xref: /netbsd-src/sys/netinet/ip_flow.c (revision 4472dbe5e3bd91ef2540bada7a7ca7384627ff9b)
1 /*	$NetBSD: ip_flow.c,v 1.14 1999/10/17 23:38:45 sommerfeld 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/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/domain.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/errno.h>
48 #include <sys/time.h>
49 #include <sys/kernel.h>
50 #include <sys/proc.h>
51 #include <sys/pool.h>
52 
53 #include <vm/vm.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 	    0, NULL, NULL, M_IPFLOW);
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 	int error;
150 	int iplen;
151 
152 	/*
153 	 * Are we forwarding packets?  Big enough for an IP packet?
154 	 */
155 	if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip))
156 		return 0;
157 
158 	/*
159 	 * Was packet recieved as a link-level multicast or broadcast?
160 	 * If so, don't try to fast forward..
161 	 */
162 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
163 		return 0;
164 
165 	/*
166 	 * IP header with no option and valid version and length
167 	 */
168 	ip = mtod(m, struct ip *);
169 	iplen = ntohs(ip->ip_len);
170 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
171 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
172 		return 0;
173 	/*
174 	 * Find a flow.
175 	 */
176 	if ((ipf = ipflow_lookup(ip)) == NULL)
177 		return 0;
178 
179 	/*
180 	 * Veryify the IP header checksum.
181 	 */
182 	if (in_cksum(m, sizeof(struct ip)) != 0)
183 		return 0;
184 
185 	/*
186 	 * Route and interface still up?
187 	 */
188 	rt = ipf->ipf_ro.ro_rt;
189 	if ((rt->rt_flags & RTF_UP) == 0 ||
190 	    (rt->rt_ifp->if_flags & IFF_UP) == 0)
191 		return 0;
192 
193 	/*
194 	 * Packet size OK?  TTL?
195 	 */
196 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
197 		return 0;
198 
199 	/*
200 	 * Everything checks out and so we can forward this packet.
201 	 * Modify the TTL and incrementally change the checksum.
202 	 *
203 	 * This method of adding the checksum works on either endian CPU.
204 	 * If htons() is inlined, all the arithmetic is folded; otherwise
205 	 * the htons()s are combined by CSE due to the __const__ attribute.
206 	 */
207 	ip->ip_ttl -= IPTTLDEC;
208 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
209 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
210 	else
211 		ip->ip_sum += htons(IPTTLDEC << 8);
212 
213 	/*
214 	 * Trim the packet in case it's too long..
215 	 */
216 	if (m->m_pkthdr.len > iplen) {
217 		if (m->m_len == m->m_pkthdr.len) {
218 			m->m_len = iplen;
219 			m->m_pkthdr.len = iplen;
220 		} else
221 			m_adj(m, iplen - m->m_pkthdr.len);
222 	}
223 
224 	/*
225 	 * Send the packet on it's way.  All we can get back is ENOBUFS
226 	 */
227 	ipf->ipf_uses++;
228 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
229 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m,
230 	    &ipf->ipf_ro.ro_dst, rt)) != 0) {
231 		if (error == ENOBUFS)
232 			ipf->ipf_dropped++;
233 		else
234 			ipf->ipf_errors++;
235 	}
236 	return 1;
237 }
238 
239 static void
240 ipflow_addstats(
241 	struct ipflow *ipf)
242 {
243 	ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
244 	ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
245 	ipstat.ips_forward += ipf->ipf_uses;
246 	ipstat.ips_fastforward += ipf->ipf_uses;
247 }
248 
249 static void
250 ipflow_free(
251 	struct ipflow *ipf)
252 {
253 	int s;
254 	/*
255 	 * Remove the flow from the hash table (at elevated IPL).
256 	 * Once it's off the list, we can deal with it at normal
257 	 * network IPL.
258 	 */
259 	s = splimp();
260 	IPFLOW_REMOVE(ipf);
261 	splx(s);
262 	ipflow_addstats(ipf);
263 	RTFREE(ipf->ipf_ro.ro_rt);
264 	ipflow_inuse--;
265 	pool_put(&ipflow_pool, ipf);
266 }
267 
268 struct ipflow *
269 ipflow_reap(
270 	int just_one)
271 {
272 	while (just_one || ipflow_inuse > ip_maxflows) {
273 		struct ipflow *ipf, *maybe_ipf = NULL;
274 		int s;
275 
276 		ipf = LIST_FIRST(&ipflowlist);
277 		while (ipf != NULL) {
278 			/*
279 			 * If this no longer points to a valid route
280 			 * reclaim it.
