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  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
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/10.1/sys/netinet/tcp_subr.c 285979 2015-07-28 19:59:11Z delphij $");
   34 
   35 #include "opt_compat.h"
   36 #include "opt_inet.h"
   37 #include "opt_inet6.h"
   38 #include "opt_ipsec.h"
   39 #include "opt_kdtrace.h"
   40 #include "opt_tcpdebug.h"
   41 
   42 #include <sys/param.h>
   43 #include <sys/systm.h>
   44 #include <sys/callout.h>
   45 #include <sys/hhook.h>
   46 #include <sys/kernel.h>
   47 #include <sys/khelp.h>
   48 #include <sys/sysctl.h>
   49 #include <sys/jail.h>
   50 #include <sys/malloc.h>
   51 #include <sys/mbuf.h>
   52 #ifdef INET6
   53 #include <sys/domain.h>
   54 #endif
   55 #include <sys/priv.h>
   56 #include <sys/proc.h>
   57 #include <sys/sdt.h>
   58 #include <sys/socket.h>
   59 #include <sys/socketvar.h>
   60 #include <sys/protosw.h>
   61 #include <sys/random.h>
   62 
   63 #include <vm/uma.h>
   64 
   65 #include <net/route.h>
   66 #include <net/if.h>
   67 #include <net/vnet.h>
   68 
   69 #include <netinet/cc.h>
   70 #include <netinet/in.h>
   71 #include <netinet/in_kdtrace.h>
   72 #include <netinet/in_pcb.h>
   73 #include <netinet/in_systm.h>
   74 #include <netinet/in_var.h>
   75 #include <netinet/ip.h>
   76 #include <netinet/ip_icmp.h>
   77 #include <netinet/ip_var.h>
   78 #ifdef INET6
   79 #include <netinet/ip6.h>
   80 #include <netinet6/in6_pcb.h>
   81 #include <netinet6/ip6_var.h>
   82 #include <netinet6/scope6_var.h>
   83 #include <netinet6/nd6.h>
   84 #endif
   85 
   86 #include <netinet/tcp_fsm.h>
   87 #include <netinet/tcp_seq.h>
   88 #include <netinet/tcp_timer.h>
   89 #include <netinet/tcp_var.h>
   90 #include <netinet/tcp_syncache.h>
   91 #ifdef INET6
   92 #include <netinet6/tcp6_var.h>
   93 #endif
   94 #include <netinet/tcpip.h>
   95 #ifdef TCPDEBUG
   96 #include <netinet/tcp_debug.h>
   97 #endif
   98 #ifdef INET6
   99 #include <netinet6/ip6protosw.h>
  100 #endif
  101 #ifdef TCP_OFFLOAD
  102 #include <netinet/tcp_offload.h>
  103 #endif
  104 
  105 #ifdef IPSEC
  106 #include <netipsec/ipsec.h>
  107 #include <netipsec/xform.h>
  108 #ifdef INET6
  109 #include <netipsec/ipsec6.h>
  110 #endif
  111 #include <netipsec/key.h>
  112 #include <sys/syslog.h>
  113 #endif /*IPSEC*/
  114 
  115 #include <machine/in_cksum.h>
  116 #include <sys/md5.h>
  117 
  118 #include <security/mac/mac_framework.h>
  119 
  120 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
  121 #ifdef INET6
  122 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
  123 #endif
  124 
  125 static int
  126 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
  127 {
  128         int error, new;
  129 
  130         new = V_tcp_mssdflt;
  131         error = sysctl_handle_int(oidp, &new, 0, req);
  132         if (error == 0 && req->newptr) {
  133                 if (new < TCP_MINMSS)
  134                         error = EINVAL;
  135                 else
  136                         V_tcp_mssdflt = new;
  137         }
  138         return (error);
  139 }
  140 
  141 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
  142     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
  143     &sysctl_net_inet_tcp_mss_check, "I",
  144     "Default TCP Maximum Segment Size");
  145 
  146 #ifdef INET6
  147 static int
  148 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
  149 {
  150         int error, new;
  151 
  152         new = V_tcp_v6mssdflt;
  153         error = sysctl_handle_int(oidp, &new, 0, req);
  154         if (error == 0 && req->newptr) {
  155                 if (new < TCP_MINMSS)
  156                         error = EINVAL;
  157                 else
  158                         V_tcp_v6mssdflt = new;
  159         }
  160         return (error);
  161 }
  162 
  163 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
  164     CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
  165     &sysctl_net_inet_tcp_mss_v6_check, "I",
  166    "Default TCP Maximum Segment Size for IPv6");
  167 #endif /* INET6 */
  168 
  169 /*
  170  * Minimum MSS we accept and use. This prevents DoS attacks where
  171  * we are forced to a ridiculous low MSS like 20 and send hundreds
  172  * of packets instead of one. The effect scales with the available
  173  * bandwidth and quickly saturates the CPU and network interface
  174  * with packet generation and sending. Set to zero to disable MINMSS
  175  * checking. This setting prevents us from sending too small packets.
  176  */
  177 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
  178 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
  179      &VNET_NAME(tcp_minmss), 0,
  180     "Minimum TCP Maximum Segment Size");
  181 
  182 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
  183 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
  184     &VNET_NAME(tcp_do_rfc1323), 0,
  185     "Enable rfc1323 (high performance TCP) extensions");
  186 
  187 static int      tcp_log_debug = 0;
  188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
  189     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
  190 
  191 static int      tcp_tcbhashsize = 0;
  192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
  193     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
  194 
  195 static int      do_tcpdrain = 1;
  196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
  197     "Enable tcp_drain routine for extra help when low on mbufs");
  198 
  199 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
  200     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
  201 
  202 static VNET_DEFINE(int, icmp_may_rst) = 1;
  203 #define V_icmp_may_rst                  VNET(icmp_may_rst)
  204 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
  205     &VNET_NAME(icmp_may_rst), 0,
  206     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
  207 
  208 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
  209 #define V_tcp_isn_reseed_interval       VNET(tcp_isn_reseed_interval)
  210 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
  211     &VNET_NAME(tcp_isn_reseed_interval), 0,
  212     "Seconds between reseeding of ISN secret");
  213 
  214 static int      tcp_soreceive_stream = 0;
  215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
  216     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
  217 
  218 #ifdef TCP_SIGNATURE
  219 static int      tcp_sig_checksigs = 1;
  220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
  221     &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
  222 #endif
  223 
  224 VNET_DEFINE(uma_zone_t, sack_hole_zone);
  225 #define V_sack_hole_zone                VNET(sack_hole_zone)
  226 
  227 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
  228 
  229 static struct inpcb *tcp_notify(struct inpcb *, int);
  230 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
  231 static char *   tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
  232                     void *ip4hdr, const void *ip6hdr);
  233 
  234 /*
  235  * Target size of TCP PCB hash tables. Must be a power of two.
  236  *
  237  * Note that this can be overridden by the kernel environment
  238  * variable net.inet.tcp.tcbhashsize
  239  */
  240 #ifndef TCBHASHSIZE
  241 #define TCBHASHSIZE     0
  242 #endif
  243 
  244 /*
  245  * XXX
  246  * Callouts should be moved into struct tcp directly.  They are currently
  247  * separate because the tcpcb structure is exported to userland for sysctl
  248  * parsing purposes, which do not know about callouts.
  249  */
  250 struct tcpcb_mem {
  251         struct  tcpcb           tcb;
  252         struct  tcp_timer       tt;
  253         struct  cc_var          ccv;
  254         struct  osd             osd;
  255 };
  256 
  257 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
  258 #define V_tcpcb_zone                    VNET(tcpcb_zone)
  259 
  260 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
  261 static struct mtx isn_mtx;
  262 
  263 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
  264 #define ISN_LOCK()      mtx_lock(&isn_mtx)
  265 #define ISN_UNLOCK()    mtx_unlock(&isn_mtx)
  266 
  267 /*
  268  * TCP initialization.
  269  */
  270 static void
  271 tcp_zone_change(void *tag)
  272 {
  273 
  274         uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
  275         uma_zone_set_max(V_tcpcb_zone, maxsockets);
  276         tcp_tw_zone_change();
  277 }
  278 
  279 static int
  280 tcp_inpcb_init(void *mem, int size, int flags)
  281 {
  282         struct inpcb *inp = mem;
  283 
  284         INP_LOCK_INIT(inp, "inp", "tcpinp");
  285         return (0);
  286 }
  287 
  288 /*
  289  * Take a value and get the next power of 2 that doesn't overflow.
  290  * Used to size the tcp_inpcb hash buckets.
  291  */
  292 static int
  293 maketcp_hashsize(int size)
  294 {
  295         int hashsize;
  296 
  297         /*
  298          * auto tune.
  299          * get the next power of 2 higher than maxsockets.
  300          */
  301         hashsize = 1 << fls(size);
  302         /* catch overflow, and just go one power of 2 smaller */
  303         if (hashsize < size) {
  304                 hashsize = 1 << (fls(size) - 1);
  305         }
  306         return (hashsize);
  307 }
  308 
  309 void
  310 tcp_init(void)
  311 {
  312         const char *tcbhash_tuneable;
  313         int hashsize;
  314 
  315         tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
  316 
  317         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
  318             &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
  319                 printf("%s: WARNING: unable to register helper hook\n", __func__);
  320         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
  321             &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
  322                 printf("%s: WARNING: unable to register helper hook\n", __func__);
  323 
  324         hashsize = TCBHASHSIZE;
  325         TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
  326         if (hashsize == 0) {
  327                 /*
  328                  * Auto tune the hash size based on maxsockets.
  329                  * A perfect hash would have a 1:1 mapping
  330                  * (hashsize = maxsockets) however it's been
  331                  * suggested that O(2) average is better.
  332                  */
  333                 hashsize = maketcp_hashsize(maxsockets / 4);
  334                 /*
  335                  * Our historical default is 512,
  336                  * do not autotune lower than this.
  337                  */
  338                 if (hashsize < 512)
  339                         hashsize = 512;
  340                 if (bootverbose)
  341                         printf("%s: %s auto tuned to %d\n", __func__,
  342                             tcbhash_tuneable, hashsize);
  343         }
  344         /*
  345          * We require a hashsize to be a power of two.
  346          * Previously if it was not a power of two we would just reset it
  347          * back to 512, which could be a nasty surprise if you did not notice
  348          * the error message.
  349          * Instead what we do is clip it to the closest power of two lower
  350          * than the specified hash value.
