xref: /freebsd/sys/netinet/siftr.c (revision 1d386b48a555f61cb7325543adbbb5c3f3407a66)
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 #include <sys/param.h>
65 #include <sys/alq.h>
66 #include <sys/errno.h>
67 #include <sys/eventhandler.h>
68 #include <sys/hash.h>
69 #include <sys/kernel.h>
70 #include <sys/kthread.h>
71 #include <sys/lock.h>
72 #include <sys/mbuf.h>
73 #include <sys/module.h>
74 #include <sys/mutex.h>
75 #include <sys/pcpu.h>
76 #include <sys/proc.h>
77 #include <sys/sbuf.h>
78 #include <sys/sdt.h>
79 #include <sys/smp.h>
80 #include <sys/socket.h>
81 #include <sys/socketvar.h>
82 #include <sys/sysctl.h>
83 #include <sys/unistd.h>
84 
85 #include <net/if.h>
86 #include <net/if_var.h>
87 #include <net/pfil.h>
88 #include <net/route.h>
89 
90 #include <netinet/in.h>
91 #include <netinet/in_kdtrace.h>
92 #include <netinet/in_fib.h>
93 #include <netinet/in_pcb.h>
94 #include <netinet/in_systm.h>
95 #include <netinet/in_var.h>
96 #include <netinet/ip.h>
97 #include <netinet/ip_var.h>
98 #include <netinet/tcp_var.h>
99 
100 #ifdef SIFTR_IPV6
101 #include <netinet/ip6.h>
102 #include <netinet6/ip6_var.h>
103 #include <netinet6/in6_fib.h>
104 #include <netinet6/in6_pcb.h>
105 #endif /* SIFTR_IPV6 */
106 
107 #include <machine/in_cksum.h>
108 
109 /*
110  * Three digit version number refers to X.Y.Z where:
111  * X is the major version number
112  * Y is bumped to mark backwards incompatible changes
113  * Z is bumped to mark backwards compatible changes
114  */
115 #define V_MAJOR		1
116 #define V_BACKBREAK	3
117 #define V_BACKCOMPAT	0
118 #define MODVERSION	__CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT))
119 #define MODVERSION_STR	__XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \
120     __XSTRING(V_BACKCOMPAT)
121 
122 #define HOOK 0
123 #define UNHOOK 1
124 #define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536
125 #define SYS_NAME "FreeBSD"
126 #define PACKET_TAG_SIFTR 100
127 #define PACKET_COOKIE_SIFTR 21749576
128 #define SIFTR_LOG_FILE_MODE 0644
129 #define SIFTR_DISABLE 0
130 #define SIFTR_ENABLE 1
131 
132 /*
133  * Hard upper limit on the length of log messages. Bump this up if you add new
134  * data fields such that the line length could exceed the below value.
135  */
136 #define MAX_LOG_MSG_LEN 300
137 /* XXX: Make this a sysctl tunable. */
138 #define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN)
139 
140 #ifdef SIFTR_IPV6
141 #define SIFTR_IPMODE 6
142 #else
143 #define SIFTR_IPMODE 4
144 #endif
145 
146 static MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR");
147 static MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode",
148     "SIFTR pkt_node struct");
149 static MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode",
150     "SIFTR flow_hash_node struct");
151 
152 /* Used as links in the pkt manager queue. */
153 struct pkt_node {
154 	/* Timestamp of pkt as noted in the pfil hook. */
155 	struct timeval		tval;
156 	/* Direction pkt is travelling. */
157 	enum {
158 		DIR_IN = 0,
159 		DIR_OUT = 1,
160 	}			direction;
161 	/* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */
162 	uint8_t			ipver;
163 	/* Local TCP port. */
164 	uint16_t		lport;
165 	/* Foreign TCP port. */
166 	uint16_t		fport;
167 	/* Local address. */
168 	union in_dependaddr	laddr;
169 	/* Foreign address. */
170 	union in_dependaddr	faddr;
171 	/* Congestion Window (bytes). */
172 	uint32_t		snd_cwnd;
173 	/* Sending Window (bytes). */
174 	uint32_t		snd_wnd;
175 	/* Receive Window (bytes). */
176 	uint32_t		rcv_wnd;
177 	/* More tcpcb flags storage */
178 	uint32_t		t_flags2;
179 	/* Slow Start Threshold (bytes). */
180 	uint32_t		snd_ssthresh;
181 	/* Current state of the TCP FSM. */
182 	int			conn_state;
183 	/* Max Segment Size (bytes). */
184 	uint32_t		mss;
185 	/* Smoothed RTT (usecs). */
186 	uint32_t		srtt;
187 	/* Is SACK enabled? */
188 	u_char			sack_enabled;
189 	/* Window scaling for snd window. */
190 	u_char			snd_scale;
191 	/* Window scaling for recv window. */
192 	u_char			rcv_scale;
193 	/* TCP control block flags. */
194 	u_int			t_flags;
195 	/* Retransmission timeout (usec). */
196 	uint32_t		rto;
197 	/* Size of the TCP send buffer in bytes. */
198 	u_int			snd_buf_hiwater;
199 	/* Current num bytes in the send socket buffer. */
200 	u_int			snd_buf_cc;
201 	/* Size of the TCP receive buffer in bytes. */
202 	u_int			rcv_buf_hiwater;
203 	/* Current num bytes in the receive socket buffer. */
204 	u_int			rcv_buf_cc;
205 	/* Number of bytes inflight that we are waiting on ACKs for. */
206 	u_int			sent_inflight_bytes;
207 	/* Number of segments currently in the reassembly queue. */
208 	int			t_segqlen;
209 	/* Flowid for the connection. */
210 	u_int			flowid;
211 	/* Flow type for the connection. */
212 	u_int			flowtype;
213 	/* Link to next pkt_node in the list. */
214 	STAILQ_ENTRY(pkt_node)	nodes;
215 };
216 
217 struct flow_info
218 {
219 #ifdef SIFTR_IPV6
220 	char	laddr[INET6_ADDRSTRLEN];	/* local IP address */
221 	char	faddr[INET6_ADDRSTRLEN];	/* foreign IP address */
222 #else
223 	char	laddr[INET_ADDRSTRLEN];		/* local IP address */
224 	char	faddr[INET_ADDRSTRLEN];		/* foreign IP address */
225 #endif
226 	uint16_t	lport;			/* local TCP port */
227 	uint16_t	fport;			/* foreign TCP port */
228 	uint32_t	key;			/* flowid of the connection */
229 };
230 
231 struct flow_hash_node
232 {
233 	uint16_t counter;
234 	struct flow_info const_info;		/* constant connection info */
