xref: /netbsd-src/sys/netinet/ip_flow.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d) 
1 /*	$NetBSD: ip_flow.c,v 1.59 2010/04/01 00:24:41 tls 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: ip_flow.c,v 1.59 2010/04/01 00:24:41 tls Exp $");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/malloc.h>
38 #include <sys/mbuf.h>
39 #include <sys/domain.h>
40 #include <sys/protosw.h>
41 #include <sys/socket.h>
42 #include <sys/socketvar.h>
43 #include <sys/errno.h>
44 #include <sys/time.h>
45 #include <sys/kernel.h>
46 #include <sys/pool.h>
47 #include <sys/sysctl.h>
48 
49 #include <net/if.h>
50 #include <net/if_dl.h>
51 #include <net/route.h>
52 #include <net/pfil.h>
53 
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/ip.h>
57 #include <netinet/in_pcb.h>
58 #include <netinet/in_var.h>
59 #include <netinet/ip_var.h>
60 #include <netinet/ip_private.h>
61 
62 /*
63  * Similar code is very well commented in netinet6/ip6_flow.c
64  */
65 
66 struct ipflow {
67 	LIST_ENTRY(ipflow) ipf_list;	/* next in active list */
68 	LIST_ENTRY(ipflow) ipf_hash;	/* next ipflow in bucket */
69 	struct in_addr ipf_dst;		/* destination address */
70 	struct in_addr ipf_src;		/* source address */
71 	uint8_t ipf_tos;		/* type-of-service */
72 	struct route ipf_ro;		/* associated route entry */
73 	u_long ipf_uses;		/* number of uses in this period */
74 	u_long ipf_last_uses;		/* number of uses in last period */
75 	u_long ipf_dropped;		/* ENOBUFS retured by if_output */
76 	u_long ipf_errors;		/* other errors returned by if_output */
77 	u_int ipf_timer;		/* lifetime timer */
78 	time_t ipf_start;		/* creation time */
79 };
80 
81 #define	IPFLOW_HASHBITS		6	/* should not be a multiple of 8 */
82 
83 static struct pool ipflow_pool;
84 
85 LIST_HEAD(ipflowhead, ipflow);
86 
87 #define	IPFLOW_TIMER		(5 * PR_SLOWHZ)
88 #define	IPFLOW_DEFAULT_HASHSIZE	(1 << IPFLOW_HASHBITS)
89 
90 static struct ipflowhead *ipflowtable = NULL;
91 static struct ipflowhead ipflowlist;
92 static int ipflow_inuse;
93 
94 #define	IPFLOW_INSERT(bucket, ipf) \
95 do { \
96 	LIST_INSERT_HEAD((bucket), (ipf), ipf_hash); \
97 	LIST_INSERT_HEAD(&ipflowlist, (ipf), ipf_list); \
98 } while (/*CONSTCOND*/ 0)
99 
100 #define	IPFLOW_REMOVE(ipf) \
101 do { \
102 	LIST_REMOVE((ipf), ipf_hash); \
103 	LIST_REMOVE((ipf), ipf_list); \
104 } while (/*CONSTCOND*/ 0)
105 
106 #ifndef IPFLOW_MAX
107 #define	IPFLOW_MAX		256
108 #endif
109 int ip_maxflows = IPFLOW_MAX;
110 int ip_hashsize = IPFLOW_DEFAULT_HASHSIZE;
111 
112 static size_t
113 ipflow_hash(const struct ip *ip)
114 {
115 	size_t hash = ip->ip_tos;
116 	size_t idx;
117 
118 	for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS) {
119 		hash += (ip->ip_dst.s_addr >> (32 - idx)) +
120 		    (ip->ip_src.s_addr >> idx);
121 	}
122 
123 	return hash & (ip_hashsize-1);
124 }
125 
126 static struct ipflow *
127 ipflow_lookup(const struct ip *ip)
128 {
129 	size_t hash;
130 	struct ipflow *ipf;
131 
132 	hash = ipflow_hash(ip);
133 
134 	LIST_FOREACH(ipf, &ipflowtable[hash], ipf_hash) {
135 		if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
136 		    && ip->ip_src.s_addr == ipf->ipf_src.s_addr
137 		    && ip->ip_tos == ipf->ipf_tos)
138 			break;
139 	}
140 	return ipf;
141 }
142 
143 void
144 ipflow_poolinit(void)
145 {
146 
147 	pool_init(&ipflow_pool, sizeof(struct ipflow), 0, 0, 0, "ipflowpl",
