xref: /netbsd-src/sys/netinet/ip_flow.c (revision 5e4c038a45edbc7d63b7c2daa76e29f88b64a4e3)
1 /*	$NetBSD: ip_flow.c,v 1.23 2002/03/08 20:48:43 thorpej 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 1.23 2002/03/08 20:48:43 thorpej 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 struct pool ipflow_pool;
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 (0)
84 
85 #define	IPFLOW_REMOVE(ipf) \
86 do { \
87 	LIST_REMOVE((ipf), ipf_hash); \
88 	LIST_REMOVE((ipf), ipf_list); \
89 } while (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 	pool_init(&ipflow_pool, sizeof(struct ipflow), 0, 0, 0, "ipflowpl",
135 	    NULL);
136 
137 	LIST_INIT(&ipflowlist);
138 	for (i = 0; i < IPFLOW_HASHSIZE; i++)
139 		LIST_INIT(&ipflowtable[i]);
140 }
141 
142 int
143 ipflow_fastforward(
144 	struct mbuf *m)
145 {
146 	struct ip *ip;
147 	struct ipflow *ipf;
148 	struct rtentry *rt;
149 	struct sockaddr *dst;
150 	int error;
151 	int iplen;
152 
153 	/*
154 	 * Are we forwarding packets?  Big enough for an IP packet?
155 	 */
156 	if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip))
157 		return 0;
158 
159 	/*
160 	 * Was packet received as a link-level multicast or broadcast?
161 	 * If so, don't try to fast forward..
162 	 */
163 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
164 		return 0;
165 
166 	/*
167 	 * IP header with no option and valid version and length
168 	 */
169 	ip = mtod(m, struct ip *);
170 	iplen = ntohs(ip->ip_len);
171 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
172 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
173 		return 0;
174 	/*
175 	 * Find a flow.
176 	 */
177 	if ((ipf = ipflow_lookup(ip)) == NULL)
178 		return 0;
179 
180 	/*
181 	 * Verify the IP header checksum.
182 	 */
183 	switch (m->m_pkthdr.csum_flags &
184 		((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_IPv4) |
185 		 M_CSUM_IPv4_BAD)) {
186 	case M_CSUM_IPv4|M_CSUM_IPv4_BAD:
187 		return (0);
188 
189 	case M_CSUM_IPv4:
190 		/* Checksum was okay. */
191 		break;
192 
193 	default:
194 		/* Must compute it ourselves. */
195 		if (in_cksum(m, sizeof(struct ip)) != 0)
196 			return (0);
197 		break;
198 	}
199 
200 	/*
201 	 * Route and interface still up?
202 	 */
203 	rt = ipf->ipf_ro.ro_rt;
204 	if ((rt->rt_flags & RTF_UP) == 0 ||
205 	    (rt->rt_ifp->if_flags & IFF_UP) == 0)
206 		return 0;
207 
208 	/*
209 	 * Packet size OK?  TTL?
210 	 */
211 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
212 		return 0;
213 
214 	/*
215 	 * Clear any in-bound checksum flags for this packet.
216 	 */
217 	m->m_pkthdr.csum_flags = 0;
218 
219 	/*
220 	 * Everything checks out and so we can forward this packet.
221 	 * Modify the TTL and incrementally change the checksum.
222 	 *
223 	 * This method of adding the checksum works on either endian CPU.
224 	 * If htons() is inlined, all the arithmetic is folded; otherwise
225 	 * the htons()s are combined by CSE due to the __const__ attribute.
226 	 *
227 	 * Don't bother using HW checksumming here -- the incremental
228 	 * update is pretty fast.
229 	 */
230 	ip->ip_ttl -= IPTTLDEC;
231 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
232 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
233 	else
234 		ip->ip_sum += htons(IPTTLDEC << 8);
235 
236 	/*
237 	 * Trim the packet in case it's too long..
238 	 */
239 	if (m->m_pkthdr.len > iplen) {
240 		if (m->m_len == m->m_pkthdr.len) {
241 			m->m_len = iplen;
242 			m->m_pkthdr.len = iplen;
243 		} else
244 			m_adj(m, iplen - m->m_pkthdr.len);
245 	}
246 
247 	/*
248 	 * Send the packet on it's way.  All we can get back is ENOBUFS
249 	 */
250 	ipf->ipf_uses++;
251 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
252 
253 	if (rt->rt_flags & RTF_GATEWAY)
254 		dst = rt->rt_gateway;
255 	else
256 		dst = &ipf->ipf_ro.ro_dst;
257 
258 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
259 		if (error == ENOBUFS)
260 			ipf->ipf_dropped++;
261 		else
262 			ipf->ipf_errors++;
263 	}
264 	return 1;
265 }
266 
267 static void
268 ipflow_addstats(
269 	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_forward += ipf->ipf_uses;
274 	ipstat.ips_fastforward += ipf->ipf_uses;
275 }
276 
277 static void
278 ipflow_free(
279 	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 	pool_put(&ipflow_pool, ipf);
294 }
295 
296 struct ipflow *
297 ipflow_reap(
298 	int just_one)
299 {
300 	while (just_one || ipflow_inuse > ip_maxflows) {
301 		struct ipflow *ipf, *maybe_ipf = NULL;
302 		int s;
303 
304 		ipf = LIST_FIRST(&ipflowlist);
305 		while (ipf != NULL) {
306 			/*
307 			 * If this no longer points to a valid route
308 			 * reclaim it.
