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 1.5 2004/12/21 02:54:15 hsu Exp $
  38  */
  39
  40 #include <sys/param.h>
  41 #include <sys/systm.h>
  42 #include <sys/malloc.h>
  43 #include <sys/mbuf.h>
  44 #include <sys/globaldata.h>
  45 #include <sys/thread.h>
  46 #include <sys/protosw.h>
  47 #include <sys/socket.h>
  48 #include <sys/kernel.h>
  49
  50 #include <sys/sysctl.h>
  51
  52 #include <net/if.h>
  53 #include <net/route.h>
  54
  55 #include <netinet/in.h>
  56 #include <netinet/in_systm.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 static LIST_HEAD(ipflowhead, ipflow) ipflows[IPFLOW_HASHSIZE];
  66 static int ipflow_inuse;
  67 #define IPFLOW_MAX              256
  68
  69 static int ipflow_active = 0;
  70 SYSCTL_INT(_net_inet_ip, IPCTL_FASTFORWARDING, fastforwarding, CTLFLAG_RW,
  71     &ipflow_active, 0, "Enable flow-based IP forwarding");
  72
  73 static MALLOC_DEFINE(M_IPFLOW, "ip_flow", "IP flow");
  74
  75 static unsigned
  76 ipflow_hash(
  77         struct in_addr dst,
  78         struct in_addr src,
  79         unsigned tos)
  80 {
  81         unsigned hash = tos;
  82         int idx;
  83         for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS)
  84                 hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx);
  85         return hash & (IPFLOW_HASHSIZE-1);
  86 }
  87
  88 static struct ipflow *
  89 ipflow_lookup(
  90         const struct ip *ip)
  91 {
  92         unsigned hash;
  93         struct ipflow *ipf;
  94
  95         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
  96
  97         ipf = LIST_FIRST(&ipflows[hash]);
  98         while (ipf != NULL) {
  99                 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
 100                     && ip->ip_src.s_addr == ipf->ipf_src.s_addr
 101                     && ip->ip_tos == ipf->ipf_tos)
 102                         break;
 103                 ipf = LIST_NEXT(ipf, ipf_next);
 104         }
 105         return ipf;
 106 }
 107
 108 int
 109 ipflow_fastforward(
 110         struct mbuf *m)
 111 {
 112         struct ip *ip;
 113         struct ipflow *ipf;
 114         struct rtentry *rt;
 115         struct sockaddr *dst;
 116         int error;
 117
 118         /*
 119          * Are we forwarding packets?  Big enough for an IP packet?
 120          */
 121         if (!ipforwarding || !ipflow_active || m->m_len < sizeof(struct ip))
 122                 return 0;
 123         /*
 124          * IP header with no option and valid version and length
 125          */
 126         ip = mtod(m, struct ip *);
 127         if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2)
 128             || ntohs(ip->ip_len) > m->m_pkthdr.len)
 129                 return 0;
 130         /*
 131          * Find a flow.
 132          */
 133         if ((ipf = ipflow_lookup(ip)) == NULL)
 134                 return 0;
 135
 136         /*
 137          * Route and interface still up?
 138          */
 139         rt = ipf->ipf_ro.ro_rt;
 140         if ((rt->rt_flags & RTF_UP) == 0 || (rt->rt_ifp->if_flags & IFF_UP) == 0)
 141                 return 0;
 142
 143         /*
 144          * Packet size OK?  TTL?
 145          */
 146         if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
 147                 return 0;
 148
 149         /*
 150          * Everything checks out and so we can forward this packet.
 151          * Modify the TTL and incrementally change the checksum.
 152          */
 153         ip->ip_ttl -= IPTTLDEC;
 154         if (ip->ip_sum >= htons(0xffff - (IPTTLDEC << 8))) {
 155                 ip->ip_sum += htons(IPTTLDEC << 8) + 1;
 156         } else {
 157                 ip->ip_sum += htons(IPTTLDEC << 8);
 158         }
 159
 160         /*
 161          * Send the packet on its way.  All we can get back is ENOBUFS
 162          */
 163         ipf->ipf_uses++;
 164         ipf->ipf_timer = IPFLOW_TIMER;
 165
 166         if (rt->rt_flags & RTF_GATEWAY)
 167                 dst = rt->rt_gateway;
 168         else
 169                 dst = &ipf->ipf_ro.ro_dst;
 170         if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
 171                 if (error == ENOBUFS)
 172                         ipf->ipf_dropped++;
 173                 else
 174                         ipf->ipf_errors++;
 175         }
 176         return 1;
 177 }
 178 \f
 179 static void
 180 ipflow_addstats(
 181         struct ipflow *ipf)
 182 {
 183         ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 184         ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
 185         ipstat.ips_forward += ipf->ipf_uses;
 186         ipstat.ips_fastforward += ipf->ipf_uses;
 187 }
 188
 189 static void
 190 ipflow_free(
 191         struct ipflow *ipf)
 192 {
 193         int s;
 194
 195         /*
 196          * Remove the flow from the hash table (at elevated IPL).
