The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/netgraph/netflow/netflow.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 2004-2005 Gleb Smirnoff <glebius@FreeBSD.org>
    3  * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net>
    4  * All rights reserved.
    5  *
    6  * Redistribution and use in source and binary forms, with or without
    7  * modification, are permitted provided that the following conditions
    8  * are met:
    9  * 1. Redistributions of source code must retain the above copyright
   10  *    notice, this list of conditions and the following disclaimer.
   11  * 2. Redistributions in binary form must reproduce the above copyright
   12  *    notice, this list of conditions and the following disclaimer in the
   13  *    documentation and/or other materials provided with the distribution.
   14  *
   15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   25  * SUCH DAMAGE.
   26  *
   27  * $SourceForge: netflow.c,v 1.41 2004/09/05 11:41:10 glebius Exp $
   28  */
   29 
   30 static const char rcs_id[] =
   31     "@(#) $FreeBSD: releng/8.3/sys/netgraph/netflow/netflow.c 212734 2010-09-16 14:30:32Z glebius $";
   32 
   33 #include <sys/param.h>
   34 #include <sys/kernel.h>
   35 #include <sys/limits.h>
   36 #include <sys/mbuf.h>
   37 #include <sys/syslog.h>
   38 #include <sys/systm.h>
   39 #include <sys/socket.h>
   40 
   41 #include <machine/atomic.h>
   42 
   43 #include <net/if.h>
   44 #include <net/route.h>
   45 #include <netinet/in.h>
   46 #include <netinet/in_systm.h>
   47 #include <netinet/ip.h>
   48 #include <netinet/tcp.h>
   49 #include <netinet/udp.h>
   50 
   51 #include <netgraph/ng_message.h>
   52 #include <netgraph/netgraph.h>
   53 
   54 #include <netgraph/netflow/netflow.h>
   55 #include <netgraph/netflow/ng_netflow.h>
   56 
   57 #define NBUCKETS        (65536)         /* must be power of 2 */
   58 
   59 /* This hash is for TCP or UDP packets. */
   60 #define FULL_HASH(addr1, addr2, port1, port2)   \
   61         (((addr1 ^ (addr1 >> 16) ^              \
   62         htons(addr2 ^ (addr2 >> 16))) ^         \
   63         port1 ^ htons(port2)) &                 \
   64         (NBUCKETS - 1))
   65 
   66 /* This hash is for all other IP packets. */
   67 #define ADDR_HASH(addr1, addr2)                 \
   68         ((addr1 ^ (addr1 >> 16) ^               \
   69         htons(addr2 ^ (addr2 >> 16))) &         \
   70         (NBUCKETS - 1))
   71 
   72 /* Macros to shorten logical constructions */
   73 /* XXX: priv must exist in namespace */
   74 #define INACTIVE(fle)   (time_uptime - fle->f.last > priv->info.nfinfo_inact_t)
   75 #define AGED(fle)       (time_uptime - fle->f.first > priv->info.nfinfo_act_t)
   76 #define ISFREE(fle)     (fle->f.packets == 0)
   77 
   78 /*
   79  * 4 is a magical number: statistically number of 4-packet flows is
   80  * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP
   81  * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case
   82  * of reachable host and 4-packet otherwise.
   83  */
   84 #define SMALL(fle)      (fle->f.packets <= 4)
   85 
   86 /*
   87  * Cisco uses milliseconds for uptime. Bad idea, since it overflows
   88  * every 48+ days. But we will do same to keep compatibility. This macro
   89  * does overflowable multiplication to 1000.
   90  */
   91 #define MILLIUPTIME(t)  (((t) << 9) +   /* 512 */       \
   92                          ((t) << 8) +   /* 256 */       \
   93                          ((t) << 7) +   /* 128 */       \
   94                          ((t) << 6) +   /* 64  */       \
   95                          ((t) << 5) +   /* 32  */       \
   96                          ((t) << 3))    /* 8   */
   97 
   98 MALLOC_DECLARE(M_NETFLOW_HASH);
   99 MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash");
  100 
  101 static int export_add(item_p, struct flow_entry *);
  102 static int export_send(priv_p, item_p, int flags);
  103 
  104 /* Generate hash for a given flow record. */
  105 static __inline uint32_t
  106 ip_hash(struct flow_rec *r)
  107 {
  108         switch (r->r_ip_p) {
  109         case IPPROTO_TCP:
  110         case IPPROTO_UDP:
  111                 return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr,
  112                     r->r_sport, r->r_dport);
  113         default:
  114                 return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr);
  115         }
  116 }
  117 
  118 /* This is callback from uma(9), called on alloc. */
  119 static int
  120 uma_ctor_flow(void *mem, int size, void *arg, int how)
  121 {
  122         priv_p priv = (priv_p )arg;
  123 
  124         if (atomic_load_acq_32(&priv->info.nfinfo_used) >= CACHESIZE)
  125                 return (ENOMEM);
  126 
  127         atomic_add_32(&priv->info.nfinfo_used, 1);
  128 
  129         return (0);
  130 }
  131 
  132 /* This is callback from uma(9), called on free. */
  133 static void
  134 uma_dtor_flow(void *mem, int size, void *arg)
  135 {
  136         priv_p priv = (priv_p )arg;
  137 
  138         atomic_subtract_32(&priv->info.nfinfo_used, 1);
  139 }
  140 
  141 /*
  142  * Detach export datagram from priv, if there is any.
