xref: /netbsd-src/sys/netinet/ip_flow.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: ip_flow.c,v 1.19 2001/06/12 15:17:28 wiz 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 <uvm/uvm_extern.h>
54 
55 #include <sys/sysctl.h>
56 
57 #include <net/if.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
60 #include <net/pfil.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip_var.h>
68 
69 struct pool ipflow_pool;
70 
71 LIST_HEAD(ipflowhead, ipflow);
72 
73 #define	IPFLOW_TIMER		(5 * PR_SLOWHZ)
74 #define	IPFLOW_HASHSIZE		(1 << IPFLOW_HASHBITS)
75 
76 static struct ipflowhead ipflowtable[IPFLOW_HASHSIZE];
77 static struct ipflowhead ipflowlist;
78 static int ipflow_inuse;
79 
80 #define	IPFLOW_INSERT(bucket, ipf) \
81 do { \
82 	LIST_INSERT_HEAD((bucket), (ipf), ipf_hash); \
83 	LIST_INSERT_HEAD(&ipflowlist, (ipf), ipf_list); \
84 } while (0)
85 
86 #define	IPFLOW_REMOVE(ipf) \
87 do { \
88 	LIST_REMOVE((ipf), ipf_hash); \
89 	LIST_REMOVE((ipf), ipf_list); \
90 } while (0)
91 
92 #ifndef IPFLOW_MAX
93 #define	IPFLOW_MAX		256
94 #endif
95 int ip_maxflows = IPFLOW_MAX;
96 
97 static unsigned
98 ipflow_hash(
99 	struct in_addr dst,
100 	struct in_addr src,
101 	unsigned tos)
102 {
103 	unsigned hash = tos;
104 	int idx;
105 	for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS)
106 		hash += (dst.s_addr >> (32 - idx)) + (src.s_addr >> idx);
107 	return hash & (IPFLOW_HASHSIZE-1);
108 }
109 
110 static struct ipflow *
111 ipflow_lookup(
112 	const struct ip *ip)
113 {
114 	unsigned hash;
115 	struct ipflow *ipf;
116 
117 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
118 
119 	ipf = LIST_FIRST(&ipflowtable[hash]);
120 	while (ipf != NULL) {
121 		if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
122 		    && ip->ip_src.s_addr == ipf->ipf_src.s_addr
123 		    && ip->ip_tos == ipf->ipf_tos)
124 			break;
125 		ipf = LIST_NEXT(ipf, ipf_hash);
126 	}
127 	return ipf;
128 }
129 
130 void
131 ipflow_init()
132 {
133 	int i;
134 
135 	pool_init(&ipflow_pool, sizeof(struct ipflow), 0, 0, 0, "ipflowpl",
136 	    0, NULL, NULL, M_IPFLOW);
137 
138 	LIST_INIT(&ipflowlist);
139 	for (i = 0; i < IPFLOW_HASHSIZE; i++)
140 		LIST_INIT(&ipflowtable[i]);
141 }
142 
143 int
144 ipflow_fastforward(
145 	struct mbuf *m)
146 {
147 	struct ip *ip;
148 	struct ipflow *ipf;
149 	struct rtentry *rt;
150 	struct sockaddr *dst;
151 	int error;
152 	int iplen;
153 
154 	/*
155 	 * Are we forwarding packets?  Big enough for an IP packet?
156 	 */
157 	if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip))
158 		return 0;
159 
160 	/*
161 	 * Was packet received as a link-level multicast or broadcast?
162 	 * If so, don't try to fast forward..
163 	 */
164 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
165 		return 0;
166 
167 	/*
168 	 * IP header with no option and valid version and length
169 	 */
170 	ip = mtod(m, struct ip *);
171 	iplen = ntohs(ip->ip_len);
172 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
173 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
174 		return 0;
175 	/*
176 	 * Find a flow.
177 	 */
178 	if ((ipf = ipflow_lookup(ip)) == NULL)
179 		return 0;
180 
181 	/*
182 	 * Verify the IP header checksum.
183 	 */
184 	switch (m->m_pkthdr.csum_flags &
185 		((m->m_pkthdr.rcvif->if_csum_flags & M_CSUM_IPv4) |
186 		 M_CSUM_IPv4_BAD)) {
187 	case M_CSUM_IPv4|M_CSUM_IPv4_BAD:
188 		return (0);
189 
190 	case M_CSUM_IPv4:
191 		/* Checksum was okay. */
192 		break;
193 
194 	default:
195 		/* Must compute it ourselves. */
196 		if (in_cksum(m, sizeof(struct ip)) != 0)
197 			return (0);
198 		break;
199 	}
200 
201 	/*
202 	 * Route and interface still up?
