xref: /freebsd/sys/netinet/siftr.c (revision cded07a8783dedbb0634ebde78764f935f26467d)
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's Centre for
42  * Advanced Internet Architectures, Melbourne, Australia, which was made
43  * possible in part by a grant from the Cisco University Research Program Fund
44  * at Community Foundation Silicon Valley. More details are available at:
45  *   http://caia.swin.edu.au/urp/newtcp/
46  *
47  * Work on SIFTR v1.2.x was sponsored by the FreeBSD Foundation as part of
48  * the "Enhancing the FreeBSD TCP Implementation" project 2008-2009.
49  * More details are available at:
50  *   http://www.freebsdfoundation.org/
51  *   http://caia.swin.edu.au/freebsd/etcp09/
52  *
53  * Lawrence Stewart is the current maintainer, and all contact regarding
54  * SIFTR should be directed to him via email: lastewart@swin.edu.au
55  *
56  * Initial release date: June 2007
57  * Most recent update: September 2010
58  ******************************************************/
59 
60 #include <sys/cdefs.h>
61 __FBSDID("$FreeBSD$");
62 
63 #include <sys/param.h>
64 #include <sys/alq.h>
65 #include <sys/errno.h>
66 #include <sys/hash.h>
67 #include <sys/kernel.h>
68 #include <sys/kthread.h>
69 #include <sys/lock.h>
70 #include <sys/mbuf.h>
71 #include <sys/module.h>
72 #include <sys/mutex.h>
73 #include <sys/pcpu.h>
74 #include <sys/proc.h>
75 #include <sys/sbuf.h>
76 #include <sys/smp.h>
77 #include <sys/socket.h>
78 #include <sys/socketvar.h>
79 #include <sys/sysctl.h>
80 #include <sys/unistd.h>
81 
82 #include <net/if.h>
83 #include <net/pfil.h>
84 
85 #include <netinet/in.h>
86 #include <netinet/in_pcb.h>
87 #include <netinet/in_systm.h>
88 #include <netinet/in_var.h>
89 #include <netinet/ip.h>
90 #include <netinet/tcp_var.h>
91 
92 #ifdef SIFTR_IPV6
93 #include <netinet/ip6.h>
94 #include <netinet6/in6_pcb.h>
95 #endif /* SIFTR_IPV6 */
96 
97 #include <machine/in_cksum.h>
98 
99 /*
100  * Three digit version number refers to X.Y.Z where:
101  * X is the major version number
102  * Y is bumped to mark backwards incompatible changes
103  * Z is bumped to mark backwards compatible changes
104  */
105 #define V_MAJOR		1
106 #define V_BACKBREAK	2
107 #define V_BACKCOMPAT	4
108 #define MODVERSION	__CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT))
109 #define MODVERSION_STR	__XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \
110     __XSTRING(V_BACKCOMPAT)
111 
112 #define HOOK 0
113 #define UNHOOK 1
114 #define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536
115 #define SYS_NAME "FreeBSD"
116 #define PACKET_TAG_SIFTR 100
117 #define PACKET_COOKIE_SIFTR 21749576
118 #define SIFTR_LOG_FILE_MODE 0644
119 #define SIFTR_DISABLE 0
120 #define SIFTR_ENABLE 1
121 
122 /*
123  * Hard upper limit on the length of log messages. Bump this up if you add new
124  * data fields such that the line length could exceed the below value.
125  */
126 #define MAX_LOG_MSG_LEN 200
127 /* XXX: Make this a sysctl tunable. */
128 #define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN)
129 
130 /*
131  * 1 byte for IP version
132  * IPv4: src/dst IP (4+4) + src/dst port (2+2) = 12 bytes
133  * IPv6: src/dst IP (16+16) + src/dst port (2+2) = 36 bytes
134  */
135 #ifdef SIFTR_IPV6
136 #define FLOW_KEY_LEN 37
137 #else
138 #define FLOW_KEY_LEN 13
139 #endif
140 
141 #ifdef SIFTR_IPV6
142 #define SIFTR_IPMODE 6
143 #else
144 #define SIFTR_IPMODE 4
145 #endif
146 
147 /* useful macros */
148 #define CAST_PTR_INT(X) (*((int*)(X)))
149 
150 #define UPPER_SHORT(X)	(((X) & 0xFFFF0000) >> 16)
151 #define LOWER_SHORT(X)	((X) & 0x0000FFFF)
152 
153 #define FIRST_OCTET(X)	(((X) & 0xFF000000) >> 24)
154 #define SECOND_OCTET(X)	(((X) & 0x00FF0000) >> 16)
155 #define THIRD_OCTET(X)	(((X) & 0x0000FF00) >> 8)
156 #define FOURTH_OCTET(X)	((X) & 0x000000FF)
157 
158 MALLOC_DECLARE(M_SIFTR);
159 MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR");
160 
161 MALLOC_DECLARE(M_SIFTR_PKTNODE);
162 MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode", "SIFTR pkt_node struct");
163 
164 MALLOC_DECLARE(M_SIFTR_HASHNODE);
165 MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode", "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 	/* Link to next pkt_node in the list. */
231 	STAILQ_ENTRY(pkt_node)	nodes;
232 };
233 
234 struct flow_hash_node
235 {
236 	uint16_t counter;
237 	uint8_t key[FLOW_KEY_LEN];
238 	LIST_ENTRY(flow_hash_node) nodes;
239 };
240 
241 struct siftr_stats
242 {
243 	/* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */
244 	uint64_t n_in;
245 	uint64_t n_out;
246 	/* # pkts skipped due to failed malloc calls. */
247 	uint32_t nskip_in_malloc;
248 	uint32_t nskip_out_malloc;
249 	/* # pkts skipped due to failed mtx acquisition. */
250 	uint32_t nskip_in_mtx;
251 	uint32_t nskip_out_mtx;
252 	/* # pkts skipped due to failed inpcb lookups. */
253 	uint32_t nskip_in_inpcb;
254 	uint32_t nskip_out_inpcb;
255 	/* # pkts skipped due to failed tcpcb lookups. */
256 	uint32_t nskip_in_tcpcb;
257 	uint32_t nskip_out_tcpcb;
258 	/* # pkts skipped due to stack reinjection. */
259 	uint32_t nskip_in_dejavu;
260 	uint32_t nskip_out_dejavu;
261 };
262 
263 static DPCPU_DEFINE(struct siftr_stats, ss);
264 
265 static volatile unsigned int siftr_exit_pkt_manager_thread = 0;
266 static unsigned int siftr_enabled = 0;
267 static unsigned int siftr_pkts_per_log = 1;
268 static unsigned int siftr_generate_hashes = 0;
269 /* static unsigned int siftr_binary_log = 0; */
270 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log";
271 static u_long siftr_hashmask;
272 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue);
273 LIST_HEAD(listhead, flow_hash_node) *counter_hash;
274 static int wait_for_pkt;
275 static struct alq *siftr_alq = NULL;
276 static struct mtx siftr_pkt_queue_mtx;
277 static struct mtx siftr_pkt_mgr_mtx;
278 static struct thread *siftr_pkt_manager_thr = NULL;
279 /*
280  * pfil.h defines PFIL_IN as 1 and PFIL_OUT as 2,
281  * which we use as an index into this array.
