xref: /netbsd-src/sys/netinet/ip_flow.c (revision 5b84b3983f71fd20a534cfa5d1556623a8aaa717)
1 /*	$NetBSD: ip_flow.c,v 1.29 2005/02/03 22:43:34 perry 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 1.29 2005/02/03 22:43:34 perry 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 	ipf = LIST_FIRST(&ipflowtable[hash]);
115 	while (ipf != NULL) {
116 		if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
117 		    && ip->ip_src.s_addr == ipf->ipf_src.s_addr
118 		    && ip->ip_tos == ipf->ipf_tos)
119 			break;
120 		ipf = LIST_NEXT(ipf, ipf_hash);
121 	}
122 	return ipf;
123 }
124 
125 void
126 ipflow_init(void)
127 {
128 	int i;
129 
130 	LIST_INIT(&ipflowlist);
131 	for (i = 0; i < IPFLOW_HASHSIZE; i++)
132 		LIST_INIT(&ipflowtable[i]);
133 }
134 
135 int
136 ipflow_fastforward(struct mbuf *m)
137 {
138 	struct ip *ip, ip_store;
139 	struct ipflow *ipf;
140 	struct rtentry *rt;
141 	struct sockaddr *dst;
142 	int error;
143 	int iplen;
144 
145 	/*
146 	 * Are we forwarding packets?  Big enough for an IP packet?
147 	 */
148 	if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip))
149 		return 0;
150 
151 	/*
152 	 * Was packet received as a link-level multicast or broadcast?
153 	 * If so, don't try to fast forward..
154 	 */
155 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
156 		return 0;
157 
158 	/*
159 	 * IP header with no option and valid version and length
160 	 */
161 	if (IP_HDR_ALIGNED_P(mtod(m, caddr_t)))
162 		ip = mtod(m, struct ip *);
163 	else {
164 		memcpy(&ip_store, mtod(m, caddr_t), sizeof(ip_store));
165 		ip = &ip_store;
166 	}
167 	iplen = ntohs(ip->ip_len);
168 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
169 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
170 		return 0;
171 	/*
172 	 * Find a flow.
173 	 */
174 	if ((ipf = ipflow_lookup(ip)) == NULL)
175 		return 0;
176 
177 	/*
178 	 * Verify the IP header checksum.
179 	 */
180 	switch (m->m_pkthdr.csum_flags &
181 		((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_IPv4) |
182 		 M_CSUM_IPv4_BAD)) {
183 	case M_CSUM_IPv4|M_CSUM_IPv4_BAD:
184 		return (0);
185 
186 	case M_CSUM_IPv4:
187 		/* Checksum was okay. */
188 		break;
189 
190 	default:
191 		/* Must compute it ourselves. */
192 		if (in_cksum(m, sizeof(struct ip)) != 0)
193 			return (0);
194 		break;
195 	}
196 
197 	/*
198 	 * Route and interface still up?
199 	 */
200 	rt = ipf->ipf_ro.ro_rt;
201 	if ((rt->rt_flags & RTF_UP) == 0 ||
202 	    (rt->rt_ifp->if_flags & IFF_UP) == 0)
203 		return 0;
204 
205 	/*
206 	 * Packet size OK?  TTL?
207 	 */
208 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
209 		return 0;
210 
211 	/*
212 	 * Clear any in-bound checksum flags for this packet.
213 	 */
214 	m->m_pkthdr.csum_flags = 0;
215 
216 	/*
217 	 * Everything checks out and so we can forward this packet.
218 	 * Modify the TTL and incrementally change the checksum.
219 	 *
220 	 * This method of adding the checksum works on either endian CPU.
221 	 * If htons() is inlined, all the arithmetic is folded; otherwise
222 	 * the htons()s are combined by CSE due to the __const__ attribute.
223 	 *
224 	 * Don't bother using HW checksumming here -- the incremental
225 	 * update is pretty fast.
