xref: /freebsd/sys/netinet/siftr.c (revision 81f08f303805bcfd0a32fb34835956740c3c9a81)
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 #define MAX_LOG_BATCH_SIZE 3
138 /* XXX: Make this a sysctl tunable. */
139 #define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN)
140 
141 #ifdef SIFTR_IPV6
142 #define SIFTR_IPMODE 6
143 #else
144 #define SIFTR_IPMODE 4
145 #endif
146 
147 static MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR");
148 static MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode",
149     "SIFTR pkt_node struct");
150 static MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode",
151     "SIFTR flow_hash_node struct");
152 
153 /* Used as links in the pkt manager queue. */
154 struct pkt_node {
155 	/* Timestamp of pkt as noted in the pfil hook. */
156 	struct timeval		tval;
157 	/* Direction pkt is travelling. */
158 	enum {
159 		DIR_IN = 0,
160 		DIR_OUT = 1,
161 	}			direction;
162 	/* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */
163 	uint8_t			ipver;
164 	/* Local TCP port. */
165 	uint16_t		lport;
166 	/* Foreign TCP port. */
167 	uint16_t		fport;
168 	/* Local address. */
169 	union in_dependaddr	laddr;
170 	/* Foreign address. */
171 	union in_dependaddr	faddr;
172 	/* Congestion Window (bytes). */
173 	uint32_t		snd_cwnd;
174 	/* Sending Window (bytes). */
175 	uint32_t		snd_wnd;
176 	/* Receive Window (bytes). */
177 	uint32_t		rcv_wnd;
178 	/* More tcpcb flags storage */
179 	uint32_t		t_flags2;
180 	/* Slow Start Threshold (bytes). */
181 	uint32_t		snd_ssthresh;
182 	/* Current state of the TCP FSM. */
183 	int			conn_state;
184 	/* Max Segment Size (bytes). */
185 	uint32_t		mss;
186 	/* Smoothed RTT (usecs). */
187 	uint32_t		srtt;
188 	/* Is SACK enabled? */
189 	u_char			sack_enabled;
190 	/* Window scaling for snd window. */
191 	u_char			snd_scale;
192 	/* Window scaling for recv window. */
193 	u_char			rcv_scale;
194 	/* TCP control block flags. */
195 	u_int			t_flags;
196 	/* Retransmission timeout (usec). */
197 	uint32_t		rto;
198 	/* Size of the TCP send buffer in bytes. */
199 	u_int			snd_buf_hiwater;
200 	/* Current num bytes in the send socket buffer. */
201 	u_int			snd_buf_cc;
202 	/* Size of the TCP receive buffer in bytes. */
203 	u_int			rcv_buf_hiwater;
204 	/* Current num bytes in the receive socket buffer. */
205 	u_int			rcv_buf_cc;
206 	/* Number of bytes inflight that we are waiting on ACKs for. */
207 	u_int			sent_inflight_bytes;
208 	/* Number of segments currently in the reassembly queue. */
209 	int			t_segqlen;
210 	/* Flowid for the connection. */
211 	u_int			flowid;
212 	/* Flow type for the connection. */
213 	u_int			flowtype;
214 	/* Link to next pkt_node in the list. */
215 	STAILQ_ENTRY(pkt_node)	nodes;
216 };
217 
218 struct flow_info
219 {
220 #ifdef SIFTR_IPV6
221 	char	laddr[INET6_ADDRSTRLEN];	/* local IP address */
222 	char	faddr[INET6_ADDRSTRLEN];	/* foreign IP address */
223 #else
224 	char	laddr[INET_ADDRSTRLEN];		/* local IP address */
225 	char	faddr[INET_ADDRSTRLEN];		/* foreign IP address */
226 #endif
227 	uint16_t	lport;			/* local TCP port */
228 	uint16_t	fport;			/* foreign TCP port */
229 	uint32_t	key;			/* flowid of the connection */
230 };
231 
232 struct flow_hash_node
233 {
234 	uint16_t counter;
235 	struct flow_info const_info;		/* constant connection info */
236 	LIST_ENTRY(flow_hash_node) nodes;
237 };
238 
239 struct siftr_stats
240 {
241 	/* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */
242 	uint64_t n_in;
243 	uint64_t n_out;
244 	/* # pkts skipped due to failed malloc calls. */
245 	uint32_t nskip_in_malloc;
246 	uint32_t nskip_out_malloc;
247 	/* # pkts skipped due to failed inpcb lookups. */
248 	uint32_t nskip_in_inpcb;
249 	uint32_t nskip_out_inpcb;
250 	/* # pkts skipped due to failed tcpcb lookups. */
251 	uint32_t nskip_in_tcpcb;
252 	uint32_t nskip_out_tcpcb;
253 	/* # pkts skipped due to stack reinjection. */
254 	uint32_t nskip_in_dejavu;
255 	uint32_t nskip_out_dejavu;
256 };
257 
258 DPCPU_DEFINE_STATIC(struct siftr_stats, ss);
259 
260 static volatile unsigned int siftr_exit_pkt_manager_thread = 0;
261 static unsigned int siftr_enabled = 0;
262 static unsigned int siftr_pkts_per_log = 1;
263 static uint16_t     siftr_port_filter = 0;
