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_syncache.c

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 2001 McAfee, Inc.
    3  * Copyright (c) 2006 Andre Oppermann, Internet Business Solutions AG
    4  * All rights reserved.
    5  *
    6  * This software was developed for the FreeBSD Project by Jonathan Lemon
    7  * and McAfee Research, the Security Research Division of McAfee, Inc. under
    8  * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
    9  * DARPA CHATS research program.
   10  *
   11  * Redistribution and use in source and binary forms, with or without
   12  * modification, are permitted provided that the following conditions
   13  * are met:
   14  * 1. Redistributions of source code must retain the above copyright
   15  *    notice, this list of conditions and the following disclaimer.
   16  * 2. Redistributions in binary form must reproduce the above copyright
   17  *    notice, this list of conditions and the following disclaimer in the
   18  *    documentation and/or other materials provided with the distribution.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   30  * SUCH DAMAGE.
   31  */
   32 
   33 #include <sys/cdefs.h>
   34 __FBSDID("$FreeBSD: releng/9.0/sys/netinet/tcp_syncache.c 222748 2011-06-06 12:55:02Z rwatson $");
   35 
   36 #include "opt_inet.h"
   37 #include "opt_inet6.h"
   38 #include "opt_ipsec.h"
   39 #include "opt_pcbgroup.h"
   40 
   41 #include <sys/param.h>
   42 #include <sys/systm.h>
   43 #include <sys/kernel.h>
   44 #include <sys/sysctl.h>
   45 #include <sys/limits.h>
   46 #include <sys/lock.h>
   47 #include <sys/mutex.h>
   48 #include <sys/malloc.h>
   49 #include <sys/mbuf.h>
   50 #include <sys/md5.h>
   51 #include <sys/proc.h>           /* for proc0 declaration */
   52 #include <sys/random.h>
   53 #include <sys/socket.h>
   54 #include <sys/socketvar.h>
   55 #include <sys/syslog.h>
   56 #include <sys/ucred.h>
   57 
   58 #include <vm/uma.h>
   59 
   60 #include <net/if.h>
   61 #include <net/route.h>
   62 #include <net/vnet.h>
   63 
   64 #include <netinet/in.h>
   65 #include <netinet/in_systm.h>
   66 #include <netinet/ip.h>
   67 #include <netinet/in_var.h>
   68 #include <netinet/in_pcb.h>
   69 #include <netinet/ip_var.h>
   70 #include <netinet/ip_options.h>
   71 #ifdef INET6
   72 #include <netinet/ip6.h>
   73 #include <netinet/icmp6.h>
   74 #include <netinet6/nd6.h>
   75 #include <netinet6/ip6_var.h>
   76 #include <netinet6/in6_pcb.h>
   77 #endif
   78 #include <netinet/tcp.h>
   79 #include <netinet/tcp_fsm.h>
   80 #include <netinet/tcp_seq.h>
   81 #include <netinet/tcp_timer.h>
   82 #include <netinet/tcp_var.h>
   83 #include <netinet/tcp_syncache.h>
   84 #include <netinet/tcp_offload.h>
   85 #ifdef INET6
   86 #include <netinet6/tcp6_var.h>
   87 #endif
   88 
   89 #ifdef IPSEC
   90 #include <netipsec/ipsec.h>
   91 #ifdef INET6
   92 #include <netipsec/ipsec6.h>
   93 #endif
   94 #include <netipsec/key.h>
   95 #endif /*IPSEC*/
   96 
   97 #include <machine/in_cksum.h>
   98 
   99 #include <security/mac/mac_framework.h>
  100 
  101 static VNET_DEFINE(int, tcp_syncookies) = 1;
  102 #define V_tcp_syncookies                VNET(tcp_syncookies)
  103 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
  104     &VNET_NAME(tcp_syncookies), 0,
  105     "Use TCP SYN cookies if the syncache overflows");
  106 
  107 static VNET_DEFINE(int, tcp_syncookiesonly) = 0;
  108 #define V_tcp_syncookiesonly            VNET(tcp_syncookiesonly)
  109 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
  110     &VNET_NAME(tcp_syncookiesonly), 0,
  111     "Use only TCP SYN cookies");
  112 
  113 #ifdef TCP_OFFLOAD_DISABLE
  114 #define TOEPCB_ISSET(sc) (0)
  115 #else
  116 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL)
  117 #endif
  118 
  119 static void      syncache_drop(struct syncache *, struct syncache_head *);
  120 static void      syncache_free(struct syncache *);
  121 static void      syncache_insert(struct syncache *, struct syncache_head *);
  122 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
  123 static int       syncache_respond(struct syncache *);
  124 static struct    socket *syncache_socket(struct syncache *, struct socket *,
  125                     struct mbuf *m);
  126 static void      syncache_timeout(struct syncache *sc, struct syncache_head *sch,
  127                     int docallout);
  128 static void      syncache_timer(void *);
  129 static void      syncookie_generate(struct syncache_head *, struct syncache *,
  130                     u_int32_t *);
  131 static struct syncache
  132                 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
  133                     struct syncache *, struct tcpopt *, struct tcphdr *,
  134                     struct socket *);
  135 
  136 /*
  137  * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
  138  * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
  139  * the odds are that the user has given up attempting to connect by then.
  140  */
  141 #define SYNCACHE_MAXREXMTS              3
  142 
  143 /* Arbitrary values */
  144 #define TCP_SYNCACHE_HASHSIZE           512
  145 #define TCP_SYNCACHE_BUCKETLIMIT        30
  146 
  147 static VNET_DEFINE(struct tcp_syncache, tcp_syncache);
  148 #define V_tcp_syncache                  VNET(tcp_syncache)
  149 
  150 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
  151 
  152 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
  153     &VNET_NAME(tcp_syncache.bucket_limit), 0,
  154     "Per-bucket hash limit for syncache");
  155 
  156 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
  157     &VNET_NAME(tcp_syncache.cache_limit), 0,
  158     "Overall entry limit for syncache");
  159 
  160 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
  161     &VNET_NAME(tcp_syncache.cache_count), 0,
  162     "Current number of entries in syncache");
  163 
  164 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
  165     &VNET_NAME(tcp_syncache.hashsize), 0,
  166     "Size of TCP syncache hashtable");
  167 
  168 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
  169     &VNET_NAME(tcp_syncache.rexmt_limit), 0,
  170     "Limit on SYN/ACK retransmissions");
  171 
  172 VNET_DEFINE(int, tcp_sc_rst_sock_fail) = 1;
  173 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail,
  174     CTLFLAG_RW, &VNET_NAME(tcp_sc_rst_sock_fail), 0,
  175     "Send reset on socket allocation failure");
  176 
  177 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
  178 
  179 #define SYNCACHE_HASH(inc, mask)                                        \
  180         ((V_tcp_syncache.hash_secret ^                                  \
  181           (inc)->inc_faddr.s_addr ^                                     \
  182           ((inc)->inc_faddr.s_addr >> 16) ^                             \
  183           (inc)->inc_fport ^ (inc)->inc_lport) & mask)
  184 
  185 #define SYNCACHE_HASH6(inc, mask)                                       \
  186         ((V_tcp_syncache.hash_secret ^                                  \
  187           (inc)->inc6_faddr.s6_addr32[0] ^                              \
  188           (inc)->inc6_faddr.s6_addr32[3] ^                              \
  189           (inc)->inc_fport ^ (inc)->inc_lport) & mask)
  190 
  191 #define ENDPTS_EQ(a, b) (                                               \
  192         (a)->ie_fport == (b)->ie_fport &&                               \
  193         (a)->ie_lport == (b)->ie_lport &&                               \
  194         (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr &&                 \
  195         (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr                    \
  196 )
  197 
  198 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
  199 
  200 #define SCH_LOCK(sch)           mtx_lock(&(sch)->sch_mtx)
  201 #define SCH_UNLOCK(sch)         mtx_unlock(&(sch)->sch_mtx)
  202 #define SCH_LOCK_ASSERT(sch)    mtx_assert(&(sch)->sch_mtx, MA_OWNED)
  203 
  204 /*
  205  * Requires the syncache entry to be already removed from the bucket list.