  143  * If there is no, allocate a new one.
  144  */
  145 static item_p
  146 get_export_dgram(priv_p priv)
  147 {
  148         item_p  item = NULL;
  149 
  150         mtx_lock(&priv->export_mtx);
  151         if (priv->export_item != NULL) {
  152                 item = priv->export_item;
  153                 priv->export_item = NULL;
  154         }
  155         mtx_unlock(&priv->export_mtx);
  156 
  157         if (item == NULL) {
  158                 struct netflow_v5_export_dgram *dgram;
  159                 struct mbuf *m;
  160 
  161                 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
  162                 if (m == NULL)
  163                         return (NULL);
  164                 item = ng_package_data(m, NG_NOFLAGS);
  165                 if (item == NULL)
  166                         return (NULL);
  167                 dgram = mtod(m, struct netflow_v5_export_dgram *);
  168                 dgram->header.count = 0;
  169                 dgram->header.version = htons(NETFLOW_V5);
  170                 dgram->header.pad = 0;
  171 
  172         }
  173 
  174         return (item);
  175 }
  176 
  177 /*
  178  * Re-attach incomplete datagram back to priv.
  179  * If there is already another one, then send incomplete. */
  180 static void
  181 return_export_dgram(priv_p priv, item_p item, int flags)
  182 {
  183         /*
  184          * It may happen on SMP, that some thread has already
  185          * put its item there, in this case we bail out and
  186          * send what we have to collector.
  187          */
  188         mtx_lock(&priv->export_mtx);
  189         if (priv->export_item == NULL) {
  190                 priv->export_item = item;
  191                 mtx_unlock(&priv->export_mtx);
  192         } else {
  193                 mtx_unlock(&priv->export_mtx);
  194                 export_send(priv, item, flags);
  195         }
  196 }
  197 
  198 /*
  199  * The flow is over. Call export_add() and free it. If datagram is
  200  * full, then call export_send().
  201  */
  202 static __inline void
  203 expire_flow(priv_p priv, item_p *item, struct flow_entry *fle, int flags)
  204 {
  205         if (*item == NULL)
  206                 *item = get_export_dgram(priv);
  207         if (*item == NULL) {
  208                 atomic_add_32(&priv->info.nfinfo_export_failed, 1);
  209                 uma_zfree_arg(priv->zone, fle, priv);
  210                 return;
  211         }
  212         if (export_add(*item, fle) > 0) {
  213                 export_send(priv, *item, flags);
  214                 *item = NULL;
  215         }
  216         uma_zfree_arg(priv->zone, fle, priv);
  217 }
  218 
  219 /* Get a snapshot of node statistics */
  220 void
  221 ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i)
  222 {
  223         /* XXX: atomic */
  224         memcpy((void *)i, (void *)&priv->info, sizeof(priv->info));
  225 }
  226 
  227 /*
  228  * Insert a record into defined slot.
  229  *
  230  * First we get for us a free flow entry, then fill in all
  231  * possible fields in it.
  232  *
  233  * TODO: consider dropping hash mutex while filling in datagram,
  234  * as this was done in previous version. Need to test & profile
  235  * to be sure.
  236  */
  237 static __inline int
  238 hash_insert(priv_p priv, struct flow_hash_entry  *hsh, struct flow_rec *r,
  239         int plen, uint8_t tcp_flags)
  240 {
  241         struct flow_entry *fle;
  242         struct sockaddr_in sin;
  243         struct rtentry *rt;
  244 
  245         mtx_assert(&hsh->mtx, MA_OWNED);
  246 
  247         fle = uma_zalloc_arg(priv->zone, priv, M_NOWAIT);
  248         if (fle == NULL) {
  249                 atomic_add_32(&priv->info.nfinfo_alloc_failed, 1);
  250                 return (ENOMEM);
  251         }
  252 
  253         /*
  254          * Now fle is totally ours. It is detached from all lists,
  255          * we can safely edit it.
