1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2007-2009 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010, The FreeBSD Foundation 7 * All rights reserved. 8 * 9 * Portions of this software were developed at the Centre for Advanced 10 * Internet Architectures, Swinburne University of Technology, Melbourne, 11 * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 /****************************************************** 36 * Statistical Information For TCP Research (SIFTR) 37 * 38 * A FreeBSD kernel module that adds very basic intrumentation to the 39 * TCP stack, allowing internal stats to be recorded to a log file 40 * for experimental, debugging and performance analysis purposes. 41 * 42 * SIFTR was first released in 2007 by James Healy and Lawrence Stewart whilst 43 * working on the NewTCP research project at Swinburne University of 44 * Technology's Centre for Advanced Internet Architectures, Melbourne, 45 * Australia, which was made possible in part by a grant from the Cisco 46 * University Research Program Fund at Community Foundation Silicon Valley. 47 * More details are available at: 48 * http://caia.swin.edu.au/urp/newtcp/ 49 * 50 * Work on SIFTR v1.2.x was sponsored by the FreeBSD Foundation as part of 51 * the "Enhancing the FreeBSD TCP Implementation" project 2008-2009. 52 * More details are available at: 53 * http://www.freebsdfoundation.org/ 54 * http://caia.swin.edu.au/freebsd/etcp09/ 55 * 56 * Lawrence Stewart is the current maintainer, and all contact regarding 57 * SIFTR should be directed to him via email: lastewart@swin.edu.au 58 * 59 * Initial release date: June 2007 60 * Most recent update: September 2010 61 ******************************************************/ 62 63 #include <sys/cdefs.h> 64 __FBSDID("$FreeBSD$"); 65 66 #include <sys/param.h> 67 #include <sys/alq.h> 68 #include <sys/errno.h> 69 #include <sys/eventhandler.h> 70 #include <sys/hash.h> 71 #include <sys/kernel.h> 72 #include <sys/kthread.h> 73 #include <sys/lock.h> 74 #include <sys/mbuf.h> 75 #include <sys/module.h> 76 #include <sys/mutex.h> 77 #include <sys/pcpu.h> 78 #include <sys/proc.h> 79 #include <sys/sbuf.h> 80 #include <sys/sdt.h> 81 #include <sys/smp.h> 82 #include <sys/socket.h> 83 #include <sys/socketvar.h> 84 #include <sys/sysctl.h> 85 #include <sys/unistd.h> 86 87 #include <net/if.h> 88 #include <net/if_var.h> 89 #include <net/pfil.h> 90 #include <net/route.h> 91 92 #include <netinet/in.h> 93 #include <netinet/in_kdtrace.h> 94 #include <netinet/in_fib.h> 95 #include <netinet/in_pcb.h> 96 #include <netinet/in_systm.h> 97 #include <netinet/in_var.h> 98 #include <netinet/ip.h> 99 #include <netinet/ip_var.h> 100 #include <netinet/tcp_var.h> 101 102 #ifdef SIFTR_IPV6 103 #include <netinet/ip6.h> 104 #include <netinet6/ip6_var.h> 105 #include <netinet6/in6_fib.h> 106 #include <netinet6/in6_pcb.h> 107 #endif /* SIFTR_IPV6 */ 108 109 #include <machine/in_cksum.h> 110 111 /* 112 * Three digit version number refers to X.Y.Z where: 113 * X is the major version number 114 * Y is bumped to mark backwards incompatible changes 115 * Z is bumped to mark backwards compatible changes 116 */ 117 #define V_MAJOR 1 118 #define V_BACKBREAK 3 119 #define V_BACKCOMPAT 0 120 #define MODVERSION __CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT)) 121 #define MODVERSION_STR __XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \ 122 __XSTRING(V_BACKCOMPAT) 123 124 #define HOOK 0 125 #define UNHOOK 1 126 #define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536 127 #define SYS_NAME "FreeBSD" 128 #define PACKET_TAG_SIFTR 100 129 #define PACKET_COOKIE_SIFTR 21749576 130 #define SIFTR_LOG_FILE_MODE 0644 131 #define SIFTR_DISABLE 0 132 #define SIFTR_ENABLE 1 133 134 /* 135 * Hard upper limit on the length of log messages. Bump this up if you add new 136 * data fields such that the line length could exceed the below value. 137 */ 138 #define MAX_LOG_MSG_LEN 300 139 /* XXX: Make this a sysctl tunable. */ 140 #define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN) 141 142 #ifdef SIFTR_IPV6 143 #define SIFTR_IPMODE 6 144 #else 145 #define SIFTR_IPMODE 4 146 #endif 147 148 static MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR"); 149 static MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode", 150 "SIFTR pkt_node struct"); 151 static MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode", 152 "SIFTR flow_hash_node struct"); 153 154 /* Used as links in the pkt manager queue. */ 155 struct pkt_node { 156 /* Timestamp of pkt as noted in the pfil hook. */ 157 struct timeval tval; 158 /* Direction pkt is travelling. */ 159 enum { 160 DIR_IN = 0, 161 DIR_OUT = 1, 162 } direction; 163 /* Congestion Window (bytes). */ 164 uint32_t snd_cwnd; 165 /* Sending Window (bytes). */ 166 uint32_t snd_wnd; 167 /* Receive Window (bytes). */ 168 uint32_t rcv_wnd; 169 /* More tcpcb