  351          */
  352         if (!powerof2(hashsize)) {
  353                 int oldhashsize = hashsize;
  354 
  355                 hashsize = maketcp_hashsize(hashsize);
  356                 /* prevent absurdly low value */
  357                 if (hashsize < 16)
  358                         hashsize = 16;
  359                 printf("%s: WARNING: TCB hash size not a power of 2, "
  360                     "clipped from %d to %d.\n", __func__, oldhashsize,
  361                     hashsize);
  362         }
  363         in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
  364             "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
  365             IPI_HASHFIELDS_4TUPLE);
  366 
  367         /*
  368          * These have to be type stable for the benefit of the timers.
  369          */
  370         V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
  371             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
  372         uma_zone_set_max(V_tcpcb_zone, maxsockets);
  373         uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
  374 
  375         tcp_tw_init();
  376         syncache_init();
  377         tcp_hc_init();
  378 
  379         TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
  380         V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
  381             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
  382 
  383         /* Skip initialization of globals for non-default instances. */
  384         if (!IS_DEFAULT_VNET(curvnet))
  385                 return;
  386 
  387         tcp_reass_global_init();
  388 
  389         /* XXX virtualize those bellow? */
  390         tcp_delacktime = TCPTV_DELACK;
  391         tcp_keepinit = TCPTV_KEEP_INIT;
  392         tcp_keepidle = TCPTV_KEEP_IDLE;
  393         tcp_keepintvl = TCPTV_KEEPINTVL;
  394         tcp_maxpersistidle = TCPTV_KEEP_IDLE;
  395         tcp_msl = TCPTV_MSL;
  396         tcp_rexmit_min = TCPTV_MIN;
  397         if (tcp_rexmit_min < 1)
  398                 tcp_rexmit_min = 1;
  399         tcp_rexmit_slop = TCPTV_CPU_VAR;
  400         tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
  401         tcp_tcbhashsize = hashsize;
  402 
  403         TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
  404         if (tcp_soreceive_stream) {
  405 #ifdef INET
  406                 tcp_usrreqs.pru_soreceive = soreceive_stream;
  407 #endif
  408 #ifdef INET6
  409                 tcp6_usrreqs.pru_soreceive = soreceive_stream;
  410 #endif /* INET6 */
  411         }
  412 
  413 #ifdef INET6
  414 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
  415 #else /* INET6 */
  416 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
  417 #endif /* INET6 */
  418         if (max_protohdr < TCP_MINPROTOHDR)
  419                 max_protohdr = TCP_MINPROTOHDR;
  420         if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
  421                 panic("tcp_init");
  422 #undef TCP_MINPROTOHDR
  423 
  424         ISN_LOCK_INIT();
  425         EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
  426                 SHUTDOWN_PRI_DEFAULT);
  427         EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
  428                 EVENTHANDLER_PRI_ANY);
  429 }
  430 
  431 #ifdef VIMAGE
  432 void
  433 tcp_destroy(void)
  434 {
  435 
  436         tcp_hc_destroy();
  437         syncache_destroy();
  438         tcp_tw_destroy();
  439         in_pcbinfo_destroy(&V_tcbinfo);
  440         uma_zdestroy(V_sack_hole_zone);
  441         uma_zdestroy(V_tcpcb_zone);
  442 }
  443 #endif
  444 
  445 void
  446 tcp_fini(void *xtp)
  447 {
  448 
  449 }
  450 
  451 /*
  452  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
  453  * tcp_template used to store this data in mbufs, but we now recopy it out
  454  * of the tcpcb each time to conserve mbufs.
  455  */
  456 void
  457 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
  458 {
  459         struct tcphdr *th = (struct tcphdr *)tcp_ptr;
  460 
  461         INP_WLOCK_ASSERT(inp);
  462 
  463 #ifdef INET6
  464         if ((inp->inp_vflag & INP_IPV6) != 0) {
  465                 struct ip6_hdr *ip6;
  466 
  467                 ip6 = (struct ip6_hdr *)ip_ptr;
  468                 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
  469                         (inp->inp_flow & IPV6_FLOWINFO_MASK);
  470                 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
  471                         (IPV6_VERSION & IPV6_VERSION_MASK);
  472                 ip6->ip6_nxt = IPPROTO_TCP;
  473                 ip6->ip6_plen = htons(sizeof(struct tcphdr));
  474                 ip6->ip6_src = inp->in6p_laddr;
  475                 ip6->ip6_dst = inp->in6p_faddr;
  476         }
  477 #endif /* INET6 */
  478 #if defined(INET6) && defined(INET)
  479         else
  480 #endif
  481 #ifdef INET
  482         {
  483                 struct ip *ip;
  484 
  485                 ip = (struct ip *)ip_ptr;
  486                 ip->ip_v = IPVERSION;
  487                 ip->ip_hl = 5;
  488                 ip->ip_tos = inp->inp_ip_tos;
  489                 ip->ip_len = 0;
  490                 ip->ip_id = 0;
  491                 ip->ip_off = 0;
  492                 ip->ip_ttl = inp->inp_ip_ttl;
  493                 ip->ip_sum = 0;
  494                 ip->ip_p = IPPROTO_TCP;
  495                 ip->ip_src = inp->inp_laddr;
  496                 ip->ip_dst = inp->inp_faddr;
  497         }
  498 #endif /* INET */
  499         th->th_sport = inp->inp_lport;
  500         th->th_dport = inp->inp_fport;
  501         th->th_seq = 0;
  502         th->th_ack = 0;
  503         th->th_x2 = 0;
  504         th->th_off = 5;
  505         th->th_flags = 0;
  506         th->th_win = 0;
  507         th->th_urp = 0;
  508         th->th_sum = 0;         /* in_pseudo() is called later for ipv4 */
  509 }
  510 
  511 /*
  512  * Create template to be used to send tcp packets on a connection.
  513  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
  514  * use for this function is in keepalives, which use tcp_respond.
  515  */
  516 struct tcptemp *
  517 tcpip_maketemplate(struct inpcb *inp)
  518 {
  519         struct tcptemp *t;
  520 
  521         t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
  522         if (t == NULL)
  523                 return (NULL);
  524         tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
  525         return (t);
  526 }
  527 
  528 /*
  529  * Send a single message to the TCP at address specified by
  530  * the given TCP/IP header.  If m == NULL, then we make a copy
  531  * of the tcpiphdr at ti and send directly to the addressed host.
  532  * This is used to force keep alive messages out using the TCP
  533  * template for a connection.  If flags are given then we send
  534  * a message back to the TCP which originated the * segment ti,
  535  * and discard the mbuf containing it and any other attached mbufs.
  536  *
  537  * In any case the ack and sequence number of the transmitted
  538  * segment are as specified by the parameters.
  539  *
  540  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
  541  */
  542 void
  543 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
  544     tcp_seq ack, tcp_seq seq, int flags)
  545 {
  546         int tlen;
  547         int win = 0;
  548         struct ip *ip;
  549         struct tcphdr *nth;
  550 #ifdef INET6
  551         struct ip6_hdr *ip6;
  552         int isipv6;
  553 #endif /* INET6 */
  554         int ipflags = 0;
  555         struct inpcb *inp;
  556 
  557         KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
  558 
  559 #ifdef INET6
  560         isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
  561         ip6 = ipgen;
  562 #endif /* INET6 */
  563         ip = ipgen;
  564 
  565         if (tp != NULL) {
  566                 inp = tp->t_inpcb;
  567                 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
  568                 INP_WLOCK_ASSERT(inp);
  569         } else
  570                 inp = NULL;
  571 
  572         if (tp != NULL) {
  573                 if (!(flags & TH_RST)) {
  574                         win = sbspace(&inp->inp_socket->so_rcv);
  575                         if (win > (long)TCP_MAXWIN << tp->rcv_scale)
  576                                 win = (long)TCP_MAXWIN << tp->rcv_scale;
  577                 }
  578         }
  579         if (m == NULL) {
  580                 m = m_gethdr(M_NOWAIT, MT_DATA);
  581                 if (m == NULL)
  582                         return;
  583                 tlen = 0;
  584                 m->m_data += max_linkhdr;
  585 #ifdef INET6
  586                 if (isipv6) {
  587                         bcopy((caddr_t)ip6, mtod(m, caddr_t),
  588                               sizeof(struct ip6_hdr));
  589                         ip6 = mtod(m, struct ip6_hdr *);
  590                         nth = (struct tcphdr *)(ip6 + 1);
  591                 } else
  592 #endif /* INET6 */
  593                 {
  594                         bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
  595                         ip = mtod(m, struct ip *);
  596                         nth = (struct tcphdr *)(ip + 1);
  597                 }
  598                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
  599                 flags = TH_ACK;
  600         } else {
  601                 /*
  602                  *  reuse the mbuf. 
  603                  * XXX MRT We inherrit the FIB, which is lucky.
  604                  */
  605                 m_freem(m->m_next);
  606                 m->m_next = NULL;
  607                 m->m_data = (caddr_t)ipgen;
  608                 /* m_len is set later */
  609                 tlen = 0;
  610 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
  611 #ifdef INET6
  612                 if (isipv6) {
  613                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
  614                         nth = (struct tcphdr *)(ip6 + 1);
  615                 } else
  616 #endif /* INET6 */
  617                 {
  618                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
  619                         nth = (struct tcphdr *)(ip + 1);
  620                 }
  621                 if (th != nth) {
  622                         /*
  623                          * this is usually a case when an extension header
  624                          * exists between the IPv6 header and the
  625                          * TCP header.
  626                          */
  627                         nth->th_sport = th->th_sport;
  628                         nth->th_dport = th->th_dport;
  629                 }
  630                 xchg(nth->th_dport, nth->th_sport, uint16_t);
  631 #undef xchg
  632         }
  633 #ifdef INET6
  634         if (isipv6) {
  635                 ip6->ip6_flow = 0;
  636                 ip6->ip6_vfc = IPV6_VERSION;
  637                 ip6->ip6_nxt = IPPROTO_TCP;
  638                 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
  639                 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
  640         }
  641 #endif
  642 #if defined(INET) && defined(INET6)
  643         else
  644 #endif
  645 #ifdef INET
  646         {
  647                 tlen += sizeof (struct tcpiphdr);
  648                 ip->ip_len = htons(tlen);
  649                 ip->ip_ttl = V_ip_defttl;
  650                 if (V_path_mtu_discovery)
  651                         ip->ip_off |= htons(IP_DF);
  652         }
  653 #endif
  654         m->m_len = tlen;
  655         m->m_pkthdr.len = tlen;
  656         m->m_pkthdr.rcvif = NULL;
  657 #ifdef MAC
  658         if (inp != NULL) {
  659                 /*
  660                  * Packet is associated with a socket, so allow the
  661                  * label of the response to reflect the socket label.
  662                  */
  663                 INP_WLOCK_ASSERT(inp);
  664                 mac_inpcb_create_mbuf(inp, m);
  665         } else {
  666                 /*
  667                  * Packet is not associated with a socket, so possibly
  668                  * update the label in place.