282  */
283 static char direction[3] = {'\0', 'i','o'};
284 
285 /* Required function prototypes. */
286 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS);
287 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS);
288 
289 
290 /* Declare the net.inet.siftr sysctl tree and populate it. */
291 
292 SYSCTL_DECL(_net_inet_siftr);
293 
294 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW, NULL,
295     "siftr related settings");
296 
297 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, CTLTYPE_UINT|CTLFLAG_RW,
298     &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU",
299     "switch siftr module operations on/off");
300 
301 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, CTLTYPE_STRING|CTLFLAG_RW,
302     &siftr_logfile, sizeof(siftr_logfile), &siftr_sysctl_logfile_name_handler,
303     "A", "file to save siftr log messages to");
304 
305 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW,
306     &siftr_pkts_per_log, 1,
307     "number of packets between generating a log message");
308 
309 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, genhashes, CTLFLAG_RW,
310     &siftr_generate_hashes, 0,
311     "enable packet hash generation");
312 
313 /* XXX: TODO
314 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW,
315     &siftr_binary_log, 0,
316     "write log files in binary instead of ascii");
317 */
318 
319 
320 /* Begin functions. */
321 
322 static void
323 siftr_process_pkt(struct pkt_node * pkt_node)
324 {
325 	struct flow_hash_node *hash_node;
326 	struct listhead *counter_list;
327 	struct siftr_stats *ss;
328 	struct ale *log_buf;
329 	uint8_t key[FLOW_KEY_LEN];
330 	uint8_t found_match, key_offset;
331 
332 	hash_node = NULL;
333 	ss = DPCPU_PTR(ss);
334 	found_match = 0;
335 	key_offset = 1;
336 
337 	/*
338 	 * Create the key that will be used to create a hash index
339 	 * into our hash table. Our key consists of:
340 	 * ipversion, localip, localport, foreignip, foreignport
341 	 */
342 	key[0] = pkt_node->ipver;
343 	memcpy(key + key_offset, &pkt_node->ip_laddr,
344 	    sizeof(pkt_node->ip_laddr));
345 	key_offset += sizeof(pkt_node->ip_laddr);
346 	memcpy(key + key_offset, &pkt_node->tcp_localport,
347 	    sizeof(pkt_node->tcp_localport));
348 	key_offset += sizeof(pkt_node->tcp_localport);
349 	memcpy(key + key_offset, &pkt_node->ip_faddr,
350 	    sizeof(pkt_node->ip_faddr));
351 	key_offset += sizeof(pkt_node->ip_faddr);
352 	memcpy(key + key_offset, &pkt_node->tcp_foreignport,
353 	    sizeof(pkt_node->tcp_foreignport));
354 
355 	counter_list = counter_hash +
356 	    (hash32_buf(key, sizeof(key), 0) & siftr_hashmask);
357 
358 	/*
359 	 * If the list is not empty i.e. the hash index has
360 	 * been used by another flow previously.
361 	 */
362 	if (LIST_FIRST(counter_list) != NULL) {
363 		/*
364 		 * Loop through the hash nodes in the list.
365 		 * There should normally only be 1 hash node in the list,
366 		 * except if there have been collisions at the hash index
367 		 * computed by hash32_buf().
368 		 */
369 		LIST_FOREACH(hash_node, counter_list, nodes) {
370 			/*
371 			 * Check if the key for the pkt we are currently
372 			 * processing is the same as the key stored in the
373 			 * hash node we are currently processing.
374 			 * If they are the same, then we've found the
375 			 * hash node that stores the counter for the flow
376 			 * the pkt belongs to.
377 			 */
378 			if (memcmp(hash_node->key, key, sizeof(key)) == 0) {
379 				found_match = 1;
380 				break;
381 			}
382 		}
383 	}
384 
385 	/* If this flow hash hasn't been seen before or we have a collision. */
386 	if (hash_node == NULL || !found_match) {
387 		/* Create a new hash node to store the flow's counter. */
388 		hash_node = malloc(sizeof(struct flow_hash_node),
389 		    M_SIFTR_HASHNODE, M_WAITOK);
390 
391 		if (hash_node != NULL) {
392 			/* Initialise our new hash node list entry. */
393 			hash_node->counter = 0;
394 			memcpy(hash_node->key, key, sizeof(key));
395 			LIST_INSERT_HEAD(counter_list, hash_node, nodes);
396 		} else {
397 			/* Malloc failed. */
398 			if (pkt_node->direction == PFIL_IN)
399 				ss->nskip_in_malloc++;
400 			else
401 				ss->nskip_out_malloc++;
402 
403 			return;
404 		}
405 	} else if (siftr_pkts_per_log > 1) {
406 		/*
407 		 * Taking the remainder of the counter divided
408 		 * by the current value of siftr_pkts_per_log
409 		 * and storing that in counter provides a neat
410 		 * way to modulate the frequency of log
411 		 * messages being written to the log file.
412 		 */
413 		hash_node->counter = (hash_node->counter + 1) %
414 		    siftr_pkts_per_log;
415 
416 		/*
417 		 * If we have not seen enough packets since the last time
418 		 * we wrote a log message for this connection, return.
