xref: /freebsd/sys/netinet/siftr.c (revision 788ca347b816afd83b2885e0c79aeeb88649b2ab)
1 /*-
2  * Copyright (c) 2007-2009
3  * 	Swinburne University of Technology, Melbourne, Australia.
4  * Copyright (c) 2009-2010, The FreeBSD Foundation
5  * All rights reserved.
6  *
7  * Portions of this software were developed at the Centre for Advanced
8  * Internet Architectures, Swinburne University of Technology, Melbourne,
9  * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /******************************************************
34  * Statistical Information For TCP Research (SIFTR)
35  *
36  * A FreeBSD kernel module that adds very basic intrumentation to the
37  * TCP stack, allowing internal stats to be recorded to a log file
38  * for experimental, debugging and performance analysis purposes.
39  *
40  * SIFTR was first released in 2007 by James Healy and Lawrence Stewart whilst
41  * working on the NewTCP research project at Swinburne University of
42  * Technology's Centre for Advanced Internet Architectures, Melbourne,
43  * Australia, which was made possible in part by a grant from the Cisco
44  * University Research Program Fund at Community Foundation Silicon Valley.
45  * More details are available at:
46  *   http://caia.swin.edu.au/urp/newtcp/
47  *
48  * Work on SIFTR v1.2.x was sponsored by the FreeBSD Foundation as part of
49  * the "Enhancing the FreeBSD TCP Implementation" project 2008-2009.
50  * More details are available at:
51  *   http://www.freebsdfoundation.org/
52  *   http://caia.swin.edu.au/freebsd/etcp09/
53  *
54  * Lawrence Stewart is the current maintainer, and all contact regarding
55  * SIFTR should be directed to him via email: lastewart@swin.edu.au
56  *
57  * Initial release date: June 2007
58  * Most recent update: September 2010
59  ******************************************************/
60 
61 #include <sys/cdefs.h>
62 __FBSDID("$FreeBSD$");
63 
64 #include <sys/param.h>
65 #include <sys/alq.h>
66 #include <sys/errno.h>
67 #include <sys/eventhandler.h>
68 #include <sys/hash.h>
69 #include <sys/kernel.h>
70 #include <sys/kthread.h>
71 #include <sys/lock.h>
72 #include <sys/mbuf.h>
73 #include <sys/module.h>
74 #include <sys/mutex.h>
75 #include <sys/pcpu.h>
76 #include <sys/proc.h>
77 #include <sys/sbuf.h>
78 #include <sys/smp.h>
79 #include <sys/socket.h>
80 #include <sys/socketvar.h>
81 #include <sys/sysctl.h>
82 #include <sys/unistd.h>
83 
84 #include <net/if.h>
85 #include <net/if_var.h>
86 #include <net/pfil.h>
87 
88 #include <netinet/in.h>
89 #include <netinet/in_pcb.h>
90 #include <netinet/in_systm.h>
91 #include <netinet/in_var.h>
92 #include <netinet/ip.h>
93 #include <netinet/tcp_var.h>
94 
95 #ifdef SIFTR_IPV6
96 #include <netinet/ip6.h>
97 #include <netinet6/in6_pcb.h>
98 #endif /* SIFTR_IPV6 */
99 
100 #include <machine/in_cksum.h>
101 
102 /*
103  * Three digit version number refers to X.Y.Z where:
104  * X is the major version number
105  * Y is bumped to mark backwards incompatible changes
106  * Z is bumped to mark backwards compatible changes
107  */
108 #define V_MAJOR		1
109 #define V_BACKBREAK	2
110 #define V_BACKCOMPAT	4
111 #define MODVERSION	__CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT))
112 #define MODVERSION_STR	__XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \
113     __XSTRING(V_BACKCOMPAT)
114 
115 #define HOOK 0
116 #define UNHOOK 1
117 #define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536
118 #define SYS_NAME "FreeBSD"
119 #define PACKET_TAG_SIFTR 100
120 #define PACKET_COOKIE_SIFTR 21749576
121 #define SIFTR_LOG_FILE_MODE 0644
122 #define SIFTR_DISABLE 0
123 #define SIFTR_ENABLE 1
124 
125 /*
126  * Hard upper limit on the length of log messages. Bump this up if you add new
127  * data fields such that the line length could exceed the below value.
128  */
129 #define MAX_LOG_MSG_LEN 200
130 /* XXX: Make this a sysctl tunable. */
131 #define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN)
132 
133 /*
134  * 1 byte for IP version
135  * IPv4: src/dst IP (4+4) + src/dst port (2+2) = 12 bytes
136  * IPv6: src/dst IP (16+16) + src/dst port (2+2) = 36 bytes
137  */
138 #ifdef SIFTR_IPV6
139 #define FLOW_KEY_LEN 37
140 #else
141 #define FLOW_KEY_LEN 13
142 #endif
143 
144 #ifdef SIFTR_IPV6
145 #define SIFTR_IPMODE 6
146 #else
147 #define SIFTR_IPMODE 4
148 #endif
149 
150 /* useful macros */
151 #define CAST_PTR_INT(X) (*((int*)(X)))
152 
153 #define UPPER_SHORT(X)	(((X) & 0xFFFF0000) >> 16)
154 #define LOWER_SHORT(X)	((X) & 0x0000FFFF)
155 
156 #define FIRST_OCTET(X)	(((X) & 0xFF000000) >> 24)
157 #define SECOND_OCTET(X)	(((X) & 0x00FF0000) >> 16)
158 #define THIRD_OCTET(X)	(((X) & 0x0000FF00) >> 8)
159 #define FOURTH_OCTET(X)	((X) & 0x000000FF)
160 
161 static MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR");
162 static MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode",
163     "SIFTR pkt_node struct");
164 static MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode",
165     "SIFTR flow_hash_node struct");
166 
167 /* Used as links in the pkt manager queue. */
168 struct pkt_node {
169 	/* Timestamp of pkt as noted in the pfil hook. */
170 	struct timeval		tval;
171 	/* Direction pkt is travelling; either PFIL_IN or PFIL_OUT. */
172 	uint8_t			direction;
173 	/* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */
174 	uint8_t			ipver;
175 	/* Hash of the pkt which triggered the log message. */
176 	uint32_t		hash;
177 	/* Local/foreign IP address. */
178 #ifdef SIFTR_IPV6
179 	uint32_t		ip_laddr[4];
180 	uint32_t		ip_faddr[4];
181 #else
182 	uint8_t			ip_laddr[4];
183 	uint8_t			ip_faddr[4];
184 #endif
185 	/* Local TCP port. */
186 	uint16_t		tcp_localport;
187 	/* Foreign TCP port. */
188 	uint16_t		tcp_foreignport;
189 	/* Congestion Window (bytes). */
190 	u_long			snd_cwnd;
191 	/* Sending Window (bytes). */
192 	u_long			snd_wnd;
193 	/* Receive Window (bytes). */
194 	u_long			rcv_wnd;
195 	/* Unused (was: Bandwidth Controlled Window (bytes)). */
196 	u_long			snd_bwnd;
197 	/* Slow Start Threshold (bytes). */
198 	u_long			snd_ssthresh;
199 	/* Current state of the TCP FSM. */
200 	int			conn_state;
201 	/* Max Segment Size (bytes). */
202 	u_int			max_seg_size;
203 	/*
204 	 * Smoothed RTT stored as found in the TCP control block
205 	 * in units of (TCP_RTT_SCALE*hz).