  669                  */
  670                 mac_netinet_tcp_reply(m);
  671         }
  672 #endif
  673         nth->th_seq = htonl(seq);
  674         nth->th_ack = htonl(ack);
  675         nth->th_x2 = 0;
  676         nth->th_off = sizeof (struct tcphdr) >> 2;
  677         nth->th_flags = flags;
  678         if (tp != NULL)
  679                 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
  680         else
  681                 nth->th_win = htons((u_short)win);
  682         nth->th_urp = 0;
  683 
  684         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
  685 #ifdef INET6
  686         if (isipv6) {
  687                 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
  688                 nth->th_sum = in6_cksum_pseudo(ip6,
  689                     tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
  690                 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
  691                     NULL, NULL);
  692         }
  693 #endif /* INET6 */
  694 #if defined(INET6) && defined(INET)
  695         else
  696 #endif
  697 #ifdef INET
  698         {
  699                 m->m_pkthdr.csum_flags = CSUM_TCP;
  700                 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
  701                     htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
  702         }
  703 #endif /* INET */
  704 #ifdef TCPDEBUG
  705         if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
  706                 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
  707 #endif
  708         if (flags & TH_RST)
  709                 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
  710                     tp, nth);
  711 
  712         TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
  713 #ifdef INET6
  714         if (isipv6)
  715                 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
  716 #endif /* INET6 */
  717 #if defined(INET) && defined(INET6)
  718         else
  719 #endif
  720 #ifdef INET
  721                 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
  722 #endif
  723 }
  724 
  725 /*
  726  * Create a new TCP control block, making an
  727  * empty reassembly queue and hooking it to the argument
  728  * protocol control block.  The `inp' parameter must have
  729  * come from the zone allocator set up in tcp_init().
  730  */
  731 struct tcpcb *
  732 tcp_newtcpcb(struct inpcb *inp)
  733 {
  734         struct tcpcb_mem *tm;
  735         struct tcpcb *tp;
  736 #ifdef INET6
  737         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
  738 #endif /* INET6 */
  739 
  740         tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
  741         if (tm == NULL)
  742                 return (NULL);
  743         tp = &tm->tcb;
  744 
  745         /* Initialise cc_var struct for this tcpcb. */
  746         tp->ccv = &tm->ccv;
  747         tp->ccv->type = IPPROTO_TCP;
  748         tp->ccv->ccvc.tcp = tp;
  749 
  750         /*
  751          * Use the current system default CC algorithm.
  752          */
  753         CC_LIST_RLOCK();
  754         KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
  755         CC_ALGO(tp) = CC_DEFAULT();
  756         CC_LIST_RUNLOCK();
  757 
  758         if (CC_ALGO(tp)->cb_init != NULL)
  759                 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
  760                         uma_zfree(V_tcpcb_zone, tm);
  761                         return (NULL);
  762                 }
  763 
  764         tp->osd = &tm->osd;
  765         if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
  766                 uma_zfree(V_tcpcb_zone, tm);
  767                 return (NULL);
  768         }
  769 
  770 #ifdef VIMAGE
  771         tp->t_vnet = inp->inp_vnet;
  772 #endif
  773         tp->t_timers = &tm->tt;
  774         /*      LIST_INIT(&tp->t_segq); */      /* XXX covered by M_ZERO */
  775         tp->t_maxseg = tp->t_maxopd =
  776 #ifdef INET6
  777                 isipv6 ? V_tcp_v6mssdflt :
  778 #endif /* INET6 */
  779                 V_tcp_mssdflt;
  780 
  781         /* Set up our timeouts. */
  782         callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
  783         callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
  784         callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
  785         callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
  786         callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
  787 
  788         if (V_tcp_do_rfc1323)
  789                 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
  790         if (V_tcp_do_sack)
  791                 tp->t_flags |= TF_SACK_PERMIT;
  792         TAILQ_INIT(&tp->snd_holes);
  793         tp->t_inpcb = inp;      /* XXX */
  794         /*
  795          * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
  796          * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
  797          * reasonable initial retransmit time.
  798          */
  799         tp->t_srtt = TCPTV_SRTTBASE;
  800         tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
  801         tp->t_rttmin = tcp_rexmit_min;
  802         tp->t_rxtcur = TCPTV_RTOBASE;
  803         tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
  804         tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
  805         tp->t_rcvtime = ticks;
  806         /*
  807          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
  808          * because the socket may be bound to an IPv6 wildcard address,
  809          * which may match an IPv4-mapped IPv6 address.
  810          */
  811         inp->inp_ip_ttl = V_ip_defttl;
  812         inp->inp_ppcb = tp;
  813         return (tp);            /* XXX */
  814 }
  815 
  816 /*
  817  * Switch the congestion control algorithm back to NewReno for any active
  818  * control blocks using an algorithm which is about to go away.
  819  * This ensures the CC framework can allow the unload to proceed without leaving
  820  * any dangling pointers which would trigger a panic.
  821  * Returning non-zero would inform the CC framework that something went wrong
  822  * and it would be unsafe to allow the unload to proceed. However, there is no
  823  * way for this to occur with this implementation so we always return zero.
  824  */
  825 int
  826 tcp_ccalgounload(struct cc_algo *unload_algo)
  827 {
  828         struct cc_algo *tmpalgo;
  829         struct inpcb *inp;
  830         struct tcpcb *tp;
  831         VNET_ITERATOR_DECL(vnet_iter);
  832 
  833         /*
  834          * Check all active control blocks across all network stacks and change
  835          * any that are using "unload_algo" back to NewReno. If "unload_algo"
  836          * requires cleanup code to be run, call it.
  837          */
  838         VNET_LIST_RLOCK();
  839         VNET_FOREACH(vnet_iter) {
  840                 CURVNET_SET(vnet_iter);
  841                 INP_INFO_RLOCK(&V_tcbinfo);
  842                 /*
  843                  * New connections already part way through being initialised
  844                  * with the CC algo we're removing will not race with this code
  845                  * because the INP_INFO_WLOCK is held during initialisation. We
  846                  * therefore don't enter the loop below until the connection
  847                  * list has stabilised.
  848                  */
  849                 LIST_FOREACH(inp, &V_tcb, inp_list) {
  850                         INP_WLOCK(inp);
  851                         /* Important to skip tcptw structs. */
  852                         if (!(inp->inp_flags & INP_TIMEWAIT) &&
  853                             (tp = intotcpcb(inp)) != NULL) {
  854                                 /*
  855                                  * By holding INP_WLOCK here, we are assured
  856                                  * that the connection is not currently
  857                                  * executing inside the CC module's functions
  858                                  * i.e. it is safe to make the switch back to
  859                                  * NewReno.
  860                                  */
  861                                 if (CC_ALGO(tp) == unload_algo) {
  862                                         tmpalgo = CC_ALGO(tp);
  863                                         /* NewReno does not require any init. */
  864                                         CC_ALGO(tp) = &newreno_cc_algo;
  865                                         if (tmpalgo->cb_destroy != NULL)
  866                                                 tmpalgo->cb_destroy(tp->ccv);
  867                                 }
  868                         }
  869                         INP_WUNLOCK(inp);
  870                 }
  871                 INP_INFO_RUNLOCK(&V_tcbinfo);
  872                 CURVNET_RESTORE();
  873         }
  874         VNET_LIST_RUNLOCK();
  875 
  876         return (0);
  877 }
  878 
  879 /*
  880  * Drop a TCP connection, reporting
  881  * the specified error.  If connection is synchronized,
  882  * then send a RST to peer.
  883  */
  884 struct tcpcb *
  885 tcp_drop(struct tcpcb *tp, int errno)
  886 {
  887         struct socket *so = tp->t_inpcb->inp_socket;
  888 
  889         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
  890         INP_WLOCK_ASSERT(tp->t_inpcb);
  891 
  892         if (TCPS_HAVERCVDSYN(tp->t_state)) {
  893                 tcp_state_change(tp, TCPS_CLOSED);
  894                 (void) tcp_output(tp);
  895                 TCPSTAT_INC(tcps_drops);
  896         } else
  897                 TCPSTAT_INC(tcps_conndrops);
  898         if (errno == ETIMEDOUT && tp->t_softerror)
  899                 errno = tp->t_softerror;
  900         so->so_error = errno;
  901         return (tcp_close(tp));
  902 }
  903 
  904 void
  905 tcp_discardcb(struct tcpcb *tp)
  906 {
  907         struct inpcb *inp = tp->t_inpcb;
  908         struct socket *so = inp->inp_socket;
  909 #ifdef INET6
  910         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
  911 #endif /* INET6 */
  912 
  913         INP_WLOCK_ASSERT(inp);
  914 
  915         /*
  916          * Make sure that all of our timers are stopped before we delete the
  917          * PCB.
  918          *
  919          * XXXRW: Really, we would like to use callout_drain() here in order
  920          * to avoid races experienced in tcp_timer.c where a timer is already
  921          * executing at this point.  However, we can't, both because we're
  922          * running in a context where we can't sleep, and also because we
  923          * hold locks required by the timers.  What we instead need to do is
  924          * test to see if callout_drain() is required, and if so, defer some
  925          * portion of the remainder of tcp_discardcb() to an asynchronous
  926          * context that can callout_drain() and then continue.  Some care
  927          * will be required to ensure that no further processing takes place
  928          * on the tcpcb, even though it hasn't been freed (a flag?).
  929          */
  930         callout_stop(&tp->t_timers->tt_rexmt);
  931         callout_stop(&tp->t_timers->tt_persist);
  932         callout_stop(&tp->t_timers->tt_keep);
  933         callout_stop(&tp->t_timers->tt_2msl);
  934         callout_stop(&tp->t_timers->tt_delack);
  935 
  936         /*
  937          * If we got enough samples through the srtt filter,
  938          * save the rtt and rttvar in the routing entry.
  939          * 'Enough' is arbitrarily defined as 4 rtt samples.
  940          * 4 samples is enough for the srtt filter to converge
  941          * to within enough % of the correct value; fewer samples
  942          * and we could save a bogus rtt. The danger is not high
  943          * as tcp quickly recovers from everything.
  944          * XXX: Works very well but needs some more statistics!
  945          */
  946         if (tp->t_rttupdated >= 4) {
  947                 struct hc_metrics_lite metrics;
  948                 u_long ssthresh;
  949 
  950                 bzero(&metrics, sizeof(metrics));
  951                 /*
  952                  * Update the ssthresh always when the conditions below
  953                  * are satisfied. This gives us better new start value
  954                  * for the congestion avoidance for new connections.
  955                  * ssthresh is only set if packet loss occured on a session.
  956                  *
  957                  * XXXRW: 'so' may be NULL here, and/or socket buffer may be
  958                  * being torn down.  Ideally this code would not use 'so'.
