xref: /netbsd-src/sys/netinet/ip_flow.c (revision ce2c90c7c172d95d2402a5b3d96d8f8e6d138a21)
1 /*	$NetBSD: ip_flow.c,v 1.36 2006/10/06 03:20:47 mrg 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 1.36 2006/10/06 03:20:47 mrg 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(struct in_addr dst,	struct in_addr src, unsigned tos)
98 {
99 	unsigned hash = tos;
100 	int idx;
101 	for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS)
102 		hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx);
103 	return hash & (IPFLOW_HASHSIZE-1);
104 }
105 
106 static struct ipflow *
107 ipflow_lookup(const struct ip *ip)
108 {
109 	unsigned hash;
110 	struct ipflow *ipf;
111 
112 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
113 
114 	LIST_FOREACH(ipf, &ipflowtable[hash], ipf_hash) {
115 		if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
116 		    && ip->ip_src.s_addr == ipf->ipf_src.s_addr
117 		    && ip->ip_tos == ipf->ipf_tos)
118 			break;
119 	}
120 	return ipf;
121 }
122 
123 void
124 ipflow_init(void)
125 {
126 	int i;
127 
128 	LIST_INIT(&ipflowlist);
129 	for (i = 0; i < IPFLOW_HASHSIZE; i++)
130 		LIST_INIT(&ipflowtable[i]);
131 }
132 
133 int
134 ipflow_fastforward(struct mbuf *m)
135 {
136 	struct ip *ip, ip_store;
137 	struct ipflow *ipf;
138 	struct rtentry *rt;
139 	struct sockaddr *dst;
140 	int error;
141 	int iplen;
142 
143 	/*
144 	 * Are we forwarding packets?  Big enough for an IP packet?
145 	 */
146 	if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip))
147 		return 0;
148 
149 	/*
150 	 * Was packet received as a link-level multicast or broadcast?
151 	 * If so, don't try to fast forward..
152 	 */
153 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
154 		return 0;
155 
156 	/*
157 	 * IP header with no option and valid version and length
158 	 */
159 	if (IP_HDR_ALIGNED_P(mtod(m, caddr_t)))
160 		ip = mtod(m, struct ip *);
161 	else {
162 		memcpy(&ip_store, mtod(m, caddr_t), sizeof(ip_store));
163 		ip = &ip_store;
164 	}
165 	iplen = ntohs(ip->ip_len);
166 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
167 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
168 		return 0;
169 	/*
170 	 * Find a flow.
171 	 */
172 	if ((ipf = ipflow_lookup(ip)) == NULL)
173 		return 0;
174 
175 	/*
176 	 * Verify the IP header checksum.
177 	 */
178 	switch (m->m_pkthdr.csum_flags &
179 		((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_IPv4) |
180 		 M_CSUM_IPv4_BAD)) {
181 	case M_CSUM_IPv4|M_CSUM_IPv4_BAD:
182 		return (0);
183 
184 	case M_CSUM_IPv4:
185 		/* Checksum was okay. */
186 		break;
187 
188 	default:
189 		/* Must compute it ourselves. */
190 		if (in_cksum(m, sizeof(struct ip)) != 0)
191 			return (0);
192 		break;
193 	}
194 
195 	/*
196 	 * Route and interface still up?
197 	 */
198 	rt = ipf->ipf_ro.ro_rt;
199 	if ((rt->rt_flags & RTF_UP) == 0 ||
200 	    (rt->rt_ifp->if_flags & IFF_UP) == 0)
201 		return 0;
202 
203 	/*
204 	 * Packet size OK?  TTL?
205 	 */
206 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
207 		return 0;
208 
209 	/*
210 	 * Clear any in-bound checksum flags for this packet.
211 	 */
212 	m->m_pkthdr.csum_flags = 0;
213 
214 	/*
215 	 * Everything checks out and so we can forward this packet.
216 	 * Modify the TTL and incrementally change the checksum.
217 	 *
218 	 * This method of adding the checksum works on either endian CPU.
219 	 * If htons() is inlined, all the arithmetic is folded; otherwise
220 	 * the htons()s are combined by CSE due to the const attribute.
221 	 *
222 	 * Don't bother using HW checksumming here -- the incremental
223 	 * update is pretty fast.
224 	 */
225 	ip->ip_ttl -= IPTTLDEC;
226 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
227 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
228 	else
229 		ip->ip_sum += htons(IPTTLDEC << 8);
230 
231 	/*
232 	 * Done modifying the header; copy it back, if necessary.
233 	 */
234 	if (IP_HDR_ALIGNED_P(mtod(m, caddr_t)) == 0)
235 		memcpy(mtod(m, caddr_t), &ip_store, sizeof(ip_store));
236 
237 	/*
238 	 * Trim the packet in case it's too long..
