xref: /netbsd-src/sys/netinet/ip_flow.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: ip_flow.c,v 1.28 2004/04/25 16:42:42 simonb Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by the 3am Software Foundry ("3am").  It was developed by Matt Thomas.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: ip_flow.c,v 1.28 2004/04/25 16:42:42 simonb Exp $");
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/domain.h>
47 #include <sys/protosw.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/errno.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/pool.h>
54 #include <sys/sysctl.h>
55 
56 #include <net/if.h>
57 #include <net/if_dl.h>
58 #include <net/route.h>
59 #include <net/pfil.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/ip.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip_var.h>
67 
68 POOL_INIT(ipflow_pool, sizeof(struct ipflow), 0, 0, 0, "ipflowpl", NULL);
69 
70 LIST_HEAD(ipflowhead, ipflow);
71 
72 #define	IPFLOW_TIMER		(5 * PR_SLOWHZ)
73 #define	IPFLOW_HASHSIZE		(1 << IPFLOW_HASHBITS)
74 
75 static struct ipflowhead ipflowtable[IPFLOW_HASHSIZE];
76 static struct ipflowhead ipflowlist;
77 static int ipflow_inuse;
78 
79 #define	IPFLOW_INSERT(bucket, ipf) \
80 do { \
81 	LIST_INSERT_HEAD((bucket), (ipf), ipf_hash); \
82 	LIST_INSERT_HEAD(&ipflowlist, (ipf), ipf_list); \
83 } while (/*CONSTCOND*/ 0)
84 
85 #define	IPFLOW_REMOVE(ipf) \
86 do { \
87 	LIST_REMOVE((ipf), ipf_hash); \
88 	LIST_REMOVE((ipf), ipf_list); \
89 } while (/*CONSTCOND*/ 0)
90 
91 #ifndef IPFLOW_MAX
92 #define	IPFLOW_MAX		256
93 #endif
94 int ip_maxflows = IPFLOW_MAX;
95 
96 static unsigned
97 ipflow_hash(
98 	struct in_addr dst,
99 	struct in_addr src,
100 	unsigned tos)
101 {
102 	unsigned hash = tos;
103 	int idx;
104 	for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS)
105 		hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx);
106 	return hash & (IPFLOW_HASHSIZE-1);
107 }
108 
109 static struct ipflow *
110 ipflow_lookup(
111 	const struct ip *ip)
112 {
113 	unsigned hash;
114 	struct ipflow *ipf;
115 
116 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
117 
118 	ipf = LIST_FIRST(&ipflowtable[hash]);
119 	while (ipf != NULL) {
120 		if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
121 		    && ip->ip_src.s_addr == ipf->ipf_src.s_addr
122 		    && ip->ip_tos == ipf->ipf_tos)
123 			break;
124 		ipf = LIST_NEXT(ipf, ipf_hash);
125 	}
126 	return ipf;
127 }
128 
129 void
130 ipflow_init()
131 {
132 	int i;
133 
134 	LIST_INIT(&ipflowlist);
135 	for (i = 0; i < IPFLOW_HASHSIZE; i++)
136 		LIST_INIT(&ipflowtable[i]);
137 }
138 
139 int
140 ipflow_fastforward(
141 	struct mbuf *m)
142 {
143 	struct ip *ip, ip_store;
144 	struct ipflow *ipf;
145 	struct rtentry *rt;
146 	struct sockaddr *dst;
147 	int error;
148 	int iplen;
149 
150 	/*
151 	 * Are we forwarding packets?  Big enough for an IP packet?
152 	 */
153 	if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip))
154 		return 0;
155 
156 	/*
157 	 * Was packet received as a link-level multicast or broadcast?
158 	 * If so, don't try to fast forward..
159 	 */
160 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
161 		return 0;
162 
163 	/*
164 	 * IP header with no option and valid version and length
165 	 */
166 	if (IP_HDR_ALIGNED_P(mtod(m, caddr_t)))
167 		ip = mtod(m, struct ip *);
168 	else {
169 		memcpy(&ip_store, mtod(m, caddr_t), sizeof(ip_store));
170 		ip = &ip_store;
171 	}
172 	iplen = ntohs(ip->ip_len);
173 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
174 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
175 		return 0;
176 	/*
177 	 * Find a flow.
