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 * $FreeBSD: src/sys/netgraph/netflow/netflow.c,v 1.29 2008/05/09 23:02:57 julian Exp $ 29 */ 30 31 #include <sys/param.h> 32 #include <sys/kernel.h> 33 #include <sys/limits.h> 34 #include <sys/mbuf.h> 35 #include <sys/syslog.h> 36 #include <sys/systm.h> 37 #include <sys/socket.h> 38 39 #include <machine/atomic.h> 40 41 #include <net/if.h> 42 #include <net/route.h> 43 #include <netinet/in.h> 44 #include <netinet/in_systm.h> 45 #include <netinet/ip.h> 46 #include <netinet/tcp.h> 47 #include <netinet/udp.h> 48 49 #include "ng_message.h" 50 #include "netgraph.h" 51 52 #include "netflow/netflow.h" 53 #include "netflow/ng_netflow.h" 54 55 #define NBUCKETS (65536) /* must be power of 2 */ 56 57 /* This hash is for TCP or UDP packets. */ 58 #define FULL_HASH(addr1, addr2, port1, port2) \ 59 (((addr1 ^ (addr1 >> 16) ^ \ 60 htons(addr2 ^ (addr2 >> 16))) ^ \ 61 port1 ^ htons(port2)) & \ 62 (NBUCKETS - 1)) 63 64 /* This hash is for all other IP packets. */ 65 #define ADDR_HASH(addr1, addr2) \ 66 ((addr1 ^ (addr1 >> 16) ^ \ 67 htons(addr2 ^ (addr2 >> 16))) & \ 68 (NBUCKETS - 1)) 69 70 /* Macros to shorten logical constructions */ 71 /* XXX: priv must exist in namespace */ 72 #define INACTIVE(fle) (time_uptime - fle->f.last > priv->info.nfinfo_inact_t) 73 #define AGED(fle) (time_uptime - fle->f.first > priv->info.nfinfo_act_t) 74 #define ISFREE(fle) (fle->f.packets == 0) 75 76 /* 77 * 4 is a magical number: statistically number of 4-packet flows is 78 * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP 79 * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case 80 * of reachable host and 4-packet otherwise. 81 */ 82 #define SMALL(fle) (fle->f.packets <= 4) 83 84 /* 85 * Cisco uses milliseconds for uptime. Bad idea, since it overflows 86 * every 48+ days. But we will do same to keep compatibility. This macro 87 * does overflowable multiplication to 1000. 88 */ 89 #define MILLIUPTIME(t) (((t) << 9) + /* 512 */ \ 90 ((t) << 8) + /* 256 */ \ 91 ((t) << 7) + /* 128 */ \ 92 ((t) << 6) + /* 64 */ \ 93 ((t) << 5) + /* 32 */ \ 94 ((t) << 3)) /* 8 */ 95 96 MALLOC_DECLARE(M_NETFLOW_HASH); 97 MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash"); 98 99 static int export_add(item_p, struct flow_entry *); 100 static int export_send(priv_p, item_p, int flags); 101 102 /* Generate hash for a given flow record. */ 103 static __inline uint32_t 104 ip_hash(struct flow_rec *r) 105 { 106 switch (r->r_ip_p) { 107 case IPPROTO_TCP: 108 case IPPROTO_UDP: 109 return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr, 110 r->r_sport, r->r_dport); 111 default: 112 return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr); 113 } 114 } 115 116 /* This is callback from uma(9), called on alloc. */ 117 static int 118 uma_ctor_flow(void *mem, int size, void *arg, int how) 119 { 120 priv_p priv = (priv_p )arg; 121 122 if (atomic_load_acq_32(&priv->info.nfinfo_used) >= CACHESIZE) 123 return (ENOMEM); 124 125 atomic_add_32(&priv->info.nfinfo_used, 1); 126 127 return (0); 128 } 129 130 /* This is callback from uma(9), called on free. */ 131 static void 132 uma_dtor_flow(void *mem, int size, void *arg) 133 { 134 priv_p priv = (priv_p )arg; 135 136 atomic_subtract_32(&priv->info.nfinfo_used, 1); 137 } 138 139 /* 140 * Detach export datagram from priv, if there is any. 141 * If there is no, allocate a new one. 142 */ 143 static item_p 144 get_export_dgram(priv_p priv) 145 { 146 item_p item = NULL; 147 148 mtx_lock(&priv->export_mtx); 149 if (priv->export_item != NULL) { 150 item = priv->export_item; 151 priv->export_item = NULL; 152 } 153 mtx_unlock(&priv->export_mtx); 154 155 if (item == NULL) { 156 struct netflow_v5_export_dgram *dgram; 157 struct mbuf *m; 158 159 m = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR); 160 if (m == NULL) 161 return (NULL); 162 item = ng_package_data(m, NG_NOFLAGS); 163 if (item == NULL) 164 return (NULL); 165 dgram = mtod(m, struct netflow_v5_export_dgram *); 166 dgram->header.count = 0; 167 dgram->header.version = htons(NETFLOW_V5); 168 169 } 170 171 return (item); 172 } 173 174 /* 175 * Re-attach incomplete datagram back to priv. 176 * If there is already another one, then send incomplete. */ 177 static void 178 return_export_dgram(priv_p priv, item_p item, int flags) 179 { 180 /* 181 * It may happen on SMP, that some thread has already 182 * put its item there, in this case we bail out and 183 * send what we have to collector. 184 */ 185 mtx_lock(&priv->export_mtx); 186 if (priv->export_item == NULL) { 187 priv->export_item = item; 188 mtx_unlock(&priv->export_mtx); 189 } else { 190 mtx_unlock(&priv->export_mtx); 191 export_send(priv, item, flags); 192 } 193 } 194 195 /* 196 * The flow is over. Call export_add() and free it. If datagram is 197 * full, then call export_send(). 198 */ 199 static __inline void 200 expire_flow(priv_p priv, item_p *item, struct flow_entry *fle, int flags) 201 { 202 if (*item == NULL) 203 *item = get_export_dgram(priv); 204 if (*item == NULL) { 205 atomic_add_32(&priv->info.nfinfo_export_failed, 1); 206 uma_zfree_arg(priv->zone, fle, priv); 207 return; 208 } 209 if (export_add(*item, fle) > 0) { 210 export_send(priv, *item, flags); 211 *item = NULL; 212 } 213 uma_zfree_arg(priv->zone, fle, priv); 214 } 215 216 /* Get a snapshot of node statistics */ 217 void 218 ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i) 219 { 220 /* XXX: atomic */ 221 memcpy((void *)i, (void *)&priv->info, sizeof(priv->info)); 222 } 223 224 /* 225 * Insert a record into defined slot. 226 * 227 * First we get for us a free flow entry, then fill in all 228 * possible fields in it. 229 * 230 * TODO: consider dropping hash mutex while filling in datagram, 231 * as this was done in previous version. Need to test & profile 232 * to be sure. 233 */ 234 static __inline int 235 hash_insert(priv_p priv, struct flow_hash_entry *hsh, struct flow_rec *r, 236 int plen, uint8_t tcp_flags) 237 { 238 struct flow_entry *fle; 239 struct sockaddr_in sin; 240 struct rtentry *rt; 241 242 KKASSERT(mtx_owned(&hsh->mtx)); 243 244 fle = uma_zalloc_arg(priv->zone, priv, M_WAITOK | M_NULLOK); 245 if (fle == NULL) { 246 atomic_add_32(&priv->info.nfinfo_alloc_failed, 1); 247 return (ENOMEM); 248 } 249 250 /* 251 * Now fle is totally ours. It is detached from all lists, 252 * we can safely edit it. 253 */ 254 255 bcopy(r, &fle->f.r, sizeof(struct flow_rec)); 256 