xref: /freebsd/sys/netinet/siftr.c (revision cd9da8d072e4dff4c1eda7307547c6a5a1e5bd35)
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		tcp_localport;
167 	/* Foreign TCP port. */
168 	uint16_t		tcp_foreignport;
169 	/* Congestion Window (bytes). */
170 	uint32_t		snd_cwnd;
171 	/* Sending Window (bytes). */
172 	uint32_t		snd_wnd;
173 	/* Receive Window (bytes). */
174 	uint32_t		rcv_wnd;
175 	/* More tcpcb flags storage */
176 	uint32_t		t_flags2;
177 	/* Slow Start Threshold (bytes). */
178 	uint32_t		snd_ssthresh;
179 	/* Current state of the TCP FSM. */
180 	int			conn_state;
181 	/* Max Segment Size (bytes). */
182 	u_int			max_seg_size;
183 	/* Smoothed RTT (usecs). */
184 	uint32_t		srtt;
185 	/* Is SACK enabled? */
186 	u_char			sack_enabled;
187 	/* Window scaling for snd window. */
188 	u_char			snd_scale;
189 	/* Window scaling for recv window. */
190 	u_char			rcv_scale;
191 	/* TCP control block flags. */
192 	u_int			flags;
193 	/* Retransmission timeout (usec). */
194 	uint32_t		rto;
195 	/* Size of the TCP send buffer in bytes. */
196 	u_int			snd_buf_hiwater;
197 	/* Current num bytes in the send socket buffer. */
198 	u_int			snd_buf_cc;
199 	/* Size of the TCP receive buffer in bytes. */
200 	u_int			rcv_buf_hiwater;
201 	/* Current num bytes in the receive socket buffer. */
202 	u_int			rcv_buf_cc;
203 	/* Number of bytes inflight that we are waiting on ACKs for. */
204 	u_int			sent_inflight_bytes;
205 	/* Number of segments currently in the reassembly queue. */
206 	int			t_segqlen;
207 	/* Flowid for the connection. */
208 	u_int			flowid;
209 	/* Flow type for the connection. */
210 	u_int			flowtype;
211 	/* Link to next pkt_node in the list. */
212 	STAILQ_ENTRY(pkt_node)	nodes;
213 };
214 
215 struct flow_info
216 {
217 #ifdef SIFTR_IPV6
218 	char	laddr[INET6_ADDRSTRLEN];	/* local IP address */
219 	char	faddr[INET6_ADDRSTRLEN];	/* foreign IP address */
220 #else
221 	char	laddr[INET_ADDRSTRLEN];		/* local IP address */
222 	char	faddr[INET_ADDRSTRLEN];		/* foreign IP address */
223 #endif
224 	uint16_t	lport;			/* local TCP port */
225 	uint16_t	fport;			/* foreign TCP port */
226 	uint8_t		ipver;			/* IP version */
227 	uint32_t	key;			/* flowid of the connection */
228 };
229 
230 struct flow_hash_node
231 {
232 	uint16_t counter;
233 	struct flow_info const_info;		/* constant connection info */
234 	LIST_ENTRY(flow_hash_node) nodes;
235 };
236 
237 struct siftr_stats
238 {
239 	/* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */
240 	uint64_t n_in;
241 	uint64_t n_out;
242 	/* # pkts skipped due to failed malloc calls. */
243 	uint32_t nskip_in_malloc;
244 	uint32_t nskip_out_malloc;
245 	/* # pkts skipped due to failed inpcb lookups. */
246 	uint32_t nskip_in_inpcb;
247 	uint32_t nskip_out_inpcb;
248 	/* # pkts skipped due to failed tcpcb lookups. */
249 	uint32_t nskip_in_tcpcb;
250 	uint32_t nskip_out_tcpcb;
251 	/* # pkts skipped due to stack reinjection. */
252 	uint32_t nskip_in_dejavu;
253 	uint32_t nskip_out_dejavu;
254 };
255 
256 DPCPU_DEFINE_STATIC(struct siftr_stats, ss);
257 
258 static volatile unsigned int siftr_exit_pkt_manager_thread = 0;
259 static unsigned int siftr_enabled = 0;
260 static unsigned int siftr_pkts_per_log = 1;
261 static uint16_t     siftr_port_filter = 0;
262 /* static unsigned int siftr_binary_log = 0; */
263 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log";
264 static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log";
265 static u_long siftr_hashmask;
266 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue);
267 LIST_HEAD(listhead, flow_hash_node) *counter_hash;
268 static int wait_for_pkt;
269 static struct alq *siftr_alq = NULL;
270 static struct mtx siftr_pkt_queue_mtx;
271 static struct mtx siftr_pkt_mgr_mtx;
272 static struct thread *siftr_pkt_manager_thr = NULL;
273 static char direction[2] = {'i','o'};
274 
275 /* Required function prototypes. */
276 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS);
277 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS);
278 
279 /* Declare the net.inet.siftr sysctl tree and populate it. */
280 
281 SYSCTL_DECL(_net_inet_siftr);
282 
283 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
284     "siftr related settings");
285 
286 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled,
287     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
288     &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU",
289     "switch siftr module operations on/off");
290 
291 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile,
292     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &siftr_logfile_shadow,
293     sizeof(siftr_logfile_shadow), &siftr_sysctl_logfile_name_handler, "A",
294     "file to save siftr log messages to");
295 
296 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW,
297     &siftr_pkts_per_log, 1,
298     "number of packets between generating a log message");
299 
300 SYSCTL_U16(_net_inet_siftr, OID_AUTO, port_filter, CTLFLAG_RW,
301     &siftr_port_filter, 0,
302     "enable packet filter on a TCP port");
303 
304 /* XXX: TODO
305 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW,
306     &siftr_binary_log, 0,
307     "write log files in binary instead of ascii");
308 */
309 
310 /* Begin functions. */
311 
312 static inline struct flow_hash_node *
313 siftr_find_flow(struct listhead *counter_list, uint32_t id)
314 {
315 	struct flow_hash_node *hash_node;
316 	/*
317 	 * If the list is not empty i.e. the hash index has
318 	 * been used by another flow previously.