264 /* static unsigned int siftr_binary_log = 0; */
265 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log";
266 static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log";
267 static u_long siftr_hashmask;
268 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue);
269 LIST_HEAD(listhead, flow_hash_node) *counter_hash;
270 static int wait_for_pkt;
271 static struct alq *siftr_alq = NULL;
272 static struct mtx siftr_pkt_queue_mtx;
273 static struct mtx siftr_pkt_mgr_mtx;
274 static struct thread *siftr_pkt_manager_thr = NULL;
275 static char direction[2] = {'i','o'};
276 static eventhandler_tag siftr_shutdown_tag;
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 *
siftr_find_flow(struct listhead * counter_list,uint32_t id)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 *
siftr_new_hash_node(struct flow_info info,int dir,struct siftr_stats * ss)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 int
siftr_process_pkt(struct pkt_node * pkt_node,char * buf)376 siftr_process_pkt(struct pkt_node * pkt_node, char *buf)
377 {
378 	struct flow_hash_node *hash_node;
379 	struct listhead *counter_list;
380 	int ret_sz;
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 0;
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 0;
407 	}
408 
409 	/* Construct a log message. */
410 	ret_sz = snprintf(buf, MAX_LOG_MSG_LEN,
411 	    "%c,%jd.%06ld,%s,%hu,%s,%hu,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,%u,"
412 	    "%u,%u,%u,%u,%u,%u,%u,%u\n",
413 	    direction[pkt_node->direction],
414 	    (intmax_t)pkt_node->tval.tv_sec,
415 	    pkt_node->tval.tv_usec,
416 	    hash_node->const_info.laddr,
417 	    hash_node->const_info.lport,
418 	    hash_node->const_info.faddr,
419 	    hash_node->const_info.fport,
420 	    pkt_node->snd_ssthresh,
421 	    pkt_node->snd_cwnd,
422 	    pkt_node->t_flags2,
423 	    pkt_node->snd_wnd,
424 	    pkt_node->rcv_wnd,
425 	    pkt_node->snd_scale,
426 	    pkt_node->rcv_scale,
427 	    pkt_node->conn_state,
428 	    pkt_node->mss,
429 	    pkt_node->srtt,
430 	    pkt_node->sack_enabled,
431 	    pkt_node->t_flags,
432 	    pkt_node->rto,
433 	    pkt_node->snd_buf_hiwater,
434 	    pkt_node->snd_buf_cc,
435 	    pkt_node->rcv_buf_hiwater,
436 	    pkt_node->rcv_buf_cc,
437 	    pkt_node->sent_inflight_bytes,
438 	    pkt_node->t_segqlen,
439 	    pkt_node->flowid,
440 	    pkt_node->flowtype);
441 
442 	return ret_sz;
443 }
444 
445 static void
siftr_pkt_manager_thread(void * arg)446 siftr_pkt_manager_thread(void *arg)
447 {
448 	STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
449 	    STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
450 	struct pkt_node *pkt_node;
451 	uint8_t draining;
452 	struct ale *log_buf;
453 	int ret_sz, cnt = 0;
454 	char *bufp;
455 
456 	draining = 2;
457 
458 	mtx_lock(&siftr_pkt_mgr_mtx);
459 
460 	/* draining == 0 when queue has been flushed and it's safe to exit. */
461 	while (draining) {
462 		/*
463 		 * Sleep until we are signalled to wake because thread has
464 		 * been told to exit or until 1 tick has passed.
465 		 */
466 		mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
467 		    1);
468 
469 		/* Gain exclusive access to the pkt_node queue. */
470 		mtx_lock(&siftr_pkt_queue_mtx);
471 
472 		/*
473 		 * Move pkt_queue to tmp_pkt_queue, which leaves
474 		 * pkt_queue empty and ready to receive more pkt_nodes.
475 		 */
476 		STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
477 
478 		/*
479 		 * We've finished making changes to the list. Unlock it
480 		 * so the pfil hooks can continue queuing pkt_nodes.
481 		 */
482 		mtx_unlock(&siftr_pkt_queue_mtx);
483 
484 		/*
485 		 * We can't hold a mutex whilst calling siftr_process_pkt
486 		 * because ALQ might sleep waiting for buffer space.
487 		 */
488 		mtx_unlock(&siftr_pkt_mgr_mtx);
489 
490 		while ((pkt_node = STAILQ_FIRST(&tmp_pkt_queue)) != NULL) {
491 
492 			log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN *
493 			    ((STAILQ_NEXT(pkt_node, nodes) != NULL) ?
494 				MAX_LOG_BATCH_SIZE : 1),
495 			    ALQ_WAITOK);
496 
497 			if (log_buf != NULL) {
498 				log_buf->ae_bytesused = 0;
499 				bufp = log_buf->ae_data;
500 			} else {
501 				/*
502 				 * Should only happen if the ALQ is shutting
503 				 * down.