  959                  */
  960                 ssthresh = tp->snd_ssthresh;
  961                 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
  962                         /*
  963                          * convert the limit from user data bytes to
  964                          * packets then to packet data bytes.
  965                          */
  966                         ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
  967                         if (ssthresh < 2)
  968                                 ssthresh = 2;
  969                         ssthresh *= (u_long)(tp->t_maxseg +
  970 #ifdef INET6
  971                             (isipv6 ? sizeof (struct ip6_hdr) +
  972                                 sizeof (struct tcphdr) :
  973 #endif
  974                                 sizeof (struct tcpiphdr)
  975 #ifdef INET6
  976                             )
  977 #endif
  978                             );
  979                 } else
  980                         ssthresh = 0;
  981                 metrics.rmx_ssthresh = ssthresh;
  982 
  983                 metrics.rmx_rtt = tp->t_srtt;
  984                 metrics.rmx_rttvar = tp->t_rttvar;
  985                 metrics.rmx_cwnd = tp->snd_cwnd;
  986                 metrics.rmx_sendpipe = 0;
  987                 metrics.rmx_recvpipe = 0;
  988 
  989                 tcp_hc_update(&inp->inp_inc, &metrics);
  990         }
  991 
  992         /* free the reassembly queue, if any */
  993         tcp_reass_flush(tp);
  994 
  995 #ifdef TCP_OFFLOAD
  996         /* Disconnect offload device, if any. */
  997         if (tp->t_flags & TF_TOE)
  998                 tcp_offload_detach(tp);
  999 #endif
 1000                 
 1001         tcp_free_sackholes(tp);
 1002 
 1003         /* Allow the CC algorithm to clean up after itself. */
 1004         if (CC_ALGO(tp)->cb_destroy != NULL)
 1005                 CC_ALGO(tp)->cb_destroy(tp->ccv);
 1006 
 1007         khelp_destroy_osd(tp->osd);
 1008 
 1009         CC_ALGO(tp) = NULL;
 1010         inp->inp_ppcb = NULL;
 1011         tp->t_inpcb = NULL;
 1012         uma_zfree(V_tcpcb_zone, tp);
 1013 }
 1014 
 1015 /*
 1016  * Attempt to close a TCP control block, marking it as dropped, and freeing
 1017  * the socket if we hold the only reference.
 1018  */
 1019 struct tcpcb *
 1020 tcp_close(struct tcpcb *tp)
 1021 {
 1022         struct inpcb *inp = tp->t_inpcb;
 1023         struct socket *so;
 1024 
 1025         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 1026         INP_WLOCK_ASSERT(inp);
 1027 
 1028 #ifdef TCP_OFFLOAD
 1029         if (tp->t_state == TCPS_LISTEN)
 1030                 tcp_offload_listen_stop(tp);
 1031 #endif
 1032         in_pcbdrop(inp);
 1033         TCPSTAT_INC(tcps_closed);
 1034         KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
 1035         so = inp->inp_socket;
 1036         soisdisconnected(so);
 1037         if (inp->inp_flags & INP_SOCKREF) {
 1038                 KASSERT(so->so_state & SS_PROTOREF,
 1039                     ("tcp_close: !SS_PROTOREF"));
 1040                 inp->inp_flags &= ~INP_SOCKREF;
 1041                 INP_WUNLOCK(inp);
 1042                 ACCEPT_LOCK();
 1043                 SOCK_LOCK(so);
 1044                 so->so_state &= ~SS_PROTOREF;
 1045                 sofree(so);
 1046                 return (NULL);
 1047         }
 1048         return (tp);
 1049 }
 1050 
 1051 void
 1052 tcp_drain(void)
 1053 {
 1054         VNET_ITERATOR_DECL(vnet_iter);
 1055 
 1056         if (!do_tcpdrain)
 1057                 return;
 1058 
 1059         VNET_LIST_RLOCK_NOSLEEP();
 1060         VNET_FOREACH(vnet_iter) {
 1061                 CURVNET_SET(vnet_iter);
 1062                 struct inpcb *inpb;
 1063                 struct tcpcb *tcpb;
 1064 
 1065         /*
 1066          * Walk the tcpbs, if existing, and flush the reassembly queue,
 1067          * if there is one...
 1068          * XXX: The "Net/3" implementation doesn't imply that the TCP
 1069          *      reassembly queue should be flushed, but in a situation
 1070          *      where we're really low on mbufs, this is potentially
 1071          *      useful.
 1072          */
 1073                 INP_INFO_RLOCK(&V_tcbinfo);
 1074                 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
 1075                         if (inpb->inp_flags & INP_TIMEWAIT)
 1076                                 continue;
 1077                         INP_WLOCK(inpb);
 1078                         if ((tcpb = intotcpcb(inpb)) != NULL) {
 1079                                 tcp_reass_flush(tcpb);
 1080                                 tcp_clean_sackreport(tcpb);
 1081                         }
 1082                         INP_WUNLOCK(inpb);
 1083                 }
 1084                 INP_INFO_RUNLOCK(&V_tcbinfo);
 1085                 CURVNET_RESTORE();
 1086         }
 1087         VNET_LIST_RUNLOCK_NOSLEEP();
 1088 }
 1089 
 1090 /*
 1091  * Notify a tcp user of an asynchronous error;
 1092  * store error as soft error, but wake up user
 1093  * (for now, won't do anything until can select for soft error).
 1094  *
 1095  * Do not wake up user since there currently is no mechanism for
 1096  * reporting soft errors (yet - a kqueue filter may be added).
 1097  */
 1098 static struct inpcb *
 1099 tcp_notify(struct inpcb *inp, int error)
 1100 {
 1101         struct tcpcb *tp;
 1102 
 1103         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 1104         INP_WLOCK_ASSERT(inp);
 1105 
 1106         if ((inp->inp_flags & INP_TIMEWAIT) ||
 1107             (inp->inp_flags & INP_DROPPED))
 1108                 return (inp);
 1109 
 1110         tp = intotcpcb(inp);
 1111         KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
 1112 
 1113         /*
 1114          * Ignore some errors if we are hooked up.
 1115          * If connection hasn't completed, has retransmitted several times,
 1116          * and receives a second error, give up now.  This is better
 1117          * than waiting a long time to establish a connection that
 1118          * can never complete.
 1119          */
 1120         if (tp->t_state == TCPS_ESTABLISHED &&
 1121             (error == EHOSTUNREACH || error == ENETUNREACH ||
 1122              error == EHOSTDOWN)) {
 1123                 return (inp);
 1124         } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
 1125             tp->t_softerror) {
 1126                 tp = tcp_drop(tp, error);
 1127                 if (tp != NULL)
 1128                         return (inp);
 1129                 else
 1130                         return (NULL);
 1131         } else {
 1132                 tp->t_softerror = error;
 1133                 return (inp);
 1134         }
 1135 #if 0
 1136         wakeup( &so->so_timeo);
 1137         sorwakeup(so);
 1138         sowwakeup(so);
 1139 #endif
 1140 }
 1141 
 1142 static int
 1143 tcp_pcblist(SYSCTL_HANDLER_ARGS)
 1144 {
 1145         int error, i, m, n, pcb_count;
 1146         struct inpcb *inp, **inp_list;
 1147         inp_gen_t gencnt;
 1148         struct xinpgen xig;
 1149 
 1150         /*
 1151          * The process of preparing the TCB list is too time-consuming and
 1152          * resource-intensive to repeat twice on every request.
 1153          */
 1154         if (req->oldptr == NULL) {
 1155                 n = V_tcbinfo.ipi_count + syncache_pcbcount();
 1156                 n += imax(n / 8, 10);
 1157                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
 1158                 return (0);
 1159         }
 1160 
 1161         if (req->newptr != NULL)
 1162                 return (EPERM);
 1163 
 1164         /*
 1165          * OK, now we're committed to doing something.
 1166          */
 1167         INP_INFO_RLOCK(&V_tcbinfo);
 1168         gencnt = V_tcbinfo.ipi_gencnt;
 1169         n = V_tcbinfo.ipi_count;
 1170         INP_INFO_RUNLOCK(&V_tcbinfo);
 1171 
 1172         m = syncache_pcbcount();
 1173 
 1174         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
 1175                 + (n + m) * sizeof(struct xtcpcb));
 1176         if (error != 0)
 1177                 return (error);
 1178 
 1179         xig.xig_len = sizeof xig;
 1180         xig.xig_count = n + m;
 1181         xig.xig_gen = gencnt;
 1182         xig.xig_sogen = so_gencnt;
 1183         error = SYSCTL_OUT(req, &xig, sizeof xig);
 1184         if (error)
 1185                 return (error);
 1186 
 1187         error = syncache_pcblist(req, m, &pcb_count);
 1188         if (error)
 1189                 return (error);
 1190 
 1191         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
 1192         if (inp_list == NULL)
 1193                 return (ENOMEM);
 1194 
 1195         INP_INFO_RLOCK(&V_tcbinfo);
 1196         for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
 1197             inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
 1198                 INP_WLOCK(inp);
 1199                 if (inp->inp_gencnt <= gencnt) {
 1200                         /*
 1201                          * XXX: This use of cr_cansee(), introduced with
 1202                          * TCP state changes, is not quite right, but for
 1203                          * now, better than nothing.
 1204                          */
 1205                         if (inp->inp_flags & INP_TIMEWAIT) {
 1206                                 if (intotw(inp) != NULL)
 1207                                         error = cr_cansee(req->td->td_ucred,
 1208                                             intotw(inp)->tw_cred);
 1209                                 else
 1210                                         error = EINVAL; /* Skip this inp. */
 1211                         } else
 1212                                 error = cr_canseeinpcb(req->td->td_ucred, inp);
 1213                         if (error == 0) {
 1214                                 in_pcbref(inp);
 1215                                 inp_list[i++] = inp;
 1216                         }
 1217                 }
 1218                 INP_WUNLOCK(inp);
 1219         }
 1220         INP_INFO_RUNLOCK(&V_tcbinfo);
 1221         n = i;
 1222 
 1223         error = 0;
 1224         for (i = 0; i < n; i++) {
 1225                 inp = inp_list[i];
 1226                 INP_RLOCK(inp);
 1227                 if (inp->inp_gencnt <= gencnt) {
 1228                         struct xtcpcb xt;
 1229                         void *inp_ppcb;
 1230 
 1231                         bzero(&xt, sizeof(xt));
 1232                         xt.xt_len = sizeof xt;
 1233                         /* XXX should avoid extra copy */
 1234                         bcopy(inp, &xt.xt_inp, sizeof *inp);
 1235                         inp_ppcb = inp->inp_ppcb;
 1236                         if (inp_ppcb == NULL)
 1237                                 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
 1238                         else if (inp->inp_flags & INP_TIMEWAIT) {
 1239                                 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
 1240                                 xt.xt_tp.t_state = TCPS_TIME_WAIT;
 1241                         } else {
 1242                                 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
 1243                                 if (xt.xt_tp.t_timers)
 1244                                         tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
 1245                         }
 1246                         if (inp->inp_socket != NULL)
 1247                                 sotoxsocket(inp->inp_socket, &xt.xt_socket);
 1248                         else {
 1249                                 bzero(&xt.xt_socket, sizeof xt.xt_socket);
 1250                                 xt.xt_socket.xso_protocol = IPPROTO_TCP;
 1251                         }
 1252                         xt.xt_inp.inp_gencnt = inp->inp_gencnt;
 1253                         INP_RUNLOCK(inp);
 1254                         error = SYSCTL_OUT(req, &xt, sizeof xt);
 1255                 } else
 1256                         INP_RUNLOCK(inp);
 1257         }
 1258         INP_INFO_WLOCK(&V_tcbinfo);
 1259         for (i = 0; i < n; i++) {
 1260                 inp = inp_list[i];
 1261                 INP_RLOCK(inp);
 1262                 if (!in_pcbrele_rlocked(inp))
 1263                         INP_RUNLOCK(inp);
 1264         }
 1265         INP_INFO_WUNLOCK(&V_tcbinfo);
 1266 
 1267         if (!error) {
 1268                 /*
 1269                  * Give the user an updated idea of our state.
