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 1.18 2008/10/26 08:32:10 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>
48 #include <sys/in_cksum.h>
50 #include <machine/smp.h>
53 #include <net/route.h>
54 #include <net/netisr.h>
55 #include <net/netmsg2.h>
57 #include <netinet/in.h>
58 #include <netinet/ip.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip_var.h>
61 #include <netinet/ip_flow.h>
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 #define IPFLOW_MAX 256
68 #define IPFLOW_RTENTRY_ISDOWN(rt) \
69 (((rt)->rt_flags & RTF_UP) == 0 || ((rt)->rt_ifp->if_flags & IFF_UP) == 0)
72 LIST_ENTRY(ipflow) ipf_next; /* next ipflow in bucket */
73 struct in_addr ipf_dst; /* destination address */
74 struct in_addr ipf_src; /* source address */
76 uint8_t ipf_tos; /* type-of-service */
77 struct route ipf_ro; /* associated route entry */
78 u_long ipf_uses; /* number of uses in this period */
80 int ipf_timer; /* remaining lifetime of this entry */
81 u_long ipf_dropped; /* ENOBUFS returned by if_output */
82 u_long ipf_errors; /* other errors returned by if_output */
83 u_long ipf_last_uses; /* number of uses in last period */
86 #define ipflow_inuse ipflow_inuse_pcpu[mycpuid]
87 #define ipflows ipflows_pcpu[mycpuid]
89 static LIST_HEAD(ipflowhead, ipflow) ipflows_pcpu[MAXCPU][IPFLOW_HASHSIZE];
90 static int ipflow_inuse_pcpu[MAXCPU];
91 static struct netmsg ipflow_timo_netmsgs[MAXCPU];
92 static int ipflow_active = 0;
94 SYSCTL_NODE(_net_inet_ip, OID_AUTO, ipflow, CTLFLAG_RW, 0, "ip flow");
95 SYSCTL_INT(_net_inet_ip, IPCTL_FASTFORWARDING, fastforwarding, CTLFLAG_RW,
96 &ipflow_active, 0, "Enable flow-based IP forwarding");
98 static MALLOC_DEFINE(M_IPFLOW, "ip_flow", "IP flow");
101 ipflow_hash(struct in_addr dst, struct in_addr src, unsigned tos)
106 for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS)
107 hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx);
108 return hash & (IPFLOW_HASHSIZE-1);
111 static struct ipflow *
112 ipflow_lookup(const struct ip *ip)
117 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
120 ipf = LIST_FIRST(&ipflows[hash]);
121 while (ipf != NULL) {
122 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr &&
123 ip->ip_src.s_addr == ipf->ipf_src.s_addr &&
124 ip->ip_tos == ipf->ipf_tos)
126 ipf = LIST_NEXT(ipf, ipf_next);
134 ipflow_fastforward(struct mbuf *m)
139 struct sockaddr *dst;
144 * Are we forwarding packets? Big enough for an IP packet?
146 if (!ipforwarding || !ipflow_active || m->m_len < sizeof(struct ip))
150 * IP header with no option and valid version and length
152 ip = mtod(m, struct ip *);
153 if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
154 ntohs(ip->ip_len) > m->m_pkthdr.len)
160 ipf = ipflow_lookup(ip);
165 * Verify the IP header checksum.
167 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
168 if (!(m->m_pkthdr.csum_flags & CSUM_IP_VALID))
171 /* Must compute it ourselves. */
172 if (in_cksum_hdr(ip) != 0)
177 * Route and interface still up?
179 rt = ipf->ipf_ro.ro_rt;
180 if (IPFLOW_RTENTRY_ISDOWN(rt))
185 * Packet size OK? TTL?
187 if (m->m_pkthdr.len > ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
191 * Clear any in-bound checksum flags for this packet.
193 m->m_pkthdr.csum_flags = 0;
196 * Everything checks out and so we can forward this packet.
197 * Modify the TTL and incrementally change the checksum.
199 * This method of adding the checksum works on either endian CPU.
200 * If htons() is inlined, all the arithmetic is folded; otherwise
201 * the htons()s are combined by CSE due to the __const__ attribute.
203 * Don't bother using HW checksumming here -- the incremental
204 * update is pretty fast.
206 ip->ip_ttl -= IPTTLDEC;
207 if (ip->ip_sum >= (uint16_t)~htons(IPTTLDEC << 8))
208 ip->ip_sum -= ~htons(IPTTLDEC << 8);
210 ip->ip_sum += htons(IPTTLDEC << 8);
213 * Send the packet on its way. All we can get back is ENOBUFS
216 ipf->ipf_timer = IPFLOW_TIMER;
218 if (rt->rt_flags & RTF_GATEWAY)
219 dst = rt->rt_gateway;
221 dst = &ipf->ipf_ro.ro_dst;
223 error = ifp->if_output(ifp, m, dst, rt);
225 if (error == ENOBUFS)
234 ipflow_addstats(struct ipflow *ipf)
236 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
237 ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
238 ipstat.ips_forward += ipf->ipf_uses;
239 ipstat.ips_fastforward += ipf->ipf_uses;
243 ipflow_free(struct ipflow *ipf)
246 * Remove the flow from the hash table (at elevated IPL).
