Home | History | Annotate | Line # | Download | only in netinet
ip_flow.c revision 1.62 
      1 /*	$NetBSD: ip_flow.c,v 1.62 2014/03/19 10:54:20 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 1.62 2014/03/19 10:54:20 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 #define	IPFLOW_HASHBITS		6	/* should not be a multiple of 8 */
     67 
     68 static struct pool ipflow_pool;
     69 
     70 LIST_HEAD(ipflowhead, ipflow);
     71 
     72 #define	IPFLOW_TIMER		(5 * PR_SLOWHZ)
     73 #define	IPFLOW_DEFAULT_HASHSIZE	(1 << IPFLOW_HASHBITS)
     74 
     75 static struct ipflowhead *ipflowtable = NULL;
     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 int ip_hashsize = IPFLOW_DEFAULT_HASHSIZE;
     96 
     97 static size_t
     98 ipflow_hash(const struct ip *ip)
     99 {
    100 	size_t hash = ip->ip_tos;
    101 	size_t idx;
    102 
    103 	for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS) {
    104 		hash += (ip->ip_dst.s_addr >> (32 - idx)) +
    105 		    (ip->ip_src.s_addr >> idx);
    106 	}
    107 
    108 	return hash & (ip_hashsize-1);
    109 }
    110 
    111 static struct ipflow *
    112 ipflow_lookup(const struct ip *ip)
    113 {
    114 	size_t hash;
    115 	struct ipflow *ipf;
    116 
    117 	hash = ipflow_hash(ip);
    118 
    119 	LIST_FOREACH(ipf, &ipflowtable[hash], ipf_hash) {
    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 	}
    125 	return ipf;
    126 }
    127 
    128 void
    129 ipflow_poolinit(void)
    130 {
    131 
    132 	pool_init(&ipflow_pool, sizeof(struct ipflow), 0, 0, 0, "ipflowpl",
    133 	    NULL, IPL_NET);
    134 }
    135 
    136 int
    137 ipflow_init(int table_size)
    138 {
    139 	struct ipflowhead *new_table;
    140 	size_t i;
    141 
    142 	new_table = (struct ipflowhead *)malloc(sizeof(struct ipflowhead) *
    143 	    table_size, M_RTABLE, M_NOWAIT);
    144 
    145 	if (new_table == NULL)
    146 		return 1;
    147 
    148 	if (ipflowtable != NULL)
    149 		free(ipflowtable, M_RTABLE);
    150 
    151 	ipflowtable = new_table;
    152 	ip_hashsize = table_size;
    153 
    154 	LIST_INIT(&ipflowlist);
    155 	for (i = 0; i < ip_hashsize; i++)
    156 		LIST_INIT(&ipflowtable[i]);
    157 
    158 	return 0;
    159 }
    160 
    161 int
    162 ipflow_fastforward(struct mbuf *m)
    163 {
    164 	struct ip *ip;
    165 	struct ip ip_store;
    166 	struct ipflow *ipf;
    167 	struct rtentry *rt;
    168 	const struct sockaddr *dst;
    169 	int error;
    170 	int iplen;
    171 
    172 	/*
    173 	 * Are we forwarding packets?  Big enough for an IP packet?
    174 	 */
    175 	if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip))
    176 		return 0;
    177 
    178 	/*
    179 	 * Was packet received as a link-level multicast or broadcast?
    180 	 * If so, don't try to fast forward..
    181 	 */
    182 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
    183 		return 0;
    184 
    185 	/*
    186 	 * IP header with no option and valid version and length
    187 	 */
    188 	if (IP_HDR_ALIGNED_P(mtod(m, const void *)))
    189 		ip = mtod(m, struct ip *);
    190 	else {
    191 		memcpy(&ip_store, mtod(m, const void *), sizeof(ip_store));
    192 		ip = &ip_store;
    193 	}
    194 	iplen = ntohs(ip->ip_len);
    195 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
    196 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
    197 		return 0;
    198 	/*
    199 	 * Find a flow.
    200 	 */
    201 	if ((ipf = ipflow_lookup(ip)) == NULL)
    202 		return 0;
    203 
    204 	/*
    205 	 * Verify the IP header checksum.
    206 	 */
    207 	switch (m->m_pkthdr.csum_flags &
    208 		((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_IPv4) |
    209 		 M_CSUM_IPv4_BAD)) {
    210 	case M_CSUM_IPv4|M_CSUM_IPv4_BAD:
    211 		return (0);
    212 
    213 	case M_CSUM_IPv4:
    214 		/* Checksum was okay. */
    215 		break;
    216 
    217 	default:
    218 		/* Must compute it ourselves. */
    219 		if (in_cksum(m, sizeof(struct ip)) != 0)
    220 			return (0);
    221 		break;
    222 	}
    223 
    224 	/*
    225 	 * Route and interface still up?