504 				 */
505 				bufp = NULL;
506 			}
507 
508 			/* Flush all pkt_nodes to the log file. */
509 			STAILQ_FOREACH(pkt_node, &tmp_pkt_queue, nodes) {
510 				if (log_buf != NULL) {
511 					ret_sz = siftr_process_pkt(pkt_node,
512 								   bufp);
513 					bufp += ret_sz;
514 					log_buf->ae_bytesused += ret_sz;
515 				}
516 				if (++cnt >= MAX_LOG_BATCH_SIZE)
517 					break;
518 			}
519 			if (log_buf != NULL) {
520 				alq_post_flags(siftr_alq, log_buf, 0);
521 			}
522 			for (;cnt > 0; cnt--) {
523 				pkt_node = STAILQ_FIRST(&tmp_pkt_queue);
524 				STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
525 				free(pkt_node, M_SIFTR_PKTNODE);
526 			}
527 		}
528 
529 		KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
530 		    ("SIFTR tmp_pkt_queue not empty after flush"));
531 
532 		mtx_lock(&siftr_pkt_mgr_mtx);
533 
534 		/*
535 		 * If siftr_exit_pkt_manager_thread gets set during the window
536 		 * where we are draining the tmp_pkt_queue above, there might
537 		 * still be pkts in pkt_queue that need to be drained.
538 		 * Allow one further iteration to occur after
539 		 * siftr_exit_pkt_manager_thread has been set to ensure
540 		 * pkt_queue is completely empty before we kill the thread.
541 		 *
542 		 * siftr_exit_pkt_manager_thread is set only after the pfil
543 		 * hooks have been removed, so only 1 extra iteration
544 		 * is needed to drain the queue.
545 		 */
546 		if (siftr_exit_pkt_manager_thread)
547 			draining--;
548 	}
549 
550 	mtx_unlock(&siftr_pkt_mgr_mtx);
551 
552 	/* Calls wakeup on this thread's struct thread ptr. */
553 	kthread_exit();
554 }
555 
556 /*
557  * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
558  * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
559  * Return value >0 means the caller should skip processing this mbuf.
560  */
561 static inline int
siftr_chkreinject(struct mbuf * m,int dir,struct siftr_stats * ss)562 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
563 {
564 	if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
565 	    != NULL) {
566 		if (dir == PFIL_IN)
567 			ss->nskip_in_dejavu++;
568 		else
569 			ss->nskip_out_dejavu++;
570 
571 		return (1);
572 	} else {
573 		struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
574 		    PACKET_TAG_SIFTR, 0, M_NOWAIT);
575 		if (tag == NULL) {
576 			if (dir == PFIL_IN)
577 				ss->nskip_in_malloc++;
578 			else
579 				ss->nskip_out_malloc++;
580 
581 			return (1);
582 		}
583 
584 		m_tag_prepend(m, tag);
585 	}
586 
587 	return (0);
588 }
589 
590 /*
591  * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
592  * otherwise.
593  */
594 static inline struct inpcb *
siftr_findinpcb(int ipver,struct ip * ip,struct mbuf * m,uint16_t sport,uint16_t dport,int dir,struct siftr_stats * ss)595 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
596     uint16_t dport, int dir, struct siftr_stats *ss)
597 {
598 	struct inpcb *inp;
599 
600 	if (dir == PFIL_IN)
601 		inp = (ipver == INP_IPV4 ?
602 		    in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
603 		    dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
604 		    :
605 #ifdef SIFTR_IPV6
606 		    in6_pcblookup(&V_tcbinfo,
607 		    &((struct ip6_hdr *)ip)->ip6_src, sport,
608 		    &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB,
609 		    m->m_pkthdr.rcvif)
610 #else
611 		    NULL
612 #endif
613 		    );
614 
615 	else
616 		inp = (ipver == INP_IPV4 ?
617 		    in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
618 		    sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
619 		    :
620 #ifdef SIFTR_IPV6
621 		    in6_pcblookup(&V_tcbinfo,
622 		    &((struct ip6_hdr *)ip)->ip6_dst, dport,
623 		    &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB,
624 		    m->m_pkthdr.rcvif)
625 #else
626 		    NULL
627 #endif
628 		    );
629 
630 	/* If we can't find the inpcb, bail. */
631 	if (inp == NULL) {
632 		if (dir == PFIL_IN)
633 			ss->nskip_in_inpcb++;
634 		else
635 			ss->nskip_out_inpcb++;
636 	}
637 
638 	return (inp);
639 }
640 
641 static inline uint32_t
siftr_get_flowid(struct inpcb * inp,int ipver,uint32_t * phashtype)642 siftr_get_flowid(struct inpcb *inp, int ipver, uint32_t *phashtype)
643 {
644 	if (inp->inp_flowid == 0) {
645 #ifdef SIFTR_IPV6
646 		if (ipver == INP_IPV6) {
647 			return fib6_calc_packet_hash(&inp->in6p_laddr,
648 						     &inp->in6p_faddr,
649 						     inp->inp_lport,
650 						     inp->inp_fport,
651 						     IPPROTO_TCP,
652 						     phashtype);
653 		} else
654 #endif
655 		{
656 			return fib4_calc_packet_hash(inp->inp_laddr,
657 						     inp->inp_faddr,
658 						     inp->inp_lport,
659 						     inp->inp_fport,
660 						     IPPROTO_TCP,
661 						     phashtype);
662 		}
663 	} else {
664 		*phashtype = inp->inp_flowtype;
665 		return inp->inp_flowid;
666 	}
667 }
668 
669 static inline void
siftr_siftdata(struct pkt_node * pn,struct inpcb * inp,struct tcpcb * tp,int ipver,int dir,int inp_locally_locked)670 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
671     int ipver, int dir, int inp_locally_locked)
672 {
673 	pn->ipver = ipver;
674 	pn->lport = inp->inp_lport;
675 	pn->fport = inp->inp_fport;
676 	pn->laddr = inp->inp_inc.inc_ie.ie_dependladdr;
677 	pn->faddr = inp->inp_inc.inc_ie.ie_dependfaddr;
678 	pn->snd_cwnd = tp->snd_cwnd;
679 	pn->snd_wnd = tp->snd_wnd;
680 	pn->rcv_wnd = tp->rcv_wnd;
681 	pn->t_flags2 = tp->t_flags2;
682 	pn->snd_ssthresh = tp->snd_ssthresh;
683 	pn->snd_scale = tp->snd_scale;
684 	pn->rcv_scale = tp->rcv_scale;
685 	pn->conn_state = tp->t_state;
686 	pn->mss = tp->t_maxseg;
687 	pn->srtt = ((uint64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
688 	pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
689 	pn->t_flags = tp->t_flags;
690 	pn->rto = tp->t_rxtcur * tick;
691 	pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
692 	pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd);
693 	pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
694 	pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv);
695 	pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
696 	pn->t_segqlen = tp->t_segqlen;
697 
698 	/* We've finished accessing the tcb so release the lock. */
699 	if (inp_locally_locked)
700 		INP_RUNLOCK(inp);
701 
702 	pn->direction = (dir == PFIL_IN ? DIR_IN : DIR_OUT);
703 
704 	/*
705 	 * Significantly more accurate than using getmicrotime(), but slower!