 1270                  * If the generation differs from what we told
 1271                  * her before, she knows that something happened
 1272                  * while we were processing this request, and it
 1273                  * might be necessary to retry.
 1274                  */
 1275                 INP_INFO_RLOCK(&V_tcbinfo);
 1276                 xig.xig_gen = V_tcbinfo.ipi_gencnt;
 1277                 xig.xig_sogen = so_gencnt;
 1278                 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
 1279                 INP_INFO_RUNLOCK(&V_tcbinfo);
 1280                 error = SYSCTL_OUT(req, &xig, sizeof xig);
 1281         }
 1282         free(inp_list, M_TEMP);
 1283         return (error);
 1284 }
 1285 
 1286 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
 1287     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
 1288     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
 1289 
 1290 #ifdef INET
 1291 static int
 1292 tcp_getcred(SYSCTL_HANDLER_ARGS)
 1293 {
 1294         struct xucred xuc;
 1295         struct sockaddr_in addrs[2];
 1296         struct inpcb *inp;
 1297         int error;
 1298 
 1299         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 1300         if (error)
 1301                 return (error);
 1302         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 1303         if (error)
 1304                 return (error);
 1305         inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
 1306             addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
 1307         if (inp != NULL) {
 1308                 if (inp->inp_socket == NULL)
 1309                         error = ENOENT;
 1310                 if (error == 0)
 1311                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 1312                 if (error == 0)
 1313                         cru2x(inp->inp_cred, &xuc);
 1314                 INP_RUNLOCK(inp);
 1315         } else
 1316                 error = ENOENT;
 1317         if (error == 0)
 1318                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 1319         return (error);
 1320 }
 1321 
 1322 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
 1323     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
 1324     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
 1325 #endif /* INET */
 1326 
 1327 #ifdef INET6
 1328 static int
 1329 tcp6_getcred(SYSCTL_HANDLER_ARGS)
 1330 {
 1331         struct xucred xuc;
 1332         struct sockaddr_in6 addrs[2];
 1333         struct inpcb *inp;
 1334         int error;
 1335 #ifdef INET
 1336         int mapped = 0;
 1337 #endif
 1338 
 1339         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 1340         if (error)
 1341                 return (error);
 1342         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 1343         if (error)
 1344                 return (error);
 1345         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
 1346             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
 1347                 return (error);
 1348         }
 1349         if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
 1350 #ifdef INET
 1351                 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
 1352                         mapped = 1;
 1353                 else
 1354 #endif
 1355                         return (EINVAL);
 1356         }
 1357 
 1358 #ifdef INET
 1359         if (mapped == 1)
 1360                 inp = in_pcblookup(&V_tcbinfo,
 1361                         *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
 1362                         addrs[1].sin6_port,
 1363                         *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
 1364                         addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
 1365         else
 1366 #endif
 1367                 inp = in6_pcblookup(&V_tcbinfo,
 1368                         &addrs[1].sin6_addr, addrs[1].sin6_port,
 1369                         &addrs[0].sin6_addr, addrs[0].sin6_port,
 1370                         INPLOOKUP_RLOCKPCB, NULL);
 1371         if (inp != NULL) {
 1372                 if (inp->inp_socket == NULL)
 1373                         error = ENOENT;
 1374                 if (error == 0)
 1375                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 1376                 if (error == 0)
 1377                         cru2x(inp->inp_cred, &xuc);
 1378                 INP_RUNLOCK(inp);
 1379         } else
 1380                 error = ENOENT;
 1381         if (error == 0)
 1382                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 1383         return (error);
 1384 }
 1385 
 1386 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
 1387     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
 1388     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
 1389 #endif /* INET6 */
 1390 
 1391 
 1392 #ifdef INET
 1393 void
 1394 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 1395 {
 1396         struct ip *ip = vip;
 1397         struct tcphdr *th;
 1398         struct in_addr faddr;
 1399         struct inpcb *inp;
 1400         struct tcpcb *tp;
 1401         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
 1402         struct icmp *icp;
 1403         struct in_conninfo inc;
 1404         tcp_seq icmp_tcp_seq;
 1405         int mtu;
 1406 
 1407         faddr = ((struct sockaddr_in *)sa)->sin_addr;
 1408         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
 1409                 return;
 1410 
 1411         if (cmd == PRC_MSGSIZE)
 1412                 notify = tcp_mtudisc_notify;
 1413         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
 1414                 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
 1415                 notify = tcp_drop_syn_sent;
 1416         /*
 1417          * Redirects don't need to be handled up here.
 1418          */
 1419         else if (PRC_IS_REDIRECT(cmd))
 1420                 return;
 1421         /*
 1422          * Source quench is depreciated.
 1423          */
 1424         else if (cmd == PRC_QUENCH)
 1425                 return;
 1426         /*
 1427          * Hostdead is ugly because it goes linearly through all PCBs.
 1428          * XXX: We never get this from ICMP, otherwise it makes an
 1429          * excellent DoS attack on machines with many connections.
 1430          */
 1431         else if (cmd == PRC_HOSTDEAD)
 1432                 ip = NULL;
 1433         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
 1434                 return;
 1435         if (ip != NULL) {
 1436                 icp = (struct icmp *)((caddr_t)ip
 1437                                       - offsetof(struct icmp, icmp_ip));
 1438                 th = (struct tcphdr *)((caddr_t)ip
 1439                                        + (ip->ip_hl << 2));
 1440                 INP_INFO_WLOCK(&V_tcbinfo);
 1441                 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
 1442                     ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
 1443                 if (inp != NULL)  {
 1444                         if (!(inp->inp_flags & INP_TIMEWAIT) &&
 1445                             !(inp->inp_flags & INP_DROPPED) &&
 1446                             !(inp->inp_socket == NULL)) {
 1447                                 icmp_tcp_seq = htonl(th->th_seq);
 1448                                 tp = intotcpcb(inp);
 1449                                 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
 1450                                     SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
 1451                                         if (cmd == PRC_MSGSIZE) {
 1452                                             /*
 1453                                              * MTU discovery:
 1454                                              * If we got a needfrag set the MTU
 1455                                              * in the route to the suggested new
 1456                                              * value (if given) and then notify.
 1457                                              */
 1458                                             bzero(&inc, sizeof(inc));
 1459                                             inc.inc_faddr = faddr;
 1460                                             inc.inc_fibnum =
 1461                                                 inp->inp_inc.inc_fibnum;
 1462 
 1463                                             mtu = ntohs(icp->icmp_nextmtu);
 1464                                             /*
 1465                                              * If no alternative MTU was
 1466                                              * proposed, try the next smaller
 1467                                              * one.
 1468                                              */
 1469                                             if (!mtu)
 1470                                                 mtu = ip_next_mtu(
 1471                                                  ntohs(ip->ip_len), 1);
 1472                                             if (mtu < V_tcp_minmss
 1473                                                  + sizeof(struct tcpiphdr))
 1474                                                 mtu = V_tcp_minmss
 1475                                                  + sizeof(struct tcpiphdr);
 1476                                             /*
 1477                                              * Only cache the MTU if it
 1478                                              * is smaller than the interface
 1479                                              * or route MTU.  tcp_mtudisc()
 1480                                              * will do right thing by itself.
 1481                                              */
 1482                                             if (mtu <= tcp_maxmtu(&inc, NULL))
 1483                                                 tcp_hc_updatemtu(&inc, mtu);
 1484                                             tcp_mtudisc(inp, mtu);
 1485                                         } else
 1486                                                 inp = (*notify)(inp,
 1487                                                     inetctlerrmap[cmd]);
 1488                                 }
 1489                         }
 1490                         if (inp != NULL)
 1491                                 INP_WUNLOCK(inp);
 1492                 } else {
 1493                         bzero(&inc, sizeof(inc));
 1494                         inc.inc_fport = th->th_dport;
 1495                         inc.inc_lport = th->th_sport;
 1496                         inc.inc_faddr = faddr;
 1497                         inc.inc_laddr = ip->ip_src;
 1498                         syncache_unreach(&inc, th);
 1499                 }
 1500                 INP_INFO_WUNLOCK(&V_tcbinfo);
 1501         } else
 1502                 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
 1503 }
 1504 #endif /* INET */
 1505 
 1506 #ifdef INET6
 1507 void
 1508 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 1509 {
 1510         struct tcphdr th;
 1511         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
 1512         struct ip6_hdr *ip6;
 1513         struct mbuf *m;
 1514         struct ip6ctlparam *ip6cp = NULL;
 1515         const struct sockaddr_in6 *sa6_src = NULL;
 1516         int off;
 1517         struct tcp_portonly {
 1518                 u_int16_t th_sport;
 1519                 u_int16_t th_dport;
 1520         } *thp;
 1521 
 1522         if (sa->sa_family != AF_INET6 ||
 1523             sa->sa_len != sizeof(struct sockaddr_in6))
 1524                 return;
 1525 
 1526         if (cmd == PRC_MSGSIZE)
 1527                 notify = tcp_mtudisc_notify;
 1528         else if (!PRC_IS_REDIRECT(cmd) &&
 1529                  ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
 1530                 return;
 1531         /* Source quench is depreciated. */
 1532         else if (cmd == PRC_QUENCH)
 1533                 return;
 1534 
 1535         /* if the parameter is from icmp6, decode it. */
 1536         if (d != NULL) {
 1537                 ip6cp = (struct ip6ctlparam *)d;
 1538                 m = ip6cp->ip6c_m;
 1539                 ip6 = ip6cp->ip6c_ip6;
 1540                 off = ip6cp->ip6c_off;
 1541                 sa6_src = ip6cp->ip6c_src;
 1542         } else {
 1543                 m = NULL;
 1544                 ip6 = NULL;
 1545                 off = 0;        /* fool gcc */
 1546                 sa6_src = &sa6_any;
 1547         }
 1548 
 1549         if (ip6 != NULL) {
 1550                 struct in_conninfo inc;
 1551                 /*
 1552                  * XXX: We assume that when IPV6 is non NULL,
 1553                  * M and OFF are valid.