fle->f.bytes = plen; 257 fle->f.packets = 1; 258 fle->f.tcp_flags = tcp_flags; 259 260 fle->f.first = fle->f.last = time_uptime; 261 262 /* 263 * First we do route table lookup on destination address. So we can 264 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases. 265 */ 266 bzero(&sin, sizeof(sin)); 267 sin.sin_len = sizeof(struct sockaddr_in); 268 sin.sin_family = AF_INET; 269 sin.sin_addr = fle->f.r.r_dst; 270 /* XXX MRT 0 as a default.. need the m here to get fib */ 271 rt = rtalloc1_fib((struct sockaddr *)&sin, 0, RTF_CLONING, 0); 272 if (rt != NULL) { 273 fle->f.fle_o_ifx = rt->rt_ifp->if_index; 274 275 if (rt->rt_flags & RTF_GATEWAY && 276 rt->rt_gateway->sa_family == AF_INET) 277 fle->f.next_hop = 278 ((struct sockaddr_in *)(rt->rt_gateway))->sin_addr; 279 280 if (rt_mask(rt)) 281 fle->f.dst_mask = bitcount32(((struct sockaddr_in *) 282 rt_mask(rt))->sin_addr.s_addr); 283 else if (rt->rt_flags & RTF_HOST) 284 /* Give up. We can't determine mask :( */ 285 fle->f.dst_mask = 32; 286 287 RTFREE_LOCKED(rt); 288 } 289 290 /* Do route lookup on source address, to fill in src_mask. */ 291 bzero(&sin, sizeof(sin)); 292 sin.sin_len = sizeof(struct sockaddr_in); 293 sin.sin_family = AF_INET; 294 sin.sin_addr = fle->f.r.r_src; 295 /* XXX MRT 0 as a default revisit. need the mbuf for fib*/ 296 rt = rtalloc1_fib((struct sockaddr *)&sin, 0, RTF_CLONING, 0); 297 if (rt != NULL) { 298 if (rt_mask(rt)) 299 fle->f.src_mask = bitcount32(((struct sockaddr_in *) 300 rt_mask(rt))->sin_addr.s_addr); 301 else if (rt->rt_flags & RTF_HOST) 302 /* Give up. We can't determine mask :( */ 303 fle->f.src_mask = 32; 304 305 RTFREE_LOCKED(rt); 306 } 307 308 /* Push new flow at the and of hash. */ 309 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 310 311 return (0); 312 } 313 314 315 /* 316 * Non-static functions called from ng_netflow.c 317 */ 318 319 /* Allocate memory and set up flow cache */ 320 int 321 ng_netflow_cache_init(priv_p priv) 322 { 323 struct flow_hash_entry *hsh; 324 int i; 325 326 /* Initialize cache UMA zone. */ 327 priv->zone = uma_zcreate("NetFlow cache", sizeof(struct flow_entry), 328 uma_ctor_flow, uma_dtor_flow, NULL, NULL, UMA_ALIGN_CACHE, 0); 329 uma_zone_set_max(priv->zone, CACHESIZE); 330 331 /* Allocate hash. */ 332 priv->hash = kmalloc(NBUCKETS * sizeof(struct flow_hash_entry), 333 M_NETFLOW_HASH, M_WAITOK | M_ZERO); 334 335 /* Initialize hash. */ 336 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) { 337 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF); 338 TAILQ_INIT(&hsh->head); 339 } 340 341 mtx_init(&priv->export_mtx, "export dgram lock", NULL, MTX_DEF); 342 343 return (0); 344 } 345 346 /* Free all flow cache memory. Called from node close method. */ 347 void 348 ng_netflow_cache_flush(priv_p priv) 349 { 350 struct flow_entry *fle, *fle1; 351 struct flow_hash_entry *hsh; 352 item_p item = NULL; 353 int i; 354 355 /* 356 * We are going to free probably billable data. 357 * Expire everything before freeing it. 358 * No locking is required since callout is already drained. 