419 		 */
420 		if (hash_node->counter > 0)
421 			return;
422 	}
423 
424 	log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK);
425 
426 	if (log_buf == NULL)
427 		return; /* Should only happen if the ALQ is shutting down. */
428 
429 #ifdef SIFTR_IPV6
430 	pkt_node->ip_laddr[3] = ntohl(pkt_node->ip_laddr[3]);
431 	pkt_node->ip_faddr[3] = ntohl(pkt_node->ip_faddr[3]);
432 
433 	if (pkt_node->ipver == INP_IPV6) { /* IPv6 packet */
434 		pkt_node->ip_laddr[0] = ntohl(pkt_node->ip_laddr[0]);
435 		pkt_node->ip_laddr[1] = ntohl(pkt_node->ip_laddr[1]);
436 		pkt_node->ip_laddr[2] = ntohl(pkt_node->ip_laddr[2]);
437 		pkt_node->ip_faddr[0] = ntohl(pkt_node->ip_faddr[0]);
438 		pkt_node->ip_faddr[1] = ntohl(pkt_node->ip_faddr[1]);
439 		pkt_node->ip_faddr[2] = ntohl(pkt_node->ip_faddr[2]);
440 
441 		/* Construct an IPv6 log message. */
442 		log_buf->ae_bytesused = snprintf(log_buf->ae_data,
443 		    MAX_LOG_MSG_LEN,
444 		    "%c,0x%08x,%zd.%06ld,%x:%x:%x:%x:%x:%x:%x:%x,%u,%x:%x:%x:"
445 		    "%x:%x:%x:%x:%x,%u,%ld,%ld,%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,"
446 		    "%u,%d,%u,%u,%u,%u,%u,%u\n",
447 		    direction[pkt_node->direction],
448 		    pkt_node->hash,
449 		    pkt_node->tval.tv_sec,
450 		    pkt_node->tval.tv_usec,
451 		    UPPER_SHORT(pkt_node->ip_laddr[0]),
452 		    LOWER_SHORT(pkt_node->ip_laddr[0]),
453 		    UPPER_SHORT(pkt_node->ip_laddr[1]),
454 		    LOWER_SHORT(pkt_node->ip_laddr[1]),
455 		    UPPER_SHORT(pkt_node->ip_laddr[2]),
456 		    LOWER_SHORT(pkt_node->ip_laddr[2]),
457 		    UPPER_SHORT(pkt_node->ip_laddr[3]),
458 		    LOWER_SHORT(pkt_node->ip_laddr[3]),
459 		    ntohs(pkt_node->tcp_localport),
460 		    UPPER_SHORT(pkt_node->ip_faddr[0]),
461 		    LOWER_SHORT(pkt_node->ip_faddr[0]),
462 		    UPPER_SHORT(pkt_node->ip_faddr[1]),
463 		    LOWER_SHORT(pkt_node->ip_faddr[1]),
464 		    UPPER_SHORT(pkt_node->ip_faddr[2]),
465 		    LOWER_SHORT(pkt_node->ip_faddr[2]),
466 		    UPPER_SHORT(pkt_node->ip_faddr[3]),
467 		    LOWER_SHORT(pkt_node->ip_faddr[3]),
468 		    ntohs(pkt_node->tcp_foreignport),
469 		    pkt_node->snd_ssthresh,
470 		    pkt_node->snd_cwnd,
471 		    pkt_node->snd_bwnd,
472 		    pkt_node->snd_wnd,
473 		    pkt_node->rcv_wnd,
474 		    pkt_node->snd_scale,
475 		    pkt_node->rcv_scale,
476 		    pkt_node->conn_state,
477 		    pkt_node->max_seg_size,
478 		    pkt_node->smoothed_rtt,
479 		    pkt_node->sack_enabled,
480 		    pkt_node->flags,
481 		    pkt_node->rxt_length,
482 		    pkt_node->snd_buf_hiwater,
483 		    pkt_node->snd_buf_cc,
484 		    pkt_node->rcv_buf_hiwater,
485 		    pkt_node->rcv_buf_cc,
486 		    pkt_node->sent_inflight_bytes,
487 		    pkt_node->t_segqlen);
488 	} else { /* IPv4 packet */
489 		pkt_node->ip_laddr[0] = FIRST_OCTET(pkt_node->ip_laddr[3]);
490 		pkt_node->ip_laddr[1] = SECOND_OCTET(pkt_node->ip_laddr[3]);
491 		pkt_node->ip_laddr[2] = THIRD_OCTET(pkt_node->ip_laddr[3]);
492 		pkt_node->ip_laddr[3] = FOURTH_OCTET(pkt_node->ip_laddr[3]);
493 		pkt_node->ip_faddr[0] = FIRST_OCTET(pkt_node->ip_faddr[3]);
494 		pkt_node->ip_faddr[1] = SECOND_OCTET(pkt_node->ip_faddr[3]);
495 		pkt_node->ip_faddr[2] = THIRD_OCTET(pkt_node->ip_faddr[3]);
496 		pkt_node->ip_faddr[3] = FOURTH_OCTET(pkt_node->ip_faddr[3]);
497 #endif /* SIFTR_IPV6 */
498 
499 		/* Construct an IPv4 log message. */
500 		log_buf->ae_bytesused = snprintf(log_buf->ae_data,
501 		    MAX_LOG_MSG_LEN,
502 		    "%c,0x%08x,%jd.%06ld,%u.%u.%u.%u,%u,%u.%u.%u.%u,%u,%ld,%ld,"
503 		    "%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,%u,%d,%u,%u,%u,%u,%u,%u\n",
504 		    direction[pkt_node->direction],
505 		    pkt_node->hash,
506 		    (intmax_t)pkt_node->tval.tv_sec,
507 		    pkt_node->tval.tv_usec,
508 		    pkt_node->ip_laddr[0],
509 		    pkt_node->ip_laddr[1],
510 		    pkt_node->ip_laddr[2],
511 		    pkt_node->ip_laddr[3],
512 		    ntohs(pkt_node->tcp_localport),
513 		    pkt_node->ip_faddr[0],
514 		    pkt_node->ip_faddr[1],
515 		    pkt_node->ip_faddr[2],
516 		    pkt_node->ip_faddr[3],
517 		    ntohs(pkt_node->tcp_foreignport),
518 		    pkt_node->snd_ssthresh,
519 		    pkt_node->snd_cwnd,
520 		    pkt_node->snd_bwnd,
521 		    pkt_node->snd_wnd,
522 		    pkt_node->rcv_wnd,
523 		    pkt_node->snd_scale,
524 		    pkt_node->rcv_scale,
525 		    pkt_node->conn_state,
526 		    pkt_node->max_seg_size,
527 		    pkt_node->smoothed_rtt,
528 		    pkt_node->sack_enabled,
529 		    pkt_node->flags,
530 		    pkt_node->rxt_length,
531 		    pkt_node->snd_buf_hiwater,
532 		    pkt_node->snd_buf_cc,
533 		    pkt_node->rcv_buf_hiwater,
534 		    pkt_node->rcv_buf_cc,
535 		    pkt_node->sent_inflight_bytes,
536 		    pkt_node->t_segqlen);
537 #ifdef SIFTR_IPV6
538 	}
539 #endif
540 
541 	alq_post_flags(siftr_alq, log_buf, 0);
542 }
543 
544 
545 static void
546 siftr_pkt_manager_thread(void *arg)
547 {
548 	STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
549 	    STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
550 	struct pkt_node *pkt_node, *pkt_node_temp;
551 	uint8_t draining;
552 
553 	draining = 2;
554 
555 	mtx_lock(&siftr_pkt_mgr_mtx);
556 
557 	/* draining == 0 when queue has been flushed and it's safe to exit. */
558 	while (draining) {
559 		/*
560 		 * Sleep until we are signalled to wake because thread has
561 		 * been told to exit or until 1 tick has passed.
562 		 */
563 		mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
564 		    1);
565 
566 		/* Gain exclusive access to the pkt_node queue. */
567 		mtx_lock(&siftr_pkt_queue_mtx);
568 
569 		/*
570 		 * Move pkt_queue to tmp_pkt_queue, which leaves
571 		 * pkt_queue empty and ready to receive more pkt_nodes.
572 		 */
573 		STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
574 
575 		/*
576 		 * We've finished making changes to the list. Unlock it
577 		 * so the pfil hooks can continue queuing pkt_nodes.
578 		 */
579 		mtx_unlock(&siftr_pkt_queue_mtx);
580 
581 		/*
582 		 * We can't hold a mutex whilst calling siftr_process_pkt
583 		 * because ALQ might sleep waiting for buffer space.