178 	 */
179 	if ((ipf = ipflow_lookup(ip)) == NULL)
180 		return 0;
181 
182 	/*
183 	 * Verify the IP header checksum.
184 	 */
185 	switch (m->m_pkthdr.csum_flags &
186 		((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_IPv4) |
187 		 M_CSUM_IPv4_BAD)) {
188 	case M_CSUM_IPv4|M_CSUM_IPv4_BAD:
189 		return (0);
190 
191 	case M_CSUM_IPv4:
192 		/* Checksum was okay. */
193 		break;
194 
195 	default:
196 		/* Must compute it ourselves. */
197 		if (in_cksum(m, sizeof(struct ip)) != 0)
198 			return (0);
199 		break;
200 	}
201 
202 	/*
203 	 * Route and interface still up?
204 	 */
205 	rt = ipf->ipf_ro.ro_rt;
206 	if ((rt->rt_flags & RTF_UP) == 0 ||
207 	    (rt->rt_ifp->if_flags & IFF_UP) == 0)
208 		return 0;
209 
210 	/*
211 	 * Packet size OK?  TTL?
212 	 */
213 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
214 		return 0;
215 
216 	/*
217 	 * Clear any in-bound checksum flags for this packet.
218 	 */
219 	m->m_pkthdr.csum_flags = 0;
220 
221 	/*
222 	 * Everything checks out and so we can forward this packet.
223 	 * Modify the TTL and incrementally change the checksum.
224 	 *
225 	 * This method of adding the checksum works on either endian CPU.
226 	 * If htons() is inlined, all the arithmetic is folded; otherwise
227 	 * the htons()s are combined by CSE due to the __const__ attribute.
228 	 *
229 	 * Don't bother using HW checksumming here -- the incremental
230 	 * update is pretty fast.
231 	 */
232 	ip->ip_ttl -= IPTTLDEC;
233 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
234 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
235 	else
236 		ip->ip_sum += htons(IPTTLDEC << 8);
237 
238 	/*
239 	 * Done modifying the header; copy it back, if necessary.
240 	 */
241 	if (IP_HDR_ALIGNED_P(mtod(m, caddr_t)) == 0)
242 		memcpy(mtod(m, caddr_t), &ip_store, sizeof(ip_store));
243 
244 	/*
245 	 * Trim the packet in case it's too long..
246 	 */
247 	if (m->m_pkthdr.len > iplen) {
248 		if (m->m_len == m->m_pkthdr.len) {
249 			m->m_len = iplen;
250 			m->m_pkthdr.len = iplen;
251 		} else
252 			m_adj(m, iplen - m->m_pkthdr.len);
253 	}
254 
255 	/*
256 	 * Send the packet on it's way.  All we can get back is ENOBUFS
257 	 */
258 	ipf->ipf_uses++;
259 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
260 
261 	if (rt->rt_flags & RTF_GATEWAY)
262 		dst = rt->rt_gateway;
263 	else
264 		dst = &ipf->ipf_ro.ro_dst;
265 
266 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
267 		if (error == ENOBUFS)
268 			ipf->ipf_dropped++;
269 		else
270 			ipf->ipf_errors++;
271 	}
272 	return 1;
273 }
274 
275 static void
276 ipflow_addstats(
277 	struct ipflow *ipf)
278 {
279 	ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
280 	ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
281 	ipstat.ips_forward += ipf->ipf_uses;
282 	ipstat.ips_fastforward += ipf->ipf_uses;
283 }
284 
285 static void
286 ipflow_free(
287 	struct ipflow *ipf)
288 {
289 	int s;
290 	/*
291 	 * Remove the flow from the hash table (at elevated IPL).
292 	 * Once it's off the list, we can deal with it at normal
293 	 * network IPL.
294 	 */
295 	s = splnet();
296 	IPFLOW_REMOVE(ipf);
297 	splx(s);
298 	ipflow_addstats(ipf);
299 	RTFREE(ipf->ipf_ro.ro_rt);
300 	ipflow_inuse--;
301 	pool_put(&ipflow_pool, ipf);
302 }
303 
304 struct ipflow *
305 ipflow_reap(
306 	int just_one)
307 {
308 	while (just_one || ipflow_inuse > ip_maxflows) {
309 		struct ipflow *ipf, *maybe_ipf = NULL;
310 		int s;
311 
312 		ipf = LIST_FIRST(&ipflowlist);
313 		while (ipf != NULL) {
314 			/*
315 			 * If this no longer points to a valid route
316 			 * reclaim it.