235 	LIST_ENTRY(flow_hash_node) nodes;
236 };
237 
238 struct siftr_stats
239 {
240 	/* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */
241 	uint64_t n_in;
242 	uint64_t n_out;
243 	/* # pkts skipped due to failed malloc calls. */
244 	uint32_t nskip_in_malloc;
245 	uint32_t nskip_out_malloc;
246 	/* # pkts skipped due to failed inpcb lookups. */
247 	uint32_t nskip_in_inpcb;
248 	uint32_t nskip_out_inpcb;
249 	/* # pkts skipped due to failed tcpcb lookups. */
250 	uint32_t nskip_in_tcpcb;
251 	uint32_t nskip_out_tcpcb;
252 	/* # pkts skipped due to stack reinjection. */
253 	uint32_t nskip_in_dejavu;
254 	uint32_t nskip_out_dejavu;
255 };
256 
257 DPCPU_DEFINE_STATIC(struct siftr_stats, ss);
258 
259 static volatile unsigned int siftr_exit_pkt_manager_thread = 0;
260 static unsigned int siftr_enabled = 0;
261 static unsigned int siftr_pkts_per_log = 1;
262 static uint16_t     siftr_port_filter = 0;
263 /* static unsigned int siftr_binary_log = 0; */
264 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log";
265 static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log";
266 static u_long siftr_hashmask;
267 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue);
268 LIST_HEAD(listhead, flow_hash_node) *counter_hash;
269 static int wait_for_pkt;
270 static struct alq *siftr_alq = NULL;
271 static struct mtx siftr_pkt_queue_mtx;
272 static struct mtx siftr_pkt_mgr_mtx;
273 static struct thread *siftr_pkt_manager_thr = NULL;
274 static char direction[2] = {'i','o'};
275 
276 /* Required function prototypes. */
277 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS);
278 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS);
279 
280 /* Declare the net.inet.siftr sysctl tree and populate it. */
281 
282 SYSCTL_DECL(_net_inet_siftr);
283 
284 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
285     "siftr related settings");
286 
287 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled,
288     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
289     &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU",
290     "switch siftr module operations on/off");
291 
292 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile,
293     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &siftr_logfile_shadow,
294     sizeof(siftr_logfile_shadow), &siftr_sysctl_logfile_name_handler, "A",
295     "file to save siftr log messages to");
296 
297 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW,
298     &siftr_pkts_per_log, 1,
299     "number of packets between generating a log message");
300 
301 SYSCTL_U16(_net_inet_siftr, OID_AUTO, port_filter, CTLFLAG_RW,
302     &siftr_port_filter, 0,
303     "enable packet filter on a TCP port");
304 
305 /* XXX: TODO
306 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW,
307     &siftr_binary_log, 0,
308     "write log files in binary instead of ascii");
309 */
310 
311 /* Begin functions. */
312 
313 static inline struct flow_hash_node *
314 siftr_find_flow(struct listhead *counter_list, uint32_t id)
315 {
316 	struct flow_hash_node *hash_node;
317 	/*
318 	 * If the list is not empty i.e. the hash index has
319 	 * been used by another flow previously.
320 	 */
321 	if (LIST_FIRST(counter_list) != NULL) {
322 		/*
323 		 * Loop through the hash nodes in the list.
324 		 * There should normally only be 1 hash node in the list.
325 		 */
326 		LIST_FOREACH(hash_node, counter_list, nodes) {
327 			/*
328 			 * Check if the key for the pkt we are currently
329 			 * processing is the same as the key stored in the
330 			 * hash node we are currently processing.
331 			 * If they are the same, then we've found the
332 			 * hash node that stores the counter for the flow
333 			 * the pkt belongs to.
334 			 */
335 			if (hash_node->const_info.key == id) {
336 				return hash_node;
337 			}
338 		}
339 	}
340 
341 	return NULL;
342 }
343 
344 static inline struct flow_hash_node *
345 siftr_new_hash_node(struct flow_info info, int dir,
346 		    struct siftr_stats *ss)
347 {
348 	struct flow_hash_node *hash_node;
349 	struct listhead *counter_list;
350 
351 	counter_list = counter_hash + (info.key & siftr_hashmask);
352 	/* Create a new hash node to store the flow's constant info. */
353 	hash_node = malloc(sizeof(struct flow_hash_node), M_SIFTR_HASHNODE,
354 			   M_NOWAIT|M_ZERO);
355 
356 	if (hash_node != NULL) {
357 		/* Initialise our new hash node list entry. */
358 		hash_node->counter = 0;
359 		hash_node->const_info = info;
360 		LIST_INSERT_HEAD(counter_list, hash_node, nodes);
361 		return hash_node;
362 	} else {
363 		/* malloc failed */
364 		if (dir == DIR_IN)
365 			ss->nskip_in_malloc++;
366 		else
367 			ss->nskip_out_malloc++;
368 
369 		return NULL;
370 	}
371 }
372 
373 static void
374 siftr_process_pkt(struct pkt_node * pkt_node)
375 {
376 	struct flow_hash_node *hash_node;
377 	struct listhead *counter_list;
378 	struct ale *log_buf;
379 
380 	if (pkt_node->flowid == 0) {
381 		panic("%s: flowid not available", __func__);
382 	}
383 
384 	counter_list = counter_hash + (pkt_node->flowid & siftr_hashmask);
385 	hash_node = siftr_find_flow(counter_list, pkt_node->flowid);
386 
387 	if (hash_node == NULL) {
388 		return;
389 	} else if (siftr_pkts_per_log > 1) {
390 		/*
391 		 * Taking the remainder of the counter divided
392 		 * by the current value of siftr_pkts_per_log
393 		 * and storing that in counter provides a neat
394 		 * way to modulate the frequency of log
395 		 * messages being written to the log file.
