xref: /freebsd/sys/netinet/siftr.c (revision 0fac350c54d0a72f5341e15021efcde63eb58a4b)
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/param.h>
64 #include <sys/alq.h>
65 #include <sys/errno.h>
66 #include <sys/eventhandler.h>
67 #include <sys/hash.h>
68 #include <sys/kernel.h>
69 #include <sys/kthread.h>
70 #include <sys/lock.h>
71 #include <sys/mbuf.h>
72 #include <sys/module.h>
73 #include <sys/mutex.h>
74 #include <sys/pcpu.h>
75 #include <sys/proc.h>
76 #include <sys/reboot.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 int
374 siftr_process_pkt(struct pkt_node * pkt_node, char *buf)
375 {
376 	struct flow_hash_node *hash_node;
377 	struct listhead *counter_list;
378 	int ret_sz;
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 0;
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 0;
405 	}
406 
407 	/* Construct a log message. */
408 	ret_sz = snprintf(buf, MAX_LOG_MSG_LEN,
409 	    "%c,%jd.%06ld,%s,%hu,%s,%hu,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,"
410 	    "%u,%u,%u,%u,%u,%u,%u,%u\n",
411 	    direction[pkt_node->direction],
412 	    (intmax_t)pkt_node->tval.tv_sec,
413 	    pkt_node->tval.tv_usec,
414 	    hash_node->const_info.laddr,
415 	    hash_node->const_info.lport,
416 	    hash_node->const_info.faddr,
417 	    hash_node->const_info.fport,
418 	    pkt_node->snd_ssthresh,
419 	    pkt_node->snd_cwnd,
420 	    pkt_node->t_flags2,
421 	    pkt_node->snd_wnd,
422 	    pkt_node->rcv_wnd,
423 	    pkt_node->snd_scale,
424 	    pkt_node->rcv_scale,
425 	    pkt_node->conn_state,
426 	    pkt_node->mss,
427 	    pkt_node->srtt,
428 	    pkt_node->sack_enabled,
429 	    pkt_node->t_flags,
430 	    pkt_node->rto,
431 	    pkt_node->snd_buf_hiwater,
432 	    pkt_node->snd_buf_cc,
433 	    pkt_node->rcv_buf_hiwater,
434 	    pkt_node->rcv_buf_cc,
435 	    pkt_node->sent_inflight_bytes,
436 	    pkt_node->t_segqlen,
437 	    pkt_node->flowid,
438 	    pkt_node->flowtype);
439 
440 	return ret_sz;
441 }
442 
443 static void
444 siftr_pkt_manager_thread(void *arg)
445 {
446 	STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
447 	    STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
448 	struct pkt_node *pkt_node, *pkt_node_temp;
449 	uint8_t draining;
450 	struct ale *log_buf;
451 	int ret_sz, cnt;
452 	char *bufp;
453 
454 	draining = 2;
455 
456 	mtx_lock(&siftr_pkt_mgr_mtx);
457 
458 	/* draining == 0 when queue has been flushed and it's safe to exit. */
459 	while (draining) {
460 		/*
461 		 * Sleep until we are signalled to wake because thread has
462 		 * been told to exit or until 1 tick has passed.
463 		 */
464 		mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
465 		    1);
466 
467 		/* Gain exclusive access to the pkt_node queue. */
468 		mtx_lock(&siftr_pkt_queue_mtx);
469 
470 		/*
471 		 * Move pkt_queue to tmp_pkt_queue, which leaves
472 		 * pkt_queue empty and ready to receive more pkt_nodes.
473 		 */
474 		STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
475 
476 		/*
477 		 * We've finished making changes to the list. Unlock it
478 		 * so the pfil hooks can continue queuing pkt_nodes.
479 		 */
480 		mtx_unlock(&siftr_pkt_queue_mtx);
481 
482 		/*
483 		 * We can't hold a mutex whilst calling siftr_process_pkt
484 		 * because ALQ might sleep waiting for buffer space.
485 		 */
486 		mtx_unlock(&siftr_pkt_mgr_mtx);
487 
488 try_again:
489 		pkt_node = STAILQ_FIRST(&tmp_pkt_queue);
490 		if (pkt_node != NULL) {
491 			if (STAILQ_NEXT(pkt_node, nodes) != NULL) {
492 				cnt = 3;
493 			} else {
494 				cnt = 1;
495 			}
496 
497 			log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN * cnt,
498 					   ALQ_WAITOK);
499 
500 			if (log_buf != NULL) {
501 				log_buf->ae_bytesused = 0;
502 				bufp = log_buf->ae_data;
503 			} else {
504 				/*
505 				 * Should only happen if the ALQ is shutting
506 				 * down.
