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 1.27 2008/10/28 07:09:26 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 || \
70 ((rt)->rt_ifp->if_flags & IFF_UP) == 0)
73 LIST_ENTRY(ipflow) ipf_hash; /* next ipflow in hash bucket */
74 LIST_ENTRY(ipflow) ipf_list; /* next ipflow in list */
76 struct in_addr ipf_dst; /* destination address */
77 struct in_addr ipf_src; /* source address */
78 uint8_t ipf_tos; /* type-of-service */
80 uint8_t ipf_flags; /* see IPFLOW_FLAG_ */
81 uint8_t ipf_pad[2]; /* explicit pad */
82 int ipf_refcnt; /* reference count */
84 struct route ipf_ro; /* associated route entry */
85 u_long ipf_uses; /* number of uses in this period */
87 int ipf_timer; /* remaining lifetime of this entry */
88 u_long ipf_dropped; /* ENOBUFS returned by if_output */
89 u_long ipf_errors; /* other errors returned by if_output */
90 u_long ipf_last_uses; /* number of uses in last period */
92 LIST_HEAD(ipflowhead, ipflow);
94 #define IPFLOW_FLAG_ONLIST 0x1
96 #define ipflow_inuse ipflow_inuse_pcpu[mycpuid]
97 #define ipflowtable ipflowtable_pcpu[mycpuid]
98 #define ipflowlist ipflowlist_pcpu[mycpuid]
100 static struct ipflowhead ipflowtable_pcpu[MAXCPU][IPFLOW_HASHSIZE];
101 static struct ipflowhead ipflowlist_pcpu[MAXCPU];
102 static int ipflow_inuse_pcpu[MAXCPU];
103 static struct netmsg ipflow_timo_netmsgs[MAXCPU];
104 static int ipflow_active = 0;
106 #define IPFLOW_REF(ipf) \
108 KKASSERT((ipf)->ipf_refcnt > 0); \
109 (ipf)->ipf_refcnt++; \
112 #define IPFLOW_FREE(ipf) \
114 KKASSERT((ipf)->ipf_refcnt > 0); \
115 (ipf)->ipf_refcnt--; \
116 if ((ipf)->ipf_refcnt == 0) \
117 ipflow_free((ipf)); \
120 #define IPFLOW_INSERT(bucket, ipf) \
122 KKASSERT(((ipf)->ipf_flags & IPFLOW_FLAG_ONLIST) == 0); \
123 (ipf)->ipf_flags |= IPFLOW_FLAG_ONLIST; \
124 LIST_INSERT_HEAD((bucket), (ipf), ipf_hash); \
125 LIST_INSERT_HEAD(&ipflowlist, (ipf), ipf_list); \
128 #define IPFLOW_REMOVE(ipf) \
130 KKASSERT((ipf)->ipf_flags & IPFLOW_FLAG_ONLIST); \
131 (ipf)->ipf_flags &= ~IPFLOW_FLAG_ONLIST; \
132 LIST_REMOVE((ipf), ipf_hash); \
133 LIST_REMOVE((ipf), ipf_list); \
136 SYSCTL_NODE(_net_inet_ip, OID_AUTO, ipflow, CTLFLAG_RW, 0, "ip flow");
137 SYSCTL_INT(_net_inet_ip, IPCTL_FASTFORWARDING, fastforwarding, CTLFLAG_RW,
138 &ipflow_active, 0, "Enable flow-based IP forwarding");
140 static MALLOC_DEFINE(M_IPFLOW, "ip_flow", "IP flow");
142 static void ipflow_free(struct ipflow *);
145 ipflow_hash(struct in_addr dst, struct in_addr src, unsigned tos)
150 for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS)
151 hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx);
152 return hash & (IPFLOW_HASHSIZE-1);
155 static struct ipflow *
156 ipflow_lookup(const struct ip *ip)
161 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
162 LIST_FOREACH(ipf, &ipflowtable[hash], ipf_hash) {
163 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr &&
164 ip->ip_src.s_addr == ipf->ipf_src.s_addr &&
165 ip->ip_tos == ipf->ipf_tos)
172 ipflow_fastforward(struct mbuf *m)
177 struct sockaddr *dst;
182 * Are we forwarding packets?
