1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2007-2009
5 * Swinburne University of Technology, Melbourne, Australia.
6 * Copyright (c) 2009-2010, The FreeBSD Foundation
7 * All rights reserved.
8 *
9 * Portions of this software were developed at the Centre for Advanced
10 * Internet Architectures, Swinburne University of Technology, Melbourne,
11 * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 */
34
35 /******************************************************
36 * Statistical Information For TCP Research (SIFTR)
37 *
38 * A FreeBSD kernel module that adds very basic intrumentation to the
39 * TCP stack, allowing internal stats to be recorded to a log file
40 * for experimental, debugging and performance analysis purposes.
41 *
42 * SIFTR was first released in 2007 by James Healy and Lawrence Stewart whilst
43 * working on the NewTCP research project at Swinburne University of
44 * Technology's Centre for Advanced Internet Architectures, Melbourne,
45 * Australia, which was made possible in part by a grant from the Cisco
46 * University Research Program Fund at Community Foundation Silicon Valley.
47 * More details are available at:
48 * http://caia.swin.edu.au/urp/newtcp/
49 *
50 * Work on SIFTR v1.2.x was sponsored by the FreeBSD Foundation as part of
51 * the "Enhancing the FreeBSD TCP Implementation" project 2008-2009.
52 * More details are available at:
53 * http://www.freebsdfoundation.org/
54 * http://caia.swin.edu.au/freebsd/etcp09/
55 *
56 * Lawrence Stewart is the current maintainer, and all contact regarding
57 * SIFTR should be directed to him via email: [email protected]
58 *
59 * Initial release date: June 2007
60 * Most recent update: September 2010
61 ******************************************************/
62
63 #include <sys/cdefs.h>
64 __FBSDID("$FreeBSD$");
65
66 #include <sys/param.h>
67 #include <sys/alq.h>
68 #include <sys/errno.h>
69 #include <sys/eventhandler.h>
70 #include <sys/hash.h>
71 #include <sys/kernel.h>
72 #include <sys/kthread.h>
73 #include <sys/lock.h>
74 #include <sys/mbuf.h>
75 #include <sys/module.h>
76 #include <sys/mutex.h>
77 #include <sys/pcpu.h>
78 #include <sys/proc.h>
79 #include <sys/sbuf.h>
80 #include <sys/sdt.h>
81 #include <sys/smp.h>
82 #include <sys/socket.h>
83 #include <sys/socketvar.h>
84 #include <sys/sysctl.h>
85 #include <sys/unistd.h>
86
87 #include <net/if.h>
88 #include <net/if_var.h>
89 #include <net/pfil.h>
90
91 #include <netinet/in.h>
92 #include <netinet/in_kdtrace.h>
93 #include <netinet/in_pcb.h>
94 #include <netinet/in_systm.h>
95 #include <netinet/in_var.h>
96 #include <netinet/ip.h>
97 #include <netinet/tcp_var.h>
98
99 #ifdef SIFTR_IPV6
100 #include <netinet/ip6.h>
101 #include <netinet6/in6_pcb.h>
102 #endif /* SIFTR_IPV6 */
103
104 #include <machine/in_cksum.h>
105
106 /*
107 * Three digit version number refers to X.Y.Z where:
108 * X is the major version number
109 * Y is bumped to mark backwards incompatible changes
110 * Z is bumped to mark backwards compatible changes
111 */
112 #define V_MAJOR 1
113 #define V_BACKBREAK 2
114 #define V_BACKCOMPAT 4
115 #define MODVERSION __CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT))
116 #define MODVERSION_STR __XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \
117 __XSTRING(V_BACKCOMPAT)
118
119 #define HOOK 0
120 #define UNHOOK 1
121 #define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536
122 #define SYS_NAME "FreeBSD"
123 #define PACKET_TAG_SIFTR 100
124 #define PACKET_COOKIE_SIFTR 21749576
125 #define SIFTR_LOG_FILE_MODE 0644
126 #define SIFTR_DISABLE 0
127 #define SIFTR_ENABLE 1
128
129 /*
130 * Hard upper limit on the length of log messages. Bump this up if you add new
131 * data fields such that the line length could exceed the below value.
