sys/netinet/ip_flow.c

   1 /*-
   2  * Copyright (c) 1998 The NetBSD Foundation, Inc.
   3  * All rights reserved.
   4  *
   5  * This code is derived from software contributed to The NetBSD Foundation
   6  * by the 3am Software Foundry ("3am").  It was developed by Matt Thomas.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the NetBSD
  19  *      Foundation, Inc. and its contributors.
  20  * 4. Neither the name of The NetBSD Foundation nor the names of its
  21  *    contributors may be used to endorse or promote products derived
  22  *    from this software without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
  25  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
  26  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  27  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
  28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  31  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  32  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  33  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  34  * POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * $FreeBSD: src/sys/netinet/ip_flow.c,v 1.9.2.2 2001/11/04 17:35:31 luigi Exp $
  37  * $DragonFly: src/sys/netinet/ip_flow.c,v 1.14 2008/05/14 11:59:24 sephe Exp $
  38  */
  39
  40 #include <sys/param.h>
  41 #include <sys/kernel.h>
  42 #include <sys/malloc.h>
  43 #include <sys/mbuf.h>
  44 #include <sys/protosw.h>
  45 #include <sys/socket.h>
  46 #include <sys/sysctl.h>
  47 #include <sys/thread2.h>
  48
  49 #include <machine/smp.h>
  50
  51 #include <net/if.h>
  52 #include <net/route.h>
  53 #include <net/netisr.h>
  54 #include <net/netmsg2.h>
  55
  56 #include <netinet/in.h>
  57 #include <netinet/ip.h>
  58 #include <netinet/in_var.h>
  59 #include <netinet/ip_var.h>
  60 #include <netinet/ip_flow.h>
  61
  62 #define IPFLOW_TIMER            (5 * PR_SLOWHZ)
  63 #define IPFLOW_HASHBITS         6       /* should not be a multiple of 8 */
  64 #define IPFLOW_HASHSIZE         (1 << IPFLOW_HASHBITS)
  65 #define IPFLOW_MAX              256
  66
  67 #define IPFLOW_RTENTRY_ISDOWN(rt) \
  68         (((rt)->rt_flags & RTF_UP) == 0 || ((rt)->rt_ifp->if_flags & IFF_UP) == 0)
  69
  70 #define ipflow_inuse            ipflow_inuse_pcpu[mycpuid]
  71 #define ipflows                 ipflows_pcpu[mycpuid]
  72
  73 static LIST_HEAD(ipflowhead, ipflow) ipflows_pcpu[MAXCPU][IPFLOW_HASHSIZE];
  74 static int              ipflow_inuse_pcpu[MAXCPU];
  75 static struct netmsg    ipflow_timo_netmsgs[MAXCPU];
  76 static int              ipflow_active = 0;
  77
  78 SYSCTL_NODE(_net_inet_ip, OID_AUTO, ipflow, CTLFLAG_RW, 0, "ip flow");
  79 SYSCTL_INT(_net_inet_ip, IPCTL_FASTFORWARDING, fastforwarding, CTLFLAG_RW,
  80            &ipflow_active, 0, "Enable flow-based IP forwarding");
  81
  82 static MALLOC_DEFINE(M_IPFLOW, "ip_flow", "IP flow");
  83
  84 static unsigned
  85 ipflow_hash(struct in_addr dst, struct in_addr src, unsigned tos)
  86 {
  87         unsigned hash = tos;
  88         int idx;
  89
  90         for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS)
  91                 hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx);
  92         return hash & (IPFLOW_HASHSIZE-1);
  93 }
  94
  95 static struct ipflow *
  96 ipflow_lookup(const struct ip *ip)
  97 {
  98         unsigned hash;
  99         struct ipflow *ipf;
 100
 101         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
 102
 103         crit_enter();
 104         ipf = LIST_FIRST(&ipflows[hash]);
 105         while (ipf != NULL) {
 106                 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr &&
 107                     ip->ip_src.s_addr == ipf->ipf_src.s_addr &&
 108                     ip->ip_tos == ipf->ipf_tos)
 109                         break;
 110                 ipf = LIST_NEXT(ipf, ipf_next);
 111         }
 112         crit_exit();
 113
 114         return ipf;
 115 }
 116
 117 int
 118 ipflow_fastforward(struct mbuf *m, lwkt_serialize_t serializer)
 119 {
 120         struct ip *ip;
 121         struct ipflow *ipf;
 122         struct rtentry *rt;
 123         struct sockaddr *dst;
 124         struct ifnet *ifp;
 125         int error;
 126
 127         /*
 128          * Are we forwarding packets?  Big enough for an IP packet?
