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/7.4/sys/netinet/tcp_syncache.c 211601 2010-08-22 08:47:51Z andre $");
   35 
   36 #include "opt_inet.h"
   37 #include "opt_inet6.h"
   38 #include "opt_ipsec.h"
   39 #include "opt_mac.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 
   63 #include <netinet/in.h>
   64 #include <netinet/in_systm.h>
   65 #include <netinet/ip.h>
   66 #include <netinet/in_var.h>
   67 #include <netinet/in_pcb.h>
   68 #include <netinet/ip_var.h>
   69 #include <netinet/ip_options.h>
   70 #ifdef INET6
   71 #include <netinet/ip6.h>
   72 #include <netinet/icmp6.h>
   73 #include <netinet6/nd6.h>
   74 #include <netinet6/ip6_var.h>
   75 #include <netinet6/in6_pcb.h>
   76 #endif
   77 #include <netinet/tcp.h>
   78 #include <netinet/tcp_fsm.h>
   79 #include <netinet/tcp_seq.h>
   80 #include <netinet/tcp_timer.h>
   81 #include <netinet/tcp_var.h>
   82 #include <netinet/tcp_syncache.h>
   83 #include <netinet/tcp_offload.h>
   84 #ifdef INET6
   85 #include <netinet6/tcp6_var.h>
   86 #endif
   87 
   88 #ifdef IPSEC
   89 #include <netipsec/ipsec.h>
   90 #ifdef INET6
   91 #include <netipsec/ipsec6.h>
   92 #endif
   93 #include <netipsec/key.h>
   94 #endif /*IPSEC*/
   95 
   96 #include <machine/in_cksum.h>
   97 
   98 #include <security/mac/mac_framework.h>
   99 
  100 static int tcp_syncookies = 1;
  101 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
  102     &tcp_syncookies, 0,
  103     "Use TCP SYN cookies if the syncache overflows");
  104 
  105 static int tcp_syncookiesonly = 0;
  106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
  107     &tcp_syncookiesonly, 0,
  108     "Use only TCP SYN cookies");
  109 
  110 #define SYNCOOKIE_SECRET_SIZE   8       /* dwords */
  111 #define SYNCOOKIE_LIFETIME      16      /* seconds */
  112 
  113 struct syncache {
  114         TAILQ_ENTRY(syncache)   sc_hash;
  115         struct          in_conninfo sc_inc;     /* addresses */
  116         int             sc_rxttime;             /* retransmit time */
  117         u_int16_t       sc_rxmits;              /* retransmit counter */
  118         u_int32_t       sc_tsreflect;           /* timestamp to reflect */
  119         u_int32_t       sc_ts;                  /* our timestamp to send */
  120         u_int32_t       sc_tsoff;               /* ts offset w/ syncookies */
  121         u_int32_t       sc_flowlabel;           /* IPv6 flowlabel */
  122         tcp_seq         sc_irs;                 /* seq from peer */
  123         tcp_seq         sc_iss;                 /* our ISS */
  124         struct          mbuf *sc_ipopts;        /* source route */
  125         u_int16_t       sc_peer_mss;            /* peer's MSS */
  126         u_int16_t       sc_wnd;                 /* advertised window */
  127         u_int8_t        sc_ip_ttl;              /* IPv4 TTL */
  128         u_int8_t        sc_ip_tos;              /* IPv4 TOS */
  129         u_int8_t        sc_requested_s_scale:4,
  130                         sc_requested_r_scale:4;
  131         u_int8_t        sc_flags;
  132 #ifndef TCP_OFFLOAD_DISABLE
  133         struct toe_usrreqs *sc_tu;              /* TOE operations */
  134         void            *sc_toepcb;             /* TOE protocol block */
  135 #endif                  
  136 #ifdef MAC
  137         struct label    *sc_label;              /* MAC label reference */
  138 #endif
  139         struct ucred    *sc_cred;               /* cred cache for jail checks */
  140 };
  141 
  142 /*
  143  * Flags for the sc_flags field.
  144  */
  145 #define SCF_NOOPT       0x01                    /* no TCP options */
  146 #define SCF_WINSCALE    0x02                    /* negotiated window scaling */
  147 #define SCF_TIMESTAMP   0x04                    /* negotiated timestamps */
  148                                                 /* MSS is implicit */
  149 #define SCF_UNREACH     0x10                    /* icmp unreachable received */
  150 #define SCF_SIGNATURE   0x20                    /* send MD5 digests */
  151 #define SCF_SACK        0x80                    /* send SACK option */
  152 
  153 #ifdef TCP_OFFLOAD_DISABLE
  154 #define TOEPCB_ISSET(sc) (0)
  155 #else
  156 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL)
  157 #endif
  158 
  159 
  160 struct syncache_head {
  161         struct mtx      sch_mtx;
  162         TAILQ_HEAD(sch_head, syncache)  sch_bucket;
  163         struct callout  sch_timer;
  164         int             sch_nextc;
  165         u_int           sch_length;
  166         u_int           sch_oddeven;
  167         u_int32_t       sch_secbits_odd[SYNCOOKIE_SECRET_SIZE];
  168         u_int32_t       sch_secbits_even[SYNCOOKIE_SECRET_SIZE];
  169         u_int           sch_reseed;             /* time_uptime, seconds */
  170 };
  171 
  172 static void      syncache_drop(struct syncache *, struct syncache_head *);
  173 static void      syncache_free(struct syncache *);
  174 static void      syncache_insert(struct syncache *, struct syncache_head *);
  175 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
  176 static int       syncache_respond(struct syncache *);
  177 static struct    socket *syncache_socket(struct syncache *, struct socket *,
  178                     struct mbuf *m);
  179 static void      syncache_timeout(struct syncache *sc, struct syncache_head *sch,
  180                     int docallout);
  181 static void      syncache_timer(void *);
  182 static void      syncookie_generate(struct syncache_head *, struct syncache *,
  183                     u_int32_t *);
  184 static struct syncache
  185                 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
  186                     struct syncache *, struct tcpopt *, struct tcphdr *,
  187                     struct socket *);
  188 
  189 /*
  190  * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
  191  * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
  192  * the odds are that the user has given up attempting to connect by then.
  193  */
  194 #define SYNCACHE_MAXREXMTS              3
  195 
  196 /* Arbitrary values */
  197 #define TCP_SYNCACHE_HASHSIZE           512
  198 #define TCP_SYNCACHE_BUCKETLIMIT        30
  199 
  200 struct tcp_syncache {
  201         struct  syncache_head *hashbase;
  202         uma_zone_t zone;
  203         u_int   hashsize;
  204         u_int   hashmask;
  205         u_int   bucket_limit;
  206         u_int   cache_count;            /* XXX: unprotected */
  207         u_int   cache_limit;
  208         u_int   rexmt_limit;
  209         u_int   hash_secret;
  210 };
  211 static struct tcp_syncache tcp_syncache;
  212 
  213 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
  214 
  215 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
  216      &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
  217 
  218 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
  219      &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
  220 
  221 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
  222      &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
  223 
  224 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
  225      &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
  226 
  227 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
  228      &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
  229 
  230 int     tcp_sc_rst_sock_fail = 1;
  231 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, CTLFLAG_RW,
  232      &tcp_sc_rst_sock_fail, 0, "Send reset on socket allocation failure");
  233 
  234 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
  235 
  236 #define SYNCACHE_HASH(inc, mask)                                        \
  237         ((tcp_syncache.hash_secret ^                                    \
  238           (inc)->inc_faddr.s_addr ^                                     \
  239           ((inc)->inc_faddr.s_addr >> 16) ^                             \
  240           (inc)->inc_fport ^ (inc)->inc_lport) & mask)
  241 
  242 #define SYNCACHE_HASH6(inc, mask)                                       \
  243         ((tcp_syncache.hash_secret ^                                    \
  244           (inc)->inc6_faddr.s6_addr32[0] ^                              \
  245           (inc)->inc6_faddr.s6_addr32[3] ^                              \
  246           (inc)->inc_fport ^ (inc)->inc_lport) & mask)
  247 
  248 #define ENDPTS_EQ(a, b) (                                               \
  249         (a)->ie_fport == (b)->ie_fport &&                               \
  250         (a)->ie_lport == (b)->ie_lport &&                               \
  251         (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr &&                 \
  252         (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr                    \
  253 )
  254 
  255 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
  256 
  257 #define SCH_LOCK(sch)           mtx_lock(&(sch)->sch_mtx)
  258 #define SCH_UNLOCK(sch)         mtx_unlock(&(sch)->sch_mtx)
  259 #define SCH_LOCK_ASSERT(sch)    mtx_assert(&(sch)->sch_mtx, MA_OWNED)
  260 
  261 /*
  262  * Requires the syncache entry to be already removed from the bucket list.
