xref: /freebsd/sys/netinet/siftr.c (revision 66fd12cf4896eb08ad8e7a2627537f84ead84dd3)
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 	/* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */
164 	uint8_t			ipver;
165 	/* Local TCP port. */
166 	uint16_t		lport;
167 	/* Foreign TCP port. */
168 	uint16_t		fport;
169 	/* Local address. */
170 	union in_dependaddr	laddr;
171 	/* Foreign address. */
172 	union in_dependaddr	faddr;
173 	/* Congestion Window (bytes). */
174 	uint32_t		snd_cwnd;
175 	/* Sending Window (bytes). */
176 	uint32_t		snd_wnd;
177 	/* Receive Window (bytes). */
178 	uint32_t		rcv_wnd;
179 	/* More tcpcb flags storage */
180 	uint32_t		t_flags2;
181 	/* Slow Start Threshold (bytes). */
182 	uint32_t		snd_ssthresh;
183 	/* Current state of the TCP FSM. */
184 	int			conn_state;
185 	/* Max Segment Size (bytes). */
186 	uint32_t		mss;
187 	/* Smoothed RTT (usecs). */
188 	uint32_t		srtt;
189 	/* Is SACK enabled? */
190 	u_char			sack_enabled;
191 	/* Window scaling for snd window. */
192 	u_char			snd_scale;
193 	/* Window scaling for recv window. */
194 	u_char			rcv_scale;
195 	/* TCP control block flags. */
196 	u_int			t_flags;
197 	/* Retransmission timeout (usec). */
198 	uint32_t		rto;
199 	/* Size of the TCP send buffer in bytes. */
200 	u_int			snd_buf_hiwater;
201 	/* Current num bytes in the send socket buffer. */
202 	u_int			snd_buf_cc;
203 	/* Size of the TCP receive buffer in bytes. */
204 	u_int			rcv_buf_hiwater;
205 	/* Current num bytes in the receive socket buffer. */
206 	u_int			rcv_buf_cc;
207 	/* Number of bytes inflight that we are waiting on ACKs for. */
208 	u_int			sent_inflight_bytes;
209 	/* Number of segments currently in the reassembly queue. */
210 	int			t_segqlen;
211 	/* Flowid for the connection. */
212 	u_int			flowid;
213 	/* Flow type for the connection. */
214 	u_int			flowtype;
215 	/* Link to next pkt_node in the list. */
216 	STAILQ_ENTRY(pkt_node)	nodes;
217 };
218 
219 struct flow_info
220 {
221 #ifdef SIFTR_IPV6
222 	char	laddr[INET6_ADDRSTRLEN];	/* local IP address */
223 	char	faddr[INET6_ADDRSTRLEN];	/* foreign IP address */
224 #else
225 	char	laddr[INET_ADDRSTRLEN];		/* local IP address */
226 	char	faddr[INET_ADDRSTRLEN];		/* foreign IP address */
227 #endif
228 	uint16_t	lport;			/* local TCP port */
229 	uint16_t	fport;			/* foreign TCP port */
230 	uint32_t	key;			/* flowid of the connection */
231 };
232 
233 struct flow_hash_node
234 {
235 	uint16_t counter;
236 	struct flow_info const_info;		/* constant connection info */
237 	LIST_ENTRY(flow_hash_node) nodes;
238 };
239 
240 struct siftr_stats
241 {
242 	/* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */
243 	uint64_t n_in;
244 	uint64_t n_out;
245 	/* # pkts skipped due to failed malloc calls. */
246 	uint32_t nskip_in_malloc;
247 	uint32_t nskip_out_malloc;
248 	/* # pkts skipped due to failed inpcb lookups. */
249 	uint32_t nskip_in_inpcb;
250 	uint32_t nskip_out_inpcb;
251 	/* # pkts skipped due to failed tcpcb lookups. */
252 	uint32_t nskip_in_tcpcb;
253 	uint32_t nskip_out_tcpcb;
254 	/* # pkts skipped due to stack reinjection. */
255 	uint32_t nskip_in_dejavu;
256 	uint32_t nskip_out_dejavu;
257 };
258 
259 DPCPU_DEFINE_STATIC(struct siftr_stats, ss);
260 
261 static volatile unsigned int siftr_exit_pkt_manager_thread = 0;
262 static unsigned int siftr_enabled = 0;
263 static unsigned int siftr_pkts_per_log = 1;
264 static uint16_t     siftr_port_filter = 0;
265 /* static unsigned int siftr_binary_log = 0; */
266 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log";
267 static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log";
268 static u_long siftr_hashmask;
269 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue);
270 LIST_HEAD(listhead, flow_hash_node) *counter_hash;
271 static int wait_for_pkt;
272 static struct alq *siftr_alq = NULL;
273 static struct mtx siftr_pkt_queue_mtx;
274 static struct mtx siftr_pkt_mgr_mtx;
275 static struct thread *siftr_pkt_manager_thr = NULL;
276 static char direction[2] = {'i','o'};
277 
278 /* Required function prototypes. */
279 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS);
280 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS);
281 
282 /* Declare the net.inet.siftr sysctl tree and populate it. */
283 
284 SYSCTL_DECL(_net_inet_siftr);
285 
286 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
287     "siftr related settings");
288 
289 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled,
290     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
291     &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU",
292     "switch siftr module operations on/off");
293 
294 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile,
295     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &siftr_logfile_shadow,
296     sizeof(siftr_logfile_shadow), &siftr_sysctl_logfile_name_handler, "A",
297     "file to save siftr log messages to");
298 
299 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW,
300     &siftr_pkts_per_log, 1,
301     "number of packets between generating a log message");
302 
303 SYSCTL_U16(_net_inet_siftr, OID_AUTO, port_filter, CTLFLAG_RW,
304     &siftr_port_filter, 0,
305     "enable packet filter on a TCP port");
306 
307 /* XXX: TODO
308 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW,
309     &siftr_binary_log, 0,
310     "write log files in binary instead of ascii");
311 */
312 
313 /* Begin functions. */
314 
315 static inline struct flow_hash_node *
316 siftr_find_flow(struct listhead *counter_list, uint32_t id)
317 {
318 	struct flow_hash_node *hash_node;
319 	/*
320 	 * If the list is not empty i.e. the hash index has
321 	 * been used by another flow previously.