flags storage */ 170 uint32_t t_flags2; 171 /* Slow Start Threshold (bytes). */ 172 uint32_t snd_ssthresh; 173 /* Current state of the TCP FSM. */ 174 int conn_state; 175 /* Max Segment Size (bytes). */ 176 u_int max_seg_size; 177 /* Smoothed RTT (usecs). */ 178 uint32_t srtt; 179 /* Is SACK enabled? */ 180 u_char sack_enabled; 181 /* Window scaling for snd window. */ 182 u_char snd_scale; 183 /* Window scaling for recv window. */ 184 u_char rcv_scale; 185 /* TCP control block flags. */ 186 u_int flags; 187 /* Retransmission timeout (usec). */ 188 uint32_t rto; 189 /* Size of the TCP send buffer in bytes. */ 190 u_int snd_buf_hiwater; 191 /* Current num bytes in the send socket buffer. */ 192 u_int snd_buf_cc; 193 /* Size of the TCP receive buffer in bytes. */ 194 u_int rcv_buf_hiwater; 195 /* Current num bytes in the receive socket buffer. */ 196 u_int rcv_buf_cc; 197 /* Number of bytes inflight that we are waiting on ACKs for. */ 198 u_int sent_inflight_bytes; 199 /* Number of segments currently in the reassembly queue. */ 200 int t_segqlen; 201 /* Flowid for the connection. */ 202 u_int flowid; 203 /* Flow type for the connection. */ 204 u_int flowtype; 205 /* Link to next pkt_node in the list. */ 206 STAILQ_ENTRY(pkt_node) nodes; 207 }; 208 209 struct flow_info 210 { 211 #ifdef SIFTR_IPV6 212 char laddr[INET6_ADDRSTRLEN]; /* local IP address */ 213 char faddr[INET6_ADDRSTRLEN]; /* foreign IP address */ 214 #else 215 char laddr[INET_ADDRSTRLEN]; /* local IP address */ 216 char faddr[INET_ADDRSTRLEN]; /* foreign IP address */ 217 #endif 218 uint16_t lport; /* local TCP port */ 219 uint16_t fport; /* foreign TCP port */ 220 uint8_t ipver; /* IP version */ 221 uint32_t key; /* flowid of the connection */ 222 }; 223 224 struct flow_hash_node 225 { 226 uint16_t counter; 227 struct flow_info const_info; /* constant connection info */ 228 LIST_ENTRY(flow_hash_node) nodes; 229 }; 230 231 struct siftr_stats 232 { 233 /* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */ 234 uint64_t n_in; 235 uint64_t n_out; 236 /* # pkts skipped due to failed malloc calls. */ 237 uint32_t nskip_in_malloc; 238 uint32_t nskip_out_malloc; 239 /* # pkts skipped due to failed inpcb lookups. */ 240 uint32_t nskip_in_inpcb; 241 uint32_t nskip_out_inpcb; 242 /* # pkts skipped due to failed tcpcb lookups. */ 243 uint32_t nskip_in_tcpcb; 244 uint32_t nskip_out_tcpcb; 245 /* # pkts skipped due to stack reinjection. */ 246 uint32_t nskip_in_dejavu; 247 uint32_t nskip_out_dejavu; 248 }; 249 250 DPCPU_DEFINE_STATIC(struct siftr_stats, ss); 251 252 static volatile unsigned int siftr_exit_pkt_manager_thread = 0; 253 static unsigned int siftr_enabled = 0; 254 static unsigned int siftr_pkts_per_log = 1; 255 static uint16_t siftr_port_filter = 0; 256 /* static unsigned int siftr_binary_log = 0; */ 257 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log"; 258 static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log"; 259 static u_long siftr_hashmask; 260 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue); 261 LIST_HEAD(listhead, flow_hash_node) *counter_hash; 262 static int wait_for_pkt; 263 static struct alq *siftr_alq = NULL; 264 static struct mtx siftr_pkt_queue_mtx; 265 static struct mtx siftr_pkt_mgr_mtx; 266 static struct thread *siftr_pkt_manager_thr = NULL; 267 static char direction[2] = {'i','o'}; 268 269 /* Required function prototypes. */ 270 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS); 271 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS); 272 273 /* Declare the net.inet.siftr sysctl tree and populate it. */ 274 275 SYSCTL_DECL(_net_inet_siftr); 276 277 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 278 "siftr related settings"); 279 280 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, 281 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 282 &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU", 283 "switch siftr module operations on/off"); 284 285 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, 286 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &siftr_logfile_shadow, 287 sizeof(siftr_logfile_shadow), &siftr_sysctl_logfile_name_handler, "A", 288 "file to save siftr log messages to"); 289 290 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW, 291 &siftr_pkts_per_log, 1, 292 "number of packets between generating a log message"); 293 294 SYSCTL_U16(_net_inet_siftr, OID_AUTO, port_filter, CTLFLAG_RW, 295 &siftr_port_filter, 0, 296 "enable packet filter on a TCP port"); 297 298 /* XXX: TODO 299 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW, 300 &siftr_binary_log, 0, 301 "write log files in binary instead of ascii"); 302 */ 303 304 /* Begin functions. */ 305 306 static inline struct flow_hash_node * 307 siftr_find_flow(struct listhead *counter_list, uint32_t id) 308 { 309 struct flow_hash_node *hash_node; 310 /* 311 * If the list is not empty i.e. the hash index has 312 * been used by another flow previously. 