  263  */
  264 static void
  265 syncache_free(struct syncache *sc)
  266 {
  267         if (sc->sc_ipopts)
  268                 (void) m_free(sc->sc_ipopts);
  269         if (sc->sc_cred)
  270                 crfree(sc->sc_cred);
  271 #ifdef MAC
  272         mac_destroy_syncache(&sc->sc_label);
  273 #endif
  274 
  275         uma_zfree(tcp_syncache.zone, sc);
  276 }
  277 
  278 void
  279 syncache_init(void)
  280 {
  281         int i;
  282 
  283         tcp_syncache.cache_count = 0;
  284         tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
  285         tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
  286         tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
  287         tcp_syncache.hash_secret = arc4random();
  288 
  289         TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
  290             &tcp_syncache.hashsize);
  291         TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
  292             &tcp_syncache.bucket_limit);
  293         if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) {
  294                 printf("WARNING: syncache hash size is not a power of 2.\n");
  295                 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
  296         }
  297         tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
  298 
  299         /* Set limits. */
  300         tcp_syncache.cache_limit =
  301             tcp_syncache.hashsize * tcp_syncache.bucket_limit;
  302         TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
  303             &tcp_syncache.cache_limit);
  304 
  305         /* Allocate the hash table. */
  306         MALLOC(tcp_syncache.hashbase, struct syncache_head *,
  307             tcp_syncache.hashsize * sizeof(struct syncache_head),
  308             M_SYNCACHE, M_WAITOK | M_ZERO);
  309 
  310         /* Initialize the hash buckets. */
  311         for (i = 0; i < tcp_syncache.hashsize; i++) {
  312                 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
  313                 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
  314                          NULL, MTX_DEF);
  315                 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer,
  316                          &tcp_syncache.hashbase[i].sch_mtx, 0);
  317                 tcp_syncache.hashbase[i].sch_length = 0;
  318         }
  319 
  320         /* Create the syncache entry zone. */
  321         tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
  322             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
  323         uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit);
  324 }
  325 
  326 /*
  327  * Inserts a syncache entry into the specified bucket row.
  328  * Locks and unlocks the syncache_head autonomously.
  329  */
  330 static void
  331 syncache_insert(struct syncache *sc, struct syncache_head *sch)
  332 {
  333         struct syncache *sc2;
  334 
  335         SCH_LOCK(sch);
  336 
  337         /*
  338          * Make sure that we don't overflow the per-bucket limit.
  339          * If the bucket is full, toss the oldest element.
  340          */
  341         if (sch->sch_length >= tcp_syncache.bucket_limit) {
  342                 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
  343                         ("sch->sch_length incorrect"));
  344                 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
  345                 syncache_drop(sc2, sch);
  346                 tcpstat.tcps_sc_bucketoverflow++;
  347         }
  348 
  349         /* Put it into the bucket. */
  350         TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
  351         sch->sch_length++;
  352 
  353         /* Reinitialize the bucket row's timer. */
  354         if (sch->sch_length == 1)
  355                 sch->sch_nextc = ticks + INT_MAX;
  356         syncache_timeout(sc, sch, 1);
  357 
  358         SCH_UNLOCK(sch);
  359 
  360         tcp_syncache.cache_count++;
  361         tcpstat.tcps_sc_added++;
  362 }
  363 
  364 /*
  365  * Remove and free entry from syncache bucket row.
  366  * Expects locked syncache head.
  367  */
  368 static void
  369 syncache_drop(struct syncache *sc, struct syncache_head *sch)
  370 {
  371 
  372         SCH_LOCK_ASSERT(sch);
  373 
  374         TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
  375         sch->sch_length--;
  376 
  377 #ifndef TCP_OFFLOAD_DISABLE
  378         if (sc->sc_tu)
  379                 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
  380 #endif              
  381         syncache_free(sc);
  382         tcp_syncache.cache_count--;
  383 }
  384 
  385 /*
  386  * Engage/reengage time on bucket row.
  387  */
  388 static void
  389 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
  390 {
  391         sc->sc_rxttime = ticks +
  392                 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
  393         sc->sc_rxmits++;
  394         if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
  395                 sch->sch_nextc = sc->sc_rxttime;
  396                 if (docallout)
  397                         callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
  398                             syncache_timer, (void *)sch);
  399         }
  400 }
  401 
  402 /*
  403  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
  404  * If we have retransmitted an entry the maximum number of times, expire it.
  405  * One separate timer for each bucket row.
  406  */
  407 static void
  408 syncache_timer(void *xsch)
  409 {
  410         struct syncache_head *sch = (struct syncache_head *)xsch;
  411         struct syncache *sc, *nsc;
  412         int tick = ticks;
  413         char *s;
  414 
  415         /* NB: syncache_head has already been locked by the callout. */
  416         SCH_LOCK_ASSERT(sch);
  417 
  418         /*
  419          * In the following cycle we may remove some entries and/or
  420          * advance some timeouts, so re-initialize the bucket timer.
  421          */
  422         sch->sch_nextc = tick + INT_MAX;
  423 
  424         TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
  425                 /*
  426                  * We do not check if the listen socket still exists
  427                  * and accept the case where the listen socket may be
  428                  * gone by the time we resend the SYN/ACK.  We do
  429                  * not expect this to happens often. If it does,
  430                  * then the RST will be sent by the time the remote
  431                  * host does the SYN/ACK->ACK.
  432                  */
  433                 if (TSTMP_GT(sc->sc_rxttime, tick)) {
  434                         if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
  435                                 sch->sch_nextc = sc->sc_rxttime;
  436                         continue;
  437                 }
  438 
  439                 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) {
  440                         if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  441                                 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
  442                                     "giving up and removing syncache entry\n",
  443                                     s, __func__);
  444                                 free(s, M_TCPLOG);
  445                         }
  446                         syncache_drop(sc, sch);
  447                         tcpstat.tcps_sc_stale++;
  448                         continue;
  449                 }
  450                 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  451                         log(LOG_DEBUG, "%s; %s: Response timeout, "
  452                             "retransmitting (%u) SYN|ACK\n",
  453                             s, __func__, sc->sc_rxmits);
  454                         free(s, M_TCPLOG);
  455                 }
  456 
  457                 (void) syncache_respond(sc);
  458                 tcpstat.tcps_sc_retransmitted++;
  459                 syncache_timeout(sc, sch, 0);
  460         }
  461         if (!TAILQ_EMPTY(&(sch)->sch_bucket))
  462                 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
  463                         syncache_timer, (void *)(sch));
  464 }
  465 
  466 /*
  467  * Find an entry in the syncache.
  468  * Returns always with locked syncache_head plus a matching entry or NULL.
