xref: /netbsd-src/sys/netinet/ip_flow.c (revision 6a493d6bc668897c91594964a732d38505b70cbb) 
1 /*	$NetBSD: ip_flow.c,v 1.60 2012/01/19 13:13:48 liamjfoy 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 1.60 2012/01/19 13:13:48 liamjfoy 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 };
79 
80 #define	IPFLOW_HASHBITS		6	/* should not be a multiple of 8 */
81 
82 static struct pool ipflow_pool;
83 
84 LIST_HEAD(ipflowhead, ipflow);
85 
86 #define	IPFLOW_TIMER		(5 * PR_SLOWHZ)
87 #define	IPFLOW_DEFAULT_HASHSIZE	(1 << IPFLOW_HASHBITS)
88 
89 static struct ipflowhead *ipflowtable = NULL;
90 static struct ipflowhead ipflowlist;
91 static int ipflow_inuse;
92 
93 #define	IPFLOW_INSERT(bucket, ipf) \
94 do { \
95 	LIST_INSERT_HEAD((bucket), (ipf), ipf_hash); \
96 	LIST_INSERT_HEAD(&ipflowlist, (ipf), ipf_list); \
97 } while (/*CONSTCOND*/ 0)
98 
99 #define	IPFLOW_REMOVE(ipf) \
100 do { \
101 	LIST_REMOVE((ipf), ipf_hash); \
102 	LIST_REMOVE((ipf), ipf_list); \
103 } while (/*CONSTCOND*/ 0)
104 
105 #ifndef IPFLOW_MAX
106 #define	IPFLOW_MAX		256
107 #endif
108 int ip_maxflows = IPFLOW_MAX;
109 int ip_hashsize = IPFLOW_DEFAULT_HASHSIZE;
110 
111 static size_t
112 ipflow_hash(const struct ip *ip)
113 {
114 	size_t hash = ip->ip_tos;
115 	size_t idx;
116 
117 	for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS) {
118 		hash += (ip->ip_dst.s_addr >> (32 - idx)) +
119 		    (ip->ip_src.s_addr >> idx);
120 	}
121 
122 	return hash & (ip_hashsize-1);
123 }
124 
125 static struct ipflow *
126 ipflow_lookup(const struct ip *ip)
127 {
128 	size_t hash;
129 	struct ipflow *ipf;
130 
131 	hash = ipflow_hash(ip);
132 
133 	LIST_FOREACH(ipf, &ipflowtable[hash], ipf_hash) {
134 		if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
135 		    && ip->ip_src.s_addr == ipf->ipf_src.s_addr
136 		    && ip->ip_tos == ipf->ipf_tos)
137 			break;
138 	}
139 	return ipf;
140 }
141 
142 void
143 ipflow_poolinit(void)
144 {
145 
146 	pool_init(&ipflow_pool, sizeof(struct ipflow), 0, 0, 0, "ipflowpl",
147 	    NULL, IPL_NET);
148 }
149 
150 int
151 ipflow_init(int table_size)
152 {
153 	struct ipflowhead *new_table;
154 	size_t i;
155 
156 	new_table = (struct ipflowhead *)malloc(sizeof(struct ipflowhead) *
157 	    table_size, M_RTABLE, M_NOWAIT);
158 
159 	if (new_table == NULL)
160 		return 1;
161 
162 	if (ipflowtable != NULL)
163 		free(ipflowtable, M_RTABLE);
164 
165 	ipflowtable = new_table;
166 	ip_hashsize = table_size;
167 
168 	LIST_INIT(&ipflowlist);
169 	for (i = 0; i < ip_hashsize; i++)
170 		LIST_INIT(&ipflowtable[i]);
171 
172 	return 0;
173 }
174 
175 int
176 ipflow_fastforward(struct mbuf *m)
177 {
178 	struct ip *ip;
179 	struct ip ip_store;
180 	struct ipflow *ipf;
181 	struct rtentry *rt;
182 	const struct sockaddr *dst;
183 	int error;
184 	int iplen;
185 
186 	/*
187 	 * Are we forwarding packets?  Big enough for an IP packet?
188 	 */
189 	if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip))
190 		return 0;
191 
192 	/*
193 	 * Was packet received as a link-level multicast or broadcast?
194 	 * If so, don't try to fast forward..
195 	 */
196 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
197 		return 0;
198 
199 	/*
200 	 * IP header with no option and valid version and length
201 	 */
202 	if (IP_HDR_ALIGNED_P(mtod(m, const void *)))
203 		ip = mtod(m, struct ip *);
204 	else {
205 		memcpy(&ip_store, mtod(m, const void *), sizeof(ip_store));
206 		ip = &ip_store;
207 	}
208 	iplen = ntohs(ip->ip_len);
209 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
210 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
211 		return 0;
212 	/*
213 	 * Find a flow.
214 	 */
215 	if ((ipf = ipflow_lookup(ip)) == NULL)
216 		return 0;
217 
218 	/*
219 	 * Verify the IP header checksum.
