bsd/netinet/ip_flow.c

   1 /*
   2  * Copyright (c) 2000-2008 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /*-
  29  * Copyright (c) 1998 The NetBSD Foundation, Inc.
  30  * All rights reserved.
  31  *
  32  * This code is derived from software contributed to The NetBSD Foundation
  33  * by the 3am Software Foundry ("3am").  It was developed by Matt Thomas.
  34  *
  35  * Redistribution and use in source and binary forms, with or without
  36  * modification, are permitted provided that the following conditions
  37  * are met:
  38  * 1. Redistributions of source code must retain the above copyright
  39  *    notice, this list of conditions and the following disclaimer.
  40  * 2. Redistributions in binary form must reproduce the above copyright
  41  *    notice, this list of conditions and the following disclaimer in the
  42  *    documentation and/or other materials provided with the distribution.
  43  * 3. All advertising materials mentioning features or use of this software
  44  *    must display the following acknowledgement:
  45  *      This product includes software developed by the NetBSD
  46  *      Foundation, Inc. and its contributors.
  47  * 4. Neither the name of The NetBSD Foundation nor the names of its
  48  *    contributors may be used to endorse or promote products derived
  49  *    from this software without specific prior written permission.
  50  *
  51  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
  52  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
  53  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  54  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
  55  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  56  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  57  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  58  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  59  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  60  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  61  * POSSIBILITY OF SUCH DAMAGE.
  62  *
  63  * $FreeBSD: src/sys/netinet/ip_flow.c,v 1.9.2.1 2001/08/08 08:20:35 ru Exp $
  64  */
  65
  66 #include <sys/param.h>
  67 #include <sys/systm.h>
  68 #include <sys/malloc.h>
  69 #include <sys/mbuf.h>
  70 #include <sys/protosw.h>
  71 #include <sys/socket.h>
  72 #include <sys/kernel.h>
  73
  74 #include <sys/sysctl.h>
  75 #include <libkern/OSAtomic.h>
  76
  77 #include <net/if.h>
  78 #include <net/route.h>
  79
  80 #include <netinet/in.h>
  81 #include <netinet/in_systm.h>
  82 #include <netinet/ip.h>
  83 #include <netinet/in_var.h>
  84 #include <netinet/ip_var.h>
  85 #include <netinet/ip_flow.h>
  86 #include <net/dlil.h>
  87
  88 #if IPFLOW
  89
  90 #define IPFLOW_TIMER            (5 * PR_SLOWHZ)
  91 #define IPFLOW_HASHBITS         6       /* should not be a multiple of 8 */
  92 #define IPFLOW_HASHSIZE         (1 << IPFLOW_HASHBITS)
  93 static LIST_HEAD(ipflowhead, ipflow) ipflows[IPFLOW_HASHSIZE];
  94 static int ipflow_inuse;
  95 #define IPFLOW_MAX              256
  96
  97 #ifdef __APPLE__
  98 #define M_IPFLOW M_TEMP
  99 #endif
 100
 101 static int ipflow_active = 0;
 102 SYSCTL_INT(_net_inet_ip, IPCTL_FASTFORWARDING, fastforwarding, CTLFLAG_RW,
 103     &ipflow_active, 0, "Enable flow-based IP forwarding");
 104
 105 #ifndef __APPLE__
 106 static MALLOC_DEFINE(M_IPFLOW, "ip_flow", "IP flow");
 107 #endif
 108
 109 static unsigned
 110 ipflow_hash(
 111         struct in_addr dst,
 112         struct in_addr src,
 113         unsigned tos)
 114 {
 115         unsigned hash = tos;
 116         int idx;
 117         for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS)
 118                 hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx);
 119         return hash & (IPFLOW_HASHSIZE-1);
 120 }
 121
 122 static struct ipflow *
 123 ipflow_lookup(
 124         const struct ip *ip)
 125 {
 126         unsigned hash;
 127         struct ipflow *ipf;
 128
 129         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
 130
 131         ipf = LIST_FIRST(&ipflows[hash]);
 132         while (ipf != NULL) {
 133                 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
 134                     && ip->ip_src.s_addr == ipf->ipf_src.s_addr
 135                     && ip->ip_tos == ipf->ipf_tos)
 136                         break;
 137                 ipf = LIST_NEXT(ipf, ipf_next);
 138         }
 139         return ipf;
 140 }
 141
 142 int
 143 ipflow_fastforward(
 144         struct mbuf *m)
 145 {
 146         struct ip *ip;
 147         struct ipflow *ipf;
 148         struct rtentry *rt;
 149         struct sockaddr *dst;
 150         int error;
 151
 152         /*
 153          * Are we forwarding packets?  Big enough for an IP packet?