584 		 */
585 		mtx_unlock(&siftr_pkt_mgr_mtx);
586 
587 		/* Flush all pkt_nodes to the log file. */
588 		STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes,
589 		    pkt_node_temp) {
590 			siftr_process_pkt(pkt_node);
591 			STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
592 			free(pkt_node, M_SIFTR_PKTNODE);
593 		}
594 
595 		KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
596 		    ("SIFTR tmp_pkt_queue not empty after flush"));
597 
598 		mtx_lock(&siftr_pkt_mgr_mtx);
599 
600 		/*
601 		 * If siftr_exit_pkt_manager_thread gets set during the window
602 		 * where we are draining the tmp_pkt_queue above, there might
603 		 * still be pkts in pkt_queue that need to be drained.
604 		 * Allow one further iteration to occur after
605 		 * siftr_exit_pkt_manager_thread has been set to ensure
606 		 * pkt_queue is completely empty before we kill the thread.
607 		 *
608 		 * siftr_exit_pkt_manager_thread is set only after the pfil
609 		 * hooks have been removed, so only 1 extra iteration
610 		 * is needed to drain the queue.
611 		 */
612 		if (siftr_exit_pkt_manager_thread)
613 			draining--;
614 	}
615 
616 	mtx_unlock(&siftr_pkt_mgr_mtx);
617 
618 	/* Calls wakeup on this thread's struct thread ptr. */
619 	kthread_exit();
620 }
621 
622 
623 static uint32_t
624 hash_pkt(struct mbuf *m, uint32_t offset)
625 {
626 	uint32_t hash;
627 
628 	hash = 0;
629 
630 	while (m != NULL && offset > m->m_len) {
631 		/*
632 		 * The IP packet payload does not start in this mbuf, so
633 		 * need to figure out which mbuf it starts in and what offset
634 		 * into the mbuf's data region the payload starts at.
635 		 */
636 		offset -= m->m_len;
637 		m = m->m_next;
638 	}
639 
640 	while (m != NULL) {
641 		/* Ensure there is data in the mbuf */
642 		if ((m->m_len - offset) > 0)
643 			hash = hash32_buf(m->m_data + offset,
644 			    m->m_len - offset, hash);
645 
646 		m = m->m_next;
647 		offset = 0;
648         }
649 
650 	return (hash);
651 }
652 
653 
654 /*
655  * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
656  * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
657  * Return value >0 means the caller should skip processing this mbuf.
658  */
659 static inline int
660 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
661 {
662 	if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
663 	    != NULL) {
664 		if (dir == PFIL_IN)
665 			ss->nskip_in_dejavu++;
666 		else
667 			ss->nskip_out_dejavu++;
668 
669 		return (1);
670 	} else {
671 		struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
672 		    PACKET_TAG_SIFTR, 0, M_NOWAIT);
673 		if (tag == NULL) {
674 			if (dir == PFIL_IN)
675 				ss->nskip_in_malloc++;
676 			else
677 				ss->nskip_out_malloc++;
678 
679 			return (1);
680 		}
681 
682 		m_tag_prepend(m, tag);
683 	}
684 
685 	return (0);
686 }
687 
688 
689 /*
690  * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
691  * otherwise.
692  */
693 static inline struct inpcb *
694 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
695     uint16_t dport, int dir, struct siftr_stats *ss)
696 {
697 	struct inpcb *inp;
698 
699 	/* We need the tcbinfo lock. */
700 	INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
701 	INP_INFO_RLOCK(&V_tcbinfo);
702 
703 	if (dir == PFIL_IN)
704 		inp = (ipver == INP_IPV4 ?
705 		    in_pcblookup_hash(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
706 		    dport, 0, m->m_pkthdr.rcvif)
707 		    :
708 #ifdef SIFTR_IPV6
709 		    in6_pcblookup_hash(&V_tcbinfo,
710 		    &((struct ip6_hdr *)ip)->ip6_src, sport,
711 		    &((struct ip6_hdr *)ip)->ip6_dst, dport, 0,
712 		    m->m_pkthdr.rcvif)
713 #else
714 		    NULL
715 #endif
716 		    );
717 
718 	else
719 		inp = (ipver == INP_IPV4 ?
720 		    in_pcblookup_hash(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
721 		    sport, 0, m->m_pkthdr.rcvif)
722 		    :
723 #ifdef SIFTR_IPV6
724 		    in6_pcblookup_hash(&V_tcbinfo,
725 		    &((struct ip6_hdr *)ip)->ip6_dst, dport,
726 		    &((struct ip6_hdr *)ip)->ip6_src, sport, 0,
727 		    m->m_pkthdr.rcvif)
728 #else
729 		    NULL
730 #endif
731 		    );
732 
733 	/* If we can't find the inpcb, bail. */
734 	if (inp == NULL) {
735 		if (dir == PFIL_IN)
736 			ss->nskip_in_inpcb++;
737 		else
738 			ss->nskip_out_inpcb++;
739 	} else {
740 		/* Acquire the inpcb lock. */
741 		INP_UNLOCK_ASSERT(inp);
742 		INP_RLOCK(inp);
743 	}
744 	INP_INFO_RUNLOCK(&V_tcbinfo);
745 
746 	return (inp);
747 }
748 
749 
750 static inline void
751 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
752     int ipver, int dir, int inp_locally_locked)
753 {
754 #ifdef SIFTR_IPV6
755 	if (ipver == INP_IPV4) {
756 		pn->ip_laddr[3] = inp->inp_laddr.s_addr;
757 		pn->ip_faddr[3] = inp->inp_faddr.s_addr;
758 #else
759 		*((uint32_t *)pn->ip_laddr) = inp->inp_laddr.s_addr;
760 		*((uint32_t *)pn->ip_faddr) = inp->inp_faddr.s_addr;
761 #endif
762 #ifdef SIFTR_IPV6
763 	} else {
764 		pn->ip_laddr[0] = inp->in6p_laddr.s6_addr32[0];
765 		pn->ip_laddr[1] = inp->in6p_laddr.s6_addr32[1];
766 		pn->ip_laddr[2] = inp->in6p_laddr.s6_addr32[2];
767 		pn->ip_laddr[3] = inp->in6p_laddr.s6_addr32[3];
768 		pn->ip_faddr[0] = inp->in6p_faddr.s6_addr32[0];
769 		pn->ip_faddr[1] = inp->in6p_faddr.s6_addr32[1];
770 		pn->ip_faddr[2] = inp->in6p_faddr.s6_addr32[2];
771 		pn->ip_faddr[3] = inp->in6p_faddr.s6_addr32[3];
772 	}
773 #endif
774 	pn->tcp_localport = inp->inp_lport;
775 	pn->tcp_foreignport = inp->inp_fport;
776 	pn->snd_cwnd = tp->snd_cwnd;
777 	pn->snd_wnd = tp->snd_wnd;
778 	pn->rcv_wnd = tp->rcv_wnd;
779 	pn->snd_bwnd = 0;		/* Unused, kept for compat. */
780 	pn->snd_ssthresh = tp->snd_ssthresh;
781 	pn->snd_scale = tp->snd_scale;
782 	pn->rcv_scale = tp->rcv_scale;
783 	pn->conn_state = tp->t_state;
784 	pn->max_seg_size = tp->t_maxseg;
785 	pn->smoothed_rtt = tp->t_srtt;
786 	pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
787 	pn->flags = tp->t_flags;
788 	pn->rxt_length = tp->t_rxtcur;
789 	pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
790 	pn->snd_buf_cc = inp->inp_socket->so_snd.sb_cc;
791 	pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
792 	pn->rcv_buf_cc = inp->inp_socket->so_rcv.sb_cc;
793 	pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
794 	pn->t_segqlen = tp->t_segqlen;
795 
796 	/* We've finished accessing the tcb so release the lock. */
797 	if (inp_locally_locked)
798 		INP_RUNLOCK(inp);
799 
800 	pn->ipver = ipver;
801 	pn->direction = dir;
802 
803 	/*
804 	 * Significantly more accurate than using getmicrotime(), but slower!