226 	 */
227 	ip->ip_ttl -= IPTTLDEC;
228 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
229 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
230 	else
231 		ip->ip_sum += htons(IPTTLDEC << 8);
232 
233 	/*
234 	 * Done modifying the header; copy it back, if necessary.
235 	 */
236 	if (IP_HDR_ALIGNED_P(mtod(m, caddr_t)) == 0)
237 		memcpy(mtod(m, caddr_t), &ip_store, sizeof(ip_store));
238 
239 	/*
240 	 * Trim the packet in case it's too long..
241 	 */
242 	if (m->m_pkthdr.len > iplen) {
243 		if (m->m_len == m->m_pkthdr.len) {
244 			m->m_len = iplen;
245 			m->m_pkthdr.len = iplen;
246 		} else
247 			m_adj(m, iplen - m->m_pkthdr.len);
248 	}
249 
250 	/*
251 	 * Send the packet on it's way.  All we can get back is ENOBUFS
252 	 */
253 	ipf->ipf_uses++;
254 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
255 
256 	if (rt->rt_flags & RTF_GATEWAY)
257 		dst = rt->rt_gateway;
258 	else
259 		dst = &ipf->ipf_ro.ro_dst;
260 
261 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
262 		if (error == ENOBUFS)
263 			ipf->ipf_dropped++;
264 		else
265 			ipf->ipf_errors++;
266 	}
267 	return 1;
268 }
269 
270 static void
271 ipflow_addstats(struct ipflow *ipf)
272 {
273 	ipf->ipf_ro.ro_rt->rt_use += ipf->ipf_uses;
274 	ipstat.ips_cantforward += ipf->ipf_errors + ipf->ipf_dropped;
275 	ipstat.ips_forward += ipf->ipf_uses;
276 	ipstat.ips_fastforward += ipf->ipf_uses;
277 }
278 
279 static void
280 ipflow_free(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(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(void)
346 {
347 	struct ipflow *ipf, *next_ipf;
348 
349 	ipf = LIST_FIRST(&ipflowlist);
350 	while (ipf != NULL) {
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_forward += ipf->ipf_uses;
358 			ipstat.ips_fastforward += ipf->ipf_uses;
359 			ipf->ipf_uses = 0;
360 		}
361 		ipf = next_ipf;
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 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
389 			if (ipf == NULL)
390 				return;
391 			ipflow_inuse++;
392 		}
393 		bzero((caddr_t) ipf, sizeof(*ipf));
394 	} else {
395 		s = splnet();
396 		IPFLOW_REMOVE(ipf);
397 		splx(s);
398 		ipflow_addstats(ipf);
399 		RTFREE(ipf->ipf_ro.ro_rt);
400 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
401 		ipf->ipf_errors = ipf->ipf_dropped = 0;
402 	}
403 
404 	/*
405 	 * Fill in the updated information.
406 	 */
407 	ipf->ipf_ro = *ro;
408 	ro->ro_rt->rt_refcnt++;
409 	ipf->ipf_dst = ip->ip_dst;
410 	ipf->ipf_src = ip->ip_src;
411 	ipf->ipf_tos = ip->ip_tos;
412 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
413 	ipf->ipf_start = time.tv_sec;
414 	/*
415 	 * Insert into the approriate bucket of the flow table.
416 	 */
417 	hash = ipflow_hash(ip->ip_dst, ip->ip_src, ip->ip_tos);
418 	s = splnet();
419 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
420 	splx(s);
421 }
422 
423 void
424 ipflow_invalidate_all(void)
425 {
426 	struct ipflow *ipf, *next_ipf;
427 	int s;
428 
429 	s = splnet();
430 	ipf = LIST_FIRST(&ipflowlist);
431 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
432 		next_ipf = LIST_NEXT(ipf, ipf_list);
433 		ipflow_free(ipf);
434 	}
435 	splx(s);
436 }
437