  206  */
  207 static void
  208 syncache_free(struct syncache *sc)
  209 {
  210 
  211         if (sc->sc_ipopts)
  212                 (void) m_free(sc->sc_ipopts);
  213         if (sc->sc_cred)
  214                 crfree(sc->sc_cred);
  215 #ifdef MAC
  216         mac_syncache_destroy(&sc->sc_label);
  217 #endif
  218 
  219         uma_zfree(V_tcp_syncache.zone, sc);
  220 }
  221 
  222 void
  223 syncache_init(void)
  224 {
  225         int i;
  226 
  227         V_tcp_syncache.cache_count = 0;
  228         V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
  229         V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
  230         V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
  231         V_tcp_syncache.hash_secret = arc4random();
  232 
  233         TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
  234             &V_tcp_syncache.hashsize);
  235         TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
  236             &V_tcp_syncache.bucket_limit);
  237         if (!powerof2(V_tcp_syncache.hashsize) ||
  238             V_tcp_syncache.hashsize == 0) {
  239                 printf("WARNING: syncache hash size is not a power of 2.\n");
  240                 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
  241         }
  242         V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1;
  243 
  244         /* Set limits. */
  245         V_tcp_syncache.cache_limit =
  246             V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit;
  247         TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
  248             &V_tcp_syncache.cache_limit);
  249 
  250         /* Allocate the hash table. */
  251         V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize *
  252             sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO);
  253 
  254         /* Initialize the hash buckets. */
  255         for (i = 0; i < V_tcp_syncache.hashsize; i++) {
  256 #ifdef VIMAGE
  257                 V_tcp_syncache.hashbase[i].sch_vnet = curvnet;
  258 #endif
  259                 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
  260                 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
  261                          NULL, MTX_DEF);
  262                 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
  263                          &V_tcp_syncache.hashbase[i].sch_mtx, 0);
  264                 V_tcp_syncache.hashbase[i].sch_length = 0;
  265         }
  266 
  267         /* Create the syncache entry zone. */
  268         V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
  269             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
  270         uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
  271 }
  272 
  273 #ifdef VIMAGE
  274 void
  275 syncache_destroy(void)
  276 {
  277         struct syncache_head *sch;
  278         struct syncache *sc, *nsc;
  279         int i;
  280 
  281         /* Cleanup hash buckets: stop timers, free entries, destroy locks. */
  282         for (i = 0; i < V_tcp_syncache.hashsize; i++) {
  283 
  284                 sch = &V_tcp_syncache.hashbase[i];
  285                 callout_drain(&sch->sch_timer);
  286 
  287                 SCH_LOCK(sch);
  288                 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc)
  289                         syncache_drop(sc, sch);
  290                 SCH_UNLOCK(sch);
  291                 KASSERT(TAILQ_EMPTY(&sch->sch_bucket),
  292                     ("%s: sch->sch_bucket not empty", __func__));
  293                 KASSERT(sch->sch_length == 0, ("%s: sch->sch_length %d not 0",
  294                     __func__, sch->sch_length));
  295                 mtx_destroy(&sch->sch_mtx);
  296         }
  297 
  298         KASSERT(V_tcp_syncache.cache_count == 0, ("%s: cache_count %d not 0",
  299             __func__, V_tcp_syncache.cache_count));
  300 
  301         /* Free the allocated global resources. */
  302         uma_zdestroy(V_tcp_syncache.zone);
  303         free(V_tcp_syncache.hashbase, M_SYNCACHE);
  304 }
  305 #endif
  306 
  307 /*
  308  * Inserts a syncache entry into the specified bucket row.
  309  * Locks and unlocks the syncache_head autonomously.
  310  */
  311 static void
  312 syncache_insert(struct syncache *sc, struct syncache_head *sch)
  313 {
  314         struct syncache *sc2;
  315 
  316         SCH_LOCK(sch);
  317 
  318         /*
  319          * Make sure that we don't overflow the per-bucket limit.
  320          * If the bucket is full, toss the oldest element.
  321          */
  322         if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
  323                 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
  324                         ("sch->sch_length incorrect"));
  325                 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
  326                 syncache_drop(sc2, sch);
  327                 TCPSTAT_INC(tcps_sc_bucketoverflow);
  328         }
  329 
  330         /* Put it into the bucket. */
  331         TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
  332         sch->sch_length++;
  333 
  334         /* Reinitialize the bucket row's timer. */
  335         if (sch->sch_length == 1)
  336                 sch->sch_nextc = ticks + INT_MAX;
  337         syncache_timeout(sc, sch, 1);
  338 
  339         SCH_UNLOCK(sch);
  340 
  341         V_tcp_syncache.cache_count++;
  342         TCPSTAT_INC(tcps_sc_added);
  343 }
  344 
  345 /*
  346  * Remove and free entry from syncache bucket row.
  347  * Expects locked syncache head.
  348  */
  349 static void
  350 syncache_drop(struct syncache *sc, struct syncache_head *sch)
  351 {
  352 
  353         SCH_LOCK_ASSERT(sch);
  354 
  355         TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
  356         sch->sch_length--;
  357 
  358 #ifndef TCP_OFFLOAD_DISABLE
  359         if (sc->sc_tu)
  360                 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
  361 #endif              
  362         syncache_free(sc);
  363         V_tcp_syncache.cache_count--;
  364 }
  365 
  366 /*
  367  * Engage/reengage time on bucket row.
  368  */
  369 static void
  370 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
  371 {
  372         sc->sc_rxttime = ticks +
  373                 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
  374         sc->sc_rxmits++;
  375         if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
  376                 sch->sch_nextc = sc->sc_rxttime;
  377                 if (docallout)
  378                         callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
  379                             syncache_timer, (void *)sch);
  380         }
  381 }
  382 
  383 /*
  384  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
  385  * If we have retransmitted an entry the maximum number of times, expire it.
  386  * One separate timer for each bucket row.
  387  */
  388 static void
  389 syncache_timer(void *xsch)
  390 {
  391         struct syncache_head *sch = (struct syncache_head *)xsch;
  392         struct syncache *sc, *nsc;
  393         int tick = ticks;
  394         char *s;
  395 
  396         CURVNET_SET(sch->sch_vnet);
  397 
  398         /* NB: syncache_head has already been locked by the callout. */
  399         SCH_LOCK_ASSERT(sch);
  400 
  401         /*
  402          * In the following cycle we may remove some entries and/or
  403          * advance some timeouts, so re-initialize the bucket timer.
  404          */
  405         sch->sch_nextc = tick + INT_MAX;
  406 
  407         TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
  408                 /*
  409                  * We do not check if the listen socket still exists
  410                  * and accept the case where the listen socket may be
  411                  * gone by the time we resend the SYN/ACK.  We do
  412                  * not expect this to happens often. If it does,
  413                  * then the RST will be sent by the time the remote
  414                  * host does the SYN/ACK->ACK.
  415                  */
  416                 if (TSTMP_GT(sc->sc_rxttime, tick)) {
  417                         if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
  418                                 sch->sch_nextc = sc->sc_rxttime;
  419                         continue;
  420                 }
  421                 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
  422                         if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  423                                 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
  424                                     "giving up and removing syncache entry\n",
  425                                     s, __func__);
  426                                 free(s, M_TCPLOG);
  427                         }
  428                         syncache_drop(sc, sch);
  429                         TCPSTAT_INC(tcps_sc_stale);
  430                         continue;
  431                 }
  432                 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  433                         log(LOG_DEBUG, "%s; %s: Response timeout, "
  434                             "retransmitting (%u) SYN|ACK\n",
  435                             s, __func__, sc->sc_rxmits);
  436                         free(s, M_TCPLOG);
  437                 }
  438 
  439                 (void) syncache_respond(sc);
  440                 TCPSTAT_INC(tcps_sc_retransmitted);
  441                 syncache_timeout(sc, sch, 0);
  442         }
  443         if (!TAILQ_EMPTY(&(sch)->sch_bucket))
  444                 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
  445                         syncache_timer, (void *)(sch));
  446         CURVNET_RESTORE();
  447 }
  448 
  449 /*
  450  * Find an entry in the syncache.
  451  * Returns always with locked syncache_head plus a matching entry or NULL.
  452  */
  453 struct syncache *
  454 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
  455 {
  456         struct syncache *sc;
  457         struct syncache_head *sch;
  458 
  459 #ifdef INET6
  460         if (inc->inc_flags & INC_ISIPV6) {
  461                 sch = &V_tcp_syncache.hashbase[
  462                     SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
  463                 *schp = sch;
  464 
  465                 SCH_LOCK(sch);
  466 
  467                 /* Circle through bucket row to find matching entry. */
  468                 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
  469                         if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
  470                                 return (sc);
  471                 }
  472         } else
  473 #endif
  474         {
  475                 sch = &V_tcp_syncache.hashbase[
  476                     SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
  477                 *schp = sch;
  478 
  479                 SCH_LOCK(sch);
  480 
  481                 /* Circle through bucket row to find matching entry. */
  482                 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
  483 #ifdef INET6
  484                         if (sc->sc_inc.inc_flags & INC_ISIPV6)
  485                                 continue;
  486 #endif
  487                         if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
  488                                 return (sc);
  489                 }
  490         }
  491         SCH_LOCK_ASSERT(*schp);
  492         return (NULL);                  /* always returns with locked sch */
  493 }
  494 
  495 /*
  496  * This function is called when we get a RST for a
  497  * non-existent connection, so that we can see if the
  498  * connection is in the syn cache.  If it is, zap it.