706 	 * Gives true microsecond resolution at the expense of a hit to
707 	 * maximum pps throughput processing when SIFTR is loaded and enabled.
708 	 */
709 	microtime(&pn->tval);
710 	TCP_PROBE1(siftr, pn);
711 }
712 
713 /*
714  * pfil hook that is called for each IPv4 packet making its way through the
715  * stack in either direction.
716  * The pfil subsystem holds a non-sleepable mutex somewhere when
717  * calling our hook function, so we can't sleep at all.
718  * It's very important to use the M_NOWAIT flag with all function calls
719  * that support it so that they won't sleep, otherwise you get a panic.
720  */
721 static pfil_return_t
siftr_chkpkt(struct mbuf ** m,struct ifnet * ifp,int flags,void * ruleset __unused,struct inpcb * inp)722 siftr_chkpkt(struct mbuf **m, struct ifnet *ifp, int flags,
723     void *ruleset __unused, struct inpcb *inp)
724 {
725 	struct pkt_node *pn;
726 	struct ip *ip;
727 	struct tcphdr *th;
728 	struct tcpcb *tp;
729 	struct siftr_stats *ss;
730 	unsigned int ip_hl;
731 	int inp_locally_locked, dir;
732 	uint32_t hash_id, hash_type;
733 	struct listhead *counter_list;
734 	struct flow_hash_node *hash_node;
735 
736 	inp_locally_locked = 0;
737 	dir = PFIL_DIR(flags);
738 	ss = DPCPU_PTR(ss);
739 
740 	/*
741 	 * m_pullup is not required here because ip_{input|output}
742 	 * already do the heavy lifting for us.
743 	 */
744 
745 	ip = mtod(*m, struct ip *);
746 
747 	/* Only continue processing if the packet is TCP. */
748 	if (ip->ip_p != IPPROTO_TCP)
749 		goto ret;
750 
751 	/*
752 	 * Create a tcphdr struct starting at the correct offset
753 	 * in the IP packet. ip->ip_hl gives the ip header length
754 	 * in 4-byte words, so multiply it to get the size in bytes.
755 	 */
756 	ip_hl = (ip->ip_hl << 2);
757 	th = (struct tcphdr *)((caddr_t)ip + ip_hl);
758 
759 	/*
760 	 * Only pkts selected by the tcp port filter
761 	 * can be inserted into the pkt_queue
762 	 */
763 	if ((siftr_port_filter != 0) &&
764 	    (siftr_port_filter != ntohs(th->th_sport)) &&
765 	    (siftr_port_filter != ntohs(th->th_dport))) {
766 		goto ret;
767 	}
768 
769 	/*
770 	 * If a kernel subsystem reinjects packets into the stack, our pfil
771 	 * hook will be called multiple times for the same packet.
772 	 * Make sure we only process unique packets.
773 	 */
774 	if (siftr_chkreinject(*m, dir, ss))
775 		goto ret;
776 
777 	if (dir == PFIL_IN)
778 		ss->n_in++;
779 	else
780 		ss->n_out++;
781 
782 	/*
783 	 * If the pfil hooks don't provide a pointer to the
784 	 * inpcb, we need to find it ourselves and lock it.
785 	 */
786 	if (!inp) {
787 		/* Find the corresponding inpcb for this pkt. */
788 		inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
789 		    th->th_dport, dir, ss);
790 
791 		if (inp == NULL)
792 			goto ret;
793 		else
794 			inp_locally_locked = 1;
795 	}
796 
797 	INP_LOCK_ASSERT(inp);
798 
799 	/* Find the TCP control block that corresponds with this packet */
800 	tp = intotcpcb(inp);
801 
802 	/*
803 	 * If we can't find the TCP control block (happens occasionaly for a
804 	 * packet sent during the shutdown phase of a TCP connection), or the
805 	 * TCP control block has not initialized (happens during TCPS_SYN_SENT),
806 	 * bail.