805 	 * Gives true microsecond resolution at the expense of a hit to
806 	 * maximum pps throughput processing when SIFTR is loaded and enabled.
807 	 */
808 	microtime(&pn->tval);
809 }
810 
811 
812 /*
813  * pfil hook that is called for each IPv4 packet making its way through the
814  * stack in either direction.
815  * The pfil subsystem holds a non-sleepable mutex somewhere when
816  * calling our hook function, so we can't sleep at all.
817  * It's very important to use the M_NOWAIT flag with all function calls
818  * that support it so that they won't sleep, otherwise you get a panic.
819  */
820 static int
821 siftr_chkpkt(void *arg, struct mbuf **m, struct ifnet *ifp, int dir,
822     struct inpcb *inp)
823 {
824 	struct pkt_node *pn;
825 	struct ip *ip;
826 	struct tcphdr *th;
827 	struct tcpcb *tp;
828 	struct siftr_stats *ss;
829 	unsigned int ip_hl;
830 	int inp_locally_locked;
831 
832 	inp_locally_locked = 0;
833 	ss = DPCPU_PTR(ss);
834 
835 	/*
836 	 * m_pullup is not required here because ip_{input|output}
837 	 * already do the heavy lifting for us.
838 	 */
839 
840 	ip = mtod(*m, struct ip *);
841 
842 	/* Only continue processing if the packet is TCP. */
843 	if (ip->ip_p != IPPROTO_TCP)
844 		goto ret;
845 
846 	/*
847 	 * If a kernel subsystem reinjects packets into the stack, our pfil
848 	 * hook will be called multiple times for the same packet.
849 	 * Make sure we only process unique packets.
850 	 */
851 	if (siftr_chkreinject(*m, dir, ss))
852 		goto ret;
853 
854 	if (dir == PFIL_IN)
855 		ss->n_in++;
856 	else
857 		ss->n_out++;
858 
859 	/*
860 	 * Create a tcphdr struct starting at the correct offset
861 	 * in the IP packet. ip->ip_hl gives the ip header length
862 	 * in 4-byte words, so multiply it to get the size in bytes.
863 	 */
864 	ip_hl = (ip->ip_hl << 2);
865 	th = (struct tcphdr *)((caddr_t)ip + ip_hl);
866 
867 	/*
868 	 * If the pfil hooks don't provide a pointer to the
869 	 * inpcb, we need to find it ourselves and lock it.
870 	 */
871 	if (!inp) {
872 		/* Find the corresponding inpcb for this pkt. */
873 		inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
874 		    th->th_dport, dir, ss);
875 
876 		if (inp == NULL)
877 			goto ret;
878 		else
879 			inp_locally_locked = 1;
880 	}
881 
882 	INP_LOCK_ASSERT(inp);
883 
884 	/* Find the TCP control block that corresponds with this packet */
885 	tp = intotcpcb(inp);
886 
887 	/*
888 	 * If we can't find the TCP control block (happens occasionaly for a
889 	 * packet sent during the shutdown phase of a TCP connection),
890 	 * or we're in the timewait state, bail
891 	 */
892 	if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
893 		if (dir == PFIL_IN)
894 			ss->nskip_in_tcpcb++;
895 		else
896 			ss->nskip_out_tcpcb++;
897 
898 		goto inp_unlock;
899 	}
900 
901 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
902 
903 	if (pn == NULL) {
904 		if (dir == PFIL_IN)
905 			ss->nskip_in_malloc++;
906 		else
907 			ss->nskip_out_malloc++;
908 
909 		goto inp_unlock;
910 	}
911 
912 	siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
913 
914 	if (siftr_generate_hashes) {
915 		if ((*m)->m_pkthdr.csum_flags & CSUM_TCP) {
916 			/*
917 			 * For outbound packets, the TCP checksum isn't
918 			 * calculated yet. This is a problem for our packet
919 			 * hashing as the receiver will calc a different hash
920 			 * to ours if we don't include the correct TCP checksum
921 			 * in the bytes being hashed. To work around this
922 			 * problem, we manually calc the TCP checksum here in
923 			 * software. We unset the CSUM_TCP flag so the lower
924 			 * layers don't recalc it.
925 			 */
926 			(*m)->m_pkthdr.csum_flags &= ~CSUM_TCP;
927 
928 			/*
929 			 * Calculate the TCP checksum in software and assign
930 			 * to correct TCP header field, which will follow the
931 			 * packet mbuf down the stack. The trick here is that
932 			 * tcp_output() sets th->th_sum to the checksum of the
933 			 * pseudo header for us already. Because of the nature
934 			 * of the checksumming algorithm, we can sum over the
935 			 * entire IP payload (i.e. TCP header and data), which
936 			 * will include the already calculated pseduo header
937 			 * checksum, thus giving us the complete TCP checksum.
938 			 *
939 			 * To put it in simple terms, if checksum(1,2,3,4)=10,
940 			 * then checksum(1,2,3,4,5) == checksum(10,5).
941 			 * This property is what allows us to "cheat" and
942 			 * checksum only the IP payload which has the TCP
943 			 * th_sum field populated with the pseudo header's
944 			 * checksum, and not need to futz around checksumming
945 			 * pseudo header bytes and TCP header/data in one hit.
946 			 * Refer to RFC 1071 for more info.
947 			 *
948 			 * NB: in_cksum_skip(struct mbuf *m, int len, int skip)
949 			 * in_cksum_skip 2nd argument is NOT the number of
950 			 * bytes to read from the mbuf at "skip" bytes offset
951 			 * from the start of the mbuf (very counter intuitive!).
952 			 * The number of bytes to read is calculated internally
953 			 * by the function as len-skip i.e. to sum over the IP
954 			 * payload (TCP header + data) bytes, it is INCORRECT
955 			 * to call the function like this:
956 			 * in_cksum_skip(at, ip->ip_len - offset, offset)
957 			 * Rather, it should be called like this:
958 			 * in_cksum_skip(at, ip->ip_len, offset)
959 			 * which means read "ip->ip_len - offset" bytes from
960 			 * the mbuf cluster "at" at offset "offset" bytes from
961 			 * the beginning of the "at" mbuf's data pointer.
962 			 */
963 			th->th_sum = in_cksum_skip(*m, ip->ip_len, ip_hl);
964 		}
965 
966 		/*
967 		 * XXX: Having to calculate the checksum in software and then
968 		 * hash over all bytes is really inefficient. Would be nice to
969 		 * find a way to create the hash and checksum in the same pass
970 		 * over the bytes.