239 	 */
240 	if (m->m_pkthdr.len > iplen) {
241 		if (m->m_len == m->m_pkthdr.len) {
242 			m->m_len = iplen;
243 			m->m_pkthdr.len = iplen;
244 		} else
245 			m_adj(m, iplen - m->m_pkthdr.len);
246 	}
247 
248 	/*
249 	 * Send the packet on it's way.  All we can get back is ENOBUFS
250 	 */
251 	ipf->ipf_uses++;
252 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
253 
254 	if (rt->rt_flags & RTF_GATEWAY)
255 		dst = rt->rt_gateway;
256 	else
257 		dst = &ipf->ipf_ro.ro_dst;
258 
259 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
260 		if (error == ENOBUFS)
261 			ipf->ipf_dropped++;
262 		else
263 			ipf->ipf_errors++;
264 	}
265 	return 1;
266 }
267 
268 static void
269 ipflow_addstats(struct ipflow *ipf)
270 {
271 	ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
272 	ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
273 	ipstat.ips_total += ipf->ipf_uses;
274 	ipstat.ips_forward += ipf->ipf_uses;
275 	ipstat.ips_fastforward += ipf->ipf_uses;
276 }
277 
278 static void
279 ipflow_free(struct ipflow *ipf)
280 {
281 	int s;
282 	/*
283 	 * Remove the flow from the hash table (at elevated IPL).
284 	 * Once it's off the list, we can deal with it at normal
285 	 * network IPL.
286 	 */
287 	s = splnet();
288 	IPFLOW_REMOVE(ipf);
289 	splx(s);
290 	ipflow_addstats(ipf);
291 	RTFREE(ipf->ipf_ro.ro_rt);
292 	ipflow_inuse--;
293 	s = splnet();
294 	pool_put(&ipflow_pool, ipf);
295 	splx(s);
296 }
297 
298 struct ipflow *
299 ipflow_reap(int just_one)
300 {
301 	while (just_one || ipflow_inuse > ip_maxflows) {
302 		struct ipflow *ipf, *maybe_ipf = NULL;
303 		int s;
304 
305 		ipf = LIST_FIRST(&ipflowlist);
306 		while (ipf != NULL) {
307 			/*
308 			 * If this no longer points to a valid route
309 			 * reclaim it.
310 			 */
311 			if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
312 				goto done;
313 			/*
314 			 * choose the one that's been least recently
315 			 * used or has had the least uses in the
316 			 * last 1.5 intervals.
317 			 */
318 			if (maybe_ipf == NULL ||
319 			    ipf->ipf_timer < maybe_ipf->ipf_timer ||
320 			    (ipf->ipf_timer == maybe_ipf->ipf_timer &&
321 			     ipf->ipf_last_uses + ipf->ipf_uses <
322 			         maybe_ipf->ipf_last_uses +
323 			         maybe_ipf->ipf_uses))
324 				maybe_ipf = ipf;
325 			ipf = LIST_NEXT(ipf, ipf_list);
326 		}
327 		ipf = maybe_ipf;
328 	    done:
329 		/*
330 		 * Remove the entry from the flow table.
331 		 */
332 		s = splnet();
333 		IPFLOW_REMOVE(ipf);
334 		splx(s);
335 		ipflow_addstats(ipf);
336 		RTFREE(ipf->ipf_ro.ro_rt);
337 		if (just_one)
338 			return ipf;
339 		pool_put(&ipflow_pool, ipf);
340 		ipflow_inuse--;
341 	}
342 	return NULL;
343 }
344 
345 void
346 ipflow_slowtimo(void)
347 {
348 	struct ipflow *ipf, *next_ipf;
349 
350 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
351 		next_ipf = LIST_NEXT(ipf, ipf_list);
352 		if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer)) {
353 			ipflow_free(ipf);
354 		} else {
355 			ipf->ipf_last_uses = ipf->ipf_uses;
356 			ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
357 			ipstat.ips_total += ipf->ipf_uses;
358 			ipstat.ips_forward += ipf->ipf_uses;
359 			ipstat.ips_fastforward += ipf->ipf_uses;
360 			ipf->ipf_uses = 0;
361 		}
362 	}
363 }
364 
365 void
366 ipflow_create(const struct route *ro, struct mbuf *m)
367 {
368 	const struct ip *const ip = mtod(m, struct ip *);
369 	struct ipflow *ipf;
370 	unsigned hash;
371 	int s;
372 
373 	/*
374 	 * Don't create cache entries for ICMP messages.
375 	 */
376 	if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP)
377 		return;
378 	/*
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).
382 	 */
383 	ipf = ipflow_lookup(ip);
384 	if (ipf == NULL) {
385 		if (ipflow_inuse >= ip_maxflows) {
386 			ipf = ipflow_reap(1);
387 		} else {
388 			s = splnet();
389 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
390 			splx(s);
391 			if (ipf == NULL)
392 				return;
393 			ipflow_inuse++;
394 		}
395 		bzero((caddr_t) ipf, sizeof(*ipf));
396 	} else {
397 		s = splnet();
398 		IPFLOW_REMOVE(ipf);
399 		splx(s);
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;
404 	}
405 
406 	/*
407 	 * Fill in the updated information.
408 	 */
409 	ipf->ipf_ro = *ro;
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 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
415 	ipf->ipf_start = time_uptime;
416 	/*
417 	 * Insert into the approriate bucket of the flow table.
418 	 */
419 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
420 	s = splnet();
421 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
422 	splx(s);
423 }
424 
425 void
426 ipflow_invalidate_all(void)
427 {
428 	struct ipflow *ipf, *next_ipf;
429 	int s;
430 
431 	s = splnet();
432 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
433 		next_ipf = LIST_NEXT(ipf, ipf_list);
434 		ipflow_free(ipf);
435 	}
436 	splx(s);
437 }
438