *          specified, and also does a good job preventing
 1857          *          over-queueing of packets over choke points like modems
 1858          *          (at least for the transmit side).
 1859          *
 1860          *      (2) Is able to handle changing network loads (bandwidth
 1861          *          drops so bwnd drops, bandwidth increases so bwnd
 1862          *          increases).
 1863          *
 1864          *      (3) Theoretically should stabilize in the face of multiple
 1865          *          connections implementing the same algorithm (this may need
 1866          *          a little work).
 1867          *
 1868          *      (4) Stability value (defaults to 20 = 2 maximal packets) can
 1869          *          be adjusted with a sysctl but typically only needs to be
 1870          *          on very slow connections.  A value no smaller then 5
 1871          *          should be used, but only reduce this default if you have
 1872          *          no other choice.
 1873          */
 1874 #define USERTT  ((tp->t_srtt + tp->t_rttbest) / 2)
 1875         bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
 1876 #undef USERTT
 1877 
 1878         if (tcp_inflight_debug > 0) {
 1879                 static int ltime;
 1880                 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
 1881                         ltime = ticks;
 1882                         printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
 1883                             tp,
 1884                             bw,
 1885                             tp->t_rttbest,
 1886                             tp->t_srtt,
 1887                             bwnd
 1888                         );
 1889                 }
 1890         }
 1891         if ((long)bwnd < V_tcp_inflight_min)
 1892                 bwnd = V_tcp_inflight_min;
 1893         if (bwnd > V_tcp_inflight_max)
 1894                 bwnd = V_tcp_inflight_max;
 1895         if ((long)bwnd < tp->t_maxseg * 2)
 1896                 bwnd = tp->t_maxseg * 2;
 1897         tp->snd_bwnd = bwnd;
 1898 }
 1899 
 1900 #ifdef TCP_SIGNATURE
 1901 /*
 1902  * Callback function invoked by m_apply() to digest TCP segment data
 1903  * contained within an mbuf chain.
 1904  */
 1905 static int
 1906 tcp_signature_apply(void *fstate, void *data, u_int len)
 1907 {
 1908 
 1909         MD5Update(fstate, (u_char *)data, len);
 1910         return (0);
 1911 }
 1912 
 1913 /*
 1914  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
 1915  *
 1916  * Parameters:
 1917  * m            pointer to head of mbuf chain
 1918  * _unused      
 1919  * len          length of TCP segment data, excluding options
 1920  * optlen       length of TCP segment options
 1921  * buf          pointer to storage for computed MD5 digest
 1922  * direction    direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
 1923  *
 1924  * We do this over ip, tcphdr, segment data, and the key in the SADB.
 1925  * When called from tcp_input(), we can be sure that th_sum has been
 1926  * zeroed out and verified already.
 1927  *
 1928  * Return 0 if successful, otherwise return -1.
 1929  *
 1930  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
 1931  * search with the destination IP address, and a 'magic SPI' to be
 1932  * determined by the application. This is hardcoded elsewhere to 1179
 1933  * right now. Another branch of this code exists which uses the SPD to
 1934  * specify per-application flows but it is unstable.