359 */ 360 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) 361 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 362 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 363 expire_flow(priv, &item, fle, NG_QUEUE); 364 } 365 366 if (item != NULL) 367 export_send(priv, item, NG_QUEUE); 368 369 uma_zdestroy(priv->zone); 370 371 /* Destroy hash mutexes. */ 372 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) 373 mtx_destroy(&hsh->mtx); 374 375 /* Free hash memory. */ 376 if (priv->hash) 377 kfree(priv->hash, M_NETFLOW_HASH); 378 379 mtx_destroy(&priv->export_mtx); 380 } 381 382 /* Insert packet from into flow cache. */ 383 int 384 ng_netflow_flow_add(priv_p priv, struct ip *ip, iface_p iface, 385 struct ifnet *ifp) 386 { 387 register struct flow_entry *fle, *fle1; 388 struct flow_hash_entry *hsh; 389 struct flow_rec r; 390 item_p item = NULL; 391 int hlen, plen; 392 int error = 0; 393 uint8_t tcp_flags = 0; 394 395 /* Try to fill flow_rec r */ 396 bzero(&r, sizeof(r)); 397 /* check version */ 398 if (ip->ip_v != IPVERSION) 399 return (EINVAL); 400 401 /* verify min header length */ 402 hlen = ip->ip_hl << 2; 403 404 if (hlen < sizeof(struct ip)) 405 return (EINVAL); 406 407 r.r_src = ip->ip_src; 408 r.r_dst = ip->ip_dst; 409 410 /* save packet length */ 411 plen = ntohs(ip->ip_len); 412 413 r.r_ip_p = ip->ip_p; 414 r.r_tos = ip->ip_tos; 415 416 /* Configured in_ifx overrides mbuf's */ 417 if (iface->info.ifinfo_index == 0) { 418 if (ifp != NULL) 419 r.r_i_ifx = ifp->if_index; 420 } else 421 r.r_i_ifx = iface->info.ifinfo_index; 422 423 /* 424 * XXX NOTE: only first fragment of fragmented TCP, UDP and 425 * ICMP packet will be recorded with proper s_port and d_port. 426 * Following fragments will be recorded simply as IP packet with 427 * ip_proto = ip->ip_p and s_port, d_port set to zero. 428 * I know, it looks like bug. But I don't want to re-implement 429 * ip packet assebmling here. Anyway, (in)famous trafd works this way - 430 * and nobody complains yet :) 431 */ 432 if ((ip->ip_off & htons(IP_OFFMASK)) == 0) 433 switch(r.r_ip_p) { 434 case IPPROTO_TCP: 435 { 436 register struct tcphdr *tcp; 437 438 tcp = (struct tcphdr *)((caddr_t )ip + hlen); 439 r.r_sport = tcp->th_sport; 440 r.r_dport = tcp->th_dport; 441 tcp_flags = tcp->th_flags; 442 break; 443 } 444 case IPPROTO_UDP: 445 r.r_ports = *(uint32_t *)((caddr_t )ip + hlen); 446 break; 447 } 448 449 /* Update node statistics. XXX: race... */ 450 priv->info.nfinfo_packets ++; 451 priv->info.nfinfo_bytes += plen; 452 453 /* Find hash slot. */ 454 hsh = &priv->hash[ip_hash(&r)]; 455 456 mtx_lock(&hsh->mtx); 457 458 /* 459 * Go through hash and find our entry. If we encounter an 460 * entry, that should be expired, purge it. We do a reverse 461 * search since most active entries are first, and most 462 * searches are done on most active entries. 463 */ 464 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) { 465 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0) 466 break; 467 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) { 468 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 469 expire_flow(priv, &item, fle, NG_QUEUE); 470 atomic_add_32(&priv->info.nfinfo_act_exp, 1); 471 } 472 } 473 474 if (fle) { /* An existent entry. */ 475 476 fle->f.bytes += plen; 477 fle->f.packets ++; 478 fle->f.tcp_flags |= tcp_flags; 479 