220 	 */
221 	switch (m->m_pkthdr.csum_flags &
222 		((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_IPv4) |
223 		 M_CSUM_IPv4_BAD)) {
224 	case M_CSUM_IPv4|M_CSUM_IPv4_BAD:
225 		return (0);
226 
227 	case M_CSUM_IPv4:
228 		/* Checksum was okay. */
229 		break;
230 
231 	default:
232 		/* Must compute it ourselves. */
233 		if (in_cksum(m, sizeof(struct ip)) != 0)
234 			return (0);
235 		break;
236 	}
237 
238 	/*
239 	 * Route and interface still up?
240 	 */
241 	if ((rt = rtcache_validate(&ipf->ipf_ro)) == NULL ||
242 	    (rt->rt_ifp->if_flags & IFF_UP) == 0)
243 		return 0;
244 
245 	/*
246 	 * Packet size OK?  TTL?
247 	 */
248 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
249 		return 0;
250 
251 	/*
252 	 * Clear any in-bound checksum flags for this packet.
253 	 */
254 	m->m_pkthdr.csum_flags = 0;
255 
256 	/*
257 	 * Everything checks out and so we can forward this packet.
258 	 * Modify the TTL and incrementally change the checksum.
259 	 *
260 	 * This method of adding the checksum works on either endian CPU.
261 	 * If htons() is inlined, all the arithmetic is folded; otherwise
262 	 * the htons()s are combined by CSE due to the const attribute.
263 	 *
264 	 * Don't bother using HW checksumming here -- the incremental
265 	 * update is pretty fast.
266 	 */
267 	ip->ip_ttl -= IPTTLDEC;
268 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
269 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
270 	else
271 		ip->ip_sum += htons(IPTTLDEC << 8);
272 
273 	/*
274 	 * Done modifying the header; copy it back, if necessary.
275 	 *
276 	 * XXX Use m_copyback_cow(9) here? --dyoung
277 	 */
278 	if (IP_HDR_ALIGNED_P(mtod(m, void *)) == 0)
279 		memcpy(mtod(m, void *), &ip_store, sizeof(ip_store));
280 
281 	/*
282 	 * Trim the packet in case it's too long..
283 	 */
284 	if (m->m_pkthdr.len > iplen) {
285 		if (m->m_len == m->m_pkthdr.len) {
286 			m->m_len = iplen;
287 			m->m_pkthdr.len = iplen;
288 		} else
289 			m_adj(m, iplen - m->m_pkthdr.len);
290 	}
291 
292 	/*
293 	 * Send the packet on it's way.  All we can get back is ENOBUFS
294 	 */
295 	ipf->ipf_uses++;
296 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
297 
298 	if (rt->rt_flags & RTF_GATEWAY)
299 		dst = rt->rt_gateway;
300 	else
301 		dst = rtcache_getdst(&ipf->ipf_ro);
302 
303 	KERNEL_LOCK(1, NULL);
304 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
305 		if (error == ENOBUFS)
306 			ipf->ipf_dropped++;
307 		else
308 			ipf->ipf_errors++;
309 	}
310 	KERNEL_UNLOCK_ONE(NULL);
311 	return 1;
312 }
313 
314 static void
315 ipflow_addstats(struct ipflow *ipf)
316 {
317 	struct rtentry *rt;
318 	uint64_t *ips;
319 
320 	if ((rt = rtcache_validate(&ipf->ipf_ro)) != NULL)
321 		rt->rt_use += ipf->ipf_uses;
322 
323 	ips = IP_STAT_GETREF();
324 	ips[IP_STAT_CANTFORWARD] += ipf->ipf_errors + ipf->ipf_dropped;
325 	ips[IP_STAT_TOTAL] += ipf->ipf_uses;
326 	ips[IP_STAT_FORWARD] += ipf->ipf_uses;
327 	ips[IP_STAT_FASTFORWARD] += ipf->ipf_uses;
328 	IP_STAT_PUTREF();
329 }
330 
331 static void
332 ipflow_free(struct ipflow *ipf)
333 {
334 	int s;
335 	/*
336 	 * Remove the flow from the hash table (at elevated IPL).
337 	 * Once it's off the list, we can deal with it at normal
338 	 * network IPL.
339 	 */
340 	s = splnet();
341 	IPFLOW_REMOVE(ipf);
342 	splx(s);
343 	ipflow_addstats(ipf);
344 	rtcache_free(&ipf->ipf_ro);
345 	ipflow_inuse--;
346 	s = splnet();
347 	pool_put(&ipflow_pool, ipf);
348 	splx(s);
349 }
350 
351 static struct ipflow *
352 ipflow_reap(bool just_one)
353 {
354 	while (just_one || ipflow_inuse > ip_maxflows) {
355 		struct ipflow *ipf, *maybe_ipf = NULL;
356 		int s;
357 
358 		ipf = LIST_FIRST(&ipflowlist);
359 		while (ipf != NULL) {
360 			/*
361 			 * If this no longer points to a valid route
362 			 * reclaim it.