317 			 */
318 			if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
319 				goto done;
320 			/*
321 			 * choose the one that's been least recently
322 			 * used or has had the least uses in the
323 			 * last 1.5 intervals.
324 			 */
325 			if (maybe_ipf == NULL ||
326 			    ipf->ipf_timer < maybe_ipf->ipf_timer ||
327 			    (ipf->ipf_timer == maybe_ipf->ipf_timer &&
328 			     ipf->ipf_last_uses + ipf->ipf_uses <
329 			         maybe_ipf->ipf_last_uses +
330 			         maybe_ipf->ipf_uses))
331 				maybe_ipf = ipf;
332 			ipf = LIST_NEXT(ipf, ipf_list);
333 		}
334 		ipf = maybe_ipf;
335 	    done:
336 		/*
337 		 * Remove the entry from the flow table.
338 		 */
339 		s = splnet();
340 		IPFLOW_REMOVE(ipf);
341 		splx(s);
342 		ipflow_addstats(ipf);
343 		RTFREE(ipf->ipf_ro.ro_rt);
344 		if (just_one)
345 			return ipf;
346 		pool_put(&ipflow_pool, ipf);
347 		ipflow_inuse--;
348 	}
349 	return NULL;
350 }
351 
352 void
353 ipflow_slowtimo(
354 	void)
355 {
356 	struct ipflow *ipf, *next_ipf;
357 
358 	ipf = LIST_FIRST(&ipflowlist);
359 	while (ipf != NULL) {
360 		next_ipf = LIST_NEXT(ipf, ipf_list);
361 		if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer)) {
362 			ipflow_free(ipf);
363 		} else {
364 			ipf->ipf_last_uses = ipf->ipf_uses;
365 			ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
366 			ipstat.ips_forward += ipf->ipf_uses;
367 			ipstat.ips_fastforward += ipf->ipf_uses;
368 			ipf->ipf_uses = 0;
369 		}
370 		ipf = next_ipf;
371 	}
372 }
373 
374 void
375 ipflow_create(
376 	const struct route *ro,
377 	struct mbuf *m)
378 {
379 	const struct ip *const ip = mtod(m, struct ip *);
380 	struct ipflow *ipf;
381 	unsigned hash;
382 	int s;
383 
384 	/*
385 	 * Don't create cache entries for ICMP messages.
386 	 */
387 	if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP)
388 		return;
389 	/*
390 	 * See if an existing flow struct exists.  If so remove it from it's
391 	 * list and free the old route.  If not, try to malloc a new one
392 	 * (if we aren't at our limit).
393 	 */
394 	ipf = ipflow_lookup(ip);
395 	if (ipf == NULL) {
396 		if (ipflow_inuse >= ip_maxflows) {
397 			ipf = ipflow_reap(1);
398 		} else {
399 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
400 			if (ipf == NULL)
401 				return;
402 			ipflow_inuse++;
403 		}
404 		bzero((caddr_t) ipf, sizeof(*ipf));
405 	} else {
406 		s = splnet();
407 		IPFLOW_REMOVE(ipf);
408 		splx(s);
409 		ipflow_addstats(ipf);
410 		RTFREE(ipf->ipf_ro.ro_rt);
411 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
412 		ipf->ipf_errors = ipf->ipf_dropped = 0;
413 	}
414 
415 	/*
416 	 * Fill in the updated information.
417 	 */
418 	ipf->ipf_ro = *ro;
419 	ro->ro_rt->rt_refcnt++;
420 	ipf->ipf_dst = ip->ip_dst;
421 	ipf->ipf_src = ip->ip_src;
422 	ipf->ipf_tos = ip->ip_tos;
423 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
424 	ipf->ipf_start = time.tv_sec;
425 	/*
426 	 * Insert into the approriate bucket of the flow table.
427 	 */
428 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
429 	s = splnet();
430 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
431 	splx(s);
432 }
433 
434 void
435 ipflow_invalidate_all(
436 	void)
437 {
438 	struct ipflow *ipf, *next_ipf;
439 	int s;
440 
441 	s = splnet();
442 	ipf = LIST_FIRST(&ipflowlist);
443 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
444 		next_ipf = LIST_NEXT(ipf, ipf_list);
445 		ipflow_free(ipf);
446 	}
447 	splx(s);
448 }
449