206 	 */
207 	int			smoothed_rtt;
208 	/* Is SACK enabled? */
209 	u_char			sack_enabled;
210 	/* Window scaling for snd window. */
211 	u_char			snd_scale;
212 	/* Window scaling for recv window. */
213 	u_char			rcv_scale;
214 	/* TCP control block flags. */
215 	u_int			flags;
216 	/* Retransmit timeout length. */
217 	int			rxt_length;
218 	/* Size of the TCP send buffer in bytes. */
219 	u_int			snd_buf_hiwater;
220 	/* Current num bytes in the send socket buffer. */
221 	u_int			snd_buf_cc;
222 	/* Size of the TCP receive buffer in bytes. */
223 	u_int			rcv_buf_hiwater;
224 	/* Current num bytes in the receive socket buffer. */
225 	u_int			rcv_buf_cc;
226 	/* Number of bytes inflight that we are waiting on ACKs for. */
227 	u_int			sent_inflight_bytes;
228 	/* Number of segments currently in the reassembly queue. */
229 	int			t_segqlen;
230 	/* Flowid for the connection. */
231 	u_int			flowid;
232 	/* Flow type for the connection. */
233 	u_int			flowtype;
234 	/* Link to next pkt_node in the list. */
235 	STAILQ_ENTRY(pkt_node)	nodes;
236 };
237 
238 struct flow_hash_node
239 {
240 	uint16_t counter;
241 	uint8_t key[FLOW_KEY_LEN];
242 	LIST_ENTRY(flow_hash_node) nodes;
243 };
244 
245 struct siftr_stats
246 {
247 	/* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */
248 	uint64_t n_in;
249 	uint64_t n_out;
250 	/* # pkts skipped due to failed malloc calls. */
251 	uint32_t nskip_in_malloc;
252 	uint32_t nskip_out_malloc;
253 	/* # pkts skipped due to failed mtx acquisition. */
254 	uint32_t nskip_in_mtx;
255 	uint32_t nskip_out_mtx;
256 	/* # pkts skipped due to failed inpcb lookups. */
257 	uint32_t nskip_in_inpcb;
258 	uint32_t nskip_out_inpcb;
259 	/* # pkts skipped due to failed tcpcb lookups. */
260 	uint32_t nskip_in_tcpcb;
261 	uint32_t nskip_out_tcpcb;
262 	/* # pkts skipped due to stack reinjection. */
263 	uint32_t nskip_in_dejavu;
264 	uint32_t nskip_out_dejavu;
265 };
266 
267 static DPCPU_DEFINE(struct siftr_stats, ss);
268 
269 static volatile unsigned int siftr_exit_pkt_manager_thread = 0;
270 static unsigned int siftr_enabled = 0;
271 static unsigned int siftr_pkts_per_log = 1;
272 static unsigned int siftr_generate_hashes = 0;
273 /* static unsigned int siftr_binary_log = 0; */
274 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log";
275 static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log";
276 static u_long siftr_hashmask;
277 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue);
278 LIST_HEAD(listhead, flow_hash_node) *counter_hash;
279 static int wait_for_pkt;
280 static struct alq *siftr_alq = NULL;
281 static struct mtx siftr_pkt_queue_mtx;
282 static struct mtx siftr_pkt_mgr_mtx;
283 static struct thread *siftr_pkt_manager_thr = NULL;
284 /*
285  * pfil.h defines PFIL_IN as 1 and PFIL_OUT as 2,
286  * which we use as an index into this array.
287  */
288 static char direction[3] = {'\0', 'i','o'};
289 
290 /* Required function prototypes. */
291 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS);
292 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS);
293 
294 
295 /* Declare the net.inet.siftr sysctl tree and populate it. */
296 
297 SYSCTL_DECL(_net_inet_siftr);
298 
299 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW, NULL,
300     "siftr related settings");
301 
302 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, CTLTYPE_UINT|CTLFLAG_RW,
303     &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU",
304     "switch siftr module operations on/off");
305 
306 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, CTLTYPE_STRING|CTLFLAG_RW,
307     &siftr_logfile_shadow, sizeof(siftr_logfile_shadow), &siftr_sysctl_logfile_name_handler,
308     "A", "file to save siftr log messages to");
309 
310 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW,
311     &siftr_pkts_per_log, 1,
312     "number of packets between generating a log message");
313 
314 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, genhashes, CTLFLAG_RW,
315     &siftr_generate_hashes, 0,
316     "enable packet hash generation");
317 
318 /* XXX: TODO
319 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW,
320     &siftr_binary_log, 0,
321     "write log files in binary instead of ascii");
322 */
323 
324 
325 /* Begin functions. */
326 
327 static void
328 siftr_process_pkt(struct pkt_node * pkt_node)
329 {
330 	struct flow_hash_node *hash_node;
331 	struct listhead *counter_list;
332 	struct siftr_stats *ss;
333 	struct ale *log_buf;
334 	uint8_t key[FLOW_KEY_LEN];
335 	uint8_t found_match, key_offset;
336 
337 	hash_node = NULL;
338 	ss = DPCPU_PTR(ss);
339 	found_match = 0;
340 	key_offset = 1;
341 
342 	/*
343 	 * Create the key that will be used to create a hash index
344 	 * into our hash table. Our key consists of:
345 	 * ipversion, localip, localport, foreignip, foreignport
346 	 */
347 	key[0] = pkt_node->ipver;
348 	memcpy(key + key_offset, &pkt_node->ip_laddr,
349 	    sizeof(pkt_node->ip_laddr));
350 	key_offset += sizeof(pkt_node->ip_laddr);
351 	memcpy(key + key_offset, &pkt_node->tcp_localport,
352 	    sizeof(pkt_node->tcp_localport));
353 	key_offset += sizeof(pkt_node->tcp_localport);
354 	memcpy(key + key_offset, &pkt_node->ip_faddr,
355 	    sizeof(pkt_node->ip_faddr));
356 	key_offset += sizeof(pkt_node->ip_faddr);
357 	memcpy(key + key_offset, &pkt_node->tcp_foreignport,
358 	    sizeof(pkt_node->tcp_foreignport));
359 
360 	counter_list = counter_hash +
361 	    (hash32_buf(key, sizeof(key), 0) & siftr_hashmask);
362 
363 	/*
364 	 * If the list is not empty i.e. the hash index has
365 	 * been used by another flow previously.
366 	 */
367 	if (LIST_FIRST(counter_list) != NULL) {
368 		/*
369 		 * Loop through the hash nodes in the list.
370 		 * There should normally only be 1 hash node in the list,
371 		 * except if there have been collisions at the hash index
372 		 * computed by hash32_buf().
373 		 */
374 		LIST_FOREACH(hash_node, counter_list, nodes) {
375 			/*
376 			 * Check if the key for the pkt we are currently
377 			 * processing is the same as the key stored in the
378 			 * hash node we are currently processing.
379 			 * If they are the same, then we've found the
380 			 * hash node that stores the counter for the flow
381 			 * the pkt belongs to.