148 	    NULL, IPL_NET);
149 }
150 
151 int
152 ipflow_init(int table_size)
153 {
154 	struct ipflowhead *new_table;
155 	size_t i;
156 
157 	new_table = (struct ipflowhead *)malloc(sizeof(struct ipflowhead) *
158 	    table_size, M_RTABLE, M_NOWAIT);
159 
160 	if (new_table == NULL)
161 		return 1;
162 
163 	if (ipflowtable != NULL)
164 		free(ipflowtable, M_RTABLE);
165 
166 	ipflowtable = new_table;
167 	ip_hashsize = table_size;
168 
169 	LIST_INIT(&ipflowlist);
170 	for (i = 0; i < ip_hashsize; i++)
171 		LIST_INIT(&ipflowtable[i]);
172 
173 	return 0;
174 }
175 
176 int
177 ipflow_fastforward(struct mbuf *m)
178 {
179 	struct ip *ip;
180 	struct ip ip_store;
181 	struct ipflow *ipf;
182 	struct rtentry *rt;
183 	const struct sockaddr *dst;
184 	int error;
185 	int iplen;
186 
187 	/*
188 	 * Are we forwarding packets?  Big enough for an IP packet?
189 	 */
190 	if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip))
191 		return 0;
192 
193 	/*
194 	 * Was packet received as a link-level multicast or broadcast?
195 	 * If so, don't try to fast forward..
196 	 */
197 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
198 		return 0;
199 
200 	/*
201 	 * IP header with no option and valid version and length
202 	 */
203 	if (IP_HDR_ALIGNED_P(mtod(m, const void *)))
204 		ip = mtod(m, struct ip *);
205 	else {
206 		memcpy(&ip_store, mtod(m, const void *), sizeof(ip_store));
207 		ip = &ip_store;
208 	}
209 	iplen = ntohs(ip->ip_len);
210 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
211 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
212 		return 0;
213 	/*
214 	 * Find a flow.
215 	 */
216 	if ((ipf = ipflow_lookup(ip)) == NULL)
217 		return 0;
218 
219 	/*
220 	 * Verify the IP header checksum.
221 	 */
222 	switch (m->m_pkthdr.csum_flags &
223 		((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_IPv4) |
224 		 M_CSUM_IPv4_BAD)) {
225 	case M_CSUM_IPv4|M_CSUM_IPv4_BAD:
226 		return (0);
227 
228 	case M_CSUM_IPv4:
229 		/* Checksum was okay. */
230 		break;
231 
232 	default:
233 		/* Must compute it ourselves. */
234 		if (in_cksum(m, sizeof(struct ip)) != 0)
235 			return (0);
236 		break;
237 	}
238 
239 	/*
240 	 * Route and interface still up?
241 	 */
242 	if ((rt = rtcache_validate(&ipf->ipf_ro)) == NULL ||
243 	    (rt->rt_ifp->if_flags & IFF_UP) == 0)
244 		return 0;
245 
246 	/*
247 	 * Packet size OK?  TTL?
248 	 */
249 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
250 		return 0;
251 
252 	/*
253 	 * Clear any in-bound checksum flags for this packet.
254 	 */
255 	m->m_pkthdr.csum_flags = 0;
256 
257 	/*
258 	 * Everything checks out and so we can forward this packet.
259 	 * Modify the TTL and incrementally change the checksum.
260 	 *
261 	 * This method of adding the checksum works on either endian CPU.
262 	 * If htons() is inlined, all the arithmetic is folded; otherwise
263 	 * the htons()s are combined by CSE due to the const attribute.
264 	 *
265 	 * Don't bother using HW checksumming here -- the incremental
266 	 * update is pretty fast.
267 	 */
268 	ip->ip_ttl -= IPTTLDEC;
269 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
270 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
271 	else
272 		ip->ip_sum += htons(IPTTLDEC << 8);
273 
274 	/*
275 	 * Done modifying the header; copy it back, if necessary.