184 if (!ipforwarding || !ipflow_active)
188 * Was packet received as a link-level multicast or broadcast?
189 * If so, don't try to fast forward..
191 if (m->m_flags & (M_BCAST | M_MCAST))
194 /* length checks already done in ip_mport() */
195 KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
196 ip = mtod(m, struct ip *);
199 * IP header with no option and valid version
201 if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2))
204 iplen = ntohs(ip->ip_len);
205 /* length checks already done in ip_mport() */
206 KASSERT(iplen >= sizeof(struct ip),
207 ("total length less then header length"));
208 KASSERT(m->m_pkthdr.len >= iplen, ("mbuf too short"));
213 ipf = ipflow_lookup(ip);
218 * Verify the IP header checksum.
220 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
221 if (!(m->m_pkthdr.csum_flags & CSUM_IP_VALID))
224 /* Must compute it ourselves. */
225 if (in_cksum_hdr(ip) != 0)
230 * Route and interface still up?
232 rt = ipf->ipf_ro.ro_rt;
233 if (IPFLOW_RTENTRY_ISDOWN(rt))
238 * Packet size OK? TTL?
240 if (m->m_pkthdr.len > ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
244 * Clear any in-bound checksum flags for this packet.
246 m->m_pkthdr.csum_flags = 0;
249 * Everything checks out and so we can forward this packet.
250 * Modify the TTL and incrementally change the checksum.
252 * This method of adding the checksum works on either endian CPU.
253 * If htons() is inlined, all the arithmetic is folded; otherwise
254 * the htons()s are combined by CSE due to the __const__ attribute.
256 * Don't bother using HW checksumming here -- the incremental
257 * update is pretty fast.
259 ip->ip_ttl -= IPTTLDEC;
260 if (ip->ip_sum >= (uint16_t)~htons(IPTTLDEC << 8))
261 ip->ip_sum -= ~htons(IPTTLDEC << 8);
263 ip->ip_sum += htons(IPTTLDEC << 8);
266 * Trim the packet in case it's too long..
268 if (m->m_pkthdr.len > iplen) {
269 if (m->m_len == m->m_pkthdr.len) {
271 m->m_pkthdr.len = iplen;
273 m_adj(m, iplen - m->m_pkthdr.len);
278 * Send the packet on its way. All we can get back is ENOBUFS
281 ipf->ipf_timer = IPFLOW_TIMER;
283 if (rt->rt_flags & RTF_GATEWAY)
284 dst = rt->rt_gateway;
286 dst = &ipf->ipf_ro.ro_dst;
289 * Reference count this ipflow, before the possible blocking
290 * ifnet.if_output(), so this ipflow will not be changed or
291 * reaped behind our back.