132 */
133 #define MAX_LOG_MSG_LEN 200
134 /* XXX: Make this a sysctl tunable. */
135 #define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN)
136
137 /*
138 * 1 byte for IP version
139 * IPv4: src/dst IP (4+4) + src/dst port (2+2) = 12 bytes
140 * IPv6: src/dst IP (16+16) + src/dst port (2+2) = 36 bytes
141 */
142 #ifdef SIFTR_IPV6
143 #define FLOW_KEY_LEN 37
144 #else
145 #define FLOW_KEY_LEN 13
146 #endif
147
148 #ifdef SIFTR_IPV6
149 #define SIFTR_IPMODE 6
150 #else
151 #define SIFTR_IPMODE 4
152 #endif
153
154 /* useful macros */
155 #define UPPER_SHORT(X) (((X) & 0xFFFF0000) >> 16)
156 #define LOWER_SHORT(X) ((X) & 0x0000FFFF)
157
158 #define FIRST_OCTET(X) (((X) & 0xFF000000) >> 24)
159 #define SECOND_OCTET(X) (((X) & 0x00FF0000) >> 16)
160 #define THIRD_OCTET(X) (((X) & 0x0000FF00) >> 8)
161 #define FOURTH_OCTET(X) ((X) & 0x000000FF)
162
163 static MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR");
164 static MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode",
165 "SIFTR pkt_node struct");
166 static MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode",
167 "SIFTR flow_hash_node struct");
168
169 /* Used as links in the pkt manager queue. */
170 struct pkt_node {
171 /* Timestamp of pkt as noted in the pfil hook. */
172 struct timeval tval;
173 /* Direction pkt is travelling; either PFIL_IN or PFIL_OUT. */
174 uint8_t direction;
175 /* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */
176 uint8_t ipver;
177 /* Hash of the pkt which triggered the log message. */
178 uint32_t hash;
179 /* Local/foreign IP address. */
180 #ifdef SIFTR_IPV6
181 uint32_t ip_laddr[4];
182 uint32_t ip_faddr[4];
183 #else
184 uint8_t ip_laddr[4];
185 uint8_t ip_faddr[4];
186 #endif
187 /* Local TCP port. */
188 uint16_t tcp_localport;
189 /* Foreign TCP port. */
190 uint16_t tcp_foreignport;
191 /* Congestion Window (bytes). */
192 u_long snd_cwnd;
193 /* Sending Window (bytes). */
194 u_long snd_wnd;
195 /* Receive Window (bytes). */
196 u_long rcv_wnd;
197 /* Unused (was: Bandwidth Controlled Window (bytes)). */
198 u_long snd_bwnd;
199 /* Slow Start Threshold (bytes). */
200 u_long snd_ssthresh;
201 /* Current state of the TCP FSM. */
202 int conn_state;
203 /* Max Segment Size (bytes). */
204 u_int max_seg_size;
205 /*
206 * Smoothed RTT stored as found in the TCP control block
207 * in units of (TCP_RTT_SCALE*hz).
208 */
209 int smoothed_rtt;
210 /* Is SACK enabled? */
211 u_char sack_enabled;
212 /* Window scaling for snd window. */
213 u_char snd_scale;
214 /* Window scaling for recv window. */
215 u_char rcv_scale;
216 /* TCP control block flags. */
217 u_int flags;
218 /* Retransmit timeout length. */
219 int rxt_length;
220 /* Size of the TCP send buffer in bytes. */
221 u_int snd_buf_hiwater;
222 /* Current num bytes in the send socket buffer. */
223 u_int snd_buf_cc;
224 /* Size of the TCP receive buffer in bytes. */
225 u_int rcv_buf_hiwater;
226 /* Current num bytes in the receive socket buffer. */
227 u_int rcv_buf_cc;
228 /* Number of bytes inflight that we are waiting on ACKs for. */
229 u_int sent_inflight_bytes;
230 /* Number of segments currently in the reassembly queue. */
231 int t_segqlen;
232 /* Flowid for the connection. */
233 u_int flowid;
234 /* Flow type for the connection. */
235 u_int flowtype;
236 /* Link to next pkt_node in the list. */
237 STAILQ_ENTRY(pkt_node) nodes;
238 };
239
240 struct flow_hash_node
241 {
242 uint16_t counter;
243 uint8_t key[FLOW_KEY_LEN];
244 LIST_ENTRY(flow_hash_node) nodes;
245 };
246
247 struct siftr_stats
248 {
249 /* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */
250 uint64_t n_in;
251 uint64_t n_out;
252 /* # pkts skipped due to failed malloc calls. */
253 uint32_t nskip_in_malloc;
254 uint32_t nskip_out_malloc;
255 /* # pkts skipped due to failed mtx acquisition. */
256 uint32_t nskip_in_mtx;
257 uint32_t nskip_out_mtx;
258 /* # pkts skipped due to failed inpcb lookups. */
259 uint32_t nskip_in_inpcb;
260 uint32_t nskip_out_inpcb;
261 /* # pkts skipped due to failed tcpcb lookups. */
262 uint32_t nskip_in_tcpcb;
263 uint32_t nskip_out_tcpcb;
264 /* # pkts skipped due to stack reinjection. */
265 uint32_t nskip_in_dejavu;
266 uint32_t nskip_out_dejavu;
267 };
268
269 DPCPU_DEFINE_STATIC(struct siftr_stats, ss);
270
271 static volatile unsigned int siftr_exit_pkt_manager_thread = 0;
272 static unsigned int siftr_enabled = 0;
273 static unsigned int siftr_pkts_per_log = 1;
274 static unsigned int siftr_generate_hashes = 0;
275 static uint16_t siftr_port_filter = 0;
276 /* static unsigned int siftr_binary_log = 0; */
277 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log";
278 static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log";
279 static u_long siftr_hashmask;
280 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue);
281 LIST_HEAD(listhead, flow_hash_node) *counter_hash;
282 static int wait_for_pkt;
283 static struct alq *siftr_alq = NULL;
284 static struct mtx siftr_pkt_queue_mtx;
285 static struct mtx siftr_pkt_mgr_mtx;
286 static struct thread *siftr_pkt_manager_thr = NULL;
287 /*
288 * pfil.h defines PFIL_IN as 1 and PFIL_OUT as 2,
289 * which we use as an index into this array.