396 		 */
397 		hash_node->counter = (hash_node->counter + 1) %
398 				     siftr_pkts_per_log;
399 		/*
400 		 * If we have not seen enough packets since the last time
401 		 * we wrote a log message for this connection, return.
402 		 */
403 		if (hash_node->counter > 0)
404 			return;
405 	}
406 
407 	log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK);
408 
409 	if (log_buf == NULL)
410 		return; /* Should only happen if the ALQ is shutting down. */
411 
412 	/* Construct a log message. */
413 	log_buf->ae_bytesused = snprintf(log_buf->ae_data, MAX_LOG_MSG_LEN,
414 	    "%c,%jd.%06ld,%s,%hu,%s,%hu,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,"
415 	    "%u,%u,%u,%u,%u,%u,%u,%u\n",
416 	    direction[pkt_node->direction],
417 	    (intmax_t)pkt_node->tval.tv_sec,
418 	    pkt_node->tval.tv_usec,
419 	    hash_node->const_info.laddr,
420 	    hash_node->const_info.lport,
421 	    hash_node->const_info.faddr,
422 	    hash_node->const_info.fport,
423 	    pkt_node->snd_ssthresh,
424 	    pkt_node->snd_cwnd,
425 	    pkt_node->t_flags2,
426 	    pkt_node->snd_wnd,
427 	    pkt_node->rcv_wnd,
428 	    pkt_node->snd_scale,
429 	    pkt_node->rcv_scale,
430 	    pkt_node->conn_state,
431 	    pkt_node->mss,
432 	    pkt_node->srtt,
433 	    pkt_node->sack_enabled,
434 	    pkt_node->t_flags,
435 	    pkt_node->rto,
436 	    pkt_node->snd_buf_hiwater,
437 	    pkt_node->snd_buf_cc,
438 	    pkt_node->rcv_buf_hiwater,
439 	    pkt_node->rcv_buf_cc,
440 	    pkt_node->sent_inflight_bytes,
441 	    pkt_node->t_segqlen,
442 	    pkt_node->flowid,
443 	    pkt_node->flowtype);
444 
445 	alq_post_flags(siftr_alq, log_buf, 0);
446 }
447 
448 static void
449 siftr_pkt_manager_thread(void *arg)
450 {
451 	STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
452 	    STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
453 	struct pkt_node *pkt_node, *pkt_node_temp;
454 	uint8_t draining;
455 
456 	draining = 2;
457 
458 	mtx_lock(&siftr_pkt_mgr_mtx);
459 
460 	/* draining == 0 when queue has been flushed and it's safe to exit. */
461 	while (draining) {
462 		/*
463 		 * Sleep until we are signalled to wake because thread has
464 		 * been told to exit or until 1 tick has passed.
465 		 */
466 		mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
467 		    1);
468 
469 		/* Gain exclusive access to the pkt_node queue. */
470 		mtx_lock(&siftr_pkt_queue_mtx);
471 
472 		/*
473 		 * Move pkt_queue to tmp_pkt_queue, which leaves
474 		 * pkt_queue empty and ready to receive more pkt_nodes.
475 		 */
476 		STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
477 
478 		/*
479 		 * We've finished making changes to the list. Unlock it
480 		 * so the pfil hooks can continue queuing pkt_nodes.
481 		 */
482 		mtx_unlock(&siftr_pkt_queue_mtx);
483 
484 		/*
485 		 * We can't hold a mutex whilst calling siftr_process_pkt
486 		 * because ALQ might sleep waiting for buffer space.
487 		 */
488 		mtx_unlock(&siftr_pkt_mgr_mtx);
489 
490 		/* Flush all pkt_nodes to the log file. */
491 		STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes,
492 		    pkt_node_temp) {
493 			siftr_process_pkt(pkt_node);
494 			STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
495 			free(pkt_node, M_SIFTR_PKTNODE);
496 		}
497 
498 		KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
499 		    ("SIFTR tmp_pkt_queue not empty after flush"));
500 
501 		mtx_lock(&siftr_pkt_mgr_mtx);
502 
503 		/*
504 		 * If siftr_exit_pkt_manager_thread gets set during the window
505 		 * where we are draining the tmp_pkt_queue above, there might
506 		 * still be pkts in pkt_queue that need to be drained.
507 		 * Allow one further iteration to occur after
508 		 * siftr_exit_pkt_manager_thread has been set to ensure
509 		 * pkt_queue is completely empty before we kill the thread.
510 		 *
511 		 * siftr_exit_pkt_manager_thread is set only after the pfil
512 		 * hooks have been removed, so only 1 extra iteration
513 		 * is needed to drain the queue.
514 		 */
515 		if (siftr_exit_pkt_manager_thread)
516 			draining--;
517 	}
518 
519 	mtx_unlock(&siftr_pkt_mgr_mtx);
520 
521 	/* Calls wakeup on this thread's struct thread ptr. */
522 	kthread_exit();
523 }
524 
525 /*
526  * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
527  * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
528  * Return value >0 means the caller should skip processing this mbuf.
