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 1.12 2008/04/03 13:43:29 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         lwkt_serialize_enter(ifp->if_serializer);
 187         error = ifp->if_output(ifp, m, dst, rt);
 188         if (error) {
 189                 if (error == ENOBUFS)
 190                         ipf->ipf_dropped++;
 191                 else
 192                         ipf->ipf_errors++;
 193         }
 194         lwkt_serialize_exit(ifp->if_serializer);
 195
 196         if (serializer)
 197                 lwkt_serialize_enter(serializer);
 198         return 1;
 199 }
 200
 201 static void
 202 ipflow_addstats(struct ipflow *ipf)
 203 {
 204         ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 205         ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
 206         ipstat.ips_forward += ipf->ipf_uses;
 207         ipstat.ips_fastforward += ipf->ipf_uses;
 208 }
 209
 210 static void
 211 ipflow_free(struct ipflow *ipf)
 212 {
 213         /*
 214          * Remove the flow from the hash table (at elevated IPL).
 215          * Once it's off the list, we can deal with it at normal
 216          * network IPL.
 217          */
 218         crit_enter();
 219         LIST_REMOVE(ipf, ipf_next);
 220
 221         KKASSERT(ipflow_inuse > 0);
 222         ipflow_inuse--;
 223         crit_exit();
 224
 225         ipflow_addstats(ipf);
 226         RTFREE(ipf->ipf_ro.ro_rt);
 227         kfree(ipf, M_IPFLOW);
 228 }
 229
 230 static struct ipflow *
 231 ipflow_reap(void)
 232 {
 233         struct ipflow *ipf, *maybe_ipf = NULL;
 234         int idx;
 235
 236         crit_enter();
 237         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 238                 ipf = LIST_FIRST(&ipflows[idx]);
 239                 while (ipf != NULL) {
 240                         /*
 241                          * If this no longer points to a valid route
 242                          * reclaim it.
 243                          */
 244                         if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
 245                                 goto done;
 246
 247                         /*
 248                          * choose the one that's been least recently used
 249                          * or has had the least uses in the last 1.5
 250                          * intervals.
 251                          */
 252                         if (maybe_ipf == NULL ||
 253                             ipf->ipf_timer < maybe_ipf->ipf_timer ||
 254                             (ipf->ipf_timer == maybe_ipf->ipf_timer &&
 255                              ipf->ipf_last_uses + ipf->ipf_uses <
 256                              maybe_ipf->ipf_last_uses + maybe_ipf->ipf_uses))
 257                                 maybe_ipf = ipf;
 258                         ipf = LIST_NEXT(ipf, ipf_next);
 259                 }
 260         }
 261         ipf = maybe_ipf;
 262 done:
 263         /*
 264          * Remove the entry from the flow table.
 265          */
 266         LIST_REMOVE(ipf, ipf_next);
 267         crit_exit();
 268
 269         ipflow_addstats(ipf);
 270         RTFREE(ipf->ipf_ro.ro_rt);
 271         return ipf;
 272 }
 273
 274 static void
 275 ipflow_timo_dispatch(struct netmsg *nmsg)
 276 {
 277         struct ipflow *ipf;
 278         int idx;
 279
 280         crit_enter();
 281         lwkt_replymsg(&nmsg->nm_lmsg, 0);       /* reply ASAP */
 282
 283         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 284                 ipf = LIST_FIRST(&ipflows[idx]);
 285                 while (ipf != NULL) {
 286                         struct ipflow *next_ipf = LIST_NEXT(ipf, ipf_next);
 287
 288                         if (--ipf->ipf_timer == 0) {
 289                                 ipflow_free(ipf);
 290                         } else {
 291                                 ipf->ipf_last_uses = ipf->ipf_uses;
 292                                 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 293                                 ipstat.ips_forward += ipf->ipf_uses;
 294                                 ipstat.ips_fastforward += ipf->ipf_uses;
 295                                 ipf->ipf_uses = 0;
 296                         }
 297                         ipf = next_ipf;
 298                 }
 299         }
 300         crit_exit();
 301 }
 302
 303 static void
 304 ipflow_timo_ipi(void *arg __unused)
 305 {
 306         struct lwkt_msg *msg = &ipflow_timo_netmsgs[mycpuid].nm_lmsg;
 307
 308         crit_enter();
 309         if (msg->ms_flags & MSGF_DONE)
 310                 lwkt_sendmsg(cpu_portfn(mycpuid), msg);
 311         crit_exit();
 312 }
 313
 314 void
 315 ipflow_slowtimo(void)
 316 {
 317         lwkt_send_ipiq_mask(smp_active_mask, ipflow_timo_ipi, NULL);
 318 }
 319
 320 static void
 321 ipflow_create_oncpu(const struct route *ro, struct mbuf *m)
 322 {
 323         const struct ip *const ip = mtod(m, struct ip *);
 324         struct ipflow *ipf;
 325         unsigned hash;
 326
 327         /*
 328          * See if an existing flow struct exists.  If so remove it from it's
 329          * list and free the old route.  If not, try to malloc a new one
 330          * (if we aren't at our limit).