313 */ 314 if (LIST_FIRST(counter_list) != NULL) { 315 /* 316 * Loop through the hash nodes in the list. 317 * There should normally only be 1 hash node in the list. 318 */ 319 LIST_FOREACH(hash_node, counter_list, nodes) { 320 /* 321 * Check if the key for the pkt we are currently 322 * processing is the same as the key stored in the 323 * hash node we are currently processing. 324 * If they are the same, then we've found the 325 * hash node that stores the counter for the flow 326 * the pkt belongs to. 327 */ 328 if (hash_node->const_info.key == id) { 329 return hash_node; 330 } 331 } 332 } 333 334 return NULL; 335 } 336 337 static inline struct flow_hash_node * 338 siftr_new_hash_node(struct flow_info info, int dir, 339 struct siftr_stats *ss) 340 { 341 struct flow_hash_node *hash_node; 342 struct listhead *counter_list; 343 344 counter_list = counter_hash + (info.key & siftr_hashmask); 345 /* Create a new hash node to store the flow's constant info. */ 346 hash_node = malloc(sizeof(struct flow_hash_node), M_SIFTR_HASHNODE, 347 M_NOWAIT|M_ZERO); 348 349 if (hash_node != NULL) { 350 /* Initialise our new hash node list entry. */ 351 hash_node->counter = 0; 352 hash_node->const_info = info; 353 LIST_INSERT_HEAD(counter_list, hash_node, nodes); 354 return hash_node; 355 } else { 356 /* malloc failed */ 357 if (dir == DIR_IN) 358 ss->nskip_in_malloc++; 359 else 360 ss->nskip_out_malloc++; 361 362 return NULL; 363 } 364 } 365 366 static void 367 siftr_process_pkt(struct pkt_node * pkt_node) 368 { 369 struct flow_hash_node *hash_node; 370 struct listhead *counter_list; 371 struct ale *log_buf; 372 373 if (pkt_node->flowid == 0) { 374 panic("%s: flowid not available", __func__); 375 } 376 377 counter_list = counter_hash + (pkt_node->flowid & siftr_hashmask); 378 hash_node = siftr_find_flow(counter_list, pkt_node->flowid); 379 380 if (hash_node == NULL) { 381 return; 382 } 383 384 log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK); 385 386 if (log_buf == NULL) 387 return; /* Should only happen if the ALQ is shutting down. */ 388 389 /* Construct a log message. */ 390 log_buf->ae_bytesused = snprintf(log_buf->ae_data, MAX_LOG_MSG_LEN, 391 "%c,%zd.%06ld,%s,%hu,%s,%hu,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u," 392 "%u,%u,%u,%u,%u,%u,%u,%u\n", 393 direction[pkt_node->direction], 394 pkt_node->tval.tv_sec, 395 pkt_node->tval.tv_usec, 396 hash_node->const_info.laddr, 397 hash_node->const_info.lport, 398 hash_node->const_info.faddr, 399 hash_node->const_info.fport, 400 pkt_node->snd_ssthresh, 401 pkt_node->snd_cwnd, 402 pkt_node->t_flags2, 403 pkt_node->snd_wnd, 404 pkt_node->rcv_wnd, 405 pkt_node->snd_scale, 406 pkt_node->rcv_scale, 407 pkt_node->conn_state, 408 pkt_node->max_seg_size, 409 pkt_node->srtt, 410 pkt_node->sack_enabled, 411 pkt_node->flags, 412 pkt_node->rto, 413 pkt_node->snd_buf_hiwater, 414 pkt_node->snd_buf_cc, 415 pkt_node->rcv_buf_hiwater, 416 pkt_node->rcv_buf_cc, 417 pkt_node->sent_inflight_bytes, 418 pkt_node->t_segqlen, 419 pkt_node->flowid, 420 pkt_node->flowtype); 421 422 alq_post_flags(siftr_alq, log_buf, 0); 423 } 424 425 static void 426 siftr_pkt_manager_thread(void *arg) 427 { 428 STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue = 429 STAILQ_HEAD_INITIALIZER(tmp_pkt_queue); 430 struct pkt_node *pkt_node, *pkt_node_temp; 431 uint8_t draining; 432 433 draining = 2; 434 435 mtx_lock(&siftr_pkt_mgr_mtx); 436 437 /* draining == 0 when queue has been flushed and it's safe to exit. */ 438 while (draining) { 439 /* 440 * Sleep until we are signalled to wake because thread has 441 * been told to exit or until 1 tick has passed. 442 */ 443 mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait", 444 1); 445 446 /* Gain exclusive access to the pkt_node queue. */ 447 mtx_lock(&siftr_pkt_queue_mtx); 448 449 /* 450 * Move pkt_queue to tmp_pkt_queue, which leaves 451 * pkt_queue empty and ready to receive more pkt_nodes. 452 */ 453 STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue); 454 455 /* 456 * We've finished making changes to the list. Unlock it 457 * so the pfil hooks can continue queuing pkt_nodes. 458 */ 459 mtx_unlock(&siftr_pkt_queue_mtx); 460 461 /* 462 * We can't hold a mutex whilst calling siftr_process_pkt 463 * because ALQ might sleep waiting for buffer space. 464 */ 465 mtx_unlock(&siftr_pkt_mgr_mtx); 466 467 /* Flush all pkt_nodes to the log file. */ 468 STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes, 469 pkt_node_temp) { 470 siftr_process_pkt(pkt_node); 471 STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes); 472 free(pkt_node, M_SIFTR_PKTNODE); 473 } 474 475 KASSERT(STAILQ_EMPTY(&tmp_pkt_queue), 476 ("SIFTR tmp_pkt_queue not empty after flush")); 477 478 mtx_lock(&siftr_pkt_mgr_mtx); 479 480 /* 481 * If siftr_exit_pkt_manager_thread gets set during the window 482 * where we are draining the tmp_pkt_queue above, there might 483 * still be pkts in pkt_queue that need to be drained. 