 1554                  */
 1555 
 1556                 /* check if we can safely examine src and dst ports */
 1557                 if (m->m_pkthdr.len < off + sizeof(*thp))
 1558                         return;
 1559 
 1560                 bzero(&th, sizeof(th));
 1561                 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
 1562 
 1563                 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
 1564                     (struct sockaddr *)ip6cp->ip6c_src,
 1565                     th.th_sport, cmd, NULL, notify);
 1566 
 1567                 bzero(&inc, sizeof(inc));
 1568                 inc.inc_fport = th.th_dport;
 1569                 inc.inc_lport = th.th_sport;
 1570                 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
 1571                 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
 1572                 inc.inc_flags |= INC_ISIPV6;
 1573                 INP_INFO_WLOCK(&V_tcbinfo);
 1574                 syncache_unreach(&inc, &th);
 1575                 INP_INFO_WUNLOCK(&V_tcbinfo);
 1576         } else
 1577                 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
 1578                               0, cmd, NULL, notify);
 1579 }
 1580 #endif /* INET6 */
 1581 
 1582 
 1583 /*
 1584  * Following is where TCP initial sequence number generation occurs.
 1585  *
 1586  * There are two places where we must use initial sequence numbers:
 1587  * 1.  In SYN-ACK packets.
 1588  * 2.  In SYN packets.
 1589  *
 1590  * All ISNs for SYN-ACK packets are generated by the syncache.  See
 1591  * tcp_syncache.c for details.
 1592  *
 1593  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
 1594  * depends on this property.  In addition, these ISNs should be
 1595  * unguessable so as to prevent connection hijacking.  To satisfy
 1596  * the requirements of this situation, the algorithm outlined in
 1597  * RFC 1948 is used, with only small modifications.
 1598  *
 1599  * Implementation details:
 1600  *
 1601  * Time is based off the system timer, and is corrected so that it
 1602  * increases by one megabyte per second.  This allows for proper
 1603  * recycling on high speed LANs while still leaving over an hour
 1604  * before rollover.
 1605  *
 1606  * As reading the *exact* system time is too expensive to be done
 1607  * whenever setting up a TCP connection, we increment the time
 1608  * offset in two ways.  First, a small random positive increment
 1609  * is added to isn_offset for each connection that is set up.
 1610  * Second, the function tcp_isn_tick fires once per clock tick
 1611  * and increments isn_offset as necessary so that sequence numbers
 1612  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
 1613  * random positive increments serve only to ensure that the same
 1614  * exact sequence number is never sent out twice (as could otherwise
 1615  * happen when a port is recycled in less than the system tick
 1616  * interval.)
 1617  *
 1618  * net.inet.tcp.isn_reseed_interval controls the number of seconds
 1619  * between seeding of isn_secret.  This is normally set to zero,
 1620  * as reseeding should not be necessary.
 1621  *
 1622  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
 1623  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
 1624  * general, this means holding an exclusive (write) lock.
 1625  */
 1626 
 1627 #define ISN_BYTES_PER_SECOND 1048576
 1628 #define ISN_STATIC_INCREMENT 4096
 1629 #define ISN_RANDOM_INCREMENT (4096 - 1)
 1630 
 1631 static VNET_DEFINE(u_char, isn_secret[32]);
 1632 static VNET_DEFINE(int, isn_last);
 1633 static VNET_DEFINE(int, isn_last_reseed);
 1634 static VNET_DEFINE(u_int32_t, isn_offset);
 1635 static VNET_DEFINE(u_int32_t, isn_offset_old);
 1636 
 1637 #define V_isn_secret                    VNET(isn_secret)
 1638 #define V_isn_last                      VNET(isn_last)
 1639 #define V_isn_last_reseed               VNET(isn_last_reseed)
 1640 #define V_isn_offset                    VNET(isn_offset)
 1641 #define V_isn_offset_old                VNET(isn_offset_old)
 1642 
 1643 tcp_seq
 1644 tcp_new_isn(struct tcpcb *tp)
 1645 {
 1646         MD5_CTX isn_ctx;
 1647         u_int32_t md5_buffer[4];
 1648         tcp_seq new_isn;
 1649         u_int32_t projected_offset;
 1650 
 1651         INP_WLOCK_ASSERT(tp->t_inpcb);
 1652 
 1653         ISN_LOCK();
 1654         /* Seed if this is the first use, reseed if requested. */
 1655         if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
 1656              (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
 1657                 < (u_int)ticks))) {
 1658                 read_random(&V_isn_secret, sizeof(V_isn_secret));
 1659                 V_isn_last_reseed = ticks;
 1660         }
 1661 
 1662         /* Compute the md5 hash and return the ISN. */
 1663         MD5Init(&isn_ctx);
 1664         MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
 1665         MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
 1666 #ifdef INET6
 1667         if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
 1668                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
 1669                           sizeof(struct in6_addr));
 1670                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
 1671                           sizeof(struct in6_addr));
 1672         } else
 1673 #endif
 1674         {
 1675                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
 1676                           sizeof(struct in_addr));
 1677                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
 1678                           sizeof(struct in_addr));
 1679         }
 1680         MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
 1681         MD5Final((u_char *) &md5_buffer, &isn_ctx);
 1682         new_isn = (tcp_seq) md5_buffer[0];
 1683         V_isn_offset += ISN_STATIC_INCREMENT +
 1684                 (arc4random() & ISN_RANDOM_INCREMENT);
 1685         if (ticks != V_isn_last) {
 1686                 projected_offset = V_isn_offset_old +
 1687                     ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
 1688                 if (SEQ_GT(projected_offset, V_isn_offset))
 1689                         V_isn_offset = projected_offset;
 1690                 V_isn_offset_old = V_isn_offset;
 1691                 V_isn_last = ticks;
 1692         }
 1693         new_isn += V_isn_offset;
 1694         ISN_UNLOCK();
 1695         return (new_isn);
 1696 }
 1697 
 1698 /*
 1699  * When a specific ICMP unreachable message is received and the
 1700  * connection state is SYN-SENT, drop the connection.  This behavior
 1701  * is controlled by the icmp_may_rst sysctl.
 1702  */
 1703 struct inpcb *
 1704 tcp_drop_syn_sent(struct inpcb *inp, int errno)
 1705 {
 1706         struct tcpcb *tp;
 1707 
 1708         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 1709         INP_WLOCK_ASSERT(inp);
 1710 
 1711         if ((inp->inp_flags & INP_TIMEWAIT) ||
 1712             (inp->inp_flags & INP_DROPPED))
 1713                 return (inp);
 1714 
 1715         tp = intotcpcb(inp);
 1716         if (tp->t_state != TCPS_SYN_SENT)
 1717                 return (inp);
 1718 
 1719         tp = tcp_drop(tp, errno);
 1720         if (tp != NULL)
 1721                 return (inp);
 1722         else
 1723                 return (NULL);
 1724 }
 1725 
 1726 /*
 1727  * When `need fragmentation' ICMP is received, update our idea of the MSS
 1728  * based on the new value. Also nudge TCP to send something, since we
 1729  * know the packet we just sent was dropped.
 1730  * This duplicates some code in the tcp_mss() function in tcp_input.c.
 1731  */
 1732 static struct inpcb *
 1733 tcp_mtudisc_notify(struct inpcb *inp, int error)
 1734 {
 1735 
 1736         return (tcp_mtudisc(inp, -1));
 1737 }
 1738 
 1739 struct inpcb *
 1740 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
 1741 {
 1742         struct tcpcb *tp;
 1743         struct socket *so;
 1744 
 1745         INP_WLOCK_ASSERT(inp);
 1746         if ((inp->inp_flags & INP_TIMEWAIT) ||
 1747             (inp->inp_flags & INP_DROPPED))
 1748                 return (inp);
 1749 
 1750         tp = intotcpcb(inp);
 1751         KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
 1752 
 1753         tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
 1754   
 1755         so = inp->inp_socket;
 1756         SOCKBUF_LOCK(&so->so_snd);
 1757         /* If the mss is larger than the socket buffer, decrease the mss. */
 1758         if (so->so_snd.sb_hiwat < tp->t_maxseg)
 1759                 tp->t_maxseg = so->so_snd.sb_hiwat;
 1760         SOCKBUF_UNLOCK(&so->so_snd);
 1761 
 1762         TCPSTAT_INC(tcps_mturesent);
 1763         tp->t_rtttime = 0;
 1764         tp->snd_nxt = tp->snd_una;
 1765         tcp_free_sackholes(tp);
 1766         tp->snd_recover = tp->snd_max;
 1767         if (tp->t_flags & TF_SACK_PERMIT)
 1768                 EXIT_FASTRECOVERY(tp->t_flags);
 1769         tcp_output(tp);
 1770         return (inp);
 1771 }
 1772 
 1773 #ifdef INET
 1774 /*
 1775  * Look-up the routing entry to the peer of this inpcb.  If no route
 1776  * is found and it cannot be allocated, then return 0.  This routine
 1777  * is called by TCP routines that access the rmx structure and by
 1778  * tcp_mss_update to get the peer/interface MTU.