322 	 */
323 	if (LIST_FIRST(counter_list) != NULL) {
324 		/*
325 		 * Loop through the hash nodes in the list.
326 		 * There should normally only be 1 hash node in the list.
327 		 */
328 		LIST_FOREACH(hash_node, counter_list, nodes) {
329 			/*
330 			 * Check if the key for the pkt we are currently
331 			 * processing is the same as the key stored in the
332 			 * hash node we are currently processing.
333 			 * If they are the same, then we've found the
334 			 * hash node that stores the counter for the flow
335 			 * the pkt belongs to.
336 			 */
337 			if (hash_node->const_info.key == id) {
338 				return hash_node;
339 			}
340 		}
341 	}
342 
343 	return NULL;
344 }
345 
346 static inline struct flow_hash_node *
347 siftr_new_hash_node(struct flow_info info, int dir,
348 		    struct siftr_stats *ss)
349 {
350 	struct flow_hash_node *hash_node;
351 	struct listhead *counter_list;
352 
353 	counter_list = counter_hash + (info.key & siftr_hashmask);
354 	/* Create a new hash node to store the flow's constant info. */
355 	hash_node = malloc(sizeof(struct flow_hash_node), M_SIFTR_HASHNODE,
356 			   M_NOWAIT|M_ZERO);
357 
358 	if (hash_node != NULL) {
359 		/* Initialise our new hash node list entry. */
360 		hash_node->counter = 0;
361 		hash_node->const_info = info;
362 		LIST_INSERT_HEAD(counter_list, hash_node, nodes);
363 		return hash_node;
364 	} else {
365 		/* malloc failed */
366 		if (dir == DIR_IN)
367 			ss->nskip_in_malloc++;
368 		else
369 			ss->nskip_out_malloc++;
370 
371 		return NULL;
372 	}
373 }
374 
375 static void
376 siftr_process_pkt(struct pkt_node * pkt_node)
377 {
378 	struct flow_hash_node *hash_node;
379 	struct listhead *counter_list;
380 	struct ale *log_buf;
381 
382 	if (pkt_node->flowid == 0) {
383 		panic("%s: flowid not available", __func__);
384 	}
385 
386 	counter_list = counter_hash + (pkt_node->flowid & siftr_hashmask);
387 	hash_node = siftr_find_flow(counter_list, pkt_node->flowid);
388 
389 	if (hash_node == NULL) {
390 		return;
391 	} else if (siftr_pkts_per_log > 1) {
392 		/*
393 		 * Taking the remainder of the counter divided
394 		 * by the current value of siftr_pkts_per_log
395 		 * and storing that in counter provides a neat
396 		 * way to modulate the frequency of log
397 		 * messages being written to the log file.
398 		 */
399 		hash_node->counter = (hash_node->counter + 1) %
400 				     siftr_pkts_per_log;
401 		/*
402 		 * If we have not seen enough packets since the last time
403 		 * we wrote a log message for this connection, return.
404 		 */
405 		if (hash_node->counter > 0)
406 			return;
407 	}
408 
409 	log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK);
410 
411 	if (log_buf == NULL)
412 		return; /* Should only happen if the ALQ is shutting down. */
413 
414 	/* Construct a log message. */
415 	log_buf->ae_bytesused = snprintf(log_buf->ae_data, MAX_LOG_MSG_LEN,
416 	    "%c,%jd.%06ld,%s,%hu,%s,%hu,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,"
417 	    "%u,%u,%u,%u,%u,%u,%u,%u\n",
418 	    direction[pkt_node->direction],
419 	    (intmax_t)pkt_node->tval.tv_sec,
420 	    pkt_node->tval.tv_usec,
421 	    hash_node->const_info.laddr,
422 	    hash_node->const_info.lport,
423 	    hash_node->const_info.faddr,
424 	    hash_node->const_info.fport,
425 	    pkt_node->snd_ssthresh,
426 	    pkt_node->snd_cwnd,
427 	    pkt_node->t_flags2,
428 	    pkt_node->snd_wnd,
429 	    pkt_node->rcv_wnd,
430 	    pkt_node->snd_scale,
431 	    pkt_node->rcv_scale,
432 	    pkt_node->conn_state,
433 	    pkt_node->mss,
434 	    pkt_node->srtt,
435 	    pkt_node->sack_enabled,
436 	    pkt_node->t_flags,
437 	    pkt_node->rto,
438 	    pkt_node->snd_buf_hiwater,
439 	    pkt_node->snd_buf_cc,
440 	    pkt_node->rcv_buf_hiwater,
441 	    pkt_node->rcv_buf_cc,
442 	    pkt_node->sent_inflight_bytes,
443 	    pkt_node->t_segqlen,
444 	    pkt_node->flowid,
445 	    pkt_node->flowtype);
446 
447 	alq_post_flags(siftr_alq, log_buf, 0);
448 }
449 
450 static void
451 siftr_pkt_manager_thread(void *arg)
452 {
453 	STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
454 	    STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
455 	struct pkt_node *pkt_node, *pkt_node_temp;
456 	uint8_t draining;
457 
458 	draining = 2;
459 
460 	mtx_lock(&siftr_pkt_mgr_mtx);
461 
462 	/* draining == 0 when queue has been flushed and it's safe to exit. */
463 	while (draining) {
464 		/*
465 		 * Sleep until we are signalled to wake because thread has
466 		 * been told to exit or until 1 tick has passed.
