bsd/netinet/ip_flow.c

   1 /*
   2  * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
   7  *
   8  * This file contains Original Code and/or Modifications of Original Code
   9  * as defined in and that are subject to the Apple Public Source License
  10  * Version 2.0 (the 'License'). You may not use this file except in
  11  * compliance with the License. Please obtain a copy of the License at
  12  * http://www.opensource.apple.com/apsl/ and read it before using this
  13  * file.
  14  *
  15  * The Original Code and all software distributed under the License are
  16  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  17  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  18  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  20  * Please see the License for the specific language governing rights and
  21  * limitations under the License.
  22  *
  23  * @APPLE_LICENSE_HEADER_END@
  24  */
  25 /*-
  26  * Copyright (c) 1998 The NetBSD Foundation, Inc.
  27  * All rights reserved.
  28  *
  29  * This code is derived from software contributed to The NetBSD Foundation
  30  * by the 3am Software Foundry ("3am").  It was developed by Matt Thomas.
  31  *
  32  * Redistribution and use in source and binary forms, with or without
  33  * modification, are permitted provided that the following conditions
  34  * are met:
  35  * 1. Redistributions of source code must retain the above copyright
  36  *    notice, this list of conditions and the following disclaimer.
  37  * 2. Redistributions in binary form must reproduce the above copyright
  38  *    notice, this list of conditions and the following disclaimer in the
  39  *    documentation and/or other materials provided with the distribution.
  40  * 3. All advertising materials mentioning features or use of this software
  41  *    must display the following acknowledgement:
  42  *      This product includes software developed by the NetBSD
  43  *      Foundation, Inc. and its contributors.
  44  * 4. Neither the name of The NetBSD Foundation nor the names of its
  45  *    contributors may be used to endorse or promote products derived
  46  *    from this software without specific prior written permission.
  47  *
  48  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
  49  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
  50  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  51  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
  52  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  53  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  54  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  55  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  56  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  57  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  58  * POSSIBILITY OF SUCH DAMAGE.
  59  *
  60  * $FreeBSD: src/sys/netinet/ip_flow.c,v 1.9.2.1 2001/08/08 08:20:35 ru Exp $
  61  */
  62
  63 #include <sys/param.h>
  64 #include <sys/systm.h>
  65 #include <sys/malloc.h>
  66 #include <sys/mbuf.h>
  67 #include <sys/protosw.h>
  68 #include <sys/socket.h>
  69 #include <sys/kernel.h>
  70
  71 #include <sys/sysctl.h>
  72
  73 #include <net/if.h>
  74 #include <net/route.h>
  75
  76 #include <netinet/in.h>
  77 #include <netinet/in_systm.h>
  78 #include <netinet/ip.h>
  79 #include <netinet/in_var.h>
  80 #include <netinet/ip_var.h>
  81 #include <netinet/ip_flow.h>
  82 #include <net/dlil.h>
  83
  84 #define IPFLOW_TIMER            (5 * PR_SLOWHZ)
  85 #define IPFLOW_HASHBITS         6       /* should not be a multiple of 8 */
  86 #define IPFLOW_HASHSIZE         (1 << IPFLOW_HASHBITS)
  87 static LIST_HEAD(ipflowhead, ipflow) ipflows[IPFLOW_HASHSIZE];
  88 static int ipflow_inuse;
  89 #define IPFLOW_MAX              256
  90
  91 #ifdef __APPLE__
  92 #define M_IPFLOW M_TEMP
  93 #endif
  94
  95 static int ipflow_active = 0;
  96 SYSCTL_INT(_net_inet_ip, IPCTL_FASTFORWARDING, fastforwarding, CTLFLAG_RW,
  97     &ipflow_active, 0, "Enable flow-based IP forwarding");
  98
  99 #ifndef __APPLE__
 100 static MALLOC_DEFINE(M_IPFLOW, "ip_flow", "IP flow");
 101 #endif
 102
 103 static unsigned
 104 ipflow_hash(
 105         struct in_addr dst,
 106         struct in_addr src,
 107         unsigned tos)
 108 {
 109         unsigned hash = tos;
 110         int idx;
 111         for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS)
 112                 hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx);
 113         return hash & (IPFLOW_HASHSIZE-1);
 114 }
 115
 116 static struct ipflow *
 117 ipflow_lookup(
 118         const struct ip *ip)
 119 {
 120         unsigned hash;
 121         struct ipflow *ipf;
 122
 123         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
 124
 125         ipf = LIST_FIRST(&ipflows[hash]);
 126         while (ipf != NULL) {
 127                 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
 128                     && ip->ip_src.s_addr == ipf->ipf_src.s_addr
 129                     && ip->ip_tos == ipf->ipf_tos)
 130                         break;
 131                 ipf = LIST_NEXT(ipf, ipf_next);
 132         }
 133         return ipf;
 134 }
 135
 136 int
 137 ipflow_fastforward(
 138         struct mbuf *m)
 139 {
 140         struct ip *ip;
 141         struct ipflow *ipf;
 142         struct rtentry *rt;
 143         struct sockaddr *dst;
 144         int error;
 145
 146         /*
 147          * Are we forwarding packets?  Big enough for an IP packet?
