xref: /netbsd-src/sys/netinet/ip_flow.c (revision 08c81a9c2dc8c7300e893321eb65c0925d60871c)
1 /*	$NetBSD: ip_flow.c,v 1.25 2002/06/30 22:40:34 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 1.25 2002/06/30 22:40:34 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, ip_store;
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 	if (IP_HDR_ALIGNED_P(mtod(m, caddr_t)))
170 		ip = mtod(m, struct ip *);
171 	else {
172 		memcpy(&ip_store, mtod(m, caddr_t), sizeof(ip_store));
173 		ip = &ip_store;
174 	}
175 	iplen = ntohs(ip->ip_len);
176 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
177 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
178 		return 0;
179 	/*
180 	 * Find a flow.
181 	 */
182 	if ((ipf = ipflow_lookup(ip)) == NULL)
183 		return 0;
184 
185 	/*
186 	 * Verify the IP header checksum.
187 	 */
188 	switch (m->m_pkthdr.csum_flags &
189 		((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_IPv4) |
190 		 M_CSUM_IPv4_BAD)) {
191 	case M_CSUM_IPv4|M_CSUM_IPv4_BAD:
192 		return (0);
193 
194 	case M_CSUM_IPv4:
195 		/* Checksum was okay. */
196 		break;
197 
198 	default:
199 		/* Must compute it ourselves. */
200 		if (in_cksum(m, sizeof(struct ip)) != 0)
201 			return (0);
202 		break;
203 	}
204 
205 	/*
206 	 * Route and interface still up?
207 	 */
208 	rt = ipf->ipf_ro.ro_rt;
209 	if ((rt->rt_flags & RTF_UP) == 0 ||
210 	    (rt->rt_ifp->if_flags & IFF_UP) == 0)
211 		return 0;
212 
213 	/*
214 	 * Packet size OK?  TTL?
215 	 */
216 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
217 		return 0;
218 
219 	/*
220 	 * Clear any in-bound checksum flags for this packet.
221 	 */
222 	m->m_pkthdr.csum_flags = 0;
223 
224 	/*
225 	 * Everything checks out and so we can forward this packet.
226 	 * Modify the TTL and incrementally change the checksum.
227 	 *
228 	 * This method of adding the checksum works on either endian CPU.
229 	 * If htons() is inlined, all the arithmetic is folded; otherwise
230 	 * the htons()s are combined by CSE due to the __const__ attribute.
231 	 *
232 	 * Don't bother using HW checksumming here -- the incremental
233 	 * update is pretty fast.
234 	 */
235 	ip->ip_ttl -= IPTTLDEC;
236 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
237 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
238 	else
239 		ip->ip_sum += htons(IPTTLDEC << 8);
240 
241 	/*
242 	 * Done modifying the header; copy it back, if necessary.
243 	 */
244 	if (IP_HDR_ALIGNED_P(mtod(m, caddr_t)) == 0)
245 		memcpy(mtod(m, caddr_t), &ip_store, sizeof(ip_store));
246 
247 	/*
248 	 * Trim the packet in case it's too long..
249 	 */
250 	if (m->m_pkthdr.len > iplen) {
251 		if (m->m_len == m->m_pkthdr.len) {
252 			m->m_len = iplen;
253 			m->m_pkthdr.len = iplen;
254 		} else
255 			m_adj(m, iplen - m->m_pkthdr.len);
256 	}
257 
258 	/*
259 	 * Send the packet on it's way.  All we can get back is ENOBUFS
260 	 */
261 	ipf->ipf_uses++;
262 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
263 
264 	if (rt->rt_flags & RTF_GATEWAY)
265 		dst = rt->rt_gateway;
266 	else
267 		dst = &ipf->ipf_ro.ro_dst;
268 
269 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
270 		if (error == ENOBUFS)
271 			ipf->ipf_dropped++;
272 		else
273 			ipf->ipf_errors++;
274 	}
275 	return 1;
276 }
277 
278 static void
279 ipflow_addstats(
280 	struct ipflow *ipf)
281 {
282 	ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
283 	ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
284 	ipstat.ips_forward += ipf->ipf_uses;
285 	ipstat.ips_fastforward += ipf->ipf_uses;
286 }
287 
288 static void
289 ipflow_free(
290 	struct ipflow *ipf)
291 {
292 	int s;
293 	/*
294 	 * Remove the flow from the hash table (at elevated IPL).
295 	 * Once it's off the list, we can deal with it at normal
296 	 * network IPL.