529  */
530 static inline int
531 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
532 {
533 	if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
534 	    != NULL) {
535 		if (dir == PFIL_IN)
536 			ss->nskip_in_dejavu++;
537 		else
538 			ss->nskip_out_dejavu++;
539 
540 		return (1);
541 	} else {
542 		struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
543 		    PACKET_TAG_SIFTR, 0, M_NOWAIT);
544 		if (tag == NULL) {
545 			if (dir == PFIL_IN)
546 				ss->nskip_in_malloc++;
547 			else
548 				ss->nskip_out_malloc++;
549 
550 			return (1);
551 		}
552 
553 		m_tag_prepend(m, tag);
554 	}
555 
556 	return (0);
557 }
558 
559 /*
560  * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
561  * otherwise.
562  */
563 static inline struct inpcb *
564 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
565     uint16_t dport, int dir, struct siftr_stats *ss)
566 {
567 	struct inpcb *inp;
568 
569 	/* We need the tcbinfo lock. */
570 	INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
571 
572 	if (dir == PFIL_IN)
573 		inp = (ipver == INP_IPV4 ?
574 		    in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
575 		    dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
576 		    :
577 #ifdef SIFTR_IPV6
578 		    in6_pcblookup(&V_tcbinfo,
579 		    &((struct ip6_hdr *)ip)->ip6_src, sport,
580 		    &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB,
581 		    m->m_pkthdr.rcvif)
582 #else
583 		    NULL
584 #endif
585 		    );
586 
587 	else
588 		inp = (ipver == INP_IPV4 ?
589 		    in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
590 		    sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
591 		    :
592 #ifdef SIFTR_IPV6
593 		    in6_pcblookup(&V_tcbinfo,
594 		    &((struct ip6_hdr *)ip)->ip6_dst, dport,
595 		    &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB,
596 		    m->m_pkthdr.rcvif)
597 #else
598 		    NULL
599 #endif
600 		    );
601 
602 	/* If we can't find the inpcb, bail. */
603 	if (inp == NULL) {
604 		if (dir == PFIL_IN)
605 			ss->nskip_in_inpcb++;
606 		else
607 			ss->nskip_out_inpcb++;
608 	}
609 
610 	return (inp);
611 }
612 
613 static inline uint32_t
614 siftr_get_flowid(struct inpcb *inp, int ipver, uint32_t *phashtype)
615 {
616 	if (inp->inp_flowid == 0) {
617 #ifdef SIFTR_IPV6
618 		if (ipver == INP_IPV6) {
619 			return fib6_calc_packet_hash(&inp->in6p_laddr,
620 						     &inp->in6p_faddr,
621 						     inp->inp_lport,
622 						     inp->inp_fport,
623 						     IPPROTO_TCP,
624 						     phashtype);
625 		} else
626 #endif
627 		{
628 			return fib4_calc_packet_hash(inp->inp_laddr,
629 						     inp->inp_faddr,
630 						     inp->inp_lport,
631 						     inp->inp_fport,
632 						     IPPROTO_TCP,
633 						     phashtype);
634 		}
635 	} else {
636 		*phashtype = inp->inp_flowtype;
637 		return inp->inp_flowid;
638 	}
639 }
640 
641 static inline void
642 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
643     int ipver, int dir, int inp_locally_locked)
644 {
645 	pn->ipver = ipver;
646 	pn->lport = inp->inp_lport;
647 	pn->fport = inp->inp_fport;
648 	pn->laddr = inp->inp_inc.inc_ie.ie_dependladdr;
649 	pn->faddr = inp->inp_inc.inc_ie.ie_dependfaddr;
650 	pn->snd_cwnd = tp->snd_cwnd;
651 	pn->snd_wnd = tp->snd_wnd;
652 	pn->rcv_wnd = tp->rcv_wnd;
653 	pn->t_flags2 = tp->t_flags2;
654 	pn->snd_ssthresh = tp->snd_ssthresh;
655 	pn->snd_scale = tp->snd_scale;
656 	pn->rcv_scale = tp->rcv_scale;
657 	pn->conn_state = tp->t_state;
658 	pn->mss = tp->t_maxseg;
659 	pn->srtt = ((uint64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
660 	pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
661 	pn->t_flags = tp->t_flags;
662 	pn->rto = tp->t_rxtcur * tick;
663 	pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
664 	pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd);
665 	pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
666 	pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv);
667 	pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
668 	pn->t_segqlen = tp->t_segqlen;
669 
670 	/* We've finished accessing the tcb so release the lock. */
671 	if (inp_locally_locked)
672 		INP_RUNLOCK(inp);
673 
674 	pn->direction = (dir == PFIL_IN ? DIR_IN : DIR_OUT);
675 
676 	/*
677 	 * Significantly more accurate than using getmicrotime(), but slower!
678 	 * Gives true microsecond resolution at the expense of a hit to
679 	 * maximum pps throughput processing when SIFTR is loaded and enabled.
680 	 */
681 	microtime(&pn->tval);
682 	TCP_PROBE1(siftr, pn);
683 }
684 
685 /*
686  * pfil hook that is called for each IPv4 packet making its way through the
687  * stack in either direction.
688  * The pfil subsystem holds a non-sleepable mutex somewhere when
689  * calling our hook function, so we can't sleep at all.
690  * It's very important to use the M_NOWAIT flag with all function calls
691  * that support it so that they won't sleep, otherwise you get a panic.