507 				 */
508 				bufp = NULL;
509 			}
510 
511 			/* Flush all pkt_nodes to the log file. */
512 			STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes,
513 			    pkt_node_temp) {
514 				if (log_buf != NULL) {
515 					ret_sz = siftr_process_pkt(pkt_node,
516 								   bufp);
517 					bufp += ret_sz;
518 					log_buf->ae_bytesused += ret_sz;
519 					cnt--;
520 				}
521 
522 				STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
523 				free(pkt_node, M_SIFTR_PKTNODE);
524 
525 				if (cnt <= 0 && !STAILQ_EMPTY(&tmp_pkt_queue)) {
526 					alq_post_flags(siftr_alq, log_buf, 0);
527 					goto try_again;
528 				}
529 			}
530 			if (log_buf != NULL) {
531 				alq_post_flags(siftr_alq, log_buf, 0);
532 			}
533 		}
534 
535 		KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
536 		    ("SIFTR tmp_pkt_queue not empty after flush"));
537 
538 		mtx_lock(&siftr_pkt_mgr_mtx);
539 
540 		/*
541 		 * If siftr_exit_pkt_manager_thread gets set during the window
542 		 * where we are draining the tmp_pkt_queue above, there might
543 		 * still be pkts in pkt_queue that need to be drained.
544 		 * Allow one further iteration to occur after
545 		 * siftr_exit_pkt_manager_thread has been set to ensure
546 		 * pkt_queue is completely empty before we kill the thread.
547 		 *
548 		 * siftr_exit_pkt_manager_thread is set only after the pfil
549 		 * hooks have been removed, so only 1 extra iteration
550 		 * is needed to drain the queue.
551 		 */
552 		if (siftr_exit_pkt_manager_thread)
553 			draining--;
554 	}
555 
556 	mtx_unlock(&siftr_pkt_mgr_mtx);
557 
558 	/* Calls wakeup on this thread's struct thread ptr. */
559 	kthread_exit();
560 }
561 
562 /*
563  * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
564  * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
565  * Return value >0 means the caller should skip processing this mbuf.
566  */
567 static inline int
568 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
569 {
570 	if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
571 	    != NULL) {
572 		if (dir == PFIL_IN)
573 			ss->nskip_in_dejavu++;
574 		else
575 			ss->nskip_out_dejavu++;
576 
577 		return (1);
578 	} else {
579 		struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
580 		    PACKET_TAG_SIFTR, 0, M_NOWAIT);
581 		if (tag == NULL) {
582 			if (dir == PFIL_IN)
583 				ss->nskip_in_malloc++;
584 			else
585 				ss->nskip_out_malloc++;
586 
587 			return (1);
588 		}
589 
590 		m_tag_prepend(m, tag);
591 	}
592 
593 	return (0);
594 }
595 
596 /*
597  * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
598  * otherwise.
599  */
600 static inline struct inpcb *
601 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
602     uint16_t dport, int dir, struct siftr_stats *ss)
603 {
604 	struct inpcb *inp;
605 
606 	/* We need the tcbinfo lock. */
607 	INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
608 
609 	if (dir == PFIL_IN)
610 		inp = (ipver == INP_IPV4 ?
611 		    in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
612 		    dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
613 		    :
614 #ifdef SIFTR_IPV6
615 		    in6_pcblookup(&V_tcbinfo,
616 		    &((struct ip6_hdr *)ip)->ip6_src, sport,
617 		    &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB,
618 		    m->m_pkthdr.rcvif)
619 #else
620 		    NULL
621 #endif
622 		    );
623 
624 	else
625 		inp = (ipver == INP_IPV4 ?