807 	 */
808 	if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) {
809 		if (dir == PFIL_IN)
810 			ss->nskip_in_tcpcb++;
811 		else
812 			ss->nskip_out_tcpcb++;
813 
814 		goto inp_unlock;
815 	}
816 
817 	hash_id = siftr_get_flowid(inp, INP_IPV4, &hash_type);
818 	counter_list = counter_hash + (hash_id & siftr_hashmask);
819 	hash_node = siftr_find_flow(counter_list, hash_id);
820 
821 	/* If this flow hasn't been seen before, we create a new entry. */
822 	if (hash_node == NULL) {
823 		struct flow_info info;
824 
825 		inet_ntoa_r(inp->inp_laddr, info.laddr);
826 		inet_ntoa_r(inp->inp_faddr, info.faddr);
827 		info.lport = ntohs(inp->inp_lport);
828 		info.fport = ntohs(inp->inp_fport);
829 		info.key = hash_id;
830 
831 		hash_node = siftr_new_hash_node(info, dir, ss);
832 	}
833 
834 	if (hash_node == NULL) {
835 		goto inp_unlock;
836 	}
837 
838 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
839 
840 	if (pn == NULL) {
841 		if (dir == PFIL_IN)
842 			ss->nskip_in_malloc++;
843 		else
844 			ss->nskip_out_malloc++;
845 
846 		goto inp_unlock;
847 	}
848 
849 	pn->flowid = hash_id;
850 	pn->flowtype = hash_type;
851 
852 	siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
853 
854 	mtx_lock(&siftr_pkt_queue_mtx);
855 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
856 	mtx_unlock(&siftr_pkt_queue_mtx);
857 	goto ret;
858 
859 inp_unlock:
860 	if (inp_locally_locked)
861 		INP_RUNLOCK(inp);
862 
863 ret:
864 	return (PFIL_PASS);
865 }
866 
867 #ifdef SIFTR_IPV6
868 static pfil_return_t
siftr_chkpkt6(struct mbuf ** m,struct ifnet * ifp,int flags,void * ruleset __unused,struct inpcb * inp)869 siftr_chkpkt6(struct mbuf **m, struct ifnet *ifp, int flags,
870     void *ruleset __unused, struct inpcb *inp)
871 {
872 	struct pkt_node *pn;
873 	struct ip6_hdr *ip6;
874 	struct tcphdr *th;
875 	struct tcpcb *tp;
876 	struct siftr_stats *ss;
877 	unsigned int ip6_hl;
878 	int inp_locally_locked, dir;
879 	uint32_t hash_id, hash_type;
880 	struct listhead *counter_list;
881 	struct flow_hash_node *hash_node;
882 
883 	inp_locally_locked = 0;
884 	dir = PFIL_DIR(flags);
885 	ss = DPCPU_PTR(ss);
886 
887 	/*
888 	 * m_pullup is not required here because ip6_{input|output}
889 	 * already do the heavy lifting for us.
890 	 */
891 
892 	ip6 = mtod(*m, struct ip6_hdr *);
893 
894 	/*
895 	 * Only continue processing if the packet is TCP
896 	 * XXX: We should follow the next header fields
897 	 * as shown on Pg 6 RFC 2460, but right now we'll
898 	 * only check pkts that have no extension headers.
899 	 */
900 	if (ip6->ip6_nxt != IPPROTO_TCP)
901 		goto ret6;
902 
903 	/*
904 	 * Create a tcphdr struct starting at the correct offset
905 	 * in the ipv6 packet.
906 	 */
907 	ip6_hl = sizeof(struct ip6_hdr);
908 	th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
909 
910 	/*
911 	 * Only pkts selected by the tcp port filter
912 	 * can be inserted into the pkt_queue
913 	 */
914 	if ((siftr_port_filter != 0) &&
915 	    (siftr_port_filter != ntohs(th->th_sport)) &&
916 	    (siftr_port_filter != ntohs(th->th_dport))) {
917 		goto ret6;
918 	}
919 
920 	/*
921 	 * If a kernel subsystem reinjects packets into the stack, our pfil
922 	 * hook will be called multiple times for the same packet.
923 	 * Make sure we only process unique packets.
924 	 */
925 	if (siftr_chkreinject(*m, dir, ss))
926 		goto ret6;
927 
928 	if (dir == PFIL_IN)
929 		ss->n_in++;
930 	else
931 		ss->n_out++;
932 
933 	/*
934 	 * For inbound packets, the pfil hooks don't provide a pointer to the
935 	 * inpcb, so we need to find it ourselves and lock it.
936 	 */
937 	if (!inp) {
938 		/* Find the corresponding inpcb for this pkt. */
939 		inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
940 		    th->th_sport, th->th_dport, dir, ss);
941 
942 		if (inp == NULL)
943 			goto ret6;
944 		else
945 			inp_locally_locked = 1;
946 	}
947 
948 	/* Find the TCP control block that corresponds with this packet. */
949 	tp = intotcpcb(inp);
950 
951 	/*
952 	 * If we can't find the TCP control block (happens occasionaly for a
953 	 * packet sent during the shutdown phase of a TCP connection), or the
954 	 * TCP control block has not initialized (happens during TCPS_SYN_SENT),
955 	 * bail.