fle->f.last = time_uptime; 480 481 /* 482 * We have the following reasons to expire flow in active way: 483 * - it hit active timeout 484 * - a TCP connection closed 485 * - it is going to overflow counter 486 */ 487 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) || 488 (fle->f.bytes >= (UINT_MAX - IF_MAXMTU)) ) { 489 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 490 expire_flow(priv, &item, fle, NG_QUEUE); 491 atomic_add_32(&priv->info.nfinfo_act_exp, 1); 492 } else { 493 /* 494 * It is the newest, move it to the tail, 495 * if it isn't there already. Next search will 496 * locate it quicker. 497 */ 498 if (fle != TAILQ_LAST(&hsh->head, fhead)) { 499 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 500 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash); 501 } 502 } 503 } else /* A new flow entry. */ 504 error = hash_insert(priv, hsh, &r, plen, tcp_flags); 505 506 mtx_unlock(&hsh->mtx); 507 508 if (item != NULL) 509 return_export_dgram(priv, item, NG_QUEUE); 510 511 return (error); 512 } 513 514 /* 515 * Return records from cache to userland. 516 * 517 * TODO: matching particular IP should be done in kernel, here. 518 */ 519 int 520 ng_netflow_flow_show(priv_p priv, uint32_t last, struct ng_mesg *resp) 521 { 522 struct flow_hash_entry *hsh; 523 struct flow_entry *fle; 524 struct ngnf_flows *data; 525 int i; 526 527 data = (struct ngnf_flows *)resp->data; 528 data->last = 0; 529 data->nentries = 0; 530 531 /* Check if this is a first run */ 532 if (last == 0) { 533 hsh = priv->hash; 534 i = 0; 535 } else { 536 if (last > NBUCKETS-1) 537 return (EINVAL); 538 hsh = priv->hash + last; 539 i = last; 540 } 541 542 /* 543 * We will transfer not more than NREC_AT_ONCE. More data 544 * will come in next message. 545 * We send current hash index to userland, and userland should 546 * return it back to us. Then, we will restart with new entry. 547 * 548 * The resulting cache snapshot is inaccurate for the 549 * following reasons: 550 * - we skip locked hash entries 551 * - we bail out, if someone wants our entry 552 * - we skip rest of entry, when hit NREC_AT_ONCE 553 */ 554 for (; i < NBUCKETS; hsh++, i++) { 555 if (mtx_trylock(&hsh->mtx) == 0) 556 continue; 557 558 TAILQ_FOREACH(fle, &hsh->head, fle_hash) { 559 if (mtx_contested(&hsh->mtx) 560 break; 561 562 bcopy(&fle->f, &(data->entries[data->nentries]), 563 sizeof(fle->f)); 564 data->nentries++; 565 if (data->nentries == NREC_AT_ONCE) { 566 mtx_unlock(&hsh->mtx); 567 if (++i < NBUCKETS) 568 data->last = i; 569 return (0); 570 } 571 } 572 mtx_unlock(&hsh->mtx); 573 } 574 575 return (0); 576 } 577 578 /* We have full datagram in privdata. Send it to export hook. */ 579 static int 580 export_send(priv_p priv, item_p item, int flags) 581 { 582 struct mbuf *m = NGI_M(item); 583 struct netflow_v5_export_dgram *dgram = mtod(m, 584 struct netflow_v5_export_dgram *); 585 struct netflow_v5_header *header = &dgram->header; 586 struct timespec ts; 587 int error = 0; 588 589 /* Fill mbuf header. */ 590 m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) * 591 header->count + sizeof(struct netflow_v5_header); 592 593 /* Fill export header. */ 594 header->sys_uptime = htonl(MILLIUPTIME(time_uptime)); 