319 	 */
320 	if (LIST_FIRST(counter_list) != NULL) {
321 		/*
322 		 * Loop through the hash nodes in the list.
323 		 * There should normally only be 1 hash node in the list.
324 		 */
325 		LIST_FOREACH(hash_node, counter_list, nodes) {
326 			/*
327 			 * Check if the key for the pkt we are currently
328 			 * processing is the same as the key stored in the
329 			 * hash node we are currently processing.
330 			 * If they are the same, then we've found the
331 			 * hash node that stores the counter for the flow
332 			 * the pkt belongs to.
333 			 */
334 			if (hash_node->const_info.key == id) {
335 				return hash_node;
336 			}
337 		}
338 	}
339 
340 	return NULL;
341 }
342 
343 static inline struct flow_hash_node *
344 siftr_new_hash_node(struct flow_info info, int dir,
345 		    struct siftr_stats *ss)
346 {
347 	struct flow_hash_node *hash_node;
348 	struct listhead *counter_list;
349 
350 	counter_list = counter_hash + (info.key & siftr_hashmask);
351 	/* Create a new hash node to store the flow's constant info. */
352 	hash_node = malloc(sizeof(struct flow_hash_node), M_SIFTR_HASHNODE,
353 			   M_NOWAIT|M_ZERO);
354 
355 	if (hash_node != NULL) {
356 		/* Initialise our new hash node list entry. */
357 		hash_node->counter = 0;
358 		hash_node->const_info = info;
359 		LIST_INSERT_HEAD(counter_list, hash_node, nodes);
360 		return hash_node;
361 	} else {
362 		/* malloc failed */
363 		if (dir == DIR_IN)
364 			ss->nskip_in_malloc++;
365 		else
366 			ss->nskip_out_malloc++;
367 
368 		return NULL;
369 	}
370 }
371 
372 static void
373 siftr_process_pkt(struct pkt_node * pkt_node)
374 {
375 	struct flow_hash_node *hash_node;
376 	struct listhead *counter_list;
377 	struct ale *log_buf;
378 
379 	if (pkt_node->flowid == 0) {
380 		panic("%s: flowid not available", __func__);
381 	}
382 
383 	counter_list = counter_hash + (pkt_node->flowid & siftr_hashmask);
384 	hash_node = siftr_find_flow(counter_list, pkt_node->flowid);
385 
386 	if (hash_node == NULL) {
387 		return;
388 	} else if (siftr_pkts_per_log > 1) {
389 		/*
390 		 * Taking the remainder of the counter divided
391 		 * by the current value of siftr_pkts_per_log
392 		 * and storing that in counter provides a neat
393 		 * way to modulate the frequency of log
394 		 * messages being written to the log file.
395 		 */
396 		hash_node->counter = (hash_node->counter + 1) %
397 				     siftr_pkts_per_log;
398 		/*
399 		 * If we have not seen enough packets since the last time
400 		 * we wrote a log message for this connection, return.
401 		 */
402 		if (hash_node->counter > 0)
403 			return;
404 	}
405 
406 	log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK);
407 
408 	if (log_buf == NULL)
409 		return; /* Should only happen if the ALQ is shutting down. */
410 
411 	/* Construct a log message. */
412 	log_buf->ae_bytesused = snprintf(log_buf->ae_data, MAX_LOG_MSG_LEN,
413 	    "%c,%jd.%06ld,%s,%hu,%s,%hu,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,"
414 	    "%u,%u,%u,%u,%u,%u,%u,%u\n",
415 	    direction[pkt_node->direction],
416 	    (intmax_t)pkt_node->tval.tv_sec,
417 	    pkt_node->tval.tv_usec,
418 	    hash_node->const_info.laddr,
419 	    hash_node->const_info.lport,
420 	    hash_node->const_info.faddr,
421 	    hash_node->const_info.fport,
422 	    pkt_node->snd_ssthresh,
423 	    pkt_node->snd_cwnd,
424 	    pkt_node->t_flags2,
425 	    pkt_node->snd_wnd,
426 	    pkt_node->rcv_wnd,
427 	    pkt_node->snd_scale,
428 	    pkt_node->rcv_scale,
429 	    pkt_node->conn_state,
430 	    pkt_node->max_seg_size,
431 	    pkt_node->srtt,
432 	    pkt_node->sack_enabled,
433 	    pkt_node->flags,
434 	    pkt_node->rto,
435 	    pkt_node->snd_buf_hiwater,
436 	    pkt_node->snd_buf_cc,
437 	    pkt_node->rcv_buf_hiwater,
438 	    pkt_node->rcv_buf_cc,
439 	    pkt_node->sent_inflight_bytes,
440 	    pkt_node->t_segqlen,
441 	    pkt_node->flowid,
442 	    pkt_node->flowtype);
443 
444 	alq_post_flags(siftr_alq, log_buf, 0);
445 }
446 
447 static void
448 siftr_pkt_manager_thread(void *arg)
449 {
450 	STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
451 	    STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
452 	struct pkt_node *pkt_node, *pkt_node_temp;
453 	uint8_t draining;
454 
455 	draining = 2;
456 
457 	mtx_lock(&siftr_pkt_mgr_mtx);
458 
459 	/* draining == 0 when queue has been flushed and it's safe to exit. */
460 	while (draining) {
461 		/*
462 		 * Sleep until we are signalled to wake because thread has
463 		 * been told to exit or until 1 tick has passed.