692  */
693 static pfil_return_t
694 siftr_chkpkt(struct mbuf **m, struct ifnet *ifp, int flags,
695     void *ruleset __unused, struct inpcb *inp)
696 {
697 	struct pkt_node *pn;
698 	struct ip *ip;
699 	struct tcphdr *th;
700 	struct tcpcb *tp;
701 	struct siftr_stats *ss;
702 	unsigned int ip_hl;
703 	int inp_locally_locked, dir;
704 	uint32_t hash_id, hash_type;
705 	struct listhead *counter_list;
706 	struct flow_hash_node *hash_node;
707 
708 	inp_locally_locked = 0;
709 	dir = PFIL_DIR(flags);
710 	ss = DPCPU_PTR(ss);
711 
712 	/*
713 	 * m_pullup is not required here because ip_{input|output}
714 	 * already do the heavy lifting for us.
715 	 */
716 
717 	ip = mtod(*m, struct ip *);
718 
719 	/* Only continue processing if the packet is TCP. */
720 	if (ip->ip_p != IPPROTO_TCP)
721 		goto ret;
722 
723 	/*
724 	 * Create a tcphdr struct starting at the correct offset
725 	 * in the IP packet. ip->ip_hl gives the ip header length
726 	 * in 4-byte words, so multiply it to get the size in bytes.
727 	 */
728 	ip_hl = (ip->ip_hl << 2);
729 	th = (struct tcphdr *)((caddr_t)ip + ip_hl);
730 
731 	/*
732 	 * Only pkts selected by the tcp port filter
733 	 * can be inserted into the pkt_queue
734 	 */
735 	if ((siftr_port_filter != 0) &&
736 	    (siftr_port_filter != ntohs(th->th_sport)) &&
737 	    (siftr_port_filter != ntohs(th->th_dport))) {
738 		goto ret;
739 	}
740 
741 	/*
742 	 * If a kernel subsystem reinjects packets into the stack, our pfil
743 	 * hook will be called multiple times for the same packet.
744 	 * Make sure we only process unique packets.
745 	 */
746 	if (siftr_chkreinject(*m, dir, ss))
747 		goto ret;
748 
749 	if (dir == PFIL_IN)
750 		ss->n_in++;
751 	else
752 		ss->n_out++;
753 
754 	/*
755 	 * If the pfil hooks don't provide a pointer to the
756 	 * inpcb, we need to find it ourselves and lock it.
757 	 */
758 	if (!inp) {
759 		/* Find the corresponding inpcb for this pkt. */
760 		inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
761 		    th->th_dport, dir, ss);
762 
763 		if (inp == NULL)
764 			goto ret;
765 		else
766 			inp_locally_locked = 1;
767 	}
768 
769 	INP_LOCK_ASSERT(inp);
770 
771 	/* Find the TCP control block that corresponds with this packet */
772 	tp = intotcpcb(inp);
773 
774 	/*
775 	 * If we can't find the TCP control block (happens occasionaly for a
776 	 * packet sent during the shutdown phase of a TCP connection), or the
777 	 * TCP control block has not initialized (happens during TCPS_SYN_SENT),
778 	 * bail.
779 	 */
780 	if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) {
781 		if (dir == PFIL_IN)
782 			ss->nskip_in_tcpcb++;
783 		else
784 			ss->nskip_out_tcpcb++;
785 
786 		goto inp_unlock;
787 	}
788 
789 	hash_id = siftr_get_flowid(inp, INP_IPV4, &hash_type);
790 	counter_list = counter_hash + (hash_id & siftr_hashmask);
791 	hash_node = siftr_find_flow(counter_list, hash_id);
792 
793 	/* If this flow hasn't been seen before, we create a new entry. */
794 	if (hash_node == NULL) {
795 		struct flow_info info;
796 
797 		inet_ntoa_r(inp->inp_laddr, info.laddr);
798 		inet_ntoa_r(inp->inp_faddr, info.faddr);
799 		info.lport = ntohs(inp->inp_lport);
800 		info.fport = ntohs(inp->inp_fport);
801 		info.key = hash_id;
802 
803 		hash_node = siftr_new_hash_node(info, dir, ss);
804 	}
805 
806 	if (hash_node == NULL) {
807 		goto inp_unlock;
808 	}
809 
810 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
811 
812 	if (pn == NULL) {
813 		if (dir == PFIL_IN)
814 			ss->nskip_in_malloc++;
815 		else
816 			ss->nskip_out_malloc++;
817 
818 		goto inp_unlock;
819 	}
820 
821 	pn->flowid = hash_id;
822 	pn->flowtype = hash_type;
823 
824 	siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
825 
826 	mtx_lock(&siftr_pkt_queue_mtx);
827 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
828 	mtx_unlock(&siftr_pkt_queue_mtx);
829 	goto ret;
830 
831 inp_unlock:
832 	if (inp_locally_locked)
833 		INP_RUNLOCK(inp);
834 
835 ret:
836 	return (PFIL_PASS);
837 }
838 
839 #ifdef SIFTR_IPV6
840 static pfil_return_t
841 siftr_chkpkt6(struct mbuf **m, struct ifnet *ifp, int flags,
842     void *ruleset __unused, struct inpcb *inp)
843 {
844 	struct pkt_node *pn;
845 	struct ip6_hdr *ip6;
846 	struct tcphdr *th;
847 	struct tcpcb *tp;
848 	struct siftr_stats *ss;
849 	unsigned int ip6_hl;
850 	int inp_locally_locked, dir;
851 	uint32_t hash_id, hash_type;
852 	struct listhead *counter_list;
853 	struct flow_hash_node *hash_node;
854 
855 	inp_locally_locked = 0;
856 	dir = PFIL_DIR(flags);
857 	ss = DPCPU_PTR(ss);
858 
859 	/*
860 	 * m_pullup is not required here because ip6_{input|output}
861 	 * already do the heavy lifting for us.