  499  */
  500 void
  501 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
  502 {
  503         struct syncache *sc;
  504         struct syncache_head *sch;
  505         char *s = NULL;
  506 
  507         sc = syncache_lookup(inc, &sch);        /* returns locked sch */
  508         SCH_LOCK_ASSERT(sch);
  509 
  510         /*
  511          * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
  512          * See RFC 793 page 65, section SEGMENT ARRIVES.
  513          */
  514         if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
  515                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  516                         log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
  517                             "FIN flag set, segment ignored\n", s, __func__);
  518                 TCPSTAT_INC(tcps_badrst);
  519                 goto done;
  520         }
  521 
  522         /*
  523          * No corresponding connection was found in syncache.
  524          * If syncookies are enabled and possibly exclusively
  525          * used, or we are under memory pressure, a valid RST
  526          * may not find a syncache entry.  In that case we're
  527          * done and no SYN|ACK retransmissions will happen.
  528          * Otherwise the RST was misdirected or spoofed.
  529          */
  530         if (sc == NULL) {
  531                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  532                         log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
  533                             "syncache entry (possibly syncookie only), "
  534                             "segment ignored\n", s, __func__);
  535                 TCPSTAT_INC(tcps_badrst);
  536                 goto done;
  537         }
  538 
  539         /*
  540          * If the RST bit is set, check the sequence number to see
  541          * if this is a valid reset segment.
  542          * RFC 793 page 37:
  543          *   In all states except SYN-SENT, all reset (RST) segments
  544          *   are validated by checking their SEQ-fields.  A reset is
  545          *   valid if its sequence number is in the window.
  546          *
  547          *   The sequence number in the reset segment is normally an
  548          *   echo of our outgoing acknowlegement numbers, but some hosts
  549          *   send a reset with the sequence number at the rightmost edge
  550          *   of our receive window, and we have to handle this case.
  551          */
  552         if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
  553             SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
  554                 syncache_drop(sc, sch);
  555                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  556                         log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
  557                             "connection attempt aborted by remote endpoint\n",
  558                             s, __func__);
  559                 TCPSTAT_INC(tcps_sc_reset);
  560         } else {
  561                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  562                         log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
  563                             "IRS %u (+WND %u), segment ignored\n",
  564                             s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
  565                 TCPSTAT_INC(tcps_badrst);
  566         }
  567 
  568 done:
  569         if (s != NULL)
  570                 free(s, M_TCPLOG);
  571         SCH_UNLOCK(sch);
  572 }
  573 
  574 void
  575 syncache_badack(struct in_conninfo *inc)
  576 {
  577         struct syncache *sc;
  578         struct syncache_head *sch;
  579 
  580         sc = syncache_lookup(inc, &sch);        /* returns locked sch */
  581         SCH_LOCK_ASSERT(sch);
  582         if (sc != NULL) {
  583                 syncache_drop(sc, sch);
  584                 TCPSTAT_INC(tcps_sc_badack);
  585         }
  586         SCH_UNLOCK(sch);
  587 }
  588 
  589 void
  590 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
  591 {
  592         struct syncache *sc;
  593         struct syncache_head *sch;
  594 
  595         sc = syncache_lookup(inc, &sch);        /* returns locked sch */
  596         SCH_LOCK_ASSERT(sch);
  597         if (sc == NULL)
  598                 goto done;
  599 
  600         /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
  601         if (ntohl(th->th_seq) != sc->sc_iss)
  602                 goto done;
  603 
  604         /*
  605          * If we've rertransmitted 3 times and this is our second error,
  606          * we remove the entry.  Otherwise, we allow it to continue on.
  607          * This prevents us from incorrectly nuking an entry during a
  608          * spurious network outage.
  609          *
  610          * See tcp_notify().
  611          */
  612         if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
  613                 sc->sc_flags |= SCF_UNREACH;
  614                 goto done;
  615         }
  616         syncache_drop(sc, sch);
  617         TCPSTAT_INC(tcps_sc_unreach);
  618 done:
  619         SCH_UNLOCK(sch);
  620 }
  621 
  622 /*
  623  * Build a new TCP socket structure from a syncache entry.
  624  */
  625 static struct socket *
  626 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
  627 {
  628         struct inpcb *inp = NULL;
  629         struct socket *so;
  630         struct tcpcb *tp;
  631         int error;
  632         char *s;
  633 
  634         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
  635 
  636         /*
  637          * Ok, create the full blown connection, and set things up
  638          * as they would have been set up if we had created the
  639          * connection when the SYN arrived.  If we can't create
  640          * the connection, abort it.
  641          */
  642         so = sonewconn(lso, SS_ISCONNECTED);
  643         if (so == NULL) {
  644                 /*
  645                  * Drop the connection; we will either send a RST or
  646                  * have the peer retransmit its SYN again after its
  647                  * RTO and try again.
  648                  */
  649                 TCPSTAT_INC(tcps_listendrop);
  650                 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  651                         log(LOG_DEBUG, "%s; %s: Socket create failed "
  652                             "due to limits or memory shortage\n",
  653                             s, __func__);
  654                         free(s, M_TCPLOG);
  655                 }
  656                 goto abort2;
  657         }
  658 #ifdef MAC
  659         mac_socketpeer_set_from_mbuf(m, so);
  660 #endif
  661 
  662         inp = sotoinpcb(so);
  663         inp->inp_inc.inc_fibnum = so->so_fibnum;
  664         INP_WLOCK(inp);
  665         INP_HASH_WLOCK(&V_tcbinfo);
  666 
  667         /* Insert new socket into PCB hash list. */
  668         inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
  669 #ifdef INET6
  670         if (sc->sc_inc.inc_flags & INC_ISIPV6) {
  671                 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
  672         } else {
  673                 inp->inp_vflag &= ~INP_IPV6;
  674                 inp->inp_vflag |= INP_IPV4;
  675 #endif
  676                 inp->inp_laddr = sc->sc_inc.inc_laddr;
  677 #ifdef INET6
  678         }
  679 #endif
  680 
  681         /*
  682          * Install in the reservation hash table for now, but don't yet
  683          * install a connection group since the full 4-tuple isn't yet
  684          * configured.
  685          */
  686         inp->inp_lport = sc->sc_inc.inc_lport;
  687         if ((error = in_pcbinshash_nopcbgroup(inp)) != 0) {
  688                 /*
  689                  * Undo the assignments above if we failed to
  690                  * put the PCB on the hash lists.
  691                  */
  692 #ifdef INET6
  693                 if (sc->sc_inc.inc_flags & INC_ISIPV6)
  694                         inp->in6p_laddr = in6addr_any;
  695                 else
  696 #endif
  697                         inp->inp_laddr.s_addr = INADDR_ANY;
  698                 inp->inp_lport = 0;
  699                 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  700                         log(LOG_DEBUG, "%s; %s: in_pcbinshash failed "
  701                             "with error %i\n",
  702                             s, __func__, error);
  703                         free(s, M_TCPLOG);
  704                 }
  705                 INP_HASH_WUNLOCK(&V_tcbinfo);
  706                 goto abort;
  707         }
  708 #ifdef IPSEC
  709         /* Copy old policy into new socket's. */
  710         if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
  711                 printf("syncache_socket: could not copy policy\n");
  712 #endif
  713 #ifdef INET6
  714         if (sc->sc_inc.inc_flags & INC_ISIPV6) {
  715                 struct inpcb *oinp = sotoinpcb(lso);
  716                 struct in6_addr laddr6;
  717                 struct sockaddr_in6 sin6;
  718                 /*
  719                  * Inherit socket options from the listening socket.
  720                  * Note that in6p_inputopts are not (and should not be)
  721                  * copied, since it stores previously received options and is
  722                  * used to detect if each new option is different than the
  723                  * previous one and hence should be passed to a user.
  724                  * If we copied in6p_inputopts, a user would not be able to
  725                  * receive options just after calling the accept system call.
