ip_flow.c revision 1.3
1 /* $NetBSD: ip_flow.c,v 1.3 1998/05/04 19:24:53 matt Exp $ */ 2 3 /*- 4 * Copyright (c) 1998 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by the 3am Software Foundry ("3am"). It was developed by Matt Thomas. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/malloc.h> 42 #include <sys/mbuf.h> 43 #include <sys/domain.h> 44 #include <sys/protosw.h> 45 #include <sys/socket.h> 46 #include <sys/socketvar.h> 47 #include <sys/errno.h> 48 #include <sys/time.h> 49 #include <sys/kernel.h> 50 #include <sys/proc.h> 51 52 #include <vm/vm.h> 53 #include <sys/sysctl.h> 54 55 #include <net/if.h> 56 #include <net/if_dl.h> 57 #include <net/route.h> 58 #include <net/pfil.h> 59 60 #include <netinet/in.h> 61 #include <netinet/in_systm.h> 62 #include <netinet/ip.h> 63 #include <netinet/in_pcb.h> 64 #include <netinet/in_var.h> 65 #include <netinet/ip_var.h> 66 67 #define IPFLOW_TIMER (5 * PR_SLOWHZ) 68 #define IPFLOW_HASHSIZE (1 << IPFLOW_HASHBITS) 69 static LIST_HEAD(ipflowhead, ipflow) ipflows[IPFLOW_HASHSIZE]; 70 static int ipflow_inuse; 71 #ifndef IPFLOW_MAX 72 #define IPFLOW_MAX 256 73 #endif 74 int ip_maxflows = IPFLOW_MAX; 75 76 static unsigned 77 ipflow_hash( 78 struct in_addr dst, 79 struct in_addr src, 80 unsigned tos) 81 { 82 unsigned hash = tos; 83 int idx; 84 for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS) 85 hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx); 86 return hash & (IPFLOW_HASHSIZE-1); 87 } 88 89 static struct ipflow * 90 ipflow_lookup( 91 const struct ip *ip) 92 { 93 unsigned hash; 94 struct ipflow *ipf; 95 96 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos); 97 98 ipf = LIST_FIRST(&ipflows[hash]); 99 while (ipf != NULL) { 100 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr 101 && ip->ip_src.s_addr == ipf->ipf_src.s_addr 102 && ip->ip_tos == ipf->ipf_tos) 103 break; 104 ipf = LIST_NEXT(ipf, ipf_next); 105 } 106 return ipf; 107 } 108 109 int 110 ipflow_fastforward( 111 struct mbuf *m) 112 { 113 struct ip *ip; 114 struct ipflow *ipf; 115 struct rtentry *rt; 116 int error; 117 118 /* 119 * Are we forwarding packets? Big enough for an IP packet? 120 */ 121 if (!ipforwarding || ipflow_inuse == 0 || 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 * Veryify the IP header checksum. 138 */ 139 if (in_cksum(m, sizeof(struct ip)) != 0) 140 return 0; 141 142 /* 143 * Route and interface still up? 144 */ 145 rt = ipf->ipf_ro.ro_rt; 146 if ((rt->rt_flags & RTF_UP) == 0 || (rt->rt_ifp->if_flags & IFF_UP) == 0) 147 return 0; 148 149 /* 150 * Packet size OK? TTL? 151 */ 152 if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC) 153 return 0; 154 155 /* 156 * Everything checks out and so we can forward this packet. 157 * Modify the TTL and incrementally change the checksum. 158 * On little endian machine, the TTL is in LSB position 159 * (so we can simply add) while on big-endian it's in the 160 * MSB position (so we have to do two calculation; the first 161 * is the add and second is to wrap the results into 17 bits, 162 * 16 bits and a carry). 163 */ 164 ip->ip_ttl -= IPTTLDEC; 165 if (ip->ip_sum >= htons(0xffff - (IPTTLDEC << 8))) { 166 ip->ip_sum += htons(IPTTLDEC << 8) + 1; 167 } else { 168 ip->ip_sum += htons(IPTTLDEC << 8); 169 } 170 171 /* 172 * Send the packet on it's way. All we can get back is ENOBUFS 173 */ 174 ipf->ipf_uses++; 175 ipf->ipf_timer = IPFLOW_TIMER; 176 if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, &ipf->ipf_ro.ro_dst, rt)) != 0) { 177 if (error == ENOBUFS) 178 ipf->ipf_dropped++; 179 else 180 ipf->ipf_errors++; 181 } 182 return 1; 183 } 184 185 static void 187 ipflow_addstats( 188 struct ipflow *ipf) 189 { 190 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses; 191 ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped; 192 ipstat.ips_forward += ipf->ipf_uses; 193 ipstat.ips_fastforward += ipf->ipf_uses; 194 } 195 196 static void 197 ipflow_free( 198 struct ipflow *ipf) 199 { 200 int s; 201 /* 202 * Remove the flow from the hash table (at elevated IPL). 