290 */
291 static char direction[3] = {'\0', 'i','o'};
292
293 /* Required function prototypes. */
294 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS);
295 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS);
296
297
298 /* Declare the net.inet.siftr sysctl tree and populate it. */
299
300 SYSCTL_DECL(_net_inet_siftr);
301
302 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW, NULL,
303 "siftr related settings");
304
305 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, CTLTYPE_UINT|CTLFLAG_RW,
306 &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU",
307 "switch siftr module operations on/off");
308
309 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, CTLTYPE_STRING|CTLFLAG_RW,
310 &siftr_logfile_shadow, sizeof(siftr_logfile_shadow), &siftr_sysctl_logfile_name_handler,
311 "A", "file to save siftr log messages to");
312
313 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW,
314 &siftr_pkts_per_log, 1,
315 "number of packets between generating a log message");
316
317 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, genhashes, CTLFLAG_RW,
318 &siftr_generate_hashes, 0,
319 "enable packet hash generation");
320
321 SYSCTL_U16(_net_inet_siftr, OID_AUTO, port_filter, CTLFLAG_RW,
322 &siftr_port_filter, 0,
323 "enable packet filter on a TCP port");
324
325 /* XXX: TODO
326 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW,
327 &siftr_binary_log, 0,
328 "write log files in binary instead of ascii");
329 */
330
331
332 /* Begin functions. */
333
334 static void
siftr_process_pkt(struct pkt_node * pkt_node)335 siftr_process_pkt(struct pkt_node * pkt_node)
336 {
337 struct flow_hash_node *hash_node;
338 struct listhead *counter_list;
339 struct siftr_stats *ss;
340 struct ale *log_buf;
341 uint8_t key[FLOW_KEY_LEN];
342 uint8_t found_match, key_offset;
343
344 hash_node = NULL;
345 ss = DPCPU_PTR(ss);
346 found_match = 0;
347 key_offset = 1;
348
349 /*
350 * Create the key that will be used to create a hash index
351 * into our hash table. Our key consists of:
352 * ipversion, localip, localport, foreignip, foreignport
353 */
354 key[0] = pkt_node->ipver;
355 memcpy(key + key_offset, &pkt_node->ip_laddr,
356 sizeof(pkt_node->ip_laddr));
357 key_offset += sizeof(pkt_node->ip_laddr);
358 memcpy(key + key_offset, &pkt_node->tcp_localport,
359 sizeof(pkt_node->tcp_localport));
360 key_offset += sizeof(pkt_node->tcp_localport);
361 memcpy(key + key_offset, &pkt_node->ip_faddr,
362 sizeof(pkt_node->ip_faddr));
363 key_offset += sizeof(pkt_node->ip_faddr);
364 memcpy(key + key_offset, &pkt_node->tcp_foreignport,
365 sizeof(pkt_node->tcp_foreignport));
366
367 counter_list = counter_hash +
368 (hash32_buf(key, sizeof(key), 0) & siftr_hashmask);
369
370 /*
371 * If the list is not empty i.e. the hash index has
372 * been used by another flow previously.
373 */
374 if (LIST_FIRST(counter_list) != NULL) {
375 /*
376 * Loop through the hash nodes in the list.
377 * There should normally only be 1 hash node in the list,
378 * except if there have been collisions at the hash index
379 * computed by hash32_buf().
380 */
381 LIST_FOREACH(hash_node, counter_list, nodes) {
382 /*
383 * Check if the key for the pkt we are currently
384 * processing is the same as the key stored in the
385 * hash node we are currently processing.
386 * If they are the same, then we've found the
387 * hash node that stores the counter for the flow
388 * the pkt belongs to.
389 */
390 if (memcmp(hash_node->key, key, sizeof(key)) == 0) {
391 found_match = 1;
392 break;
393 }
394 }
395 }
396
397 /* If this flow hash hasn't been seen before or we have a collision. */
398 if (hash_node == NULL || !found_match) {
399 /* Create a new hash node to store the flow's counter. */
400 hash_node = malloc(sizeof(struct flow_hash_node),
401 M_SIFTR_HASHNODE, M_WAITOK);
402
403 if (hash_node != NULL) {
404 /* Initialise our new hash node list entry. */
405 hash_node->counter = 0;
406 memcpy(hash_node->key, key, sizeof(key));
407 LIST_INSERT_HEAD(counter_list, hash_node, nodes);
408 } else {
409 /* Malloc failed. */
410 if (pkt_node->direction == PFIL_IN)
411 ss->nskip_in_malloc++;
412 else
413 ss->nskip_out_malloc++;
414
415 return;
416 }
417 } else if (siftr_pkts_per_log > 1) {
418 /*
419 * Taking the remainder of the counter divided
420 * by the current value of siftr_pkts_per_log
421 * and storing that in counter provides a neat
422 * way to modulate the frequency of log
423 * messages being written to the log file.
424 */
425 hash_node->counter = (hash_node->counter + 1) %
426 siftr_pkts_per_log;
427
428 /*
429 * If we have not seen enough packets since the last time
430 * we wrote a log message for this connection, return.
