2 * Copyright (c) 1998 The NetBSD Foundation, Inc.
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.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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.
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 1.16 2008/10/26 07:11:28 sephe Exp $
40 #include <sys/param.h>
41 #include <sys/kernel.h>
42 #include <sys/malloc.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/sysctl.h>
47 #include <sys/thread2.h>
49 #include <machine/smp.h>
52 #include <net/route.h>
53 #include <net/netisr.h>
54 #include <net/netmsg2.h>
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>
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
67 #define IPFLOW_RTENTRY_ISDOWN(rt) \
68 (((rt)->rt_flags & RTF_UP) == 0 || ((rt)->rt_ifp->if_flags & IFF_UP) == 0)
71 LIST_ENTRY(ipflow) ipf_next; /* next ipflow in bucket */
72 struct in_addr ipf_dst; /* destination address */
73 struct in_addr ipf_src; /* source address */
75 uint8_t ipf_tos; /* type-of-service */
76 struct route ipf_ro; /* associated route entry */
77 u_long ipf_uses; /* number of uses in this period */
79 int ipf_timer; /* remaining lifetime of this entry */
80 u_long ipf_dropped; /* ENOBUFS returned by if_output */
81 u_long ipf_errors; /* other errors returned by if_output */
82 u_long ipf_last_uses; /* number of uses in last period */
85 #define ipflow_inuse ipflow_inuse_pcpu[mycpuid]
86 #define ipflows ipflows_pcpu[mycpuid]
88 static LIST_HEAD(ipflowhead, ipflow) ipflows_pcpu[MAXCPU][IPFLOW_HASHSIZE];
89 static int ipflow_inuse_pcpu[MAXCPU];
90 static struct netmsg ipflow_timo_netmsgs[MAXCPU];
91 static int ipflow_active = 0;
93 SYSCTL_NODE(_net_inet_ip, OID_AUTO, ipflow, CTLFLAG_RW, 0, "ip flow");
94 SYSCTL_INT(_net_inet_ip, IPCTL_FASTFORWARDING, fastforwarding, CTLFLAG_RW,
95 &ipflow_active, 0, "Enable flow-based IP forwarding");
97 static MALLOC_DEFINE(M_IPFLOW, "ip_flow", "IP flow");
100 ipflow_hash(struct in_addr dst, struct in_addr src, unsigned tos)
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);
110 static struct ipflow *
111 ipflow_lookup(const struct ip *ip)
116 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
119 ipf = LIST_FIRST(&ipflows[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)
125 ipf = LIST_NEXT(ipf, ipf_next);
133 ipflow_fastforward(struct mbuf *m)
138 struct sockaddr *dst;
143 * Are we forwarding packets? Big enough for an IP packet?
145 if (!ipforwarding || !ipflow_active || m->m_len < sizeof(struct ip))
149 * IP header with no option and valid version and length
151 ip = mtod(m, struct ip *);
152 if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
153 ntohs(ip->ip_len) > m->m_pkthdr.len)
159 ipf = ipflow_lookup(ip);
164 * Route and interface still up?
166 rt = ipf->ipf_ro.ro_rt;
167 if (IPFLOW_RTENTRY_ISDOWN(rt))
172 * Packet size OK? TTL?
174 if (m->m_pkthdr.len > ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
178 * Everything checks out and so we can forward this packet.
179 * Modify the TTL and incrementally change the checksum.
181 ip->ip_ttl -= IPTTLDEC;
182 if (ip->ip_sum >= htons(0xffff - (IPTTLDEC << 8)))
183 ip->ip_sum += htons(IPTTLDEC << 8) + 1;
185 ip->ip_sum += htons(IPTTLDEC << 8);
188 * Send the packet on its way. All we can get back is ENOBUFS
191 ipf->ipf_timer = IPFLOW_TIMER;
193 if (rt->rt_flags & RTF_GATEWAY)
194 dst = rt->rt_gateway;
196 dst = &ipf->ipf_ro.ro_dst;
198 error = ifp->if_output(ifp, m, dst, rt);
200 if (error == ENOBUFS)
209 ipflow_addstats(struct ipflow *ipf)
211 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
212 ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
213 ipstat.ips_forward += ipf->ipf_uses;
214 ipstat.ips_fastforward += ipf->ipf_uses;
218 ipflow_free(struct ipflow *ipf)
221 * Remove the flow from the hash table (at elevated IPL).