467 		 */
468 		mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
469 		    1);
470 
471 		/* Gain exclusive access to the pkt_node queue. */
472 		mtx_lock(&siftr_pkt_queue_mtx);
473 
474 		/*
475 		 * Move pkt_queue to tmp_pkt_queue, which leaves
476 		 * pkt_queue empty and ready to receive more pkt_nodes.
477 		 */
478 		STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
479 
480 		/*
481 		 * We've finished making changes to the list. Unlock it
482 		 * so the pfil hooks can continue queuing pkt_nodes.
483 		 */
484 		mtx_unlock(&siftr_pkt_queue_mtx);
485 
486 		/*
487 		 * We can't hold a mutex whilst calling siftr_process_pkt
488 		 * because ALQ might sleep waiting for buffer space.
489 		 */
490 		mtx_unlock(&siftr_pkt_mgr_mtx);
491 
492 		/* Flush all pkt_nodes to the log file. */
493 		STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes,
494 		    pkt_node_temp) {
495 			siftr_process_pkt(pkt_node);
496 			STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
497 			free(pkt_node, M_SIFTR_PKTNODE);
498 		}
499 
500 		KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
501 		    ("SIFTR tmp_pkt_queue not empty after flush"));
502 
503 		mtx_lock(&siftr_pkt_mgr_mtx);
504 
505 		/*
506 		 * If siftr_exit_pkt_manager_thread gets set during the window
507 		 * where we are draining the tmp_pkt_queue above, there might
508 		 * still be pkts in pkt_queue that need to be drained.
509 		 * Allow one further iteration to occur after
510 		 * siftr_exit_pkt_manager_thread has been set to ensure
511 		 * pkt_queue is completely empty before we kill the thread.
512 		 *
513 		 * siftr_exit_pkt_manager_thread is set only after the pfil
514 		 * hooks have been removed, so only 1 extra iteration
515 		 * is needed to drain the queue.
516 		 */
517 		if (siftr_exit_pkt_manager_thread)
518 			draining--;
519 	}
520 
521 	mtx_unlock(&siftr_pkt_mgr_mtx);
522 
523 	/* Calls wakeup on this thread's struct thread ptr. */
524 	kthread_exit();
525 }
526 
527 /*
528  * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
529  * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
530  * Return value >0 means the caller should skip processing this mbuf.
531  */
532 static inline int
533 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
534 {
535 	if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
536 	    != NULL) {
537 		if (dir == PFIL_IN)
538 			ss->nskip_in_dejavu++;
539 		else
540 			ss->nskip_out_dejavu++;
541 
542 		return (1);
543 	} else {
544 		struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
545 		    PACKET_TAG_SIFTR, 0, M_NOWAIT);
546 		if (tag == NULL) {
547 			if (dir == PFIL_IN)
548 				ss->nskip_in_malloc++;
549 			else
550 				ss->nskip_out_malloc++;
551 
552 			return (1);
553 		}
554 
555 		m_tag_prepend(m, tag);
556 	}
557 
558 	return (0);
559 }
560 
561 /*
562  * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
563  * otherwise.
564  */
565 static inline struct inpcb *
566 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
567     uint16_t dport, int dir, struct siftr_stats *ss)
568 {
569 	struct inpcb *inp;
570 
571 	/* We need the tcbinfo lock. */
572 	INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
573 
574 	if (dir == PFIL_IN)
575 		inp = (ipver == INP_IPV4 ?
576 		    in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
577 		    dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
578 		    :
579 #ifdef SIFTR_IPV6
580 		    in6_pcblookup(&V_tcbinfo,
581 		    &((struct ip6_hdr *)ip)->ip6_src, sport,
582 		    &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB,
583 		    m->m_pkthdr.rcvif)
584 #else
585 		    NULL
586 #endif
587 		    );
588 
589 	else
590 		inp = (ipver == INP_IPV4 ?