382 			 */
383 			if (memcmp(hash_node->key, key, sizeof(key)) == 0) {
384 				found_match = 1;
385 				break;
386 			}
387 		}
388 	}
389 
390 	/* If this flow hash hasn't been seen before or we have a collision. */
391 	if (hash_node == NULL || !found_match) {
392 		/* Create a new hash node to store the flow's counter. */
393 		hash_node = malloc(sizeof(struct flow_hash_node),
394 		    M_SIFTR_HASHNODE, M_WAITOK);
395 
396 		if (hash_node != NULL) {
397 			/* Initialise our new hash node list entry. */
398 			hash_node->counter = 0;
399 			memcpy(hash_node->key, key, sizeof(key));
400 			LIST_INSERT_HEAD(counter_list, hash_node, nodes);
401 		} else {
402 			/* Malloc failed. */
403 			if (pkt_node->direction == PFIL_IN)
404 				ss->nskip_in_malloc++;
405 			else
406 				ss->nskip_out_malloc++;
407 
408 			return;
409 		}
410 	} else if (siftr_pkts_per_log > 1) {
411 		/*
412 		 * Taking the remainder of the counter divided
413 		 * by the current value of siftr_pkts_per_log
414 		 * and storing that in counter provides a neat
415 		 * way to modulate the frequency of log
416 		 * messages being written to the log file.
417 		 */
418 		hash_node->counter = (hash_node->counter + 1) %
419 		    siftr_pkts_per_log;
420 
421 		/*
422 		 * If we have not seen enough packets since the last time
423 		 * we wrote a log message for this connection, return.
424 		 */
425 		if (hash_node->counter > 0)
426 			return;
427 	}
428 
429 	log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK);
430 
431 	if (log_buf == NULL)
432 		return; /* Should only happen if the ALQ is shutting down. */
433 
434 #ifdef SIFTR_IPV6
435 	pkt_node->ip_laddr[3] = ntohl(pkt_node->ip_laddr[3]);
436 	pkt_node->ip_faddr[3] = ntohl(pkt_node->ip_faddr[3]);
437 
438 	if (pkt_node->ipver == INP_IPV6) { /* IPv6 packet */
439 		pkt_node->ip_laddr[0] = ntohl(pkt_node->ip_laddr[0]);
440 		pkt_node->ip_laddr[1] = ntohl(pkt_node->ip_laddr[1]);
441 		pkt_node->ip_laddr[2] = ntohl(pkt_node->ip_laddr[2]);
442 		pkt_node->ip_faddr[0] = ntohl(pkt_node->ip_faddr[0]);
443 		pkt_node->ip_faddr[1] = ntohl(pkt_node->ip_faddr[1]);
444 		pkt_node->ip_faddr[2] = ntohl(pkt_node->ip_faddr[2]);
445 
446 		/* Construct an IPv6 log message. */
447 		log_buf->ae_bytesused = snprintf(log_buf->ae_data,
448 		    MAX_LOG_MSG_LEN,
449 		    "%c,0x%08x,%zd.%06ld,%x:%x:%x:%x:%x:%x:%x:%x,%u,%x:%x:%x:"
450 		    "%x:%x:%x:%x:%x,%u,%ld,%ld,%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,"
451 		    "%u,%d,%u,%u,%u,%u,%u,%u,%u,%u\n",
452 		    direction[pkt_node->direction],
453 		    pkt_node->hash,
454 		    pkt_node->tval.tv_sec,
455 		    pkt_node->tval.tv_usec,
456 		    UPPER_SHORT(pkt_node->ip_laddr[0]),
457 		    LOWER_SHORT(pkt_node->ip_laddr[0]),
458 		    UPPER_SHORT(pkt_node->ip_laddr[1]),
459 		    LOWER_SHORT(pkt_node->ip_laddr[1]),
460 		    UPPER_SHORT(pkt_node->ip_laddr[2]),
461 		    LOWER_SHORT(pkt_node->ip_laddr[2]),
462 		    UPPER_SHORT(pkt_node->ip_laddr[3]),
463 		    LOWER_SHORT(pkt_node->ip_laddr[3]),
464 		    ntohs(pkt_node->tcp_localport),
465 		    UPPER_SHORT(pkt_node->ip_faddr[0]),
466 		    LOWER_SHORT(pkt_node->ip_faddr[0]),
467 		    UPPER_SHORT(pkt_node->ip_faddr[1]),
468 		    LOWER_SHORT(pkt_node->ip_faddr[1]),
469 		    UPPER_SHORT(pkt_node->ip_faddr[2]),
470 		    LOWER_SHORT(pkt_node->ip_faddr[2]),
471 		    UPPER_SHORT(pkt_node->ip_faddr[3]),
472 		    LOWER_SHORT(pkt_node->ip_faddr[3]),
473 		    ntohs(pkt_node->tcp_foreignport),
474 		    pkt_node->snd_ssthresh,
475 		    pkt_node->snd_cwnd,
476 		    pkt_node->snd_bwnd,
477 		    pkt_node->snd_wnd,
478 		    pkt_node->rcv_wnd,
479 		    pkt_node->snd_scale,
480 		    pkt_node->rcv_scale,
481 		    pkt_node->conn_state,
482 		    pkt_node->max_seg_size,
483 		    pkt_node->smoothed_rtt,
484 		    pkt_node->sack_enabled,
485 		    pkt_node->flags,
486 		    pkt_node->rxt_length,
487 		    pkt_node->snd_buf_hiwater,
488 		    pkt_node->snd_buf_cc,
489 		    pkt_node->rcv_buf_hiwater,
490 		    pkt_node->rcv_buf_cc,
491 		    pkt_node->sent_inflight_bytes,
492 		    pkt_node->t_segqlen,
493 		    pkt_node->flowid,
494 		    pkt_node->flowtype);
495 	} else { /* IPv4 packet */
496 		pkt_node->ip_laddr[0] = FIRST_OCTET(pkt_node->ip_laddr[3]);
497 		pkt_node->ip_laddr[1] = SECOND_OCTET(pkt_node->ip_laddr[3]);
498 		pkt_node->ip_laddr[2] = THIRD_OCTET(pkt_node->ip_laddr[3]);
499 		pkt_node->ip_laddr[3] = FOURTH_OCTET(pkt_node->ip_laddr[3]);
500 		pkt_node->ip_faddr[0] = FIRST_OCTET(pkt_node->ip_faddr[3]);
501 		pkt_node->ip_faddr[1] = SECOND_OCTET(pkt_node->ip_faddr[3]);
502 		pkt_node->ip_faddr[2] = THIRD_OCTET(pkt_node->ip_faddr[3]);
503 		pkt_node->ip_faddr[3] = FOURTH_OCTET(pkt_node->ip_faddr[3]);
504 #endif /* SIFTR_IPV6 */
505 
506 		/* Construct an IPv4 log message. */
507 		log_buf->ae_bytesused = snprintf(log_buf->ae_data,
508 		    MAX_LOG_MSG_LEN,
509 		    "%c,0x%08x,%jd.%06ld,%u.%u.%u.%u,%u,%u.%u.%u.%u,%u,%ld,%ld,"
510 		    "%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,%u,%d,%u,%u,%u,%u,%u,%u,%u,%u\n",
511 		    direction[pkt_node->direction],
512 		    pkt_node->hash,
513 		    (intmax_t)pkt_node->tval.tv_sec,
514 		    pkt_node->tval.tv_usec,
515 		    pkt_node->ip_laddr[0],
516 		    pkt_node->ip_laddr[1],
517 		    pkt_node->ip_laddr[2],
518 		    pkt_node->ip_laddr[3],
519 		    ntohs(pkt_node->tcp_localport),
520 		    pkt_node->ip_faddr[0],
521 		    pkt_node->ip_faddr[1],
522 		    pkt_node->ip_faddr[2],
523 		    pkt_node->ip_faddr[3],
524 		    ntohs(pkt_node->tcp_foreignport),
525 		    pkt_node->snd_ssthresh,
526 		    pkt_node->snd_cwnd,
527 		    pkt_node->snd_bwnd,
528 		    pkt_node->snd_wnd,
529 		    pkt_node->rcv_wnd,
530 		    pkt_node->snd_scale,
531 		    pkt_node->rcv_scale,
532 		    pkt_node->conn_state,
533 		    pkt_node->max_seg_size,
534 		    pkt_node->smoothed_rtt,
535 		    pkt_node->sack_enabled,
536 		    pkt_node->flags,
537 		    pkt_node->rxt_length,
538 		    pkt_node->snd_buf_hiwater,
539 		    pkt_node->snd_buf_cc,
540 		    pkt_node->rcv_buf_hiwater,
541 		    pkt_node->rcv_buf_cc,
542 		    pkt_node->sent_inflight_bytes,
543 		    pkt_node->t_segqlen,
544 		    pkt_node->flowid,
545 		    pkt_node->flowtype);
546 #ifdef SIFTR_IPV6
547 	}
548 #endif
549 
550 	alq_post_flags(siftr_alq, log_buf, 0);
551 }
552 
553 
554 static void
555 siftr_pkt_manager_thread(void *arg)
556 {
557 	STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
558 	    STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
559 	struct pkt_node *pkt_node, *pkt_node_temp;
560 	uint8_t draining;
561 
562 	draining = 2;
563 
564 	mtx_lock(&siftr_pkt_mgr_mtx);
565 
566 	/* draining == 0 when queue has been flushed and it's safe to exit. */
567 	while (draining) {
568 		/*
569 		 * Sleep until we are signalled to wake because thread has
570 		 * been told to exit or until 1 tick has passed.