431 */
432 if (hash_node->counter > 0)
433 return;
434 }
435
436 log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK);
437
438 if (log_buf == NULL)
439 return; /* Should only happen if the ALQ is shutting down. */
440
441 #ifdef SIFTR_IPV6
442 pkt_node->ip_laddr[3] = ntohl(pkt_node->ip_laddr[3]);
443 pkt_node->ip_faddr[3] = ntohl(pkt_node->ip_faddr[3]);
444
445 if (pkt_node->ipver == INP_IPV6) { /* IPv6 packet */
446 pkt_node->ip_laddr[0] = ntohl(pkt_node->ip_laddr[0]);
447 pkt_node->ip_laddr[1] = ntohl(pkt_node->ip_laddr[1]);
448 pkt_node->ip_laddr[2] = ntohl(pkt_node->ip_laddr[2]);
449 pkt_node->ip_faddr[0] = ntohl(pkt_node->ip_faddr[0]);
450 pkt_node->ip_faddr[1] = ntohl(pkt_node->ip_faddr[1]);
451 pkt_node->ip_faddr[2] = ntohl(pkt_node->ip_faddr[2]);
452
453 /* Construct an IPv6 log message. */
454 log_buf->ae_bytesused = snprintf(log_buf->ae_data,
455 MAX_LOG_MSG_LEN,
456 "%c,0x%08x,%zd.%06ld,%x:%x:%x:%x:%x:%x:%x:%x,%u,%x:%x:%x:"
457 "%x:%x:%x:%x:%x,%u,%ld,%ld,%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,"
458 "%u,%d,%u,%u,%u,%u,%u,%u,%u,%u\n",
459 direction[pkt_node->direction],
460 pkt_node->hash,
461 pkt_node->tval.tv_sec,
462 pkt_node->tval.tv_usec,
463 UPPER_SHORT(pkt_node->ip_laddr[0]),
464 LOWER_SHORT(pkt_node->ip_laddr[0]),
465 UPPER_SHORT(pkt_node->ip_laddr[1]),
466 LOWER_SHORT(pkt_node->ip_laddr[1]),
467 UPPER_SHORT(pkt_node->ip_laddr[2]),
468 LOWER_SHORT(pkt_node->ip_laddr[2]),
469 UPPER_SHORT(pkt_node->ip_laddr[3]),
470 LOWER_SHORT(pkt_node->ip_laddr[3]),
471 ntohs(pkt_node->tcp_localport),
472 UPPER_SHORT(pkt_node->ip_faddr[0]),
473 LOWER_SHORT(pkt_node->ip_faddr[0]),
474 UPPER_SHORT(pkt_node->ip_faddr[1]),
475 LOWER_SHORT(pkt_node->ip_faddr[1]),
476 UPPER_SHORT(pkt_node->ip_faddr[2]),
477 LOWER_SHORT(pkt_node->ip_faddr[2]),
478 UPPER_SHORT(pkt_node->ip_faddr[3]),
479 LOWER_SHORT(pkt_node->ip_faddr[3]),
480 ntohs(pkt_node->tcp_foreignport),
481 pkt_node->snd_ssthresh,
482 pkt_node->snd_cwnd,
483 pkt_node->snd_bwnd,
484 pkt_node->snd_wnd,
485 pkt_node->rcv_wnd,
486 pkt_node->snd_scale,
487 pkt_node->rcv_scale,
488 pkt_node->conn_state,
489 pkt_node->max_seg_size,
490 pkt_node->smoothed_rtt,
491 pkt_node->sack_enabled,
492 pkt_node->flags,
493 pkt_node->rxt_length,
494 pkt_node->snd_buf_hiwater,
495 pkt_node->snd_buf_cc,
496 pkt_node->rcv_buf_hiwater,
497 pkt_node->rcv_buf_cc,
498 pkt_node->sent_inflight_bytes,
499 pkt_node->t_segqlen,
500 pkt_node->flowid,
501 pkt_node->flowtype);
502 } else { /* IPv4 packet */
503 pkt_node->ip_laddr[0] = FIRST_OCTET(pkt_node->ip_laddr[3]);
504 pkt_node->ip_laddr[1] = SECOND_OCTET(pkt_node->ip_laddr[3]);
505 pkt_node->ip_laddr[2] = THIRD_OCTET(pkt_node->ip_laddr[3]);
506 pkt_node->ip_laddr[3] = FOURTH_OCTET(pkt_node->ip_laddr[3]);
507 pkt_node->ip_faddr[0] = FIRST_OCTET(pkt_node->ip_faddr[3]);
508 pkt_node->ip_faddr[1] = SECOND_OCTET(pkt_node->ip_faddr[3]);
509 pkt_node->ip_faddr[2] = THIRD_OCTET(pkt_node->ip_faddr[3]);
510 pkt_node->ip_faddr[3] = FOURTH_OCTET(pkt_node->ip_faddr[3]);
511 #endif /* SIFTR_IPV6 */
512
513 /* Construct an IPv4 log message. */
514 log_buf->ae_bytesused = snprintf(log_buf->ae_data,
515 MAX_LOG_MSG_LEN,
516 "%c,0x%08x,%jd.%06ld,%u.%u.%u.%u,%u,%u.%u.%u.%u,%u,%ld,%ld,"
517 "%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,%u,%d,%u,%u,%u,%u,%u,%u,%u,%u\n",
518 direction[pkt_node->direction],
519 pkt_node->hash,
520 (intmax_t)pkt_node->tval.tv_sec,
521 pkt_node->tval.tv_usec,
522 pkt_node->ip_laddr[0],
523 pkt_node->ip_laddr[1],
524 pkt_node->ip_laddr[2],
525 pkt_node->ip_laddr[3],
526 ntohs(pkt_node->tcp_localport),
527 pkt_node->ip_faddr[0],
528 pkt_node->ip_faddr[1],
529 pkt_node->ip_faddr[2],
530 pkt_node->ip_faddr[3],
531 ntohs(pkt_node->tcp_foreignport),
532 pkt_node->snd_ssthresh,
533 pkt_node->snd_cwnd,
534 pkt_node->snd_bwnd,
535 pkt_node->snd_wnd,
536 pkt_node->rcv_wnd,
537 pkt_node->snd_scale,
538 pkt_node->rcv_scale,
539 pkt_node->conn_state,
540 pkt_node->max_seg_size,
541 pkt_node->smoothed_rtt,
542 pkt_node->sack_enabled,
543 pkt_node->flags,
544 pkt_node->rxt_length,
545 pkt_node->snd_buf_hiwater,
546 pkt_node->snd_buf_cc,
547 pkt_node->rcv_buf_hiwater,
548 pkt_node->rcv_buf_cc,
549 pkt_node->sent_inflight_bytes,
550 pkt_node->t_segqlen,