  256          */
  257 
  258         bcopy(r, &fle->f.r, sizeof(struct flow_rec));
  259         fle->f.bytes = plen;
  260         fle->f.packets = 1;
  261         fle->f.tcp_flags = tcp_flags;
  262 
  263         fle->f.first = fle->f.last = time_uptime;
  264 
  265         /*
  266          * First we do route table lookup on destination address. So we can
  267          * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
  268          */
  269         bzero(&sin, sizeof(sin));
  270         sin.sin_len = sizeof(struct sockaddr_in);
  271         sin.sin_family = AF_INET;
  272         sin.sin_addr = fle->f.r.r_dst;
  273         /* XXX MRT 0 as a default.. need the m here to get fib */
  274         rt = rtalloc1_fib((struct sockaddr *)&sin, 0, 0, 0);
  275         if (rt != NULL) {
  276                 fle->f.fle_o_ifx = rt->rt_ifp->if_index;
  277 
  278                 if (rt->rt_flags & RTF_GATEWAY &&
  279                     rt->rt_gateway->sa_family == AF_INET)
  280                         fle->f.next_hop =
  281                             ((struct sockaddr_in *)(rt->rt_gateway))->sin_addr;
  282 
  283                 if (rt_mask(rt))
  284                         fle->f.dst_mask = bitcount32(((struct sockaddr_in *)
  285                             rt_mask(rt))->sin_addr.s_addr);
  286                 else if (rt->rt_flags & RTF_HOST)
  287                         /* Give up. We can't determine mask :( */
  288                         fle->f.dst_mask = 32;
  289 
  290                 RTFREE_LOCKED(rt);
  291         }
  292 
  293         /* Do route lookup on source address, to fill in src_mask. */
  294         bzero(&sin, sizeof(sin));
  295         sin.sin_len = sizeof(struct sockaddr_in);
  296         sin.sin_family = AF_INET;
  297         sin.sin_addr = fle->f.r.r_src;
  298         /* XXX MRT 0 as a default  revisit.  need the mbuf for fib*/
  299         rt = rtalloc1_fib((struct sockaddr *)&sin, 0, 0, 0);
  300         if (rt != NULL) {
  301                 if (rt_mask(rt))
  302                         fle->f.src_mask = bitcount32(((struct sockaddr_in *)
  303                             rt_mask(rt))->sin_addr.s_addr);
  304                 else if (rt->rt_flags & RTF_HOST)
  305                         /* Give up. We can't determine mask :( */
  306                         fle->f.src_mask = 32;
  307 
  308                 RTFREE_LOCKED(rt);
  309         }
  310 
  311         /* Push new flow at the and of hash. */
  312         TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
  313 
  314         return (0);
  315 }
  316 
  317 
  318 /*
  319  * Non-static functions called from ng_netflow.c
  320  */
  321 
  322 /* Allocate memory and set up flow cache */
  323 int
  324 ng_netflow_cache_init(priv_p priv)
  325 {
  326         struct flow_hash_entry  *hsh;
  327         int i;
  328 
  329         /* Initialize cache UMA zone. */
  330         priv->zone = uma_zcreate("NetFlow cache", sizeof(struct flow_entry),
  331             uma_ctor_flow, uma_dtor_flow, NULL, NULL, UMA_ALIGN_CACHE, 0);
  332         uma_zone_set_max(priv->zone, CACHESIZE);
  333 
  334         /* Allocate hash. */
  335         priv->hash = malloc(NBUCKETS * sizeof(struct flow_hash_entry),
  336             M_NETFLOW_HASH, M_WAITOK | M_ZERO);
  337 
  338         if (priv->hash == NULL) {
  339                 uma_zdestroy(priv->zone);
  340                 return (ENOMEM);
  341         }
  342 
  343         /* Initialize hash. */
  344         for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) {
  345                 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
  346                 TAILQ_INIT(&hsh->head);
  347         }
  348 
  349         mtx_init(&priv->export_mtx, "export dgram lock", NULL, MTX_DEF);
  350 
  351         return (0);
  352 }
  353 
  354 /* Free all flow cache memory. Called from node close method. */
  355 void
  356 ng_netflow_cache_flush(priv_p priv)
  357 {
  358         struct flow_entry       *fle, *fle1;
  359         struct flow_hash_entry  *hsh;
  360         item_p                  item = NULL;
  361         int i;
  362 
  363         /*
  364          * We are going to free probably billable data.