  726                  */
  727                 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
  728                 if (oinp->in6p_outputopts)
  729                         inp->in6p_outputopts =
  730                             ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
  731 
  732                 sin6.sin6_family = AF_INET6;
  733                 sin6.sin6_len = sizeof(sin6);
  734                 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
  735                 sin6.sin6_port = sc->sc_inc.inc_fport;
  736                 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
  737                 laddr6 = inp->in6p_laddr;
  738                 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
  739                         inp->in6p_laddr = sc->sc_inc.inc6_laddr;
  740                 if ((error = in6_pcbconnect_mbuf(inp, (struct sockaddr *)&sin6,
  741                     thread0.td_ucred, m)) != 0) {
  742                         inp->in6p_laddr = laddr6;
  743                         if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  744                                 log(LOG_DEBUG, "%s; %s: in6_pcbconnect failed "
  745                                     "with error %i\n",
  746                                     s, __func__, error);
  747                                 free(s, M_TCPLOG);
  748                         }
  749                         INP_HASH_WUNLOCK(&V_tcbinfo);
  750                         goto abort;
  751                 }
  752                 /* Override flowlabel from in6_pcbconnect. */
  753                 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
  754                 inp->inp_flow |= sc->sc_flowlabel;
  755         }
  756 #endif /* INET6 */
  757 #if defined(INET) && defined(INET6)
  758         else
  759 #endif
  760 #ifdef INET
  761         {
  762                 struct in_addr laddr;
  763                 struct sockaddr_in sin;
  764 
  765                 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
  766                 
  767                 if (inp->inp_options == NULL) {
  768                         inp->inp_options = sc->sc_ipopts;
  769                         sc->sc_ipopts = NULL;
  770                 }
  771 
  772                 sin.sin_family = AF_INET;
  773                 sin.sin_len = sizeof(sin);
  774                 sin.sin_addr = sc->sc_inc.inc_faddr;
  775                 sin.sin_port = sc->sc_inc.inc_fport;
  776                 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
  777                 laddr = inp->inp_laddr;
  778                 if (inp->inp_laddr.s_addr == INADDR_ANY)
  779                         inp->inp_laddr = sc->sc_inc.inc_laddr;
  780                 if ((error = in_pcbconnect_mbuf(inp, (struct sockaddr *)&sin,
  781                     thread0.td_ucred, m)) != 0) {
  782                         inp->inp_laddr = laddr;
  783                         if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  784                                 log(LOG_DEBUG, "%s; %s: in_pcbconnect failed "
  785                                     "with error %i\n",
  786                                     s, __func__, error);
  787                                 free(s, M_TCPLOG);
  788                         }
  789                         INP_HASH_WUNLOCK(&V_tcbinfo);
  790                         goto abort;
  791                 }
  792         }
  793 #endif /* INET */
  794         INP_HASH_WUNLOCK(&V_tcbinfo);
  795         tp = intotcpcb(inp);
  796         tp->t_state = TCPS_SYN_RECEIVED;
  797         tp->iss = sc->sc_iss;
  798         tp->irs = sc->sc_irs;
  799         tcp_rcvseqinit(tp);
  800         tcp_sendseqinit(tp);
  801         tp->snd_wl1 = sc->sc_irs;
  802         tp->snd_max = tp->iss + 1;
  803         tp->snd_nxt = tp->iss + 1;
  804         tp->rcv_up = sc->sc_irs + 1;
  805         tp->rcv_wnd = sc->sc_wnd;
  806         tp->rcv_adv += tp->rcv_wnd;
  807         tp->last_ack_sent = tp->rcv_nxt;
  808 
  809         tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
  810         if (sc->sc_flags & SCF_NOOPT)
  811                 tp->t_flags |= TF_NOOPT;
  812         else {
  813                 if (sc->sc_flags & SCF_WINSCALE) {
  814                         tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
  815                         tp->snd_scale = sc->sc_requested_s_scale;
  816                         tp->request_r_scale = sc->sc_requested_r_scale;
  817                 }
  818                 if (sc->sc_flags & SCF_TIMESTAMP) {
  819                         tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
  820                         tp->ts_recent = sc->sc_tsreflect;
  821                         tp->ts_recent_age = ticks;
  822                         tp->ts_offset = sc->sc_tsoff;
  823                 }
  824 #ifdef TCP_SIGNATURE
  825                 if (sc->sc_flags & SCF_SIGNATURE)
  826                         tp->t_flags |= TF_SIGNATURE;
  827 #endif
  828                 if (sc->sc_flags & SCF_SACK)
  829                         tp->t_flags |= TF_SACK_PERMIT;
  830         }
  831 
  832         if (sc->sc_flags & SCF_ECN)
  833                 tp->t_flags |= TF_ECN_PERMIT;
  834 
  835         /*
  836          * Set up MSS and get cached values from tcp_hostcache.
  837          * This might overwrite some of the defaults we just set.
  838          */
  839         tcp_mss(tp, sc->sc_peer_mss);
  840 
  841         /*
  842          * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
  843          * NB: sc_rxmits counts all SYN,ACK transmits, not just retransmits.
  844          */
  845         if (sc->sc_rxmits > 1)
  846                 tp->snd_cwnd = tp->t_maxseg;
  847         tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
  848 
  849         INP_WUNLOCK(inp);
  850 
  851         TCPSTAT_INC(tcps_accepts);
  852         return (so);
  853 
  854 abort:
  855         INP_WUNLOCK(inp);
  856 abort2:
  857         if (so != NULL)
  858                 soabort(so);
  859         return (NULL);
  860 }
  861 
  862 /*
  863  * This function gets called when we receive an ACK for a
  864  * socket in the LISTEN state.  We look up the connection
  865  * in the syncache, and if its there, we pull it out of
  866  * the cache and turn it into a full-blown connection in
  867  * the SYN-RECEIVED state.
  868  */
  869 int
  870 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
  871     struct socket **lsop, struct mbuf *m)
  872 {
  873         struct syncache *sc;
  874         struct syncache_head *sch;
  875         struct syncache scs;
  876         char *s;
  877 
  878         /*
  879          * Global TCP locks are held because we manipulate the PCB lists
  880          * and create a new socket.
  881          */
  882         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
  883         KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
  884             ("%s: can handle only ACK", __func__));
  885 
  886         sc = syncache_lookup(inc, &sch);        /* returns locked sch */
  887         SCH_LOCK_ASSERT(sch);
  888         if (sc == NULL) {
  889                 /*
  890                  * There is no syncache entry, so see if this ACK is
  891                  * a returning syncookie.  To do this, first:
  892                  *  A. See if this socket has had a syncache entry dropped in
  893                  *     the past.  We don't want to accept a bogus syncookie
  894                  *     if we've never received a SYN.
  895                  *  B. check that the syncookie is valid.  If it is, then
  896                  *     cobble up a fake syncache entry, and return.
  897                  */
  898                 if (!V_tcp_syncookies) {
  899                         SCH_UNLOCK(sch);
  900                         if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  901                                 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
  902                                     "segment rejected (syncookies disabled)\n",
  903                                     s, __func__);
  904                         goto failed;
  905                 }
  906                 bzero(&scs, sizeof(scs));
  907                 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
  908                 SCH_UNLOCK(sch);
  909                 if (sc == NULL) {
  910                         if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  911                                 log(LOG_DEBUG, "%s; %s: Segment failed "
  912                                     "SYNCOOKIE authentication, segment rejected "
  913                                     "(probably spoofed)\n", s, __func__);
  914                         goto failed;
  915                 }
  916         } else {
  917                 /* Pull out the entry to unlock the bucket row. */
  918                 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
  919                 sch->sch_length--;
  920                 V_tcp_syncache.cache_count--;
  921                 SCH_UNLOCK(sch);
  922         }
  923 
  924         /*
  925          * Segment validation:
  926          * ACK must match our initial sequence number + 1 (the SYN|ACK).
  927          */
  928         if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
  929                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  930                         log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
  931                             "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
  932                 goto failed;
  933         }
  934 
  935         /*
  936          * The SEQ must fall in the window starting at the received
  937          * initial receive sequence number + 1 (the SYN).
  938          */
  939         if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
  940             SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
  941             !TOEPCB_ISSET(sc)) {
  942                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  943                         log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
  944                             "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
  945                 goto failed;
  946         }
  947 
  948         if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
  949                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  950                         log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
  951                             "segment rejected\n", s, __func__);
  952                 goto failed;
  953         }
  954         /*
  955          * If timestamps were negotiated the reflected timestamp
  956          * must be equal to what we actually sent in the SYN|ACK.