484 * Allow one further iteration to occur after 485 * siftr_exit_pkt_manager_thread has been set to ensure 486 * pkt_queue is completely empty before we kill the thread. 487 * 488 * siftr_exit_pkt_manager_thread is set only after the pfil 489 * hooks have been removed, so only 1 extra iteration 490 * is needed to drain the queue. 491 */ 492 if (siftr_exit_pkt_manager_thread) 493 draining--; 494 } 495 496 mtx_unlock(&siftr_pkt_mgr_mtx); 497 498 /* Calls wakeup on this thread's struct thread ptr. */ 499 kthread_exit(); 500 } 501 502 /* 503 * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that 504 * it's a reinjected packet and return. If it doesn't, tag the mbuf and return. 505 * Return value >0 means the caller should skip processing this mbuf. 506 */ 507 static inline int 508 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss) 509 { 510 if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL) 511 != NULL) { 512 if (dir == PFIL_IN) 513 ss->nskip_in_dejavu++; 514 else 515 ss->nskip_out_dejavu++; 516 517 return (1); 518 } else { 519 struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR, 520 PACKET_TAG_SIFTR, 0, M_NOWAIT); 521 if (tag == NULL) { 522 if (dir == PFIL_IN) 523 ss->nskip_in_malloc++; 524 else 525 ss->nskip_out_malloc++; 526 527 return (1); 528 } 529 530 m_tag_prepend(m, tag); 531 } 532 533 return (0); 534 } 535 536 /* 537 * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL 538 * otherwise. 539 */ 540 static inline struct inpcb * 541 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport, 542 uint16_t dport, int dir, struct siftr_stats *ss) 543 { 544 struct inpcb *inp; 545 546 /* We need the tcbinfo lock. */ 547 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 548 549 if (dir == PFIL_IN) 550 inp = (ipver == INP_IPV4 ? 551 in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst, 552 dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif) 553 : 554 #ifdef SIFTR_IPV6 555 in6_pcblookup(&V_tcbinfo, 556 &((struct ip6_hdr *)ip)->ip6_src, sport, 557 &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB, 558 m->m_pkthdr.rcvif) 559 #else 560 NULL 561 #endif 562 ); 563 564 else 565 inp = (ipver == INP_IPV4 ? 566 in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src, 567 sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif) 568 : 569 #ifdef SIFTR_IPV6 570 in6_pcblookup(&V_tcbinfo, 571 &((struct ip6_hdr *)ip)->ip6_dst, dport, 572 &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB, 573 m->m_pkthdr.rcvif) 574 #else 575 NULL 576 #endif 577 ); 578 579 /* If we can't find the inpcb, bail. */ 580 if (inp == NULL) { 581 if (dir == PFIL_IN) 582 ss->nskip_in_inpcb++; 583 else 584 ss->nskip_out_inpcb++; 585 } 586 587 return (inp); 588 } 589 590 static inline uint32_t 591 siftr_get_flowid(struct inpcb *inp, int ipver, uint32_t *phashtype) 592 { 593 if (inp->inp_flowid == 0) { 594 #ifdef SIFTR_IPV6 595 if (ipver == INP_IPV6) { 596 return fib6_calc_packet_hash(&inp->in6p_laddr, 597 &inp->in6p_faddr, 598 inp->inp_lport, 599 inp->inp_fport, 600 IPPROTO_TCP, 601 phashtype); 602 } else 603 #endif 604 { 605 return fib4_calc_packet_hash(inp->inp_laddr, 606 inp->inp_faddr, 607 inp->inp_lport, 608 inp->inp_fport, 609 IPPROTO_TCP, 610 phashtype); 611 } 612 } else { 613 *phashtype = inp->inp_flowtype; 614 return inp->inp_flowid; 615 } 616 } 617 618 static inline void 619 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp, 620 int ipver, int dir, int inp_locally_locked) 621 { 622 pn->snd_cwnd = tp->snd_cwnd; 623 pn->snd_wnd = tp->snd_wnd; 624 pn->rcv_wnd = tp->rcv_wnd; 625 pn->t_flags2 = tp->t_flags2; 626 pn->snd_ssthresh = tp->snd_ssthresh; 627 pn->snd_scale = tp->snd_scale; 628 pn->rcv_scale = tp->rcv_scale; 629 pn->conn_state = tp->t_state; 630 pn->max_seg_size = tp->t_maxseg; 631 pn->srtt = ((uint64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT; 632 pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0; 633 pn->flags = tp->t_flags; 634 pn->rto = tp->t_rxtcur * tick; 635 pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat; 636 pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd); 637 pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat; 638 pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv); 639 pn->sent_inflight_bytes = tp->snd_max - tp->snd_una; 640 pn->t_segqlen = tp->t_segqlen; 641 642 /* We've finished accessing the tcb so release the lock. */ 643 if (inp_locally_locked) 644 INP_RUNLOCK(inp); 645 646 pn->direction = (dir == PFIL_IN ? DIR_IN : DIR_OUT); 647 648 /* 649 * Significantly more accurate than using getmicrotime(), but slower! 650 * Gives true microsecond resolution at the expense of a hit to 651 * maximum pps throughput processing when SIFTR is loaded and enabled. 