 331          */
 332         ipf = ipflow_lookup(ip);
 333         if (ipf == NULL) {
 334                 if (ipflow_inuse == IPFLOW_MAX) {
 335                         ipf = ipflow_reap();
 336                 } else {
 337                         ipf = kmalloc(sizeof(*ipf), M_IPFLOW,
 338                                       M_INTWAIT | M_NULLOK);
 339                         if (ipf == NULL)
 340                                 return;
 341                         ipflow_inuse++;
 342                 }
 343                 bzero(ipf, sizeof(*ipf));
 344         } else if (IPFLOW_RTENTRY_ISDOWN(ipf->ipf_ro.ro_rt)) {
 345                 crit_enter();
 346                 LIST_REMOVE(ipf, ipf_next);
 347                 crit_exit();
 348
 349                 ipflow_addstats(ipf);
 350                 RTFREE(ipf->ipf_ro.ro_rt);
 351                 ipf->ipf_uses = ipf->ipf_last_uses = 0;
 352                 ipf->ipf_errors = ipf->ipf_dropped = 0;
 353         } else {
 354                 /*
 355                  * The route entry cached in ipf is still up,
 356                  * this could happen while the ipf installation
 357                  * is in transition state.
 358                  * XXX should not happen on UP box
 359                  */
 360                 return;
 361         }
 362
 363         /*
 364          * Fill in the updated information.
 365          */
 366         ipf->ipf_ro = *ro;
 367         ro->ro_rt->rt_refcnt++;
 368         ipf->ipf_dst = ip->ip_dst;
 369         ipf->ipf_src = ip->ip_src;
 370         ipf->ipf_tos = ip->ip_tos;
 371         ipf->ipf_timer = IPFLOW_TIMER;
 372
 373         /*
 374          * Insert into the approriate bucket of the flow table.
 375          */
 376         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
 377         crit_enter();
 378         LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next);
 379         crit_exit();
 380 }
 381
 382 #ifdef SMP
 383
 384 static void
 385 ipflow_create_dispatch(struct netmsg *nmsg)
 386 {
 387         struct netmsg_packet *nmp = (struct netmsg_packet *)nmsg;
 388         struct sockaddr_in *sin;
 389         struct route ro;
 390         int nextcpu;
 391
 392         bzero(&ro, sizeof(ro));
 393         sin = (struct sockaddr_in *)&ro.ro_dst;
 394         sin->sin_family = AF_INET;
 395         sin->sin_len = sizeof(struct sockaddr_in);
 396         sin->sin_addr.s_addr = (in_addr_t)nmsg->nm_lmsg.u.ms_result32;
 397
 398         rtalloc_ign(&ro, RTF_PRCLONING);
 399         if (ro.ro_rt != NULL) {
 400                 ipflow_create_oncpu(&ro, nmp->nm_packet);
 401                 RTFREE(ro.ro_rt);
 402         }
 403
 404         nextcpu = mycpuid + 1;
 405         if (nextcpu < ncpus)
 406                 lwkt_forwardmsg(cpu_portfn(nextcpu), &nmsg->nm_lmsg);
 407         else
 408                 m_freem(nmp->nm_packet);
 409 }
 410
 411 #endif  /* SMP */
 412
 413 void
 414 ipflow_create(const struct route *ro, struct mbuf *m)
 415 {
 416         const struct ip *const ip = mtod(m, struct ip *);
 417 #ifdef SMP
 418         struct netmsg_packet *nmp;
 419         struct netmsg *nmsg;
 420         int nextcpu;
 421 #endif
 422
 423         /*
 424          * Don't create cache entries for ICMP messages.
 425          */
 426         if (!ipflow_active || ip->ip_p == IPPROTO_ICMP) {
 427                 m_freem(m);
 428                 return;
 429         }
 430
 431 #ifdef SMP
 432         nmp = &m->m_hdr.mh_netmsg;
 433         nmsg = &nmp->nm_netmsg;
 434
 435         netmsg_init(nmsg, &netisr_apanic_rport, 0, ipflow_create_dispatch);
 436         nmp->nm_packet = m;
 437         nmsg->nm_lmsg.u.ms_result32 =
 438                 ((const struct sockaddr_in *)&ro->ro_dst)->sin_addr.s_addr;
 439
 440         if (mycpuid == 0) {
 441                 ipflow_create_oncpu(ro, m);
 442                 nextcpu = 1;
 443         } else {
 444                 nextcpu = 0;
 445         }
 446         if (nextcpu < ncpus)
 447                 lwkt_sendmsg(cpu_portfn(nextcpu), &nmsg->nm_lmsg);
 448         else
 449                 m_freem(m);
 450 #else
 451         ipflow_create_oncpu(ro, m);
 452         m_freem(m);
 453 #endif
 454 }
 455
 456 static void
 457 ipflow_init(void)
 458 {
 459         char oid_name[32];
 460         int i;
 461
 462         for (i = 0; i < ncpus; ++i) {
 463                 netmsg_init(&ipflow_timo_netmsgs[i], &netisr_adone_rport, 0,
 464                             ipflow_timo_dispatch);
 465
 466                 ksnprintf(oid_name, sizeof(oid_name), "inuse%d", i);
 467
 468                 SYSCTL_ADD_INT(NULL,
 469                 SYSCTL_STATIC_CHILDREN(_net_inet_ip_ipflow),
 470                 OID_AUTO, oid_name, CTLFLAG_RD, &ipflow_inuse_pcpu[i], 0,
 471                 "# of ip flow being used");
 472         }
 473 }
 474 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, ipflow_init, 0);