971 		 */
972 		pn->hash = hash_pkt(*m, ip_hl);
973 	}
974 
975 	mtx_lock(&siftr_pkt_queue_mtx);
976 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
977 	mtx_unlock(&siftr_pkt_queue_mtx);
978 	goto ret;
979 
980 inp_unlock:
981 	if (inp_locally_locked)
982 		INP_RUNLOCK(inp);
983 
984 ret:
985 	/* Returning 0 ensures pfil will not discard the pkt */
986 	return (0);
987 }
988 
989 
990 #ifdef SIFTR_IPV6
991 static int
992 siftr_chkpkt6(void *arg, struct mbuf **m, struct ifnet *ifp, int dir,
993     struct inpcb *inp)
994 {
995 	struct pkt_node *pn;
996 	struct ip6_hdr *ip6;
997 	struct tcphdr *th;
998 	struct tcpcb *tp;
999 	struct siftr_stats *ss;
1000 	unsigned int ip6_hl;
1001 	int inp_locally_locked;
1002 
1003 	inp_locally_locked = 0;
1004 	ss = DPCPU_PTR(ss);
1005 
1006 	/*
1007 	 * m_pullup is not required here because ip6_{input|output}
1008 	 * already do the heavy lifting for us.
1009 	 */
1010 
1011 	ip6 = mtod(*m, struct ip6_hdr *);
1012 
1013 	/*
1014 	 * Only continue processing if the packet is TCP
1015 	 * XXX: We should follow the next header fields
1016 	 * as shown on Pg 6 RFC 2460, but right now we'll
1017 	 * only check pkts that have no extension headers.
1018 	 */
1019 	if (ip6->ip6_nxt != IPPROTO_TCP)
1020 		goto ret6;
1021 
1022 	/*
1023 	 * If a kernel subsystem reinjects packets into the stack, our pfil
1024 	 * hook will be called multiple times for the same packet.
1025 	 * Make sure we only process unique packets.
1026 	 */
1027 	if (siftr_chkreinject(*m, dir, ss))
1028 		goto ret6;
1029 
1030 	if (dir == PFIL_IN)
1031 		ss->n_in++;
1032 	else
1033 		ss->n_out++;
1034 
1035 	ip6_hl = sizeof(struct ip6_hdr);
1036 
1037 	/*
1038 	 * Create a tcphdr struct starting at the correct offset
1039 	 * in the ipv6 packet. ip->ip_hl gives the ip header length
1040 	 * in 4-byte words, so multiply it to get the size in bytes.
1041 	 */
1042 	th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
1043 
1044 	/*
1045 	 * For inbound packets, the pfil hooks don't provide a pointer to the
1046 	 * inpcb, so we need to find it ourselves and lock it.
1047 	 */
1048 	if (!inp) {
1049 		/* Find the corresponding inpcb for this pkt. */
1050 		inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
1051 		    th->th_sport, th->th_dport, dir, ss);
1052 
1053 		if (inp == NULL)
1054 			goto ret6;
1055 		else
1056 			inp_locally_locked = 1;
1057 	}
1058 
1059 	/* Find the TCP control block that corresponds with this packet. */
1060 	tp = intotcpcb(inp);
1061 
1062 	/*
1063 	 * If we can't find the TCP control block (happens occasionaly for a
1064 	 * packet sent during the shutdown phase of a TCP connection),
1065 	 * or we're in the timewait state, bail.
1066 	 */
1067 	if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
1068 		if (dir == PFIL_IN)
1069 			ss->nskip_in_tcpcb++;
1070 		else
1071 			ss->nskip_out_tcpcb++;
1072 
1073 		goto inp_unlock6;
1074 	}
1075 
1076 	pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
1077 
1078 	if (pn == NULL) {
1079 		if (dir == PFIL_IN)
1080 			ss->nskip_in_malloc++;
1081 		else
1082 			ss->nskip_out_malloc++;
1083 
1084 		goto inp_unlock6;
1085 	}
1086 
1087 	siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
1088 
1089 	/* XXX: Figure out how to generate hashes for IPv6 packets. */
1090 
1091 	mtx_lock(&siftr_pkt_queue_mtx);
1092 	STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
1093 	mtx_unlock(&siftr_pkt_queue_mtx);
1094 	goto ret6;
1095 
1096 inp_unlock6:
1097 	if (inp_locally_locked)
1098 		INP_RUNLOCK(inp);
1099 
1100 ret6:
1101 	/* Returning 0 ensures pfil will not discard the pkt. */
1102 	return (0);
1103 }
1104 #endif /* #ifdef SIFTR_IPV6 */
1105 
1106 
1107 static int
1108 siftr_pfil(int action)
1109 {
1110 	struct pfil_head *pfh_inet;
1111 #ifdef SIFTR_IPV6
1112 	struct pfil_head *pfh_inet6;
1113 #endif
1114 	VNET_ITERATOR_DECL(vnet_iter);
1115 
1116 	VNET_LIST_RLOCK();
1117 	VNET_FOREACH(vnet_iter) {
1118 		CURVNET_SET(vnet_iter);
1119 		pfh_inet = pfil_head_get(PFIL_TYPE_AF, AF_INET);
1120 #ifdef SIFTR_IPV6
1121 		pfh_inet6 = pfil_head_get(PFIL_TYPE_AF, AF_INET6);
1122 #endif
1123 
1124 		if (action == HOOK) {
1125 			pfil_add_hook(siftr_chkpkt, NULL,
1126 			    PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet);
1127 #ifdef SIFTR_IPV6
1128 			pfil_add_hook(siftr_chkpkt6, NULL,
1129 			    PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6);
1130 #endif
1131 		} else if (action == UNHOOK) {
1132 			pfil_remove_hook(siftr_chkpkt, NULL,
1133 			    PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet);
1134 #ifdef SIFTR_IPV6
1135 			pfil_remove_hook(siftr_chkpkt6, NULL,
1136 			    PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6);
1137 #endif
1138 		}
1139 		CURVNET_RESTORE();
1140 	}
1141 	VNET_LIST_RUNLOCK();
1142 
1143 	return (0);
1144 }
1145 
1146 
1147 static int
1148 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1149 {
1150 	struct alq *new_alq;
1151 	int error;
1152 
1153 	if (req->newptr == NULL)
1154 		goto skip;
1155 
1156 	/* If old filename and new filename are different. */
1157 	if (strncmp(siftr_logfile, (char *)req->newptr, PATH_MAX)) {
1158 
1159 		error = alq_open(&new_alq, req->newptr, curthread->td_ucred,
1160 		    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1161 
1162 		/* Bail if unable to create new alq. */
1163 		if (error)
1164 			return (1);
1165 
1166 		/*
1167 		 * If disabled, siftr_alq == NULL so we simply close
1168 		 * the alq as we've proved it can be opened.
1169 		 * If enabled, close the existing alq and switch the old
1170 		 * for the new.