  469  */
  470 struct syncache *
  471 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
  472 {
  473         struct syncache *sc;
  474         struct syncache_head *sch;
  475 
  476 #ifdef INET6
  477         if (inc->inc_flags & INC_ISIPV6) {
  478                 sch = &tcp_syncache.hashbase[
  479                     SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
  480                 *schp = sch;
  481 
  482                 SCH_LOCK(sch);
  483 
  484                 /* Circle through bucket row to find matching entry. */
  485                 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
  486                         if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
  487                                 return (sc);
  488                 }
  489         } else
  490 #endif
  491         {
  492                 sch = &tcp_syncache.hashbase[
  493                     SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
  494                 *schp = sch;
  495 
  496                 SCH_LOCK(sch);
  497 
  498                 /* Circle through bucket row to find matching entry. */
  499                 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
  500 #ifdef INET6
  501                         if (sc->sc_inc.inc_flags & INC_ISIPV6)
  502                                 continue;
  503 #endif
  504                         if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
  505                                 return (sc);
  506                 }
  507         }
  508         SCH_LOCK_ASSERT(*schp);
  509         return (NULL);                  /* always returns with locked sch */
  510 }
  511 
  512 /*
  513  * This function is called when we get a RST for a
  514  * non-existent connection, so that we can see if the
  515  * connection is in the syn cache.  If it is, zap it.
  516  */
  517 void
  518 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
  519 {
  520         struct syncache *sc;
  521         struct syncache_head *sch;
  522         char *s = NULL;
  523 
  524         sc = syncache_lookup(inc, &sch);        /* returns locked sch */
  525         SCH_LOCK_ASSERT(sch);
  526 
  527         /*
  528          * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
  529          * See RFC 793 page 65, section SEGMENT ARRIVES.
  530          */
  531         if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
  532                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  533                         log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
  534                             "FIN flag set, segment ignored\n", s, __func__);
  535                 tcpstat.tcps_badrst++;
  536                 goto done;
  537         }
  538 
  539         /*
  540          * No corresponding connection was found in syncache.
  541          * If syncookies are enabled and possibly exclusively
  542          * used, or we are under memory pressure, a valid RST
  543          * may not find a syncache entry.  In that case we're
  544          * done and no SYN|ACK retransmissions will happen.
  545          * Otherwise the the RST was misdirected or spoofed.
  546          */
  547         if (sc == NULL) {
  548                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  549                         log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
  550                             "syncache entry (possibly syncookie only), "
  551                             "segment ignored\n", s, __func__);
  552                 tcpstat.tcps_badrst++;
  553                 goto done;
  554         }
  555 
  556         /*
  557          * If the RST bit is set, check the sequence number to see
  558          * if this is a valid reset segment.
  559          * RFC 793 page 37:
  560          *   In all states except SYN-SENT, all reset (RST) segments
  561          *   are validated by checking their SEQ-fields.  A reset is
  562          *   valid if its sequence number is in the window.
  563          *
  564          *   The sequence number in the reset segment is normally an
  565          *   echo of our outgoing acknowlegement numbers, but some hosts
  566          *   send a reset with the sequence number at the rightmost edge
  567          *   of our receive window, and we have to handle this case.
  568          */
  569         if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
  570             SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
  571                 syncache_drop(sc, sch);
  572                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  573                         log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
  574                             "connection attempt aborted by remote endpoint\n",
  575                             s, __func__);
  576                 tcpstat.tcps_sc_reset++;
  577         } else {
  578                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  579                         log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
  580                             "IRS %u (+WND %u), segment ignored\n",
  581                             s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
  582                 tcpstat.tcps_badrst++;
  583         }
  584 
  585 done:
  586         if (s != NULL)
  587                 free(s, M_TCPLOG);
  588         SCH_UNLOCK(sch);
  589 }
  590 
  591 void
  592 syncache_badack(struct in_conninfo *inc)
  593 {
  594         struct syncache *sc;
  595         struct syncache_head *sch;
  596 
  597         sc = syncache_lookup(inc, &sch);        /* returns locked sch */
  598         SCH_LOCK_ASSERT(sch);
  599         if (sc != NULL) {
  600                 syncache_drop(sc, sch);
  601                 tcpstat.tcps_sc_badack++;
  602         }
  603         SCH_UNLOCK(sch);
  604 }
  605 
  606 void
  607 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
  608 {
  609         struct syncache *sc;
  610         struct syncache_head *sch;
  611 
  612         sc = syncache_lookup(inc, &sch);        /* returns locked sch */
  613         SCH_LOCK_ASSERT(sch);
  614         if (sc == NULL)
  615                 goto done;
  616 
  617         /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
  618         if (ntohl(th->th_seq) != sc->sc_iss)
  619                 goto done;
  620 
  621         /*
  622          * If we've rertransmitted 3 times and this is our second error,
  623          * we remove the entry.  Otherwise, we allow it to continue on.
  624          * This prevents us from incorrectly nuking an entry during a
  625          * spurious network outage.
  626          *
  627          * See tcp_notify().
  628          */
  629         if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
  630                 sc->sc_flags |= SCF_UNREACH;
  631                 goto done;
  632         }
  633         syncache_drop(sc, sch);
  634         tcpstat.tcps_sc_unreach++;
  635 done:
  636         SCH_UNLOCK(sch);
  637 }
  638 
  639 /*
  640  * Build a new TCP socket structure from a syncache entry.
  641  */
  642 static struct socket *
  643 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
  644 {
  645         struct inpcb *inp = NULL;
  646         struct socket *so;
  647         struct tcpcb *tp;
  648         int error = 0;
  649         char *s;
  650 
  651         INP_INFO_WLOCK_ASSERT(&tcbinfo);
  652 
  653         /*
  654          * Ok, create the full blown connection, and set things up
  655          * as they would have been set up if we had created the
  656          * connection when the SYN arrived.  If we can't create
  657          * the connection, abort it.
  658          */
  659         so = sonewconn(lso, SS_ISCONNECTED);
  660         if (so == NULL) {
  661                 /*
  662                  * Drop the connection; we will either send a RST or
  663                  * have the peer retransmit its SYN again after its
  664                  * RTO and try again.
  665                  */
  666                 tcpstat.tcps_listendrop++;
  667                 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  668                         log(LOG_DEBUG, "%s; %s: Socket create failed "
  669                             "due to limits or memory shortage\n",
  670                             s, __func__);
  671                         free(s, M_TCPLOG);
  672                 }
  673                 goto abort2;
  674         }
  675 #ifdef MAC
  676         SOCK_LOCK(so);
  677         mac_set_socket_peer_from_mbuf(m, so);
  678         SOCK_UNLOCK(so);
  679 #endif
  680 
  681         inp = sotoinpcb(so);
  682         inp->inp_inc.inc_fibnum = so->so_fibnum;
  683         INP_WLOCK(inp);
  684 
  685         /* Insert new socket into PCB hash list. */
  686         inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
  687 #ifdef INET6
  688         if (sc->sc_inc.inc_flags & INC_ISIPV6) {
  689                 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
  690         } else {
  691                 inp->inp_vflag &= ~INP_IPV6;
  692                 inp->inp_vflag |= INP_IPV4;
  693 #endif
  694                 inp->inp_laddr = sc->sc_inc.inc_laddr;
  695 #ifdef INET6
  696         }
  697 #endif
  698         inp->inp_lport = sc->sc_inc.inc_lport;
  699         if ((error = in_pcbinshash(inp)) != 0) {
  700                 /*
  701                  * Undo the assignments above if we failed to
  702                  * put the PCB on the hash lists.
  703                  */
  704 #ifdef INET6
  705                 if (sc->sc_inc.inc_flags & INC_ISIPV6)
  706                         inp->in6p_laddr = in6addr_any;
  707                 else
  708 #endif
  709                         inp->inp_laddr.s_addr = INADDR_ANY;
  710                 inp->inp_lport = 0;
  711                 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  712                         log(LOG_DEBUG, "%s; %s: in_pcbinshash failed "
  713                             "with error %i\n",
  714                             s, __func__, error);
  715                         free(s, M_TCPLOG);
  716                 }
  717                 goto abort;
  718         }
  719 #ifdef IPSEC
  720         /* Copy old policy into new socket's. */
  721         if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
  722                 printf("syncache_socket: could not copy policy\n");
  723 #endif
  724 #ifdef INET6
  725         if (sc->sc_inc.inc_flags & INC_ISIPV6) {
  726                 struct inpcb *oinp = sotoinpcb(lso);
  727                 struct in6_addr laddr6;
  728                 struct sockaddr_in6 sin6;
  729                 /*
  730                  * Inherit socket options from the listening socket.
