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 $
39 #include <sys/param.h>
40 #include <sys/kernel.h>
41 #include <sys/malloc.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/sysctl.h>
46 #include <sys/thread2.h>
47 #include <sys/in_cksum.h>
49 #include <machine/smp.h>
52 #include <net/if_var.h>
53 #include <net/route.h>
54 #include <net/netisr2.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_TIMEOUT_FREQ 2 /* 2/second */
64 #define IPFLOW_TIMEOUT (hz / IPFLOW_TIMEOUT_FREQ)
66 #define IPFLOW_TIMER (5 * IPFLOW_TIMEOUT_FREQ)
67 #define IPFLOW_HASHBITS 6 /* should not be a multiple of 8 */
68 #define IPFLOW_HASHSIZE (1 << IPFLOW_HASHBITS)
69 #define IPFLOW_MAX 256
71 #define IPFLOW_RTENTRY_ISDOWN(rt) \
72 (((rt)->rt_flags & RTF_UP) == 0 || \
73 ((rt)->rt_ifp->if_flags & IFF_UP) == 0)
75 struct netmsg_ipfaddr
{
76 struct netmsg_base base
;
77 struct in_addr ipf_addr
;
81 LIST_ENTRY(ipflow
) ipf_hash
; /* next ipflow in hash bucket */
82 LIST_ENTRY(ipflow
) ipf_list
; /* next ipflow in list */
84 struct in_addr ipf_dst
; /* destination address */
85 struct in_addr ipf_src
; /* source address */
86 uint8_t ipf_tos
; /* type-of-service */
88 uint8_t ipf_flags
; /* see IPFLOW_FLAG_ */
89 uint8_t ipf_pad
[2]; /* explicit pad */
90 int ipf_timer
; /* remaining lifetime of this entry */
92 struct route ipf_ro
; /* associated route entry */
93 u_long ipf_uses
; /* number of uses in this period */
95 u_long ipf_dropped
; /* ENOBUFS returned by if_output */
96 u_long ipf_errors
; /* other errors returned by if_output */
97 u_long ipf_last_uses
; /* number of uses in last period */
99 LIST_HEAD(ipflowhead
, ipflow
);
101 #define IPFLOW_FLAG_ONLIST 0x1
104 struct ipflowhead ipf_table
[IPFLOW_HASHSIZE
];
105 struct ipflowhead ipf_list
;
107 struct callout ipf_timeo
;
108 struct netmsg_base ipf_timeo_netmsg
;
111 static struct ipflow_pcpu
*ipflow_pcpu_data
[MAXCPU
];
112 static int ipflow_active
= 0;
114 #define IPFLOW_INSERT(pcpu, bucket, ipf) \
116 KKASSERT(((ipf)->ipf_flags & IPFLOW_FLAG_ONLIST) == 0); \
117 (ipf)->ipf_flags |= IPFLOW_FLAG_ONLIST; \
118 LIST_INSERT_HEAD((bucket), (ipf), ipf_hash); \
119 LIST_INSERT_HEAD(&(pcpu)->ipf_list, (ipf), ipf_list); \
122 #define IPFLOW_REMOVE(ipf) \
124 KKASSERT((ipf)->ipf_flags & IPFLOW_FLAG_ONLIST); \
125 (ipf)->ipf_flags &= ~IPFLOW_FLAG_ONLIST; \
126 LIST_REMOVE((ipf), ipf_hash); \
127 LIST_REMOVE((ipf), ipf_list); \
130 SYSCTL_NODE(_net_inet_ip
, OID_AUTO
, ipflow
, CTLFLAG_RW
, 0, "ip flow");
131 SYSCTL_INT(_net_inet_ip
, IPCTL_FASTFORWARDING
, fastforwarding
, CTLFLAG_RW
,
132 &ipflow_active
, 0, "Enable flow-based IP forwarding");
134 static MALLOC_DEFINE(M_IPFLOW
, "ip_flow", "IP flow");
136 static void ipflow_free(struct ipflow_pcpu
*, struct ipflow
*);
137 static void ipflow_timeo(void *);
140 ipflow_hash(struct in_addr dst
, struct in_addr src
, unsigned tos
)
142 unsigned hash
= tos
+ src
.s_addr
;
145 for (idx
= IPFLOW_HASHBITS
; idx
< 32; idx
+= IPFLOW_HASHBITS
)
146 hash
+= (dst
.s_addr
>> (32 - idx
)) + (src
.s_addr
>> idx
);
147 return hash
& (IPFLOW_HASHSIZE
-1);
150 static struct ipflow
*
151 ipflow_lookup(struct ipflow_pcpu
*pcpu
, const struct ip
*ip
)
156 hash
= ipflow_hash(ip
->ip_dst
, ip
->ip_src
, ip
->ip_tos
);
157 LIST_FOREACH(ipf
, &pcpu
->ipf_table
[hash
], ipf_hash
) {
158 if (ip
->ip_dst
.s_addr
== ipf
->ipf_dst
.s_addr
&&
159 ip
->ip_src
.s_addr
== ipf
->ipf_src
.s_addr
&&
160 ip
->ip_tos
== ipf
->ipf_tos
)
167 ipflow_fastforward(struct mbuf
*m
)
172 struct sockaddr
*dst
;
176 ASSERT_NETISR_NCPUS(mycpuid
);
179 * Are we forwarding packets?