956 	 */
957 	if (tp == NULL || tp->t_state < TCPS_ESTABLISHED) {
958 		if (dir == PFIL_IN)
959 			ss->nskip_in_tcpcb++;
960 		else
961 			ss->nskip_out_tcpcb++;
962 
963 		goto inp_unlock6;
964 	}
965 
966 	hash_id = siftr_get_flowid(inp, INP_IPV6, &hash_type);
967 	counter_list = counter_hash + (hash_id & siftr_hashmask);
968 	hash_node = siftr_find_flow(counter_list, hash_id);
969 
970 	/* If this flow hasn't been seen before, we create a new entry. */
971 	if (!hash_node) {
972 		struct flow_info info;
973 
974 		ip6_sprintf(info.laddr, &inp->in6p_laddr);
975 		ip6_sprintf(info.faddr, &inp->in6p_faddr);
976 		info.lport = ntohs(inp->inp_lport);
977 		info.fport = ntohs(inp->inp_fport);
978 		info.key = hash_id;
979 
980 		hash_node = siftr_new_hash_node(info, dir, ss);
981 	}
982 
983 	if (!hash_node) {
984 		goto inp_unlock6;
985 	}
986 
987 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
988 
989 	if (pn == NULL) {
990 		if (dir == PFIL_IN)
991 			ss->nskip_in_malloc++;
992 		else
993 			ss->nskip_out_malloc++;
994 
995 		goto inp_unlock6;
996 	}
997 
998 	pn->flowid = hash_id;
999 	pn->flowtype = hash_type;
1000 
1001 	siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
1002 
1003 	mtx_lock(&siftr_pkt_queue_mtx);
1004 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
1005 	mtx_unlock(&siftr_pkt_queue_mtx);
1006 	goto ret6;
1007 
1008 inp_unlock6:
1009 	if (inp_locally_locked)
1010 		INP_RUNLOCK(inp);
1011 
1012 ret6:
1013 	return (PFIL_PASS);
1014 }
1015 #endif /* #ifdef SIFTR_IPV6 */
1016 
1017 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet_hook);
1018 #define	V_siftr_inet_hook	VNET(siftr_inet_hook)
1019 #ifdef SIFTR_IPV6
1020 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet6_hook);
1021 #define	V_siftr_inet6_hook	VNET(siftr_inet6_hook)
1022 #endif
1023 static int
siftr_pfil(int action)1024 siftr_pfil(int action)
1025 {
1026 	struct pfil_hook_args pha = {
1027 		.pa_version = PFIL_VERSION,
1028 		.pa_flags = PFIL_IN | PFIL_OUT,
1029 		.pa_modname = "siftr",
1030 		.pa_rulname = "default",
1031 	};
1032 	struct pfil_link_args pla = {
1033 		.pa_version = PFIL_VERSION,
1034 		.pa_flags = PFIL_IN | PFIL_OUT | PFIL_HEADPTR | PFIL_HOOKPTR,
1035 	};
1036 
1037 	VNET_ITERATOR_DECL(vnet_iter);
1038 
1039 	VNET_LIST_RLOCK();
1040 	VNET_FOREACH(vnet_iter) {
1041 		CURVNET_SET(vnet_iter);
1042 
1043 		if (action == HOOK) {
1044 			pha.pa_mbuf_chk = siftr_chkpkt;
1045 			pha.pa_type = PFIL_TYPE_IP4;
1046 			V_siftr_inet_hook = pfil_add_hook(&pha);
1047 			pla.pa_hook = V_siftr_inet_hook;
1048 			pla.pa_head = V_inet_pfil_head;
1049 			(void)pfil_link(&pla);
1050 #ifdef SIFTR_IPV6
1051 			pha.pa_mbuf_chk = siftr_chkpkt6;
1052 			pha.pa_type = PFIL_TYPE_IP6;
1053 			V_siftr_inet6_hook = pfil_add_hook(&pha);
1054 			pla.pa_hook = V_siftr_inet6_hook;
1055 			pla.pa_head = V_inet6_pfil_head;
1056 			(void)pfil_link(&pla);
1057 #endif
1058 		} else if (action == UNHOOK) {
1059 			pfil_remove_hook(V_siftr_inet_hook);
1060 #ifdef SIFTR_IPV6
1061 			pfil_remove_hook(V_siftr_inet6_hook);
1062 #endif
1063 		}
1064 		CURVNET_RESTORE();
1065 	}
1066 	VNET_LIST_RUNLOCK();
1067 
1068 	return (0);
1069 }
1070 
1071 static int
siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)1072 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1073 {
1074 	struct alq *new_alq;
1075 	int error;
1076 
1077 	error = sysctl_handle_string(oidp, arg1, arg2, req);
1078 
1079 	/* Check for error or same filename */
1080 	if (error != 0 || req->newptr == NULL ||
1081 	    strncmp(siftr_logfile, arg1, arg2) == 0)
1082 		goto done;
1083 
1084 	/* file name changed */
1085 	error = alq_open(&new_alq, arg1, curthread->td_ucred,
1086 	    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1087 	if (error != 0)
1088 		goto done;
1089 
1090 	/*
1091 	 * If disabled, siftr_alq == NULL so we simply close
1092 	 * the alq as we've proved it can be opened.
1093 	 * If enabled, close the existing alq and switch the old
1094 	 * for the new.