363 			 */
364 			if (rtcache_validate(&ipf->ipf_ro) == NULL)
365 				goto done;
366 			/*
367 			 * choose the one that's been least recently
368 			 * used or has had the least uses in the
369 			 * last 1.5 intervals.
370 			 */
371 			if (maybe_ipf == NULL ||
372 			    ipf->ipf_timer < maybe_ipf->ipf_timer ||
373 			    (ipf->ipf_timer == maybe_ipf->ipf_timer &&
374 			     ipf->ipf_last_uses + ipf->ipf_uses <
375 			         maybe_ipf->ipf_last_uses +
376 			         maybe_ipf->ipf_uses))
377 				maybe_ipf = ipf;
378 			ipf = LIST_NEXT(ipf, ipf_list);
379 		}
380 		ipf = maybe_ipf;
381 	    done:
382 		/*
383 		 * Remove the entry from the flow table.
384 		 */
385 		s = splnet();
386 		IPFLOW_REMOVE(ipf);
387 		splx(s);
388 		ipflow_addstats(ipf);
389 		rtcache_free(&ipf->ipf_ro);
390 		if (just_one)
391 			return ipf;
392 		pool_put(&ipflow_pool, ipf);
393 		ipflow_inuse--;
394 	}
395 	return NULL;
396 }
397 
398 void
399 ipflow_prune(void)
400 {
401 
402 	(void) ipflow_reap(false);
403 }
404 
405 void
406 ipflow_slowtimo(void)
407 {
408 	struct rtentry *rt;
409 	struct ipflow *ipf, *next_ipf;
410 	uint64_t *ips;
411 
412 	mutex_enter(softnet_lock);
413 	KERNEL_LOCK(1, NULL);
414 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
415 		next_ipf = LIST_NEXT(ipf, ipf_list);
416 		if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer) ||
417 		    (rt = rtcache_validate(&ipf->ipf_ro)) == NULL) {
418 			ipflow_free(ipf);
419 		} else {
420 			ipf->ipf_last_uses = ipf->ipf_uses;
421 			rt->rt_use += ipf->ipf_uses;
422 			ips = IP_STAT_GETREF();
423 			ips[IP_STAT_TOTAL] += ipf->ipf_uses;
424 			ips[IP_STAT_FORWARD] += ipf->ipf_uses;
425 			ips[IP_STAT_FASTFORWARD] += ipf->ipf_uses;
426 			IP_STAT_PUTREF();
427 			ipf->ipf_uses = 0;
428 		}
429 	}
430 	KERNEL_UNLOCK_ONE(NULL);
431 	mutex_exit(softnet_lock);
432 }
433 
434 void
435 ipflow_create(const struct route *ro, struct mbuf *m)
436 {
437 	const struct ip *const ip = mtod(m, const struct ip *);
438 	struct ipflow *ipf;
439 	size_t hash;
440 	int s;
441 
442 	/*
443 	 * Don't create cache entries for ICMP messages.
444 	 */
445 	if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP)
446 		return;
447 	/*
448 	 * See if an existing flow struct exists.  If so remove it from it's
449 	 * list and free the old route.  If not, try to malloc a new one
450 	 * (if we aren't at our limit).
451 	 */
452 	ipf = ipflow_lookup(ip);
453 	if (ipf == NULL) {
454 		if (ipflow_inuse >= ip_maxflows) {
455 			ipf = ipflow_reap(true);
456 		} else {
457 			s = splnet();
458 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
459 			splx(s);
460 			if (ipf == NULL)
461 				return;
462 			ipflow_inuse++;
463 		}
464 		memset(ipf, 0, sizeof(*ipf));
465 	} else {
466 		s = splnet();
467 		IPFLOW_REMOVE(ipf);
468 		splx(s);
469 		ipflow_addstats(ipf);
470 		rtcache_free(&ipf->ipf_ro);
471 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
472 		ipf->ipf_errors = ipf->ipf_dropped = 0;
473 	}
474 
475 	/*
476 	 * Fill in the updated information.
477 	 */
478 	rtcache_copy(&ipf->ipf_ro, ro);
479 	ipf->ipf_dst = ip->ip_dst;
480 	ipf->ipf_src = ip->ip_src;
481 	ipf->ipf_tos = ip->ip_tos;
482 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
483 
484 	/*
485 	 * Insert into the approriate bucket of the flow table.
486 	 */
487 	hash = ipflow_hash(ip);
488 	s = splnet();
489 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
490 	splx(s);
491 }
492 
493 int
494 ipflow_invalidate_all(int new_size)
495 {
496 	struct ipflow *ipf, *next_ipf;
497 	int s, error;
498 
499 	error = 0;
500 	s = splnet();
501 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
502 		next_ipf = LIST_NEXT(ipf, ipf_list);
503 		ipflow_free(ipf);
504 	}
505 
506 	if (new_size)
507 		error = ipflow_init(new_size);
508 	splx(s);
509 
510 	return error;
511 }
512