 1779  */
 1780 u_long
 1781 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
 1782 {
 1783         struct route sro;
 1784         struct sockaddr_in *dst;
 1785         struct ifnet *ifp;
 1786         u_long maxmtu = 0;
 1787 
 1788         KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
 1789 
 1790         bzero(&sro, sizeof(sro));
 1791         if (inc->inc_faddr.s_addr != INADDR_ANY) {
 1792                 dst = (struct sockaddr_in *)&sro.ro_dst;
 1793                 dst->sin_family = AF_INET;
 1794                 dst->sin_len = sizeof(*dst);
 1795                 dst->sin_addr = inc->inc_faddr;
 1796                 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
 1797         }
 1798         if (sro.ro_rt != NULL) {
 1799                 ifp = sro.ro_rt->rt_ifp;
 1800                 if (sro.ro_rt->rt_mtu == 0)
 1801                         maxmtu = ifp->if_mtu;
 1802                 else
 1803                         maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
 1804 
 1805                 /* Report additional interface capabilities. */
 1806                 if (cap != NULL) {
 1807                         if (ifp->if_capenable & IFCAP_TSO4 &&
 1808                             ifp->if_hwassist & CSUM_TSO) {
 1809                                 cap->ifcap |= CSUM_TSO;
 1810                                 cap->tsomax = ifp->if_hw_tsomax;
 1811                         }
 1812                 }
 1813                 RTFREE(sro.ro_rt);
 1814         }
 1815         return (maxmtu);
 1816 }
 1817 #endif /* INET */
 1818 
 1819 #ifdef INET6
 1820 u_long
 1821 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
 1822 {
 1823         struct route_in6 sro6;
 1824         struct ifnet *ifp;
 1825         u_long maxmtu = 0;
 1826 
 1827         KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
 1828 
 1829         bzero(&sro6, sizeof(sro6));
 1830         if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
 1831                 sro6.ro_dst.sin6_family = AF_INET6;
 1832                 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
 1833                 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
 1834                 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
 1835         }
 1836         if (sro6.ro_rt != NULL) {
 1837                 ifp = sro6.ro_rt->rt_ifp;
 1838                 if (sro6.ro_rt->rt_mtu == 0)
 1839                         maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
 1840                 else
 1841                         maxmtu = min(sro6.ro_rt->rt_mtu,
 1842                                      IN6_LINKMTU(sro6.ro_rt->rt_ifp));
 1843 
 1844                 /* Report additional interface capabilities. */
 1845                 if (cap != NULL) {
 1846                         if (ifp->if_capenable & IFCAP_TSO6 &&
 1847                             ifp->if_hwassist & CSUM_TSO) {
 1848                                 cap->ifcap |= CSUM_TSO;
 1849                                 cap->tsomax = ifp->if_hw_tsomax;
 1850                         }
 1851                 }
 1852                 RTFREE(sro6.ro_rt);
 1853         }
 1854 
 1855         return (maxmtu);
 1856 }
 1857 #endif /* INET6 */
 1858 
 1859 #ifdef IPSEC
 1860 /* compute ESP/AH header size for TCP, including outer IP header. */
 1861 size_t
 1862 ipsec_hdrsiz_tcp(struct tcpcb *tp)
 1863 {
 1864         struct inpcb *inp;
 1865         struct mbuf *m;
 1866         size_t hdrsiz;
 1867         struct ip *ip;
 1868 #ifdef INET6
 1869         struct ip6_hdr *ip6;
 1870 #endif
 1871         struct tcphdr *th;
 1872 
 1873         if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
 1874                 return (0);
 1875         m = m_gethdr(M_NOWAIT, MT_DATA);
 1876         if (!m)
 1877                 return (0);
 1878 
 1879 #ifdef INET6
 1880         if ((inp->inp_vflag & INP_IPV6) != 0) {
 1881                 ip6 = mtod(m, struct ip6_hdr *);
 1882                 th = (struct tcphdr *)(ip6 + 1);
 1883                 m->m_pkthdr.len = m->m_len =
 1884                         sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
 1885                 tcpip_fillheaders(inp, ip6, th);
 1886                 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
 1887         } else
 1888 #endif /* INET6 */
 1889         {
 1890                 ip = mtod(m, struct ip *);
 1891                 th = (struct tcphdr *)(ip + 1);
 1892                 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
 1893                 tcpip_fillheaders(inp, ip, th);
 1894                 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
 1895         }
 1896 
 1897         m_free(m);
 1898         return (hdrsiz);
 1899 }
 1900 #endif /* IPSEC */
 1901 
 1902 #ifdef TCP_SIGNATURE
 1903 /*
 1904  * Callback function invoked by m_apply() to digest TCP segment data
 1905  * contained within an mbuf chain.
 1906  */
 1907 static int
 1908 tcp_signature_apply(void *fstate, void *data, u_int len)
 1909 {
 1910 
 1911         MD5Update(fstate, (u_char *)data, len);
 1912         return (0);
 1913 }
 1914 
 1915 /*
 1916  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
 1917  *
 1918  * Parameters:
 1919  * m            pointer to head of mbuf chain
 1920  * _unused      
 1921  * len          length of TCP segment data, excluding options
 1922  * optlen       length of TCP segment options
 1923  * buf          pointer to storage for computed MD5 digest
 1924  * direction    direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
 1925  *
 1926  * We do this over ip, tcphdr, segment data, and the key in the SADB.
 1927  * When called from tcp_input(), we can be sure that th_sum has been
 1928  * zeroed out and verified already.
 1929  *
 1930  * Return 0 if successful, otherwise return -1.
 1931  *
 1932  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
 1933  * search with the destination IP address, and a 'magic SPI' to be
 1934  * determined by the application. This is hardcoded elsewhere to 1179
 1935  * right now. Another branch of this code exists which uses the SPD to
 1936  * specify per-application flows but it is unstable.
 1937  */
 1938 int
 1939 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
 1940     u_char *buf, u_int direction)
 1941 {
 1942         union sockaddr_union dst;
 1943 #ifdef INET
 1944         struct ippseudo ippseudo;
 1945 #endif
 1946         MD5_CTX ctx;
 1947         int doff;
 1948         struct ip *ip;
 1949 #ifdef INET
 1950         struct ipovly *ipovly;
 1951 #endif
 1952         struct secasvar *sav;
 1953         struct tcphdr *th;
 1954 #ifdef INET6
 1955         struct ip6_hdr *ip6;
 1956         struct in6_addr in6;
 1957         char ip6buf[INET6_ADDRSTRLEN];
 1958         uint32_t plen;
 1959         uint16_t nhdr;
 1960 #endif
 1961         u_short savecsum;
 1962 
 1963         KASSERT(m != NULL, ("NULL mbuf chain"));
 1964         KASSERT(buf != NULL, ("NULL signature pointer"));
 1965 
 1966         /* Extract the destination from the IP header in the mbuf. */
 1967         bzero(&dst, sizeof(union sockaddr_union));
 1968         ip = mtod(m, struct ip *);
 1969 #ifdef INET6
 1970         ip6 = NULL;     /* Make the compiler happy. */
 1971 #endif
 1972         switch (ip->ip_v) {
 1973 #ifdef INET
 1974         case IPVERSION:
 1975                 dst.sa.sa_len = sizeof(struct sockaddr_in);
 1976                 dst.sa.sa_family = AF_INET;
 1977                 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
 1978                     ip->ip_src : ip->ip_dst;
 1979                 break;
 1980 #endif
 1981 #ifdef INET6
 1982         case (IPV6_VERSION >> 4):
 1983                 ip6 = mtod(m, struct ip6_hdr *);
 1984                 dst.sa.sa_len = sizeof(struct sockaddr_in6);
 1985                 dst.sa.sa_family = AF_INET6;
 1986                 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
 1987                     ip6->ip6_src : ip6->ip6_dst;
 1988                 break;
 1989 #endif
 1990         default:
 1991                 return (EINVAL);
 1992                 /* NOTREACHED */
 1993                 break;
 1994         }
 1995 
 1996         /* Look up an SADB entry which matches the address of the peer. */
 1997         sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
 1998         if (sav == NULL) {
 1999                 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
 2000                     (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
 2001 #ifdef INET6
 2002                         (ip->ip_v == (IPV6_VERSION >> 4)) ?
 2003                             ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
 2004 #endif
 2005                         "(unsupported)"));
 2006                 return (EINVAL);
 2007         }
 2008 
 2009         MD5Init(&ctx);
 2010         /*
 2011          * Step 1: Update MD5 hash with IP(v6) pseudo-header.
 2012          *
 2013          * XXX The ippseudo header MUST be digested in network byte order,
 2014          * or else we'll fail the regression test. Assume all fields we've
 2015          * been doing arithmetic on have been in host byte order.
 2016          * XXX One cannot depend on ipovly->ih_len here. When called from
 2017          * tcp_output(), the underlying ip_len member has not yet been set.
 2018          */
 2019         switch (ip->ip_v) {
 2020 #ifdef INET
 2021         case IPVERSION:
 2022                 ipovly = (struct ipovly *)ip;
 2023                 ippseudo.ippseudo_src = ipovly->ih_src;
 2024                 ippseudo.ippseudo_dst = ipovly->ih_dst;
 2025                 ippseudo.ippseudo_pad = 0;
 2026                 ippseudo.ippseudo_p = IPPROTO_TCP;
 2027                 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
 2028                     optlen);
 2029                 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
 2030 
 2031                 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
 2032                 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
 2033                 break;
 2034 #endif
 2035 #ifdef INET6
 2036         /*
 2037          * RFC 2385, 2.0  Proposal
 2038          * For IPv6, the pseudo-header is as described in RFC 2460, namely the
 2039          * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
 2040          * extended next header value (to form 32 bits), and 32-bit segment
 2041          * length.
 2042          * Note: Upper-Layer Packet Length comes before Next Header.
 2043          */
 2044         case (IPV6_VERSION >> 4):
 2045                 in6 = ip6->ip6_src;
 2046                 in6_clearscope(&in6);
 2047                 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
 2048                 in6 = ip6->ip6_dst;
 2049                 in6_clearscope(&in6);
 2050                 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
 2051                 plen = htonl(len + sizeof(struct tcphdr) + optlen);
 2052                 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
 2053                 nhdr = 0;
 2054                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
 2055                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
 2056                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
 2057                 nhdr = IPPROTO_TCP;
 2058                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
 2059 
 2060                 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
 2061                 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
 2062                 break;
 2063 #endif
 2064         default:
 2065                 return (EINVAL);
 2066                 /* NOTREACHED */
 2067                 break;
 2068         }
 2069 
 2070 
 2071         /*
 2072          * Step 2: Update MD5 hash with TCP header, excluding options.
 2073          * The TCP checksum must be set to zero.
 2074          */
 2075         savecsum = th->th_sum;
 2076         th->th_sum = 0;
 2077         MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
 2078         th->th_sum = savecsum;
 2079 
 2080         /*
 2081          * Step 3: Update MD5 hash with TCP segment data.
 2082          *         Use m_apply() to avoid an early m_pullup().
 2083          */
 2084         if (len > 0)
 2085                 m_apply(m, doff, len, tcp_signature_apply, &ctx);
 2086 
 2087         /*
 2088          * Step 4: Update MD5 hash with shared secret.
 2089          */
 2090         MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
 2091         MD5Final(buf, &ctx);
 2092 
 2093         key_sa_recordxfer(sav, m);
 2094         KEY_FREESAV(&sav);
 2095         return (0);
 2096 }
 2097 
 2098 /*
 2099  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
 2100  *
 2101  * Parameters:
 2102  * m            pointer to head of mbuf chain
 2103  * len          length of TCP segment data, excluding options
 2104  * optlen       length of TCP segment options
 2105  * buf          pointer to storage for computed MD5 digest
 2106  * direction    direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
 2107  *
 2108  * Return 1 if successful, otherwise return 0.