571 		 */
572 		mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
573 		    1);
574 
575 		/* Gain exclusive access to the pkt_node queue. */
576 		mtx_lock(&siftr_pkt_queue_mtx);
577 
578 		/*
579 		 * Move pkt_queue to tmp_pkt_queue, which leaves
580 		 * pkt_queue empty and ready to receive more pkt_nodes.
581 		 */
582 		STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
583 
584 		/*
585 		 * We've finished making changes to the list. Unlock it
586 		 * so the pfil hooks can continue queuing pkt_nodes.
587 		 */
588 		mtx_unlock(&siftr_pkt_queue_mtx);
589 
590 		/*
591 		 * We can't hold a mutex whilst calling siftr_process_pkt
592 		 * because ALQ might sleep waiting for buffer space.
593 		 */
594 		mtx_unlock(&siftr_pkt_mgr_mtx);
595 
596 		/* Flush all pkt_nodes to the log file. */
597 		STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes,
598 		    pkt_node_temp) {
599 			siftr_process_pkt(pkt_node);
600 			STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
601 			free(pkt_node, M_SIFTR_PKTNODE);
602 		}
603 
604 		KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
605 		    ("SIFTR tmp_pkt_queue not empty after flush"));
606 
607 		mtx_lock(&siftr_pkt_mgr_mtx);
608 
609 		/*
610 		 * If siftr_exit_pkt_manager_thread gets set during the window
611 		 * where we are draining the tmp_pkt_queue above, there might
612 		 * still be pkts in pkt_queue that need to be drained.
613 		 * Allow one further iteration to occur after
614 		 * siftr_exit_pkt_manager_thread has been set to ensure
615 		 * pkt_queue is completely empty before we kill the thread.
616 		 *
617 		 * siftr_exit_pkt_manager_thread is set only after the pfil
618 		 * hooks have been removed, so only 1 extra iteration
619 		 * is needed to drain the queue.
620 		 */
621 		if (siftr_exit_pkt_manager_thread)
622 			draining--;
623 	}
624 
625 	mtx_unlock(&siftr_pkt_mgr_mtx);
626 
627 	/* Calls wakeup on this thread's struct thread ptr. */
628 	kthread_exit();
629 }
630 
631 
632 static uint32_t
633 hash_pkt(struct mbuf *m, uint32_t offset)
634 {
635 	uint32_t hash;
636 
637 	hash = 0;
638 
639 	while (m != NULL && offset > m->m_len) {
640 		/*
641 		 * The IP packet payload does not start in this mbuf, so
642 		 * need to figure out which mbuf it starts in and what offset
643 		 * into the mbuf's data region the payload starts at.
644 		 */
645 		offset -= m->m_len;
646 		m = m->m_next;
647 	}
648 
649 	while (m != NULL) {
650 		/* Ensure there is data in the mbuf */
651 		if ((m->m_len - offset) > 0)
652 			hash = hash32_buf(m->m_data + offset,
653 			    m->m_len - offset, hash);
654 
655 		m = m->m_next;
656 		offset = 0;
657         }
658 
659 	return (hash);
660 }
661 
662 
663 /*
664  * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
665  * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
666  * Return value >0 means the caller should skip processing this mbuf.
667  */
668 static inline int
669 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
670 {
671 	if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
672 	    != NULL) {
673 		if (dir == PFIL_IN)
674 			ss->nskip_in_dejavu++;
675 		else
676 			ss->nskip_out_dejavu++;
677 
678 		return (1);
679 	} else {
680 		struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
681 		    PACKET_TAG_SIFTR, 0, M_NOWAIT);
682 		if (tag == NULL) {
683 			if (dir == PFIL_IN)
684 				ss->nskip_in_malloc++;
685 			else
686 				ss->nskip_out_malloc++;
687 
688 			return (1);
689 		}
690 
691 		m_tag_prepend(m, tag);
692 	}
693 
694 	return (0);
695 }
696 
697 
698 /*
699  * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
700  * otherwise.
701  */
702 static inline struct inpcb *
703 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
704     uint16_t dport, int dir, struct siftr_stats *ss)
705 {
706 	struct inpcb *inp;
707 
708 	/* We need the tcbinfo lock. */
709 	INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
710 
711 	if (dir == PFIL_IN)
712 		inp = (ipver == INP_IPV4 ?
713 		    in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
714 		    dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
715 		    :
716 #ifdef SIFTR_IPV6
717 		    in6_pcblookup(&V_tcbinfo,
718 		    &((struct ip6_hdr *)ip)->ip6_src, sport,
719 		    &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB,
720 		    m->m_pkthdr.rcvif)
721 #else
722 		    NULL
723 #endif
724 		    );
725 
726 	else
727 		inp = (ipver == INP_IPV4 ?