626 		    in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
627 		    sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
628 		    :
629 #ifdef SIFTR_IPV6
630 		    in6_pcblookup(&V_tcbinfo,
631 		    &((struct ip6_hdr *)ip)->ip6_dst, dport,
632 		    &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB,
633 		    m->m_pkthdr.rcvif)
634 #else
635 		    NULL
636 #endif
637 		    );
638 
639 	/* If we can't find the inpcb, bail. */
640 	if (inp == NULL) {
641 		if (dir == PFIL_IN)
642 			ss->nskip_in_inpcb++;
643 		else
644 			ss->nskip_out_inpcb++;
645 	}
646 
647 	return (inp);
648 }
649 
650 static inline uint32_t
651 siftr_get_flowid(struct inpcb *inp, int ipver, uint32_t *phashtype)
652 {
653 	if (inp->inp_flowid == 0) {
654 #ifdef SIFTR_IPV6
655 		if (ipver == INP_IPV6) {
656 			return fib6_calc_packet_hash(&inp->in6p_laddr,
657 						     &inp->in6p_faddr,
658 						     inp->inp_lport,
659 						     inp->inp_fport,
660 						     IPPROTO_TCP,
661 						     phashtype);
662 		} else
663 #endif
664 		{
665 			return fib4_calc_packet_hash(inp->inp_laddr,
666 						     inp->inp_faddr,
667 						     inp->inp_lport,
668 						     inp->inp_fport,
669 						     IPPROTO_TCP,
670 						     phashtype);
671 		}
672 	} else {
673 		*phashtype = inp->inp_flowtype;
674 		return inp->inp_flowid;
675 	}
676 }
677 
678 static inline void
679 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
680     int ipver, int dir, int inp_locally_locked)
681 {
682 	pn->ipver = ipver;
683 	pn->lport = inp->inp_lport;
684 	pn->fport = inp->inp_fport;
685 	pn->laddr = inp->inp_inc.inc_ie.ie_dependladdr;
686 	pn->faddr = inp->inp_inc.inc_ie.ie_dependfaddr;
687 	pn->snd_cwnd = tp->snd_cwnd;
688 	pn->snd_wnd = tp->snd_wnd;
689 	pn->rcv_wnd = tp->rcv_wnd;
690 	pn->t_flags2 = tp->t_flags2;
691 	pn->snd_ssthresh = tp->snd_ssthresh;
692 	pn->snd_scale = tp->snd_scale;
693 	pn->rcv_scale = tp->rcv_scale;
694 	pn->conn_state = tp->t_state;
695 	pn->mss = tp->t_maxseg;
696 	pn->srtt = ((uint64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
697 	pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
698 	pn->t_flags = tp->t_flags;
699 	pn->rto = tp->t_rxtcur * tick;
700 	pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
701 	pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd);
702 	pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
703 	pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv);
704 	pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
705 	pn->t_segqlen = tp->t_segqlen;
706 
707 	/* We've finished accessing the tcb so release the lock. */
708 	if (inp_locally_locked)
709 		INP_RUNLOCK(inp);
710 
711 	pn->direction = (dir == PFIL_IN ? DIR_IN : DIR_OUT);
712 
713 	/*
714 	 * Significantly more accurate than using getmicrotime(), but slower!
715 	 * Gives true microsecond resolution at the expense of a hit to
716 	 * maximum pps throughput processing when SIFTR is loaded and enabled.
717 	 */
718 	microtime(&pn->tval);
719 	TCP_PROBE1(siftr, pn);
720 }
721 
722 /*
723  * pfil hook that is called for each IPv4 packet making its way through the
724  * stack in either direction.
725  * The pfil subsystem holds a non-sleepable mutex somewhere when
726  * calling our hook function, so we can't sleep at all.
727  * It's very important to use the M_NOWAIT flag with all function calls
728  * that support it so that they won't sleep, otherwise you get a panic.
729  */
730 static pfil_return_t
731 siftr_chkpkt(struct mbuf **m, struct ifnet *ifp, int flags,
732     void *ruleset __unused, struct inpcb *inp)
733 {
734 	struct pkt_node *pn;
735 	struct ip *ip;
736 	struct tcphdr *th;
737 	struct tcpcb *tp;
738 	struct siftr_stats *ss;
739 	unsigned int ip_hl;
740 	int inp_locally_locked, dir;
741 	uint32_t hash_id, hash_type;
742 	struct listhead *counter_list;
743 	struct flow_hash_node *hash_node;
744 
745 	inp_locally_locked = 0;
746 	dir = PFIL_DIR(flags);
747 	ss = DPCPU_PTR(ss);
748 
749 	/*
750 	 * m_pullup is not required here because ip_{input|output}
751 	 * already do the heavy lifting for us.
752 	 */
753 
754 	ip = mtod(*m, struct ip *);
755 
756 	/* Only continue processing if the packet is TCP. */
757 	if (ip->ip_p != IPPROTO_TCP)
758 		goto ret;
759 
760 	/*
761 	 * Create a tcphdr struct starting at the correct offset
762 	 * in the IP packet. ip->ip_hl gives the ip header length
763 	 * in 4-byte words, so multiply it to get the size in bytes.