1095 	 */
1096 	if (siftr_alq == NULL) {
1097 		alq_close(new_alq);
1098 	} else {
1099 		alq_close(siftr_alq);
1100 		siftr_alq = new_alq;
1101 	}
1102 
1103 	/* Update filename upon success */
1104 	strlcpy(siftr_logfile, arg1, arg2);
1105 done:
1106 	return (error);
1107 }
1108 
1109 static int
siftr_manage_ops(uint8_t action)1110 siftr_manage_ops(uint8_t action)
1111 {
1112 	struct siftr_stats totalss;
1113 	struct timeval tval;
1114 	struct flow_hash_node *counter, *tmp_counter;
1115 	struct sbuf *s;
1116 	int i, error;
1117 	uint32_t bytes_to_write, total_skipped_pkts;
1118 
1119 	error = 0;
1120 	total_skipped_pkts = 0;
1121 
1122 	/* Init an autosizing sbuf that initially holds 200 chars. */
1123 	if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1124 		return (-1);
1125 
1126 	if (action == SIFTR_ENABLE && siftr_pkt_manager_thr == NULL) {
1127 		/*
1128 		 * Create our alq
1129 		 * XXX: We should abort if alq_open fails!
1130 		 */
1131 		alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1132 		    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1133 
1134 		STAILQ_INIT(&pkt_queue);
1135 
1136 		DPCPU_ZERO(ss);
1137 
1138 		siftr_exit_pkt_manager_thread = 0;
1139 
1140 		kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1141 		    &siftr_pkt_manager_thr, RFNOWAIT, 0,
1142 		    "siftr_pkt_manager_thr");
1143 
1144 		siftr_pfil(HOOK);
1145 
1146 		microtime(&tval);
1147 
1148 		sbuf_printf(s,
1149 		    "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1150 		    "siftrver=%s\tsysname=%s\tsysver=%u\tipmode=%u\n",
1151 		    (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR,
1152 		    SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1153 
1154 		sbuf_finish(s);
1155 		alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1156 
1157 	} else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1158 		/*
1159 		 * Remove the pfil hook functions. All threads currently in
1160 		 * the hook functions are allowed to exit before siftr_pfil()
1161 		 * returns.
1162 		 */
1163 		siftr_pfil(UNHOOK);
1164 
1165 		/* This will block until the pkt manager thread unlocks it. */
1166 		mtx_lock(&siftr_pkt_mgr_mtx);
1167 
1168 		/* Tell the pkt manager thread that it should exit now. */
1169 		siftr_exit_pkt_manager_thread = 1;
1170 
1171 		/*
1172 		 * Wake the pkt_manager thread so it realises that
1173 		 * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1174 		 * The wakeup won't be delivered until we unlock
1175 		 * siftr_pkt_mgr_mtx so this isn't racy.
1176 		 */
1177 		wakeup(&wait_for_pkt);
1178 
1179 		/* Wait for the pkt_manager thread to exit. */
1180 		mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1181 		    "thrwait", 0);
1182 
1183 		siftr_pkt_manager_thr = NULL;
1184 		mtx_unlock(&siftr_pkt_mgr_mtx);
1185 
1186 		totalss.n_in = DPCPU_VARSUM(ss, n_in);
1187 		totalss.n_out = DPCPU_VARSUM(ss, n_out);
1188 		totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1189 		totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1190 		totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1191 		totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1192 		totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1193 		totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1194 
1195 		total_skipped_pkts = totalss.nskip_in_malloc +
1196 		    totalss.nskip_out_malloc + totalss.nskip_in_tcpcb +
1197 		    totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1198 		    totalss.nskip_out_inpcb;
1199 
1200 		microtime(&tval);
1201 
1202 		sbuf_printf(s,
1203 		    "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1204 		    "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1205 		    "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1206 		    "num_outbound_skipped_pkts_malloc=%u\t"
1207 		    "num_inbound_skipped_pkts_tcpcb=%u\t"
1208 		    "num_outbound_skipped_pkts_tcpcb=%u\t"
1209 		    "num_inbound_skipped_pkts_inpcb=%u\t"
1210 		    "num_outbound_skipped_pkts_inpcb=%u\t"
1211 		    "total_skipped_tcp_pkts=%u\tflow_list=",
1212 		    (intmax_t)tval.tv_sec,
1213 		    tval.tv_usec,
1214 		    (uintmax_t)totalss.n_in,
1215 		    (uintmax_t)totalss.n_out,
1216 		    (uintmax_t)(totalss.n_in + totalss.n_out),
1217 		    totalss.nskip_in_malloc,
1218 		    totalss.nskip_out_malloc,
1219 		    totalss.nskip_in_tcpcb,
1220 		    totalss.nskip_out_tcpcb,
1221 		    totalss.nskip_in_inpcb,
1222 		    totalss.nskip_out_inpcb,
1223 		    total_skipped_pkts);
1224 
1225 		/*
1226 		 * Iterate over the flow hash, printing a summary of each
1227 		 * flow seen and freeing any malloc'd memory.
1228 		 * The hash consists of an array of LISTs (man 3 queue).