728 		    in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
729 		    sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
730 		    :
731 #ifdef SIFTR_IPV6
732 		    in6_pcblookup(&V_tcbinfo,
733 		    &((struct ip6_hdr *)ip)->ip6_dst, dport,
734 		    &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB,
735 		    m->m_pkthdr.rcvif)
736 #else
737 		    NULL
738 #endif
739 		    );
740 
741 	/* If we can't find the inpcb, bail. */
742 	if (inp == NULL) {
743 		if (dir == PFIL_IN)
744 			ss->nskip_in_inpcb++;
745 		else
746 			ss->nskip_out_inpcb++;
747 	}
748 
749 	return (inp);
750 }
751 
752 
753 static inline void
754 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
755     int ipver, int dir, int inp_locally_locked)
756 {
757 #ifdef SIFTR_IPV6
758 	if (ipver == INP_IPV4) {
759 		pn->ip_laddr[3] = inp->inp_laddr.s_addr;
760 		pn->ip_faddr[3] = inp->inp_faddr.s_addr;
761 #else
762 		*((uint32_t *)pn->ip_laddr) = inp->inp_laddr.s_addr;
763 		*((uint32_t *)pn->ip_faddr) = inp->inp_faddr.s_addr;
764 #endif
765 #ifdef SIFTR_IPV6
766 	} else {
767 		pn->ip_laddr[0] = inp->in6p_laddr.s6_addr32[0];
768 		pn->ip_laddr[1] = inp->in6p_laddr.s6_addr32[1];
769 		pn->ip_laddr[2] = inp->in6p_laddr.s6_addr32[2];
770 		pn->ip_laddr[3] = inp->in6p_laddr.s6_addr32[3];
771 		pn->ip_faddr[0] = inp->in6p_faddr.s6_addr32[0];
772 		pn->ip_faddr[1] = inp->in6p_faddr.s6_addr32[1];
773 		pn->ip_faddr[2] = inp->in6p_faddr.s6_addr32[2];
774 		pn->ip_faddr[3] = inp->in6p_faddr.s6_addr32[3];
775 	}
776 #endif
777 	pn->tcp_localport = inp->inp_lport;
778 	pn->tcp_foreignport = inp->inp_fport;
779 	pn->snd_cwnd = tp->snd_cwnd;
780 	pn->snd_wnd = tp->snd_wnd;
781 	pn->rcv_wnd = tp->rcv_wnd;
782 	pn->snd_bwnd = 0;		/* Unused, kept for compat. */
783 	pn->snd_ssthresh = tp->snd_ssthresh;
784 	pn->snd_scale = tp->snd_scale;
785 	pn->rcv_scale = tp->rcv_scale;
786 	pn->conn_state = tp->t_state;
787 	pn->max_seg_size = tp->t_maxseg;
788 	pn->smoothed_rtt = tp->t_srtt;
789 	pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
790 	pn->flags = tp->t_flags;
791 	pn->rxt_length = tp->t_rxtcur;
792 	pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
793 	pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd);
794 	pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
795 	pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv);
796 	pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
797 	pn->t_segqlen = tp->t_segqlen;
798 	pn->flowid = inp->inp_flowid;
799 	pn->flowtype = inp->inp_flowtype;
800 
801 	/* We've finished accessing the tcb so release the lock. */
802 	if (inp_locally_locked)
803 		INP_RUNLOCK(inp);
804 
805 	pn->ipver = ipver;
806 	pn->direction = dir;
807 
808 	/*
809 	 * Significantly more accurate than using getmicrotime(), but slower!
810 	 * Gives true microsecond resolution at the expense of a hit to
811 	 * maximum pps throughput processing when SIFTR is loaded and enabled.
812 	 */
813 	microtime(&pn->tval);
814 }
815 
816 
817 /*
818  * pfil hook that is called for each IPv4 packet making its way through the
819  * stack in either direction.
820  * The pfil subsystem holds a non-sleepable mutex somewhere when
821  * calling our hook function, so we can't sleep at all.
822  * It's very important to use the M_NOWAIT flag with all function calls
823  * that support it so that they won't sleep, otherwise you get a panic.
824  */
825 static int
826 siftr_chkpkt(void *arg, struct mbuf **m, struct ifnet *ifp, int dir,
827     struct inpcb *inp)
828 {
829 	struct pkt_node *pn;
830 	struct ip *ip;
831 	struct tcphdr *th;
832 	struct tcpcb *tp;
833 	struct siftr_stats *ss;
834 	unsigned int ip_hl;
835 	int inp_locally_locked;
836 
837 	inp_locally_locked = 0;
838 	ss = DPCPU_PTR(ss);
839 
840 	/*
841 	 * m_pullup is not required here because ip_{input|output}
842 	 * already do the heavy lifting for us.
843 	 */
844 
845 	ip = mtod(*m, struct ip *);
846 
847 	/* Only continue processing if the packet is TCP. */
848 	if (ip->ip_p != IPPROTO_TCP)
849 		goto ret;
850 
851 	/*
852 	 * If a kernel subsystem reinjects packets into the stack, our pfil
853 	 * hook will be called multiple times for the same packet.
854 	 * Make sure we only process unique packets.
855 	 */
856 	if (siftr_chkreinject(*m, dir, ss))
857 		goto ret;
858 
859 	if (dir == PFIL_IN)
860 		ss->n_in++;
861 	else
862 		ss->n_out++;
863 
864 	/*
865 	 * Create a tcphdr struct starting at the correct offset
866 	 * in the IP packet. ip->ip_hl gives the ip header length
867 	 * in 4-byte words, so multiply it to get the size in bytes.
868 	 */
869 	ip_hl = (ip->ip_hl << 2);
870 	th = (struct tcphdr *)((caddr_t)ip + ip_hl);
871 
872 	/*
873 	 * If the pfil hooks don't provide a pointer to the
874 	 * inpcb, we need to find it ourselves and lock it.
875 	 */
876 	if (!inp) {
877 		/* Find the corresponding inpcb for this pkt. */
878 		inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
879 		    th->th_dport, dir, ss);
880 
881 		if (inp == NULL)
882 			goto ret;
883 		else
884 			inp_locally_locked = 1;
885 	}
886 
887 	INP_LOCK_ASSERT(inp);
888 
889 	/* Find the TCP control block that corresponds with this packet */
890 	tp = intotcpcb(inp);
891 
892 	/*
893 	 * If we can't find the TCP control block (happens occasionaly for a
894 	 * packet sent during the shutdown phase of a TCP connection),
895 	 * or we're in the timewait state, bail
896 	 */
897 	if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
898 		if (dir == PFIL_IN)
899 			ss->nskip_in_tcpcb++;
900 		else
901 			ss->nskip_out_tcpcb++;
902 
903 		goto inp_unlock;
904 	}
905 
906 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
907 
908 	if (pn == NULL) {
909 		if (dir == PFIL_IN)
910 			ss->nskip_in_malloc++;
911 		else
912 			ss->nskip_out_malloc++;
913 
914 		goto inp_unlock;
915 	}
916 
917 	siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
918 
919 	if (siftr_generate_hashes) {
920 		if ((*m)->m_pkthdr.csum_flags & CSUM_TCP) {
921 			/*
922 			 * For outbound packets, the TCP checksum isn't
923 			 * calculated yet. This is a problem for our packet
924 			 * hashing as the receiver will calc a different hash
925 			 * to ours if we don't include the correct TCP checksum
926 			 * in the bytes being hashed. To work around this
927 			 * problem, we manually calc the TCP checksum here in
928 			 * software. We unset the CSUM_TCP flag so the lower
929 			 * layers don't recalc it.
930 			 */
931 			(*m)->m_pkthdr.csum_flags &= ~CSUM_TCP;
932 
933 			/*
934 			 * Calculate the TCP checksum in software and assign
935 			 * to correct TCP header field, which will follow the
936 			 * packet mbuf down the stack. The trick here is that
937 			 * tcp_output() sets th->th_sum to the checksum of the
938 			 * pseudo header for us already. Because of the nature
939 			 * of the checksumming algorithm, we can sum over the
940 			 * entire IP payload (i.e. TCP header and data), which
941 			 * will include the already calculated pseduo header
942 			 * checksum, thus giving us the complete TCP checksum.
943 			 *
944 			 * To put it in simple terms, if checksum(1,2,3,4)=10,
945 			 * then checksum(1,2,3,4,5) == checksum(10,5).
946 			 * This property is what allows us to "cheat" and
947 			 * checksum only the IP payload which has the TCP
948 			 * th_sum field populated with the pseudo header's
949 			 * checksum, and not need to futz around checksumming
950 			 * pseudo header bytes and TCP header/data in one hit.