551 pkt_node->flowid,
552 pkt_node->flowtype);
553 #ifdef SIFTR_IPV6
554 }
555 #endif
556
557 alq_post_flags(siftr_alq, log_buf, 0);
558 }
559
560
561 static void
siftr_pkt_manager_thread(void * arg)562 siftr_pkt_manager_thread(void *arg)
563 {
564 STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
565 STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
566 struct pkt_node *pkt_node, *pkt_node_temp;
567 uint8_t draining;
568
569 draining = 2;
570
571 mtx_lock(&siftr_pkt_mgr_mtx);
572
573 /* draining == 0 when queue has been flushed and it's safe to exit. */
574 while (draining) {
575 /*
576 * Sleep until we are signalled to wake because thread has
577 * been told to exit or until 1 tick has passed.
578 */
579 mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
580 1);
581
582 /* Gain exclusive access to the pkt_node queue. */
583 mtx_lock(&siftr_pkt_queue_mtx);
584
585 /*
586 * Move pkt_queue to tmp_pkt_queue, which leaves
587 * pkt_queue empty and ready to receive more pkt_nodes.
588 */
589 STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
590
591 /*
592 * We've finished making changes to the list. Unlock it
593 * so the pfil hooks can continue queuing pkt_nodes.
594 */
595 mtx_unlock(&siftr_pkt_queue_mtx);
596
597 /*
598 * We can't hold a mutex whilst calling siftr_process_pkt
599 * because ALQ might sleep waiting for buffer space.
600 */
601 mtx_unlock(&siftr_pkt_mgr_mtx);
602
603 /* Flush all pkt_nodes to the log file. */
604 STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes,
605 pkt_node_temp) {
606 siftr_process_pkt(pkt_node);
607 STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
608 free(pkt_node, M_SIFTR_PKTNODE);
609 }
610
611 KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
612 ("SIFTR tmp_pkt_queue not empty after flush"));
613
614 mtx_lock(&siftr_pkt_mgr_mtx);
615
616 /*
617 * If siftr_exit_pkt_manager_thread gets set during the window
618 * where we are draining the tmp_pkt_queue above, there might
619 * still be pkts in pkt_queue that need to be drained.
620 * Allow one further iteration to occur after
621 * siftr_exit_pkt_manager_thread has been set to ensure
622 * pkt_queue is completely empty before we kill the thread.
623 *
624 * siftr_exit_pkt_manager_thread is set only after the pfil
625 * hooks have been removed, so only 1 extra iteration
626 * is needed to drain the queue.
627 */
628 if (siftr_exit_pkt_manager_thread)
629 draining--;
630 }
631
632 mtx_unlock(&siftr_pkt_mgr_mtx);
633
634 /* Calls wakeup on this thread's struct thread ptr. */
635 kthread_exit();
636 }
637
638
639 static uint32_t
hash_pkt(struct mbuf * m,uint32_t offset)640 hash_pkt(struct mbuf *m, uint32_t offset)
641 {
642 uint32_t hash;
643
644 hash = 0;
645
646 while (m != NULL && offset > m->m_len) {
647 /*
648 * The IP packet payload does not start in this mbuf, so
649 * need to figure out which mbuf it starts in and what offset
650 * into the mbuf's data region the payload starts at.
651 */
652 offset -= m->m_len;
653 m = m->m_next;
654 }
655
656 while (m != NULL) {
657 /* Ensure there is data in the mbuf */
658 if ((m->m_len - offset) > 0)
659 hash = hash32_buf(m->m_data + offset,
660 m->m_len - offset, hash);
661
662 m = m->m_next;
663 offset = 0;
664 }
665
666 return (hash);
667 }
668
669
670 /*
671 * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
672 * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
673 * Return value >0 means the caller should skip processing this mbuf.
674 */
675 static inline int
siftr_chkreinject(struct mbuf * m,int dir,struct siftr_stats * ss)676 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
677 {
678 if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
679 != NULL) {
680 if (dir == PFIL_IN)
681 ss->nskip_in_dejavu++;
682 else
683 ss->nskip_out_dejavu++;
684
685 return (1);
686 } else {
687 struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
688 PACKET_TAG_SIFTR, 0, M_NOWAIT);
689 if (tag == NULL) {
690 if (dir == PFIL_IN)
691 ss->nskip_in_malloc++;
692 else
693 ss->nskip_out_malloc++;
694
695 return (1);
696 }
697
698 m_tag_prepend(m, tag);
699 }
700
701 return (0);
702 }
703
704
705 /*
706 * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
707 * otherwise.