  957          */
  958         if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
  959             !TOEPCB_ISSET(sc)) {
  960                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  961                         log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
  962                             "segment rejected\n",
  963                             s, __func__, to->to_tsecr, sc->sc_ts);
  964                 goto failed;
  965         }
  966 
  967         *lsop = syncache_socket(sc, *lsop, m);
  968 
  969         if (*lsop == NULL)
  970                 TCPSTAT_INC(tcps_sc_aborted);
  971         else
  972                 TCPSTAT_INC(tcps_sc_completed);
  973 
  974 /* how do we find the inp for the new socket? */
  975         if (sc != &scs)
  976                 syncache_free(sc);
  977         return (1);
  978 failed:
  979         if (sc != NULL && sc != &scs)
  980                 syncache_free(sc);
  981         if (s != NULL)
  982                 free(s, M_TCPLOG);
  983         *lsop = NULL;
  984         return (0);
  985 }
  986 
  987 int
  988 tcp_offload_syncache_expand(struct in_conninfo *inc, struct toeopt *toeo,
  989     struct tcphdr *th, struct socket **lsop, struct mbuf *m)
  990 {
  991         struct tcpopt to;
  992         int rc;
  993 
  994         bzero(&to, sizeof(struct tcpopt));
  995         to.to_mss = toeo->to_mss;
  996         to.to_wscale = toeo->to_wscale;
  997         to.to_flags = toeo->to_flags;
  998         
  999         INP_INFO_WLOCK(&V_tcbinfo);
 1000         rc = syncache_expand(inc, &to, th, lsop, m);
 1001         INP_INFO_WUNLOCK(&V_tcbinfo);
 1002 
 1003         return (rc);
 1004 }
 1005 
 1006 /*
 1007  * Given a LISTEN socket and an inbound SYN request, add
 1008  * this to the syn cache, and send back a segment:
 1009  *      <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
 1010  * to the source.
 1011  *
 1012  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
 1013  * Doing so would require that we hold onto the data and deliver it
 1014  * to the application.  However, if we are the target of a SYN-flood
 1015  * DoS attack, an attacker could send data which would eventually
 1016  * consume all available buffer space if it were ACKed.  By not ACKing
 1017  * the data, we avoid this DoS scenario.
 1018  */
 1019 static void
 1020 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
 1021     struct inpcb *inp, struct socket **lsop, struct mbuf *m,
 1022     struct toe_usrreqs *tu, void *toepcb)
 1023 {
 1024         struct tcpcb *tp;
 1025         struct socket *so;
 1026         struct syncache *sc = NULL;
 1027         struct syncache_head *sch;
 1028         struct mbuf *ipopts = NULL;
 1029         u_int32_t flowtmp;
 1030         u_int ltflags;
 1031         int win, sb_hiwat, ip_ttl, ip_tos;
 1032         char *s;
 1033 #ifdef INET6
 1034         int autoflowlabel = 0;
 1035 #endif
 1036 #ifdef MAC
 1037         struct label *maclabel;
 1038 #endif
 1039         struct syncache scs;
 1040         struct ucred *cred;
 1041 
 1042         INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
 1043         INP_WLOCK_ASSERT(inp);                  /* listen socket */
 1044         KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
 1045             ("%s: unexpected tcp flags", __func__));
 1046 
 1047         /*
 1048          * Combine all so/tp operations very early to drop the INP lock as
 1049          * soon as possible.
 1050          */
 1051         so = *lsop;
 1052         tp = sototcpcb(so);
 1053         cred = crhold(so->so_cred);
 1054 
 1055 #ifdef INET6
 1056         if ((inc->inc_flags & INC_ISIPV6) &&
 1057             (inp->inp_flags & IN6P_AUTOFLOWLABEL))
 1058                 autoflowlabel = 1;
 1059 #endif
 1060         ip_ttl = inp->inp_ip_ttl;
 1061         ip_tos = inp->inp_ip_tos;
 1062         win = sbspace(&so->so_rcv);
 1063         sb_hiwat = so->so_rcv.sb_hiwat;
 1064         ltflags = (tp->t_flags & (TF_NOOPT | TF_SIGNATURE));
 1065 
 1066         /* By the time we drop the lock these should no longer be used. */
 1067         so = NULL;
 1068         tp = NULL;
 1069 
 1070 #ifdef MAC
 1071         if (mac_syncache_init(&maclabel) != 0) {
 1072                 INP_WUNLOCK(inp);
 1073                 INP_INFO_WUNLOCK(&V_tcbinfo);
 1074                 goto done;
 1075         } else
 1076                 mac_syncache_create(maclabel, inp);
 1077 #endif
 1078         INP_WUNLOCK(inp);
 1079         INP_INFO_WUNLOCK(&V_tcbinfo);
 1080 
 1081         /*
 1082          * Remember the IP options, if any.
 1083          */
 1084 #ifdef INET6
 1085         if (!(inc->inc_flags & INC_ISIPV6))
 1086 #endif
 1087 #ifdef INET
 1088                 ipopts = (m) ? ip_srcroute(m) : NULL;
 1089 #else
 1090                 ipopts = NULL;
 1091 #endif
 1092 
 1093         /*
 1094          * See if we already have an entry for this connection.
 1095          * If we do, resend the SYN,ACK, and reset the retransmit timer.
 1096          *
 1097          * XXX: should the syncache be re-initialized with the contents
 1098          * of the new SYN here (which may have different options?)
 1099          *
 1100          * XXX: We do not check the sequence number to see if this is a
 1101          * real retransmit or a new connection attempt.  The question is
 1102          * how to handle such a case; either ignore it as spoofed, or
 1103          * drop the current entry and create a new one?
 1104          */
 1105         sc = syncache_lookup(inc, &sch);        /* returns locked entry */
 1106         SCH_LOCK_ASSERT(sch);
 1107         if (sc != NULL) {
 1108 #ifndef TCP_OFFLOAD_DISABLE
 1109                 if (sc->sc_tu)
 1110                         sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
 1111                             sc->sc_toepcb);
 1112 #endif              
 1113                 TCPSTAT_INC(tcps_sc_dupsyn);
 1114                 if (ipopts) {
 1115                         /*
 1116                          * If we were remembering a previous source route,
 1117                          * forget it and use the new one we've been given.
 1118                          */
 1119                         if (sc->sc_ipopts)
 1120                                 (void) m_free(sc->sc_ipopts);
 1121                         sc->sc_ipopts = ipopts;
 1122                 }
 1123                 /*
 1124                  * Update timestamp if present.
 1125                  */
 1126                 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
 1127                         sc->sc_tsreflect = to->to_tsval;
 1128                 else
 1129                         sc->sc_flags &= ~SCF_TIMESTAMP;
 1130 #ifdef MAC
 1131                 /*
 1132                  * Since we have already unconditionally allocated label
 1133                  * storage, free it up.  The syncache entry will already
 1134                  * have an initialized label we can use.
 1135                  */
 1136                 mac_syncache_destroy(&maclabel);
 1137 #endif
 1138                 /* Retransmit SYN|ACK and reset retransmit count. */
 1139                 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
 1140                         log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
 1141                             "resetting timer and retransmitting SYN|ACK\n",
 1142                             s, __func__);
 1143                         free(s, M_TCPLOG);
 1144                 }
 1145                 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
 1146                         sc->sc_rxmits = 0;
 1147                         syncache_timeout(sc, sch, 1);
 1148                         TCPSTAT_INC(tcps_sndacks);
 1149                         TCPSTAT_INC(tcps_sndtotal);
 1150                 }
 1151                 SCH_UNLOCK(sch);
 1152                 goto done;
 1153         }
 1154 
 1155         sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
 1156         if (sc == NULL) {
 1157                 /*
 1158                  * The zone allocator couldn't provide more entries.
 1159                  * Treat this as if the cache was full; drop the oldest
 1160                  * entry and insert the new one.
 1161                  */
 1162                 TCPSTAT_INC(tcps_sc_zonefail);
 1163                 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
 1164                         syncache_drop(sc, sch);
 1165                 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
 1166                 if (sc == NULL) {
 1167                         if (V_tcp_syncookies) {
 1168                                 bzero(&scs, sizeof(scs));
 1169                                 sc = &scs;
 1170                         } else {
 1171                                 SCH_UNLOCK(sch);
 1172                                 if (ipopts)
 1173                                         (void) m_free(ipopts);
 1174                                 goto done;
 1175                         }
 1176                 }
 1177         }
 1178         
 1179         /*
 1180          * Fill in the syncache values.
 1181          */
 1182 #ifdef MAC
 1183         sc->sc_label = maclabel;
 1184 #endif
 1185         sc->sc_cred = cred;
 1186         cred = NULL;
 1187         sc->sc_ipopts = ipopts;
 1188         bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
 1189 #ifdef INET6
 1190         if (!(inc->inc_flags & INC_ISIPV6))
 1191 #endif
 1192         {
 1193                 sc->sc_ip_tos = ip_tos;
 1194                 sc->sc_ip_ttl = ip_ttl;
 1195         }
 1196 #ifndef TCP_OFFLOAD_DISABLE     
 1197         sc->sc_tu = tu;
 1198         sc->sc_toepcb = toepcb;
 1199 #endif
 1200         sc->sc_irs = th->th_seq;
 1201         sc->sc_iss = arc4random();
 1202         sc->sc_flags = 0;
 1203         sc->sc_flowlabel = 0;
 1204 
 1205         /*
 1206          * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
 1207          * win was derived from socket earlier in the function.