 197          * Once it's off the list, we can deal with it at normal
 198          * network IPL.
 199          */
 200         s = splimp();
 201         LIST_REMOVE(ipf, ipf_next);
 202         splx(s);
 203         ipflow_addstats(ipf);
 204         RTFREE(ipf->ipf_ro.ro_rt);
 205         ipflow_inuse--;
 206         free(ipf, M_IPFLOW);
 207 }
 208
 209 static struct ipflow *
 210 ipflow_reap(
 211         void)
 212 {
 213         struct ipflow *ipf, *maybe_ipf = NULL;
 214         int idx;
 215         int s;
 216
 217         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 218                 ipf = LIST_FIRST(&ipflows[idx]);
 219                 while (ipf != NULL) {
 220                         /*
 221                          * If this no longer points to a valid route
 222                          * reclaim it.
 223                          */
 224                         if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
 225                                 goto done;
 226                         /*
 227                          * choose the one that's been least recently used
 228                          * or has had the least uses in the last 1.5
 229                          * intervals.
 230                          */
 231                         if (maybe_ipf == NULL
 232                             || ipf->ipf_timer < maybe_ipf->ipf_timer
 233                             || (ipf->ipf_timer == maybe_ipf->ipf_timer
 234                                 && ipf->ipf_last_uses + ipf->ipf_uses <
 235                                       maybe_ipf->ipf_last_uses +
 236                                         maybe_ipf->ipf_uses))
 237                                 maybe_ipf = ipf;
 238                         ipf = LIST_NEXT(ipf, ipf_next);
 239                 }
 240         }
 241         ipf = maybe_ipf;
 242     done:
 243         /*
 244          * Remove the entry from the flow table.
 245          */
 246         s = splimp();
 247         LIST_REMOVE(ipf, ipf_next);
 248         splx(s);
 249         ipflow_addstats(ipf);
 250         RTFREE(ipf->ipf_ro.ro_rt);
 251         return ipf;
 252 }
 253
 254 void
 255 ipflow_slowtimo(
 256         void)
 257 {
 258         struct ipflow *ipf;
 259         int idx;
 260
 261         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 262                 ipf = LIST_FIRST(&ipflows[idx]);
 263                 while (ipf != NULL) {
 264                         struct ipflow *next_ipf = LIST_NEXT(ipf, ipf_next);
 265                         if (--ipf->ipf_timer == 0) {
 266                                 ipflow_free(ipf);
 267                         } else {
 268                                 ipf->ipf_last_uses = ipf->ipf_uses;
 269                                 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 270                                 ipstat.ips_forward += ipf->ipf_uses;
 271                                 ipstat.ips_fastforward += ipf->ipf_uses;
 272                                 ipf->ipf_uses = 0;
 273                         }
 274                         ipf = next_ipf;
 275                 }
 276         }
 277 }
 278
 279 void
 280 ipflow_create(const struct route *ro, struct mbuf *m)
 281 {
 282         const struct ip *const ip = mtod(m, struct ip *);
 283         struct ipflow *ipf;
 284         unsigned hash;
 285         int s;
 286
 287         /*
 288          * Don't create cache entries for ICMP messages.
 289          */
 290         if (!ipflow_active || ip->ip_p == IPPROTO_ICMP)
 291                 return;
 292         /*
 293          * See if an existing flow struct exists.  If so remove it from it's
 294          * list and free the old route.  If not, try to malloc a new one
 295          * (if we aren't at our limit).
 296          */
 297         ipf = ipflow_lookup(ip);
 298         if (ipf == NULL) {
 299                 if (ipflow_inuse == IPFLOW_MAX) {
 300                         ipf = ipflow_reap();
 301                 } else {
 302                         ipf = malloc(sizeof(*ipf), M_IPFLOW,
 303                                      M_INTWAIT | M_NULLOK);
 304                         if (ipf == NULL)
 305                                 return;
 306                         ipflow_inuse++;
 307                 }
 308                 bzero((caddr_t) ipf, sizeof(*ipf));
 309         } else {
 310                 s = splimp();
 311                 LIST_REMOVE(ipf, ipf_next);
 312                 splx(s);
 313                 ipflow_addstats(ipf);
 314                 RTFREE(ipf->ipf_ro.ro_rt);
 315                 ipf->ipf_uses = ipf->ipf_last_uses = 0;
 316                 ipf->ipf_errors = ipf->ipf_dropped = 0;
 317         }
 318
 319         /*
 320          * Fill in the updated information.
 321          */
 322         ipf->ipf_ro = *ro;
 323         ro->ro_rt->rt_refcnt++;
 324         ipf->ipf_dst = ip->ip_dst;
 325         ipf->ipf_src = ip->ip_src;
 326         ipf->ipf_tos = ip->ip_tos;
 327         ipf->ipf_timer = IPFLOW_TIMER;
 328         /*
 329          * Insert into the approriate bucket of the flow table.
 330          */
 331         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
 332         s = splimp();
 333         LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next);
 334         splx(s);
 335 }