951 			 * Refer to RFC 1071 for more info.
952 			 *
953 			 * NB: in_cksum_skip(struct mbuf *m, int len, int skip)
954 			 * in_cksum_skip 2nd argument is NOT the number of
955 			 * bytes to read from the mbuf at "skip" bytes offset
956 			 * from the start of the mbuf (very counter intuitive!).
957 			 * The number of bytes to read is calculated internally
958 			 * by the function as len-skip i.e. to sum over the IP
959 			 * payload (TCP header + data) bytes, it is INCORRECT
960 			 * to call the function like this:
961 			 * in_cksum_skip(at, ip->ip_len - offset, offset)
962 			 * Rather, it should be called like this:
963 			 * in_cksum_skip(at, ip->ip_len, offset)
964 			 * which means read "ip->ip_len - offset" bytes from
965 			 * the mbuf cluster "at" at offset "offset" bytes from
966 			 * the beginning of the "at" mbuf's data pointer.
967 			 */
968 			th->th_sum = in_cksum_skip(*m, ntohs(ip->ip_len),
969 			    ip_hl);
970 		}
971 
972 		/*
973 		 * XXX: Having to calculate the checksum in software and then
974 		 * hash over all bytes is really inefficient. Would be nice to
975 		 * find a way to create the hash and checksum in the same pass
976 		 * over the bytes.
977 		 */
978 		pn->hash = hash_pkt(*m, ip_hl);
979 	}
980 
981 	mtx_lock(&siftr_pkt_queue_mtx);
982 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
983 	mtx_unlock(&siftr_pkt_queue_mtx);
984 	goto ret;
985 
986 inp_unlock:
987 	if (inp_locally_locked)
988 		INP_RUNLOCK(inp);
989 
990 ret:
991 	/* Returning 0 ensures pfil will not discard the pkt */
992 	return (0);
993 }
994 
995 
996 #ifdef SIFTR_IPV6
997 static int
998 siftr_chkpkt6(void *arg, struct mbuf **m, struct ifnet *ifp, int dir,
999     struct inpcb *inp)
1000 {
1001 	struct pkt_node *pn;
1002 	struct ip6_hdr *ip6;
1003 	struct tcphdr *th;
1004 	struct tcpcb *tp;
1005 	struct siftr_stats *ss;
1006 	unsigned int ip6_hl;
1007 	int inp_locally_locked;
1008 
1009 	inp_locally_locked = 0;
1010 	ss = DPCPU_PTR(ss);
1011 
1012 	/*
1013 	 * m_pullup is not required here because ip6_{input|output}
1014 	 * already do the heavy lifting for us.
1015 	 */
1016 
1017 	ip6 = mtod(*m, struct ip6_hdr *);
1018 
1019 	/*
1020 	 * Only continue processing if the packet is TCP
1021 	 * XXX: We should follow the next header fields
1022 	 * as shown on Pg 6 RFC 2460, but right now we'll
1023 	 * only check pkts that have no extension headers.
1024 	 */
1025 	if (ip6->ip6_nxt != IPPROTO_TCP)
1026 		goto ret6;
1027 
1028 	/*
1029 	 * If a kernel subsystem reinjects packets into the stack, our pfil
1030 	 * hook will be called multiple times for the same packet.
1031 	 * Make sure we only process unique packets.
1032 	 */
1033 	if (siftr_chkreinject(*m, dir, ss))
1034 		goto ret6;
1035 
1036 	if (dir == PFIL_IN)
1037 		ss->n_in++;
1038 	else
1039 		ss->n_out++;
1040 
1041 	ip6_hl = sizeof(struct ip6_hdr);
1042 
1043 	/*
1044 	 * Create a tcphdr struct starting at the correct offset
1045 	 * in the ipv6 packet. ip->ip_hl gives the ip header length
1046 	 * in 4-byte words, so multiply it to get the size in bytes.
1047 	 */
1048 	th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
1049 
1050 	/*
1051 	 * For inbound packets, the pfil hooks don't provide a pointer to the
1052 	 * inpcb, so we need to find it ourselves and lock it.
1053 	 */
1054 	if (!inp) {
1055 		/* Find the corresponding inpcb for this pkt. */
1056 		inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
1057 		    th->th_sport, th->th_dport, dir, ss);
1058 
1059 		if (inp == NULL)
1060 			goto ret6;
1061 		else
1062 			inp_locally_locked = 1;
1063 	}
1064 
1065 	/* Find the TCP control block that corresponds with this packet. */
1066 	tp = intotcpcb(inp);
1067 
1068 	/*
1069 	 * If we can't find the TCP control block (happens occasionaly for a
1070 	 * packet sent during the shutdown phase of a TCP connection),
1071 	 * or we're in the timewait state, bail.
1072 	 */
1073 	if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
1074 		if (dir == PFIL_IN)
1075 			ss->nskip_in_tcpcb++;
1076 		else
1077 			ss->nskip_out_tcpcb++;
1078 
1079 		goto inp_unlock6;
1080 	}
1081 
1082 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
1083 
1084 	if (pn == NULL) {
1085 		if (dir == PFIL_IN)
1086 			ss->nskip_in_malloc++;
1087 		else
1088 			ss->nskip_out_malloc++;
1089 
1090 		goto inp_unlock6;
1091 	}
1092 
1093 	siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
1094 
1095 	/* XXX: Figure out how to generate hashes for IPv6 packets. */
1096 
1097 	mtx_lock(&siftr_pkt_queue_mtx);
1098 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
1099 	mtx_unlock(&siftr_pkt_queue_mtx);
1100 	goto ret6;
1101 
1102 inp_unlock6:
1103 	if (inp_locally_locked)
1104 		INP_RUNLOCK(inp);
1105 
1106 ret6:
1107 	/* Returning 0 ensures pfil will not discard the pkt. */
1108 	return (0);
1109 }
1110 #endif /* #ifdef SIFTR_IPV6 */
1111 
1112 
1113 static int
1114 siftr_pfil(int action)
1115 {
1116 	struct pfil_head *pfh_inet;
1117 #ifdef SIFTR_IPV6
1118 	struct pfil_head *pfh_inet6;
1119 #endif
1120 	VNET_ITERATOR_DECL(vnet_iter);
1121 
1122 	VNET_LIST_RLOCK();
1123 	VNET_FOREACH(vnet_iter) {
1124 		CURVNET_SET(vnet_iter);
1125 		pfh_inet = pfil_head_get(PFIL_TYPE_AF, AF_INET);
1126 #ifdef SIFTR_IPV6
1127 		pfh_inet6 = pfil_head_get(PFIL_TYPE_AF, AF_INET6);
1128 #endif
1129 
1130 		if (action == HOOK) {
1131 			pfil_add_hook(siftr_chkpkt, NULL,
1132 			    PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet);
1133 #ifdef SIFTR_IPV6
1134 			pfil_add_hook(siftr_chkpkt6, NULL,
1135 			    PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6);
1136 #endif
1137 		} else if (action == UNHOOK) {
1138 			pfil_remove_hook(siftr_chkpkt, NULL,
1139 			    PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet);
1140 #ifdef SIFTR_IPV6
1141 			pfil_remove_hook(siftr_chkpkt6, NULL,
1142 			    PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6);
1143 #endif
1144 		}
1145 		CURVNET_RESTORE();
1146 	}
1147 	VNET_LIST_RUNLOCK();
1148 
1149 	return (0);
1150 }
1151 
1152 
1153 static int
1154 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1155 {
1156 	struct alq *new_alq;
1157 	int error;
1158 
1159 	error = sysctl_handle_string(oidp, arg1, arg2, req);
1160 
1161 	/* Check for error or same filename */
1162 	if (error != 0 || req->newptr == NULL ||
1163 	    strncmp(siftr_logfile, arg1, arg2) == 0)
1164 		goto done;
1165 
1166 	/* Filname changed */
1167 	error = alq_open(&new_alq, arg1, curthread->td_ucred,
1168 	    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1169 	if (error != 0)
1170 		goto done;
1171 
1172 	/*
1173 	 * If disabled, siftr_alq == NULL so we simply close
1174 	 * the alq as we've proved it can be opened.