 129          */
 130         if (!ipforwarding || !ipflow_active || m->m_len < sizeof(struct ip))
 131                 return 0;
 132
 133         /*
 134          * IP header with no option and valid version and length
 135          */
 136         ip = mtod(m, struct ip *);
 137         if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
 138             ntohs(ip->ip_len) > m->m_pkthdr.len)
 139                 return 0;
 140
 141         /*
 142          * Find a flow.
 143          */
 144         ipf = ipflow_lookup(ip);
 145         if (ipf == NULL)
 146                 return 0;
 147
 148         /*
 149          * Route and interface still up?
 150          */
 151         rt = ipf->ipf_ro.ro_rt;
 152         if (IPFLOW_RTENTRY_ISDOWN(rt))
 153                 return 0;
 154         ifp = rt->rt_ifp;
 155
 156         /*
 157          * Packet size OK?  TTL?
 158          */
 159         if (m->m_pkthdr.len > ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
 160                 return 0;
 161
 162         /*
 163          * Everything checks out and so we can forward this packet.
 164          * Modify the TTL and incrementally change the checksum.
 165          */
 166         ip->ip_ttl -= IPTTLDEC;
 167         if (ip->ip_sum >= htons(0xffff - (IPTTLDEC << 8)))
 168                 ip->ip_sum += htons(IPTTLDEC << 8) + 1;
 169         else
 170                 ip->ip_sum += htons(IPTTLDEC << 8);
 171
 172         /*
 173          * Send the packet on its way.  All we can get back is ENOBUFS
 174          */
 175         ipf->ipf_uses++;
 176         ipf->ipf_timer = IPFLOW_TIMER;
 177
 178         if (rt->rt_flags & RTF_GATEWAY)
 179                 dst = rt->rt_gateway;
 180         else
 181                 dst = &ipf->ipf_ro.ro_dst;
 182
 183         if (serializer)
 184                 lwkt_serialize_exit(serializer);
 185
 186         error = ifp->if_output(ifp, m, dst, rt);
 187         if (error) {
 188                 if (error == ENOBUFS)
 189                         ipf->ipf_dropped++;
 190                 else
 191                         ipf->ipf_errors++;
 192         }
 193
 194         if (serializer)
 195                 lwkt_serialize_enter(serializer);
 196         return 1;
 197 }
 198
 199 static void
 200 ipflow_addstats(struct ipflow *ipf)
 201 {
 202         ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 203         ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
 204         ipstat.ips_forward += ipf->ipf_uses;
 205         ipstat.ips_fastforward += ipf->ipf_uses;
 206 }
 207
 208 static void
 209 ipflow_free(struct ipflow *ipf)
 210 {
 211         /*
 212          * Remove the flow from the hash table (at elevated IPL).
 213          * Once it's off the list, we can deal with it at normal
 214          * network IPL.
 215          */
 216         crit_enter();
 217         LIST_REMOVE(ipf, ipf_next);
 218
 219         KKASSERT(ipflow_inuse > 0);
 220         ipflow_inuse--;
 221         crit_exit();
 222
 223         ipflow_addstats(ipf);
 224         RTFREE(ipf->ipf_ro.ro_rt);
 225         kfree(ipf, M_IPFLOW);
 226 }
 227
 228 static struct ipflow *
 229 ipflow_reap(void)
 230 {
 231         struct ipflow *ipf, *maybe_ipf = NULL;
 232         int idx;
 233
 234         crit_enter();
 235         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 236                 ipf = LIST_FIRST(&ipflows[idx]);
 237                 while (ipf != NULL) {
 238                         /*
 239                          * If this no longer points to a valid route
 240                          * reclaim it.
 241                          */
 242                         if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
 243                                 goto done;
 244
 245                         /*
 246                          * choose the one that's been least recently used
 247                          * or has had the least uses in the last 1.5
 248                          * intervals.