464 		 */
465 		mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
466 		    1);
467 
468 		/* Gain exclusive access to the pkt_node queue. */
469 		mtx_lock(&siftr_pkt_queue_mtx);
470 
471 		/*
472 		 * Move pkt_queue to tmp_pkt_queue, which leaves
473 		 * pkt_queue empty and ready to receive more pkt_nodes.
474 		 */
475 		STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
476 
477 		/*
478 		 * We've finished making changes to the list. Unlock it
479 		 * so the pfil hooks can continue queuing pkt_nodes.
480 		 */
481 		mtx_unlock(&siftr_pkt_queue_mtx);
482 
483 		/*
484 		 * We can't hold a mutex whilst calling siftr_process_pkt
485 		 * because ALQ might sleep waiting for buffer space.
486 		 */
487 		mtx_unlock(&siftr_pkt_mgr_mtx);
488 
489 		/* Flush all pkt_nodes to the log file. */
490 		STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes,
491 		    pkt_node_temp) {
492 			siftr_process_pkt(pkt_node);
493 			STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
494 			free(pkt_node, M_SIFTR_PKTNODE);
495 		}
496 
497 		KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
498 		    ("SIFTR tmp_pkt_queue not empty after flush"));
499 
500 		mtx_lock(&siftr_pkt_mgr_mtx);
501 
502 		/*
503 		 * If siftr_exit_pkt_manager_thread gets set during the window
504 		 * where we are draining the tmp_pkt_queue above, there might
505 		 * still be pkts in pkt_queue that need to be drained.
506 		 * Allow one further iteration to occur after
507 		 * siftr_exit_pkt_manager_thread has been set to ensure
508 		 * pkt_queue is completely empty before we kill the thread.
509 		 *
510 		 * siftr_exit_pkt_manager_thread is set only after the pfil
511 		 * hooks have been removed, so only 1 extra iteration
512 		 * is needed to drain the queue.
513 		 */
514 		if (siftr_exit_pkt_manager_thread)
515 			draining--;
516 	}
517 
518 	mtx_unlock(&siftr_pkt_mgr_mtx);
519 
520 	/* Calls wakeup on this thread's struct thread ptr. */
521 	kthread_exit();
522 }
523 
524 /*
525  * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
526  * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
527  * Return value >0 means the caller should skip processing this mbuf.
528  */
529 static inline int
530 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
531 {
532 	if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
533 	    != NULL) {
534 		if (dir == PFIL_IN)
535 			ss->nskip_in_dejavu++;
536 		else
537 			ss->nskip_out_dejavu++;
538 
539 		return (1);
540 	} else {
541 		struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
542 		    PACKET_TAG_SIFTR, 0, M_NOWAIT);
543 		if (tag == NULL) {
544 			if (dir == PFIL_IN)
545 				ss->nskip_in_malloc++;
546 			else
547 				ss->nskip_out_malloc++;
548 
549 			return (1);
550 		}
551 
552 		m_tag_prepend(m, tag);
553 	}
554 
555 	return (0);
556 }
557 
558 /*
559  * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
560  * otherwise.
561  */
562 static inline struct inpcb *
563 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
564     uint16_t dport, int dir, struct siftr_stats *ss)
565 {
566 	struct inpcb *inp;
567 
568 	/* We need the tcbinfo lock. */
569 	INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
570 
571 	if (dir == PFIL_IN)
572 		inp = (ipver == INP_IPV4 ?
573 		    in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
574 		    dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
575 		    :
576 #ifdef SIFTR_IPV6
577 		    in6_pcblookup(&V_tcbinfo,
578 		    &((struct ip6_hdr *)ip)->ip6_src, sport,
579 		    &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB,
580 		    m->m_pkthdr.rcvif)
581 #else
582 		    NULL
583 #endif
584 		    );
585 
586 	else
587 		inp = (ipver == INP_IPV4 ?