 154          */
 155         if (!ipforwarding || !ipflow_active || m->m_len < sizeof(struct ip))
 156                 return 0;
 157         /*
 158          * IP header with no option and valid version and length
 159          */
 160         ip = mtod(m, struct ip *);
 161         if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2)
 162             || ntohs(ip->ip_len) > m->m_pkthdr.len)
 163                 return 0;
 164         /*
 165          * Find a flow.
 166          */
 167         if ((ipf = ipflow_lookup(ip)) == NULL)
 168                 return 0;
 169
 170         /*
 171          * Route and interface still up?
 172          */
 173         rt = ipf->ipf_ro.ro_rt;
 174         if ((rt->rt_flags & RTF_UP) == 0 || (rt->rt_ifp->if_flags & IFF_UP) == 0)
 175                 return 0;
 176
 177         /*
 178          * Packet size OK?  TTL?
 179          */
 180         if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
 181                 return 0;
 182
 183         /*
 184          * Everything checks out and so we can forward this packet.
 185          * Modify the TTL and incrementally change the checksum.
 186          */
 187         ip->ip_ttl -= IPTTLDEC;
 188         if (ip->ip_sum >= htons(0xffff - (IPTTLDEC << 8))) {
 189                 ip->ip_sum += htons(IPTTLDEC << 8) + 1;
 190         } else {
 191                 ip->ip_sum += htons(IPTTLDEC << 8);
 192         }
 193
 194         /*
 195          * Send the packet on its way.  All we can get back is ENOBUFS
 196          */
 197         ipf->ipf_uses++;
 198         ipf->ipf_timer = IPFLOW_TIMER;
 199
 200         if (rt->rt_flags & RTF_GATEWAY)
 201                 dst = rt->rt_gateway;
 202         else
 203                 dst = &ipf->ipf_ro.ro_dst;
 204 #ifdef __APPLE__
 205         /* Not sure the rt_dlt is valid here !! XXX */
 206         if ((error = dlil_output(rt->rt_ifp, PF_INET, m, (caddr_t) rt, dst, 0)) != 0) {
 207
 208 #else
 209         if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
 210 #endif
 211                 if (error == ENOBUFS)
 212                         ipf->ipf_dropped++;
 213                 else
 214                         ipf->ipf_errors++;
 215         }
 216         return 1;
 217 }
 218 \f
 219 static void
 220 ipflow_addstats(
 221         struct ipflow *ipf)
 222 {
 223         ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 224         OSAddAtomic(ipf->ipf_errors + ipf->ipf_dropped, &ipstat.ips_cantforward);
 225         OSAddAtomic(ipf->ipf_uses, &ipstat.ips_forward);
 226         OSAddAtomic(ipf->ipf_uses, &ipstat.ips_fastforward);
 227 }
 228
 229 static void
 230 ipflow_free(
 231         struct ipflow *ipf)
 232 {
 233         /*
 234          * Remove the flow from the hash table (at elevated IPL).
 235          * Once it's off the list, we can deal with it at normal
 236          * network IPL.
 237          */
 238         LIST_REMOVE(ipf, ipf_next);
 239         ipflow_addstats(ipf);
 240         rtfree(ipf->ipf_ro.ro_rt);
 241         ipflow_inuse--;
 242         FREE(ipf, M_IPFLOW);
 243 }
 244
 245 static struct ipflow *
 246 ipflow_reap(
 247         void)
 248 {
 249         struct ipflow *ipf, *maybe_ipf = NULL;
 250         int idx;
 251
 252         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 253                 ipf = LIST_FIRST(&ipflows[idx]);
 254                 while (ipf != NULL) {
 255                         /*
 256                          * If this no longer points to a valid route
 257                          * reclaim it.