591 		    in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
592 		    sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
593 		    :
594 #ifdef SIFTR_IPV6
595 		    in6_pcblookup(&V_tcbinfo,
596 		    &((struct ip6_hdr *)ip)->ip6_dst, dport,
597 		    &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB,
598 		    m->m_pkthdr.rcvif)
599 #else
600 		    NULL
601 #endif
602 		    );
603 
604 	/* If we can't find the inpcb, bail. */
605 	if (inp == NULL) {
606 		if (dir == PFIL_IN)
607 			ss->nskip_in_inpcb++;
608 		else
609 			ss->nskip_out_inpcb++;
610 	}
611 
612 	return (inp);
613 }
614 
615 static inline uint32_t
616 siftr_get_flowid(struct inpcb *inp, int ipver, uint32_t *phashtype)
617 {
618 	if (inp->inp_flowid == 0) {
619 #ifdef SIFTR_IPV6
620 		if (ipver == INP_IPV6) {
621 			return fib6_calc_packet_hash(&inp->in6p_laddr,
622 						     &inp->in6p_faddr,
623 						     inp->inp_lport,
624 						     inp->inp_fport,
625 						     IPPROTO_TCP,
626 						     phashtype);
627 		} else
628 #endif
629 		{
630 			return fib4_calc_packet_hash(inp->inp_laddr,
631 						     inp->inp_faddr,
632 						     inp->inp_lport,
633 						     inp->inp_fport,
634 						     IPPROTO_TCP,
635 						     phashtype);
636 		}
637 	} else {
638 		*phashtype = inp->inp_flowtype;
639 		return inp->inp_flowid;
640 	}
641 }
642 
643 static inline void
644 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
645     int ipver, int dir, int inp_locally_locked)
646 {
647 	pn->ipver = ipver;
648 	pn->lport = inp->inp_lport;
649 	pn->fport = inp->inp_fport;
650 	pn->laddr = inp->inp_inc.inc_ie.ie_dependladdr;
651 	pn->faddr = inp->inp_inc.inc_ie.ie_dependfaddr;
652 	pn->snd_cwnd = tp->snd_cwnd;
653 	pn->snd_wnd = tp->snd_wnd;
654 	pn->rcv_wnd = tp->rcv_wnd;
655 	pn->t_flags2 = tp->t_flags2;
656 	pn->snd_ssthresh = tp->snd_ssthresh;
657 	pn->snd_scale = tp->snd_scale;
658 	pn->rcv_scale = tp->rcv_scale;
659 	pn->conn_state = tp->t_state;
660 	pn->mss = tp->t_maxseg;
661 	pn->srtt = ((uint64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
662 	pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
663 	pn->t_flags = tp->t_flags;
664 	pn->rto = tp->t_rxtcur * tick;
665 	pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
666 	pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd);
667 	pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
668 	pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv);
669 	pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
670 	pn->t_segqlen = tp->t_segqlen;
671 
672 	/* We've finished accessing the tcb so release the lock. */
673 	if (inp_locally_locked)
674 		INP_RUNLOCK(inp);
675 
676 	pn->direction = (dir == PFIL_IN ? DIR_IN : DIR_OUT);
677 
678 	/*
679 	 * Significantly more accurate than using getmicrotime(), but slower!
680 	 * Gives true microsecond resolution at the expense of a hit to
681 	 * maximum pps throughput processing when SIFTR is loaded and enabled.
682 	 */
683 	microtime(&pn->tval);
684 	TCP_PROBE1(siftr, pn);
685 }
686 
687 /*
688  * pfil hook that is called for each IPv4 packet making its way through the
689  * stack in either direction.
690  * The pfil subsystem holds a non-sleepable mutex somewhere when
691  * calling our hook function, so we can't sleep at all.
692  * It's very important to use the M_NOWAIT flag with all function calls
693  * that support it so that they won't sleep, otherwise you get a panic.
694  */
695 static pfil_return_t
696 siftr_chkpkt(struct mbuf **m, struct ifnet *ifp, int flags,
697     void *ruleset __unused, struct inpcb *inp)
698 {
699 	struct pkt_node *pn;
700 	struct ip *ip;
701 	struct tcphdr *th;
702 	struct tcpcb *tp;
703 	struct siftr_stats *ss;
704 	unsigned int ip_hl;
705 	int inp_locally_locked, dir;
706 	uint32_t hash_id, hash_type;
707 	struct listhead *counter_list;
708 	struct flow_hash_node *hash_node;
709 
710 	inp_locally_locked = 0;
711 	dir = PFIL_DIR(flags);
712 	ss = DPCPU_PTR(ss);
713 
714 	/*
715 	 * m_pullup is not required here because ip_{input|output}
716 	 * already do the heavy lifting for us.
717 	 */
718 
719 	ip = mtod(*m, struct ip *);
720 
721 	/* Only continue processing if the packet is TCP. */
722 	if (ip->ip_p != IPPROTO_TCP)
723 		goto ret;
724 
725 	/*
726 	 * Create a tcphdr struct starting at the correct offset
727 	 * in the IP packet. ip->ip_hl gives the ip header length
728 	 * in 4-byte words, so multiply it to get the size in bytes.
729 	 */
730 	ip_hl = (ip->ip_hl << 2);
731 	th = (struct tcphdr *)((caddr_t)ip + ip_hl);
732 
733 	/*
734 	 * Only pkts selected by the tcp port filter
735 	 * can be inserted into the pkt_queue
736 	 */
737 	if ((siftr_port_filter != 0) &&
738 	    (siftr_port_filter != ntohs(th->th_sport)) &&
739 	    (siftr_port_filter != ntohs(th->th_dport))) {
740 		goto ret;
741 	}
742 
743 	/*
744 	 * If a kernel subsystem reinjects packets into the stack, our pfil
745 	 * hook will be called multiple times for the same packet.