588 		    in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
589 		    sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
590 		    :
591 #ifdef SIFTR_IPV6
592 		    in6_pcblookup(&V_tcbinfo,
593 		    &((struct ip6_hdr *)ip)->ip6_dst, dport,
594 		    &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB,
595 		    m->m_pkthdr.rcvif)
596 #else
597 		    NULL
598 #endif
599 		    );
600 
601 	/* If we can't find the inpcb, bail. */
602 	if (inp == NULL) {
603 		if (dir == PFIL_IN)
604 			ss->nskip_in_inpcb++;
605 		else
606 			ss->nskip_out_inpcb++;
607 	}
608 
609 	return (inp);
610 }
611 
612 static inline uint32_t
613 siftr_get_flowid(struct inpcb *inp, int ipver, uint32_t *phashtype)
614 {
615 	if (inp->inp_flowid == 0) {
616 #ifdef SIFTR_IPV6
617 		if (ipver == INP_IPV6) {
618 			return fib6_calc_packet_hash(&inp->in6p_laddr,
619 						     &inp->in6p_faddr,
620 						     inp->inp_lport,
621 						     inp->inp_fport,
622 						     IPPROTO_TCP,
623 						     phashtype);
624 		} else
625 #endif
626 		{
627 			return fib4_calc_packet_hash(inp->inp_laddr,
628 						     inp->inp_faddr,
629 						     inp->inp_lport,
630 						     inp->inp_fport,
631 						     IPPROTO_TCP,
632 						     phashtype);
633 		}
634 	} else {
635 		*phashtype = inp->inp_flowtype;
636 		return inp->inp_flowid;
637 	}
638 }
639 
640 static inline void
641 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
642     int ipver, int dir, int inp_locally_locked)
643 {
644 	pn->ipver = ipver;
645 	pn->tcp_localport = inp->inp_lport;
646 	pn->tcp_foreignport = inp->inp_fport;
647 	pn->snd_cwnd = tp->snd_cwnd;
648 	pn->snd_wnd = tp->snd_wnd;
649 	pn->rcv_wnd = tp->rcv_wnd;
650 	pn->t_flags2 = tp->t_flags2;
651 	pn->snd_ssthresh = tp->snd_ssthresh;
652 	pn->snd_scale = tp->snd_scale;
653 	pn->rcv_scale = tp->rcv_scale;
654 	pn->conn_state = tp->t_state;
655 	pn->max_seg_size = tp->t_maxseg;
656 	pn->srtt = ((uint64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
657 	pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
658 	pn->flags = tp->t_flags;
659 	pn->rto = tp->t_rxtcur * tick;
660 	pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
661 	pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd);
662 	pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
663 	pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv);
664 	pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
665 	pn->t_segqlen = tp->t_segqlen;
666 
667 	/* We've finished accessing the tcb so release the lock. */
668 	if (inp_locally_locked)
669 		INP_RUNLOCK(inp);
670 
671 	pn->direction = (dir == PFIL_IN ? DIR_IN : DIR_OUT);
672 
673 	/*
674 	 * Significantly more accurate than using getmicrotime(), but slower!
675 	 * Gives true microsecond resolution at the expense of a hit to
676 	 * maximum pps throughput processing when SIFTR is loaded and enabled.
677 	 */
678 	microtime(&pn->tval);
679 	TCP_PROBE1(siftr, pn);
680 }
681 
682 /*
683  * pfil hook that is called for each IPv4 packet making its way through the
684  * stack in either direction.
685  * The pfil subsystem holds a non-sleepable mutex somewhere when
686  * calling our hook function, so we can't sleep at all.
687  * It's very important to use the M_NOWAIT flag with all function calls
688  * that support it so that they won't sleep, otherwise you get a panic.
689  */
690 static pfil_return_t
691 siftr_chkpkt(struct mbuf **m, struct ifnet *ifp, int flags,
692     void *ruleset __unused, struct inpcb *inp)
693 {
694 	struct pkt_node *pn;
695 	struct ip *ip;
696 	struct tcphdr *th;
697 	struct tcpcb *tp;
698 	struct siftr_stats *ss;
699 	unsigned int ip_hl;
700 	int inp_locally_locked, dir;
701 	uint32_t hash_id, hash_type;
702 	struct listhead *counter_list;
703 	struct flow_hash_node *hash_node;
704 
705 	inp_locally_locked = 0;
706 	dir = PFIL_DIR(flags);
707 	ss = DPCPU_PTR(ss);
708 
709 	/*
710 	 * m_pullup is not required here because ip_{input|output}
711 	 * already do the heavy lifting for us.
712 	 */
713 
714 	ip = mtod(*m, struct ip *);
715 
716 	/* Only continue processing if the packet is TCP. */
717 	if (ip->ip_p != IPPROTO_TCP)
718 		goto ret;
719 
720 	/*
721 	 * Create a tcphdr struct starting at the correct offset
722 	 * in the IP packet. ip->ip_hl gives the ip header length
723 	 * in 4-byte words, so multiply it to get the size in bytes.
724 	 */
725 	ip_hl = (ip->ip_hl << 2);
726 	th = (struct tcphdr *)((caddr_t)ip + ip_hl);
727 
728 	/*
729 	 * Only pkts selected by the tcp port filter
730 	 * can be inserted into the pkt_queue
731 	 */
732 	if ((siftr_port_filter != 0) &&
733 	    (siftr_port_filter != ntohs(th->th_sport)) &&
734 	    (siftr_port_filter != ntohs(th->th_dport))) {
735 		goto ret;
736 	}
737 
738 	/*
739 	 * If a kernel subsystem reinjects packets into the stack, our pfil
740 	 * hook will be called multiple times for the same packet.
741 	 * Make sure we only process unique packets.
