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 1.10 2006/09/05 00:55:48 dillon 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 #include <sys/thread2.h>
  52
  53 #include <net/if.h>
  54 #include <net/route.h>
  55
  56 #include <netinet/in.h>
  57 #include <netinet/in_systm.h>
  58 #include <netinet/ip.h>
  59 #include <netinet/in_var.h>
  60 #include <netinet/ip_var.h>
  61 #include <netinet/ip_flow.h>
  62
  63 #define IPFLOW_TIMER            (5 * PR_SLOWHZ)
  64 #define IPFLOW_HASHBITS         6       /* should not be a multiple of 8 */
  65 #define IPFLOW_HASHSIZE         (1 << IPFLOW_HASHBITS)
  66 static LIST_HEAD(ipflowhead, ipflow) ipflows[IPFLOW_HASHSIZE];
  67 static int ipflow_inuse;
  68 #define IPFLOW_MAX              256
  69
  70 static int ipflow_active = 0;
  71 SYSCTL_INT(_net_inet_ip, IPCTL_FASTFORWARDING, fastforwarding, CTLFLAG_RW,
  72     &ipflow_active, 0, "Enable flow-based IP forwarding");
  73
  74 static MALLOC_DEFINE(M_IPFLOW, "ip_flow", "IP flow");
  75
  76 static unsigned
  77 ipflow_hash(struct in_addr dst, struct in_addr src, unsigned tos)
  78 {
  79         unsigned hash = tos;
  80         int idx;
  81
  82         for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS)
  83                 hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx);
  84         return hash & (IPFLOW_HASHSIZE-1);
  85 }
  86
  87 static struct ipflow *
  88 ipflow_lookup(const struct ip *ip)
  89 {
  90         unsigned hash;
  91         struct ipflow *ipf;
  92
  93         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
  94
  95         ipf = LIST_FIRST(&ipflows[hash]);
  96         while (ipf != NULL) {
  97                 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr &&
  98                     ip->ip_src.s_addr == ipf->ipf_src.s_addr &&
  99                     ip->ip_tos == ipf->ipf_tos)
 100                         break;
 101                 ipf = LIST_NEXT(ipf, ipf_next);
 102         }
 103         return ipf;
 104 }
 105
 106 int
 107 ipflow_fastforward(struct mbuf *m, lwkt_serialize_t serializer)
 108 {
 109         struct ip *ip;
 110         struct ipflow *ipf;
 111         struct rtentry *rt;
 112         struct sockaddr *dst;
 113         int error;
 114
 115         /*
 116          * Are we forwarding packets?  Big enough for an IP packet?
 117          */
 118         if (!ipforwarding || !ipflow_active || m->m_len < sizeof(struct ip))
 119                 return 0;
 120         /*
 121          * IP header with no option and valid version and length
 122          */
 123         ip = mtod(m, struct ip *);
 124         if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
 125             ntohs(ip->ip_len) > m->m_pkthdr.len)
 126                 return 0;
 127         /*
 128          * Find a flow.
 129          */
 130         if ((ipf = ipflow_lookup(ip)) == NULL)
 131                 return 0;
 132
 133         /*
 134          * Route and interface still up?
 135          */
 136         rt = ipf->ipf_ro.ro_rt;
 137         if ((rt->rt_flags & RTF_UP) == 0 || (rt->rt_ifp->if_flags & IFF_UP) == 0)
 138                 return 0;
 139
 140         /*
 141          * Packet size OK?  TTL?
 142          */
 143         if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
 144                 return 0;
 145
 146         /*
 147          * Everything checks out and so we can forward this packet.
 148          * Modify the TTL and incrementally change the checksum.
 149          */
 150         ip->ip_ttl -= IPTTLDEC;
 151         if (ip->ip_sum >= htons(0xffff - (IPTTLDEC << 8)))
 152                 ip->ip_sum += htons(IPTTLDEC << 8) + 1;
 153         else
 154                 ip->ip_sum += htons(IPTTLDEC << 8);
 155
 156         /*
 157          * Send the packet on its way.  All we can get back is ENOBUFS
 158          */
 159         ipf->ipf_uses++;
 160         ipf->ipf_timer = IPFLOW_TIMER;
 161
 162         if (rt->rt_flags & RTF_GATEWAY)
 163                 dst = rt->rt_gateway;
 164         else
 165                 dst = &ipf->ipf_ro.ro_dst;
 166         if (serializer)
 167                 lwkt_serialize_exit(serializer);
 168         lwkt_serialize_enter(rt->rt_ifp->if_serializer);
 169         if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
 170                 if (error == ENOBUFS)
 171                         ipf->ipf_dropped++;
 172                 else
 173                         ipf->ipf_errors++;
 174         }
 175         lwkt_serialize_exit(rt->rt_ifp->if_serializer);
 176         if (serializer)
 177                 lwkt_serialize_enter(serializer);
 178         return 1;
 179 }
 180
 181 static void
 182 ipflow_addstats(struct ipflow *ipf)
 183 {
 184         ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 185         ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
 186         ipstat.ips_forward += ipf->ipf_uses;
 187         ipstat.ips_fastforward += ipf->ipf_uses;
 188 }
 189
 190 static void
 191 ipflow_free(struct ipflow *ipf)
 192 {
 193         /*
 194          * Remove the flow from the hash table (at elevated IPL).