764 	 */
765 	ip_hl = (ip->ip_hl << 2);
766 	th = (struct tcphdr *)((caddr_t)ip + ip_hl);
767 
768 	/*
769 	 * Only pkts selected by the tcp port filter
770 	 * can be inserted into the pkt_queue
771 	 */
772 	if ((siftr_port_filter != 0) &&
773 	    (siftr_port_filter != ntohs(th->th_sport)) &&
774 	    (siftr_port_filter != ntohs(th->th_dport))) {
775 		goto ret;
776 	}
777 
778 	/*
779 	 * If a kernel subsystem reinjects packets into the stack, our pfil
780 	 * hook will be called multiple times for the same packet.
781 	 * Make sure we only process unique packets.
782 	 */
783 	if (siftr_chkreinject(*m, dir, ss))
784 		goto ret;
785 
786 	if (dir == PFIL_IN)
787 		ss->n_in++;
788 	else
789 		ss->n_out++;
790 
791 	/*
792 	 * If the pfil hooks don't provide a pointer to the
793 	 * inpcb, we need to find it ourselves and lock it.
794 	 */
795 	if (!inp) {
796 		/* Find the corresponding inpcb for this pkt. */
797 		inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
798 		    th->th_dport, dir, ss);
799 
800 		if (inp == NULL)
801 			goto ret;
802 		else
803 			inp_locally_locked = 1;
804 	}
805 
806 	INP_LOCK_ASSERT(inp);
807 
808 	/* Find the TCP control block that corresponds with this packet */
809 	tp = intotcpcb(inp);
810 
811 	/*
812 	 * If we can't find the TCP control block (happens occasionaly for a
813 	 * packet sent during the shutdown phase of a TCP connection), or the
814 	 * TCP control block has not initialized (happens during TCPS_SYN_SENT),
815 	 * bail.
816 	 */
817 	if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) {
818 		if (dir == PFIL_IN)
819 			ss->nskip_in_tcpcb++;
820 		else
821 			ss->nskip_out_tcpcb++;
822 
823 		goto inp_unlock;
824 	}
825 
826 	hash_id = siftr_get_flowid(inp, INP_IPV4, &hash_type);
827 	counter_list = counter_hash + (hash_id & siftr_hashmask);
828 	hash_node = siftr_find_flow(counter_list, hash_id);
829 
830 	/* If this flow hasn't been seen before, we create a new entry. */
831 	if (hash_node == NULL) {
832 		struct flow_info info;
833 
834 		inet_ntoa_r(inp->inp_laddr, info.laddr);
835 		inet_ntoa_r(inp->inp_faddr, info.faddr);
836 		info.lport = ntohs(inp->inp_lport);
837 		info.fport = ntohs(inp->inp_fport);
838 		info.key = hash_id;
839 
840 		hash_node = siftr_new_hash_node(info, dir, ss);
841 	}
842 
843 	if (hash_node == NULL) {
844 		goto inp_unlock;
845 	}
846 
847 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
848 
849 	if (pn == NULL) {
850 		if (dir == PFIL_IN)
851 			ss->nskip_in_malloc++;
852 		else
853 			ss->nskip_out_malloc++;
854 
855 		goto inp_unlock;
856 	}
857 
858 	pn->flowid = hash_id;
859 	pn->flowtype = hash_type;
860 
861 	siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
862 
863 	mtx_lock(&siftr_pkt_queue_mtx);
864 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
865 	mtx_unlock(&siftr_pkt_queue_mtx);
866 	goto ret;
867 
868 inp_unlock:
869 	if (inp_locally_locked)
870 		INP_RUNLOCK(inp);
871 
872 ret:
873 	return (PFIL_PASS);
874 }
875 
876 #ifdef SIFTR_IPV6
877 static pfil_return_t
878 siftr_chkpkt6(struct mbuf **m, struct ifnet *ifp, int flags,
879     void *ruleset __unused, struct inpcb *inp)
880 {
881 	struct pkt_node *pn;
882 	struct ip6_hdr *ip6;
883 	struct tcphdr *th;
884 	struct tcpcb *tp;
885 	struct siftr_stats *ss;
886 	unsigned int ip6_hl;
887 	int inp_locally_locked, dir;
888 	uint32_t hash_id, hash_type;
889 	struct listhead *counter_list;
890 	struct flow_hash_node *hash_node;
891 
892 	inp_locally_locked = 0;
893 	dir = PFIL_DIR(flags);
894 	ss = DPCPU_PTR(ss);
895 
896 	/*
897 	 * m_pullup is not required here because ip6_{input|output}
898 	 * already do the heavy lifting for us.