181 if (!ipforwarding
|| !ipflow_active
)
185 * Was packet received as a link-level multicast or broadcast?
186 * If so, don't try to fast forward..
188 if (m
->m_flags
& (M_BCAST
| M_MCAST
))
191 /* length checks already done in ip_hashfn() */
192 KASSERT(m
->m_len
>= sizeof(struct ip
), ("IP header not in one mbuf"));
193 ip
= mtod(m
, struct ip
*);
196 * IP header with no option and valid version
198 if (ip
->ip_v
!= IPVERSION
|| ip
->ip_hl
!= (sizeof(struct ip
) >> 2))
201 iplen
= ntohs(ip
->ip_len
);
202 /* length checks already done in ip_hashfn() */
203 KASSERT(iplen
>= sizeof(struct ip
),
204 ("total length less then header length"));
205 KASSERT(m
->m_pkthdr
.len
>= iplen
, ("mbuf too short"));
210 ipf
= ipflow_lookup(ipflow_pcpu_data
[mycpuid
], ip
);
215 * Verify the IP header checksum.
217 if (m
->m_pkthdr
.csum_flags
& CSUM_IP_CHECKED
) {
218 if (!(m
->m_pkthdr
.csum_flags
& CSUM_IP_VALID
))
221 /* Must compute it ourselves. */
222 if (in_cksum_hdr(ip
) != 0)
227 * Route and interface still up?
229 rt
= ipf
->ipf_ro
.ro_rt
;
230 if (IPFLOW_RTENTRY_ISDOWN(rt
))
235 * Packet size OK? TTL?
237 if (m
->m_pkthdr
.len
> ifp
->if_mtu
|| ip
->ip_ttl
<= IPTTLDEC
)
241 * Clear any in-bound checksum flags for this packet.
243 m
->m_pkthdr
.csum_flags
= 0;
246 * Everything checks out and so we can forward this packet.
247 * Modify the TTL and incrementally change the checksum.
249 * This method of adding the checksum works on either endian CPU.
250 * If htons() is inlined, all the arithmetic is folded; otherwise
251 * the htons()s are combined by CSE due to the __const__ attribute.
253 * Don't bother using HW checksumming here -- the incremental
254 * update is pretty fast.
256 ip
->ip_ttl
-= IPTTLDEC
;
257 if (ip
->ip_sum
>= (uint16_t)~htons(IPTTLDEC
<< 8))
258 ip
->ip_sum
-= ~htons(IPTTLDEC
<< 8);
260 ip
->ip_sum
+= htons(IPTTLDEC
<< 8);
263 * Trim the packet in case it's too long..