  731                  * Note that in6p_inputopts are not (and should not be)
  732                  * copied, since it stores previously received options and is
  733                  * used to detect if each new option is different than the
  734                  * previous one and hence should be passed to a user.
  735                  * If we copied in6p_inputopts, a user would not be able to
  736                  * receive options just after calling the accept system call.
  737                  */
  738                 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
  739                 if (oinp->in6p_outputopts)
  740                         inp->in6p_outputopts =
  741                             ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
  742 
  743                 sin6.sin6_family = AF_INET6;
  744                 sin6.sin6_len = sizeof(sin6);
  745                 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
  746                 sin6.sin6_port = sc->sc_inc.inc_fport;
  747                 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
  748                 laddr6 = inp->in6p_laddr;
  749                 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
  750                         inp->in6p_laddr = sc->sc_inc.inc6_laddr;
  751                 if ((error = in6_pcbconnect(inp, (struct sockaddr *)&sin6,
  752                     thread0.td_ucred)) != 0) {
  753                         inp->in6p_laddr = laddr6;
  754                         if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  755                                 log(LOG_DEBUG, "%s; %s: in6_pcbconnect failed "
  756                                     "with error %i\n",
  757                                     s, __func__, error);
  758                                 free(s, M_TCPLOG);
  759                         }
  760                         goto abort;
  761                 }
  762                 /* Override flowlabel from in6_pcbconnect. */
  763                 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
  764                 inp->inp_flow |= sc->sc_flowlabel;
  765         } else
  766 #endif
  767         {
  768                 struct in_addr laddr;
  769                 struct sockaddr_in sin;
  770 
  771                 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
  772                 
  773                 if (inp->inp_options == NULL) {
  774                         inp->inp_options = sc->sc_ipopts;
  775                         sc->sc_ipopts = NULL;
  776                 }
  777 
  778                 sin.sin_family = AF_INET;
  779                 sin.sin_len = sizeof(sin);
  780                 sin.sin_addr = sc->sc_inc.inc_faddr;
  781                 sin.sin_port = sc->sc_inc.inc_fport;
  782                 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
  783                 laddr = inp->inp_laddr;
  784                 if (inp->inp_laddr.s_addr == INADDR_ANY)
  785                         inp->inp_laddr = sc->sc_inc.inc_laddr;
  786                 if ((error = in_pcbconnect(inp, (struct sockaddr *)&sin,
  787                     thread0.td_ucred)) != 0) {
  788                         inp->inp_laddr = laddr;
  789                         if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
  790                                 log(LOG_DEBUG, "%s; %s: in_pcbconnect failed "
  791                                     "with error %i\n",
  792                                     s, __func__, error);
  793                                 free(s, M_TCPLOG);
  794                         }
  795                         goto abort;
  796                 }
  797         }
  798         tp = intotcpcb(inp);
  799         tp->t_state = TCPS_SYN_RECEIVED;
  800         tp->iss = sc->sc_iss;
  801         tp->irs = sc->sc_irs;
  802         tcp_rcvseqinit(tp);
  803         tcp_sendseqinit(tp);
  804         tp->snd_wl1 = sc->sc_irs;
  805         tp->snd_max = tp->iss + 1;
  806         tp->snd_nxt = tp->iss + 1;
  807         tp->rcv_up = sc->sc_irs + 1;
  808         tp->rcv_wnd = sc->sc_wnd;
  809         tp->rcv_adv += tp->rcv_wnd;
  810         tp->last_ack_sent = tp->rcv_nxt;
  811 
  812         tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
  813         if (sc->sc_flags & SCF_NOOPT)
  814                 tp->t_flags |= TF_NOOPT;
  815         else {
  816                 if (sc->sc_flags & SCF_WINSCALE) {
  817                         tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
  818                         tp->snd_scale = sc->sc_requested_s_scale;
  819                         tp->request_r_scale = sc->sc_requested_r_scale;
  820                 }
  821                 if (sc->sc_flags & SCF_TIMESTAMP) {
  822                         tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
  823                         tp->ts_recent = sc->sc_tsreflect;
  824                         tp->ts_recent_age = ticks;
  825                         tp->ts_offset = sc->sc_tsoff;
  826                 }
  827 #ifdef TCP_SIGNATURE
  828                 if (sc->sc_flags & SCF_SIGNATURE)
  829                         tp->t_flags |= TF_SIGNATURE;
  830 #endif
  831                 if (sc->sc_flags & SCF_SACK)
  832                         tp->t_flags |= TF_SACK_PERMIT;
  833         }
  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.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(&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 (!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                 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 #if 0
  948         /*
  949          * If timestamps were present in the SYN and we accepted
  950          * them in our SYN|ACK we require them to be present from
  951          * now on.  And vice versa.
  952          *
  953          * Unfortunately, during testing of 7.0 some users found
  954          * network devices that violate this constraint, so it must
  955          * be disabled.
  956          */
  957         if ((sc->sc_flags & SCF_TIMESTAMP) && !(to->to_flags & TOF_TS)) {
  958                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  959                         log(LOG_DEBUG, "%s; %s: Timestamp missing, "
  960                             "segment rejected\n", s, __func__);
  961                 goto failed;
  962         }
  963 #endif
  964         if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
  965                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  966                         log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
  967                             "segment rejected\n", s, __func__);
  968                 goto failed;
  969         }
  970         /*
  971          * If timestamps were negotiated the reflected timestamp
  972          * must be equal to what we actually sent in the SYN|ACK.
  973          */
  974         if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
  975             !TOEPCB_ISSET(sc)) {
  976                 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
  977                         log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
  978                             "segment rejected\n",
  979                             s, __func__, to->to_tsecr, sc->sc_ts);
  980                 goto failed;
  981         }
  982 
  983         *lsop = syncache_socket(sc, *lsop, m);
  984 
  985         if (*lsop == NULL)
  986                 tcpstat.tcps_sc_aborted++;
  987         else
  988                 tcpstat.tcps_sc_completed++;
  989 
  990 /* how do we find the inp for the new socket? */
  991         if (sc != &scs)
  992                 syncache_free(sc);
  993         return (1);
  994 failed:
  995         if (sc != NULL && sc != &scs)
  996                 syncache_free(sc);
  997         if (s != NULL)
  998                 free(s, M_TCPLOG);
  999         *lsop = NULL;
 1000         return (0);
 1001 }
 1002 
 1003 int
 1004 tcp_offload_syncache_expand(struct in_conninfo *inc, struct tcpopt *to,
 1005     struct tcphdr *th, struct socket **lsop, struct mbuf *m)
 1006 {
 1007         int rc;
 1008         
 1009         INP_INFO_WLOCK(&tcbinfo);
 1010         rc = syncache_expand(inc, to, th, lsop, m);
 1011         INP_INFO_WUNLOCK(&tcbinfo);
 1012 
 1013         return (rc);
 1014 }
 1015 
 1016 /*
 1017  * Given a LISTEN socket and an inbound SYN request, add
 1018  * this to the syn cache, and send back a segment:
 1019  *      <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
 1020  * to the source.
 1021  *
 1022  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
 1023  * Doing so would require that we hold onto the data and deliver it
 1024  * to the application.  However, if we are the target of a SYN-flood
 1025  * DoS attack, an attacker could send data which would eventually
 1026  * consume all available buffer space if it were ACKed.  By not ACKing
 1027  * the data, we avoid this DoS scenario.