742 	 */
743 	if (siftr_chkreinject(*m, dir, ss))
744 		goto ret;
745 
746 	if (dir == PFIL_IN)
747 		ss->n_in++;
748 	else
749 		ss->n_out++;
750 
751 	/*
752 	 * If the pfil hooks don't provide a pointer to the
753 	 * inpcb, we need to find it ourselves and lock it.
754 	 */
755 	if (!inp) {
756 		/* Find the corresponding inpcb for this pkt. */
757 		inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
758 		    th->th_dport, dir, ss);
759 
760 		if (inp == NULL)
761 			goto ret;
762 		else
763 			inp_locally_locked = 1;
764 	}
765 
766 	INP_LOCK_ASSERT(inp);
767 
768 	/* Find the TCP control block that corresponds with this packet */
769 	tp = intotcpcb(inp);
770 
771 	/*
772 	 * If we can't find the TCP control block (happens occasionaly for a
773 	 * packet sent during the shutdown phase of a TCP connection), or the
774 	 * TCP control block has not initialized (happens during TCPS_SYN_SENT),
775 	 * bail.
776 	 */
777 	if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) {
778 		if (dir == PFIL_IN)
779 			ss->nskip_in_tcpcb++;
780 		else
781 			ss->nskip_out_tcpcb++;
782 
783 		goto inp_unlock;
784 	}
785 
786 	hash_id = siftr_get_flowid(inp, INP_IPV4, &hash_type);
787 	counter_list = counter_hash + (hash_id & siftr_hashmask);
788 	hash_node = siftr_find_flow(counter_list, hash_id);
789 
790 	/* If this flow hasn't been seen before, we create a new entry. */
791 	if (hash_node == NULL) {
792 		struct flow_info info;
793 
794 		inet_ntoa_r(inp->inp_laddr, info.laddr);
795 		inet_ntoa_r(inp->inp_faddr, info.faddr);
796 		info.lport = ntohs(inp->inp_lport);
797 		info.fport = ntohs(inp->inp_fport);
798 		info.key = hash_id;
799 		info.ipver = INP_IPV4;
800 
801 		hash_node = siftr_new_hash_node(info, dir, ss);
802 	}
803 
804 	if (hash_node == NULL) {
805 		goto inp_unlock;
806 	}
807 
808 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
809 
810 	if (pn == NULL) {
811 		if (dir == PFIL_IN)
812 			ss->nskip_in_malloc++;
813 		else
814 			ss->nskip_out_malloc++;
815 
816 		goto inp_unlock;
817 	}
818 
819 	pn->flowid = hash_id;
820 	pn->flowtype = hash_type;
821 
822 	siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
823 
824 	mtx_lock(&siftr_pkt_queue_mtx);
825 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
826 	mtx_unlock(&siftr_pkt_queue_mtx);
827 	goto ret;
828 
829 inp_unlock:
830 	if (inp_locally_locked)
831 		INP_RUNLOCK(inp);
832 
833 ret:
834 	return (PFIL_PASS);
835 }
836 
837 #ifdef SIFTR_IPV6
838 static pfil_return_t
839 siftr_chkpkt6(struct mbuf **m, struct ifnet *ifp, int flags,
840     void *ruleset __unused, struct inpcb *inp)
841 {
842 	struct pkt_node *pn;
843 	struct ip6_hdr *ip6;
844 	struct tcphdr *th;
845 	struct tcpcb *tp;
846 	struct siftr_stats *ss;
847 	unsigned int ip6_hl;
848 	int inp_locally_locked, dir;
849 	uint32_t hash_id, hash_type;
850 	struct listhead *counter_list;
851 	struct flow_hash_node *hash_node;
852 
853 	inp_locally_locked = 0;
854 	dir = PFIL_DIR(flags);
855 	ss = DPCPU_PTR(ss);
856 
857 	/*
858 	 * m_pullup is not required here because ip6_{input|output}
859 	 * already do the heavy lifting for us.
860 	 */
861 
862 	ip6 = mtod(*m, struct ip6_hdr *);
863 
864 	/*
865 	 * Only continue processing if the packet is TCP
866 	 * XXX: We should follow the next header fields
867 	 * as shown on Pg 6 RFC 2460, but right now we'll
868 	 * only check pkts that have no extension headers.
869 	 */
870 	if (ip6->ip6_nxt != IPPROTO_TCP)
871 		goto ret6;
872 
873 	/*
874 	 * Create a tcphdr struct starting at the correct offset
875 	 * in the ipv6 packet.
876 	 */
877 	ip6_hl = sizeof(struct ip6_hdr);
878 	th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
879 
880 	/*
881 	 * Only pkts selected by the tcp port filter
882 	 * can be inserted into the pkt_queue
883 	 */
884 	if ((siftr_port_filter != 0) &&
885 	    (siftr_port_filter != ntohs(th->th_sport)) &&
886 	    (siftr_port_filter != ntohs(th->th_dport))) {
887 		goto ret6;
888 	}
889 
890 	/*
891 	 * If a kernel subsystem reinjects packets into the stack, our pfil
892 	 * hook will be called multiple times for the same packet.
893 	 * Make sure we only process unique packets.
894 	 */
895 	if (siftr_chkreinject(*m, dir, ss))
896 		goto ret6;
897 
898 	if (dir == PFIL_IN)
899 		ss->n_in++;
900 	else
901 		ss->n_out++;
902 
903 	/*
904 	 * For inbound packets, the pfil hooks don't provide a pointer to the
905 	 * inpcb, so we need to find it ourselves and lock it.