  365          * Expire everything before freeing it.
  366          * No locking is required since callout is already drained.
  367          */
  368         for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++)
  369                 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
  370                         TAILQ_REMOVE(&hsh->head, fle, fle_hash);
  371                         expire_flow(priv, &item, fle, NG_QUEUE);
  372                 }
  373 
  374         if (item != NULL)
  375                 export_send(priv, item, NG_QUEUE);
  376 
  377         uma_zdestroy(priv->zone);
  378 
  379         /* Destroy hash mutexes. */
  380         for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++)
  381                 mtx_destroy(&hsh->mtx);
  382 
  383         /* Free hash memory. */
  384         if (priv->hash)
  385                 free(priv->hash, M_NETFLOW_HASH);
  386 
  387         mtx_destroy(&priv->export_mtx);
  388 }
  389 
  390 /* Insert packet from into flow cache. */
  391 int
  392 ng_netflow_flow_add(priv_p priv, struct ip *ip, unsigned int src_if_index)
  393 {
  394         register struct flow_entry      *fle, *fle1;
  395         struct flow_hash_entry          *hsh;
  396         struct flow_rec         r;
  397         item_p                  item = NULL;
  398         int                     hlen, plen;
  399         int                     error = 0;
  400         uint8_t                 tcp_flags = 0;
  401 
  402         /* Try to fill flow_rec r */
  403         bzero(&r, sizeof(r));
  404         /* check version */
  405         if (ip->ip_v != IPVERSION)
  406                 return (EINVAL);
  407 
  408         /* verify min header length */
  409         hlen = ip->ip_hl << 2;
  410 
  411         if (hlen < sizeof(struct ip))
  412                 return (EINVAL);
  413 
  414         r.r_src = ip->ip_src;
  415         r.r_dst = ip->ip_dst;
  416 
  417         /* save packet length */
  418         plen = ntohs(ip->ip_len);
  419 
  420         r.r_ip_p = ip->ip_p;
  421         r.r_tos = ip->ip_tos;
  422 
  423         r.r_i_ifx = src_if_index;
  424 
  425         /*
  426          * XXX NOTE: only first fragment of fragmented TCP, UDP and
  427          * ICMP packet will be recorded with proper s_port and d_port.
  428          * Following fragments will be recorded simply as IP packet with
  429          * ip_proto = ip->ip_p and s_port, d_port set to zero.
  430          * I know, it looks like bug. But I don't want to re-implement
  431          * ip packet assebmling here. Anyway, (in)famous trafd works this way -
  432          * and nobody complains yet :)
  433          */
  434         if ((ip->ip_off & htons(IP_OFFMASK)) == 0)
  435                 switch(r.r_ip_p) {
  436                 case IPPROTO_TCP:
  437                 {
  438                         register struct tcphdr *tcp;
  439 
  440                         tcp = (struct tcphdr *)((caddr_t )ip + hlen);
  441                         r.r_sport = tcp->th_sport;
  442                         r.r_dport = tcp->th_dport;
  443                         tcp_flags = tcp->th_flags;
  444                         break;
  445                 }
  446                         case IPPROTO_UDP:
  447                         r.r_ports = *(uint32_t *)((caddr_t )ip + hlen);
  448                         break;
  449                 }
  450 
  451         /* Update node statistics. XXX: race... */
  452         priv->info.nfinfo_packets ++;
  453         priv->info.nfinfo_bytes += plen;
  454 
  455         /* Find hash slot. */
  456         hsh = &priv->hash[ip_hash(&r)];
  457 
  458         mtx_lock(&hsh->mtx);
  459 
  460         /*
  461          * Go through hash and find our entry. If we encounter an
  462          * entry, that should be expired, purge it. We do a reverse
  463          * search since most active entries are first, and most
  464          * searches are done on most active entries.