 148          */
 149         if (!ipforwarding || !ipflow_active || m->m_len < sizeof(struct ip))
 150                 return 0;
 151         /*
 152          * IP header with no option and valid version and length
 153          */
 154         ip = mtod(m, struct ip *);
 155         if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2)
 156             || ntohs(ip->ip_len) > m->m_pkthdr.len)
 157                 return 0;
 158         /*
 159          * Find a flow.
 160          */
 161         if ((ipf = ipflow_lookup(ip)) == NULL)
 162                 return 0;
 163
 164         /*
 165          * Route and interface still up?
 166          */
 167         rt = ipf->ipf_ro.ro_rt;
 168         if ((rt->rt_flags & RTF_UP) == 0 || (rt->rt_ifp->if_flags & IFF_UP) == 0)
 169                 return 0;
 170
 171         /*
 172          * Packet size OK?  TTL?
 173          */
 174         if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
 175                 return 0;
 176
 177         /*
 178          * Everything checks out and so we can forward this packet.
 179          * Modify the TTL and incrementally change the checksum.
 180          */
 181         ip->ip_ttl -= IPTTLDEC;
 182         if (ip->ip_sum >= htons(0xffff - (IPTTLDEC << 8))) {
 183                 ip->ip_sum += htons(IPTTLDEC << 8) + 1;
 184         } else {
 185                 ip->ip_sum += htons(IPTTLDEC << 8);
 186         }
 187
 188         /*
 189          * Send the packet on its way.  All we can get back is ENOBUFS
 190          */
 191         ipf->ipf_uses++;
 192         ipf->ipf_timer = IPFLOW_TIMER;
 193
 194         if (rt->rt_flags & RTF_GATEWAY)
 195                 dst = rt->rt_gateway;
 196         else
 197                 dst = &ipf->ipf_ro.ro_dst;
 198 #ifdef __APPLE__
 199         /* Not sure the rt_dlt is valid here !! XXX */
 200         if ((error = dlil_output((u_long)rt->rt_dlt, m, (caddr_t) rt, dst, 0)) != 0) {
 201
 202 #else
 203         if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
 204 #endif
 205                 if (error == ENOBUFS)
 206                         ipf->ipf_dropped++;
 207                 else
 208                         ipf->ipf_errors++;
 209         }
 210         return 1;
 211 }
 212 \f
 213 static void
 214 ipflow_addstats(
 215         struct ipflow *ipf)
 216 {
 217         ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 218         ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
 219         ipstat.ips_forward += ipf->ipf_uses;
 220         ipstat.ips_fastforward += ipf->ipf_uses;
 221 }
 222
 223 static void
 224 ipflow_free(
 225         struct ipflow *ipf)
 226 {
 227         int s;
 228         /*
 229          * Remove the flow from the hash table (at elevated IPL).
 230          * Once it's off the list, we can deal with it at normal
 231          * network IPL.