 1028  */
 1029 static void
 1030 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
 1031     struct inpcb *inp, struct socket **lsop, struct mbuf *m,
 1032     struct toe_usrreqs *tu, void *toepcb)
 1033 {
 1034         struct tcpcb *tp;
 1035         struct socket *so;
 1036         struct syncache *sc = NULL;
 1037         struct syncache_head *sch;
 1038         struct mbuf *ipopts = NULL;
 1039         u_int32_t flowtmp;
 1040         int win, sb_hiwat, ip_ttl, ip_tos, noopt;
 1041         char *s;
 1042 #ifdef INET6
 1043         int autoflowlabel = 0;
 1044 #endif
 1045 #ifdef MAC
 1046         struct label *maclabel;
 1047 #endif
 1048         struct syncache scs;
 1049         struct ucred *cred;
 1050 
 1051         INP_INFO_WLOCK_ASSERT(&tcbinfo);
 1052         INP_WLOCK_ASSERT(inp);                  /* listen socket */
 1053         KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
 1054             ("%s: unexpected tcp flags", __func__));
 1055 
 1056         /*
 1057          * Combine all so/tp operations very early to drop the INP lock as
 1058          * soon as possible.
 1059          */
 1060         so = *lsop;
 1061         tp = sototcpcb(so);
 1062         cred = crhold(so->so_cred);
 1063 
 1064 #ifdef INET6
 1065         if ((inc->inc_flags & INC_ISIPV6) &&
 1066             (inp->inp_flags & IN6P_AUTOFLOWLABEL))
 1067                 autoflowlabel = 1;
 1068 #endif
 1069         ip_ttl = inp->inp_ip_ttl;
 1070         ip_tos = inp->inp_ip_tos;
 1071         win = sbspace(&so->so_rcv);
 1072         sb_hiwat = so->so_rcv.sb_hiwat;
 1073         noopt = (tp->t_flags & TF_NOOPT);
 1074 
 1075         /* By the time we drop the lock these should no longer be used. */
 1076         so = NULL;
 1077         tp = NULL;
 1078 
 1079 #ifdef MAC
 1080         if (mac_init_syncache(&maclabel) != 0) {
 1081                 INP_WUNLOCK(inp);
 1082                 INP_INFO_WUNLOCK(&tcbinfo);
 1083                 goto done;
 1084         } else
 1085                 mac_init_syncache_from_inpcb(maclabel, inp);
 1086 #endif
 1087         INP_WUNLOCK(inp);
 1088         INP_INFO_WUNLOCK(&tcbinfo);
 1089 
 1090         /*
 1091          * Remember the IP options, if any.
 1092          */
 1093 #ifdef INET6
 1094         if (!(inc->inc_flags & INC_ISIPV6))
 1095 #endif
 1096                 ipopts = (m) ? ip_srcroute(m) : NULL;
 1097 
 1098         /*
 1099          * See if we already have an entry for this connection.
 1100          * If we do, resend the SYN,ACK, and reset the retransmit timer.
 1101          *
 1102          * XXX: should the syncache be re-initialized with the contents
 1103          * of the new SYN here (which may have different options?)
 1104          *
 1105          * XXX: We do not check the sequence number to see if this is a
 1106          * real retransmit or a new connection attempt.  The question is
 1107          * how to handle such a case; either ignore it as spoofed, or
 1108          * drop the current entry and create a new one?
 1109          */
 1110         sc = syncache_lookup(inc, &sch);        /* returns locked entry */
 1111         SCH_LOCK_ASSERT(sch);
 1112         if (sc != NULL) {
 1113 #ifndef TCP_OFFLOAD_DISABLE
 1114                 if (sc->sc_tu)
 1115                         sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
 1116                             sc->sc_toepcb);
 1117 #endif              
 1118                 tcpstat.tcps_sc_dupsyn++;
 1119                 if (ipopts) {
 1120                         /*
 1121                          * If we were remembering a previous source route,
 1122                          * forget it and use the new one we've been given.
 1123                          */
 1124                         if (sc->sc_ipopts)
 1125                                 (void) m_free(sc->sc_ipopts);
 1126                         sc->sc_ipopts = ipopts;
 1127                 }
 1128                 /*
 1129                  * Update timestamp if present.
 1130                  */
 1131                 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
 1132                         sc->sc_tsreflect = to->to_tsval;
 1133                 else
 1134                         sc->sc_flags &= ~SCF_TIMESTAMP;
 1135 #ifdef MAC
 1136                 /*
 1137                  * Since we have already unconditionally allocated label
 1138                  * storage, free it up.  The syncache entry will already
 1139                  * have an initialized label we can use.
 1140                  */
 1141                 mac_destroy_syncache(&maclabel);
 1142                 KASSERT(sc->sc_label != NULL,
 1143                     ("%s: label not initialized", __func__));
 1144 #endif
 1145                 /* Retransmit SYN|ACK and reset retransmit count. */
 1146                 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
 1147                         log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
 1148                             "resetting timer and retransmitting SYN|ACK\n",
 1149                             s, __func__);
 1150                         free(s, M_TCPLOG);
 1151                 }
 1152                 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
 1153                         sc->sc_rxmits = 0;
 1154                         syncache_timeout(sc, sch, 1);
 1155                         tcpstat.tcps_sndacks++;
 1156                         tcpstat.tcps_sndtotal++;
 1157                 }
 1158                 SCH_UNLOCK(sch);
 1159                 goto done;
 1160         }
 1161 
 1162         sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
 1163         if (sc == NULL) {
 1164                 /*
 1165                  * The zone allocator couldn't provide more entries.
 1166                  * Treat this as if the cache was full; drop the oldest
 1167                  * entry and insert the new one.
 1168                  */
 1169                 tcpstat.tcps_sc_zonefail++;
 1170                 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
 1171                         syncache_drop(sc, sch);
 1172                 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
 1173                 if (sc == NULL) {
 1174                         if (tcp_syncookies) {
 1175                                 bzero(&scs, sizeof(scs));
 1176                                 sc = &scs;
 1177                         } else {
 1178                                 SCH_UNLOCK(sch);
 1179                                 if (ipopts)
 1180                                         (void) m_free(ipopts);
 1181                                 goto done;
 1182                         }
 1183                 }
 1184         }
 1185 
 1186         /*
 1187          * Fill in the syncache values.
 1188          */
 1189 #ifdef MAC
 1190         sc->sc_label = maclabel;
 1191 #endif
 1192         sc->sc_cred = cred;
 1193         cred = NULL;
 1194         sc->sc_ipopts = ipopts;
 1195         bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
 1196 #ifdef INET6
 1197         if (!(inc->inc_flags & INC_ISIPV6))
 1198 #endif
 1199         {
 1200                 sc->sc_ip_tos = ip_tos;
 1201                 sc->sc_ip_ttl = ip_ttl;
 1202         }
 1203 #ifndef TCP_OFFLOAD_DISABLE     
 1204         sc->sc_tu = tu;
 1205         sc->sc_toepcb = toepcb;
 1206 #endif
 1207         sc->sc_irs = th->th_seq;
 1208         sc->sc_iss = arc4random();
 1209         sc->sc_flags = 0;
 1210         sc->sc_flowlabel = 0;
 1211 
 1212         /*
 1213          * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
 1214          * win was derived from socket earlier in the function.
 1215          */
 1216         win = imax(win, 0);
 1217         win = imin(win, TCP_MAXWIN);
 1218         sc->sc_wnd = win;
 1219 
 1220         if (tcp_do_rfc1323) {
 1221                 /*
 1222                  * A timestamp received in a SYN makes
 1223                  * it ok to send timestamp requests and replies.
 1224                  */
 1225                 if (to->to_flags & TOF_TS) {
 1226                         sc->sc_tsreflect = to->to_tsval;
 1227                         sc->sc_ts = ticks;
 1228                         sc->sc_flags |= SCF_TIMESTAMP;
 1229                 }
 1230                 if (to->to_flags & TOF_SCALE) {
 1231                         int wscale = 0;
 1232 
 1233                         /*
 1234                          * Pick the smallest possible scaling factor that
 1235                          * will still allow us to scale up to sb_max, aka
 1236                          * kern.ipc.maxsockbuf.
 1237                          *
 1238                          * We do this because there are broken firewalls that
 1239                          * will corrupt the window scale option, leading to
 1240                          * the other endpoint believing that our advertised
 1241                          * window is unscaled.  At scale factors larger than
 1242                          * 5 the unscaled window will drop below 1500 bytes,
 1243                          * leading to serious problems when traversing these
 1244                          * broken firewalls.