265 if (m
->m_pkthdr
.len
> iplen
) {
266 if (m
->m_len
== m
->m_pkthdr
.len
) {
268 m
->m_pkthdr
.len
= iplen
;
270 m_adj(m
, iplen
- m
->m_pkthdr
.len
);
275 * Send the packet on its way. All we can get back is ENOBUFS
278 ipf
->ipf_timer
= IPFLOW_TIMER
;
280 if (rt
->rt_flags
& RTF_GATEWAY
)
281 dst
= rt
->rt_gateway
;
283 dst
= &ipf
->ipf_ro
.ro_dst
;
285 error
= ifp
->if_output(ifp
, m
, dst
, rt
);
287 if (error
== ENOBUFS
)
296 ipflow_addstats(struct ipflow
*ipf
)
298 ipf
->ipf_ro
.ro_rt
->rt_use
+= ipf
->ipf_uses
;
299 ipstat
.ips_cantforward
+= ipf
->ipf_errors
+ ipf
->ipf_dropped
;
300 ipstat
.ips_total
+= ipf
->ipf_uses
;
301 ipstat
.ips_forward
+= ipf
->ipf_uses
;
302 ipstat
.ips_fastforward
+= ipf
->ipf_uses
;
306 ipflow_free(struct ipflow_pcpu
*pcpu
, struct ipflow
*ipf
)
308 KKASSERT((ipf
->ipf_flags
& IPFLOW_FLAG_ONLIST
) == 0);
310 KKASSERT(pcpu
->ipf_inuse
> 0);
313 ipflow_addstats(ipf
);
314 RTFREE(ipf
->ipf_ro
.ro_rt
);
315 kfree(ipf
, M_IPFLOW
);
319 ipflow_reset(struct ipflow
*ipf
)
321 ipflow_addstats(ipf
);
322 RTFREE(ipf
->ipf_ro
.ro_rt
);
323 ipf
->ipf_uses
= ipf
->ipf_last_uses
= 0;
324 ipf
->ipf_errors
= ipf
->ipf_dropped
= 0;
327 static struct ipflow
*
328 ipflow_reap(struct ipflow_pcpu
*pcpu
)
330 struct ipflow
*ipf
, *maybe_ipf
= NULL
;
332 LIST_FOREACH(ipf
, &pcpu
->ipf_list
, ipf_list
) {
334 * If this no longer points to a valid route
337 if ((ipf
->ipf_ro
.ro_rt
->rt_flags
& RTF_UP
) == 0)
341 * choose the one that's been least recently used
342 * or has had the least uses in the last 1.5
345 if (maybe_ipf
== NULL
||
346 ipf
->ipf_timer
< maybe_ipf
->ipf_timer
||
347 (ipf
->ipf_timer
== maybe_ipf
->ipf_timer
&&
348 ipf
->ipf_last_uses
+ ipf
->ipf_uses
<
349 maybe_ipf
->ipf_last_uses
+ maybe_ipf
->ipf_uses
))
352 if (maybe_ipf
== NULL
)
358 * Remove the entry from the flow table and reset its states
366 ipflow_timeo_dispatch(netmsg_t nmsg
)
368 struct ipflow
*ipf
, *next_ipf
;
369 struct ipflow_pcpu
*pcpu
= ipflow_pcpu_data
[mycpuid
];
371 ASSERT_NETISR_NCPUS(mycpuid
);
374 netisr_replymsg(&nmsg
->base
, 0); /* reply ASAP */
377 LIST_FOREACH_MUTABLE(ipf
, &pcpu
->ipf_list
, ipf_list
, next_ipf
) {
378 if (--ipf
->ipf_timer
== 0) {
380 ipflow_free(pcpu
, ipf
);
382 ipf
->ipf_last_uses
= ipf
->ipf_uses
;
383 ipf
->ipf_ro
.ro_rt
->rt_use
+= ipf
->ipf_uses
;
384 ipstat
.ips_total
+= ipf
->ipf_uses
;
385 ipstat
.ips_forward
+= ipf
->ipf_uses
;
386 ipstat
.ips_fastforward
+= ipf
->ipf_uses
;
390 callout_reset(&pcpu
->ipf_timeo
, IPFLOW_TIMEOUT
, ipflow_timeo
, pcpu
);
394 ipflow_timeo(void *xpcpu
)
396 struct ipflow_pcpu
*pcpu
= xpcpu
;
397 struct netmsg_base
*nm
= &pcpu
->ipf_timeo_netmsg
;
400 if (nm
->lmsg
.ms_flags
& MSGF_DONE
)
401 netisr_sendmsg_oncpu(nm
);
406 ipflow_create(const struct route
*ro
, struct mbuf
*m
)
408 struct ipflow_pcpu
*pcpu
= ipflow_pcpu_data
[mycpuid
];
409 const struct ip
*const ip
= mtod(m
, struct ip
*);
413 ASSERT_NETISR_NCPUS(mycpuid
);
416 * Don't create cache entries for ICMP messages.
418 if (!ipflow_active
|| ip
->ip_p
== IPPROTO_ICMP
)
422 * See if an existing flow struct exists. If so remove it from it's
423 * list and free the old route. If not, try to malloc a new one
424 * (if we aren't at our limit).
426 ipf
= ipflow_lookup(pcpu
, ip
);
428 if (pcpu
->ipf_inuse
== IPFLOW_MAX
) {
429 ipf
= ipflow_reap(pcpu
);
433 ipf
= kmalloc(sizeof(*ipf
), M_IPFLOW
,
434 M_INTWAIT
| M_NULLOK
| M_ZERO
);
445 * Fill in the updated information.