 249                          */
 250                         if (maybe_ipf == NULL ||
 251                             ipf->ipf_timer < maybe_ipf->ipf_timer ||
 252                             (ipf->ipf_timer == maybe_ipf->ipf_timer &&
 253                              ipf->ipf_last_uses + ipf->ipf_uses <
 254                              maybe_ipf->ipf_last_uses + maybe_ipf->ipf_uses))
 255                                 maybe_ipf = ipf;
 256                         ipf = LIST_NEXT(ipf, ipf_next);
 257                 }
 258         }
 259         ipf = maybe_ipf;
 260 done:
 261         /*
 262          * Remove the entry from the flow table.
 263          */
 264         LIST_REMOVE(ipf, ipf_next);
 265         crit_exit();
 266
 267         ipflow_addstats(ipf);
 268         RTFREE(ipf->ipf_ro.ro_rt);
 269         return ipf;
 270 }
 271
 272 static void
 273 ipflow_timo_dispatch(struct netmsg *nmsg)
 274 {
 275         struct ipflow *ipf;
 276         int idx;
 277
 278         crit_enter();
 279         lwkt_replymsg(&nmsg->nm_lmsg, 0);       /* reply ASAP */
 280
 281         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 282                 ipf = LIST_FIRST(&ipflows[idx]);
 283                 while (ipf != NULL) {
 284                         struct ipflow *next_ipf = LIST_NEXT(ipf, ipf_next);
 285
 286                         if (--ipf->ipf_timer == 0) {
 287                                 ipflow_free(ipf);
 288                         } else {
 289                                 ipf->ipf_last_uses = ipf->ipf_uses;
 290                                 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 291                                 ipstat.ips_forward += ipf->ipf_uses;
 292                                 ipstat.ips_fastforward += ipf->ipf_uses;
 293                                 ipf->ipf_uses = 0;
 294                         }
 295                         ipf = next_ipf;
 296                 }
 297         }
 298         crit_exit();
 299 }
 300
 301 static void
 302 ipflow_timo_ipi(void *arg __unused)
 303 {
 304         struct lwkt_msg *msg = &ipflow_timo_netmsgs[mycpuid].nm_lmsg;
 305
 306         crit_enter();
 307         if (msg->ms_flags & MSGF_DONE)
 308                 lwkt_sendmsg(cpu_portfn(mycpuid), msg);
 309         crit_exit();
 310 }
 311
 312 void
 313 ipflow_slowtimo(void)
 314 {
 315 #ifdef SMP
 316         lwkt_send_ipiq_mask(smp_active_mask, ipflow_timo_ipi, NULL);
 317 #else
 318         ipflow_timo_ipi(NULL);
 319 #endif
 320 }
 321
 322 static void
 323 ipflow_create_oncpu(const struct route *ro, struct mbuf *m)
 324 {
 325         const struct ip *const ip = mtod(m, struct ip *);
 326         struct ipflow *ipf;
 327         unsigned hash;
 328
 329         /*
 330          * See if an existing flow struct exists.  If so remove it from it's
 331          * list and free the old route.  If not, try to malloc a new one
 332          * (if we aren't at our limit).
 333          */
 334         ipf = ipflow_lookup(ip);
 335         if (ipf == NULL) {
 336                 if (ipflow_inuse == IPFLOW_MAX) {
 337                         ipf = ipflow_reap();
 338                 } else {
 339                         ipf = kmalloc(sizeof(*ipf), M_IPFLOW,
 340                                       M_INTWAIT | M_NULLOK);
 341                         if (ipf == NULL)
 342                                 return;
 343                         ipflow_inuse++;
 344                 }
 345                 bzero(ipf, sizeof(*ipf));
 346         } else if (IPFLOW_RTENTRY_ISDOWN(ipf->ipf_ro.ro_rt)) {
 347                 crit_enter();
 348                 LIST_REMOVE(ipf, ipf_next);
 349                 crit_exit();
 350
 351                 ipflow_addstats(ipf);
 352                 RTFREE(ipf->ipf_ro.ro_rt);
 353                 ipf->ipf_uses = ipf->ipf_last_uses = 0;
 354                 ipf->ipf_errors = ipf->ipf_dropped = 0;
 355         } else {
 356                 /*
 357                  * The route entry cached in ipf is still up,
 358                  * this could happen while the ipf installation
 359                  * is in transition state.