  465          */
  466         TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
  467                 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0)
  468                         break;
  469                 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) {
  470                         TAILQ_REMOVE(&hsh->head, fle, fle_hash);
  471                         expire_flow(priv, &item, fle, NG_QUEUE);
  472                         atomic_add_32(&priv->info.nfinfo_act_exp, 1);
  473                 }
  474         }
  475 
  476         if (fle) {                      /* An existent entry. */
  477 
  478                 fle->f.bytes += plen;
  479                 fle->f.packets ++;
  480                 fle->f.tcp_flags |= tcp_flags;
  481                 fle->f.last = time_uptime;
  482 
  483                 /*
  484                  * We have the following reasons to expire flow in active way:
  485                  * - it hit active timeout
  486                  * - a TCP connection closed
  487                  * - it is going to overflow counter
  488                  */
  489                 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) ||
  490                     (fle->f.bytes >= (UINT_MAX - IF_MAXMTU)) ) {
  491                         TAILQ_REMOVE(&hsh->head, fle, fle_hash);
  492                         expire_flow(priv, &item, fle, NG_QUEUE);
  493                         atomic_add_32(&priv->info.nfinfo_act_exp, 1);
  494                 } else {
  495                         /*
  496                          * It is the newest, move it to the tail,
  497                          * if it isn't there already. Next search will
  498                          * locate it quicker.
  499                          */
  500                         if (fle != TAILQ_LAST(&hsh->head, fhead)) {
  501                                 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
  502                                 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
  503                         }
  504                 }
  505         } else                          /* A new flow entry. */
  506                 error = hash_insert(priv, hsh, &r, plen, tcp_flags);
  507 
  508         mtx_unlock(&hsh->mtx);
  509 
  510         if (item != NULL)
  511                 return_export_dgram(priv, item, NG_QUEUE);
  512 
  513         return (error);
  514 }
  515 
  516 /*
  517  * Return records from cache to userland.
  518  *
  519  * TODO: matching particular IP should be done in kernel, here.
  520  */
  521 int
  522 ng_netflow_flow_show(priv_p priv, uint32_t last, struct ng_mesg *resp)
  523 {
  524         struct flow_hash_entry *hsh;
  525         struct flow_entry *fle;
  526         struct ngnf_flows *data;
  527         int i;
  528 
  529         data = (struct ngnf_flows *)resp->data;
  530         data->last = 0;
  531         data->nentries = 0;
  532 
  533         /* Check if this is a first run */
  534         if (last == 0) {
  535                 hsh = priv->hash;
  536                 i = 0;
  537         } else {
  538                 if (last > NBUCKETS-1)
  539                         return (EINVAL);
  540                 hsh = priv->hash + last;
  541                 i = last;
  542         }
  543 
  544         /*
  545          * We will transfer not more than NREC_AT_ONCE. More data
  546          * will come in next message.
  547          * We send current hash index to userland, and userland should
  548          * return it back to us. Then, we will restart with new entry.
  549          *
  550          * The resulting cache snapshot is inaccurate for the
  551          * following reasons:
  552          *  - we skip locked hash entries
  553          *  - we bail out, if someone wants our entry
  554          *  - we skip rest of entry, when hit NREC_AT_ONCE
  555          */
  556         for (; i < NBUCKETS; hsh++, i++) {
  557                 if (mtx_trylock(&hsh->mtx) == 0)
  558                         continue;
  559 
  560                 TAILQ_FOREACH(fle, &hsh->head, fle_hash) {
  561                         if (hsh->mtx.mtx_lock & MTX_CONTESTED)
  562                                 break;
  563 
  564                         bcopy(&fle->f, &(data->entries[data->nentries]),
  565                             sizeof(fle->f));
  566                         data->nentries++;
  567                         if (data->nentries == NREC_AT_ONCE) {
  568                                 mtx_unlock(&hsh->mtx);
  569                                 if (++i < NBUCKETS)
  570                                         data->last = i;
  571                                 return (0);
  572                         }
  573                 }
  574                 mtx_unlock(&hsh->mtx);
  575         }
  576 
  577         return (0);
  578 }
  579 
  580 /* We have full datagram in privdata. Send it to export hook. */
  581 static int
  582 export_send(priv_p priv, item_p item, int flags)
  583 {
  584         struct mbuf *m = NGI_M(item);
  585         struct netflow_v5_export_dgram *dgram = mtod(m,
  586                                         struct netflow_v5_export_dgram *);
  587         struct netflow_v5_header *header = &dgram->header;
  588         struct timespec ts;