276 	 *
277 	 * XXX Use m_copyback_cow(9) here? --dyoung
278 	 */
279 	if (IP_HDR_ALIGNED_P(mtod(m, void *)) == 0)
280 		memcpy(mtod(m, void *), &ip_store, sizeof(ip_store));
281 
282 	/*
283 	 * Trim the packet in case it's too long..
284 	 */
285 	if (m->m_pkthdr.len > iplen) {
286 		if (m->m_len == m->m_pkthdr.len) {
287 			m->m_len = iplen;
288 			m->m_pkthdr.len = iplen;
289 		} else
290 			m_adj(m, iplen - m->m_pkthdr.len);
291 	}
292 
293 	/*
294 	 * Send the packet on it's way.  All we can get back is ENOBUFS
295 	 */
296 	ipf->ipf_uses++;
297 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
298 
299 	if (rt->rt_flags & RTF_GATEWAY)
300 		dst = rt->rt_gateway;
301 	else
302 		dst = rtcache_getdst(&ipf->ipf_ro);
303 
304 	KERNEL_LOCK(1, NULL);
305 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
306 		if (error == ENOBUFS)
307 			ipf->ipf_dropped++;
308 		else
309 			ipf->ipf_errors++;
310 	}
311 	KERNEL_UNLOCK_ONE(NULL);
312 	return 1;
313 }
314 
315 static void
316 ipflow_addstats(struct ipflow *ipf)
317 {
318 	struct rtentry *rt;
319 	uint64_t *ips;
320 
321 	if ((rt = rtcache_validate(&ipf->ipf_ro)) != NULL)
322 		rt->rt_use += ipf->ipf_uses;
323 
324 	ips = IP_STAT_GETREF();
325 	ips[IP_STAT_CANTFORWARD] += ipf->ipf_errors + ipf->ipf_dropped;
326 	ips[IP_STAT_TOTAL] += ipf->ipf_uses;
327 	ips[IP_STAT_FORWARD] += ipf->ipf_uses;
328 	ips[IP_STAT_FASTFORWARD] += ipf->ipf_uses;
329 	IP_STAT_PUTREF();
330 }
331 
332 static void
333 ipflow_free(struct ipflow *ipf)
334 {
335 	int s;
336 	/*
337 	 * Remove the flow from the hash table (at elevated IPL).
338 	 * Once it's off the list, we can deal with it at normal
339 	 * network IPL.
340 	 */
341 	s = splnet();
342 	IPFLOW_REMOVE(ipf);
343 	splx(s);
344 	ipflow_addstats(ipf);
345 	rtcache_free(&ipf->ipf_ro);
346 	ipflow_inuse--;
347 	s = splnet();
348 	pool_put(&ipflow_pool, ipf);
349 	splx(s);
350 }
351 
352 static struct ipflow *
353 ipflow_reap(bool just_one)
354 {
355 	while (just_one || ipflow_inuse > ip_maxflows) {
356 		struct ipflow *ipf, *maybe_ipf = NULL;
357 		int s;
358 
359 		ipf = LIST_FIRST(&ipflowlist);
360 		while (ipf != NULL) {
361 			/*
362 			 * If this no longer points to a valid route
363 			 * reclaim it.
364 			 */
365 			if (rtcache_validate(&ipf->ipf_ro) == NULL)
366 				goto done;
367 			/*
368 			 * choose the one that's been least recently
369 			 * used or has had the least uses in the
370 			 * last 1.5 intervals.
371 			 */
372 			if (maybe_ipf == NULL ||
373 			    ipf->ipf_timer < maybe_ipf->ipf_timer ||
374 			    (ipf->ipf_timer == maybe_ipf->ipf_timer &&
375 			     ipf->ipf_last_uses + ipf->ipf_uses <
376 			         maybe_ipf->ipf_last_uses +
377 			         maybe_ipf->ipf_uses))
378 				maybe_ipf = ipf;
379 			ipf = LIST_NEXT(ipf, ipf_list);
380 		}
381 		ipf = maybe_ipf;
382 	    done:
383 		/*
384 		 * Remove the entry from the flow table.