 195          * Once it's off the list, we can deal with it at normal
 196          * network IPL.
 197          */
 198         crit_enter();
 199         LIST_REMOVE(ipf, ipf_next);
 200         crit_exit();
 201         ipflow_addstats(ipf);
 202         RTFREE(ipf->ipf_ro.ro_rt);
 203         ipflow_inuse--;
 204         kfree(ipf, M_IPFLOW);
 205 }
 206
 207 static struct ipflow *
 208 ipflow_reap(void)
 209 {
 210         struct ipflow *ipf, *maybe_ipf = NULL;
 211         int idx;
 212
 213         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 214                 ipf = LIST_FIRST(&ipflows[idx]);
 215                 while (ipf != NULL) {
 216                         /*
 217                          * If this no longer points to a valid route
 218                          * reclaim it.
 219                          */
 220                         if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
 221                                 goto done;
 222                         /*
 223                          * choose the one that's been least recently used
 224                          * or has had the least uses in the last 1.5
 225                          * intervals.
 226                          */
 227                         if (maybe_ipf == NULL ||
 228                             ipf->ipf_timer < maybe_ipf->ipf_timer ||
 229                             (ipf->ipf_timer == maybe_ipf->ipf_timer &&
 230                              ipf->ipf_last_uses + ipf->ipf_uses <
 231                              maybe_ipf->ipf_last_uses + maybe_ipf->ipf_uses))
 232                                 maybe_ipf = ipf;
 233                         ipf = LIST_NEXT(ipf, ipf_next);
 234                 }
 235         }
 236         ipf = maybe_ipf;
 237 done:
 238         /*
 239          * Remove the entry from the flow table.
 240          */
 241         crit_enter();
 242         LIST_REMOVE(ipf, ipf_next);
 243         crit_exit();
 244         ipflow_addstats(ipf);
 245         RTFREE(ipf->ipf_ro.ro_rt);
 246         return ipf;
 247 }
 248
 249 void
 250 ipflow_slowtimo(void)
 251 {
 252         struct ipflow *ipf;
 253         int idx;
 254
 255         for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
 256                 ipf = LIST_FIRST(&ipflows[idx]);
 257                 while (ipf != NULL) {
 258                         struct ipflow *next_ipf = LIST_NEXT(ipf, ipf_next);
 259                         if (--ipf->ipf_timer == 0) {
 260                                 ipflow_free(ipf);
 261                         } else {
 262                                 ipf->ipf_last_uses = ipf->ipf_uses;
 263                                 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
 264                                 ipstat.ips_forward += ipf->ipf_uses;
 265                                 ipstat.ips_fastforward += ipf->ipf_uses;
 266                                 ipf->ipf_uses = 0;
 267                         }
 268                         ipf = next_ipf;
 269                 }
 270         }
 271 }
 272
 273 void
 274 ipflow_create(const struct route *ro, struct mbuf *m)
 275 {
 276         const struct ip *const ip = mtod(m, struct ip *);
 277         struct ipflow *ipf;
 278         unsigned hash;
 279
 280         /*
 281          * Don't create cache entries for ICMP messages.
 282          */
 283         if (!ipflow_active || ip->ip_p == IPPROTO_ICMP)
 284                 return;
 285         /*
 286          * See if an existing flow struct exists.  If so remove it from it's
 287          * list and free the old route.  If not, try to malloc a new one
 288          * (if we aren't at our limit).
 289          */
 290         ipf = ipflow_lookup(ip);
 291         if (ipf == NULL) {
 292                 if (ipflow_inuse == IPFLOW_MAX) {
 293                         ipf = ipflow_reap();
 294                 } else {
 295                         ipf = kmalloc(sizeof *ipf, M_IPFLOW,
 296                                      M_INTWAIT | M_NULLOK);
 297                         if (ipf == NULL)
 298                                 return;
 299                         ipflow_inuse++;
 300                 }
 301                 bzero(ipf, sizeof *ipf);
 302         } else {
 303                 crit_enter();
 304                 LIST_REMOVE(ipf, ipf_next);
 305                 crit_exit();
 306                 ipflow_addstats(ipf);
 307                 RTFREE(ipf->ipf_ro.ro_rt);
 308                 ipf->ipf_uses = ipf->ipf_last_uses = 0;
 309                 ipf->ipf_errors = ipf->ipf_dropped = 0;
 310         }
 311
 312         /*
 313          * Fill in the updated information.
 314          */
 315         ipf->ipf_ro = *ro;
 316         ro->ro_rt->rt_refcnt++;
 317         ipf->ipf_dst = ip->ip_dst;
 318         ipf->ipf_src = ip->ip_src;
 319         ipf->ipf_tos = ip->ip_tos;
 320         ipf->ipf_timer = IPFLOW_TIMER;
 321         /*
 322          * Insert into the approriate bucket of the flow table.
 323          */
 324         hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
 325         crit_enter();
 326         LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next);
 327         crit_exit();
 328 }