222 * Once it's off the list, we can deal with it at normal
226 LIST_REMOVE(ipf, ipf_next);
228 KKASSERT(ipflow_inuse > 0);
232 ipflow_addstats(ipf);
233 RTFREE(ipf->ipf_ro.ro_rt);
234 kfree(ipf, M_IPFLOW);
237 static struct ipflow *
240 struct ipflow *ipf, *maybe_ipf = NULL;
244 for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
245 ipf = LIST_FIRST(&ipflows[idx]);
246 while (ipf != NULL) {
248 * If this no longer points to a valid route
251 if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
255 * choose the one that's been least recently used
256 * or has had the least uses in the last 1.5
259 if (maybe_ipf == NULL ||
260 ipf->ipf_timer < maybe_ipf->ipf_timer ||
261 (ipf->ipf_timer == maybe_ipf->ipf_timer &&
262 ipf->ipf_last_uses + ipf->ipf_uses <
263 maybe_ipf->ipf_last_uses + maybe_ipf->ipf_uses))
265 ipf = LIST_NEXT(ipf, ipf_next);
271 * Remove the entry from the flow table.
273 LIST_REMOVE(ipf, ipf_next);
276 ipflow_addstats(ipf);
277 RTFREE(ipf->ipf_ro.ro_rt);
282 ipflow_timo_dispatch(struct netmsg *nmsg)
288 lwkt_replymsg(&nmsg->nm_lmsg, 0); /* reply ASAP */
290 for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
291 ipf = LIST_FIRST(&ipflows[idx]);
292 while (ipf != NULL) {
293 struct ipflow *next_ipf = LIST_NEXT(ipf, ipf_next);
295 if (--ipf->ipf_timer == 0) {
298 ipf->ipf_last_uses = ipf->ipf_uses;
299 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
300 ipstat.ips_forward += ipf->ipf_uses;
301 ipstat.ips_fastforward += ipf->ipf_uses;
311 ipflow_timo_ipi(void *arg __unused)
313 struct lwkt_msg *msg = &ipflow_timo_netmsgs[mycpuid].nm_lmsg;
316 if (msg->ms_flags & MSGF_DONE)
317 lwkt_sendmsg(cpu_portfn(mycpuid), msg);
322 ipflow_slowtimo(void)
325 lwkt_send_ipiq_mask(smp_active_mask, ipflow_timo_ipi, NULL);
327 ipflow_timo_ipi(NULL);
332 ipflow_create(const struct route *ro, struct mbuf *m)
334 const struct ip *const ip = mtod(m, struct ip *);
339 * Don't create cache entries for ICMP messages.
341 if (!ipflow_active || ip->ip_p == IPPROTO_ICMP)
345 * See if an existing flow struct exists. If so remove it from it's
346 * list and free the old route. If not, try to malloc a new one
347 * (if we aren't at our limit).
349 ipf = ipflow_lookup(ip);
351 if (ipflow_inuse == IPFLOW_MAX) {
354 ipf = kmalloc(sizeof(*ipf), M_IPFLOW,
355 M_INTWAIT | M_NULLOK);
360 bzero(ipf, sizeof(*ipf));
363 LIST_REMOVE(ipf, ipf_next);
366 ipflow_addstats(ipf);
367 RTFREE(ipf->ipf_ro.ro_rt);
368 ipf->ipf_uses = ipf->ipf_last_uses = 0;
369 ipf->ipf_errors = ipf->ipf_dropped = 0;
373 * Fill in the updated information.
376 ro->ro_rt->rt_refcnt++;
377 ipf->ipf_dst = ip->ip_dst;
378 ipf->ipf_src = ip->ip_src;
379 ipf->ipf_tos = ip->ip_tos;
380 ipf->ipf_timer = IPFLOW_TIMER;
383 * Insert into the approriate bucket of the flow table.
385 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
387 LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next);
397 for (i = 0; i < ncpus; ++i) {
398 netmsg_init(&ipflow_timo_netmsgs[i], &netisr_adone_rport, 0,
399 ipflow_timo_dispatch);
401 ksnprintf(oid_name, sizeof(oid_name), "inuse%d", i);
404 SYSCTL_STATIC_CHILDREN(_net_inet_ip_ipflow),
405 OID_AUTO, oid_name, CTLFLAG_RD, &ipflow_inuse_pcpu[i], 0,
406 "# of ip flow being used");
409 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, ipflow_init, 0);