746 	 * Make sure we only process unique packets.
747 	 */
748 	if (siftr_chkreinject(*m, dir, ss))
749 		goto ret;
750 
751 	if (dir == PFIL_IN)
752 		ss->n_in++;
753 	else
754 		ss->n_out++;
755 
756 	/*
757 	 * If the pfil hooks don't provide a pointer to the
758 	 * inpcb, we need to find it ourselves and lock it.
759 	 */
760 	if (!inp) {
761 		/* Find the corresponding inpcb for this pkt. */
762 		inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
763 		    th->th_dport, dir, ss);
764 
765 		if (inp == NULL)
766 			goto ret;
767 		else
768 			inp_locally_locked = 1;
769 	}
770 
771 	INP_LOCK_ASSERT(inp);
772 
773 	/* Find the TCP control block that corresponds with this packet */
774 	tp = intotcpcb(inp);
775 
776 	/*
777 	 * If we can't find the TCP control block (happens occasionaly for a
778 	 * packet sent during the shutdown phase of a TCP connection), or the
779 	 * TCP control block has not initialized (happens during TCPS_SYN_SENT),
780 	 * bail.
781 	 */
782 	if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) {
783 		if (dir == PFIL_IN)
784 			ss->nskip_in_tcpcb++;
785 		else
786 			ss->nskip_out_tcpcb++;
787 
788 		goto inp_unlock;
789 	}
790 
791 	hash_id = siftr_get_flowid(inp, INP_IPV4, &hash_type);
792 	counter_list = counter_hash + (hash_id & siftr_hashmask);
793 	hash_node = siftr_find_flow(counter_list, hash_id);
794 
795 	/* If this flow hasn't been seen before, we create a new entry. */
796 	if (hash_node == NULL) {
797 		struct flow_info info;
798 
799 		inet_ntoa_r(inp->inp_laddr, info.laddr);
800 		inet_ntoa_r(inp->inp_faddr, info.faddr);
801 		info.lport = ntohs(inp->inp_lport);
802 		info.fport = ntohs(inp->inp_fport);
803 		info.key = hash_id;
804 
805 		hash_node = siftr_new_hash_node(info, dir, ss);
806 	}
807 
808 	if (hash_node == NULL) {
809 		goto inp_unlock;
810 	}
811 
812 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
813 
814 	if (pn == NULL) {
815 		if (dir == PFIL_IN)
816 			ss->nskip_in_malloc++;
817 		else
818 			ss->nskip_out_malloc++;
819 
820 		goto inp_unlock;
821 	}
822 
823 	pn->flowid = hash_id;
824 	pn->flowtype = hash_type;
825 
826 	siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
827 
828 	mtx_lock(&siftr_pkt_queue_mtx);
829 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
830 	mtx_unlock(&siftr_pkt_queue_mtx);
831 	goto ret;
832 
833 inp_unlock:
834 	if (inp_locally_locked)
835 		INP_RUNLOCK(inp);
836 
837 ret:
838 	return (PFIL_PASS);
839 }
840 
841 #ifdef SIFTR_IPV6
842 static pfil_return_t
843 siftr_chkpkt6(struct mbuf **m, struct ifnet *ifp, int flags,
844     void *ruleset __unused, struct inpcb *inp)
845 {
846 	struct pkt_node *pn;
847 	struct ip6_hdr *ip6;
848 	struct tcphdr *th;
849 	struct tcpcb *tp;
850 	struct siftr_stats *ss;
851 	unsigned int ip6_hl;
852 	int inp_locally_locked, dir;
853 	uint32_t hash_id, hash_type;
854 	struct listhead *counter_list;
855 	struct flow_hash_node *hash_node;
856 
857 	inp_locally_locked = 0;
858 	dir = PFIL_DIR(flags);
859 	ss = DPCPU_PTR(ss);
860 
861 	/*
862 	 * m_pullup is not required here because ip6_{input|output}
863 	 * already do the heavy lifting for us.
864 	 */
865 
866 	ip6 = mtod(*m, struct ip6_hdr *);
867 
868 	/*
869 	 * Only continue processing if the packet is TCP
870 	 * XXX: We should follow the next header fields
871 	 * as shown on Pg 6 RFC 2460, but right now we'll
872 	 * only check pkts that have no extension headers.
873 	 */
874 	if (ip6->ip6_nxt != IPPROTO_TCP)
875 		goto ret6;
876 
877 	/*
878 	 * Create a tcphdr struct starting at the correct offset
879 	 * in the ipv6 packet.
880 	 */
881 	ip6_hl = sizeof(struct ip6_hdr);
882 	th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
883 
884 	/*
885 	 * Only pkts selected by the tcp port filter
886 	 * can be inserted into the pkt_queue
887 	 */
888 	if ((siftr_port_filter != 0) &&
889 	    (siftr_port_filter != ntohs(th->th_sport)) &&
890 	    (siftr_port_filter != ntohs(th->th_dport))) {
891 		goto ret6;
892 	}
893 
894 	/*
895 	 * If a kernel subsystem reinjects packets into the stack, our pfil
896 	 * hook will be called multiple times for the same packet.
897 	 * Make sure we only process unique packets.