899 	 */
900 
901 	ip6 = mtod(*m, struct ip6_hdr *);
902 
903 	/*
904 	 * Only continue processing if the packet is TCP
905 	 * XXX: We should follow the next header fields
906 	 * as shown on Pg 6 RFC 2460, but right now we'll
907 	 * only check pkts that have no extension headers.
908 	 */
909 	if (ip6->ip6_nxt != IPPROTO_TCP)
910 		goto ret6;
911 
912 	/*
913 	 * Create a tcphdr struct starting at the correct offset
914 	 * in the ipv6 packet.
915 	 */
916 	ip6_hl = sizeof(struct ip6_hdr);
917 	th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
918 
919 	/*
920 	 * Only pkts selected by the tcp port filter
921 	 * can be inserted into the pkt_queue
922 	 */
923 	if ((siftr_port_filter != 0) &&
924 	    (siftr_port_filter != ntohs(th->th_sport)) &&
925 	    (siftr_port_filter != ntohs(th->th_dport))) {
926 		goto ret6;
927 	}
928 
929 	/*
930 	 * If a kernel subsystem reinjects packets into the stack, our pfil
931 	 * hook will be called multiple times for the same packet.
932 	 * Make sure we only process unique packets.
933 	 */
934 	if (siftr_chkreinject(*m, dir, ss))
935 		goto ret6;
936 
937 	if (dir == PFIL_IN)
938 		ss->n_in++;
939 	else
940 		ss->n_out++;
941 
942 	/*
943 	 * For inbound packets, the pfil hooks don't provide a pointer to the
944 	 * inpcb, so we need to find it ourselves and lock it.
945 	 */
946 	if (!inp) {
947 		/* Find the corresponding inpcb for this pkt. */
948 		inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
949 		    th->th_sport, th->th_dport, dir, ss);
950 
951 		if (inp == NULL)
952 			goto ret6;
953 		else
954 			inp_locally_locked = 1;
955 	}
956 
957 	/* Find the TCP control block that corresponds with this packet. */
958 	tp = intotcpcb(inp);
959 
960 	/*
961 	 * If we can't find the TCP control block (happens occasionaly for a
962 	 * packet sent during the shutdown phase of a TCP connection), or the
963 	 * TCP control block has not initialized (happens during TCPS_SYN_SENT),
964 	 * bail.
965 	 */
966 	if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) {
967 		if (dir == PFIL_IN)
968 			ss->nskip_in_tcpcb++;
969 		else
970 			ss->nskip_out_tcpcb++;
971 
972 		goto inp_unlock6;
973 	}
974 
975 	hash_id = siftr_get_flowid(inp, INP_IPV6, &hash_type);
976 	counter_list = counter_hash + (hash_id & siftr_hashmask);
977 	hash_node = siftr_find_flow(counter_list, hash_id);
978 
979 	/* If this flow hasn't been seen before, we create a new entry. */
980 	if (!hash_node) {
981 		struct flow_info info;
982 
983 		ip6_sprintf(info.laddr, &inp->in6p_laddr);
984 		ip6_sprintf(info.faddr, &inp->in6p_faddr);
985 		info.lport = ntohs(inp->inp_lport);
986 		info.fport = ntohs(inp->inp_fport);
987 		info.key = hash_id;
988 
989 		hash_node = siftr_new_hash_node(info, dir, ss);
990 	}
991 
992 	if (!hash_node) {
993 		goto inp_unlock6;
994 	}
995 
996 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
997 
998 	if (pn == NULL) {
999 		if (dir == PFIL_IN)
1000 			ss->nskip_in_malloc++;
1001 		else
1002 			ss->nskip_out_malloc++;
1003 
1004 		goto inp_unlock6;
1005 	}
1006 
1007 	pn->flowid = hash_id;
1008 	pn->flowtype = hash_type;
1009 
1010 	siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
1011 
1012 	mtx_lock(&siftr_pkt_queue_mtx);
1013 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
1014 	mtx_unlock(&siftr_pkt_queue_mtx);
1015 	goto ret6;
1016 
1017 inp_unlock6:
1018 	if (inp_locally_locked)