203 	 */
204 	rt = ipf->ipf_ro.ro_rt;
205 	if ((rt->rt_flags & RTF_UP) == 0 ||
206 	    (rt->rt_ifp->if_flags & IFF_UP) == 0)
207 		return 0;
208 
209 	/*
210 	 * Packet size OK?  TTL?
211 	 */
212 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
213 		return 0;
214 
215 	/*
216 	 * Clear any in-bound checksum flags for this packet.
217 	 */
218 	m->m_pkthdr.csum_flags = 0;
219 
220 	/*
221 	 * Everything checks out and so we can forward this packet.
222 	 * Modify the TTL and incrementally change the checksum.
223 	 *
224 	 * This method of adding the checksum works on either endian CPU.
225 	 * If htons() is inlined, all the arithmetic is folded; otherwise
226 	 * the htons()s are combined by CSE due to the __const__ attribute.
227 	 *
228 	 * Don't bother using HW checksumming here -- the incremental
229 	 * update is pretty fast.
230 	 */
231 	ip->ip_ttl -= IPTTLDEC;
232 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
233 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
234 	else
235 		ip->ip_sum += htons(IPTTLDEC << 8);
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(
270 	struct ipflow *ipf)
271 {
272 	ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
273 	ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
274 	ipstat.ips_forward += ipf->ipf_uses;
275 	ipstat.ips_fastforward += ipf->ipf_uses;
276 }
277 
278 static void
279 ipflow_free(
280 	struct ipflow *ipf)
281 {
282 	int s;
283 	/*
284 	 * Remove the flow from the hash table (at elevated IPL).
285 	 * Once it's off the list, we can deal with it at normal
286 	 * network IPL.
287 	 */
288 	s = splnet();
289 	IPFLOW_REMOVE(ipf);
290 	splx(s);
291 	ipflow_addstats(ipf);
292 	RTFREE(ipf->ipf_ro.ro_rt);
293 	ipflow_inuse--;
294 	pool_put(&ipflow_pool, ipf);
295 }
296 
297 struct ipflow *
298 ipflow_reap(
299 	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(
347 	void)
348 {
349 	struct ipflow *ipf, *next_ipf;
350 
351 	ipf = LIST_FIRST(&ipflowlist);
352 	while (ipf != NULL) {
353 		next_ipf = LIST_NEXT(ipf, ipf_list);
354 		if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer)) {
355 			ipflow_free(ipf);
356 		} else {
357 			ipf->ipf_last_uses = ipf->ipf_uses;
358 			ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
359 			ipstat.ips_forward += ipf->ipf_uses;
360 			ipstat.ips_fastforward += ipf->ipf_uses;
361 			ipf->ipf_uses = 0;
362 		}
363 		ipf = next_ipf;
364 	}
365 }
366 
367 void
368 ipflow_create(
369 	const struct route *ro,
370 	struct mbuf *m)
371 {
372 	const struct ip *const ip = mtod(m, struct ip *);
373 	struct ipflow *ipf;
374 	unsigned hash;
375 	int s;
376 
377 	/*
378 	 * Don't create cache entries for ICMP messages.
379 	 */
380 	if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP)
381 		return;
382 	/*
383 	 * See if an existing flow struct exists.  If so remove it from it's
384 	 * list and free the old route.  If not, try to malloc a new one
385 	 * (if we aren't at our limit).
386 	 */
387 	ipf = ipflow_lookup(ip);
388 	if (ipf == NULL) {
389 		if (ipflow_inuse >= ip_maxflows) {
390 			ipf = ipflow_reap(1);
391 		} else {
392 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
393 			if (ipf == NULL)
394 				return;
395 			ipflow_inuse++;
396 		}
397 		bzero((caddr_t) ipf, sizeof(*ipf));
398 	} else {
399 		s = splnet();
400 		IPFLOW_REMOVE(ipf);
401 		splx(s);
402 		ipflow_addstats(ipf);
403 		RTFREE(ipf->ipf_ro.ro_rt);
404 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
405 		ipf->ipf_errors = ipf->ipf_dropped = 0;
406 	}
407 
408 	/*
409 	 * Fill in the updated information.
410 	 */
411 	ipf->ipf_ro = *ro;
412 	ro->ro_rt->rt_refcnt++;
413 	ipf->ipf_dst = ip->ip_dst;
414 	ipf->ipf_src = ip->ip_src;
415 	ipf->ipf_tos = ip->ip_tos;
416 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
417 	ipf->ipf_start = time.tv_sec;
418 	/*
419 	 * Insert into the approriate bucket of the flow table.
420 	 */
421 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
422 	s = splnet();
423 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
424 	splx(s);
425 }
426