898 	 */
899 	if (siftr_chkreinject(*m, dir, ss))
900 		goto ret6;
901 
902 	if (dir == PFIL_IN)
903 		ss->n_in++;
904 	else
905 		ss->n_out++;
906 
907 	/*
908 	 * For inbound packets, the pfil hooks don't provide a pointer to the
909 	 * inpcb, so we need to find it ourselves and lock it.
910 	 */
911 	if (!inp) {
912 		/* Find the corresponding inpcb for this pkt. */
913 		inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
914 		    th->th_sport, th->th_dport, dir, ss);
915 
916 		if (inp == NULL)
917 			goto ret6;
918 		else
919 			inp_locally_locked = 1;
920 	}
921 
922 	/* Find the TCP control block that corresponds with this packet. */
923 	tp = intotcpcb(inp);
924 
925 	/*
926 	 * If we can't find the TCP control block (happens occasionaly for a
927 	 * packet sent during the shutdown phase of a TCP connection), or the
928 	 * TCP control block has not initialized (happens during TCPS_SYN_SENT),
929 	 * bail.
930 	 */
931 	if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) {
932 		if (dir == PFIL_IN)
933 			ss->nskip_in_tcpcb++;
934 		else
935 			ss->nskip_out_tcpcb++;
936 
937 		goto inp_unlock6;
938 	}
939 
940 	hash_id = siftr_get_flowid(inp, INP_IPV6, &hash_type);
941 	counter_list = counter_hash + (hash_id & siftr_hashmask);
942 	hash_node = siftr_find_flow(counter_list, hash_id);
943 
944 	/* If this flow hasn't been seen before, we create a new entry. */
945 	if (!hash_node) {
946 		struct flow_info info;
947 
948 		ip6_sprintf(info.laddr, &inp->in6p_laddr);
949 		ip6_sprintf(info.faddr, &inp->in6p_faddr);
950 		info.lport = ntohs(inp->inp_lport);
951 		info.fport = ntohs(inp->inp_fport);
952 		info.key = hash_id;
953 
954 		hash_node = siftr_new_hash_node(info, dir, ss);
955 	}
956 
957 	if (!hash_node) {
958 		goto inp_unlock6;
959 	}
960 
961 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
962 
963 	if (pn == NULL) {
964 		if (dir == PFIL_IN)
965 			ss->nskip_in_malloc++;
966 		else
967 			ss->nskip_out_malloc++;
968 
969 		goto inp_unlock6;
970 	}
971 
972 	pn->flowid = hash_id;
973 	pn->flowtype = hash_type;
974 
975 	siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
976 
977 	mtx_lock(&siftr_pkt_queue_mtx);
978 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
979 	mtx_unlock(&siftr_pkt_queue_mtx);
980 	goto ret6;
981 
982 inp_unlock6:
983 	if (inp_locally_locked)
984 		INP_RUNLOCK(inp);
985 
986 ret6:
987 	return (PFIL_PASS);
988 }
989 #endif /* #ifdef SIFTR_IPV6 */
990 
991 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet_hook);
992 #define	V_siftr_inet_hook	VNET(siftr_inet_hook)
993 #ifdef SIFTR_IPV6
994 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet6_hook);
995 #define	V_siftr_inet6_hook	VNET(siftr_inet6_hook)
996 #endif
997 static int
998 siftr_pfil(int action)
999 {
1000 	struct pfil_hook_args pha = {
1001 		.pa_version = PFIL_VERSION,
1002 		.pa_flags = PFIL_IN | PFIL_OUT,
1003 		.pa_modname = "siftr",
1004 		.pa_rulname = "default",
1005 	};
1006 	struct pfil_link_args pla = {
1007 		.pa_version = PFIL_VERSION,
1008 		.pa_flags = PFIL_IN | PFIL_OUT | PFIL_HEADPTR | PFIL_HOOKPTR,
1009 	};
1010 
1011 	VNET_ITERATOR_DECL(vnet_iter);
1012 
1013 	VNET_LIST_RLOCK();
1014 	VNET_FOREACH(vnet_iter) {
1015 		CURVNET_SET(vnet_iter);
1016 
1017 		if (action == HOOK) {
1018 			pha.pa_mbuf_chk = siftr_chkpkt;
1019 			pha.pa_type = PFIL_TYPE_IP4;
1020 			V_siftr_inet_hook = pfil_add_hook(&pha);
1021 			pla.pa_hook = V_siftr_inet_hook;
1022 			pla.pa_head = V_inet_pfil_head;
1023 			(void)pfil_link(&pla);
1024 #ifdef SIFTR_IPV6
1025 			pha.pa_mbuf_chk = siftr_chkpkt6;
1026 			pha.pa_type = PFIL_TYPE_IP6;
1027 			V_siftr_inet6_hook = pfil_add_hook(&pha);
1028 			pla.pa_hook = V_siftr_inet6_hook;
1029 			pla.pa_head = V_inet6_pfil_head;
1030 			(void)pfil_link(&pla);
1031 #endif
1032 		} else if (action == UNHOOK) {
1033 			pfil_remove_hook(V_siftr_inet_hook);
1034 #ifdef SIFTR_IPV6
1035 			pfil_remove_hook(V_siftr_inet6_hook);
1036 #endif
1037 		}
1038 		CURVNET_RESTORE();
1039 	}
1040 	VNET_LIST_RUNLOCK();
1041 
1042 	return (0);
1043 }
1044 
1045 static int
1046 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1047 {
1048 	struct alq *new_alq;
1049 	int error;
1050 
1051 	error = sysctl_handle_string(oidp, arg1, arg2, req);
1052 
1053 	/* Check for error or same filename */
1054 	if (error != 0 || req->newptr == NULL ||
1055 	    strncmp(siftr_logfile, arg1, arg2) == 0)
1056 		goto done;
1057 
1058 	/* file name changed */
1059 	error = alq_open(&new_alq, arg1, curthread->td_ucred,
1060 	    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1061 	if (error != 0)
1062 		goto done;
1063 
1064 	/*
1065 	 * If disabled, siftr_alq == NULL so we simply close
1066 	 * the alq as we've proved it can be opened.