1019 		INP_RUNLOCK(inp);
1020 
1021 ret6:
1022 	return (PFIL_PASS);
1023 }
1024 #endif /* #ifdef SIFTR_IPV6 */
1025 
1026 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet_hook);
1027 #define	V_siftr_inet_hook	VNET(siftr_inet_hook)
1028 #ifdef SIFTR_IPV6
1029 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet6_hook);
1030 #define	V_siftr_inet6_hook	VNET(siftr_inet6_hook)
1031 #endif
1032 static int
1033 siftr_pfil(int action)
1034 {
1035 	struct pfil_hook_args pha = {
1036 		.pa_version = PFIL_VERSION,
1037 		.pa_flags = PFIL_IN | PFIL_OUT,
1038 		.pa_modname = "siftr",
1039 		.pa_rulname = "default",
1040 	};
1041 	struct pfil_link_args pla = {
1042 		.pa_version = PFIL_VERSION,
1043 		.pa_flags = PFIL_IN | PFIL_OUT | PFIL_HEADPTR | PFIL_HOOKPTR,
1044 	};
1045 
1046 	VNET_ITERATOR_DECL(vnet_iter);
1047 
1048 	VNET_LIST_RLOCK();
1049 	VNET_FOREACH(vnet_iter) {
1050 		CURVNET_SET(vnet_iter);
1051 
1052 		if (action == HOOK) {
1053 			pha.pa_mbuf_chk = siftr_chkpkt;
1054 			pha.pa_type = PFIL_TYPE_IP4;
1055 			V_siftr_inet_hook = pfil_add_hook(&pha);
1056 			pla.pa_hook = V_siftr_inet_hook;
1057 			pla.pa_head = V_inet_pfil_head;
1058 			(void)pfil_link(&pla);
1059 #ifdef SIFTR_IPV6
1060 			pha.pa_mbuf_chk = siftr_chkpkt6;
1061 			pha.pa_type = PFIL_TYPE_IP6;
1062 			V_siftr_inet6_hook = pfil_add_hook(&pha);
1063 			pla.pa_hook = V_siftr_inet6_hook;
1064 			pla.pa_head = V_inet6_pfil_head;
1065 			(void)pfil_link(&pla);
1066 #endif
1067 		} else if (action == UNHOOK) {
1068 			pfil_remove_hook(V_siftr_inet_hook);
1069 #ifdef SIFTR_IPV6
1070 			pfil_remove_hook(V_siftr_inet6_hook);
1071 #endif
1072 		}
1073 		CURVNET_RESTORE();
1074 	}
1075 	VNET_LIST_RUNLOCK();
1076 
1077 	return (0);
1078 }
1079 
1080 static int
1081 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1082 {
1083 	struct alq *new_alq;
1084 	int error;
1085 
1086 	error = sysctl_handle_string(oidp, arg1, arg2, req);
1087 
1088 	/* Check for error or same filename */
1089 	if (error != 0 || req->newptr == NULL ||
1090 	    strncmp(siftr_logfile, arg1, arg2) == 0)
1091 		goto done;
1092 
1093 	/* file name changed */
1094 	error = alq_open(&new_alq, arg1, curthread->td_ucred,
1095 	    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1096 	if (error != 0)
1097 		goto done;
1098 
1099 	/*
1100 	 * If disabled, siftr_alq == NULL so we simply close
1101 	 * the alq as we've proved it can be opened.
1102 	 * If enabled, close the existing alq and switch the old
1103 	 * for the new.
1104 	 */
1105 	if (siftr_alq == NULL) {
1106 		alq_close(new_alq);
1107 	} else {
1108 		alq_close(siftr_alq);
1109 		siftr_alq = new_alq;
1110 	}
1111 
1112 	/* Update filename upon success */
1113 	strlcpy(siftr_logfile, arg1, arg2);
1114 done:
1115 	return (error);
1116 }
1117 
1118 static int
1119 siftr_manage_ops(uint8_t action)
1120 {
1121 	struct siftr_stats totalss;
1122 	struct timeval tval;
1123 	struct flow_hash_node *counter, *tmp_counter;
1124 	struct sbuf *s;
1125 	int i, error;
1126 	uint32_t bytes_to_write, total_skipped_pkts;
1127 
1128 	error = 0;
1129 	total_skipped_pkts = 0;
1130 
1131 	/* Init an autosizing sbuf that initially holds 200 chars. */
1132 	if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1133 		return (-1);
1134 
1135 	if (action == SIFTR_ENABLE && siftr_pkt_manager_thr == NULL) {
1136 		/*
1137 		 * Create our alq
1138 		 * XXX: We should abort if alq_open fails!