1229 		 */
1230 		for (i = 0; i <= siftr_hashmask; i++) {
1231 			LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1232 			    tmp_counter) {
1233 				sbuf_printf(s, "%s;%hu-%s;%hu,",
1234 					    counter->const_info.laddr,
1235 					    counter->const_info.lport,
1236 					    counter->const_info.faddr,
1237 					    counter->const_info.fport);
1238 
1239 				free(counter, M_SIFTR_HASHNODE);
1240 			}
1241 
1242 			LIST_INIT(counter_hash + i);
1243 		}
1244 
1245 		sbuf_printf(s, "\n");
1246 		sbuf_finish(s);
1247 
1248 		i = 0;
1249 		do {
1250 			bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1251 			alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1252 			i += bytes_to_write;
1253 		} while (i < sbuf_len(s));
1254 
1255 		alq_close(siftr_alq);
1256 		siftr_alq = NULL;
1257 	} else
1258 		error = EINVAL;
1259 
1260 	sbuf_delete(s);
1261 
1262 	/*
1263 	 * XXX: Should be using ret to check if any functions fail
1264 	 * and set error appropriately
1265 	 */
1266 
1267 	return (error);
1268 }
1269 
1270 static int
siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)1271 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1272 {
1273 	int error;
1274 	uint32_t new;
1275 
1276 	new = siftr_enabled;
1277 	error = sysctl_handle_int(oidp, &new, 0, req);
1278 	if (error == 0 && req->newptr != NULL) {
1279 		if (new > 1)
1280 			return (EINVAL);
1281 		else if (new != siftr_enabled) {
1282 			if ((error = siftr_manage_ops(new)) == 0) {
1283 				siftr_enabled = new;
1284 			} else {
1285 				siftr_manage_ops(SIFTR_DISABLE);
1286 			}
1287 		}
1288 	}
1289 
1290 	return (error);
1291 }
1292 
1293 static void
siftr_shutdown_handler(void * arg,int howto)1294 siftr_shutdown_handler(void *arg, int howto)
1295 {
1296 	if ((howto & RB_NOSYNC) != 0 || SCHEDULER_STOPPED())
1297 		return;
1298 
1299 	if (siftr_enabled == 1) {
1300 		siftr_manage_ops(SIFTR_DISABLE);
1301 	}
1302 }
1303 
1304 /*
1305  * Module is being unloaded or machine is shutting down. Take care of cleanup.
1306  */
1307 static int
deinit_siftr(void)1308 deinit_siftr(void)
1309 {
1310 	/* Cleanup. */
1311 	EVENTHANDLER_DEREGISTER(shutdown_pre_sync, siftr_shutdown_tag);
1312 	siftr_manage_ops(SIFTR_DISABLE);
1313 	hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1314 	mtx_destroy(&siftr_pkt_queue_mtx);
1315 	mtx_destroy(&siftr_pkt_mgr_mtx);
1316 
1317 	return (0);
1318 }
1319 
1320 /*
1321  * Module has just been loaded into the kernel.
1322  */
1323 static int
init_siftr(void)1324 init_siftr(void)
1325 {
1326 	siftr_shutdown_tag = EVENTHANDLER_REGISTER(shutdown_pre_sync,
1327 	    siftr_shutdown_handler, NULL, SHUTDOWN_PRI_FIRST);
1328 
1329 	/* Initialise our flow counter hash table. */
1330 	counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1331 	    &siftr_hashmask);
1332 
1333 	mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1334 	mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1335 
1336 	/* Print message to the user's current terminal. */
1337 	uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1338 	    "          http://caia.swin.edu.au/urp/newtcp\n\n",
1339 	    MODVERSION_STR);
1340 
1341 	return (0);
1342 }
1343 
1344 /*
1345  * This is the function that is called to load and unload the module.
1346  * When the module is loaded, this function is called once with
1347  * "what" == MOD_LOAD
1348  * When the module is unloaded, this function is called twice with
1349  * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1350  * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1351  * this function is called once with "what" = MOD_SHUTDOWN
1352  * When the system is shut down, the handler isn't called until the very end
1353  * of the shutdown sequence i.e. after the disks have been synced.
1354  */
1355 static int
siftr_load_handler(module_t mod,int what,void * arg)1356 siftr_load_handler(module_t mod, int what, void *arg)
1357 {
1358 	int ret;
1359 
1360 	switch (what) {
1361 	case MOD_LOAD:
1362 		ret = init_siftr();
1363 		break;
1364 
1365 	case MOD_QUIESCE:
1366 	case MOD_SHUTDOWN:
1367 		ret = deinit_siftr();
1368 		break;
1369 
1370 	case MOD_UNLOAD:
1371 		ret = 0;
1372 		break;
1373 
1374 	default:
1375 		ret = EINVAL;
1376 		break;
1377 	}
1378 
1379 	return (ret);
1380 }
1381 
1382 static moduledata_t siftr_mod = {
1383 	.name = "siftr",
1384 	.evhand = siftr_load_handler,
1385 };
1386 
1387 /*
1388  * Param 1: name of the kernel module
1389  * Param 2: moduledata_t struct containing info about the kernel module
1390  *          and the execution entry point for the module
1391  * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1392  *          Defines the module initialisation order
1393  * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1394  *          Defines the initialisation order of this kld relative to others
1395  *          within the same subsystem as defined by param 3
1396  */
1397 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_LAST, SI_ORDER_ANY);
1398 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1399 MODULE_VERSION(siftr, MODVERSION);
1400