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 1.15 2008/07/27 10:06:57 sephe Exp $ 38 */ 39 40 #include <sys/param.h> 41 #include <sys/kernel.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/protosw.h> 45 #include <sys/socket.h> 46 #include <sys/sysctl.h> 47 #include <sys/thread2.h> 48 49 #include <machine/smp.h> 50 51 #include <net/if.h> 52 #include <net/route.h> 53 #include <net/netisr.h> 54 #include <net/netmsg2.h> 55 56 #include <netinet/in.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 #define IPFLOW_MAX 256 66 67 #define IPFLOW_RTENTRY_ISDOWN(rt) \ 68 (((rt)->rt_flags & RTF_UP) == 0 || ((rt)->rt_ifp->if_flags & IFF_UP) == 0) 69 70 #define ipflow_inuse ipflow_inuse_pcpu[mycpuid] 71 #define ipflows ipflows_pcpu[mycpuid] 72 73 static LIST_HEAD(ipflowhead, ipflow) ipflows_pcpu[MAXCPU][IPFLOW_HASHSIZE]; 74 static int ipflow_inuse_pcpu[MAXCPU]; 75 static struct netmsg ipflow_timo_netmsgs[MAXCPU]; 76 static int ipflow_active = 0; 77 78 SYSCTL_NODE(_net_inet_ip, OID_AUTO, ipflow, CTLFLAG_RW, 0, "ip flow"); 79 SYSCTL_INT(_net_inet_ip, IPCTL_FASTFORWARDING, fastforwarding, CTLFLAG_RW, 80 &ipflow_active, 0, "Enable flow-based IP forwarding"); 81 82 static MALLOC_DEFINE(M_IPFLOW, "ip_flow", "IP flow"); 83 84 static unsigned 85 ipflow_hash(struct in_addr dst, struct in_addr src, unsigned tos) 86 { 87 unsigned hash = tos; 88 int idx; 89 90 for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS) 91 hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx); 92 return hash & (IPFLOW_HASHSIZE-1); 93 } 94 95 static struct ipflow * 96 ipflow_lookup(const struct ip *ip) 97 { 98 unsigned hash; 99 struct ipflow *ipf; 100 101 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos); 102 103 crit_enter(); 104 ipf = LIST_FIRST(&ipflows[hash]); 105 while (ipf != NULL) { 106 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr && 107 ip->ip_src.s_addr == ipf->ipf_src.s_addr && 108 ip->ip_tos == ipf->ipf_tos) 109 break; 110 ipf = LIST_NEXT(ipf, ipf_next); 111 } 112 crit_exit(); 113 114 return ipf; 115 } 116 117 int 118 ipflow_fastforward(struct mbuf *m) 119 { 120 struct ip *ip; 121 struct ipflow *ipf; 122 struct rtentry *rt; 123 struct sockaddr *dst; 124 struct ifnet *ifp; 125 int error; 126 127 /* 128 * Are we forwarding packets? Big enough for an IP packet? 129 */ 130 if (!ipforwarding || !ipflow_active || m->m_len < sizeof(struct ip)) 131 return 0; 132 133 /* 134 * IP header with no option and valid version and length 135 */ 136 ip = mtod(m, struct ip *); 137 if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) || 138 ntohs(ip->ip_len) > m->m_pkthdr.len) 139 return 0; 140 141 /* 142 * Find a flow. 143 */ 144 ipf = ipflow_lookup(ip); 145 if (ipf == NULL) 146 return 0; 147 148 /* 149 * Route and interface still up? 150 */ 151 rt = ipf->ipf_ro.ro_rt; 152 if (IPFLOW_RTENTRY_ISDOWN(rt)) 153 return 0; 154 ifp = rt->rt_ifp; 155 156 /* 157 * Packet size OK? TTL? 158 */ 159 if (m->m_pkthdr.len > ifp->if_mtu || ip->ip_ttl <= IPTTLDEC) 160 return 0; 161 162 /* 163 * Everything checks out and so we can forward this packet. 164 * Modify the TTL and incrementally change the checksum. 165 */ 166 ip->ip_ttl -= IPTTLDEC; 167 if (ip->ip_sum >= htons(0xffff - (IPTTLDEC << 8))) 168 ip->ip_sum += htons(IPTTLDEC << 8) + 1; 169 else 170 ip->ip_sum += htons(IPTTLDEC << 8); 171 172 /* 173 * Send the packet on its way. All we can get back is ENOBUFS 174 */ 175 ipf->ipf_uses++; 176 ipf->ipf_timer = IPFLOW_TIMER; 177 178 if (rt->rt_flags & RTF_GATEWAY) 179 dst = rt->rt_gateway; 180 else 181 dst = &ipf->ipf_ro.ro_dst; 182 183 error = ifp->if_output(ifp, m, dst, rt); 184 if (error) { 185 if (error == ENOBUFS) 186 ipf->ipf_dropped++; 187 else 188 ipf->ipf_errors++; 189 } 190 return 1; 191 } 192 193 static void 194 ipflow_addstats(struct ipflow *ipf) 195 { 196 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses; 197 ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped; 198 ipstat.ips_forward += ipf->ipf_uses; 199 ipstat.ips_fastforward += ipf->ipf_uses; 200 } 201 202 static void 203 ipflow_free(struct ipflow *ipf) 204 { 205 /* 206 * Remove the flow from the hash table (at elevated IPL). 