 1245                          *
 1246                          * With the default maxsockbuf of 256K, a scale factor
 1247                          * of 3 will be chosen by this algorithm.  Those who
 1248                          * choose a larger maxsockbuf should watch out
 1249                          * for the compatiblity problems mentioned above.
 1250                          *
 1251                          * RFC1323: The Window field in a SYN (i.e., a <SYN>
 1252                          * or <SYN,ACK>) segment itself is never scaled.
 1253                          */
 1254                         while (wscale < TCP_MAX_WINSHIFT &&
 1255                             (TCP_MAXWIN << wscale) < sb_max)
 1256                                 wscale++;
 1257                         sc->sc_requested_r_scale = wscale;
 1258                         sc->sc_requested_s_scale = to->to_wscale;
 1259                         sc->sc_flags |= SCF_WINSCALE;
 1260                 }
 1261         }
 1262 #ifdef TCP_SIGNATURE
 1263         /*
 1264          * If listening socket requested TCP digests, and received SYN
 1265          * contains the option, flag this in the syncache so that
 1266          * syncache_respond() will do the right thing with the SYN+ACK.
 1267          * XXX: Currently we always record the option by default and will
 1268          * attempt to use it in syncache_respond().
 1269          */
 1270         if (to->to_flags & TOF_SIGNATURE)
 1271                 sc->sc_flags |= SCF_SIGNATURE;
 1272 #endif
 1273         if (to->to_flags & TOF_SACKPERM)
 1274                 sc->sc_flags |= SCF_SACK;
 1275         if (to->to_flags & TOF_MSS)
 1276                 sc->sc_peer_mss = to->to_mss;   /* peer mss may be zero */
 1277         if (noopt)
 1278                 sc->sc_flags |= SCF_NOOPT;
 1279 
 1280         if (tcp_syncookies) {
 1281                 syncookie_generate(sch, sc, &flowtmp);
 1282 #ifdef INET6
 1283                 if (autoflowlabel)
 1284                         sc->sc_flowlabel = flowtmp;
 1285 #endif
 1286         } else {
 1287 #ifdef INET6
 1288                 if (autoflowlabel)
 1289                         sc->sc_flowlabel =
 1290                             (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
 1291 #endif
 1292         }
 1293         SCH_UNLOCK(sch);
 1294 
 1295         /*
 1296          * Do a standard 3-way handshake.
 1297          */
 1298         if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
 1299                 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs)
 1300                         syncache_free(sc);
 1301                 else if (sc != &scs)
 1302                         syncache_insert(sc, sch);   /* locks and unlocks sch */
 1303                 tcpstat.tcps_sndacks++;
 1304                 tcpstat.tcps_sndtotal++;
 1305         } else {
 1306                 if (sc != &scs)
 1307                         syncache_free(sc);
 1308                 tcpstat.tcps_sc_dropped++;
 1309         }
 1310 
 1311 done:
 1312         if (cred != NULL)
 1313                 crfree(cred);
 1314 #ifdef MAC
 1315         if (sc == &scs)
 1316                 mac_destroy_syncache(&maclabel);
 1317 #endif
 1318         *lsop = NULL;
 1319         m_freem(m);
 1320 }
 1321 
 1322 static int
 1323 syncache_respond(struct syncache *sc)
 1324 {
 1325         struct ip *ip = NULL;
 1326         struct mbuf *m;
 1327         struct tcphdr *th;
 1328         int optlen, error;
 1329         u_int16_t hlen, tlen, mssopt;
 1330         struct tcpopt to;
 1331 #ifdef INET6
 1332         struct ip6_hdr *ip6 = NULL;
 1333 #endif
 1334 
 1335         hlen =
 1336 #ifdef INET6
 1337                (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
 1338 #endif
 1339                 sizeof(struct ip);
 1340         tlen = hlen + sizeof(struct tcphdr);
 1341 
 1342         /* Determine MSS we advertize to other end of connection. */
 1343         mssopt = tcp_mssopt(&sc->sc_inc);
 1344         if (sc->sc_peer_mss)
 1345                 mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss);
 1346 
 1347         /* XXX: Assume that the entire packet will fit in a header mbuf. */
 1348         KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
 1349             ("syncache: mbuf too small"));
 1350 
 1351         /* Create the IP+TCP header from scratch. */
 1352         m = m_gethdr(M_DONTWAIT, MT_DATA);
 1353         if (m == NULL)
 1354                 return (ENOBUFS);
 1355 #ifdef MAC
 1356         mac_create_mbuf_from_syncache(sc->sc_label, m);
 1357 #endif
 1358         m->m_data += max_linkhdr;
 1359         m->m_len = tlen;
 1360         m->m_pkthdr.len = tlen;
 1361         m->m_pkthdr.rcvif = NULL;
 1362 
 1363 #ifdef INET6
 1364         if (sc->sc_inc.inc_flags & INC_ISIPV6) {
 1365                 ip6 = mtod(m, struct ip6_hdr *);
 1366                 ip6->ip6_vfc = IPV6_VERSION;
 1367                 ip6->ip6_nxt = IPPROTO_TCP;
 1368                 ip6->ip6_src = sc->sc_inc.inc6_laddr;
 1369                 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
 1370                 ip6->ip6_plen = htons(tlen - hlen);
 1371                 /* ip6_hlim is set after checksum */
 1372                 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
 1373                 ip6->ip6_flow |= sc->sc_flowlabel;
 1374 
 1375                 th = (struct tcphdr *)(ip6 + 1);
 1376         } else
 1377 #endif
 1378         {
 1379                 ip = mtod(m, struct ip *);
 1380                 ip->ip_v = IPVERSION;
 1381                 ip->ip_hl = sizeof(struct ip) >> 2;
 1382                 ip->ip_len = tlen;
 1383                 ip->ip_id = 0;
 1384                 ip->ip_off = 0;
 1385                 ip->ip_sum = 0;
 1386                 ip->ip_p = IPPROTO_TCP;
 1387                 ip->ip_src = sc->sc_inc.inc_laddr;
 1388                 ip->ip_dst = sc->sc_inc.inc_faddr;
 1389                 ip->ip_ttl = sc->sc_ip_ttl;
 1390                 ip->ip_tos = sc->sc_ip_tos;
 1391 
 1392                 /*
 1393                  * See if we should do MTU discovery.  Route lookups are
 1394                  * expensive, so we will only unset the DF bit if:
 1395                  *
 1396                  *      1) path_mtu_discovery is disabled
 1397                  *      2) the SCF_UNREACH flag has been set
 1398                  */
 1399                 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
 1400                        ip->ip_off |= IP_DF;
 1401 
 1402                 th = (struct tcphdr *)(ip + 1);
 1403         }
 1404         th->th_sport = sc->sc_inc.inc_lport;
 1405         th->th_dport = sc->sc_inc.inc_fport;
 1406 
 1407         th->th_seq = htonl(sc->sc_iss);
 1408         th->th_ack = htonl(sc->sc_irs + 1);
 1409         th->th_off = sizeof(struct tcphdr) >> 2;
 1410         th->th_x2 = 0;
 1411         th->th_flags = TH_SYN|TH_ACK;
 1412         th->th_win = htons(sc->sc_wnd);
 1413         th->th_urp = 0;
 1414 
 1415         /* Tack on the TCP options. */
 1416         if ((sc->sc_flags & SCF_NOOPT) == 0) {
 1417                 to.to_flags = 0;
 1418 
 1419                 to.to_mss = mssopt;
 1420                 to.to_flags = TOF_MSS;
 1421                 if (sc->sc_flags & SCF_WINSCALE) {
 1422                         to.to_wscale = sc->sc_requested_r_scale;
 1423                         to.to_flags |= TOF_SCALE;
 1424                 }
 1425                 if (sc->sc_flags & SCF_TIMESTAMP) {
 1426                         /* Virgin timestamp or TCP cookie enhanced one. */
 1427                         to.to_tsval = sc->sc_ts;
 1428                         to.to_tsecr = sc->sc_tsreflect;
 1429                         to.to_flags |= TOF_TS;