448 ro
->ro_rt
->rt_refcnt
++;
449 ipf
->ipf_dst
= ip
->ip_dst
;
450 ipf
->ipf_src
= ip
->ip_src
;
451 ipf
->ipf_tos
= ip
->ip_tos
;
452 ipf
->ipf_timer
= IPFLOW_TIMER
;
455 * Insert into the approriate bucket of the flow table.
457 hash
= ipflow_hash(ip
->ip_dst
, ip
->ip_src
, ip
->ip_tos
);
458 IPFLOW_INSERT(pcpu
, &pcpu
->ipf_table
[hash
], ipf
);
462 ipflow_flush_oncpu(void)
464 struct ipflow_pcpu
*pcpu
= ipflow_pcpu_data
[mycpuid
];
467 ASSERT_NETISR_NCPUS(mycpuid
);
469 while ((ipf
= LIST_FIRST(&pcpu
->ipf_list
)) != NULL
) {
471 ipflow_free(pcpu
, ipf
);
476 ipflow_ifaddr_handler(netmsg_t nmsg
)
478 struct netmsg_ipfaddr
*amsg
= (struct netmsg_ipfaddr
*)nmsg
;
479 struct ipflow_pcpu
*pcpu
= ipflow_pcpu_data
[mycpuid
];
480 struct ipflow
*ipf
, *next_ipf
;
482 LIST_FOREACH_MUTABLE(ipf
, &pcpu
->ipf_list
, ipf_list
, next_ipf
) {
483 if (ipf
->ipf_dst
.s_addr
== amsg
->ipf_addr
.s_addr
||
484 ipf
->ipf_src
.s_addr
== amsg
->ipf_addr
.s_addr
) {
486 ipflow_free(pcpu
, ipf
);
489 netisr_forwardmsg(&nmsg
->base
, mycpuid
+ 1);
493 ipflow_ifaddr(void *arg __unused
, struct ifnet
*ifp __unused
,
494 enum ifaddr_event event
, struct ifaddr
*ifa
)
496 struct netmsg_ipfaddr amsg
;
498 if (ifa
->ifa_addr
->sa_family
!= AF_INET
)
501 /* Only add/change events need to be handled */
503 case IFADDR_EVENT_ADD
:
504 case IFADDR_EVENT_CHANGE
:
507 case IFADDR_EVENT_DELETE
:
511 netmsg_init(&amsg
.base
, NULL
, &curthread
->td_msgport
,
512 MSGF_PRIORITY
, ipflow_ifaddr_handler
);
513 amsg
.ipf_addr
= ifatoia(ifa
)->ia_addr
.sin_addr
;
515 netisr_domsg_global(&amsg
.base
);
519 ipflow_init_dispatch(netmsg_t nm
)
521 struct ipflow_pcpu
*pcpu
;
525 pcpu
= kmalloc(sizeof(*pcpu
), M_IPFLOW
, M_WAITOK
| M_ZERO
);
527 netmsg_init(&pcpu
->ipf_timeo_netmsg
, NULL
, &netisr_adone_rport
,
528 MSGF_PRIORITY
, ipflow_timeo_dispatch
);
529 callout_init_mp(&pcpu
->ipf_timeo
);
531 ksnprintf(oid_name
, sizeof(oid_name
), "inuse%d", cpuid
);
532 SYSCTL_ADD_INT(NULL
, SYSCTL_STATIC_CHILDREN(_net_inet_ip_ipflow
),
533 OID_AUTO
, oid_name
, CTLFLAG_RD
, &pcpu
->ipf_inuse
, 0,
534 "# of ip flow being used");
536 ipflow_pcpu_data
[cpuid
] = pcpu
;
538 callout_reset(&pcpu
->ipf_timeo
, IPFLOW_TIMEOUT
, ipflow_timeo
, pcpu
);
540 netisr_forwardmsg(&nm
->base
, cpuid
+ 1);
546 struct netmsg_base nm
;
548 netmsg_init(&nm
, NULL
, &curthread
->td_msgport
, 0,
549 ipflow_init_dispatch
);
550 netisr_domsg_global(&nm
);
552 EVENTHANDLER_REGISTER(ifaddr_event
, ipflow_ifaddr
, NULL
,
553 EVENTHANDLER_PRI_ANY
);
555 SYSINIT(arp
, SI_SUB_PROTO_DOMAIN
, SI_ORDER_ANY
, ipflow_init
, 0);