1067 	 * If enabled, close the existing alq and switch the old
1068 	 * for the new.
1069 	 */
1070 	if (siftr_alq == NULL) {
1071 		alq_close(new_alq);
1072 	} else {
1073 		alq_close(siftr_alq);
1074 		siftr_alq = new_alq;
1075 	}
1076 
1077 	/* Update filename upon success */
1078 	strlcpy(siftr_logfile, arg1, arg2);
1079 done:
1080 	return (error);
1081 }
1082 
1083 static int
1084 siftr_manage_ops(uint8_t action)
1085 {
1086 	struct siftr_stats totalss;
1087 	struct timeval tval;
1088 	struct flow_hash_node *counter, *tmp_counter;
1089 	struct sbuf *s;
1090 	int i, error;
1091 	uint32_t bytes_to_write, total_skipped_pkts;
1092 
1093 	error = 0;
1094 	total_skipped_pkts = 0;
1095 
1096 	/* Init an autosizing sbuf that initially holds 200 chars. */
1097 	if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1098 		return (-1);
1099 
1100 	if (action == SIFTR_ENABLE && siftr_pkt_manager_thr == NULL) {
1101 		/*
1102 		 * Create our alq
1103 		 * XXX: We should abort if alq_open fails!
1104 		 */
1105 		alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1106 		    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1107 
1108 		STAILQ_INIT(&pkt_queue);
1109 
1110 		DPCPU_ZERO(ss);
1111 
1112 		siftr_exit_pkt_manager_thread = 0;
1113 
1114 		kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1115 		    &siftr_pkt_manager_thr, RFNOWAIT, 0,
1116 		    "siftr_pkt_manager_thr");
1117 
1118 		siftr_pfil(HOOK);
1119 
1120 		microtime(&tval);
1121 
1122 		sbuf_printf(s,
1123 		    "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1124 		    "siftrver=%s\tsysname=%s\tsysver=%u\tipmode=%u\n",
1125 		    (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR,
1126 		    SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1127 
1128 		sbuf_finish(s);
1129 		alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1130 
1131 	} else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1132 		/*
1133 		 * Remove the pfil hook functions. All threads currently in
1134 		 * the hook functions are allowed to exit before siftr_pfil()
1135 		 * returns.
1136 		 */
1137 		siftr_pfil(UNHOOK);
1138 
1139 		/* This will block until the pkt manager thread unlocks it. */
1140 		mtx_lock(&siftr_pkt_mgr_mtx);
1141 
1142 		/* Tell the pkt manager thread that it should exit now. */
1143 		siftr_exit_pkt_manager_thread = 1;
1144 
1145 		/*
1146 		 * Wake the pkt_manager thread so it realises that
1147 		 * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1148 		 * The wakeup won't be delivered until we unlock
1149 		 * siftr_pkt_mgr_mtx so this isn't racy.
1150 		 */
1151 		wakeup(&wait_for_pkt);
1152 
1153 		/* Wait for the pkt_manager thread to exit. */
1154 		mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1155 		    "thrwait", 0);
1156 
1157 		siftr_pkt_manager_thr = NULL;
1158 		mtx_unlock(&siftr_pkt_mgr_mtx);
1159 
1160 		totalss.n_in = DPCPU_VARSUM(ss, n_in);
1161 		totalss.n_out = DPCPU_VARSUM(ss, n_out);
1162 		totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1163 		totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1164 		totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1165 		totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1166 		totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1167 		totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1168 
1169 		total_skipped_pkts = totalss.nskip_in_malloc +
1170 		    totalss.nskip_out_malloc + totalss.nskip_in_tcpcb +
1171 		    totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1172 		    totalss.nskip_out_inpcb;
1173 
1174 		microtime(&tval);
1175 
1176 		sbuf_printf(s,
1177 		    "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1178 		    "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1179 		    "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1180 		    "num_outbound_skipped_pkts_malloc=%u\t"
1181 		    "num_inbound_skipped_pkts_tcpcb=%u\t"
1182 		    "num_outbound_skipped_pkts_tcpcb=%u\t"
1183 		    "num_inbound_skipped_pkts_inpcb=%u\t"
1184 		    "num_outbound_skipped_pkts_inpcb=%u\t"
1185 		    "total_skipped_tcp_pkts=%u\tflow_list=",
1186 		    (intmax_t)tval.tv_sec,
1187 		    tval.tv_usec,
1188 		    (uintmax_t)totalss.n_in,
1189 		    (uintmax_t)totalss.n_out,
1190 		    (uintmax_t)(totalss.n_in + totalss.n_out),
1191 		    totalss.nskip_in_malloc,
1192 		    totalss.nskip_out_malloc,
1193 		    totalss.nskip_in_tcpcb,
1194 		    totalss.nskip_out_tcpcb,
1195 		    totalss.nskip_in_inpcb,
1196 		    totalss.nskip_out_inpcb,
1197 		    total_skipped_pkts);
1198 
1199 		/*
1200 		 * Iterate over the flow hash, printing a summary of each
1201 		 * flow seen and freeing any malloc'd memory.