 1208          */
 1209         win = imax(win, 0);
 1210         win = imin(win, TCP_MAXWIN);
 1211         sc->sc_wnd = win;
 1212 
 1213         if (V_tcp_do_rfc1323) {
 1214                 /*
 1215                  * A timestamp received in a SYN makes
 1216                  * it ok to send timestamp requests and replies.
 1217                  */
 1218                 if (to->to_flags & TOF_TS) {
 1219                         sc->sc_tsreflect = to->to_tsval;
 1220                         sc->sc_ts = ticks;
 1221                         sc->sc_flags |= SCF_TIMESTAMP;
 1222                 }
 1223                 if (to->to_flags & TOF_SCALE) {
 1224                         int wscale = 0;
 1225 
 1226                         /*
 1227                          * Pick the smallest possible scaling factor that
 1228                          * will still allow us to scale up to sb_max, aka
 1229                          * kern.ipc.maxsockbuf.
 1230                          *
 1231                          * We do this because there are broken firewalls that
 1232                          * will corrupt the window scale option, leading to
 1233                          * the other endpoint believing that our advertised
 1234                          * window is unscaled.  At scale factors larger than
 1235                          * 5 the unscaled window will drop below 1500 bytes,
 1236                          * leading to serious problems when traversing these
 1237                          * broken firewalls.
 1238                          *
 1239                          * With the default maxsockbuf of 256K, a scale factor
 1240                          * of 3 will be chosen by this algorithm.  Those who
 1241                          * choose a larger maxsockbuf should watch out
 1242                          * for the compatiblity problems mentioned above.
 1243                          *
 1244                          * RFC1323: The Window field in a SYN (i.e., a <SYN>
 1245                          * or <SYN,ACK>) segment itself is never scaled.
 1246                          */
 1247                         while (wscale < TCP_MAX_WINSHIFT &&
 1248                             (TCP_MAXWIN << wscale) < sb_max)
 1249                                 wscale++;
 1250                         sc->sc_requested_r_scale = wscale;
 1251                         sc->sc_requested_s_scale = to->to_wscale;
 1252                         sc->sc_flags |= SCF_WINSCALE;
 1253                 }
 1254         }
 1255 #ifdef TCP_SIGNATURE
 1256         /*
 1257          * If listening socket requested TCP digests, and received SYN
 1258          * contains the option, flag this in the syncache so that
 1259          * syncache_respond() will do the right thing with the SYN+ACK.
 1260          * XXX: Currently we always record the option by default and will
 1261          * attempt to use it in syncache_respond().
 1262          */
 1263         if (to->to_flags & TOF_SIGNATURE || ltflags & TF_SIGNATURE)
 1264                 sc->sc_flags |= SCF_SIGNATURE;
 1265 #endif
 1266         if (to->to_flags & TOF_SACKPERM)
 1267                 sc->sc_flags |= SCF_SACK;
 1268         if (to->to_flags & TOF_MSS)
 1269                 sc->sc_peer_mss = to->to_mss;   /* peer mss may be zero */
 1270         if (ltflags & TF_NOOPT)
 1271                 sc->sc_flags |= SCF_NOOPT;
 1272         if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
 1273                 sc->sc_flags |= SCF_ECN;
 1274 
 1275         if (V_tcp_syncookies) {
 1276                 syncookie_generate(sch, sc, &flowtmp);
 1277 #ifdef INET6
 1278                 if (autoflowlabel)
 1279                         sc->sc_flowlabel = flowtmp;
 1280 #endif
 1281         } else {
 1282 #ifdef INET6
 1283                 if (autoflowlabel)
 1284                         sc->sc_flowlabel =
 1285                             (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
 1286 #endif
 1287         }
 1288         SCH_UNLOCK(sch);
 1289 
 1290         /*
 1291          * Do a standard 3-way handshake.
 1292          */
 1293         if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
 1294                 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
 1295                         syncache_free(sc);
 1296                 else if (sc != &scs)
 1297                         syncache_insert(sc, sch);   /* locks and unlocks sch */
 1298                 TCPSTAT_INC(tcps_sndacks);
 1299                 TCPSTAT_INC(tcps_sndtotal);
 1300         } else {
 1301                 if (sc != &scs)
 1302                         syncache_free(sc);
 1303                 TCPSTAT_INC(tcps_sc_dropped);
 1304         }
 1305 
 1306 done:
 1307         if (cred != NULL)
 1308                 crfree(cred);
 1309 #ifdef MAC
 1310         if (sc == &scs)
 1311                 mac_syncache_destroy(&maclabel);
 1312 #endif
 1313         if (m) {
 1314                 
 1315                 *lsop = NULL;
 1316                 m_freem(m);
 1317         }
 1318 }
 1319 
 1320 static int
 1321 syncache_respond(struct syncache *sc)
 1322 {
 1323         struct ip *ip = NULL;
 1324         struct mbuf *m;
 1325         struct tcphdr *th = NULL;
 1326         int optlen, error = 0;  /* Make compiler happy */
 1327         u_int16_t hlen, tlen, mssopt;
 1328         struct tcpopt to;
 1329 #ifdef INET6
 1330         struct ip6_hdr *ip6 = NULL;
 1331 #endif
 1332 
 1333         hlen =
 1334 #ifdef INET6
 1335                (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
 1336 #endif
 1337                 sizeof(struct ip);
 1338         tlen = hlen + sizeof(struct tcphdr);
 1339 
 1340         /* Determine MSS we advertize to other end of connection. */
 1341         mssopt = tcp_mssopt(&sc->sc_inc);
 1342         if (sc->sc_peer_mss)
 1343                 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
 1344 
 1345         /* XXX: Assume that the entire packet will fit in a header mbuf. */
 1346         KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
 1347             ("syncache: mbuf too small"));
 1348 
 1349         /* Create the IP+TCP header from scratch. */
 1350         m = m_gethdr(M_DONTWAIT, MT_DATA);
 1351         if (m == NULL)
 1352                 return (ENOBUFS);
 1353 #ifdef MAC
 1354         mac_syncache_create_mbuf(sc->sc_label, m);
 1355 #endif
 1356         m->m_data += max_linkhdr;
 1357         m->m_len = tlen;
 1358         m->m_pkthdr.len = tlen;
 1359         m->m_pkthdr.rcvif = NULL;
 1360 
 1361 #ifdef INET6
 1362         if (sc->sc_inc.inc_flags & INC_ISIPV6) {
 1363                 ip6 = mtod(m, struct ip6_hdr *);
 1364                 ip6->ip6_vfc = IPV6_VERSION;
 1365                 ip6->ip6_nxt = IPPROTO_TCP;
 1366                 ip6->ip6_src = sc->sc_inc.inc6_laddr;
 1367                 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
 1368                 ip6->ip6_plen = htons(tlen - hlen);
 1369                 /* ip6_hlim is set after checksum */
 1370                 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
 1371                 ip6->ip6_flow |= sc->sc_flowlabel;
 1372 
 1373                 th = (struct tcphdr *)(ip6 + 1);
 1374         }
 1375 #endif
 1376 #if defined(INET6) && defined(INET)
 1377         else
 1378 #endif
 1379 #ifdef INET
 1380         {
 1381                 ip = mtod(m, struct ip *);
 1382                 ip->ip_v = IPVERSION;
 1383                 ip->ip_hl = sizeof(struct ip) >> 2;
 1384                 ip->ip_len = tlen;
 1385                 ip->ip_id = 0;
 1386                 ip->ip_off = 0;
 1387                 ip->ip_sum = 0;
 1388                 ip->ip_p = IPPROTO_TCP;
 1389                 ip->ip_src = sc->sc_inc.inc_laddr;
 1390                 ip->ip_dst = sc->sc_inc.inc_faddr;
 1391                 ip->ip_ttl = sc->sc_ip_ttl;
 1392                 ip->ip_tos = sc->sc_ip_tos;
 1393 
 1394                 /*
 1395                  * See if we should do MTU discovery.  Route lookups are
 1396                  * expensive, so we will only unset the DF bit if:
 1397                  *
 1398                  *      1) path_mtu_discovery is disabled
 1399                  *      2) the SCF_UNREACH flag has been set
 1400                  */
 1401                 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
 1402                        ip->ip_off |= IP_DF;
 1403 
 1404                 th = (struct tcphdr *)(ip + 1);
 1405         }
 1406 #endif /* INET */
 1407         th->th_sport = sc->sc_inc.inc_lport;
 1408         th->th_dport = sc->sc_inc.inc_fport;
 1409 
 1410         th->th_seq = htonl(sc->sc_iss);
 1411         th->th_ack = htonl(sc->sc_irs + 1);
 1412         th->th_off = sizeof(struct tcphdr) >> 2;
 1413         th->th_x2 = 0;
 1414         th->th_flags = TH_SYN|TH_ACK;
 1415         th->th_win = htons(sc->sc_wnd);
 1416         th->th_urp = 0;
 1417 