1139 		 */
1140 		alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1141 		    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1142 
1143 		STAILQ_INIT(&pkt_queue);
1144 
1145 		DPCPU_ZERO(ss);
1146 
1147 		siftr_exit_pkt_manager_thread = 0;
1148 
1149 		kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1150 		    &siftr_pkt_manager_thr, RFNOWAIT, 0,
1151 		    "siftr_pkt_manager_thr");
1152 
1153 		siftr_pfil(HOOK);
1154 
1155 		microtime(&tval);
1156 
1157 		sbuf_printf(s,
1158 		    "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1159 		    "siftrver=%s\tsysname=%s\tsysver=%u\tipmode=%u\n",
1160 		    (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR,
1161 		    SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1162 
1163 		sbuf_finish(s);
1164 		alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1165 
1166 	} else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1167 		/*
1168 		 * Remove the pfil hook functions. All threads currently in
1169 		 * the hook functions are allowed to exit before siftr_pfil()
1170 		 * returns.
1171 		 */
1172 		siftr_pfil(UNHOOK);
1173 
1174 		/* This will block until the pkt manager thread unlocks it. */
1175 		mtx_lock(&siftr_pkt_mgr_mtx);
1176 
1177 		/* Tell the pkt manager thread that it should exit now. */
1178 		siftr_exit_pkt_manager_thread = 1;
1179 
1180 		/*
1181 		 * Wake the pkt_manager thread so it realises that
1182 		 * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1183 		 * The wakeup won't be delivered until we unlock
1184 		 * siftr_pkt_mgr_mtx so this isn't racy.
1185 		 */
1186 		wakeup(&wait_for_pkt);
1187 
1188 		/* Wait for the pkt_manager thread to exit. */
1189 		mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1190 		    "thrwait", 0);
1191 
1192 		siftr_pkt_manager_thr = NULL;
1193 		mtx_unlock(&siftr_pkt_mgr_mtx);
1194 
1195 		totalss.n_in = DPCPU_VARSUM(ss, n_in);
1196 		totalss.n_out = DPCPU_VARSUM(ss, n_out);
1197 		totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1198 		totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1199 		totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1200 		totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1201 		totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1202 		totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1203 
1204 		total_skipped_pkts = totalss.nskip_in_malloc +
1205 		    totalss.nskip_out_malloc + totalss.nskip_in_tcpcb +
1206 		    totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1207 		    totalss.nskip_out_inpcb;
1208 
1209 		microtime(&tval);
1210 
1211 		sbuf_printf(s,
1212 		    "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1213 		    "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1214 		    "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1215 		    "num_outbound_skipped_pkts_malloc=%u\t"
1216 		    "num_inbound_skipped_pkts_tcpcb=%u\t"
1217 		    "num_outbound_skipped_pkts_tcpcb=%u\t"
1218 		    "num_inbound_skipped_pkts_inpcb=%u\t"
1219 		    "num_outbound_skipped_pkts_inpcb=%u\t"
1220 		    "total_skipped_tcp_pkts=%u\tflow_list=",
1221 		    (intmax_t)tval.tv_sec,
1222 		    tval.tv_usec,
1223 		    (uintmax_t)totalss.n_in,
1224 		    (uintmax_t)totalss.n_out,
1225 		    (uintmax_t)(totalss.n_in + totalss.n_out),
1226 		    totalss.nskip_in_malloc,
1227 		    totalss.nskip_out_malloc,
1228 		    totalss.nskip_in_tcpcb,
1229 		    totalss.nskip_out_tcpcb,
1230 		    totalss.nskip_in_inpcb,
1231 		    totalss.nskip_out_inpcb,
1232 		    total_skipped_pkts);
1233 
1234 		/*
1235 		 * Iterate over the flow hash, printing a summary of each
1236 		 * flow seen and freeing any malloc'd memory.
1237 		 * The hash consists of an array of LISTs (man 3 queue).