 1935  */
 1936 int
 1937 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
 1938     u_char *buf, u_int direction)
 1939 {
 1940         union sockaddr_union dst;
 1941         struct ippseudo ippseudo;
 1942         MD5_CTX ctx;
 1943         int doff;
 1944         struct ip *ip;
 1945         struct ipovly *ipovly;
 1946         struct secasvar *sav;
 1947         struct tcphdr *th;
 1948 #ifdef INET6
 1949         struct ip6_hdr *ip6;
 1950         struct in6_addr in6;
 1951         char ip6buf[INET6_ADDRSTRLEN];
 1952         uint32_t plen;
 1953         uint16_t nhdr;
 1954 #endif
 1955         u_short savecsum;
 1956 
 1957         KASSERT(m != NULL, ("NULL mbuf chain"));
 1958         KASSERT(buf != NULL, ("NULL signature pointer"));
 1959 
 1960         /* Extract the destination from the IP header in the mbuf. */
 1961         bzero(&dst, sizeof(union sockaddr_union));
 1962         ip = mtod(m, struct ip *);
 1963 #ifdef INET6
 1964         ip6 = NULL;     /* Make the compiler happy. */
 1965 #endif
 1966         switch (ip->ip_v) {
 1967         case IPVERSION:
 1968                 dst.sa.sa_len = sizeof(struct sockaddr_in);
 1969                 dst.sa.sa_family = AF_INET;
 1970                 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
 1971                     ip->ip_src : ip->ip_dst;
 1972                 break;
 1973 #ifdef INET6
 1974         case (IPV6_VERSION >> 4):
 1975                 ip6 = mtod(m, struct ip6_hdr *);
 1976                 dst.sa.sa_len = sizeof(struct sockaddr_in6);
 1977                 dst.sa.sa_family = AF_INET6;
 1978                 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
 1979                     ip6->ip6_src : ip6->ip6_dst;
 1980                 break;
 1981 #endif
 1982         default:
 1983                 return (EINVAL);
 1984                 /* NOTREACHED */
 1985                 break;
 1986         }
 1987 
 1988         /* Look up an SADB entry which matches the address of the peer. */
 1989         sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
 1990         if (sav == NULL) {
 1991                 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
 1992                     (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
 1993 #ifdef INET6
 1994                         (ip->ip_v == (IPV6_VERSION >> 4)) ?
 1995                             ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
 1996 #endif
 1997                         "(unsupported)"));
 1998                 return (EINVAL);
 1999         }
 2000 
 2001         MD5Init(&ctx);
 2002         /*
 2003          * Step 1: Update MD5 hash with IP(v6) pseudo-header.
 2004          *
 2005          * XXX The ippseudo header MUST be digested in network byte order,
 2006          * or else we'll fail the regression test. Assume all fields we've
 2007          * been doing arithmetic on have been in host byte order.
 2008          * XXX One cannot depend on ipovly->ih_len here. When called from
 2009          * tcp_output(), the underlying ip_len member has not yet been set.
 2010          */
 2011         switch (ip->ip_v) {
 2012         case IPVERSION:
 2013                 ipovly = (struct ipovly *)ip;
 2014                 ippseudo.ippseudo_src = ipovly->ih_src;
 2015                 ippseudo.ippseudo_dst = ipovly->ih_dst;
 2016                 ippseudo.ippseudo_pad = 0;
 2017                 ippseudo.ippseudo_p = IPPROTO_TCP;
 2018                 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
 2019                     optlen);
 2020                 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
 2021 
 2022                 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
 2023                 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
 2024                 break;
 2025 #ifdef INET6
 2026         /*
 2027          * RFC 2385, 2.0  Proposal
 2028          * For IPv6, the pseudo-header is as described in RFC 2460, namely the
 2029          * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
 2030          * extended next header value (to form 32 bits), and 32-bit segment
 2031          * length.
 2032          * Note: Upper-Layer Packet Length comes before Next Header.
 2033          */
 2034         case (IPV6_VERSION >> 4):
 2035                 in6 = ip6->ip6_src;
 2036                 in6_clearscope(&in6);
 2037                 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
 2038                 in6 = ip6->ip6_dst;
 2039                 in6_clearscope(&in6);
 2040                 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
 2041                 plen = htonl(len + sizeof(struct tcphdr) + optlen);
 2042                 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
 2043                 nhdr = 0;
 2044                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
 2045                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
 2046                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
 2047                 nhdr = IPPROTO_TCP;
 2048                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
 2049 
 2050                 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
 2051                 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
 2052                 break;
 2053 #endif
 2054         default:
 2055                 return (EINVAL);
 2056                 /* NOTREACHED */
 2057                 break;
 2058         }
 2059 
 2060 
 2061         /*
 2062          * Step 2: Update MD5 hash with TCP header, excluding options.
 2063          * The TCP checksum must be set to zero.
 2064          */
 2065         savecsum = th->th_sum;
 2066         th->th_sum = 0;
 2067         MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
 2068         th->th_sum = savecsum;
 2069 
 2070         /*
 2071          * Step 3: Update MD5 hash with TCP segment data.
 2072          *         Use m_apply() to avoid an early m_pullup().
 2073          */
 2074         if (len > 0)
 2075                 m_apply(m, doff, len, tcp_signature_apply, &ctx);
 2076 
 2077         /*
 2078          * Step 4: Update MD5 hash with shared secret.