 360                  * XXX should not happen on UP box
 361                  */
 362                 return;
 363         }
 364
 365         /*
 366          * Fill in the updated information.
 367          */
 368         ipf->ipf_ro = *ro;
 369         ro->ro_rt->rt_refcnt++;
 370         ipf->ipf_dst = ip->ip_dst;
 371         ipf->ipf_src = ip->ip_src;
 372         ipf->ipf_tos = ip->ip_tos;
 373         ipf->ipf_timer = IPFLOW_TIMER;
 374
 375         /*
 376          * Insert into the approriate bucket of the flow table.
 377          */
 378         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
 379         crit_enter();
 380         LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next);
 381         crit_exit();
 382 }
 383
 384 #ifdef SMP
 385
 386 static void
 387 ipflow_create_dispatch(struct netmsg *nmsg)
 388 {
 389         struct netmsg_packet *nmp = (struct netmsg_packet *)nmsg;
 390         struct sockaddr_in *sin;
 391         struct route ro;
 392         int nextcpu;
 393
 394         bzero(&ro, sizeof(ro));
 395         sin = (struct sockaddr_in *)&ro.ro_dst;
 396         sin->sin_family = AF_INET;
 397         sin->sin_len = sizeof(struct sockaddr_in);
 398         sin->sin_addr.s_addr = (in_addr_t)nmsg->nm_lmsg.u.ms_result32;
 399
 400         rtalloc_ign(&ro, RTF_PRCLONING);
 401         if (ro.ro_rt != NULL) {
 402                 ipflow_create_oncpu(&ro, nmp->nm_packet);
 403                 RTFREE(ro.ro_rt);
 404         }
 405
 406         nextcpu = mycpuid + 1;
 407         if (nextcpu < ncpus)
 408                 lwkt_forwardmsg(cpu_portfn(nextcpu), &nmsg->nm_lmsg);
 409         else
 410                 m_freem(nmp->nm_packet);
 411 }
 412
 413 #endif  /* SMP */
 414
 415 void
 416 ipflow_create(const struct route *ro, struct mbuf *m)
 417 {
 418         const struct ip *const ip = mtod(m, struct ip *);
 419 #ifdef SMP
 420         struct netmsg_packet *nmp;
 421         struct netmsg *nmsg;
 422         int nextcpu;
 423 #endif
 424
 425         /*
 426          * Don't create cache entries for ICMP messages.
 427          */
 428         if (!ipflow_active || ip->ip_p == IPPROTO_ICMP) {
 429                 m_freem(m);
 430                 return;
 431         }
 432
 433 #ifdef SMP
 434         nmp = &m->m_hdr.mh_netmsg;
 435         nmsg = &nmp->nm_netmsg;
 436
 437         netmsg_init(nmsg, &netisr_apanic_rport, 0, ipflow_create_dispatch);
 438         nmp->nm_packet = m;
 439         nmsg->nm_lmsg.u.ms_result32 =
 440                 ((const struct sockaddr_in *)&ro->ro_dst)->sin_addr.s_addr;
 441
 442         if (mycpuid == 0) {
 443                 ipflow_create_oncpu(ro, m);
 444                 nextcpu = 1;
 445         } else {
 446                 nextcpu = 0;
 447         }
 448         if (nextcpu < ncpus)
 449                 lwkt_sendmsg(cpu_portfn(nextcpu), &nmsg->nm_lmsg);
 450         else
 451                 m_freem(m);
 452 #else
 453         ipflow_create_oncpu(ro, m);
 454         m_freem(m);
 455 #endif
 456 }
 457
 458 static void
 459 ipflow_init(void)
 460 {
 461         char oid_name[32];
 462         int i;
 463
 464         for (i = 0; i < ncpus; ++i) {
 465                 netmsg_init(&ipflow_timo_netmsgs[i], &netisr_adone_rport, 0,
 466                             ipflow_timo_dispatch);
 467
 468                 ksnprintf(oid_name, sizeof(oid_name), "inuse%d", i);
 469
 470                 SYSCTL_ADD_INT(NULL,
 471                 SYSCTL_STATIC_CHILDREN(_net_inet_ip_ipflow),
 472                 OID_AUTO, oid_name, CTLFLAG_RD, &ipflow_inuse_pcpu[i], 0,
 473                 "# of ip flow being used");
 474         }
 475 }
 476 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, ipflow_init, 0);