1171 		 */
1172 		if (siftr_alq == NULL)
1173 			alq_close(new_alq);
1174 		else {
1175 			alq_close(siftr_alq);
1176 			siftr_alq = new_alq;
1177 		}
1178 	}
1179 
1180 skip:
1181 	return (sysctl_handle_string(oidp, arg1, arg2, req));
1182 }
1183 
1184 
1185 static int
1186 siftr_manage_ops(uint8_t action)
1187 {
1188 	struct siftr_stats totalss;
1189 	struct timeval tval;
1190 	struct flow_hash_node *counter, *tmp_counter;
1191 	struct sbuf *s;
1192 	int i, key_index, ret, error;
1193 	uint32_t bytes_to_write, total_skipped_pkts;
1194 	uint16_t lport, fport;
1195 	uint8_t *key, ipver;
1196 
1197 #ifdef SIFTR_IPV6
1198 	uint32_t laddr[4];
1199 	uint32_t faddr[4];
1200 #else
1201 	uint8_t laddr[4];
1202 	uint8_t faddr[4];
1203 #endif
1204 
1205 	error = 0;
1206 	total_skipped_pkts = 0;
1207 
1208 	/* Init an autosizing sbuf that initially holds 200 chars. */
1209 	if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1210 		return (-1);
1211 
1212 	if (action == SIFTR_ENABLE) {
1213 		/*
1214 		 * Create our alq
1215 		 * XXX: We should abort if alq_open fails!
1216 		 */
1217 		alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1218 		    SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1219 
1220 		STAILQ_INIT(&pkt_queue);
1221 
1222 		DPCPU_ZERO(ss);
1223 
1224 		siftr_exit_pkt_manager_thread = 0;
1225 
1226 		ret = kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1227 		    &siftr_pkt_manager_thr, RFNOWAIT, 0,
1228 		    "siftr_pkt_manager_thr");
1229 
1230 		siftr_pfil(HOOK);
1231 
1232 		microtime(&tval);
1233 
1234 		sbuf_printf(s,
1235 		    "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1236 		    "siftrver=%s\thz=%u\ttcp_rtt_scale=%u\tsysname=%s\t"
1237 		    "sysver=%u\tipmode=%u\n",
1238 		    (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, hz,
1239 		    TCP_RTT_SCALE, SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1240 
1241 		sbuf_finish(s);
1242 		alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1243 
1244 	} else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1245 		/*
1246 		 * Remove the pfil hook functions. All threads currently in
1247 		 * the hook functions are allowed to exit before siftr_pfil()
1248 		 * returns.
1249 		 */
1250 		siftr_pfil(UNHOOK);
1251 
1252 		/* This will block until the pkt manager thread unlocks it. */
1253 		mtx_lock(&siftr_pkt_mgr_mtx);
1254 
1255 		/* Tell the pkt manager thread that it should exit now. */
1256 		siftr_exit_pkt_manager_thread = 1;
1257 
1258 		/*
1259 		 * Wake the pkt_manager thread so it realises that
1260 		 * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1261 		 * The wakeup won't be delivered until we unlock
1262 		 * siftr_pkt_mgr_mtx so this isn't racy.
1263 		 */
1264 		wakeup(&wait_for_pkt);
1265 
1266 		/* Wait for the pkt_manager thread to exit. */
1267 		mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1268 		    "thrwait", 0);
1269 
1270 		siftr_pkt_manager_thr = NULL;
1271 		mtx_unlock(&siftr_pkt_mgr_mtx);
1272 
1273 		totalss.n_in = DPCPU_VARSUM(ss, n_in);
1274 		totalss.n_out = DPCPU_VARSUM(ss, n_out);
1275 		totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1276 		totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1277 		totalss.nskip_in_mtx = DPCPU_VARSUM(ss, nskip_in_mtx);
1278 		totalss.nskip_out_mtx = DPCPU_VARSUM(ss, nskip_out_mtx);
1279 		totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1280 		totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1281 		totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1282 		totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1283 
1284 		total_skipped_pkts = totalss.nskip_in_malloc +
1285 		    totalss.nskip_out_malloc + totalss.nskip_in_mtx +
1286 		    totalss.nskip_out_mtx + totalss.nskip_in_tcpcb +
1287 		    totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1288 		    totalss.nskip_out_inpcb;
1289 
1290 		microtime(&tval);
1291 
1292 		sbuf_printf(s,
1293 		    "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1294 		    "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1295 		    "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1296 		    "num_outbound_skipped_pkts_malloc=%u\t"
1297 		    "num_inbound_skipped_pkts_mtx=%u\t"
1298 		    "num_outbound_skipped_pkts_mtx=%u\t"
1299 		    "num_inbound_skipped_pkts_tcpcb=%u\t"
1300 		    "num_outbound_skipped_pkts_tcpcb=%u\t"
1301 		    "num_inbound_skipped_pkts_inpcb=%u\t"
1302 		    "num_outbound_skipped_pkts_inpcb=%u\t"
1303 		    "total_skipped_tcp_pkts=%u\tflow_list=",
1304 		    (intmax_t)tval.tv_sec,
1305 		    tval.tv_usec,
1306 		    (uintmax_t)totalss.n_in,
1307 		    (uintmax_t)totalss.n_out,
1308 		    (uintmax_t)(totalss.n_in + totalss.n_out),
1309 		    totalss.nskip_in_malloc,
1310 		    totalss.nskip_out_malloc,
1311 		    totalss.nskip_in_mtx,
1312 		    totalss.nskip_out_mtx,
1313 		    totalss.nskip_in_tcpcb,
1314 		    totalss.nskip_out_tcpcb,
1315 		    totalss.nskip_in_inpcb,
1316 		    totalss.nskip_out_inpcb,
1317 		    total_skipped_pkts);
1318 
1319 		/*
1320 		 * Iterate over the flow hash, printing a summary of each
1321 		 * flow seen and freeing any malloc'd memory.
1322 		 * The hash consists of an array of LISTs (man 3 queue).