 2079          */
 2080         MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
 2081         MD5Final(buf, &ctx);
 2082 
 2083         key_sa_recordxfer(sav, m);
 2084         KEY_FREESAV(&sav);
 2085         return (0);
 2086 }
 2087 #endif /* TCP_SIGNATURE */
 2088 
 2089 static int
 2090 sysctl_drop(SYSCTL_HANDLER_ARGS)
 2091 {
 2092         /* addrs[0] is a foreign socket, addrs[1] is a local one. */
 2093         struct sockaddr_storage addrs[2];
 2094         struct inpcb *inp;
 2095         struct tcpcb *tp;
 2096         struct tcptw *tw;
 2097         struct sockaddr_in *fin, *lin;
 2098 #ifdef INET6
 2099         struct sockaddr_in6 *fin6, *lin6;
 2100 #endif
 2101         int error;
 2102 
 2103         inp = NULL;
 2104         fin = lin = NULL;
 2105 #ifdef INET6
 2106         fin6 = lin6 = NULL;
 2107 #endif
 2108         error = 0;
 2109 
 2110         if (req->oldptr != NULL || req->oldlen != 0)
 2111                 return (EINVAL);
 2112         if (req->newptr == NULL)
 2113                 return (EPERM);
 2114         if (req->newlen < sizeof(addrs))
 2115                 return (ENOMEM);
 2116         error = SYSCTL_IN(req, &addrs, sizeof(addrs));
 2117         if (error)
 2118                 return (error);
 2119 
 2120         switch (addrs[0].ss_family) {
 2121 #ifdef INET6
 2122         case AF_INET6:
 2123                 fin6 = (struct sockaddr_in6 *)&addrs[0];
 2124                 lin6 = (struct sockaddr_in6 *)&addrs[1];
 2125                 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
 2126                     lin6->sin6_len != sizeof(struct sockaddr_in6))
 2127                         return (EINVAL);
 2128                 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
 2129                         if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
 2130                                 return (EINVAL);
 2131                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
 2132                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
 2133                         fin = (struct sockaddr_in *)&addrs[0];
 2134                         lin = (struct sockaddr_in *)&addrs[1];
 2135                         break;
 2136                 }
 2137                 error = sa6_embedscope(fin6, V_ip6_use_defzone);
 2138                 if (error)
 2139                         return (error);
 2140                 error = sa6_embedscope(lin6, V_ip6_use_defzone);
 2141                 if (error)
 2142                         return (error);
 2143                 break;
 2144 #endif
 2145         case AF_INET:
 2146                 fin = (struct sockaddr_in *)&addrs[0];
 2147                 lin = (struct sockaddr_in *)&addrs[1];
 2148                 if (fin->sin_len != sizeof(struct sockaddr_in) ||
 2149                     lin->sin_len != sizeof(struct sockaddr_in))
 2150                         return (EINVAL);
 2151                 break;
 2152         default:
 2153                 return (EINVAL);
 2154         }
 2155         INP_INFO_WLOCK(&V_tcbinfo);
 2156         switch (addrs[0].ss_family) {
 2157 #ifdef INET6
 2158         case AF_INET6:
 2159                 inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
 2160                     fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
 2161                     NULL);
 2162                 break;
 2163 #endif
 2164         case AF_INET:
 2165                 inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
 2166                     fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
 2167                 break;
 2168         }
 2169         if (inp != NULL) {
 2170                 INP_WLOCK(inp);
 2171                 if (inp->inp_flags & INP_TIMEWAIT) {
 2172                         /*
 2173                          * XXXRW: There currently exists a state where an
 2174                          * inpcb is present, but its timewait state has been
 2175                          * discarded.  For now, don't allow dropping of this
 2176                          * type of inpcb.
 2177                          */
 2178                         tw = intotw(inp);
 2179                         if (tw != NULL)
 2180                                 tcp_twclose(tw, 0);
 2181                         else
 2182                                 INP_WUNLOCK(inp);
 2183                 } else if (!(inp->inp_flags & INP_DROPPED) &&
 2184                            !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
 2185                         tp = intotcpcb(inp);
 2186                         tp = tcp_drop(tp, ECONNABORTED);
 2187                         if (tp != NULL)
 2188                                 INP_WUNLOCK(inp);
 2189                 } else
 2190                         INP_WUNLOCK(inp);
 2191         } else
 2192                 error = ESRCH;
 2193         INP_INFO_WUNLOCK(&V_tcbinfo);
 2194         return (error);
 2195 }
 2196 
 2197 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
 2198     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
 2199     0, sysctl_drop, "", "Drop TCP connection");
 2200 
 2201 /*
 2202  * Generate a standardized TCP log line for use throughout the
 2203  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
 2204  * allow use in the interrupt context.