1202 		 * The hash consists of an array of LISTs (man 3 queue).
1203 		 */
1204 		for (i = 0; i <= siftr_hashmask; i++) {
1205 			LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1206 			    tmp_counter) {
1207 				sbuf_printf(s, "%s;%hu-%s;%hu,",
1208 					    counter->const_info.laddr,
1209 					    counter->const_info.lport,
1210 					    counter->const_info.faddr,
1211 					    counter->const_info.fport);
1212 
1213 				free(counter, M_SIFTR_HASHNODE);
1214 			}
1215 
1216 			LIST_INIT(counter_hash + i);
1217 		}
1218 
1219 		sbuf_printf(s, "\n");
1220 		sbuf_finish(s);
1221 
1222 		i = 0;
1223 		do {
1224 			bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1225 			alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1226 			i += bytes_to_write;
1227 		} while (i < sbuf_len(s));
1228 
1229 		alq_close(siftr_alq);
1230 		siftr_alq = NULL;
1231 	} else
1232 		error = EINVAL;
1233 
1234 	sbuf_delete(s);
1235 
1236 	/*
1237 	 * XXX: Should be using ret to check if any functions fail
1238 	 * and set error appropriately
1239 	 */
1240 
1241 	return (error);
1242 }
1243 
1244 static int
1245 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1246 {
1247 	int error;
1248 	uint32_t new;
1249 
1250 	new = siftr_enabled;
1251 	error = sysctl_handle_int(oidp, &new, 0, req);
1252 	if (error == 0 && req->newptr != NULL) {
1253 		if (new > 1)
1254 			return (EINVAL);
1255 		else if (new != siftr_enabled) {
1256 			if ((error = siftr_manage_ops(new)) == 0) {
1257 				siftr_enabled = new;
1258 			} else {
1259 				siftr_manage_ops(SIFTR_DISABLE);
1260 			}
1261 		}
1262 	}
1263 
1264 	return (error);
1265 }
1266 
1267 static void
1268 siftr_shutdown_handler(void *arg)
1269 {
1270 	if (siftr_enabled == 1) {
1271 		siftr_manage_ops(SIFTR_DISABLE);
1272 	}
1273 }
1274 
1275 /*
1276  * Module is being unloaded or machine is shutting down. Take care of cleanup.
1277  */
1278 static int
1279 deinit_siftr(void)
1280 {
1281 	/* Cleanup. */
1282 	siftr_manage_ops(SIFTR_DISABLE);
1283 	hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1284 	mtx_destroy(&siftr_pkt_queue_mtx);
1285 	mtx_destroy(&siftr_pkt_mgr_mtx);
1286 
1287 	return (0);
1288 }
1289 
1290 /*
1291  * Module has just been loaded into the kernel.
1292  */
1293 static int
1294 init_siftr(void)
1295 {
1296 	EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL,
1297 	    SHUTDOWN_PRI_FIRST);
1298 
1299 	/* Initialise our flow counter hash table. */
1300 	counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1301 	    &siftr_hashmask);
1302 
1303 	mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1304 	mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1305 
1306 	/* Print message to the user's current terminal. */
1307 	uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1308 	    "          http://caia.swin.edu.au/urp/newtcp\n\n",
1309 	    MODVERSION_STR);
1310 
1311 	return (0);
1312 }
1313 
1314 /*
1315  * This is the function that is called to load and unload the module.
1316  * When the module is loaded, this function is called once with
1317  * "what" == MOD_LOAD
1318  * When the module is unloaded, this function is called twice with
1319  * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1320  * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1321  * this function is called once with "what" = MOD_SHUTDOWN
1322  * When the system is shut down, the handler isn't called until the very end
1323  * of the shutdown sequence i.e. after the disks have been synced.
1324  */
1325 static int
1326 siftr_load_handler(module_t mod, int what, void *arg)
1327 {
1328 	int ret;
1329 
1330 	switch (what) {
1331 	case MOD_LOAD:
1332 		ret = init_siftr();
1333 		break;
1334 
1335 	case MOD_QUIESCE:
1336 	case MOD_SHUTDOWN:
1337 		ret = deinit_siftr();
1338 		break;
1339 
1340 	case MOD_UNLOAD:
1341 		ret = 0;
1342 		break;
1343 
1344 	default:
1345 		ret = EINVAL;
1346 		break;
1347 	}
1348 
1349 	return (ret);
1350 }
1351 
1352 static moduledata_t siftr_mod = {
1353 	.name = "siftr",
1354 	.evhand = siftr_load_handler,
1355 };
1356 
1357 /*
1358  * Param 1: name of the kernel module
1359  * Param 2: moduledata_t struct containing info about the kernel module
1360  *          and the execution entry point for the module
1361  * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1362  *          Defines the module initialisation order
1363  * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1364  *          Defines the initialisation order of this kld relative to others
1365  *          within the same subsystem as defined by param 3
1366  */
1367 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_LAST, SI_ORDER_ANY);
1368 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1369 MODULE_VERSION(siftr, MODVERSION);
1370