281 			 */
282 			if ((ipf->ipf_ro.ro_rt->rt_flags & RTF_UP) == 0)
283 				goto done;
284 			/*
285 			 * choose the one that's been least recently
286 			 * used or has had the least uses in the
287 			 * last 1.5 intervals.
288 			 */
289 			if (maybe_ipf == NULL ||
290 			    ipf->ipf_timer < maybe_ipf->ipf_timer ||
291 			    (ipf->ipf_timer == maybe_ipf->ipf_timer &&
292 			     ipf->ipf_last_uses + ipf->ipf_uses <
293 			         maybe_ipf->ipf_last_uses +
294 			         maybe_ipf->ipf_uses))
295 				maybe_ipf = ipf;
296 			ipf = LIST_NEXT(ipf, ipf_list);
297 		}
298 		ipf = maybe_ipf;
299 	    done:
300 		/*
301 		 * Remove the entry from the flow table.
302 		 */
303 		s = splimp();
304 		IPFLOW_REMOVE(ipf);
305 		splx(s);
306 		ipflow_addstats(ipf);
307 		RTFREE(ipf->ipf_ro.ro_rt);
308 		if (just_one)
309 			return ipf;
310 		pool_put(&ipflow_pool, ipf);
311 		ipflow_inuse--;
312 	}
313 	return NULL;
314 }
315 
316 void
317 ipflow_slowtimo(
318 	void)
319 {
320 	struct ipflow *ipf, *next_ipf;
321 
322 	ipf = LIST_FIRST(&ipflowlist);
323 	while (ipf != NULL) {
324 		next_ipf = LIST_NEXT(ipf, ipf_list);
325 		if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer)) {
326 			ipflow_free(ipf);
327 		} else {
328 			ipf->ipf_last_uses = ipf->ipf_uses;
329 			ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
330 			ipstat.ips_forward += ipf->ipf_uses;
331 			ipstat.ips_fastforward += ipf->ipf_uses;
332 			ipf->ipf_uses = 0;
333 		}
334 		ipf = next_ipf;
335 	}
336 }
337 
338 void
339 ipflow_create(
340 	const struct route *ro,
341 	struct mbuf *m)
342 {
343 	const struct ip *const ip = mtod(m, struct ip *);
344 	struct ipflow *ipf;
345 	unsigned hash;
346 	int s;
347 
348 	/*
349 	 * Don't create cache entries for ICMP messages.
350 	 */
351 	if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP)
352 		return;
353 	/*
354 	 * See if an existing flow struct exists.  If so remove it from it's
355 	 * list and free the old route.  If not, try to malloc a new one
356 	 * (if we aren't at our limit).
357 	 */
358 	ipf = ipflow_lookup(ip);
359 	if (ipf == NULL) {
360 		if (ipflow_inuse >= ip_maxflows) {
361 			ipf = ipflow_reap(1);
362 		} else {
363 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
364 			if (ipf == NULL)
365 				return;
366 			ipflow_inuse++;
367 		}
368 		bzero((caddr_t) ipf, sizeof(*ipf));
369 	} else {
370 		s = splimp();
371 		IPFLOW_REMOVE(ipf);
372 		splx(s);
373 		ipflow_addstats(ipf);
374 		RTFREE(ipf->ipf_ro.ro_rt);
375 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
376 		ipf->ipf_errors = ipf->ipf_dropped = 0;
377 	}
378 
379 	/*
380 	 * Fill in the updated information.
381 	 */
382 	ipf->ipf_ro = *ro;
383 	ro->ro_rt->rt_refcnt++;
384 	ipf->ipf_dst = ip->ip_dst;
385 	ipf->ipf_src = ip->ip_src;
386 	ipf->ipf_tos = ip->ip_tos;
387 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
388 	ipf->ipf_start = time.tv_sec;
389 	/*
390 	 * Insert into the approriate bucket of the flow table.
391 	 */
392 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
393 	s = splimp();
394 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
395 	splx(s);
396 }
397