 2109  */
 2110 int
 2111 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
 2112     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
 2113 {
 2114         char tmpdigest[TCP_SIGLEN];
 2115 
 2116         if (tcp_sig_checksigs == 0)
 2117                 return (1);
 2118         if ((tcpbflag & TF_SIGNATURE) == 0) {
 2119                 if ((to->to_flags & TOF_SIGNATURE) != 0) {
 2120 
 2121                         /*
 2122                          * If this socket is not expecting signature but
 2123                          * the segment contains signature just fail.
 2124                          */
 2125                         TCPSTAT_INC(tcps_sig_err_sigopt);
 2126                         TCPSTAT_INC(tcps_sig_rcvbadsig);
 2127                         return (0);
 2128                 }
 2129 
 2130                 /* Signature is not expected, and not present in segment. */
 2131                 return (1);
 2132         }
 2133 
 2134         /*
 2135          * If this socket is expecting signature but the segment does not
 2136          * contain any just fail.
 2137          */
 2138         if ((to->to_flags & TOF_SIGNATURE) == 0) {
 2139                 TCPSTAT_INC(tcps_sig_err_nosigopt);
 2140                 TCPSTAT_INC(tcps_sig_rcvbadsig);
 2141                 return (0);
 2142         }
 2143         if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
 2144             IPSEC_DIR_INBOUND) == -1) {
 2145                 TCPSTAT_INC(tcps_sig_err_buildsig);
 2146                 TCPSTAT_INC(tcps_sig_rcvbadsig);
 2147                 return (0);
 2148         }
 2149         
 2150         if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
 2151                 TCPSTAT_INC(tcps_sig_rcvbadsig);
 2152                 return (0);
 2153         }
 2154         TCPSTAT_INC(tcps_sig_rcvgoodsig);
 2155         return (1);
 2156 }
 2157 #endif /* TCP_SIGNATURE */
 2158 
 2159 static int
 2160 sysctl_drop(SYSCTL_HANDLER_ARGS)
 2161 {
 2162         /* addrs[0] is a foreign socket, addrs[1] is a local one. */
 2163         struct sockaddr_storage addrs[2];
 2164         struct inpcb *inp;
 2165         struct tcpcb *tp;
 2166         struct tcptw *tw;
 2167         struct sockaddr_in *fin, *lin;
 2168 #ifdef INET6
 2169         struct sockaddr_in6 *fin6, *lin6;
 2170 #endif
 2171         int error;
 2172 
 2173         inp = NULL;
 2174         fin = lin = NULL;
 2175 #ifdef INET6
 2176         fin6 = lin6 = NULL;
 2177 #endif
 2178         error = 0;
 2179 
 2180         if (req->oldptr != NULL || req->oldlen != 0)
 2181                 return (EINVAL);
 2182         if (req->newptr == NULL)
 2183                 return (EPERM);
 2184         if (req->newlen < sizeof(addrs))
 2185                 return (ENOMEM);
 2186         error = SYSCTL_IN(req, &addrs, sizeof(addrs));
 2187         if (error)
 2188                 return (error);
 2189 
 2190         switch (addrs[0].ss_family) {
 2191 #ifdef INET6
 2192         case AF_INET6:
 2193                 fin6 = (struct sockaddr_in6 *)&addrs[0];
 2194                 lin6 = (struct sockaddr_in6 *)&addrs[1];
 2195                 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
 2196                     lin6->sin6_len != sizeof(struct sockaddr_in6))
 2197                         return (EINVAL);
 2198                 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
 2199                         if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
 2200                                 return (EINVAL);
 2201                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
 2202                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
 2203                         fin = (struct sockaddr_in *)&addrs[0];
 2204                         lin = (struct sockaddr_in *)&addrs[1];
 2205                         break;
 2206                 }
 2207                 error = sa6_embedscope(fin6, V_ip6_use_defzone);
 2208                 if (error)
 2209                         return (error);
 2210                 error = sa6_embedscope(lin6, V_ip6_use_defzone);
 2211                 if (error)
 2212                         return (error);
 2213                 break;
 2214 #endif
 2215 #ifdef INET
 2216         case AF_INET:
 2217                 fin = (struct sockaddr_in *)&addrs[0];
 2218                 lin = (struct sockaddr_in *)&addrs[1];
 2219                 if (fin->sin_len != sizeof(struct sockaddr_in) ||
 2220                     lin->sin_len != sizeof(struct sockaddr_in))
 2221                         return (EINVAL);
 2222                 break;
 2223 #endif
 2224         default:
 2225                 return (EINVAL);
 2226         }
 2227         INP_INFO_WLOCK(&V_tcbinfo);
 2228         switch (addrs[0].ss_family) {
 2229 #ifdef INET6
 2230         case AF_INET6:
 2231                 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
 2232                     fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
 2233                     INPLOOKUP_WLOCKPCB, NULL);
 2234                 break;
 2235 #endif
 2236 #ifdef INET
 2237         case AF_INET:
 2238                 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
 2239                     lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
 2240                 break;
 2241 #endif
 2242         }
 2243         if (inp != NULL) {
 2244                 if (inp->inp_flags & INP_TIMEWAIT) {
 2245                         /*
 2246                          * XXXRW: There currently exists a state where an
 2247                          * inpcb is present, but its timewait state has been
 2248                          * discarded.  For now, don't allow dropping of this
 2249                          * type of inpcb.
 2250                          */
 2251                         tw = intotw(inp);
 2252                         if (tw != NULL)
 2253                                 tcp_twclose(tw, 0);
 2254                         else
 2255                                 INP_WUNLOCK(inp);
 2256                 } else if (!(inp->inp_flags & INP_DROPPED) &&
 2257                            !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
 2258                         tp = intotcpcb(inp);
 2259                         tp = tcp_drop(tp, ECONNABORTED);
 2260                         if (tp != NULL)
 2261                                 INP_WUNLOCK(inp);
 2262                 } else
 2263                         INP_WUNLOCK(inp);
 2264         } else
 2265                 error = ESRCH;
 2266         INP_INFO_WUNLOCK(&V_tcbinfo);
 2267         return (error);
 2268 }
 2269 
 2270 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
 2271     CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
 2272     0, sysctl_drop, "", "Drop TCP connection");
 2273 
 2274 /*
 2275  * Generate a standardized TCP log line for use throughout the
 2276  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
 2277  * allow use in the interrupt context.
 2278  *
 2279  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
 2280  * NB: The function may return NULL if memory allocation failed.
 2281  *
 2282  * Due to header inclusion and ordering limitations the struct ip
 2283  * and ip6_hdr pointers have to be passed as void pointers.
 2284  */
 2285 char *
 2286 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
 2287     const void *ip6hdr)
 2288 {
 2289 
 2290         /* Is logging enabled? */
 2291         if (tcp_log_in_vain == 0)
 2292                 return (NULL);
 2293 
 2294         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
 2295 }
 2296 
 2297 char *
 2298 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
 2299     const void *ip6hdr)
 2300 {
 2301 
 2302         /* Is logging enabled? */
 2303         if (tcp_log_debug == 0)
 2304                 return (NULL);
 2305 
 2306         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
 2307 }
 2308 
 2309 static char *
 2310 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
 2311     const void *ip6hdr)
 2312 {
 2313         char *s, *sp;
 2314         size_t size;
 2315         struct ip *ip;
 2316 #ifdef INET6
 2317         const struct ip6_hdr *ip6;
 2318 
 2319         ip6 = (const struct ip6_hdr *)ip6hdr;
 2320 #endif /* INET6 */
 2321         ip = (struct ip *)ip4hdr;
 2322 
 2323         /*
 2324          * The log line looks like this:
 2325          * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
 2326          */
 2327         size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
 2328             sizeof(PRINT_TH_FLAGS) + 1 +
 2329 #ifdef INET6
 2330             2 * INET6_ADDRSTRLEN;
 2331 #else
 2332             2 * INET_ADDRSTRLEN;
 2333 #endif /* INET6 */
 2334 
 2335         s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
 2336         if (s == NULL)
 2337                 return (NULL);
 2338 
 2339         strcat(s, "TCP: [");
 2340         sp = s + strlen(s);
 2341 
 2342         if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
 2343                 inet_ntoa_r(inc->inc_faddr, sp);
 2344                 sp = s + strlen(s);
 2345                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
 2346                 sp = s + strlen(s);
 2347                 inet_ntoa_r(inc->inc_laddr, sp);
 2348                 sp = s + strlen(s);
 2349                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
 2350 #ifdef INET6
 2351         } else if (inc) {
 2352                 ip6_sprintf(sp, &inc->inc6_faddr);
 2353                 sp = s + strlen(s);
 2354                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
 2355                 sp = s + strlen(s);
 2356                 ip6_sprintf(sp, &inc->inc6_laddr);
 2357                 sp = s + strlen(s);
 2358                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
 2359         } else if (ip6 && th) {
 2360                 ip6_sprintf(sp, &ip6->ip6_src);
 2361                 sp = s + strlen(s);
 2362                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
 2363                 sp = s + strlen(s);
 2364                 ip6_sprintf(sp, &ip6->ip6_dst);
 2365                 sp = s + strlen(s);
 2366                 sprintf(sp, "]:%i", ntohs(th->th_dport));
 2367 #endif /* INET6 */
 2368 #ifdef INET
 2369         } else if (ip && th) {
 2370                 inet_ntoa_r(ip->ip_src, sp);
 2371                 sp = s + strlen(s);
 2372                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
 2373                 sp = s + strlen(s);
 2374                 inet_ntoa_r(ip->ip_dst, sp);
 2375                 sp = s + strlen(s);
 2376                 sprintf(sp, "]:%i", ntohs(th->th_dport));
 2377 #endif /* INET */
 2378         } else {
 2379                 free(s, M_TCPLOG);
 2380                 return (NULL);
 2381         }
 2382         sp = s + strlen(s);
 2383         if (th)
 2384                 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
 2385         if (*(s + size - 1) != '\0')
 2386                 panic("%s: string too long", __func__);
 2387         return (s);
 2388 }
 2389 
 2390 /*
 2391  * A subroutine which makes it easy to track TCP state changes with DTrace.
 2392  * This function shouldn't be called for t_state initializations that don't
 2393  * correspond to actual TCP state transitions.
 2394  */
 2395 void
 2396 tcp_state_change(struct tcpcb *tp, int newstate)
 2397 {
 2398 #if defined(KDTRACE_HOOKS)
 2399         int pstate = tp->t_state;
 2400 #endif
 2401 
 2402         tp->t_state = newstate;
 2403         TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
 2404 }

Cache object: 661f0175deb3cd30cb4b974f41470741


[ 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.