 232          */
 233         s = splimp();
 234         LIST_REMOVE(ipf, ipf_next);
 235         splx(s);
 236         ipflow_addstats(ipf);
 237         rtfree(ipf->ipf_ro.ro_rt);
 238         ipflow_inuse--;
 239         FREE(ipf, M_IPFLOW);
 240 }
 241
 242 static struct ipflow *
 243 ipflow_reap(
 244         void)
 245 {
 246         struct ipflow *ipf, *maybe_ipf = NULL;
 247         int idx;
 248         int s;
 249
 250         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 251                 ipf = LIST_FIRST(&ipflows[idx]);
 252                 while (ipf != NULL) {
 253                         /*
 254                          * If this no longer points to a valid route
 255                          * reclaim it.
 256                          */
 257                         if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
 258                                 goto done;
 259                         /*
 260                          * choose the one that's been least recently used
 261                          * or has had the least uses in the last 1.5
 262                          * intervals.
 263                          */
 264                         if (maybe_ipf == NULL
 265                             || ipf->ipf_timer < maybe_ipf->ipf_timer
 266                             || (ipf->ipf_timer == maybe_ipf->ipf_timer
 267                                 && ipf->ipf_last_uses + ipf->ipf_uses <
 268                                       maybe_ipf->ipf_last_uses +
 269                                         maybe_ipf->ipf_uses))
 270                                 maybe_ipf = ipf;
 271                         ipf = LIST_NEXT(ipf, ipf_next);
 272                 }
 273         }
 274         ipf = maybe_ipf;
 275     done:
 276         /*
 277          * Remove the entry from the flow table.
 278          */
 279         s = splimp();
 280         LIST_REMOVE(ipf, ipf_next);
 281         splx(s);
 282         ipflow_addstats(ipf);
 283         rtfree(ipf->ipf_ro.ro_rt);
 284         return ipf;
 285 }
 286
 287 void
 288 ipflow_slowtimo(
 289         void)
 290 {
 291         struct ipflow *ipf;
 292         int idx;
 293
 294         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 295                 ipf = LIST_FIRST(&ipflows[idx]);
 296                 while (ipf != NULL) {
 297                         struct ipflow *next_ipf = LIST_NEXT(ipf, ipf_next);
 298                         if (--ipf->ipf_timer == 0) {
 299                                 ipflow_free(ipf);
 300                         } else {
 301                                 ipf->ipf_last_uses = ipf->ipf_uses;
 302                                 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 303                                 ipstat.ips_forward += ipf->ipf_uses;
 304                                 ipstat.ips_fastforward += ipf->ipf_uses;
 305                                 ipf->ipf_uses = 0;
 306                         }
 307                         ipf = next_ipf;
 308                 }
 309         }
 310 }
 311
 312 void
 313 ipflow_create(
 314         const struct route *ro,
 315         struct mbuf *m)
 316 {
 317         const struct ip *const ip = mtod(m, struct ip *);
 318         struct ipflow *ipf;
 319         unsigned hash;
 320         int s;
 321
 322         /*
 323          * Don't create cache entries for ICMP messages.
 324          */
 325         if (!ipflow_active || ip->ip_p == IPPROTO_ICMP)
 326                 return;
 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 = (struct ipflow *) _MALLOC(sizeof(*ipf), M_IPFLOW,
 338                                                        M_NOWAIT);
 339                         if (ipf == NULL)
 340                                 return;
 341                         ipflow_inuse++;
 342                 }
 343                 bzero((caddr_t) ipf, sizeof(*ipf));
 344         } else {
 345                 s = splimp();
 346                 LIST_REMOVE(ipf, ipf_next);
 347                 splx(s);
 348                 ipflow_addstats(ipf);
 349                 rtfree(ipf->ipf_ro.ro_rt);
 350                 ipf->ipf_uses = ipf->ipf_last_uses = 0;
 351                 ipf->ipf_errors = ipf->ipf_dropped = 0;
 352         }
 353
 354         /*
 355          * Fill in the updated information.
 356          */
 357         ipf->ipf_ro = *ro;
 358         rtref(ro->ro_rt);
 359         ipf->ipf_dst = ip->ip_dst;
 360         ipf->ipf_src = ip->ip_src;
 361         ipf->ipf_tos = ip->ip_tos;
 362         ipf->ipf_timer = IPFLOW_TIMER;
 363         /*
 364          * Insert into the approriate bucket of the flow table.
 365          */
 366         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
 367         s = splimp();
 368         LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next);
 369         splx(s);
 370 }