906 	 */
907 	if (!inp) {
908 		/* Find the corresponding inpcb for this pkt. */
909 		inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
910 		    th->th_sport, th->th_dport, dir, ss);
911 
912 		if (inp == NULL)
913 			goto ret6;
914 		else
915 			inp_locally_locked = 1;
916 	}
917 
918 	/* Find the TCP control block that corresponds with this packet. */
919 	tp = intotcpcb(inp);
920 
921 	/*
922 	 * If we can't find the TCP control block (happens occasionaly for a
923 	 * packet sent during the shutdown phase of a TCP connection), or the
924 	 * TCP control block has not initialized (happens during TCPS_SYN_SENT),
925 	 * bail.
926 	 */
927 	if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) {
928 		if (dir == PFIL_IN)
929 			ss->nskip_in_tcpcb++;
930 		else
931 			ss->nskip_out_tcpcb++;
932 
933 		goto inp_unlock6;
934 	}
935 
936 	hash_id = siftr_get_flowid(inp, INP_IPV6, &hash_type);
937 	counter_list = counter_hash + (hash_id & siftr_hashmask);
938 	hash_node = siftr_find_flow(counter_list, hash_id);
939 
940 	/* If this flow hasn't been seen before, we create a new entry. */
941 	if (!hash_node) {
942 		struct flow_info info;
943 
944 		ip6_sprintf(info.laddr, &inp->in6p_laddr);
945 		ip6_sprintf(info.faddr, &inp->in6p_faddr);
946 		info.lport = ntohs(inp->inp_lport);
947 		info.fport = ntohs(inp->inp_fport);
948 		info.key = hash_id;
949 		info.ipver = INP_IPV6;
950 
951 		hash_node = siftr_new_hash_node(info, dir, ss);
952 	}
953 
954 	if (!hash_node) {
955 		goto inp_unlock6;
956 	}
957 
958 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
959 
960 	if (pn == NULL) {
961 		if (dir == PFIL_IN)
962 			ss->nskip_in_malloc++;
963 		else
964 			ss->nskip_out_malloc++;
965 
966 		goto inp_unlock6;
967 	}
968 
969 	pn->flowid = hash_id;
970 	pn->flowtype = hash_type;
971 
972 	siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
973 
974 	mtx_lock(&siftr_pkt_queue_mtx);
975 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
976 	mtx_unlock(&siftr_pkt_queue_mtx);
977 	goto ret6;
978 
979 inp_unlock6:
980 	if (inp_locally_locked)
981 		INP_RUNLOCK(inp);
982 
983 ret6:
984 	return (PFIL_PASS);
985 }
986 #endif /* #ifdef SIFTR_IPV6 */
987 
988 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet_hook);
989 #define	V_siftr_inet_hook	VNET(siftr_inet_hook)
990 #ifdef SIFTR_IPV6
991 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet6_hook);
992 #define	V_siftr_inet6_hook	VNET(siftr_inet6_hook)
993 #endif
994 static int
995 siftr_pfil(int action)
996 {
997 	struct pfil_hook_args pha = {
998 		.pa_version = PFIL_VERSION,
999 		.pa_flags = PFIL_IN | PFIL_OUT,
1000 		.pa_modname = "siftr",
1001 		.pa_rulname = "default",
1002 	};
1003 	struct pfil_link_args pla = {
1004 		.pa_version = PFIL_VERSION,
1005 		.pa_flags = PFIL_IN | PFIL_OUT | PFIL_HEADPTR | PFIL_HOOKPTR,
1006 	};
1007 
1008 	VNET_ITERATOR_DECL(vnet_iter);
1009 
1010 	VNET_LIST_RLOCK();
1011 	VNET_FOREACH(vnet_iter) {
1012 		CURVNET_SET(vnet_iter);
1013 
1014 		if (action == HOOK) {
1015 			pha.pa_mbuf_chk = siftr_chkpkt;
1016 			pha.pa_type = PFIL_TYPE_IP4;
1017 			V_siftr_inet_hook = pfil_add_hook(&pha);
1018 			pla.pa_hook = V_siftr_inet_hook;
1019 			pla.pa_head = V_inet_pfil_head;
1020 			(void)pfil_link(&pla);
1021 #ifdef SIFTR_IPV6
1022 			pha.pa_mbuf_chk = siftr_chkpkt6;
1023 			pha.pa_type = PFIL_TYPE_IP6;
1024 			V_siftr_inet6_hook = pfil_add_hook(&pha);
1025 			pla.pa_hook = V_siftr_inet6_hook;
1026 			pla.pa_head = V_inet6_pfil_head;
1027 			(void)pfil_link(&pla);
1028 #endif
1029 		} else if (action == UNHOOK) {
1030 			pfil_remove_hook(V_siftr_inet_hook);
1031 #ifdef SIFTR_IPV6
1032 			pfil_remove_hook(V_siftr_inet6_hook);
1033 #endif
1034 		}
1035 		CURVNET_RESTORE();
1036 	}
1037 	VNET_LIST_RUNLOCK();
1038 
1039 	return (0);
1040 }
1041 
1042 static int
1043 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1044 {
1045 	struct alq *new_alq;
1046 	int error;
1047 
1048 	error = sysctl_handle_string(oidp, arg1, arg2, req);
1049 
1050 	/* Check for error or same filename */
1051 	if (error != 0 || req->newptr == NULL ||
1052 	    strncmp(siftr_logfile, arg1, arg2) == 0)
1053 		goto done;
1054 
1055 	/* file name changed */
1056 	error = alq_open(&new_alq, arg1, curthread->td_ucred,
1057 	    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1058 	if (error != 0)
1059 		goto done;
1060 
1061 	/*
1062 	 * If disabled, siftr_alq == NULL so we simply close
1063 	 * the alq as we've proved it can be opened.