295 error = ifp->if_output(ifp, m, dst, rt);
297 if (error == ENOBUFS)
308 ipflow_addstats(struct ipflow *ipf)
310 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
311 ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
312 ipstat.ips_total += ipf->ipf_uses;
313 ipstat.ips_forward += ipf->ipf_uses;
314 ipstat.ips_fastforward += ipf->ipf_uses;
318 ipflow_free(struct ipflow *ipf)
320 KKASSERT(ipf->ipf_refcnt == 0);
321 KKASSERT((ipf->ipf_flags & IPFLOW_FLAG_ONLIST) == 0);
323 KKASSERT(ipflow_inuse > 0);
326 ipflow_addstats(ipf);
327 RTFREE(ipf->ipf_ro.ro_rt);
328 kfree(ipf, M_IPFLOW);
332 ipflow_reset(struct ipflow *ipf)
334 ipflow_addstats(ipf);
335 RTFREE(ipf->ipf_ro.ro_rt);
336 ipf->ipf_uses = ipf->ipf_last_uses = 0;
337 ipf->ipf_errors = ipf->ipf_dropped = 0;
340 static struct ipflow *
343 struct ipflow *ipf, *maybe_ipf = NULL;
345 LIST_FOREACH(ipf, &ipflowlist, ipf_list) {
347 * Skip actively used ipflow
349 if (ipf->ipf_refcnt > 1)
353 * If this no longer points to a valid route
356 if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
360 * choose the one that's been least recently used
361 * or has had the least uses in the last 1.5
364 if (maybe_ipf == NULL ||
365 ipf->ipf_timer < maybe_ipf->ipf_timer ||
366 (ipf->ipf_timer == maybe_ipf->ipf_timer &&
367 ipf->ipf_last_uses + ipf->ipf_uses <
368 maybe_ipf->ipf_last_uses + maybe_ipf->ipf_uses))
371 if (maybe_ipf == NULL)
377 * Remove the entry from the flow table and reset its states
385 ipflow_timo_dispatch(struct netmsg *nmsg)
387 struct ipflow *ipf, *next_ipf;
390 lwkt_replymsg(&nmsg->nm_lmsg, 0); /* reply ASAP */
393 LIST_FOREACH_MUTABLE(ipf, &ipflowlist, ipf_list, next_ipf) {
394 if (--ipf->ipf_timer == 0) {
398 ipf->ipf_last_uses = ipf->ipf_uses;
399 ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
400 ipstat.ips_total += ipf->ipf_uses;
401 ipstat.ips_forward += ipf->ipf_uses;
402 ipstat.ips_fastforward += ipf->ipf_uses;
409 ipflow_timo_ipi(void *arg __unused)
411 struct lwkt_msg *msg = &ipflow_timo_netmsgs[mycpuid].nm_lmsg;
414 if (msg->ms_flags & MSGF_DONE)
415 lwkt_sendmsg(cpu_portfn(mycpuid), msg);
420 ipflow_slowtimo(void)
426 for (i = 0; i < ncpus; ++i) {
427 if (ipflow_inuse_pcpu[i])
430 mask &= smp_active_mask;
432 lwkt_send_ipiq_mask(mask, ipflow_timo_ipi, NULL);
435 ipflow_timo_ipi(NULL);
440 ipflow_create(const struct route *ro, struct mbuf *m)
442 const struct ip *const ip = mtod(m, struct ip *);
447 * Don't create cache entries for ICMP messages.
449 if (!ipflow_active || ip->ip_p == IPPROTO_ICMP)
453 * See if an existing flow struct exists. If so remove it from it's
454 * list and free the old route. If not, try to malloc a new one
455 * (if we aren't at our limit).
457 ipf = ipflow_lookup(ip);
459 if (ipflow_inuse == IPFLOW_MAX) {
464 ipf = kmalloc(sizeof(*ipf), M_IPFLOW,
473 if (ipf->ipf_refcnt == 1) {
477 /* This ipflow is being used; don't change it */
478 KKASSERT(ipf->ipf_refcnt > 1);
482 /* This ipflow should not be actively used */
483 KKASSERT(ipf->ipf_refcnt == 1);
486 * Fill in the updated information.
489 ro->ro_rt->rt_refcnt++;
490 ipf->ipf_dst = ip->ip_dst;
491 ipf->ipf_src = ip->ip_src;
492 ipf->ipf_tos = ip->ip_tos;
493 ipf->ipf_timer = IPFLOW_TIMER;
496 * Insert into the approriate bucket of the flow table.
498 hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
499 IPFLOW_INSERT(&ipflowtable[hash], ipf);
508 for (i = 0; i < ncpus; ++i) {
509 netmsg_init(&ipflow_timo_netmsgs[i], &netisr_adone_rport,
510 MSGF_MPSAFE, ipflow_timo_dispatch);
512 ksnprintf(oid_name, sizeof(oid_name), "inuse%d", i);
515 SYSCTL_STATIC_CHILDREN(_net_inet_ip_ipflow),
516 OID_AUTO, oid_name, CTLFLAG_RD, &ipflow_inuse_pcpu[i], 0,
517 "# of ip flow being used");
520 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, ipflow_init, 0);