1175 	 * If enabled, close the existing alq and switch the old
1176 	 * for the new.
1177 	 */
1178 	if (siftr_alq == NULL) {
1179 		alq_close(new_alq);
1180 	} else {
1181 		alq_close(siftr_alq);
1182 		siftr_alq = new_alq;
1183 	}
1184 
1185 	/* Update filename upon success */
1186 	strlcpy(siftr_logfile, arg1, arg2);
1187 done:
1188 	return (error);
1189 }
1190 
1191 static int
1192 siftr_manage_ops(uint8_t action)
1193 {
1194 	struct siftr_stats totalss;
1195 	struct timeval tval;
1196 	struct flow_hash_node *counter, *tmp_counter;
1197 	struct sbuf *s;
1198 	int i, key_index, ret, error;
1199 	uint32_t bytes_to_write, total_skipped_pkts;
1200 	uint16_t lport, fport;
1201 	uint8_t *key, ipver;
1202 
1203 #ifdef SIFTR_IPV6
1204 	uint32_t laddr[4];
1205 	uint32_t faddr[4];
1206 #else
1207 	uint8_t laddr[4];
1208 	uint8_t faddr[4];
1209 #endif
1210 
1211 	error = 0;
1212 	total_skipped_pkts = 0;
1213 
1214 	/* Init an autosizing sbuf that initially holds 200 chars. */
1215 	if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1216 		return (-1);
1217 
1218 	if (action == SIFTR_ENABLE) {
1219 		/*
1220 		 * Create our alq
1221 		 * XXX: We should abort if alq_open fails!
1222 		 */
1223 		alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1224 		    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1225 
1226 		STAILQ_INIT(&pkt_queue);
1227 
1228 		DPCPU_ZERO(ss);
1229 
1230 		siftr_exit_pkt_manager_thread = 0;
1231 
1232 		ret = kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1233 		    &siftr_pkt_manager_thr, RFNOWAIT, 0,
1234 		    "siftr_pkt_manager_thr");
1235 
1236 		siftr_pfil(HOOK);
1237 
1238 		microtime(&tval);
1239 
1240 		sbuf_printf(s,
1241 		    "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1242 		    "siftrver=%s\thz=%u\ttcp_rtt_scale=%u\tsysname=%s\t"
1243 		    "sysver=%u\tipmode=%u\n",
1244 		    (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, hz,
1245 		    TCP_RTT_SCALE, SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1246 
1247 		sbuf_finish(s);
1248 		alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1249 
1250 	} else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1251 		/*
1252 		 * Remove the pfil hook functions. All threads currently in
1253 		 * the hook functions are allowed to exit before siftr_pfil()
1254 		 * returns.
1255 		 */
1256 		siftr_pfil(UNHOOK);
1257 
1258 		/* This will block until the pkt manager thread unlocks it. */
1259 		mtx_lock(&siftr_pkt_mgr_mtx);
1260 
1261 		/* Tell the pkt manager thread that it should exit now. */
1262 		siftr_exit_pkt_manager_thread = 1;
1263 
1264 		/*
1265 		 * Wake the pkt_manager thread so it realises that
1266 		 * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1267 		 * The wakeup won't be delivered until we unlock
1268 		 * siftr_pkt_mgr_mtx so this isn't racy.
1269 		 */
1270 		wakeup(&wait_for_pkt);
1271 
1272 		/* Wait for the pkt_manager thread to exit. */
1273 		mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1274 		    "thrwait", 0);
1275 
1276 		siftr_pkt_manager_thr = NULL;
1277 		mtx_unlock(&siftr_pkt_mgr_mtx);
1278 
1279 		totalss.n_in = DPCPU_VARSUM(ss, n_in);
1280 		totalss.n_out = DPCPU_VARSUM(ss, n_out);
1281 		totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1282 		totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1283 		totalss.nskip_in_mtx = DPCPU_VARSUM(ss, nskip_in_mtx);
1284 		totalss.nskip_out_mtx = DPCPU_VARSUM(ss, nskip_out_mtx);
1285 		totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1286 		totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1287 		totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1288 		totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1289 
1290 		total_skipped_pkts = totalss.nskip_in_malloc +
1291 		    totalss.nskip_out_malloc + totalss.nskip_in_mtx +
1292 		    totalss.nskip_out_mtx + totalss.nskip_in_tcpcb +
1293 		    totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1294 		    totalss.nskip_out_inpcb;
1295 
1296 		microtime(&tval);
1297 
1298 		sbuf_printf(s,
1299 		    "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1300 		    "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1301 		    "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1302 		    "num_outbound_skipped_pkts_malloc=%u\t"
1303 		    "num_inbound_skipped_pkts_mtx=%u\t"
1304 		    "num_outbound_skipped_pkts_mtx=%u\t"
1305 		    "num_inbound_skipped_pkts_tcpcb=%u\t"
1306 		    "num_outbound_skipped_pkts_tcpcb=%u\t"
1307 		    "num_inbound_skipped_pkts_inpcb=%u\t"
1308 		    "num_outbound_skipped_pkts_inpcb=%u\t"
1309 		    "total_skipped_tcp_pkts=%u\tflow_list=",
1310 		    (intmax_t)tval.tv_sec,
1311 		    tval.tv_usec,
1312 		    (uintmax_t)totalss.n_in,
1313 		    (uintmax_t)totalss.n_out,
1314 		    (uintmax_t)(totalss.n_in + totalss.n_out),
1315 		    totalss.nskip_in_malloc,
1316 		    totalss.nskip_out_malloc,
1317 		    totalss.nskip_in_mtx,
1318 		    totalss.nskip_out_mtx,
1319 		    totalss.nskip_in_tcpcb,
1320 		    totalss.nskip_out_tcpcb,
1321 		    totalss.nskip_in_inpcb,
1322 		    totalss.nskip_out_inpcb,
1323 		    total_skipped_pkts);
1324 
1325 		/*
1326 		 * Iterate over the flow hash, printing a summary of each
1327 		 * flow seen and freeing any malloc'd memory.
1328 		 * The hash consists of an array of LISTs (man 3 queue).
