1 /* $NetBSD: ip_flow.c,v 1.16 2000/06/30 19:43:53 thorpej 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 #include <sys/pool.h> 52 53 #include <uvm/uvm_extern.h> 54 55 #include <sys/sysctl.h> 56 57 #include <net/if.h> 58 #include <net/if_dl.h> 59 #include <net/route.h> 60 #include <net/pfil.h> 61 62 #include <netinet/in.h> 63 #include <netinet/in_systm.h> 64 #include <netinet/ip.h> 65 #include <netinet/in_pcb.h> 66 #include <netinet/in_var.h> 67 #include <netinet/ip_var.h> 68 69 struct pool ipflow_pool; 70 71 LIST_HEAD(ipflowhead, ipflow); 72 73 #define IPFLOW_TIMER (5 * PR_SLOWHZ) 74 #define IPFLOW_HASHSIZE (1 << IPFLOW_HASHBITS) 75 76 static struct ipflowhead ipflowtable[IPFLOW_HASHSIZE]; 77 static struct ipflowhead ipflowlist; 78 static int ipflow_inuse; 79 80 #define IPFLOW_INSERT(bucket, ipf) \ 81 do { \ 82 LIST_INSERT_HEAD((bucket), (ipf), ipf_hash); \ 83 LIST_INSERT_HEAD(&ipflowlist, (ipf), ipf_list); \ 84 } while (0) 85 86 #define IPFLOW_REMOVE(ipf) \ 87 do { \ 88 LIST_REMOVE((ipf), ipf_hash); \ 89 LIST_REMOVE((ipf), ipf_list); \ 90 } while (0) 91 92 #ifndef IPFLOW_MAX 93 #define IPFLOW_MAX 256 94 #endif 95 int ip_maxflows = IPFLOW_MAX; 96 97 static unsigned 98 ipflow_hash( 99 struct in_addr dst, 100 struct in_addr src, 101 unsigned tos) 102 { 103 unsigned hash = tos; 104 int idx; 105 for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS) 106 hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx); 107 return hash & (IPFLOW_HASHSIZE-1); 108 } 109 110 static struct ipflow * 111 ipflow_lookup( 112 const struct ip *ip) 113 { 114 unsigned hash; 115 struct ipflow *ipf; 116 117 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos); 118 119 ipf = LIST_FIRST(&ipflowtable[hash]); 120 while (ipf != NULL) { 121 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr 122 && ip->ip_src.s_addr == ipf->ipf_src.s_addr 123 && ip->ip_tos == ipf->ipf_tos) 124 break; 125 ipf = LIST_NEXT(ipf, ipf_hash); 126 } 127 return ipf; 128 } 129 130 void 131 ipflow_init() 132 { 133 int i; 134 135 pool_init(&ipflow_pool, sizeof(struct ipflow), 0, 0, 0, "ipflowpl", 136 0, NULL, NULL, M_IPFLOW); 137 138 LIST_INIT(&ipflowlist); 139 for (i = 0; i < IPFLOW_HASHSIZE; i++) 140 LIST_INIT(&ipflowtable[i]); 141 } 142 143 int 144 ipflow_fastforward( 145 struct mbuf *m) 146 { 147 struct ip *ip; 148 struct ipflow *ipf; 149 struct rtentry *rt; 150 struct sockaddr *dst; 151 int error; 152 int iplen; 153 154 /* 155 * Are we forwarding packets? Big enough for an IP packet? 156 */ 157 if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip)) 158 return 0; 159 160 /* 161 * Was packet recieved as a link-level multicast or broadcast? 162 * If so, don't try to fast forward.. 163 */ 164 if ((m->m_flags & (M_BCAST|M_MCAST)) != 0) 165 return 0; 166 167 /* 168 * IP header with no option and valid version and length 169 */ 170 ip = mtod(m, struct ip *); 171 iplen = ntohs(ip->ip_len); 172 if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) || 173 iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len) 174 return 0; 175 /* 176 * Find a flow. 177 */ 178 if ((ipf = ipflow_lookup(ip)) == NULL) 179 return 0; 180 181 /* 182 * Veryify the IP header checksum. 183 */ 184 if (in_cksum(m, sizeof(struct ip)) != 0) 185 return 0; 186 187 /* 188 * Route and interface still up? 189 */ 190 rt = ipf->ipf_ro.ro_rt; 191 if ((rt->rt_flags & RTF_UP) == 0 || 192 (rt->rt_ifp->if_flags & IFF_UP) == 0) 193 return 0; 194 195 /* 196 * Packet size OK? TTL? 197 */ 198 if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC) 199 return 0; 200 201 /* 202 * Everything checks out and so we can forward this packet. 203 * Modify the TTL and incrementally change the checksum. 204 * 205 * This method of adding the checksum works on either endian CPU. 206 * If htons() is inlined, all the arithmetic is folded; otherwise 207 * the htons()s are combined by CSE due to the __const__ attribute. 