708 */
709 static inline struct inpcb *
siftr_findinpcb(int ipver,struct ip * ip,struct mbuf * m,uint16_t sport,uint16_t dport,int dir,struct siftr_stats * ss)710 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
711 uint16_t dport, int dir, struct siftr_stats *ss)
712 {
713 struct inpcb *inp;
714
715 /* We need the tcbinfo lock. */
716 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
717
718 if (dir == PFIL_IN)
719 inp = (ipver == INP_IPV4 ?
720 in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
721 dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
722 :
723 #ifdef SIFTR_IPV6
724 in6_pcblookup(&V_tcbinfo,
725 &((struct ip6_hdr *)ip)->ip6_src, sport,
726 &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB,
727 m->m_pkthdr.rcvif)
728 #else
729 NULL
730 #endif
731 );
732
733 else
734 inp = (ipver == INP_IPV4 ?
735 in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
736 sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
737 :
738 #ifdef SIFTR_IPV6
739 in6_pcblookup(&V_tcbinfo,
740 &((struct ip6_hdr *)ip)->ip6_dst, dport,
741 &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB,
742 m->m_pkthdr.rcvif)
743 #else
744 NULL
745 #endif
746 );
747
748 /* If we can't find the inpcb, bail. */
749 if (inp == NULL) {
750 if (dir == PFIL_IN)
751 ss->nskip_in_inpcb++;
752 else
753 ss->nskip_out_inpcb++;
754 }
755
756 return (inp);
757 }
758
759
760 static inline void
siftr_siftdata(struct pkt_node * pn,struct inpcb * inp,struct tcpcb * tp,int ipver,int dir,int inp_locally_locked)761 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
762 int ipver, int dir, int inp_locally_locked)
763 {
764 #ifdef SIFTR_IPV6
765 if (ipver == INP_IPV4) {
766 pn->ip_laddr[3] = inp->inp_laddr.s_addr;
767 pn->ip_faddr[3] = inp->inp_faddr.s_addr;
768 #else
769 *((uint32_t *)pn->ip_laddr) = inp->inp_laddr.s_addr;
770 *((uint32_t *)pn->ip_faddr) = inp->inp_faddr.s_addr;
771 #endif
772 #ifdef SIFTR_IPV6
773 } else {
774 pn->ip_laddr[0] = inp->in6p_laddr.s6_addr32[0];
775 pn->ip_laddr[1] = inp->in6p_laddr.s6_addr32[1];
776 pn->ip_laddr[2] = inp->in6p_laddr.s6_addr32[2];
777 pn->ip_laddr[3] = inp->in6p_laddr.s6_addr32[3];
778 pn->ip_faddr[0] = inp->in6p_faddr.s6_addr32[0];
779 pn->ip_faddr[1] = inp->in6p_faddr.s6_addr32[1];
780 pn->ip_faddr[2] = inp->in6p_faddr.s6_addr32[2];
781 pn->ip_faddr[3] = inp->in6p_faddr.s6_addr32[3];
782 }
783 #endif
784 pn->tcp_localport = inp->inp_lport;
785 pn->tcp_foreignport = inp->inp_fport;
786 pn->snd_cwnd = tp->snd_cwnd;
787 pn->snd_wnd = tp->snd_wnd;
788 pn->rcv_wnd = tp->rcv_wnd;
789 pn->snd_bwnd = 0; /* Unused, kept for compat. */
790 pn->snd_ssthresh = tp->snd_ssthresh;
791 pn->snd_scale = tp->snd_scale;
792 pn->rcv_scale = tp->rcv_scale;
793 pn->conn_state = tp->t_state;
794 pn->max_seg_size = tp->t_maxseg;
795 pn->smoothed_rtt = tp->t_srtt;
796 pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
797 pn->flags = tp->t_flags;
798 pn->rxt_length = tp->t_rxtcur;
799 pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
800 pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd);
801 pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
802 pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv);
803 pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
804 pn->t_segqlen = tp->t_segqlen;
805 pn->flowid = inp->inp_flowid;
806 pn->flowtype = inp->inp_flowtype;
807
808 /* We've finished accessing the tcb so release the lock. */
809 if (inp_locally_locked)
810 INP_RUNLOCK(inp);
811
812 pn->ipver = ipver;
813 pn->direction = dir;
814
815 /*
816 * Significantly more accurate than using getmicrotime(), but slower!
817 * Gives true microsecond resolution at the expense of a hit to
818 * maximum pps throughput processing when SIFTR is loaded and enabled.
819 */
820 microtime(&pn->tval);
821 TCP_PROBE1(siftr, &pn);
822
823 }
824
825
826 /*
827 * pfil hook that is called for each IPv4 packet making its way through the
828 * stack in either direction.
829 * The pfil subsystem holds a non-sleepable mutex somewhere when
830 * calling our hook function, so we can't sleep at all.
831 * It's very important to use the M_NOWAIT flag with all function calls
832 * that support it so that they won't sleep, otherwise you get a panic.
833 */
834 static int
siftr_chkpkt(void * arg,struct mbuf ** m,struct ifnet * ifp,int dir,struct inpcb * inp)835 siftr_chkpkt(void *arg, struct mbuf **m, struct ifnet *ifp, int dir,
836 struct inpcb *inp)
837 {
838 struct pkt_node *pn;
839 struct ip *ip;
840 struct tcphdr *th;
841 struct tcpcb *tp;
842 struct siftr_stats *ss;
843 unsigned int ip_hl;
844 int inp_locally_locked;
845
846 inp_locally_locked = 0;
847 ss = DPCPU_PTR(ss);
848
849 /*
850 * m_pullup is not required here because ip_{input|output}
851 * already do the heavy lifting for us.