652 */ 653 microtime(&pn->tval); 654 TCP_PROBE1(siftr, &pn); 655 } 656 657 /* 658 * pfil hook that is called for each IPv4 packet making its way through the 659 * stack in either direction. 660 * The pfil subsystem holds a non-sleepable mutex somewhere when 661 * calling our hook function, so we can't sleep at all. 662 * It's very important to use the M_NOWAIT flag with all function calls 663 * that support it so that they won't sleep, otherwise you get a panic. 664 */ 665 static pfil_return_t 666 siftr_chkpkt(struct mbuf **m, struct ifnet *ifp, int flags, 667 void *ruleset __unused, struct inpcb *inp) 668 { 669 struct pkt_node *pn; 670 struct ip *ip; 671 struct tcphdr *th; 672 struct tcpcb *tp; 673 struct siftr_stats *ss; 674 unsigned int ip_hl; 675 int inp_locally_locked, dir; 676 uint32_t hash_id, hash_type; 677 struct listhead *counter_list; 678 struct flow_hash_node *hash_node; 679 680 inp_locally_locked = 0; 681 dir = PFIL_DIR(flags); 682 ss = DPCPU_PTR(ss); 683 684 /* 685 * m_pullup is not required here because ip_{input|output} 686 * already do the heavy lifting for us. 687 */ 688 689 ip = mtod(*m, struct ip *); 690 691 /* Only continue processing if the packet is TCP. */ 692 if (ip->ip_p != IPPROTO_TCP) 693 goto ret; 694 695 /* 696 * Create a tcphdr struct starting at the correct offset 697 * in the IP packet. ip->ip_hl gives the ip header length 698 * in 4-byte words, so multiply it to get the size in bytes. 699 */ 700 ip_hl = (ip->ip_hl << 2); 701 th = (struct tcphdr *)((caddr_t)ip + ip_hl); 702 703 /* 704 * Only pkts selected by the tcp port filter 705 * can be inserted into the pkt_queue 706 */ 707 if ((siftr_port_filter != 0) && 708 (siftr_port_filter != ntohs(th->th_sport)) && 709 (siftr_port_filter != ntohs(th->th_dport))) { 710 goto ret; 711 } 712 713 /* 714 * If a kernel subsystem reinjects packets into the stack, our pfil 715 * hook will be called multiple times for the same packet. 716 * Make sure we only process unique packets. 717 */ 718 if (siftr_chkreinject(*m, dir, ss)) 719 goto ret; 720 721 if (dir == PFIL_IN) 722 ss->n_in++; 723 else 724 ss->n_out++; 725 726 /* 727 * If the pfil hooks don't provide a pointer to the 728 * inpcb, we need to find it ourselves and lock it. 729 */ 730 if (!inp) { 731 /* Find the corresponding inpcb for this pkt. */ 732 inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport, 733 th->th_dport, dir, ss); 734 735 if (inp == NULL) 736 goto ret; 737 else 738 inp_locally_locked = 1; 739 } 740 741 INP_LOCK_ASSERT(inp); 742 743 /* Find the TCP control block that corresponds with this packet */ 744 tp = intotcpcb(inp); 745 746 /* 747 * If we can't find the TCP control block (happens occasionaly for a 748 * packet sent during the shutdown phase of a TCP connection), or the 749 * TCP control block has not initialized (happens during TCPS_SYN_SENT), 750 * bail. 751 */ 752 if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) { 753 if (dir == PFIL_IN) 754 ss->nskip_in_tcpcb++; 755 else 756 ss->nskip_out_tcpcb++; 757 758 goto inp_unlock; 759 } 760 761 hash_id = siftr_get_flowid(inp, INP_IPV4, &hash_type); 762 counter_list = counter_hash + (hash_id & siftr_hashmask); 763 hash_node = siftr_find_flow(counter_list, hash_id); 764 765 /* If this flow hasn't been seen before, we create a new entry. */ 766 if (hash_node == NULL) { 767 struct flow_info info; 768 769 inet_ntoa_r(inp->inp_laddr, info.laddr); 770 inet_ntoa_r(inp->inp_faddr, info.faddr); 771 info.lport = ntohs(inp->inp_lport); 772 info.fport = ntohs(inp->inp_fport); 773 info.key = hash_id; 774 info.ipver = INP_IPV4; 775 776 hash_node = siftr_new_hash_node(info, dir, ss); 777 } 778 779 if (hash_node == NULL) { 780 goto inp_unlock; 781 } 782 783 pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO); 784 785 if (pn == NULL) { 786 if (dir == PFIL_IN) 787 ss->nskip_in_malloc++; 788 else 789 ss->nskip_out_malloc++; 790 791 goto inp_unlock; 792 } 793 794 pn->flowid = hash_id; 795 pn->flowtype = hash_type; 796 797 siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked); 798 799 mtx_lock(&siftr_pkt_queue_mtx); 800 STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes); 801 mtx_unlock(&siftr_pkt_queue_mtx); 802 goto ret; 803 804 inp_unlock: 805 if (inp_locally_locked) 806 INP_RUNLOCK(inp); 807 808 ret: 809 return (PFIL_PASS); 810 } 811 812 #ifdef SIFTR_IPV6 813 static pfil_return_t 814 siftr_chkpkt6(struct mbuf **m, struct ifnet *ifp, int flags, 815 void *ruleset __unused, struct inpcb *inp) 816 { 817 struct pkt_node *pn; 818 struct ip6_hdr *ip6; 819 struct tcphdr *th; 820 struct tcpcb *tp; 821 struct siftr_stats *ss; 822 unsigned int ip6_hl; 823 int inp_locally_locked, dir; 824 uint32_t hash_id, hash_type; 825 struct listhead *counter_list; 826 struct flow_hash_node *hash_node; 827 828 inp_locally_locked = 0; 829 dir = PFIL_DIR(flags); 830 ss = DPCPU_PTR(ss); 831 832 /* 833 * m_pullup is not required here because ip6_{input|output} 834 * already do the heavy lifting for us. 835 */ 836 837 ip6 = mtod(*m, struct ip6_hdr *); 838 839 /* 840 * Only continue processing if the packet is TCP 841 * XXX: We should follow the next header fields 842 * as shown on Pg 6 RFC 2460, but right now we'll 843 * only check pkts that have no extension headers. 