1064 	 * If enabled, close the existing alq and switch the old
1065 	 * for the new.
1066 	 */
1067 	if (siftr_alq == NULL) {
1068 		alq_close(new_alq);
1069 	} else {
1070 		alq_close(siftr_alq);
1071 		siftr_alq = new_alq;
1072 	}
1073 
1074 	/* Update filename upon success */
1075 	strlcpy(siftr_logfile, arg1, arg2);
1076 done:
1077 	return (error);
1078 }
1079 
1080 static int
1081 siftr_manage_ops(uint8_t action)
1082 {
1083 	struct siftr_stats totalss;
1084 	struct timeval tval;
1085 	struct flow_hash_node *counter, *tmp_counter;
1086 	struct sbuf *s;
1087 	int i, error;
1088 	uint32_t bytes_to_write, total_skipped_pkts;
1089 
1090 	error = 0;
1091 	total_skipped_pkts = 0;
1092 
1093 	/* Init an autosizing sbuf that initially holds 200 chars. */
1094 	if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1095 		return (-1);
1096 
1097 	if (action == SIFTR_ENABLE && siftr_pkt_manager_thr == NULL) {
1098 		/*
1099 		 * Create our alq
1100 		 * XXX: We should abort if alq_open fails!
1101 		 */
1102 		alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1103 		    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1104 
1105 		STAILQ_INIT(&pkt_queue);
1106 
1107 		DPCPU_ZERO(ss);
1108 
1109 		siftr_exit_pkt_manager_thread = 0;
1110 
1111 		kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1112 		    &siftr_pkt_manager_thr, RFNOWAIT, 0,
1113 		    "siftr_pkt_manager_thr");
1114 
1115 		siftr_pfil(HOOK);
1116 
1117 		microtime(&tval);
1118 
1119 		sbuf_printf(s,
1120 		    "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1121 		    "siftrver=%s\tsysname=%s\tsysver=%u\tipmode=%u\n",
1122 		    (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR,
1123 		    SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1124 
1125 		sbuf_finish(s);
1126 		alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1127 
1128 	} else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1129 		/*
1130 		 * Remove the pfil hook functions. All threads currently in
1131 		 * the hook functions are allowed to exit before siftr_pfil()
1132 		 * returns.
1133 		 */
1134 		siftr_pfil(UNHOOK);
1135 
1136 		/* This will block until the pkt manager thread unlocks it. */
1137 		mtx_lock(&siftr_pkt_mgr_mtx);
1138 
1139 		/* Tell the pkt manager thread that it should exit now. */
1140 		siftr_exit_pkt_manager_thread = 1;
1141 
1142 		/*
1143 		 * Wake the pkt_manager thread so it realises that
1144 		 * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1145 		 * The wakeup won't be delivered until we unlock
1146 		 * siftr_pkt_mgr_mtx so this isn't racy.
1147 		 */
1148 		wakeup(&wait_for_pkt);
1149 
1150 		/* Wait for the pkt_manager thread to exit. */
1151 		mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1152 		    "thrwait", 0);
1153 
1154 		siftr_pkt_manager_thr = NULL;
1155 		mtx_unlock(&siftr_pkt_mgr_mtx);
1156 
1157 		totalss.n_in = DPCPU_VARSUM(ss, n_in);
1158 		totalss.n_out = DPCPU_VARSUM(ss, n_out);
1159 		totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1160 		totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1161 		totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1162 		totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1163 		totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1164 		totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1165 
1166 		total_skipped_pkts = totalss.nskip_in_malloc +
1167 		    totalss.nskip_out_malloc + totalss.nskip_in_tcpcb +
1168 		    totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1169 		    totalss.nskip_out_inpcb;
1170 
1171 		microtime(&tval);
1172 
1173 		sbuf_printf(s,
1174 		    "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1175 		    "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1176 		    "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1177 		    "num_outbound_skipped_pkts_malloc=%u\t"
1178 		    "num_inbound_skipped_pkts_tcpcb=%u\t"
1179 		    "num_outbound_skipped_pkts_tcpcb=%u\t"
1180 		    "num_inbound_skipped_pkts_inpcb=%u\t"
1181 		    "num_outbound_skipped_pkts_inpcb=%u\t"
1182 		    "total_skipped_tcp_pkts=%u\tflow_list=",
1183 		    (intmax_t)tval.tv_sec,
1184 		    tval.tv_usec,
1185 		    (uintmax_t)totalss.n_in,
1186 		    (uintmax_t)totalss.n_out,
1187 		    (uintmax_t)(totalss.n_in + totalss.n_out),
1188 		    totalss.nskip_in_malloc,
1189 		    totalss.nskip_out_malloc,
1190 		    totalss.nskip_in_tcpcb,
1191 		    totalss.nskip_out_tcpcb,
1192 		    totalss.nskip_in_inpcb,
1193 		    totalss.nskip_out_inpcb,
1194 		    total_skipped_pkts);
1195 
1196 		/*
1197 		 * Iterate over the flow hash, printing a summary of each
1198 		 * flow seen and freeing any malloc'd memory.