309 			 */
310 			if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
311 				goto done;
312 			/*
313 			 * choose the one that's been least recently
314 			 * used or has had the least uses in the
315 			 * last 1.5 intervals.
316 			 */
317 			if (maybe_ipf == NULL ||
318 			    ipf->ipf_timer < maybe_ipf->ipf_timer ||
319 			    (ipf->ipf_timer == maybe_ipf->ipf_timer &&
320 			     ipf->ipf_last_uses + ipf->ipf_uses <
321 			         maybe_ipf->ipf_last_uses +
322 			         maybe_ipf->ipf_uses))
323 				maybe_ipf = ipf;
324 			ipf = LIST_NEXT(ipf, ipf_list);
325 		}
326 		ipf = maybe_ipf;
327 	    done:
328 		/*
329 		 * Remove the entry from the flow table.
330 		 */
331 		s = splnet();
332 		IPFLOW_REMOVE(ipf);
333 		splx(s);
334 		ipflow_addstats(ipf);
335 		RTFREE(ipf->ipf_ro.ro_rt);
336 		if (just_one)
337 			return ipf;
338 		pool_put(&ipflow_pool, ipf);
339 		ipflow_inuse--;
340 	}
341 	return NULL;
342 }
343 
344 void
345 ipflow_slowtimo(
346 	void)
347 {
348 	struct ipflow *ipf, *next_ipf;
349 
350 	ipf = LIST_FIRST(&ipflowlist);
351 	while (ipf != NULL) {
352 		next_ipf = LIST_NEXT(ipf, ipf_list);
353 		if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer)) {
354 			ipflow_free(ipf);
355 		} else {
356 			ipf->ipf_last_uses = ipf->ipf_uses;
357 			ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
358 			ipstat.ips_forward += ipf->ipf_uses;
359 			ipstat.ips_fastforward += ipf->ipf_uses;
360 			ipf->ipf_uses = 0;
361 		}
362 		ipf = next_ipf;
363 	}
364 }
365 
366 void
367 ipflow_create(
368 	const struct route *ro,
369 	struct mbuf *m)
370 {
371 	const struct ip *const ip = mtod(m, struct ip *);
372 	struct ipflow *ipf;
373 	unsigned hash;
374 	int s;
375 
376 	/*
377 	 * Don't create cache entries for ICMP messages.
378 	 */
379 	if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP)
380 		return;
381 	/*
382 	 * See if an existing flow struct exists.  If so remove it from it's
383 	 * list and free the old route.  If not, try to malloc a new one
384 	 * (if we aren't at our limit).
385 	 */
386 	ipf = ipflow_lookup(ip);
387 	if (ipf == NULL) {
388 		if (ipflow_inuse >= ip_maxflows) {
389 			ipf = ipflow_reap(1);
390 		} else {
391 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
392 			if (ipf == NULL)
393 				return;
394 			ipflow_inuse++;
395 		}
396 		bzero((caddr_t) ipf, sizeof(*ipf));
397 	} else {
398 		s = splnet();
399 		IPFLOW_REMOVE(ipf);
400 		splx(s);
401 		ipflow_addstats(ipf);
402 		RTFREE(ipf->ipf_ro.ro_rt);
403 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
404 		ipf->ipf_errors = ipf->ipf_dropped = 0;
405 	}
406 
407 	/*
408 	 * Fill in the updated information.
409 	 */
410 	ipf->ipf_ro = *ro;
411 	ro->ro_rt->rt_refcnt++;
412 	ipf->ipf_dst = ip->ip_dst;
413 	ipf->ipf_src = ip->ip_src;
414 	ipf->ipf_tos = ip->ip_tos;
415 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
416 	ipf->ipf_start = time.tv_sec;
417 	/*
418 	 * Insert into the approriate bucket of the flow table.
419 	 */
420 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
421 	s = splnet();
422 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
423 	splx(s);
424 }
425