1323 		 */
1324 		for (i = 0; i < siftr_hashmask; i++) {
1325 			LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1326 			    tmp_counter) {
1327 				key = counter->key;
1328 				key_index = 1;
1329 
1330 				ipver = key[0];
1331 
1332 				memcpy(laddr, key + key_index, sizeof(laddr));
1333 				key_index += sizeof(laddr);
1334 				memcpy(&lport, key + key_index, sizeof(lport));
1335 				key_index += sizeof(lport);
1336 				memcpy(faddr, key + key_index, sizeof(faddr));
1337 				key_index += sizeof(faddr);
1338 				memcpy(&fport, key + key_index, sizeof(fport));
1339 
1340 #ifdef SIFTR_IPV6
1341 				laddr[3] = ntohl(laddr[3]);
1342 				faddr[3] = ntohl(faddr[3]);
1343 
1344 				if (ipver == INP_IPV6) {
1345 					laddr[0] = ntohl(laddr[0]);
1346 					laddr[1] = ntohl(laddr[1]);
1347 					laddr[2] = ntohl(laddr[2]);
1348 					faddr[0] = ntohl(faddr[0]);
1349 					faddr[1] = ntohl(faddr[1]);
1350 					faddr[2] = ntohl(faddr[2]);
1351 
1352 					sbuf_printf(s,
1353 					    "%x:%x:%x:%x:%x:%x:%x:%x;%u-"
1354 					    "%x:%x:%x:%x:%x:%x:%x:%x;%u,",
1355 					    UPPER_SHORT(laddr[0]),
1356 					    LOWER_SHORT(laddr[0]),
1357 					    UPPER_SHORT(laddr[1]),
1358 					    LOWER_SHORT(laddr[1]),
1359 					    UPPER_SHORT(laddr[2]),
1360 					    LOWER_SHORT(laddr[2]),
1361 					    UPPER_SHORT(laddr[3]),
1362 					    LOWER_SHORT(laddr[3]),
1363 					    ntohs(lport),
1364 					    UPPER_SHORT(faddr[0]),
1365 					    LOWER_SHORT(faddr[0]),
1366 					    UPPER_SHORT(faddr[1]),
1367 					    LOWER_SHORT(faddr[1]),
1368 					    UPPER_SHORT(faddr[2]),
1369 					    LOWER_SHORT(faddr[2]),
1370 					    UPPER_SHORT(faddr[3]),
1371 					    LOWER_SHORT(faddr[3]),
1372 					    ntohs(fport));
1373 				} else {
1374 					laddr[0] = FIRST_OCTET(laddr[3]);
1375 					laddr[1] = SECOND_OCTET(laddr[3]);
1376 					laddr[2] = THIRD_OCTET(laddr[3]);
1377 					laddr[3] = FOURTH_OCTET(laddr[3]);
1378 					faddr[0] = FIRST_OCTET(faddr[3]);
1379 					faddr[1] = SECOND_OCTET(faddr[3]);
1380 					faddr[2] = THIRD_OCTET(faddr[3]);
1381 					faddr[3] = FOURTH_OCTET(faddr[3]);
1382 #endif
1383 					sbuf_printf(s,
1384 					    "%u.%u.%u.%u;%u-%u.%u.%u.%u;%u,",
1385 					    laddr[0],
1386 					    laddr[1],
1387 					    laddr[2],
1388 					    laddr[3],
1389 					    ntohs(lport),
1390 					    faddr[0],
1391 					    faddr[1],
1392 					    faddr[2],
1393 					    faddr[3],
1394 					    ntohs(fport));
1395 #ifdef SIFTR_IPV6
1396 				}
1397 #endif
1398 
1399 				free(counter, M_SIFTR_HASHNODE);
1400 			}
1401 
1402 			LIST_INIT(counter_hash + i);
1403 		}
1404 
1405 		sbuf_printf(s, "\n");
1406 		sbuf_finish(s);
1407 
1408 		i = 0;
1409 		do {
1410 			bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1411 			alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1412 			i += bytes_to_write;
1413 		} while (i < sbuf_len(s));
1414 
1415 		alq_close(siftr_alq);
1416 		siftr_alq = NULL;
1417 	}
1418 
1419 	sbuf_delete(s);
1420 
1421 	/*
1422 	 * XXX: Should be using ret to check if any functions fail
1423 	 * and set error appropriately
1424 	 */
1425 
1426 	return (error);
1427 }
1428 
1429 
1430 static int
1431 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1432 {
1433 	if (req->newptr == NULL)
1434 		goto skip;
1435 
1436 	/* If the value passed in isn't 0 or 1, return an error. */
1437 	if (CAST_PTR_INT(req->newptr) != 0 && CAST_PTR_INT(req->newptr) != 1)
1438 		return (1);
1439 
1440 	/* If we are changing state (0 to 1 or 1 to 0). */
1441 	if (CAST_PTR_INT(req->newptr) != siftr_enabled )
1442 		if (siftr_manage_ops(CAST_PTR_INT(req->newptr))) {
1443 			siftr_manage_ops(SIFTR_DISABLE);
1444 			return (1);
1445 		}
1446 
1447 skip:
1448 	return (sysctl_handle_int(oidp, arg1, arg2, req));
1449 }
1450 
1451 
1452 static void
1453 siftr_shutdown_handler(void *arg)
1454 {
1455 	siftr_manage_ops(SIFTR_DISABLE);
1456 }
1457 
1458 
1459 /*
1460  * Module is being unloaded or machine is shutting down. Take care of cleanup.
1461  */
1462 static int
1463 deinit_siftr(void)
1464 {
1465 	/* Cleanup. */
1466 	siftr_manage_ops(SIFTR_DISABLE);
1467 	hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1468 	mtx_destroy(&siftr_pkt_queue_mtx);
1469 	mtx_destroy(&siftr_pkt_mgr_mtx);
1470 
1471 	return (0);
1472 }
1473 
1474 
1475 /*
1476  * Module has just been loaded into the kernel.
1477  */
1478 static int
1479 init_siftr(void)
1480 {
1481 	EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL,
1482 	    SHUTDOWN_PRI_FIRST);
1483 
1484 	/* Initialise our flow counter hash table. */
1485 	counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1486 	    &siftr_hashmask);
1487 
1488 	mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1489 	mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1490 
1491 	/* Print message to the user's current terminal. */
1492 	uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1493 	    "          http://caia.swin.edu.au/urp/newtcp\n\n",
1494 	    MODVERSION_STR);
1495 
1496 	return (0);
1497 }
1498 
1499 
1500 /*
1501  * This is the function that is called to load and unload the module.
1502  * When the module is loaded, this function is called once with
1503  * "what" == MOD_LOAD
1504  * When the module is unloaded, this function is called twice with
1505  * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1506  * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1507  * this function is called once with "what" = MOD_SHUTDOWN
1508  * When the system is shut down, the handler isn't called until the very end
1509  * of the shutdown sequence i.e. after the disks have been synced.
1510  */
1511 static int
1512 siftr_load_handler(module_t mod, int what, void *arg)
1513 {
1514 	int ret;
1515 
1516 	switch (what) {
1517 	case MOD_LOAD:
1518 		ret = init_siftr();
1519 		break;
1520 
1521 	case MOD_QUIESCE:
1522 	case MOD_SHUTDOWN:
1523 		ret = deinit_siftr();
1524 		break;
1525 
1526 	case MOD_UNLOAD:
1527 		ret = 0;
1528 		break;
1529 
1530 	default:
1531 		ret = EINVAL;
1532 		break;
1533 	}
1534 
1535 	return (ret);
1536 }
1537 
1538 
1539 static moduledata_t siftr_mod = {
1540 	.name = "siftr",
1541 	.evhand = siftr_load_handler,
1542 };
1543 
1544 /*
1545  * Param 1: name of the kernel module
1546  * Param 2: moduledata_t struct containing info about the kernel module
1547  *          and the execution entry point for the module
1548  * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1549  *          Defines the module initialisation order
1550  * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1551  *          Defines the initialisation order of this kld relative to others
1552  *          within the same subsystem as defined by param 3
1553  */
1554 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_SMP, SI_ORDER_ANY);
1555 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1556 MODULE_VERSION(siftr, MODVERSION);
1557