 258                          */
 259                         if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
 260                                 goto done;
 261                         /*
 262                          * choose the one that's been least recently used
 263                          * or has had the least uses in the last 1.5
 264                          * intervals.
 265                          */
 266                         if (maybe_ipf == NULL
 267                             || ipf->ipf_timer < maybe_ipf->ipf_timer
 268                             || (ipf->ipf_timer == maybe_ipf->ipf_timer
 269                                 && ipf->ipf_last_uses + ipf->ipf_uses <
 270                                       maybe_ipf->ipf_last_uses +
 271                                         maybe_ipf->ipf_uses))
 272                                 maybe_ipf = ipf;
 273                         ipf = LIST_NEXT(ipf, ipf_next);
 274                 }
 275         }
 276         ipf = maybe_ipf;
 277     done:
 278         /*
 279          * Remove the entry from the flow table.
 280          */
 281         LIST_REMOVE(ipf, ipf_next);
 282         ipflow_addstats(ipf);
 283         rtfree(ipf->ipf_ro.ro_rt);
 284         return ipf;
 285 }
 286 /* note: called under the ip_mutex lock */
 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                                 OSAddAtomic(ipf->ipf_uses, &ipstat.ips_forward);
 304                                 OSAddAtomic(ipf->ipf_uses, &ipstat.ips_fastforward);
 305                                 ipstat.ips_forward += ipf->ipf_uses;
 306                                 ipstat.ips_fastforward += ipf->ipf_uses;
 307                                 ipf->ipf_uses = 0;
 308                         }
 309                         ipf = next_ipf;
 310                 }
 311         }
 312 }
 313
 314 void
 315 ipflow_create(
 316         const struct route *ro,
 317         struct mbuf *m)
 318 {
 319         const struct ip *const ip = mtod(m, struct ip *);
 320         struct ipflow *ipf;
 321         unsigned hash;
 322
 323         /*
 324          * Don't create cache entries for ICMP messages.
 325          */
 326         if (!ipflow_active || ip->ip_p == IPPROTO_ICMP)
 327                 return;
 328         /*
 329          * See if an existing flow struct exists.  If so remove it from it's
 330          * list and free the old route.  If not, try to malloc a new one
 331          * (if we aren't at our limit).
 332          */
 333         ipf = ipflow_lookup(ip);
 334         if (ipf == NULL) {
 335                 if (ipflow_inuse == IPFLOW_MAX) {
 336                         ipf = ipflow_reap();
 337                 } else {
 338                         ipf = (struct ipflow *) _MALLOC(sizeof(*ipf), M_IPFLOW,
 339                                                        M_NOWAIT);
 340                         if (ipf == NULL)
 341                                 return;
 342                         ipflow_inuse++;
 343                 }
 344                 bzero((caddr_t) ipf, sizeof(*ipf));
 345         } else {
 346                 LIST_REMOVE(ipf, ipf_next);
 347                 ipflow_addstats(ipf);
 348                 rtfree(ipf->ipf_ro.ro_rt);
 349                 ipf->ipf_uses = ipf->ipf_last_uses = 0;
 350                 ipf->ipf_errors = ipf->ipf_dropped = 0;
 351         }
 352
 353         /*
 354          * Fill in the updated information.
 355          */
 356         ipf->ipf_ro = *ro;
 357         RT_ADDREF(ro->ro_rt);
 358         ipf->ipf_dst = ip->ip_dst;
 359         ipf->ipf_src = ip->ip_src;
 360         ipf->ipf_tos = ip->ip_tos;
 361         ipf->ipf_timer = IPFLOW_TIMER;
 362         /*
 363          * Insert into the approriate bucket of the flow table.
 364          */
 365         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
 366         LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next);
 367 }
 368 #else /* !IPFLOW */
 369 int
 370 ipflow_fastforward(struct mbuf *m)
 371 {
 372 #pragma unused(m)
 373         /*
 374          * Since this symbol is exported (albeit unsupported), just return
 375          * false to keep things (e.g. PPP) happy, in case ipflow is not
 376          * compiled in.
 377          */
 378         return (0);
 379 }
 380 #endif /* !IPFLOW */