    226 	 */
    227 	if ((rt = rtcache_validate(&ipf->ipf_ro)) == NULL ||
    228 	    (rt->rt_ifp->if_flags & IFF_UP) == 0)
    229 		return 0;
    230 
    231 	/*
    232 	 * Packet size OK?  TTL?
    233 	 */
    234 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
    235 		return 0;
    236 
    237 	/*
    238 	 * Clear any in-bound checksum flags for this packet.
    239 	 */
    240 	m->m_pkthdr.csum_flags = 0;
    241 
    242 	/*
    243 	 * Everything checks out and so we can forward this packet.
    244 	 * Modify the TTL and incrementally change the checksum.
    245 	 *
    246 	 * This method of adding the checksum works on either endian CPU.
    247 	 * If htons() is inlined, all the arithmetic is folded; otherwise
    248 	 * the htons()s are combined by CSE due to the const attribute.
    249 	 *
    250 	 * Don't bother using HW checksumming here -- the incremental
    251 	 * update is pretty fast.
    252 	 */
    253 	ip->ip_ttl -= IPTTLDEC;
    254 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
    255 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
    256 	else
    257 		ip->ip_sum += htons(IPTTLDEC << 8);
    258 
    259 	/*
    260 	 * Done modifying the header; copy it back, if necessary.
    261 	 *
    262 	 * XXX Use m_copyback_cow(9) here? --dyoung
    263 	 */
    264 	if (IP_HDR_ALIGNED_P(mtod(m, void *)) == 0)
    265 		memcpy(mtod(m, void *), &ip_store, sizeof(ip_store));
    266 
    267 	/*
    268 	 * Trim the packet in case it's too long..
    269 	 */
    270 	if (m->m_pkthdr.len > iplen) {
    271 		if (m->m_len == m->m_pkthdr.len) {
    272 			m->m_len = iplen;
    273 			m->m_pkthdr.len = iplen;
    274 		} else
    275 			m_adj(m, iplen - m->m_pkthdr.len);
    276 	}
    277 
    278 	/*
    279 	 * Send the packet on it's way.  All we can get back is ENOBUFS
    280 	 */
    281 	ipf->ipf_uses++;
    282 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
    283 
    284 	if (rt->rt_flags & RTF_GATEWAY)
    285 		dst = rt->rt_gateway;
    286 	else
    287 		dst = rtcache_getdst(&ipf->ipf_ro);
    288 
    289 	KERNEL_LOCK(1, NULL);
    290 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
    291 		if (error == ENOBUFS)
    292 			ipf->ipf_dropped++;
    293 		else
    294 			ipf->ipf_errors++;
    295 	}
    296 	KERNEL_UNLOCK_ONE(NULL);
    297 	return 1;
    298 }
    299 
    300 static void
    302 ipflow_addstats(struct ipflow *ipf)
    303 {
    304 	struct rtentry *rt;
    305 	uint64_t *ips;
    306 
    307 	if ((rt = rtcache_validate(&ipf->ipf_ro)) != NULL)
    308 		rt->rt_use += ipf->ipf_uses;
    309 
    310 	ips = IP_STAT_GETREF();
    311 	ips[IP_STAT_CANTFORWARD] += ipf->ipf_errors + ipf->ipf_dropped;
    312 	ips[IP_STAT_TOTAL] += ipf->ipf_uses;
    313 	ips[IP_STAT_FORWARD] += ipf->ipf_uses;
    314 	ips[IP_STAT_FASTFORWARD] += ipf->ipf_uses;
    315 	IP_STAT_PUTREF();
    316 }
    317 
    318 static void
    319 ipflow_free(struct ipflow *ipf)
    320 {
    321 	int s;
    322 	/*
    323 	 * Remove the flow from the hash table (at elevated IPL).
    324 	 * Once it's off the list, we can deal with it at normal
    325 	 * network IPL.
    326 	 */
    327 	s = splnet();
    328 	IPFLOW_REMOVE(ipf);
    329 	splx(s);
    330 	ipflow_addstats(ipf);
    331 	rtcache_free(&ipf->ipf_ro);
    332 	ipflow_inuse--;
    333 	s = splnet();
    334 	pool_put(&ipflow_pool, ipf);
    335 	splx(s);
    336 }
    337 
    338 struct ipflow *
    339 ipflow_reap(bool just_one)
    340 {
    341 	while (just_one || ipflow_inuse > ip_maxflows) {
    342 		struct ipflow *ipf, *maybe_ipf = NULL;
    343 		int s;
    344 
    345 		ipf = LIST_FIRST(&ipflowlist);
    346 		while (ipf != NULL) {
    347 			/*
    348 			 * If this no longer points to a valid route
    349 			 * reclaim it.
    350 			 */
    351 			if (rtcache_validate(&ipf->ipf_ro) == NULL)
    352 				goto done;
    353 			/*
    354 			 * choose the one that's been least recently
    355 			 * used or has had the least uses in the
    356 			 * last 1.5 intervals.