 1418         if (sc->sc_flags & SCF_ECN) {
 1419                 th->th_flags |= TH_ECE;
 1420                 TCPSTAT_INC(tcps_ecn_shs);
 1421         }
 1422 
 1423         /* Tack on the TCP options. */
 1424         if ((sc->sc_flags & SCF_NOOPT) == 0) {
 1425                 to.to_flags = 0;
 1426 
 1427                 to.to_mss = mssopt;
 1428                 to.to_flags = TOF_MSS;
 1429                 if (sc->sc_flags & SCF_WINSCALE) {
 1430                         to.to_wscale = sc->sc_requested_r_scale;
 1431                         to.to_flags |= TOF_SCALE;
 1432                 }
 1433                 if (sc->sc_flags & SCF_TIMESTAMP) {
 1434                         /* Virgin timestamp or TCP cookie enhanced one. */
 1435                         to.to_tsval = sc->sc_ts;
 1436                         to.to_tsecr = sc->sc_tsreflect;
 1437                         to.to_flags |= TOF_TS;
 1438                 }
 1439                 if (sc->sc_flags & SCF_SACK)
 1440                         to.to_flags |= TOF_SACKPERM;
 1441 #ifdef TCP_SIGNATURE
 1442                 if (sc->sc_flags & SCF_SIGNATURE)
 1443                         to.to_flags |= TOF_SIGNATURE;
 1444 #endif
 1445                 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
 1446 
 1447                 /* Adjust headers by option size. */
 1448                 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
 1449                 m->m_len += optlen;
 1450                 m->m_pkthdr.len += optlen;
 1451 
 1452 #ifdef TCP_SIGNATURE
 1453                 if (sc->sc_flags & SCF_SIGNATURE)
 1454                         tcp_signature_compute(m, 0, 0, optlen,
 1455                             to.to_signature, IPSEC_DIR_OUTBOUND);
 1456 #endif
 1457 #ifdef INET6
 1458                 if (sc->sc_inc.inc_flags & INC_ISIPV6)
 1459                         ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
 1460                 else
 1461 #endif
 1462                         ip->ip_len += optlen;
 1463         } else
 1464                 optlen = 0;
 1465 
 1466         M_SETFIB(m, sc->sc_inc.inc_fibnum);
 1467 #ifdef INET6
 1468         if (sc->sc_inc.inc_flags & INC_ISIPV6) {
 1469                 th->th_sum = 0;
 1470                 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
 1471                                        tlen + optlen - hlen);
 1472                 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
 1473                 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
 1474         }
 1475 #endif
 1476 #if defined(INET6) && defined(INET)
 1477         else
 1478 #endif
 1479 #ifdef INET
 1480         {
 1481                 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
 1482                     htons(tlen + optlen - hlen + IPPROTO_TCP));
 1483                 m->m_pkthdr.csum_flags = CSUM_TCP;
 1484                 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
 1485                 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
 1486         }
 1487 #endif
 1488         return (error);
 1489 }
 1490 
 1491 void
 1492 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
 1493     struct inpcb *inp, struct socket **lsop, struct mbuf *m)
 1494 {
 1495         _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
 1496 }
 1497 
 1498 void
 1499 tcp_offload_syncache_add(struct in_conninfo *inc, struct toeopt *toeo,
 1500     struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
 1501     struct toe_usrreqs *tu, void *toepcb)
 1502 {
 1503         struct tcpopt to;
 1504 
 1505         bzero(&to, sizeof(struct tcpopt));
 1506         to.to_mss = toeo->to_mss;
 1507         to.to_wscale = toeo->to_wscale;
 1508         to.to_flags = toeo->to_flags;
 1509 
 1510         INP_INFO_WLOCK(&V_tcbinfo);
 1511         INP_WLOCK(inp);
 1512 
 1513         _syncache_add(inc, &to, th, inp, lsop, NULL, tu, toepcb);
 1514 }
 1515 
 1516 /*
 1517  * The purpose of SYN cookies is to avoid keeping track of all SYN's we
 1518  * receive and to be able to handle SYN floods from bogus source addresses
 1519  * (where we will never receive any reply).  SYN floods try to exhaust all
 1520  * our memory and available slots in the SYN cache table to cause a denial
 1521  * of service to legitimate users of the local host.
 1522  *
 1523  * The idea of SYN cookies is to encode and include all necessary information
 1524  * about the connection setup state within the SYN-ACK we send back and thus
 1525  * to get along without keeping any local state until the ACK to the SYN-ACK
 1526  * arrives (if ever).  Everything we need to know should be available from
 1527  * the information we encoded in the SYN-ACK.
 1528  *
 1529  * More information about the theory behind SYN cookies and its first
 1530  * discussion and specification can be found at:
 1531  *  http://cr.yp.to/syncookies.html    (overview)
 1532  *  http://cr.yp.to/syncookies/archive (gory details)
 1533  *
 1534  * This implementation extends the orginal idea and first implementation
 1535  * of FreeBSD by using not only the initial sequence number field to store
 1536  * information but also the timestamp field if present.  This way we can
 1537  * keep track of the entire state we need to know to recreate the session in
 1538  * its original form.  Almost all TCP speakers implement RFC1323 timestamps
 1539  * these days.  For those that do not we still have to live with the known
 1540  * shortcomings of the ISN only SYN cookies.
 1541  *
 1542  * Cookie layers:
 1543  *
 1544  * Initial sequence number we send:
 1545  * 31|................................|0
 1546  *    DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
 1547  *    D = MD5 Digest (first dword)
 1548  *    M = MSS index
 1549  *    R = Rotation of secret
 1550  *    P = Odd or Even secret
 1551  *
 1552  * The MD5 Digest is computed with over following parameters:
 1553  *  a) randomly rotated secret
 1554  *  b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
 1555  *  c) the received initial sequence number from remote host
 1556  *  d) the rotation offset and odd/even bit
 1557  *
 1558  * Timestamp we send:
 1559  * 31|................................|0
 1560  *    DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
 1561  *    D = MD5 Digest (third dword) (only as filler)
 1562  *    S = Requested send window scale
 1563  *    R = Requested receive window scale
 1564  *    A = SACK allowed
 1565  *    5 = TCP-MD5 enabled (not implemented yet)
 1566  *    XORed with MD5 Digest (forth dword)
 1567  *
 1568  * The timestamp isn't cryptographically secure and doesn't need to be.
 1569  * The double use of the MD5 digest dwords ties it to a specific remote/
 1570  * local host/port, remote initial sequence number and our local time
 1571  * limited secret.  A received timestamp is reverted (XORed) and then
 1572  * the contained MD5 dword is compared to the computed one to ensure the
 1573  * timestamp belongs to the SYN-ACK we sent.  The other parameters may
 1574  * have been tampered with but this isn't different from supplying bogus
 1575  * values in the SYN in the first place.
 1576  *
 1577  * Some problems with SYN cookies remain however:
 1578  * Consider the problem of a recreated (and retransmitted) cookie.  If the
 1579  * original SYN was accepted, the connection is established.  The second
 1580  * SYN is inflight, and if it arrives with an ISN that falls within the
 1581  * receive window, the connection is killed.
 1582  *
 1583  * Notes:
 1584  * A heuristic to determine when to accept syn cookies is not necessary.
 1585  * An ACK flood would cause the syncookie verification to be attempted,
 1586  * but a SYN flood causes syncookies to be generated.  Both are of equal
 1587  * cost, so there's no point in trying to optimize the ACK flood case.
 1588  * Also, if you don't process certain ACKs for some reason, then all someone
 1589  * would have to do is launch a SYN and ACK flood at the same time, which
 1590  * would stop cookie verification and defeat the entire purpose of syncookies.
 1591  */
 1592 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
 1593 
 1594 static void
 1595 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
 1596     u_int32_t *flowlabel)
 1597 {
 1598         MD5_CTX ctx;
 1599         u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
 1600         u_int32_t data;
 1601         u_int32_t *secbits;
 1602         u_int off, pmss, mss;
 1603         int i;
 1604 
 1605         SCH_LOCK_ASSERT(sch);
 1606 
 1607         /* Which of the two secrets to use. */
 1608         secbits = sch->sch_oddeven ?
 1609                         sch->sch_secbits_odd : sch->sch_secbits_even;
 1610 
 1611         /* Reseed secret if too old. */
 1612         if (sch->sch_reseed < time_uptime) {
 1613                 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1;    /* toggle */
 1614                 secbits = sch->sch_oddeven ?