1238 		 */
1239 		for (i = 0; i <= siftr_hashmask; i++) {
1240 			LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1241 			    tmp_counter) {
1242 				sbuf_printf(s, "%s;%hu-%s;%hu,",
1243 					    counter->const_info.laddr,
1244 					    counter->const_info.lport,
1245 					    counter->const_info.faddr,
1246 					    counter->const_info.fport);
1247 
1248 				free(counter, M_SIFTR_HASHNODE);
1249 			}
1250 
1251 			LIST_INIT(counter_hash + i);
1252 		}
1253 
1254 		sbuf_printf(s, "\n");
1255 		sbuf_finish(s);
1256 
1257 		i = 0;
1258 		do {
1259 			bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1260 			alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1261 			i += bytes_to_write;
1262 		} while (i < sbuf_len(s));
1263 
1264 		alq_close(siftr_alq);
1265 		siftr_alq = NULL;
1266 	} else
1267 		error = EINVAL;
1268 
1269 	sbuf_delete(s);
1270 
1271 	/*
1272 	 * XXX: Should be using ret to check if any functions fail
1273 	 * and set error appropriately
1274 	 */
1275 
1276 	return (error);
1277 }
1278 
1279 static int
1280 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1281 {
1282 	int error;
1283 	uint32_t new;
1284 
1285 	new = siftr_enabled;
1286 	error = sysctl_handle_int(oidp, &new, 0, req);
1287 	if (error == 0 && req->newptr != NULL) {
1288 		if (new > 1)
1289 			return (EINVAL);
1290 		else if (new != siftr_enabled) {
1291 			if ((error = siftr_manage_ops(new)) == 0) {
1292 				siftr_enabled = new;
1293 			} else {
1294 				siftr_manage_ops(SIFTR_DISABLE);
1295 			}
1296 		}
1297 	}
1298 
1299 	return (error);
1300 }
1301 
1302 static void
1303 siftr_shutdown_handler(void *arg, int howto)
1304 {
1305 	if ((howto & RB_NOSYNC) != 0 || SCHEDULER_STOPPED())
1306 		return;
1307 
1308 	if (siftr_enabled == 1) {
1309 		siftr_manage_ops(SIFTR_DISABLE);
1310 	}
1311 }
1312 
1313 /*
1314  * Module is being unloaded or machine is shutting down. Take care of cleanup.
1315  */
1316 static int
1317 deinit_siftr(void)
1318 {
1319 	/* Cleanup. */
1320 	siftr_manage_ops(SIFTR_DISABLE);
1321 	hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1322 	mtx_destroy(&siftr_pkt_queue_mtx);
1323 	mtx_destroy(&siftr_pkt_mgr_mtx);
1324 
1325 	return (0);
1326 }
1327 
1328 /*
1329  * Module has just been loaded into the kernel.
1330  */
1331 static int
1332 init_siftr(void)
1333 {
1334 	EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL,
1335 	    SHUTDOWN_PRI_FIRST);
1336 
1337 	/* Initialise our flow counter hash table. */
1338 	counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1339 	    &siftr_hashmask);
1340 
1341 	mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1342 	mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1343 
1344 	/* Print message to the user's current terminal. */
1345 	uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1346 	    "          http://caia.swin.edu.au/urp/newtcp\n\n",
1347 	    MODVERSION_STR);
1348 
1349 	return (0);
1350 }
1351 
1352 /*
1353  * This is the function that is called to load and unload the module.
1354  * When the module is loaded, this function is called once with
1355  * "what" == MOD_LOAD
1356  * When the module is unloaded, this function is called twice with
1357  * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1358  * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1359  * this function is called once with "what" = MOD_SHUTDOWN
1360  * When the system is shut down, the handler isn't called until the very end
1361  * of the shutdown sequence i.e. after the disks have been synced.
1362  */
1363 static int
1364 siftr_load_handler(module_t mod, int what, void *arg)
1365 {
1366 	int ret;
1367 
1368 	switch (what) {
1369 	case MOD_LOAD:
1370 		ret = init_siftr();
1371 		break;
1372 
1373 	case MOD_QUIESCE:
1374 	case MOD_SHUTDOWN:
1375 		ret = deinit_siftr();
1376 		break;
1377 
1378 	case MOD_UNLOAD:
1379 		ret = 0;
1380 		break;
1381 
1382 	default:
1383 		ret = EINVAL;
1384 		break;
1385 	}
1386 
1387 	return (ret);
1388 }
1389 
1390 static moduledata_t siftr_mod = {
1391 	.name = "siftr",
1392 	.evhand = siftr_load_handler,
1393 };
1394 
1395 /*
1396  * Param 1: name of the kernel module
1397  * Param 2: moduledata_t struct containing info about the kernel module
1398  *          and the execution entry point for the module
1399  * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1400  *          Defines the module initialisation order
1401  * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1402  *          Defines the initialisation order of this kld relative to others
1403  *          within the same subsystem as defined by param 3
1404  */
1405 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_LAST, SI_ORDER_ANY);
1406 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1407 MODULE_VERSION(siftr, MODVERSION);
1408