852 */
853
854 ip = mtod(*m, struct ip *);
855
856 /* Only continue processing if the packet is TCP. */
857 if (ip->ip_p != IPPROTO_TCP)
858 goto ret;
859
860 /*
861 * If a kernel subsystem reinjects packets into the stack, our pfil
862 * hook will be called multiple times for the same packet.
863 * Make sure we only process unique packets.
864 */
865 if (siftr_chkreinject(*m, dir, ss))
866 goto ret;
867
868 if (dir == PFIL_IN)
869 ss->n_in++;
870 else
871 ss->n_out++;
872
873 /*
874 * Create a tcphdr struct starting at the correct offset
875 * in the IP packet. ip->ip_hl gives the ip header length
876 * in 4-byte words, so multiply it to get the size in bytes.
877 */
878 ip_hl = (ip->ip_hl << 2);
879 th = (struct tcphdr *)((caddr_t)ip + ip_hl);
880
881 /*
882 * If the pfil hooks don't provide a pointer to the
883 * inpcb, we need to find it ourselves and lock it.
884 */
885 if (!inp) {
886 /* Find the corresponding inpcb for this pkt. */
887 inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
888 th->th_dport, dir, ss);
889
890 if (inp == NULL)
891 goto ret;
892 else
893 inp_locally_locked = 1;
894 }
895
896 INP_LOCK_ASSERT(inp);
897
898 /* Find the TCP control block that corresponds with this packet */
899 tp = intotcpcb(inp);
900
901 /*
902 * If we can't find the TCP control block (happens occasionaly for a
903 * packet sent during the shutdown phase of a TCP connection),
904 * or we're in the timewait state, bail
905 */
906 if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
907 if (dir == PFIL_IN)
908 ss->nskip_in_tcpcb++;
909 else
910 ss->nskip_out_tcpcb++;
911
912 goto inp_unlock;
913 }
914
915 /*
916 * Only pkts selected by the tcp port filter
917 * can be inserted into the pkt_queue
918 */
919 if ((siftr_port_filter != 0) &&
920 (siftr_port_filter != ntohs(inp->inp_lport)) &&
921 (siftr_port_filter != ntohs(inp->inp_fport))) {
922 goto inp_unlock;
923 }
924
925 pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
926
927 if (pn == NULL) {
928 if (dir == PFIL_IN)
929 ss->nskip_in_malloc++;
930 else
931 ss->nskip_out_malloc++;
932
933 goto inp_unlock;
934 }
935
936 siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
937
938 if (siftr_generate_hashes) {
939 if ((*m)->m_pkthdr.csum_flags & CSUM_TCP) {
940 /*
941 * For outbound packets, the TCP checksum isn't
942 * calculated yet. This is a problem for our packet
943 * hashing as the receiver will calc a different hash
944 * to ours if we don't include the correct TCP checksum
945 * in the bytes being hashed. To work around this
946 * problem, we manually calc the TCP checksum here in
947 * software. We unset the CSUM_TCP flag so the lower
948 * layers don't recalc it.
949 */
950 (*m)->m_pkthdr.csum_flags &= ~CSUM_TCP;
951
952 /*
953 * Calculate the TCP checksum in software and assign
954 * to correct TCP header field, which will follow the
955 * packet mbuf down the stack. The trick here is that
956 * tcp_output() sets th->th_sum to the checksum of the
957 * pseudo header for us already. Because of the nature
958 * of the checksumming algorithm, we can sum over the
959 * entire IP payload (i.e. TCP header and data), which
960 * will include the already calculated pseduo header
961 * checksum, thus giving us the complete TCP checksum.
962 *
963 * To put it in simple terms, if checksum(1,2,3,4)=10,
964 * then checksum(1,2,3,4,5) == checksum(10,5).
965 * This property is what allows us to "cheat" and
966 * checksum only the IP payload which has the TCP
967 * th_sum field populated with the pseudo header's
968 * checksum, and not need to futz around checksumming
969 * pseudo header bytes and TCP header/data in one hit.
970 * Refer to RFC 1071 for more info.
971 *
972 * NB: in_cksum_skip(struct mbuf *m, int len, int skip)
973 * in_cksum_skip 2nd argument is NOT the number of
974 * bytes to read from the mbuf at "skip" bytes offset
975 * from the start of the mbuf (very counter intuitive!).
976 * The number of bytes to read is calculated internally
977 * by the function as len-skip i.e. to sum over the IP
978 * payload (TCP header + data) bytes, it is INCORRECT
979 * to call the function like this:
980 * in_cksum_skip(at, ip->ip_len - offset, offset)
981 * Rather, it should be called like this:
982 * in_cksum_skip(at, ip->ip_len, offset)
983 * which means read "ip->ip_len - offset" bytes from
984 * the mbuf cluster "at" at offset "offset" bytes from
985 * the beginning of the "at" mbuf's data pointer.
986 */
987 th->th_sum = in_cksum_skip(*m, ntohs(ip->ip_len),
988 ip_hl);
989 }
990
991 /*
992 * XXX: Having to calculate the checksum in software and then
993 * hash over all bytes is really inefficient. Would be nice to
994 * find a way to create the hash and checksum in the same pass
995 * over the bytes.