 1430                 }
 1431                 if (sc->sc_flags & SCF_SACK)
 1432                         to.to_flags |= TOF_SACKPERM;
 1433 #ifdef TCP_SIGNATURE
 1434                 if (sc->sc_flags & SCF_SIGNATURE)
 1435                         to.to_flags |= TOF_SIGNATURE;
 1436 #endif
 1437                 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
 1438 
 1439                 /* Adjust headers by option size. */
 1440                 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
 1441                 m->m_len += optlen;
 1442                 m->m_pkthdr.len += optlen;
 1443 
 1444 #ifdef TCP_SIGNATURE
 1445                 if (sc->sc_flags & SCF_SIGNATURE)
 1446                         tcp_signature_compute(m, 0, 0, optlen,
 1447                             to.to_signature, IPSEC_DIR_OUTBOUND);
 1448 #endif
 1449 #ifdef INET6
 1450                 if (sc->sc_inc.inc_flags & INC_ISIPV6)
 1451                         ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
 1452                 else
 1453 #endif
 1454                         ip->ip_len += optlen;
 1455         } else
 1456                 optlen = 0;
 1457 
 1458         M_SETFIB(m, sc->sc_inc.inc_fibnum);
 1459 #ifdef INET6
 1460         if (sc->sc_inc.inc_flags & INC_ISIPV6) {
 1461                 th->th_sum = 0;
 1462                 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
 1463                                        tlen + optlen - hlen);
 1464                 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
 1465                 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
 1466         } else
 1467 #endif
 1468         {
 1469                 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
 1470                     htons(tlen + optlen - hlen + IPPROTO_TCP));
 1471                 m->m_pkthdr.csum_flags = CSUM_TCP;
 1472                 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
 1473                 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
 1474         }
 1475         return (error);
 1476 }
 1477 
 1478 void
 1479 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
 1480     struct inpcb *inp, struct socket **lsop, struct mbuf *m)
 1481 {
 1482 
 1483         _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
 1484 }
 1485 
 1486 void
 1487 tcp_offload_syncache_add(struct in_conninfo *inc, struct tcpopt *to,
 1488     struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
 1489     struct toe_usrreqs *tu, void *toepcb)
 1490 {
 1491 
 1492 
 1493         INP_INFO_WLOCK(&tcbinfo);
 1494         INP_WLOCK(inp);
 1495         _syncache_add(inc, to, th, inp, lsop, NULL, tu, toepcb);
 1496         
 1497 }
 1498 
 1499 /*
 1500  * The purpose of SYN cookies is to avoid keeping track of all SYN's we
 1501  * receive and to be able to handle SYN floods from bogus source addresses
 1502  * (where we will never receive any reply).  SYN floods try to exhaust all
 1503  * our memory and available slots in the SYN cache table to cause a denial
 1504  * of service to legitimate users of the local host.
 1505  *
 1506  * The idea of SYN cookies is to encode and include all necessary information
 1507  * about the connection setup state within the SYN-ACK we send back and thus
 1508  * to get along without keeping any local state until the ACK to the SYN-ACK
 1509  * arrives (if ever).  Everything we need to know should be available from
 1510  * the information we encoded in the SYN-ACK.
 1511  *
 1512  * More information about the theory behind SYN cookies and its first
 1513  * discussion and specification can be found at:
 1514  *  http://cr.yp.to/syncookies.html    (overview)
 1515  *  http://cr.yp.to/syncookies/archive (gory details)
 1516  *
 1517  * This implementation extends the orginal idea and first implementation
 1518  * of FreeBSD by using not only the initial sequence number field to store
 1519  * information but also the timestamp field if present.  This way we can
 1520  * keep track of the entire state we need to know to recreate the session in
 1521  * its original form.  Almost all TCP speakers implement RFC1323 timestamps
 1522  * these days.  For those that do not we still have to live with the known
 1523  * shortcomings of the ISN only SYN cookies.
 1524  *
 1525  * Cookie layers:
 1526  *
 1527  * Initial sequence number we send:
 1528  * 31|................................|0
 1529  *    DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
 1530  *    D = MD5 Digest (first dword)
 1531  *    M = MSS index
 1532  *    R = Rotation of secret
 1533  *    P = Odd or Even secret
 1534  *
 1535  * The MD5 Digest is computed with over following parameters:
 1536  *  a) randomly rotated secret
 1537  *  b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
 1538  *  c) the received initial sequence number from remote host
 1539  *  d) the rotation offset and odd/even bit
 1540  *
 1541  * Timestamp we send:
 1542  * 31|................................|0
 1543  *    DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
 1544  *    D = MD5 Digest (third dword) (only as filler)
 1545  *    S = Requested send window scale
 1546  *    R = Requested receive window scale
 1547  *    A = SACK allowed
 1548  *    5 = TCP-MD5 enabled (not implemented yet)
 1549  *    XORed with MD5 Digest (forth dword)
 1550  *
 1551  * The timestamp isn't cryptographically secure and doesn't need to be.
 1552  * The double use of the MD5 digest dwords ties it to a specific remote/
 1553  * local host/port, remote initial sequence number and our local time
 1554  * limited secret.  A received timestamp is reverted (XORed) and then
 1555  * the contained MD5 dword is compared to the computed one to ensure the
 1556  * timestamp belongs to the SYN-ACK we sent.  The other parameters may
 1557  * have been tampered with but this isn't different from supplying bogus
 1558  * values in the SYN in the first place.
 1559  *
 1560  * Some problems with SYN cookies remain however:
 1561  * Consider the problem of a recreated (and retransmitted) cookie.  If the
 1562  * original SYN was accepted, the connection is established.  The second
 1563  * SYN is inflight, and if it arrives with an ISN that falls within the
 1564  * receive window, the connection is killed.
 1565  *
 1566  * Notes:
 1567  * A heuristic to determine when to accept syn cookies is not necessary.
 1568  * An ACK flood would cause the syncookie verification to be attempted,
 1569  * but a SYN flood causes syncookies to be generated.  Both are of equal
 1570  * cost, so there's no point in trying to optimize the ACK flood case.
 1571  * Also, if you don't process certain ACKs for some reason, then all someone
 1572  * would have to do is launch a SYN and ACK flood at the same time, which
 1573  * would stop cookie verification and defeat the entire purpose of syncookies.
 1574  */
 1575 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
 1576 
 1577 static void
 1578 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
 1579     u_int32_t *flowlabel)
 1580 {
 1581         MD5_CTX ctx;
 1582         u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
 1583         u_int32_t data;
 1584         u_int32_t *secbits;
 1585         u_int off, pmss, mss;
 1586         int i;
 1587 
 1588         SCH_LOCK_ASSERT(sch);
 1589 
 1590         /* Which of the two secrets to use. */
 1591         secbits = sch->sch_oddeven ?
 1592                         sch->sch_secbits_odd : sch->sch_secbits_even;
 1593 
 1594         /* Reseed secret if too old. */
 1595         if (sch->sch_reseed < time_uptime) {
 1596                 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1;    /* toggle */
 1597                 secbits = sch->sch_oddeven ?