207 * Once it's off the list, we can deal with it at normal 208 * network IPL. 209 */ 210 crit_enter(); 211 LIST_REMOVE(ipf, ipf_next); 212 213 KKASSERT(ipflow_inuse > 0); 214 ipflow_inuse--; 215 crit_exit(); 216 217 ipflow_addstats(ipf); 218 RTFREE(ipf->ipf_ro.ro_rt); 219 kfree(ipf, M_IPFLOW); 220 } 221 222 static struct ipflow * 223 ipflow_reap(void) 224 { 225 struct ipflow *ipf, *maybe_ipf = NULL; 226 int idx; 227 228 crit_enter(); 229 for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) { 230 ipf = LIST_FIRST(&ipflows[idx]); 231 while (ipf != NULL) { 232 /* 233 * If this no longer points to a valid route 234 * reclaim it. 235 */ 236 if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0) 237 goto done; 238 239 /* 240 * choose the one that's been least recently used 241 * or has had the least uses in the last 1.5 242 * intervals. 243 */ 244 if (maybe_ipf == NULL || 245 ipf->ipf_timer < maybe_ipf->ipf_timer || 246 (ipf->ipf_timer == maybe_ipf->ipf_timer && 247 ipf->ipf_last_uses + ipf->ipf_uses < 248 maybe_ipf->ipf_last_uses + maybe_ipf->ipf_uses)) 249 maybe_ipf = ipf; 250 ipf = LIST_NEXT(ipf, ipf_next); 251 } 252 } 253 ipf = maybe_ipf; 254 done: 255 /* 256 * Remove the entry from the flow table. 257 */ 258 LIST_REMOVE(ipf, ipf_next); 259 crit_exit(); 260 261 ipflow_addstats(ipf); 262 RTFREE(ipf->ipf_ro.ro_rt); 263 return ipf; 264 } 265 266 static void 267 ipflow_timo_dispatch(struct netmsg *nmsg) 268 { 269 struct ipflow *ipf; 270 int idx; 271 272 crit_enter(); 273 lwkt_replymsg(&nmsg->nm_lmsg, 0); /* reply ASAP */ 274 275 for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) { 276 ipf = LIST_FIRST(&ipflows[idx]); 277 while (ipf != NULL) { 278 struct ipflow *next_ipf = LIST_NEXT(ipf, ipf_next); 279 280 if (--ipf->ipf_timer == 0) { 281 ipflow_free(ipf); 282 } else { 283 ipf->ipf_last_uses = ipf->ipf_uses; 284 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses; 285 ipstat.ips_forward += ipf->ipf_uses; 286 ipstat.ips_fastforward += ipf->ipf_uses; 287 ipf->ipf_uses = 0; 288 } 289 ipf = next_ipf; 290 } 291 } 292 crit_exit(); 293 } 294 295 static void 296 ipflow_timo_ipi(void *arg __unused) 297 { 298 struct lwkt_msg *msg = &ipflow_timo_netmsgs[mycpuid].nm_lmsg; 299 300 crit_enter(); 301 if (msg->ms_flags & MSGF_DONE) 302 lwkt_sendmsg(cpu_portfn(mycpuid), msg); 303 crit_exit(); 304 } 305 306 void 307 ipflow_slowtimo(void) 308 { 309 #ifdef SMP 310 lwkt_send_ipiq_mask(smp_active_mask, ipflow_timo_ipi, NULL); 311 #else 312 ipflow_timo_ipi(NULL); 313 #endif 314 } 315 316 void 317 ipflow_create(const struct route *ro, 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 /* 330 * See if an existing flow struct exists. If so remove it from it's 331 * list and free the old route. If not, try to malloc a new one 332 * (if we aren't at our limit). 333 */ 334 ipf = ipflow_lookup(ip); 335 if (ipf == NULL) { 336 if (ipflow_inuse == IPFLOW_MAX) { 337 ipf = ipflow_reap(); 338 } else { 339 ipf = kmalloc(sizeof(*ipf), M_IPFLOW, 340 M_INTWAIT | M_NULLOK); 341 if (ipf == NULL) 342 return; 343 ipflow_inuse++; 344 } 345 bzero(ipf, sizeof(*ipf)); 346 } else { 347 crit_enter(); 348 LIST_REMOVE(ipf, ipf_next); 349 crit_exit(); 350 351 ipflow_addstats(ipf); 352 RTFREE(ipf->ipf_ro.ro_rt); 353 ipf->ipf_uses = ipf->ipf_last_uses = 0; 354 ipf->ipf_errors = ipf->ipf_dropped = 0; 355 } 356 357 /* 358 * Fill in the updated information. 359 */ 360 ipf->ipf_ro = *ro; 361 ro->ro_rt->rt_refcnt++; 362 ipf->ipf_dst = ip->ip_dst; 363 ipf->ipf_src = ip->ip_src; 364 ipf->ipf_tos = ip->ip_tos; 365 ipf->ipf_timer = IPFLOW_TIMER; 366 367 /* 368 * Insert into the approriate bucket of the flow table. 369 */ 370 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos); 371 crit_enter(); 372 LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next); 373 crit_exit(); 374 } 375 376 static void 377 ipflow_init(void) 378 { 379 char oid_name[32]; 380 int i; 381 382 for (i = 0; i < ncpus; ++i) { 383 netmsg_init(&ipflow_timo_netmsgs[i], &netisr_adone_rport, 0, 384 ipflow_timo_dispatch); 385 386 ksnprintf(oid_name, sizeof(oid_name), "inuse%d", i); 387 388 SYSCTL_ADD_INT(NULL, 389 SYSCTL_STATIC_CHILDREN(_net_inet_ip_ipflow), 390 OID_AUTO, oid_name, CTLFLAG_RD, &ipflow_inuse_pcpu[i], 0, 391 "# of ip flow being used"); 392 } 393 } 394 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, ipflow_init, 0); 395