844 */ 845 if (ip6->ip6_nxt != IPPROTO_TCP) 846 goto ret6; 847 848 /* 849 * Create a tcphdr struct starting at the correct offset 850 * in the ipv6 packet. 851 */ 852 ip6_hl = sizeof(struct ip6_hdr); 853 th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl); 854 855 /* 856 * Only pkts selected by the tcp port filter 857 * can be inserted into the pkt_queue 858 */ 859 if ((siftr_port_filter != 0) && 860 (siftr_port_filter != ntohs(th->th_sport)) && 861 (siftr_port_filter != ntohs(th->th_dport))) { 862 goto ret6; 863 } 864 865 /* 866 * If a kernel subsystem reinjects packets into the stack, our pfil 867 * hook will be called multiple times for the same packet. 868 * Make sure we only process unique packets. 869 */ 870 if (siftr_chkreinject(*m, dir, ss)) 871 goto ret6; 872 873 if (dir == PFIL_IN) 874 ss->n_in++; 875 else 876 ss->n_out++; 877 878 /* 879 * For inbound packets, the pfil hooks don't provide a pointer to the 880 * inpcb, so we need to find it ourselves and lock it. 881 */ 882 if (!inp) { 883 /* Find the corresponding inpcb for this pkt. */ 884 inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m, 885 th->th_sport, th->th_dport, dir, ss); 886 887 if (inp == NULL) 888 goto ret6; 889 else 890 inp_locally_locked = 1; 891 } 892 893 /* Find the TCP control block that corresponds with this packet. */ 894 tp = intotcpcb(inp); 895 896 /* 897 * If we can't find the TCP control block (happens occasionaly for a 898 * packet sent during the shutdown phase of a TCP connection), or the 899 * TCP control block has not initialized (happens during TCPS_SYN_SENT), 900 * bail. 901 */ 902 if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) { 903 if (dir == PFIL_IN) 904 ss->nskip_in_tcpcb++; 905 else 906 ss->nskip_out_tcpcb++; 907 908 goto inp_unlock6; 909 } 910 911 hash_id = siftr_get_flowid(inp, INP_IPV6, &hash_type); 912 counter_list = counter_hash + (hash_id & siftr_hashmask); 913 hash_node = siftr_find_flow(counter_list, hash_id); 914 915 /* If this flow hasn't been seen before, we create a new entry. */ 916 if (!hash_node) { 917 struct flow_info info; 918 919 ip6_sprintf(info.laddr, &inp->in6p_laddr); 920 ip6_sprintf(info.faddr, &inp->in6p_faddr); 921 info.lport = ntohs(inp->inp_lport); 922 info.fport = ntohs(inp->inp_fport); 923 info.key = hash_id; 924 info.ipver = INP_IPV6; 925 926 hash_node = siftr_new_hash_node(info, dir, ss); 927 } 928 929 if (!hash_node) { 930 goto inp_unlock6; 931 } 932 933 pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO); 934 935 if (pn == NULL) { 936 if (dir == PFIL_IN) 937 ss->nskip_in_malloc++; 938 else 939 ss->nskip_out_malloc++; 940 941 goto inp_unlock6; 942 } 943 944 pn->flowid = hash_id; 945 pn->flowtype = hash_type; 946 947 siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked); 948 949 /* XXX: Figure out how to generate hashes for IPv6 packets. */ 950 951 mtx_lock(&siftr_pkt_queue_mtx); 952 STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes); 953 mtx_unlock(&siftr_pkt_queue_mtx); 954 goto ret6; 955 956 inp_unlock6: 957 if (inp_locally_locked) 958 INP_RUNLOCK(inp); 959 960 ret6: 961 return (PFIL_PASS); 962 } 963 #endif /* #ifdef SIFTR_IPV6 */ 964 965 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet_hook); 966 #define V_siftr_inet_hook VNET(siftr_inet_hook) 967 #ifdef SIFTR_IPV6 968 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet6_hook); 969 #define V_siftr_inet6_hook VNET(siftr_inet6_hook) 970 #endif 971 static int 972 siftr_pfil(int action) 973 { 974 struct pfil_hook_args pha = { 975 .pa_version = PFIL_VERSION, 976 .pa_flags = PFIL_IN | PFIL_OUT, 977 .pa_modname = "siftr", 978 .pa_rulname = "default", 979 }; 980 struct pfil_link_args pla = { 981 .pa_version = PFIL_VERSION, 982 .pa_flags = PFIL_IN | PFIL_OUT | PFIL_HEADPTR | PFIL_HOOKPTR, 983 }; 984 985 VNET_ITERATOR_DECL(vnet_iter); 986 987 VNET_LIST_RLOCK(); 988 VNET_FOREACH(vnet_iter) { 989 CURVNET_SET(vnet_iter); 990 991 if (action == HOOK) { 992 pha.pa_mbuf_chk = siftr_chkpkt; 993 pha.pa_type = PFIL_TYPE_IP4; 994 V_siftr_inet_hook = pfil_add_hook(&pha); 995 pla.pa_hook = V_siftr_inet_hook; 996 pla.pa_head = V_inet_pfil_head; 997 (void)pfil_link(&pla); 998 #ifdef SIFTR_IPV6 999 pha.pa_mbuf_chk = siftr_chkpkt6; 1000 pha.pa_type = PFIL_TYPE_IP6; 1001 V_siftr_inet6_hook = pfil_add_hook(&pha); 1002 pla.pa_hook = V_siftr_inet6_hook; 1003 pla.pa_head = V_inet6_pfil_head; 1004 (void)pfil_link(&pla); 1005 #endif 1006 } else if (action == UNHOOK) { 1007 