  589         int error = 0;
  590 
  591         /* Fill mbuf header. */
  592         m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) *
  593            header->count + sizeof(struct netflow_v5_header);
  594 
  595         /* Fill export header. */
  596         header->sys_uptime = htonl(MILLIUPTIME(time_uptime));
  597         getnanotime(&ts);
  598         header->unix_secs  = htonl(ts.tv_sec);
  599         header->unix_nsecs = htonl(ts.tv_nsec);
  600         header->engine_type = 0;
  601         header->engine_id = 0;
  602         header->pad = 0;
  603         header->flow_seq = htonl(atomic_fetchadd_32(&priv->flow_seq,
  604             header->count));
  605         header->count = htons(header->count);
  606 
  607         if (priv->export != NULL)
  608                 NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags);
  609         else
  610                 NG_FREE_ITEM(item);
  611 
  612         return (error);
  613 }
  614 
  615 
  616 /* Add export record to dgram. */
  617 static int
  618 export_add(item_p item, struct flow_entry *fle)
  619 {
  620         struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item),
  621                                         struct netflow_v5_export_dgram *);
  622         struct netflow_v5_header *header = &dgram->header;
  623         struct netflow_v5_record *rec;
  624 
  625         rec = &dgram->r[header->count];
  626         header->count ++;
  627 
  628         KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS,
  629             ("ng_netflow: export too big"));
  630 
  631         /* Fill in export record. */
  632         rec->src_addr = fle->f.r.r_src.s_addr;
  633         rec->dst_addr = fle->f.r.r_dst.s_addr;
  634         rec->next_hop = fle->f.next_hop.s_addr;
  635         rec->i_ifx    = htons(fle->f.fle_i_ifx);
  636         rec->o_ifx    = htons(fle->f.fle_o_ifx);
  637         rec->packets  = htonl(fle->f.packets);
  638         rec->octets   = htonl(fle->f.bytes);
  639         rec->first    = htonl(MILLIUPTIME(fle->f.first));
  640         rec->last     = htonl(MILLIUPTIME(fle->f.last));
  641         rec->s_port   = fle->f.r.r_sport;
  642         rec->d_port   = fle->f.r.r_dport;
  643         rec->flags    = fle->f.tcp_flags;
  644         rec->prot     = fle->f.r.r_ip_p;
  645         rec->tos      = fle->f.r.r_tos;
  646         rec->dst_mask = fle->f.dst_mask;
  647         rec->src_mask = fle->f.src_mask;
  648         rec->pad1     = 0;
  649         rec->pad2     = 0;
  650 
  651         /* Not supported fields. */
  652         rec->src_as = rec->dst_as = 0;
  653 
  654         if (header->count == NETFLOW_V5_MAX_RECORDS)
  655                 return (1); /* end of datagram */
  656         else
  657                 return (0);     
  658 }
  659 
  660 /* Periodic flow expiry run. */
  661 void
  662 ng_netflow_expire(void *arg)
  663 {
  664         struct flow_entry       *fle, *fle1;
  665         struct flow_hash_entry  *hsh;
  666         priv_p                  priv = (priv_p )arg;
  667         item_p                  item = NULL;
  668         uint32_t                used;
  669         int                     i;
  670 
  671         /*
  672          * Going through all the cache.
  673          */
  674         for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) {
  675                 /*
  676                  * Skip entries, that are already being worked on.
  677                  */
  678                 if (mtx_trylock(&hsh->mtx) == 0)
  679                         continue;
  680 
  681                 used = atomic_load_acq_32(&priv->info.nfinfo_used);
  682                 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
  683                         /*
  684                          * Interrupt thread wants this entry!
  685                          * Quick! Quick! Bail out!
  686                          */
  687                         if (hsh->mtx.mtx_lock & MTX_CONTESTED)
  688                                 break;
  689 
  690                         /*
  691                          * Don't expire aggressively while hash collision
  692                          * ratio is predicted small.
  693                          */
  694                         if (used <= (NBUCKETS*2) && !INACTIVE(fle))
  695                                 break;
  696 
  697                         if ((INACTIVE(fle) && (SMALL(fle) ||
  698                             (used > (NBUCKETS*2)))) || AGED(fle)) {
  699                                 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
  700                                 expire_flow(priv, &item, fle, NG_NOFLAGS);
  701                                 used--;
  702                                 atomic_add_32(&priv->info.nfinfo_inact_exp, 1);
  703                         }
  704                 }
  705                 mtx_unlock(&hsh->mtx);
  706         }
  707 
  708         if (item != NULL)
  709                 return_export_dgram(priv, item, NG_NOFLAGS);
  710 
  711         /* Schedule next expire. */
  712         callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire,
  713             (void *)priv);
  714 }

Cache object: 31741cc7cc51e05145dfc342e74f4cdb


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.