595 getnanotime(&ts); 596 header->unix_secs = htonl(ts.tv_sec); 597 header->unix_nsecs = htonl(ts.tv_nsec); 598 header->engine_type = 0; 599 header->engine_id = 0; 600 header->pad = 0; 601 header->flow_seq = htonl(atomic_fetchadd_32(&priv->flow_seq, 602 header->count)); 603 header->count = htons(header->count); 604 605 if (priv->export != NULL) 606 NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags); 607 else 608 NG_FREE_ITEM(item); 609 610 return (error); 611 } 612 613 614 /* Add export record to dgram. */ 615 static int 616 export_add(item_p item, struct flow_entry *fle) 617 { 618 struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item), 619 struct netflow_v5_export_dgram *); 620 struct netflow_v5_header *header = &dgram->header; 621 struct netflow_v5_record *rec; 622 623 rec = &dgram->r[header->count]; 624 header->count ++; 625 626 KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS, 627 ("ng_netflow: export too big")); 628 629 /* Fill in export record. */ 630 rec->src_addr = fle->f.r.r_src.s_addr; 631 rec->dst_addr = fle->f.r.r_dst.s_addr; 632 rec->next_hop = fle->f.next_hop.s_addr; 633 rec->i_ifx = htons(fle->f.fle_i_ifx); 634 rec->o_ifx = htons(fle->f.fle_o_ifx); 635 rec->packets = htonl(fle->f.packets); 636 rec->octets = htonl(fle->f.bytes); 637 rec->first = htonl(MILLIUPTIME(fle->f.first)); 638 rec->last = htonl(MILLIUPTIME(fle->f.last)); 639 rec->s_port = fle->f.r.r_sport; 640 rec->d_port = fle->f.r.r_dport; 641 rec->flags = fle->f.tcp_flags; 642 rec->prot = fle->f.r.r_ip_p; 643 rec->tos = fle->f.r.r_tos; 644 rec->dst_mask = fle->f.dst_mask; 645 rec->src_mask = fle->f.src_mask; 646 647 /* Not supported fields. */ 648 rec->src_as = rec->dst_as = 0; 649 650 if (header->count == NETFLOW_V5_MAX_RECORDS) 651 return (1); /* end of datagram */ 652 else 653 return (0); 654 } 655 656 /* Periodic flow expiry run. */ 657 void 658 ng_netflow_expire(void *arg) 659 { 660 struct flow_entry *fle, *fle1; 661 struct flow_hash_entry *hsh; 662 priv_p priv = (priv_p )arg; 663 item_p item = NULL; 664 uint32_t used; 665 int i; 666 667 /* 668 * Going through all the cache. 669 */ 670 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) { 671 /* 672 * Skip entries, that are already being worked on. 673 */ 674 if (mtx_trylock(&hsh->mtx) == 0) 675 continue; 676 677 used = atomic_load_acq_32(&priv->info.nfinfo_used); 678 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) { 679 /* 680 * Interrupt thread wants this entry! 681 * Quick! Quick! Bail out! 682 */ 683 if (mtx_contested(&hsh->mtx)) 684 break; 685 686 /* 687 * Don't expire aggressively while hash collision 688 * ratio is predicted small. 689 */ 690 if (used <= (NBUCKETS*2) && !INACTIVE(fle)) 691 break; 692 693 if ((INACTIVE(fle) && (SMALL(fle) || 694 (used > (NBUCKETS*2)))) || AGED(fle)) { 695 TAILQ_REMOVE(&hsh->head, fle, fle_hash); 696 expire_flow(priv, &item, fle, NG_NOFLAGS); 697 used--; 698 atomic_add_32(&priv->info.nfinfo_inact_exp, 1); 699 } 700 } 701 mtx_unlock(&hsh->mtx); 702 } 703 704 if (item != NULL) 705 return_export_dgram(priv, item, NG_NOFLAGS); 706 707 /* Schedule next expire. */ 708 callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire, 709 (void *)priv); 710 } 711