862 	 */
863 
864 	ip6 = mtod(*m, struct ip6_hdr *);
865 
866 	/*
867 	 * Only continue processing if the packet is TCP
868 	 * XXX: We should follow the next header fields
869 	 * as shown on Pg 6 RFC 2460, but right now we'll
870 	 * only check pkts that have no extension headers.
871 	 */
872 	if (ip6->ip6_nxt != IPPROTO_TCP)
873 		goto ret6;
874 
875 	/*
876 	 * Create a tcphdr struct starting at the correct offset
877 	 * in the ipv6 packet.
878 	 */
879 	ip6_hl = sizeof(struct ip6_hdr);
880 	th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
881 
882 	/*
883 	 * Only pkts selected by the tcp port filter
884 	 * can be inserted into the pkt_queue
885 	 */
886 	if ((siftr_port_filter != 0) &&
887 	    (siftr_port_filter != ntohs(th->th_sport)) &&
888 	    (siftr_port_filter != ntohs(th->th_dport))) {
889 		goto ret6;
890 	}
891 
892 	/*
893 	 * If a kernel subsystem reinjects packets into the stack, our pfil
894 	 * hook will be called multiple times for the same packet.
895 	 * Make sure we only process unique packets.
896 	 */
897 	if (siftr_chkreinject(*m, dir, ss))
898 		goto ret6;
899 
900 	if (dir == PFIL_IN)
901 		ss->n_in++;
902 	else
903 		ss->n_out++;
904 
905 	/*
906 	 * For inbound packets, the pfil hooks don't provide a pointer to the
907 	 * inpcb, so we need to find it ourselves and lock it.
908 	 */
909 	if (!inp) {
910 		/* Find the corresponding inpcb for this pkt. */
911 		inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
912 		    th->th_sport, th->th_dport, dir, ss);
913 
914 		if (inp == NULL)
915 			goto ret6;
916 		else
917 			inp_locally_locked = 1;
918 	}
919 
920 	/* Find the TCP control block that corresponds with this packet. */
921 	tp = intotcpcb(inp);
922 
923 	/*
924 	 * If we can't find the TCP control block (happens occasionaly for a
925 	 * packet sent during the shutdown phase of a TCP connection), or the
926 	 * TCP control block has not initialized (happens during TCPS_SYN_SENT),
927 	 * bail.
928 	 */
929 	if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) {
930 		if (dir == PFIL_IN)
931 			ss->nskip_in_tcpcb++;
932 		else
933 			ss->nskip_out_tcpcb++;
934 
935 		goto inp_unlock6;
936 	}
937 
938 	hash_id = siftr_get_flowid(inp, INP_IPV6, &hash_type);
939 	counter_list = counter_hash + (hash_id & siftr_hashmask);
940 	hash_node = siftr_find_flow(counter_list, hash_id);
941 
942 	/* If this flow hasn't been seen before, we create a new entry. */
943 	if (!hash_node) {
944 		struct flow_info info;
945 
946 		ip6_sprintf(info.laddr, &inp->in6p_laddr);
947 		ip6_sprintf(info.faddr, &inp->in6p_faddr);
948 		info.lport = ntohs(inp->inp_lport);
949 		info.fport = ntohs(inp->inp_fport);
950 		info.key = hash_id;
951 
952 		hash_node = siftr_new_hash_node(info, dir, ss);
953 	}
954 
955 	if (!hash_node) {
956 		goto inp_unlock6;
957 	}
958 
959 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
960 
961 	if (pn == NULL) {
962 		if (dir == PFIL_IN)
963 			ss->nskip_in_malloc++;
964 		else
965 			ss->nskip_out_malloc++;
966 
967 		goto inp_unlock6;
968 	}
969 
970 	pn->flowid = hash_id;
971 	pn->flowtype = hash_type;
972 
973 	siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
974 
975 	mtx_lock(&siftr_pkt_queue_mtx);
976 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
977 	mtx_unlock(&siftr_pkt_queue_mtx);
978 	goto ret6;
979 
980 inp_unlock6:
981 	if (inp_locally_locked)
982 		INP_RUNLOCK(inp);
983 
984 ret6:
985 	return (PFIL_PASS);
986 }
987 #endif /* #ifdef SIFTR_IPV6 */
988 
989 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet_hook);
990 #define	V_siftr_inet_hook	VNET(siftr_inet_hook)
991 #ifdef SIFTR_IPV6
992 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet6_hook);
993 #define	V_siftr_inet6_hook	VNET(siftr_inet6_hook)
994 #endif
995 static int
996 siftr_pfil(int action)
997 {
998 	struct pfil_hook_args pha = {
999 		.pa_version = PFIL_VERSION,
1000 		.pa_flags = PFIL_IN | PFIL_OUT,
1001 		.pa_modname = "siftr",
1002 		.pa_rulname = "default",
1003 	};
1004 	struct pfil_link_args pla = {
1005 		.pa_version = PFIL_VERSION,
1006 		.pa_flags = PFIL_IN | PFIL_OUT | PFIL_HEADPTR | PFIL_HOOKPTR,
1007 	};
1008 
1009 	VNET_ITERATOR_DECL(vnet_iter);
1010 
1011 	VNET_LIST_RLOCK();
1012 	VNET_FOREACH(vnet_iter) {
1013 		CURVNET_SET(vnet_iter);
1014 
1015 		if (action == HOOK) {
1016 			pha.pa_mbuf_chk = siftr_chkpkt;
1017 			pha.pa_type = PFIL_TYPE_IP4;
1018 			V_siftr_inet_hook = pfil_add_hook(&pha);
1019 			pla.pa_hook = V_siftr_inet_hook;
1020 			pla.pa_head = V_inet_pfil_head;
1021 			(void)pfil_link(&pla);
1022 #ifdef SIFTR_IPV6
1023 			pha.pa_mbuf_chk = siftr_chkpkt6;
1024 			pha.pa_type = PFIL_TYPE_IP6;
1025 			V_siftr_inet6_hook = pfil_add_hook(&pha);
1026 			pla.pa_hook = V_siftr_inet6_hook;
1027 			pla.pa_head = V_inet6_pfil_head;
1028 			(void)pfil_link(&pla);
1029 #endif
1030 		} else if (action == UNHOOK) {
1031 			pfil_remove_hook(V_siftr_inet_hook);
1032 #ifdef SIFTR_IPV6
1033 			pfil_remove_hook(V_siftr_inet6_hook);
1034 #endif
1035 		}
1036 		CURVNET_RESTORE();
1037 	}
1038 	VNET_LIST_RUNLOCK();
1039 
1040 	return (0);
1041 }
1042 
1043 static int
1044 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1045 {
1046 	struct alq *new_alq;
1047 	int error;
1048 
1049 	error = sysctl_handle_string(oidp, arg1, arg2, req);
1050 
1051 	/* Check for error or same filename */
1052 	if (error != 0 || req->newptr == NULL ||
1053 	    strncmp(siftr_logfile, arg1, arg2) == 0)
1054 		goto done;
1055 
1056 	/* file name changed */
1057 	error = alq_open(&new_alq, arg1, curthread->td_ucred,
1058 	    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1059 	if (error != 0)
1060 		goto done;
1061 
1062 	/*
1063 	 * If disabled, siftr_alq == NULL so we simply close
1064 	 * the alq as we've proved it can be opened.