 1615                                 sch->sch_secbits_odd : sch->sch_secbits_even;
 1616                 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
 1617                         secbits[i] = arc4random();
 1618                 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
 1619         }
 1620 
 1621         /* Secret rotation offset. */
 1622         off = sc->sc_iss & 0x7;                 /* iss was randomized before */
 1623 
 1624         /* Maximum segment size calculation. */
 1625         pmss =
 1626             max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss);
 1627         for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
 1628                 if (tcp_sc_msstab[mss] <= pmss)
 1629                         break;
 1630 
 1631         /* Fold parameters and MD5 digest into the ISN we will send. */
 1632         data = sch->sch_oddeven;/* odd or even secret, 1 bit */
 1633         data |= off << 1;       /* secret offset, derived from iss, 3 bits */
 1634         data |= mss << 4;       /* mss, 3 bits */
 1635 
 1636         MD5Init(&ctx);
 1637         MD5Update(&ctx, ((u_int8_t *)secbits) + off,
 1638             SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
 1639         MD5Update(&ctx, secbits, off);
 1640         MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
 1641         MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
 1642         MD5Update(&ctx, &data, sizeof(data));
 1643         MD5Final((u_int8_t *)&md5_buffer, &ctx);
 1644 
 1645         data |= (md5_buffer[0] << 7);
 1646         sc->sc_iss = data;
 1647 
 1648 #ifdef INET6
 1649         *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
 1650 #endif
 1651 
 1652         /* Additional parameters are stored in the timestamp if present. */
 1653         if (sc->sc_flags & SCF_TIMESTAMP) {
 1654                 data =  ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
 1655                 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
 1656                 data |= sc->sc_requested_s_scale << 2;  /* SWIN scale, 4 bits */
 1657                 data |= sc->sc_requested_r_scale << 6;  /* RWIN scale, 4 bits */
 1658                 data |= md5_buffer[2] << 10;            /* more digest bits */
 1659                 data ^= md5_buffer[3];
 1660                 sc->sc_ts = data;
 1661                 sc->sc_tsoff = data - ticks;            /* after XOR */
 1662         }
 1663 
 1664         TCPSTAT_INC(tcps_sc_sendcookie);
 1665 }
 1666 
 1667 static struct syncache *
 1668 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, 
 1669     struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
 1670     struct socket *so)
 1671 {
 1672         MD5_CTX ctx;
 1673         u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
 1674         u_int32_t data = 0;
 1675         u_int32_t *secbits;
 1676         tcp_seq ack, seq;
 1677         int off, mss, wnd, flags;
 1678 
 1679         SCH_LOCK_ASSERT(sch);
 1680 
 1681         /*
 1682          * Pull information out of SYN-ACK/ACK and
 1683          * revert sequence number advances.
 1684          */
 1685         ack = th->th_ack - 1;
 1686         seq = th->th_seq - 1;
 1687         off = (ack >> 1) & 0x7;
 1688         mss = (ack >> 4) & 0x7;
 1689         flags = ack & 0x7f;
 1690 
 1691         /* Which of the two secrets to use. */
 1692         secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
 1693 
 1694         /*
 1695          * The secret wasn't updated for the lifetime of a syncookie,
 1696          * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
 1697          */
 1698         if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
 1699                 return (NULL);
 1700         }
 1701 
 1702         /* Recompute the digest so we can compare it. */
 1703         MD5Init(&ctx);
 1704         MD5Update(&ctx, ((u_int8_t *)secbits) + off,
 1705             SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
 1706         MD5Update(&ctx, secbits, off);
 1707         MD5Update(&ctx, inc, sizeof(*inc));
 1708         MD5Update(&ctx, &seq, sizeof(seq));
 1709         MD5Update(&ctx, &flags, sizeof(flags));
 1710         MD5Final((u_int8_t *)&md5_buffer, &ctx);
 1711 
 1712         /* Does the digest part of or ACK'ed ISS match? */
 1713         if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
 1714                 return (NULL);
 1715 
 1716         /* Does the digest part of our reflected timestamp match? */
 1717         if (to->to_flags & TOF_TS) {
 1718                 data = md5_buffer[3] ^ to->to_tsecr;
 1719                 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
 1720                         return (NULL);
 1721         }
 1722 
 1723         /* Fill in the syncache values. */
 1724         bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
 1725         sc->sc_ipopts = NULL;
 1726         
 1727         sc->sc_irs = seq;
 1728         sc->sc_iss = ack;
 1729 
 1730 #ifdef INET6
 1731         if (inc->inc_flags & INC_ISIPV6) {
 1732                 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
 1733                         sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
 1734         } else
 1735 #endif
 1736         {
 1737                 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
 1738                 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
 1739         }
 1740 
 1741         /* Additional parameters that were encoded in the timestamp. */
 1742         if (data) {
 1743                 sc->sc_flags |= SCF_TIMESTAMP;
 1744                 sc->sc_tsreflect = to->to_tsval;
 1745                 sc->sc_ts = to->to_tsecr;
 1746                 sc->sc_tsoff = to->to_tsecr - ticks;
 1747                 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
 1748                 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
 1749                 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
 1750                     TCP_MAX_WINSHIFT);
 1751                 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
 1752                     TCP_MAX_WINSHIFT);
 1753                 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
 1754                         sc->sc_flags |= SCF_WINSCALE;
 1755         } else
 1756                 sc->sc_flags |= SCF_NOOPT;
 1757 
 1758         wnd = sbspace(&so->so_rcv);
 1759         wnd = imax(wnd, 0);
 1760         wnd = imin(wnd, TCP_MAXWIN);
 1761         sc->sc_wnd = wnd;
 1762 
 1763         sc->sc_rxmits = 0;
 1764         sc->sc_peer_mss = tcp_sc_msstab[mss];
 1765 
 1766         TCPSTAT_INC(tcps_sc_recvcookie);
 1767         return (sc);
 1768 }
 1769 
 1770 /*
 1771  * Returns the current number of syncache entries.  This number
 1772  * will probably change before you get around to calling 
 1773  * syncache_pcblist.
 1774  */
 1775 
 1776 int
 1777 syncache_pcbcount(void)
 1778 {
 1779         struct syncache_head *sch;
 1780         int count, i;
 1781 
 1782         for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
 1783                 /* No need to lock for a read. */
 1784                 sch = &V_tcp_syncache.hashbase[i];
 1785                 count += sch->sch_length;
 1786         }
 1787         return count;
 1788 }
 1789 
 1790 /*
 1791  * Exports the syncache entries to userland so that netstat can display
 1792  * them alongside the other sockets.  This function is intended to be
 1793  * called only from tcp_pcblist.
 1794  *
 1795  * Due to concurrency on an active system, the number of pcbs exported
 1796  * may have no relation to max_pcbs.  max_pcbs merely indicates the
 1797  * amount of space the caller allocated for this function to use.
 1798  */
 1799 int
 1800 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
 1801 {
 1802         struct xtcpcb xt;
 1803         struct syncache *sc;
 1804         struct syncache_head *sch;
 1805         int count, error, i;
 1806 
 1807         for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
 1808                 sch = &V_tcp_syncache.hashbase[i];
 1809                 SCH_LOCK(sch);
 1810                 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
 1811                         if (count >= max_pcbs) {
 1812                                 SCH_UNLOCK(sch);
 1813                                 goto exit;
 1814                         }
 1815                         if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
 1816                                 continue;
 1817                         bzero(&xt, sizeof(xt));
 1818                         xt.xt_len = sizeof(xt);
 1819                         if (sc->sc_inc.inc_flags & INC_ISIPV6)
 1820                                 xt.xt_inp.inp_vflag = INP_IPV6;
 1821                         else
 1822                                 xt.xt_inp.inp_vflag = INP_IPV4;
 1823                         bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
 1824                         xt.xt_tp.t_inpcb = &xt.xt_inp;
 1825                         xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
 1826                         xt.xt_socket.xso_protocol = IPPROTO_TCP;
 1827                         xt.xt_socket.xso_len = sizeof (struct xsocket);
 1828                         xt.xt_socket.so_type = SOCK_STREAM;
 1829                         xt.xt_socket.so_state = SS_ISCONNECTING;
 1830                         error = SYSCTL_OUT(req, &xt, sizeof xt);
 1831                         if (error) {
 1832                                 SCH_UNLOCK(sch);
 1833                                 goto exit;
 1834                         }
 1835                         count++;
 1836                 }
 1837                 SCH_UNLOCK(sch);
 1838         }
 1839 exit:
 1840         *pcbs_exported = count;
 1841         return error;
 1842 }

Cache object: 956c31800b5193d84d8d0b130ea4b189


[ 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.