 2205  *
 2206  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
 2207  * NB: The function may return NULL if memory allocation failed.
 2208  *
 2209  * Due to header inclusion and ordering limitations the struct ip
 2210  * and ip6_hdr pointers have to be passed as void pointers.
 2211  */
 2212 char *
 2213 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
 2214     const void *ip6hdr)
 2215 {
 2216 
 2217         /* Is logging enabled? */
 2218         if (tcp_log_in_vain == 0)
 2219                 return (NULL);
 2220 
 2221         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
 2222 }
 2223 
 2224 char *
 2225 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
 2226     const void *ip6hdr)
 2227 {
 2228 
 2229         /* Is logging enabled? */
 2230         if (tcp_log_debug == 0)
 2231                 return (NULL);
 2232 
 2233         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
 2234 }
 2235 
 2236 static char *
 2237 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
 2238     const void *ip6hdr)
 2239 {
 2240         char *s, *sp;
 2241         size_t size;
 2242         struct ip *ip;
 2243 #ifdef INET6
 2244         const struct ip6_hdr *ip6;
 2245 
 2246         ip6 = (const struct ip6_hdr *)ip6hdr;
 2247 #endif /* INET6 */
 2248         ip = (struct ip *)ip4hdr;
 2249 
 2250         /*
 2251          * The log line looks like this:
 2252          * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
 2253          */
 2254         size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
 2255             sizeof(PRINT_TH_FLAGS) + 1 +
 2256 #ifdef INET6
 2257             2 * INET6_ADDRSTRLEN;
 2258 #else
 2259             2 * INET_ADDRSTRLEN;
 2260 #endif /* INET6 */
 2261 
 2262         s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
 2263         if (s == NULL)
 2264                 return (NULL);
 2265 
 2266         strcat(s, "TCP: [");
 2267         sp = s + strlen(s);
 2268 
 2269         if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
 2270                 inet_ntoa_r(inc->inc_faddr, sp);
 2271                 sp = s + strlen(s);
 2272                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
 2273                 sp = s + strlen(s);
 2274                 inet_ntoa_r(inc->inc_laddr, sp);
 2275                 sp = s + strlen(s);
 2276                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
 2277 #ifdef INET6
 2278         } else if (inc) {
 2279                 ip6_sprintf(sp, &inc->inc6_faddr);
 2280                 sp = s + strlen(s);
 2281                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
 2282                 sp = s + strlen(s);
 2283                 ip6_sprintf(sp, &inc->inc6_laddr);
 2284                 sp = s + strlen(s);
 2285                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
 2286         } else if (ip6 && th) {
 2287                 ip6_sprintf(sp, &ip6->ip6_src);
 2288                 sp = s + strlen(s);
 2289                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
 2290                 sp = s + strlen(s);
 2291                 ip6_sprintf(sp, &ip6->ip6_dst);
 2292                 sp = s + strlen(s);
 2293                 sprintf(sp, "]:%i", ntohs(th->th_dport));
 2294 #endif /* INET6 */
 2295         } else if (ip && th) {
 2296                 inet_ntoa_r(ip->ip_src, sp);
 2297                 sp = s + strlen(s);
 2298                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
 2299                 sp = s + strlen(s);
 2300                 inet_ntoa_r(ip->ip_dst, sp);
 2301                 sp = s + strlen(s);
 2302                 sprintf(sp, "]:%i", ntohs(th->th_dport));
 2303         } else {
 2304                 free(s, M_TCPLOG);
 2305                 return (NULL);
 2306         }
 2307         sp = s + strlen(s);
 2308         if (th)
 2309                 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
 2310         if (*(s + size - 1) != '\0')
 2311                 panic("%s: string too long", __func__);
 2312         return (s);
 2313 }

Cache object: 9533c3fcc43da507f50869af7b6c65f0


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.