203 * Once it's off the list, we can deal with it at normal 204 * network IPL. 205 */ 206 s = splimp(); 207 LIST_REMOVE(ipf, ipf_next); 208 splx(s); 209 ipflow_addstats(ipf); 210 RTFREE(ipf->ipf_ro.ro_rt); 211 ipflow_inuse--; 212 FREE(ipf, M_IPFLOW); 213 } 214 215 struct ipflow * 216 ipflow_reap( 217 int just_one) 218 { 219 while (just_one || ipflow_inuse > ip_maxflows) { 220 struct ipflow *ipf, *maybe_ipf = NULL; 221 int idx; 222 int s; 223 224 for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) { 225 ipf = LIST_FIRST(&ipflows[idx]); 226 while (ipf != NULL) { 227 /* 228 * If this no longer points to a valid route 229 * reclaim it. 230 */ 231 if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0) 232 goto done; 233 /* 234 * choose the one that's been least recently 235 * used or has had the least uses in the 236 * last 1.5 intervals. 237 */ 238 if (ipf == NULL 239 || ipf->ipf_timer < maybe_ipf->ipf_timer 240 || (ipf->ipf_timer == maybe_ipf->ipf_timer 241 && ipf->ipf_last_uses + ipf->ipf_uses < 242 maybe_ipf->ipf_last_uses + 243 maybe_ipf->ipf_uses)) 244 maybe_ipf = ipf; 245 ipf = LIST_NEXT(ipf, ipf_next); 246 } 247 } 248 ipf = maybe_ipf; 249 done: 250 /* 251 * Remove the entry from the flow table. 252 */ 253 s = splimp(); 254 LIST_REMOVE(ipf, ipf_next); 255 splx(s); 256 ipflow_addstats(ipf); 257 RTFREE(ipf->ipf_ro.ro_rt); 258 if (just_one) 259 return ipf; 260 FREE(ipf, M_IPFLOW); 261 ipflow_inuse--; 262 } 263 return NULL; 264 } 265 266 void 267 ipflow_slowtimo( 268 void) 269 { 270 struct ipflow *ipf; 271 int idx; 272 int inuse = ipflow_inuse; 273 274 /* 275 * Save ourselves some work if we know there aren't any flows. 276 */ 277 for (idx = 0; inuse > 0 && idx < IPFLOW_HASHSIZE; idx++) { 278 ipf = LIST_FIRST(&ipflows[idx]); 279 while (ipf != NULL) { 280 struct ipflow *next_ipf = LIST_NEXT(ipf, ipf_next); 281 if (--ipf->ipf_timer == 0) { 282 ipflow_free(ipf); 283 } else { 284 ipf->ipf_last_uses = ipf->ipf_uses; 285 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses; 286 ipstat.ips_forward += ipf->ipf_uses; 287 ipstat.ips_fastforward += ipf->ipf_uses; 288 ipf->ipf_uses = 0; 289 inuse--; 290 } 291 ipf = next_ipf; 292 } 293 } 294 } 295 296 void 297 ipflow_create( 298 const struct route *ro, 299 struct mbuf *m) 300 { 301 const struct ip *const ip = mtod(m, struct ip *); 302 struct ipflow *ipf; 303 unsigned hash; 304 int s; 305 306 /* 307 * Don't create cache entries for ICMP messages. 308 */ 309 if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP) 310 return; 311 /* 312 * See if an existing flow struct exists. If so remove it from it's 313 * list and free the old route. If not, try to malloc a new one 314 * (if we aren't at our limit). 315 */ 316 ipf = ipflow_lookup(ip); 317 if (ipf == NULL) { 318 if (ipflow_inuse >= ip_maxflows) { 319 ipf = ipflow_reap(1); 320 } else { 321 ipf = (struct ipflow *) malloc(sizeof(*ipf), M_IPFLOW, 322 M_NOWAIT); 323 if (ipf == NULL) 324 return; 325 ipflow_inuse++; 326 } 327 bzero((caddr_t) ipf, sizeof(*ipf)); 328 } else { 329 s = splimp(); 330 LIST_REMOVE(ipf, ipf_next); 331 splx(s); 332 ipflow_addstats(ipf); 333 RTFREE(ipf->ipf_ro.ro_rt); 334 ipf->ipf_uses = ipf->ipf_last_uses = 0; 335 ipf->ipf_errors = ipf->ipf_dropped = 0; 336 } 337 338 /* 339 * Fill in the updated information. 340 */ 341 ipf->ipf_ro = *ro; 342 ro->ro_rt->rt_refcnt++; 343 ipf->ipf_dst = ip->ip_dst; 344 ipf->ipf_src = ip->ip_src; 345 ipf->ipf_tos = ip->ip_tos; 346 ipf->ipf_timer = IPFLOW_TIMER; 347 ipf->ipf_start = time.tv_sec; 348 /* 349 * Insert into the approriate bucket of the flow table. 350 */ 351 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos); 352 s = splimp(); 353 LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next); 354 splx(s); 355 } 356
Indexes created Wed Oct 15 03:09:54 GMT 2025