pfil_remove_hook(V_siftr_inet_hook); 1008 #ifdef SIFTR_IPV6 1009 pfil_remove_hook(V_siftr_inet6_hook); 1010 #endif 1011 } 1012 CURVNET_RESTORE(); 1013 } 1014 VNET_LIST_RUNLOCK(); 1015 1016 return (0); 1017 } 1018 1019 static int 1020 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS) 1021 { 1022 struct alq *new_alq; 1023 int error; 1024 1025 error = sysctl_handle_string(oidp, arg1, arg2, req); 1026 1027 /* Check for error or same filename */ 1028 if (error != 0 || req->newptr == NULL || 1029 strncmp(siftr_logfile, arg1, arg2) == 0) 1030 goto done; 1031 1032 /* Filname changed */ 1033 error = alq_open(&new_alq, arg1, curthread->td_ucred, 1034 SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0); 1035 if (error != 0) 1036 goto done; 1037 1038 /* 1039 * If disabled, siftr_alq == NULL so we simply close 1040 * the alq as we've proved it can be opened. 1041 * If enabled, close the existing alq and switch the old 1042 * for the new. 1043 */ 1044 if (siftr_alq == NULL) { 1045 alq_close(new_alq); 1046 } else { 1047 alq_close(siftr_alq); 1048 siftr_alq = new_alq; 1049 } 1050 1051 /* Update filename upon success */ 1052 strlcpy(siftr_logfile, arg1, arg2); 1053 done: 1054 return (error); 1055 } 1056 1057 static int 1058 siftr_manage_ops(uint8_t action) 1059 { 1060 struct siftr_stats totalss; 1061 struct timeval tval; 1062 struct flow_hash_node *counter, *tmp_counter; 1063 struct sbuf *s; 1064 int i, error; 1065 uint32_t bytes_to_write, total_skipped_pkts; 1066 1067 error = 0; 1068 total_skipped_pkts = 0; 1069 1070 /* Init an autosizing sbuf that initially holds 200 chars. */ 1071 if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL) 1072 return (-1); 1073 1074 if (action == SIFTR_ENABLE && siftr_pkt_manager_thr == NULL) { 1075 /* 1076 * Create our alq 1077 * XXX: We should abort if alq_open fails! 1078 */ 1079 alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred, 1080 SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0); 1081 1082 STAILQ_INIT(&pkt_queue); 1083 1084 DPCPU_ZERO(ss); 1085 1086 siftr_exit_pkt_manager_thread = 0; 1087 1088 kthread_add(&siftr_pkt_manager_thread, NULL, NULL, 1089 &siftr_pkt_manager_thr, RFNOWAIT, 0, 1090 "siftr_pkt_manager_thr"); 1091 1092 siftr_pfil(HOOK); 1093 1094 microtime(&tval); 1095 1096 sbuf_printf(s, 1097 "enable_time_secs=%jd\tenable_time_usecs=%06ld\t" 1098 "siftrver=%s\tsysname=%s\tsysver=%u\tipmode=%u\n", 1099 (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, 1100 SYS_NAME, __FreeBSD_version, SIFTR_IPMODE); 1101 1102 sbuf_finish(s); 1103 alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK); 1104 1105 } else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) { 1106 /* 1107 * Remove the pfil hook functions. All threads currently in 1108 * the hook functions are allowed to exit before siftr_pfil() 1109 * returns. 1110 */ 1111 siftr_pfil(UNHOOK); 1112 1113 /* This will block until the pkt manager thread unlocks it. */ 1114 mtx_lock(&siftr_pkt_mgr_mtx); 1115 1116 /* Tell the pkt manager thread that it should exit now. */ 1117 siftr_exit_pkt_manager_thread = 1; 1118 1119 /* 1120 * Wake the pkt_manager thread so it realises that 1121 * siftr_exit_pkt_manager_thread == 1 and exits gracefully. 1122 * The wakeup won't be delivered until we unlock 1123 * siftr_pkt_mgr_mtx so this isn't racy. 1124 */ 1125 wakeup(&wait_for_pkt); 1126 1127 /* Wait for the pkt_manager thread to exit. */ 1128 mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT, 1129 "thrwait", 0); 1130 1131 siftr_pkt_manager_thr = NULL; 1132 mtx_unlock(&siftr_pkt_mgr_mtx); 1133 1134 totalss.n_in = DPCPU_VARSUM(ss, n_in); 1135 totalss.n_out = DPCPU_VARSUM(ss, n_out); 1136 totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc); 1137 totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc); 1138 totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb); 1139 totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb); 1140 totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb); 1141 totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb); 1142 1143 total_skipped_pkts = totalss.nskip_in_malloc + 1144 totalss.nskip_out_malloc + totalss.nskip_in_tcpcb + 1145 totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb + 1146 totalss.nskip_out_inpcb; 1147 1148 microtime(&tval); 1149 1150 sbuf_printf(s, 1151 "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t" 1152 "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t" 1153 "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t" 1154 "num_outbound_skipped_pkts_malloc=%u\t" 1155 "num_inbound_skipped_pkts_tcpcb=%u\t" 1156 "num_outbound_skipped_pkts_tcpcb=%u\t" 1157 "num_inbound_skipped_pkts_inpcb=%u\t" 1158 "num_outbound_skipped_pkts_inpcb=%u\t" 1159 "total_skipped_tcp_pkts=%u\tflow_list=", 1160 (intmax_t)tval.tv_sec, 1161 tval.tv_usec, 1162 (uintmax_t)totalss.n_in, 1163 (uintmax_t)totalss.n_out, 1164 (uintmax_t)(totalss.n_in + totalss.n_out), 1165 totalss.nskip_in_malloc, 1166 totalss.nskip_out_malloc, 1167 totalss.nskip_in_tcpcb, 1168 totalss.nskip_out_tcpcb, 1169 totalss.nskip_in_inpcb, 1170 totalss.nskip_out_inpcb, 1171 total_skipped_pkts); 1172 1173 /* 1174 * Iterate over the flow hash, printing a summary of each 1175 * flow seen and freeing any malloc'd memory. 