1065 	 * If enabled, close the existing alq and switch the old
1066 	 * for the new.
1067 	 */
1068 	if (siftr_alq == NULL) {
1069 		alq_close(new_alq);
1070 	} else {
1071 		alq_close(siftr_alq);
1072 		siftr_alq = new_alq;
1073 	}
1074 
1075 	/* Update filename upon success */
1076 	strlcpy(siftr_logfile, arg1, arg2);
1077 done:
1078 	return (error);
1079 }
1080 
1081 static int
1082 siftr_manage_ops(uint8_t action)
1083 {
1084 	struct siftr_stats totalss;
1085 	struct timeval tval;
1086 	struct flow_hash_node *counter, *tmp_counter;
1087 	struct sbuf *s;
1088 	int i, error;
1089 	uint32_t bytes_to_write, total_skipped_pkts;
1090 
1091 	error = 0;
1092 	total_skipped_pkts = 0;
1093 
1094 	/* Init an autosizing sbuf that initially holds 200 chars. */
1095 	if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1096 		return (-1);
1097 
1098 	if (action == SIFTR_ENABLE && siftr_pkt_manager_thr == NULL) {
1099 		/*
1100 		 * Create our alq
1101 		 * XXX: We should abort if alq_open fails!
1102 		 */
1103 		alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1104 		    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1105 
1106 		STAILQ_INIT(&pkt_queue);
1107 
1108 		DPCPU_ZERO(ss);
1109 
1110 		siftr_exit_pkt_manager_thread = 0;
1111 
1112 		kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1113 		    &siftr_pkt_manager_thr, RFNOWAIT, 0,
1114 		    "siftr_pkt_manager_thr");
1115 
1116 		siftr_pfil(HOOK);
1117 
1118 		microtime(&tval);
1119 
1120 		sbuf_printf(s,
1121 		    "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1122 		    "siftrver=%s\tsysname=%s\tsysver=%u\tipmode=%u\n",
1123 		    (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR,
1124 		    SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1125 
1126 		sbuf_finish(s);
1127 		alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1128 
1129 	} else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1130 		/*
1131 		 * Remove the pfil hook functions. All threads currently in
1132 		 * the hook functions are allowed to exit before siftr_pfil()
1133 		 * returns.
1134 		 */
1135 		siftr_pfil(UNHOOK);
1136 
1137 		/* This will block until the pkt manager thread unlocks it. */
1138 		mtx_lock(&siftr_pkt_mgr_mtx);
1139 
1140 		/* Tell the pkt manager thread that it should exit now. */
1141 		siftr_exit_pkt_manager_thread = 1;
1142 
1143 		/*
1144 		 * Wake the pkt_manager thread so it realises that
1145 		 * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1146 		 * The wakeup won't be delivered until we unlock
1147 		 * siftr_pkt_mgr_mtx so this isn't racy.
1148 		 */
1149 		wakeup(&wait_for_pkt);
1150 
1151 		/* Wait for the pkt_manager thread to exit. */
1152 		mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1153 		    "thrwait", 0);
1154 
1155 		siftr_pkt_manager_thr = NULL;
1156 		mtx_unlock(&siftr_pkt_mgr_mtx);
1157 
1158 		totalss.n_in = DPCPU_VARSUM(ss, n_in);
1159 		totalss.n_out = DPCPU_VARSUM(ss, n_out);
1160 		totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1161 		totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1162 		totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1163 		totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1164 		totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1165 		totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1166 
1167 		total_skipped_pkts = totalss.nskip_in_malloc +
1168 		    totalss.nskip_out_malloc + totalss.nskip_in_tcpcb +
1169 		    totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1170 		    totalss.nskip_out_inpcb;
1171 
1172 		microtime(&tval);
1173 
1174 		sbuf_printf(s,
1175 		    "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1176 		    "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1177 		    "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1178 		    "num_outbound_skipped_pkts_malloc=%u\t"
1179 		    "num_inbound_skipped_pkts_tcpcb=%u\t"
1180 		    "num_outbound_skipped_pkts_tcpcb=%u\t"
1181 		    "num_inbound_skipped_pkts_inpcb=%u\t"
1182 		    "num_outbound_skipped_pkts_inpcb=%u\t"
1183 		    "total_skipped_tcp_pkts=%u\tflow_list=",
1184 		    (intmax_t)tval.tv_sec,
1185 		    tval.tv_usec,
1186 		    (uintmax_t)totalss.n_in,
1187 		    (uintmax_t)totalss.n_out,
1188 		    (uintmax_t)(totalss.n_in + totalss.n_out),
1189 		    totalss.nskip_in_malloc,
1190 		    totalss.nskip_out_malloc,
1191 		    totalss.nskip_in_tcpcb,
1192 		    totalss.nskip_out_tcpcb,
1193 		    totalss.nskip_in_inpcb,
1194 		    totalss.nskip_out_inpcb,
1195 		    total_skipped_pkts);
1196 
1197 		/*
1198 		 * Iterate over the flow hash, printing a summary of each
1199 		 * flow seen and freeing any malloc'd memory.