 1598                                 sch->sch_secbits_odd : sch->sch_secbits_even;
 1599                 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
 1600                         secbits[i] = arc4random();
 1601                 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
 1602         }
 1603 
 1604         /* Secret rotation offset. */
 1605         off = sc->sc_iss & 0x7;                 /* iss was randomized before */
 1606 
 1607         /* Maximum segment size calculation. */
 1608         pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss);
 1609         for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
 1610                 if (tcp_sc_msstab[mss] <= pmss)
 1611                         break;
 1612 
 1613         /* Fold parameters and MD5 digest into the ISN we will send. */
 1614         data = sch->sch_oddeven;/* odd or even secret, 1 bit */
 1615         data |= off << 1;       /* secret offset, derived from iss, 3 bits */
 1616         data |= mss << 4;       /* mss, 3 bits */
 1617 
 1618         MD5Init(&ctx);
 1619         MD5Update(&ctx, ((u_int8_t *)secbits) + off,
 1620             SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
 1621         MD5Update(&ctx, secbits, off);
 1622         MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
 1623         MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
 1624         MD5Update(&ctx, &data, sizeof(data));
 1625         MD5Final((u_int8_t *)&md5_buffer, &ctx);
 1626 
 1627         data |= (md5_buffer[0] << 7);
 1628         sc->sc_iss = data;
 1629 
 1630 #ifdef INET6
 1631         *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
 1632 #endif
 1633 
 1634         /* Additional parameters are stored in the timestamp if present. */
 1635         if (sc->sc_flags & SCF_TIMESTAMP) {
 1636                 data =  ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
 1637                 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
 1638                 data |= sc->sc_requested_s_scale << 2;  /* SWIN scale, 4 bits */
 1639                 data |= sc->sc_requested_r_scale << 6;  /* RWIN scale, 4 bits */
 1640                 data |= md5_buffer[2] << 10;            /* more digest bits */
 1641                 data ^= md5_buffer[3];
 1642                 sc->sc_ts = data;
 1643                 sc->sc_tsoff = data - ticks;            /* after XOR */
 1644         }
 1645 
 1646         tcpstat.tcps_sc_sendcookie++;
 1647 }
 1648 
 1649 static struct syncache *
 1650 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, 
 1651     struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
 1652     struct socket *so)
 1653 {
 1654         MD5_CTX ctx;
 1655         u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
 1656         u_int32_t data = 0;
 1657         u_int32_t *secbits;
 1658         tcp_seq ack, seq;
 1659         int off, mss, wnd, flags;
 1660 
 1661         SCH_LOCK_ASSERT(sch);
 1662 
 1663         /*
 1664          * Pull information out of SYN-ACK/ACK and
 1665          * revert sequence number advances.
 1666          */
 1667         ack = th->th_ack - 1;
 1668         seq = th->th_seq - 1;
 1669         off = (ack >> 1) & 0x7;
 1670         mss = (ack >> 4) & 0x7;
 1671         flags = ack & 0x7f;
 1672 
 1673         /* Which of the two secrets to use. */
 1674         secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
 1675 
 1676         /*
 1677          * The secret wasn't updated for the lifetime of a syncookie,
 1678          * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
 1679          */
 1680         if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
 1681                 return (NULL);
 1682         }
 1683 
 1684         /* Recompute the digest so we can compare it. */
 1685         MD5Init(&ctx);
 1686         MD5Update(&ctx, ((u_int8_t *)secbits) + off,
 1687             SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
 1688         MD5Update(&ctx, secbits, off);
 1689         MD5Update(&ctx, inc, sizeof(*inc));
 1690         MD5Update(&ctx, &seq, sizeof(seq));
 1691         MD5Update(&ctx, &flags, sizeof(flags));
 1692         MD5Final((u_int8_t *)&md5_buffer, &ctx);
 1693 
 1694         /* Does the digest part of or ACK'ed ISS match? */
 1695         if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
 1696                 return (NULL);
 1697 
 1698         /* Does the digest part of our reflected timestamp match? */
 1699         if (to->to_flags & TOF_TS) {
 1700                 data = md5_buffer[3] ^ to->to_tsecr;
 1701                 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
 1702                         return (NULL);
 1703         }
 1704 
 1705         /* Fill in the syncache values. */
 1706         bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
 1707         sc->sc_ipopts = NULL;
 1708         
 1709         sc->sc_irs = seq;
 1710         sc->sc_iss = ack;
 1711 
 1712 #ifdef INET6
 1713         if (inc->inc_flags & INC_ISIPV6) {
 1714                 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
 1715                         sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
 1716         } else
 1717 #endif
 1718         {
 1719                 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
 1720                 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
 1721         }
 1722 
 1723         /* Additional parameters that were encoded in the timestamp. */
 1724         if (data) {
 1725                 sc->sc_flags |= SCF_TIMESTAMP;
 1726                 sc->sc_tsreflect = to->to_tsval;
 1727                 sc->sc_ts = to->to_tsecr;
 1728                 sc->sc_tsoff = to->to_tsecr - ticks;
 1729                 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
 1730                 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
 1731                 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
 1732                     TCP_MAX_WINSHIFT);
 1733                 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
 1734                     TCP_MAX_WINSHIFT);
 1735                 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
 1736                         sc->sc_flags |= SCF_WINSCALE;
 1737         } else
 1738                 sc->sc_flags |= SCF_NOOPT;
 1739 
 1740         wnd = sbspace(&so->so_rcv);
 1741         wnd = imax(wnd, 0);
 1742         wnd = imin(wnd, TCP_MAXWIN);
 1743         sc->sc_wnd = wnd;
 1744 
 1745         sc->sc_rxmits = 0;
 1746         sc->sc_peer_mss = tcp_sc_msstab[mss];
 1747 
 1748         tcpstat.tcps_sc_recvcookie++;
 1749         return (sc);
 1750 }
 1751 
 1752 /*
 1753  * Returns the current number of syncache entries.  This number
 1754  * will probably change before you get around to calling 
 1755  * syncache_pcblist.
 1756  */
 1757 
 1758 int
 1759 syncache_pcbcount(void)
 1760 {
 1761         struct syncache_head *sch;
 1762         int count, i;
 1763 
 1764         for (count = 0, i = 0; i < tcp_syncache.hashsize; i++) {
 1765                 /* No need to lock for a read. */
 1766                 sch = &tcp_syncache.hashbase[i];
 1767                 count += sch->sch_length;
 1768         }
 1769         return count;
 1770 }
 1771 
 1772 /*
 1773  * Exports the syncache entries to userland so that netstat can display
 1774  * them alongside the other sockets.  This function is intended to be
 1775  * called only from tcp_pcblist.
 1776  *
 1777  * Due to concurrency on an active system, the number of pcbs exported
 1778  * may have no relation to max_pcbs.  max_pcbs merely indicates the
 1779  * amount of space the caller allocated for this function to use.
 1780  */
 1781 int
 1782 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
 1783 {
 1784         struct xtcpcb xt;
 1785         struct syncache *sc;
 1786         struct syncache_head *sch;
 1787         int count, error, i;
 1788 
 1789         for (count = 0, error = 0, i = 0; i < tcp_syncache.hashsize; i++) {
 1790                 sch = &tcp_syncache.hashbase[i];
 1791                 SCH_LOCK(sch);
 1792                 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
 1793                         if (count >= max_pcbs) {
 1794                                 SCH_UNLOCK(sch);
 1795                                 goto exit;
 1796                         }
 1797                         if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
 1798                                 continue;
 1799                         bzero(&xt, sizeof(xt));
 1800                         xt.xt_len = sizeof(xt);
 1801                         if (sc->sc_inc.inc_flags & INC_ISIPV6)
 1802                                 xt.xt_inp.inp_vflag = INP_IPV6;
 1803                         else
 1804                                 xt.xt_inp.inp_vflag = INP_IPV4;
 1805                         bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
 1806                         xt.xt_tp.t_inpcb = &xt.xt_inp;
 1807                         xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
 1808                         xt.xt_socket.xso_protocol = IPPROTO_TCP;
 1809                         xt.xt_socket.xso_len = sizeof (struct xsocket);
 1810                         xt.xt_socket.so_type = SOCK_STREAM;
 1811                         xt.xt_socket.so_state = SS_ISCONNECTING;
 1812                         error = SYSCTL_OUT(req, &xt, sizeof xt);
 1813                         if (error) {
 1814                                 SCH_UNLOCK(sch);
 1815                                 goto exit;
 1816                         }
 1817                         count++;
 1818                 }
 1819                 SCH_UNLOCK(sch);
 1820         }
 1821 exit:
 1822         *pcbs_exported = count;
 1823         return error;
 1824 }
 1825 

Cache object: 3b0a849ef4bc0d76671283571ebe35fb


[ 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.