208 */ 209 ip->ip_ttl -= IPTTLDEC; 210 if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8)) 211 ip->ip_sum -= ~htons(IPTTLDEC << 8); 212 else 213 ip->ip_sum += htons(IPTTLDEC << 8); 214 215 /* 216 * Trim the packet in case it's too long.. 217 */ 218 if (m->m_pkthdr.len > iplen) { 219 if (m->m_len == m->m_pkthdr.len) { 220 m->m_len = iplen; 221 m->m_pkthdr.len = iplen; 222 } else 223 m_adj(m, iplen - m->m_pkthdr.len); 224 } 225 226 /* 227 * Send the packet on it's way. All we can get back is ENOBUFS 228 */ 229 ipf->ipf_uses++; 230 PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER); 231 232 if (rt->rt_flags & RTF_GATEWAY) 233 dst = rt->rt_gateway; 234 else 235 dst = &ipf->ipf_ro.ro_dst; 236 237 if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) { 238 if (error == ENOBUFS) 239 ipf->ipf_dropped++; 240 else 241 ipf->ipf_errors++; 242 } 243 return 1; 244 } 245 246 static void 247 ipflow_addstats( 248 struct ipflow *ipf) 249 { 250 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses; 251 ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped; 252 ipstat.ips_forward += ipf->ipf_uses; 253 ipstat.ips_fastforward += ipf->ipf_uses; 254 } 255 256 static void 257 ipflow_free( 258 struct ipflow *ipf) 259 { 260 int s; 261 /* 262 * Remove the flow from the hash table (at elevated IPL). 263 * Once it's off the list, we can deal with it at normal 264 * network IPL. 265 */ 266 s = splimp(); 267 IPFLOW_REMOVE(ipf); 268 splx(s); 269 ipflow_addstats(ipf); 270 RTFREE(ipf->ipf_ro.ro_rt); 271 ipflow_inuse--; 272 pool_put(&ipflow_pool, ipf); 273 } 274 275 struct ipflow * 276 ipflow_reap( 277 int just_one) 278 { 279 while (just_one || ipflow_inuse > ip_maxflows) { 280 struct ipflow *ipf, *maybe_ipf = NULL; 281 int s; 282 283 ipf = LIST_FIRST(&ipflowlist); 284 while (ipf != NULL) { 285 /* 286 * If this no longer points to a valid route 287 * reclaim it. 288 */ 289 if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0) 290 goto done; 291 /* 292 * choose the one that's been least recently 293 * used or has had the least uses in the 294 * last 1.5 intervals. 295 */ 296 if (maybe_ipf == NULL || 297 ipf->ipf_timer < maybe_ipf->ipf_timer || 298 (ipf->ipf_timer == maybe_ipf->ipf_timer && 299 ipf->ipf_last_uses + ipf->ipf_uses < 300 maybe_ipf->ipf_last_uses + 301 maybe_ipf->ipf_uses)) 302 maybe_ipf = ipf; 303 ipf = LIST_NEXT(ipf, ipf_list); 304 } 305 ipf = maybe_ipf; 306 done: 307 /* 308 * Remove the entry from the flow table. 309 */ 310 s = splimp(); 311 IPFLOW_REMOVE(ipf); 312 splx(s); 313 ipflow_addstats(ipf); 314 RTFREE(ipf->ipf_ro.ro_rt); 315 if (just_one) 316 return ipf; 317 pool_put(&ipflow_pool, ipf); 318 ipflow_inuse--; 319 } 320 return NULL; 321 } 322 323 void 324 ipflow_slowtimo( 325 void) 326 { 327 struct ipflow *ipf, *next_ipf; 328 329 ipf = LIST_FIRST(&ipflowlist); 330 while (ipf != NULL) { 331 next_ipf = LIST_NEXT(ipf, ipf_list); 332 if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer)) { 333 ipflow_free(ipf); 334 } else { 335 ipf->ipf_last_uses = ipf->ipf_uses; 336 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses; 337 ipstat.ips_forward += ipf->ipf_uses; 338 ipstat.ips_fastforward += ipf->ipf_uses; 339 ipf->ipf_uses = 0; 340 } 341 ipf = next_ipf; 342 } 343 } 344 345 void 346 ipflow_create( 347 const struct route *ro, 348 struct mbuf *m) 349 { 350 const struct ip *const ip = mtod(m, struct ip *); 351 struct ipflow *ipf; 352 unsigned hash; 353 int s; 354 355 /* 356 * Don't create cache entries for ICMP messages. 357 */ 358 if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP) 359 return; 360 /* 361 * See if an existing flow struct exists. If so remove it from it's 362 * list and free the old route. If not, try to malloc a new one 363 * (if we aren't at our limit). 364 */ 365 ipf = ipflow_lookup(ip); 366 if (ipf == NULL) { 367 if (ipflow_inuse >= ip_maxflows) { 368 ipf = ipflow_reap(1); 369 } else { 370 ipf = pool_get(&ipflow_pool, PR_NOWAIT); 371 if (ipf == NULL) 372 return; 373 ipflow_inuse++; 374 } 375 bzero((caddr_t) ipf, sizeof(*ipf)); 376 } else { 377 s = splimp(); 378 IPFLOW_REMOVE(ipf); 379 splx(s); 380 ipflow_addstats(ipf); 381 RTFREE(ipf->ipf_ro.ro_rt); 382 ipf->ipf_uses = ipf->ipf_last_uses = 0; 383 ipf->ipf_errors = ipf->ipf_dropped = 0; 384 } 385 386 /* 387 * Fill in the updated information. 388 */ 389 ipf->ipf_ro = *ro; 390 ro->ro_rt->rt_refcnt++; 391 ipf->ipf_dst = ip->ip_dst; 392 ipf->ipf_src = ip->ip_src; 393 ipf->ipf_tos = ip->ip_tos; 394 PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER); 395 ipf->ipf_start = time.tv_sec; 396 /* 397 * Insert into the approriate bucket of the flow table. 398 */ 399 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos); 400 s = splimp(); 401 IPFLOW_INSERT(&ipflowtable[hash], ipf); 402 splx(s); 403 } 404