247 * Once it's off the list, we can deal with it at normal
251 LIST_REMOVE(ipf, ipf_next);
253 KKASSERT(ipflow_inuse > 0);
257 ipflow_addstats(ipf);
258 RTFREE(ipf->ipf_ro.ro_rt);
259 kfree(ipf, M_IPFLOW);
262 static struct ipflow *
265 struct ipflow *ipf, *maybe_ipf = NULL;
269 for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
270 ipf = LIST_FIRST(&ipflows[idx]);
271 while (ipf != NULL) {
273 * If this no longer points to a valid route
276 if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
280 * choose the one that's been least recently used
281 * or has had the least uses in the last 1.5
284 if (maybe_ipf == NULL ||
285 ipf->ipf_timer < maybe_ipf->ipf_timer ||
286 (ipf->ipf_timer == maybe_ipf->ipf_timer &&
287 ipf->ipf_last_uses + ipf->ipf_uses <
288 maybe_ipf->ipf_last_uses + maybe_ipf->ipf_uses))
290 ipf = LIST_NEXT(ipf, ipf_next);
296 * Remove the entry from the flow table.
298 LIST_REMOVE(ipf, ipf_next);
301 ipflow_addstats(ipf);
302 RTFREE(ipf->ipf_ro.ro_rt);
307 ipflow_timo_dispatch(struct netmsg *nmsg)
313 lwkt_replymsg(&nmsg->nm_lmsg, 0); /* reply ASAP */
315 for (idx = 0; idx < IPFLOW_HASHSIZE; idx++) {
316 ipf = LIST_FIRST(&ipflows[idx]);
317 while (ipf != NULL) {
318 struct ipflow *next_ipf = LIST_NEXT(ipf, ipf_next);
320 if (--ipf->ipf_timer == 0) {
323 ipf->ipf_last_uses = ipf->ipf_uses;
324 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
325 ipstat.ips_forward += ipf->ipf_uses;
326 ipstat.ips_fastforward += ipf->ipf_uses;
336 ipflow_timo_ipi(void *arg __unused)
338 struct lwkt_msg *msg = &ipflow_timo_netmsgs[mycpuid].nm_lmsg;
341 if (msg->ms_flags & MSGF_DONE)
342 lwkt_sendmsg(cpu_portfn(mycpuid), msg);
347 ipflow_slowtimo(void)
353 for (i = 0; i < ncpus; ++i) {
354 if (ipflow_inuse_pcpu[i])
358 lwkt_send_ipiq_mask(mask, ipflow_timo_ipi, NULL);
361 ipflow_timo_ipi(NULL);
366 ipflow_create(const struct route *ro, struct mbuf *m)
368 const struct ip *const ip = mtod(m, struct ip *);
373 * Don't create cache entries for ICMP messages.
375 if (!ipflow_active || ip->ip_p == IPPROTO_ICMP)
379 * See if an existing flow struct exists. If so remove it from it's
380 * list and free the old route. If not, try to malloc a new one
381 * (if we aren't at our limit).
383 ipf = ipflow_lookup(ip);
385 if (ipflow_inuse == IPFLOW_MAX) {
388 ipf = kmalloc(sizeof(*ipf), M_IPFLOW,
389 M_INTWAIT | M_NULLOK);
394 bzero(ipf, sizeof(*ipf));
397 LIST_REMOVE(ipf, ipf_next);
400 ipflow_addstats(ipf);
401 RTFREE(ipf->ipf_ro.ro_rt);
402 ipf->ipf_uses = ipf->ipf_last_uses = 0;
403 ipf->ipf_errors = ipf->ipf_dropped = 0;
407 * Fill in the updated information.
410 ro->ro_rt->rt_refcnt++;
411 ipf->ipf_dst = ip->ip_dst;
412 ipf->ipf_src = ip->ip_src;
413 ipf->ipf_tos = ip->ip_tos;
414 ipf->ipf_timer = IPFLOW_TIMER;
417 * Insert into the approriate bucket of the flow table.
419 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
421 LIST_INSERT_HEAD(&ipflows[hash], ipf, ipf_next);
431 for (i = 0; i < ncpus; ++i) {
432 netmsg_init(&ipflow_timo_netmsgs[i], &netisr_adone_rport, 0,
433 ipflow_timo_dispatch);
435 ksnprintf(oid_name, sizeof(oid_name), "inuse%d", i);
438 SYSCTL_STATIC_CHILDREN(_net_inet_ip_ipflow),
439 OID_AUTO, oid_name, CTLFLAG_RD, &ipflow_inuse_pcpu[i], 0,
440 "# of ip flow being used");
443 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, ipflow_init, 0);