    357 			 */
    358 			if (maybe_ipf == NULL ||
    359 			    ipf->ipf_timer < maybe_ipf->ipf_timer ||
    360 			    (ipf->ipf_timer == maybe_ipf->ipf_timer &&
    361 			     ipf->ipf_last_uses + ipf->ipf_uses <
    362 			         maybe_ipf->ipf_last_uses +
    363 			         maybe_ipf->ipf_uses))
    364 				maybe_ipf = ipf;
    365 			ipf = LIST_NEXT(ipf, ipf_list);
    366 		}
    367 		ipf = maybe_ipf;
    368 	    done:
    369 		/*
    370 		 * Remove the entry from the flow table.
    371 		 */
    372 		s = splnet();
    373 		IPFLOW_REMOVE(ipf);
    374 		splx(s);
    375 		ipflow_addstats(ipf);
    376 		rtcache_free(&ipf->ipf_ro);
    377 		if (just_one)
    378 			return ipf;
    379 		pool_put(&ipflow_pool, ipf);
    380 		ipflow_inuse--;
    381 	}
    382 	return NULL;
    383 }
    384 
    385 void
    386 ipflow_slowtimo(void)
    387 {
    388 	struct rtentry *rt;
    389 	struct ipflow *ipf, *next_ipf;
    390 	uint64_t *ips;
    391 
    392 	mutex_enter(softnet_lock);
    393 	KERNEL_LOCK(1, NULL);
    394 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
    395 		next_ipf = LIST_NEXT(ipf, ipf_list);
    396 		if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer) ||
    397 		    (rt = rtcache_validate(&ipf->ipf_ro)) == NULL) {
    398 			ipflow_free(ipf);
    399 		} else {
    400 			ipf->ipf_last_uses = ipf->ipf_uses;
    401 			rt->rt_use += ipf->ipf_uses;
    402 			ips = IP_STAT_GETREF();
    403 			ips[IP_STAT_TOTAL] += ipf->ipf_uses;
    404 			ips[IP_STAT_FORWARD] += ipf->ipf_uses;
    405 			ips[IP_STAT_FASTFORWARD] += ipf->ipf_uses;
    406 			IP_STAT_PUTREF();
    407 			ipf->ipf_uses = 0;
    408 		}
    409 	}
    410 	KERNEL_UNLOCK_ONE(NULL);
    411 	mutex_exit(softnet_lock);
    412 }
    413 
    414 void
    415 ipflow_create(const struct route *ro, struct mbuf *m)
    416 {
    417 	const struct ip *const ip = mtod(m, const struct ip *);
    418 	struct ipflow *ipf;
    419 	size_t hash;
    420 	int s;
    421 
    422 	/*
    423 	 * Don't create cache entries for ICMP messages.
    424 	 */
    425 	if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP)
    426 		return;
    427 	/*
    428 	 * See if an existing flow struct exists.  If so remove it from it's
    429 	 * list and free the old route.  If not, try to malloc a new one
    430 	 * (if we aren't at our limit).
    431 	 */
    432 	ipf = ipflow_lookup(ip);
    433 	if (ipf == NULL) {
    434 		if (ipflow_inuse >= ip_maxflows) {
    435 			ipf = ipflow_reap(true);
    436 		} else {
    437 			s = splnet();
    438 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
    439 			splx(s);
    440 			if (ipf == NULL)
    441 				return;
    442 			ipflow_inuse++;
    443 		}
    444 		memset(ipf, 0, sizeof(*ipf));
    445 	} else {
    446 		s = splnet();
    447 		IPFLOW_REMOVE(ipf);
    448 		splx(s);
    449 		ipflow_addstats(ipf);
    450 		rtcache_free(&ipf->ipf_ro);
    451 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
    452 		ipf->ipf_errors = ipf->ipf_dropped = 0;
    453 	}
    454 
    455 	/*
    456 	 * Fill in the updated information.
    457 	 */
    458 	rtcache_copy(&ipf->ipf_ro, ro);
    459 	ipf->ipf_dst = ip->ip_dst;
    460 	ipf->ipf_src = ip->ip_src;
    461 	ipf->ipf_tos = ip->ip_tos;
    462 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
    463 
    464 	/*
    465 	 * Insert into the approriate bucket of the flow table.
    466 	 */
    467 	hash = ipflow_hash(ip);
    468 	s = splnet();
    469 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
    470 	splx(s);
    471 }
    472 
    473 int
    474 ipflow_invalidate_all(int new_size)
    475 {
    476 	struct ipflow *ipf, *next_ipf;
    477 	int s, error;
    478 
    479 	error = 0;
    480 	s = splnet();
    481 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
    482 		next_ipf = LIST_NEXT(ipf, ipf_list);
    483 		ipflow_free(ipf);
    484 	}
    485 
    486 	if (new_size)
    487 		error = ipflow_init(new_size);
    488 	splx(s);
    489 
    490 	return error;
    491 }
    492