385 		 */
386 		s = splnet();
387 		IPFLOW_REMOVE(ipf);
388 		splx(s);
389 		ipflow_addstats(ipf);
390 		rtcache_free(&ipf->ipf_ro);
391 		if (just_one)
392 			return ipf;
393 		pool_put(&ipflow_pool, ipf);
394 		ipflow_inuse--;
395 	}
396 	return NULL;
397 }
398 
399 void
400 ipflow_prune(void)
401 {
402 
403 	(void) ipflow_reap(false);
404 }
405 
406 void
407 ipflow_slowtimo(void)
408 {
409 	struct rtentry *rt;
410 	struct ipflow *ipf, *next_ipf;
411 	uint64_t *ips;
412 
413 	mutex_enter(softnet_lock);
414 	KERNEL_LOCK(1, NULL);
415 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
416 		next_ipf = LIST_NEXT(ipf, ipf_list);
417 		if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer) ||
418 		    (rt = rtcache_validate(&ipf->ipf_ro)) == NULL) {
419 			ipflow_free(ipf);
420 		} else {
421 			ipf->ipf_last_uses = ipf->ipf_uses;
422 			rt->rt_use += ipf->ipf_uses;
423 			ips = IP_STAT_GETREF();
424 			ips[IP_STAT_TOTAL] += ipf->ipf_uses;
425 			ips[IP_STAT_FORWARD] += ipf->ipf_uses;
426 			ips[IP_STAT_FASTFORWARD] += ipf->ipf_uses;
427 			IP_STAT_PUTREF();
428 			ipf->ipf_uses = 0;
429 		}
430 	}
431 	KERNEL_UNLOCK_ONE(NULL);
432 	mutex_exit(softnet_lock);
433 }
434 
435 void
436 ipflow_create(const struct route *ro, struct mbuf *m)
437 {
438 	const struct ip *const ip = mtod(m, const struct ip *);
439 	struct ipflow *ipf;
440 	size_t hash;
441 	int s;
442 
443 	/*
444 	 * Don't create cache entries for ICMP messages.
445 	 */
446 	if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP)
447 		return;
448 	/*
449 	 * See if an existing flow struct exists.  If so remove it from it's
450 	 * list and free the old route.  If not, try to malloc a new one
451 	 * (if we aren't at our limit).
452 	 */
453 	ipf = ipflow_lookup(ip);
454 	if (ipf == NULL) {
455 		if (ipflow_inuse >= ip_maxflows) {
456 			ipf = ipflow_reap(true);
457 		} else {
458 			s = splnet();
459 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
460 			splx(s);
461 			if (ipf == NULL)
462 				return;
463 			ipflow_inuse++;
464 		}
465 		memset(ipf, 0, sizeof(*ipf));
466 	} else {
467 		s = splnet();
468 		IPFLOW_REMOVE(ipf);
469 		splx(s);
470 		ipflow_addstats(ipf);
471 		rtcache_free(&ipf->ipf_ro);
472 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
473 		ipf->ipf_errors = ipf->ipf_dropped = 0;
474 	}
475 
476 	/*
477 	 * Fill in the updated information.
478 	 */
479 	rtcache_copy(&ipf->ipf_ro, ro);
480 	ipf->ipf_dst = ip->ip_dst;
481 	ipf->ipf_src = ip->ip_src;
482 	ipf->ipf_tos = ip->ip_tos;
483 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
484 	ipf->ipf_start = time_uptime;
485 	/*
486 	 * Insert into the approriate bucket of the flow table.
487 	 */
488 	hash = ipflow_hash(ip);
489 	s = splnet();
490 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
491 	splx(s);
492 }
493 
494 int
495 ipflow_invalidate_all(int new_size)
496 {
497 	struct ipflow *ipf, *next_ipf;
498 	int s, error;
499 
500 	error = 0;
501 	s = splnet();
502 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
503 		next_ipf = LIST_NEXT(ipf, ipf_list);
504 		ipflow_free(ipf);
505 	}
506 
507 	if (new_size)
508 		error = ipflow_init(new_size);
509 	splx(s);
510 
511 	return error;
512 }
513