1176 * The hash consists of an array of LISTs (man 3 queue). 1177 */ 1178 for (i = 0; i <= siftr_hashmask; i++) { 1179 LIST_FOREACH_SAFE(counter, counter_hash + i, nodes, 1180 tmp_counter) { 1181 sbuf_printf(s, "%s;%hu-%s;%hu,", 1182 counter->const_info.laddr, 1183 counter->const_info.lport, 1184 counter->const_info.faddr, 1185 counter->const_info.fport); 1186 1187 free(counter, M_SIFTR_HASHNODE); 1188 } 1189 1190 LIST_INIT(counter_hash + i); 1191 } 1192 1193 sbuf_printf(s, "\n"); 1194 sbuf_finish(s); 1195 1196 i = 0; 1197 do { 1198 bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i); 1199 alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK); 1200 i += bytes_to_write; 1201 } while (i < sbuf_len(s)); 1202 1203 alq_close(siftr_alq); 1204 siftr_alq = NULL; 1205 } else 1206 error = EINVAL; 1207 1208 sbuf_delete(s); 1209 1210 /* 1211 * XXX: Should be using ret to check if any functions fail 1212 * and set error appropriately 1213 */ 1214 1215 return (error); 1216 } 1217 1218 static int 1219 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS) 1220 { 1221 int error; 1222 uint32_t new; 1223 1224 new = siftr_enabled; 1225 error = sysctl_handle_int(oidp, &new, 0, req); 1226 if (error == 0 && req->newptr != NULL) { 1227 if (new > 1) 1228 return (EINVAL); 1229 else if (new != siftr_enabled) { 1230 if ((error = siftr_manage_ops(new)) == 0) { 1231 siftr_enabled = new; 1232 } else { 1233 siftr_manage_ops(SIFTR_DISABLE); 1234 } 1235 } 1236 } 1237 1238 return (error); 1239 } 1240 1241 static void 1242 siftr_shutdown_handler(void *arg) 1243 { 1244 if (siftr_enabled == 1) { 1245 siftr_manage_ops(SIFTR_DISABLE); 1246 } 1247 } 1248 1249 /* 1250 * Module is being unloaded or machine is shutting down. Take care of cleanup. 1251 */ 1252 static int 1253 deinit_siftr(void) 1254 { 1255 /* Cleanup. */ 1256 siftr_manage_ops(SIFTR_DISABLE); 1257 hashdestroy(counter_hash, M_SIFTR, siftr_hashmask); 1258 mtx_destroy(&siftr_pkt_queue_mtx); 1259 mtx_destroy(&siftr_pkt_mgr_mtx); 1260 1261 return (0); 1262 } 1263 1264 /* 1265 * Module has just been loaded into the kernel. 1266 */ 1267 static int 1268 init_siftr(void) 1269 { 1270 EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL, 1271 SHUTDOWN_PRI_FIRST); 1272 1273 /* Initialise our flow counter hash table. */ 1274 counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR, 1275 &siftr_hashmask); 1276 1277 mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF); 1278 mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF); 1279 1280 /* Print message to the user's current terminal. */ 1281 uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n" 1282 " http://caia.swin.edu.au/urp/newtcp\n\n", 1283 MODVERSION_STR); 1284 1285 return (0); 1286 } 1287 1288 /* 1289 * This is the function that is called to load and unload the module. 1290 * When the module is loaded, this function is called once with 1291 * "what" == MOD_LOAD 1292 * When the module is unloaded, this function is called twice with 1293 * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second 1294 * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command, 1295 * this function is called once with "what" = MOD_SHUTDOWN 1296 * When the system is shut down, the handler isn't called until the very end 1297 * of the shutdown sequence i.e. after the disks have been synced. 1298 */ 1299 static int 1300 siftr_load_handler(module_t mod, int what, void *arg) 1301 { 1302 int ret; 1303 1304 switch (what) { 1305 case MOD_LOAD: 1306 ret = init_siftr(); 1307 break; 1308 1309 case MOD_QUIESCE: 1310 case MOD_SHUTDOWN: 1311 ret = deinit_siftr(); 1312 break; 1313 1314 case MOD_UNLOAD: 1315 ret = 0; 1316 break; 1317 1318 default: 1319 ret = EINVAL; 1320 break; 1321 } 1322 1323 return (ret); 1324 } 1325 1326 static moduledata_t siftr_mod = { 1327 .name = "siftr", 1328 .evhand = siftr_load_handler, 1329 }; 1330 1331 /* 1332 * Param 1: name of the kernel module 1333 * Param 2: moduledata_t struct containing info about the kernel module 1334 * and the execution entry point for the module 1335 * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h 1336 * Defines the module initialisation order 1337 * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h 1338 * Defines the initialisation order of this kld relative to others 1339 * within the same subsystem as defined by param 3 1340 */ 1341 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_LAST, SI_ORDER_ANY); 1342 MODULE_DEPEND(siftr, alq, 1, 1, 1); 1343 MODULE_VERSION(siftr, MODVERSION); 1344