1329 		 */
1330 		for (i = 0; i <= siftr_hashmask; i++) {
1331 			LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1332 			    tmp_counter) {
1333 				key = counter->key;
1334 				key_index = 1;
1335 
1336 				ipver = key[0];
1337 
1338 				memcpy(laddr, key + key_index, sizeof(laddr));
1339 				key_index += sizeof(laddr);
1340 				memcpy(&lport, key + key_index, sizeof(lport));
1341 				key_index += sizeof(lport);
1342 				memcpy(faddr, key + key_index, sizeof(faddr));
1343 				key_index += sizeof(faddr);
1344 				memcpy(&fport, key + key_index, sizeof(fport));
1345 
1346 #ifdef SIFTR_IPV6
1347 				laddr[3] = ntohl(laddr[3]);
1348 				faddr[3] = ntohl(faddr[3]);
1349 
1350 				if (ipver == INP_IPV6) {
1351 					laddr[0] = ntohl(laddr[0]);
1352 					laddr[1] = ntohl(laddr[1]);
1353 					laddr[2] = ntohl(laddr[2]);
1354 					faddr[0] = ntohl(faddr[0]);
1355 					faddr[1] = ntohl(faddr[1]);
1356 					faddr[2] = ntohl(faddr[2]);
1357 
1358 					sbuf_printf(s,
1359 					    "%x:%x:%x:%x:%x:%x:%x:%x;%u-"
1360 					    "%x:%x:%x:%x:%x:%x:%x:%x;%u,",
1361 					    UPPER_SHORT(laddr[0]),
1362 					    LOWER_SHORT(laddr[0]),
1363 					    UPPER_SHORT(laddr[1]),
1364 					    LOWER_SHORT(laddr[1]),
1365 					    UPPER_SHORT(laddr[2]),
1366 					    LOWER_SHORT(laddr[2]),
1367 					    UPPER_SHORT(laddr[3]),
1368 					    LOWER_SHORT(laddr[3]),
1369 					    ntohs(lport),
1370 					    UPPER_SHORT(faddr[0]),
1371 					    LOWER_SHORT(faddr[0]),
1372 					    UPPER_SHORT(faddr[1]),
1373 					    LOWER_SHORT(faddr[1]),
1374 					    UPPER_SHORT(faddr[2]),
1375 					    LOWER_SHORT(faddr[2]),
1376 					    UPPER_SHORT(faddr[3]),
1377 					    LOWER_SHORT(faddr[3]),
1378 					    ntohs(fport));
1379 				} else {
1380 					laddr[0] = FIRST_OCTET(laddr[3]);
1381 					laddr[1] = SECOND_OCTET(laddr[3]);
1382 					laddr[2] = THIRD_OCTET(laddr[3]);
1383 					laddr[3] = FOURTH_OCTET(laddr[3]);
1384 					faddr[0] = FIRST_OCTET(faddr[3]);
1385 					faddr[1] = SECOND_OCTET(faddr[3]);
1386 					faddr[2] = THIRD_OCTET(faddr[3]);
1387 					faddr[3] = FOURTH_OCTET(faddr[3]);
1388 #endif
1389 					sbuf_printf(s,
1390 					    "%u.%u.%u.%u;%u-%u.%u.%u.%u;%u,",
1391 					    laddr[0],
1392 					    laddr[1],
1393 					    laddr[2],
1394 					    laddr[3],
1395 					    ntohs(lport),
1396 					    faddr[0],
1397 					    faddr[1],
1398 					    faddr[2],
1399 					    faddr[3],
1400 					    ntohs(fport));
1401 #ifdef SIFTR_IPV6
1402 				}
1403 #endif
1404 
1405 				free(counter, M_SIFTR_HASHNODE);
1406 			}
1407 
1408 			LIST_INIT(counter_hash + i);
1409 		}
1410 
1411 		sbuf_printf(s, "\n");
1412 		sbuf_finish(s);
1413 
1414 		i = 0;
1415 		do {
1416 			bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1417 			alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1418 			i += bytes_to_write;
1419 		} while (i < sbuf_len(s));
1420 
1421 		alq_close(siftr_alq);
1422 		siftr_alq = NULL;
1423 	}
1424 
1425 	sbuf_delete(s);
1426 
1427 	/*
1428 	 * XXX: Should be using ret to check if any functions fail
1429 	 * and set error appropriately
1430 	 */
1431 
1432 	return (error);
1433 }
1434 
1435 
1436 static int
1437 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1438 {
1439 	if (req->newptr == NULL)
1440 		goto skip;
1441 
1442 	/* If the value passed in isn't 0 or 1, return an error. */
1443 	if (CAST_PTR_INT(req->newptr) != 0 && CAST_PTR_INT(req->newptr) != 1)
1444 		return (1);
1445 
1446 	/* If we are changing state (0 to 1 or 1 to 0). */
1447 	if (CAST_PTR_INT(req->newptr) != siftr_enabled )
1448 		if (siftr_manage_ops(CAST_PTR_INT(req->newptr))) {
1449 			siftr_manage_ops(SIFTR_DISABLE);
1450 			return (1);
1451 		}
1452 
1453 skip:
1454 	return (sysctl_handle_int(oidp, arg1, arg2, req));
1455 }
1456 
1457 
1458 static void
1459 siftr_shutdown_handler(void *arg)
1460 {
1461 	siftr_manage_ops(SIFTR_DISABLE);
1462 }
1463 
1464 
1465 /*
1466  * Module is being unloaded or machine is shutting down. Take care of cleanup.
1467  */
1468 static int
1469 deinit_siftr(void)
1470 {
1471 	/* Cleanup. */
1472 	siftr_manage_ops(SIFTR_DISABLE);
1473 	hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1474 	mtx_destroy(&siftr_pkt_queue_mtx);
1475 	mtx_destroy(&siftr_pkt_mgr_mtx);
1476 
1477 	return (0);
1478 }
1479 
1480 
1481 /*
1482  * Module has just been loaded into the kernel.
1483  */
1484 static int
1485 init_siftr(void)
1486 {
1487 	EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL,
1488 	    SHUTDOWN_PRI_FIRST);
1489 
1490 	/* Initialise our flow counter hash table. */
1491 	counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1492 	    &siftr_hashmask);
1493 
1494 	mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1495 	mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1496 
1497 	/* Print message to the user's current terminal. */
1498 	uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1499 	    "          http://caia.swin.edu.au/urp/newtcp\n\n",
1500 	    MODVERSION_STR);
1501 
1502 	return (0);
1503 }
1504 
1505 
1506 /*
1507  * This is the function that is called to load and unload the module.
1508  * When the module is loaded, this function is called once with
1509  * "what" == MOD_LOAD
1510  * When the module is unloaded, this function is called twice with
1511  * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1512  * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1513  * this function is called once with "what" = MOD_SHUTDOWN
1514  * When the system is shut down, the handler isn't called until the very end
1515  * of the shutdown sequence i.e. after the disks have been synced.
1516  */
1517 static int
1518 siftr_load_handler(module_t mod, int what, void *arg)
1519 {
1520 	int ret;
1521 
1522 	switch (what) {
1523 	case MOD_LOAD:
1524 		ret = init_siftr();
1525 		break;
1526 
1527 	case MOD_QUIESCE:
1528 	case MOD_SHUTDOWN:
1529 		ret = deinit_siftr();
1530 		break;
1531 
1532 	case MOD_UNLOAD:
1533 		ret = 0;
1534 		break;
1535 
1536 	default:
1537 		ret = EINVAL;
1538 		break;
1539 	}
1540 
1541 	return (ret);
1542 }
1543 
1544 
1545 static moduledata_t siftr_mod = {
1546 	.name = "siftr",
1547 	.evhand = siftr_load_handler,
1548 };
1549 
1550 /*
1551  * Param 1: name of the kernel module
1552  * Param 2: moduledata_t struct containing info about the kernel module
1553  *          and the execution entry point for the module
1554  * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1555  *          Defines the module initialisation order
1556  * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1557  *          Defines the initialisation order of this kld relative to others
1558  *          within the same subsystem as defined by param 3
1559  */
1560 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_SMP, SI_ORDER_ANY);
1561 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1562 MODULE_VERSION(siftr, MODVERSION);
1563