1200 		 * The hash consists of an array of LISTs (man 3 queue).
1201 		 */
1202 		for (i = 0; i <= siftr_hashmask; i++) {
1203 			LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1204 			    tmp_counter) {
1205 				sbuf_printf(s, "%s;%hu-%s;%hu,",
1206 					    counter->const_info.laddr,
1207 					    counter->const_info.lport,
1208 					    counter->const_info.faddr,
1209 					    counter->const_info.fport);
1210 
1211 				free(counter, M_SIFTR_HASHNODE);
1212 			}
1213 
1214 			LIST_INIT(counter_hash + i);
1215 		}
1216 
1217 		sbuf_printf(s, "\n");
1218 		sbuf_finish(s);
1219 
1220 		i = 0;
1221 		do {
1222 			bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1223 			alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1224 			i += bytes_to_write;
1225 		} while (i < sbuf_len(s));
1226 
1227 		alq_close(siftr_alq);
1228 		siftr_alq = NULL;
1229 	} else
1230 		error = EINVAL;
1231 
1232 	sbuf_delete(s);
1233 
1234 	/*
1235 	 * XXX: Should be using ret to check if any functions fail
1236 	 * and set error appropriately
1237 	 */
1238 
1239 	return (error);
1240 }
1241 
1242 static int
1243 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1244 {
1245 	int error;
1246 	uint32_t new;
1247 
1248 	new = siftr_enabled;
1249 	error = sysctl_handle_int(oidp, &new, 0, req);
1250 	if (error == 0 && req->newptr != NULL) {
1251 		if (new > 1)
1252 			return (EINVAL);
1253 		else if (new != siftr_enabled) {
1254 			if ((error = siftr_manage_ops(new)) == 0) {
1255 				siftr_enabled = new;
1256 			} else {
1257 				siftr_manage_ops(SIFTR_DISABLE);
1258 			}
1259 		}
1260 	}
1261 
1262 	return (error);
1263 }
1264 
1265 static void
1266 siftr_shutdown_handler(void *arg)
1267 {
1268 	if (siftr_enabled == 1) {
1269 		siftr_manage_ops(SIFTR_DISABLE);
1270 	}
1271 }
1272 
1273 /*
1274  * Module is being unloaded or machine is shutting down. Take care of cleanup.
1275  */
1276 static int
1277 deinit_siftr(void)
1278 {
1279 	/* Cleanup. */
1280 	siftr_manage_ops(SIFTR_DISABLE);
1281 	hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1282 	mtx_destroy(&siftr_pkt_queue_mtx);
1283 	mtx_destroy(&siftr_pkt_mgr_mtx);
1284 
1285 	return (0);
1286 }
1287 
1288 /*
1289  * Module has just been loaded into the kernel.
1290  */
1291 static int
1292 init_siftr(void)
1293 {
1294 	EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL,
1295 	    SHUTDOWN_PRI_FIRST);
1296 
1297 	/* Initialise our flow counter hash table. */
1298 	counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1299 	    &siftr_hashmask);
1300 
1301 	mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1302 	mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1303 
1304 	/* Print message to the user's current terminal. */
1305 	uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1306 	    "          http://caia.swin.edu.au/urp/newtcp\n\n",
1307 	    MODVERSION_STR);
1308 
1309 	return (0);
1310 }
1311 
1312 /*
1313  * This is the function that is called to load and unload the module.
1314  * When the module is loaded, this function is called once with
1315  * "what" == MOD_LOAD
1316  * When the module is unloaded, this function is called twice with
1317  * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1318  * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1319  * this function is called once with "what" = MOD_SHUTDOWN
1320  * When the system is shut down, the handler isn't called until the very end
1321  * of the shutdown sequence i.e. after the disks have been synced.
1322  */
1323 static int
1324 siftr_load_handler(module_t mod, int what, void *arg)
1325 {
1326 	int ret;
1327 
1328 	switch (what) {
1329 	case MOD_LOAD:
1330 		ret = init_siftr();
1331 		break;
1332 
1333 	case MOD_QUIESCE:
1334 	case MOD_SHUTDOWN:
1335 		ret = deinit_siftr();
1336 		break;
1337 
1338 	case MOD_UNLOAD:
1339 		ret = 0;
1340 		break;
1341 
1342 	default:
1343 		ret = EINVAL;
1344 		break;
1345 	}
1346 
1347 	return (ret);
1348 }
1349 
1350 static moduledata_t siftr_mod = {
1351 	.name = "siftr",
1352 	.evhand = siftr_load_handler,
1353 };
1354 
1355 /*
1356  * Param 1: name of the kernel module
1357  * Param 2: moduledata_t struct containing info about the kernel module
1358  *          and the execution entry point for the module
1359  * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1360  *          Defines the module initialisation order
1361  * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1362  *          Defines the initialisation order of this kld relative to others
1363  *          within the same subsystem as defined by param 3
1364  */
1365 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_LAST, SI_ORDER_ANY);
1366 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1367 MODULE_VERSION(siftr, MODVERSION);
1368