297 	 */
298 	s = splnet();
299 	IPFLOW_REMOVE(ipf);
300 	splx(s);
301 	ipflow_addstats(ipf);
302 	RTFREE(ipf->ipf_ro.ro_rt);
303 	ipflow_inuse--;
304 	pool_put(&ipflow_pool, ipf);
305 }
306 
307 struct ipflow *
308 ipflow_reap(
309 	int just_one)
310 {
311 	while (just_one || ipflow_inuse > ip_maxflows) {
312 		struct ipflow *ipf, *maybe_ipf = NULL;
313 		int s;
314 
315 		ipf = LIST_FIRST(&ipflowlist);
316 		while (ipf != NULL) {
317 			/*
318 			 * If this no longer points to a valid route
319 			 * reclaim it.
320 			 */
321 			if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
322 				goto done;
323 			/*
324 			 * choose the one that's been least recently
325 			 * used or has had the least uses in the
326 			 * last 1.5 intervals.
327 			 */
328 			if (maybe_ipf == NULL ||
329 			    ipf->ipf_timer < maybe_ipf->ipf_timer ||
330 			    (ipf->ipf_timer == maybe_ipf->ipf_timer &&
331 			     ipf->ipf_last_uses + ipf->ipf_uses <
332 			         maybe_ipf->ipf_last_uses +
333 			         maybe_ipf->ipf_uses))
334 				maybe_ipf = ipf;
335 			ipf = LIST_NEXT(ipf, ipf_list);
336 		}
337 		ipf = maybe_ipf;
338 	    done:
339 		/*
340 		 * Remove the entry from the flow table.
341 		 */
342 		s = splnet();
343 		IPFLOW_REMOVE(ipf);
344 		splx(s);
345 		ipflow_addstats(ipf);
346 		RTFREE(ipf->ipf_ro.ro_rt);
347 		if (just_one)
348 			return ipf;
349 		pool_put(&ipflow_pool, ipf);
350 		ipflow_inuse--;
351 	}
352 	return NULL;
353 }
354 
355 void
356 ipflow_slowtimo(
357 	void)
358 {
359 	struct ipflow *ipf, *next_ipf;
360 
361 	ipf = LIST_FIRST(&ipflowlist);
362 	while (ipf != NULL) {
363 		next_ipf = LIST_NEXT(ipf, ipf_list);
364 		if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer)) {
365 			ipflow_free(ipf);
366 		} else {
367 			ipf->ipf_last_uses = ipf->ipf_uses;
368 			ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
369 			ipstat.ips_forward += ipf->ipf_uses;
370 			ipstat.ips_fastforward += ipf->ipf_uses;
371 			ipf->ipf_uses = 0;
372 		}
373 		ipf = next_ipf;
374 	}
375 }
376 
377 void
378 ipflow_create(
379 	const struct route *ro,
380 	struct mbuf *m)
381 {
382 	const struct ip *const ip = mtod(m, struct ip *);
383 	struct ipflow *ipf;
384 	unsigned hash;
385 	int s;
386 
387 	/*
388 	 * Don't create cache entries for ICMP messages.
389 	 */
390 	if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP)
391 		return;
392 	/*
393 	 * See if an existing flow struct exists.  If so remove it from it's
394 	 * list and free the old route.  If not, try to malloc a new one
395 	 * (if we aren't at our limit).
396 	 */
397 	ipf = ipflow_lookup(ip);
398 	if (ipf == NULL) {
399 		if (ipflow_inuse >= ip_maxflows) {
400 			ipf = ipflow_reap(1);
401 		} else {
402 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
403 			if (ipf == NULL)
404 				return;
405 			ipflow_inuse++;
406 		}
407 		bzero((caddr_t) ipf, sizeof(*ipf));
408 	} else {
409 		s = splnet();
410 		IPFLOW_REMOVE(ipf);
411 		splx(s);
412 		ipflow_addstats(ipf);
413 		RTFREE(ipf->ipf_ro.ro_rt);
414 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
415 		ipf->ipf_errors = ipf->ipf_dropped = 0;
416 	}
417 
418 	/*
419 	 * Fill in the updated information.
420 	 */
421 	ipf->ipf_ro = *ro;
422 	ro->ro_rt->rt_refcnt++;
423 	ipf->ipf_dst = ip->ip_dst;
424 	ipf->ipf_src = ip->ip_src;
425 	ipf->ipf_tos = ip->ip_tos;
426 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
427 	ipf->ipf_start = time.tv_sec;
428 	/*
429 	 * Insert into the approriate bucket of the flow table.
430 	 */
431 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
432 	s = splnet();
433 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
434 	splx(s);
435 }
436