1199 		 * The hash consists of an array of LISTs (man 3 queue).
1200 		 */
1201 		for (i = 0; i <= siftr_hashmask; i++) {
1202 			LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1203 			    tmp_counter) {
1204 				sbuf_printf(s, "%s;%hu-%s;%hu,",
1205 					    counter->const_info.laddr,
1206 					    counter->const_info.lport,
1207 					    counter->const_info.faddr,
1208 					    counter->const_info.fport);
1209 
1210 				free(counter, M_SIFTR_HASHNODE);
1211 			}
1212 
1213 			LIST_INIT(counter_hash + i);
1214 		}
1215 
1216 		sbuf_printf(s, "\n");
1217 		sbuf_finish(s);
1218 
1219 		i = 0;
1220 		do {
1221 			bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1222 			alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1223 			i += bytes_to_write;
1224 		} while (i < sbuf_len(s));
1225 
1226 		alq_close(siftr_alq);
1227 		siftr_alq = NULL;
1228 	} else
1229 		error = EINVAL;
1230 
1231 	sbuf_delete(s);
1232 
1233 	/*
1234 	 * XXX: Should be using ret to check if any functions fail
1235 	 * and set error appropriately
1236 	 */
1237 
1238 	return (error);
1239 }
1240 
1241 static int
1242 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1243 {
1244 	int error;
1245 	uint32_t new;
1246 
1247 	new = siftr_enabled;
1248 	error = sysctl_handle_int(oidp, &new, 0, req);
1249 	if (error == 0 && req->newptr != NULL) {
1250 		if (new > 1)
1251 			return (EINVAL);
1252 		else if (new != siftr_enabled) {
1253 			if ((error = siftr_manage_ops(new)) == 0) {
1254 				siftr_enabled = new;
1255 			} else {
1256 				siftr_manage_ops(SIFTR_DISABLE);
1257 			}
1258 		}
1259 	}
1260 
1261 	return (error);
1262 }
1263 
1264 static void
1265 siftr_shutdown_handler(void *arg)
1266 {
1267 	if (siftr_enabled == 1) {
1268 		siftr_manage_ops(SIFTR_DISABLE);
1269 	}
1270 }
1271 
1272 /*
1273  * Module is being unloaded or machine is shutting down. Take care of cleanup.
1274  */
1275 static int
1276 deinit_siftr(void)
1277 {
1278 	/* Cleanup. */
1279 	siftr_manage_ops(SIFTR_DISABLE);
1280 	hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1281 	mtx_destroy(&siftr_pkt_queue_mtx);
1282 	mtx_destroy(&siftr_pkt_mgr_mtx);
1283 
1284 	return (0);
1285 }
1286 
1287 /*
1288  * Module has just been loaded into the kernel.
1289  */
1290 static int
1291 init_siftr(void)
1292 {
1293 	EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL,
1294 	    SHUTDOWN_PRI_FIRST);
1295 
1296 	/* Initialise our flow counter hash table. */
1297 	counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1298 	    &siftr_hashmask);
1299 
1300 	mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1301 	mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1302 
1303 	/* Print message to the user's current terminal. */
1304 	uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1305 	    "          http://caia.swin.edu.au/urp/newtcp\n\n",
1306 	    MODVERSION_STR);
1307 
1308 	return (0);
1309 }
1310 
1311 /*
1312  * This is the function that is called to load and unload the module.
1313  * When the module is loaded, this function is called once with
1314  * "what" == MOD_LOAD
1315  * When the module is unloaded, this function is called twice with
1316  * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1317  * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1318  * this function is called once with "what" = MOD_SHUTDOWN
1319  * When the system is shut down, the handler isn't called until the very end
1320  * of the shutdown sequence i.e. after the disks have been synced.
1321  */
1322 static int
1323 siftr_load_handler(module_t mod, int what, void *arg)
1324 {
1325 	int ret;
1326 
1327 	switch (what) {
1328 	case MOD_LOAD:
1329 		ret = init_siftr();
1330 		break;
1331 
1332 	case MOD_QUIESCE:
1333 	case MOD_SHUTDOWN:
1334 		ret = deinit_siftr();
1335 		break;
1336 
1337 	case MOD_UNLOAD:
1338 		ret = 0;
1339 		break;
1340 
1341 	default:
1342 		ret = EINVAL;
1343 		break;
1344 	}
1345 
1346 	return (ret);
1347 }
1348 
1349 static moduledata_t siftr_mod = {
1350 	.name = "siftr",
1351 	.evhand = siftr_load_handler,
1352 };
1353 
1354 /*
1355  * Param 1: name of the kernel module
1356  * Param 2: moduledata_t struct containing info about the kernel module
1357  *          and the execution entry point for the module
1358  * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1359  *          Defines the module initialisation order
1360  * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1361  *          Defines the initialisation order of this kld relative to others
1362  *          within the same subsystem as defined by param 3
1363  */
1364 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_LAST, SI_ORDER_ANY);
1365 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1366 MODULE_VERSION(siftr, MODVERSION);
1367