996 */
997 pn->hash = hash_pkt(*m, ip_hl);
998 }
999
1000 mtx_lock(&siftr_pkt_queue_mtx);
1001 STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
1002 mtx_unlock(&siftr_pkt_queue_mtx);
1003 goto ret;
1004
1005 inp_unlock:
1006 if (inp_locally_locked)
1007 INP_RUNLOCK(inp);
1008
1009 ret:
1010 /* Returning 0 ensures pfil will not discard the pkt */
1011 return (0);
1012 }
1013
1014
1015 #ifdef SIFTR_IPV6
1016 static int
siftr_chkpkt6(void * arg,struct mbuf ** m,struct ifnet * ifp,int dir,struct inpcb * inp)1017 siftr_chkpkt6(void *arg, struct mbuf **m, struct ifnet *ifp, int dir,
1018 struct inpcb *inp)
1019 {
1020 struct pkt_node *pn;
1021 struct ip6_hdr *ip6;
1022 struct tcphdr *th;
1023 struct tcpcb *tp;
1024 struct siftr_stats *ss;
1025 unsigned int ip6_hl;
1026 int inp_locally_locked;
1027
1028 inp_locally_locked = 0;
1029 ss = DPCPU_PTR(ss);
1030
1031 /*
1032 * m_pullup is not required here because ip6_{input|output}
1033 * already do the heavy lifting for us.
1034 */
1035
1036 ip6 = mtod(*m, struct ip6_hdr *);
1037
1038 /*
1039 * Only continue processing if the packet is TCP
1040 * XXX: We should follow the next header fields
1041 * as shown on Pg 6 RFC 2460, but right now we'll
1042 * only check pkts that have no extension headers.
1043 */
1044 if (ip6->ip6_nxt != IPPROTO_TCP)
1045 goto ret6;
1046
1047 /*
1048 * If a kernel subsystem reinjects packets into the stack, our pfil
1049 * hook will be called multiple times for the same packet.
1050 * Make sure we only process unique packets.
1051 */
1052 if (siftr_chkreinject(*m, dir, ss))
1053 goto ret6;
1054
1055 if (dir == PFIL_IN)
1056 ss->n_in++;
1057 else
1058 ss->n_out++;
1059
1060 ip6_hl = sizeof(struct ip6_hdr);
1061
1062 /*
1063 * Create a tcphdr struct starting at the correct offset
1064 * in the ipv6 packet. ip->ip_hl gives the ip header length
1065 * in 4-byte words, so multiply it to get the size in bytes.
1066 */
1067 th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
1068
1069 /*
1070 * For inbound packets, the pfil hooks don't provide a pointer to the
1071 * inpcb, so we need to find it ourselves and lock it.
1072 */
1073 if (!inp) {
1074 /* Find the corresponding inpcb for this pkt. */
1075 inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
1076 th->th_sport, th->th_dport, dir, ss);
1077
1078 if (inp == NULL)
1079 goto ret6;
1080 else
1081 inp_locally_locked = 1;
1082 }
1083
1084 /* Find the TCP control block that corresponds with this packet. */
1085 tp = intotcpcb(inp);
1086
1087 /*
1088 * If we can't find the TCP control block (happens occasionaly for a
1089 * packet sent during the shutdown phase of a TCP connection),
1090 * or we're in the timewait state, bail.
1091 */
1092 if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
1093 if (dir == PFIL_IN)
1094 ss->nskip_in_tcpcb++;
1095 else
1096 ss->nskip_out_tcpcb++;
1097
1098 goto inp_unlock6;
1099 }
1100
1101 /*
1102 * Only pkts selected by the tcp port filter
1103 * can be inserted into the pkt_queue
1104 */
1105 if ((siftr_port_filter != 0) &&
1106 (siftr_port_filter != ntohs(inp->inp_lport)) &&
1107 (siftr_port_filter != ntohs(inp->inp_fport))) {
1108 goto inp_unlock6;
1109 }
1110
1111 pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
1112
1113 if (pn == NULL) {
1114 if (dir == PFIL_IN)
1115 ss->nskip_in_malloc++;
1116 else
1117 ss->nskip_out_malloc++;
1118
1119 goto inp_unlock6;
1120 }
1121
1122 siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
1123
1124 /* XXX: Figure out how to generate hashes for IPv6 packets. */
1125
1126 mtx_lock(&siftr_pkt_queue_mtx);
1127 STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
1128 mtx_unlock(&siftr_pkt_queue_mtx);
1129 goto ret6;
1130
1131 inp_unlock6:
1132 if (inp_locally_locked)
1133 INP_RUNLOCK(inp);
1134
1135 ret6:
1136 /* Returning 0 ensures pfil will not discard the pkt. */
1137 return (0);
1138 }
1139 #endif /* #ifdef SIFTR_IPV6 */
1140
1141
1142 static int
siftr_pfil(int action)1143 siftr_pfil(int action)
1144 {
1145 struct pfil_head *pfh_inet;
1146 #ifdef SIFTR_IPV6
1147 struct pfil_head *pfh_inet6;
1148 #endif
1149 VNET_ITERATOR_DECL(vnet_iter);
1150
1151 VNET_LIST_RLOCK();
1152 VNET_FOREACH(vnet_iter) {
1153 CURVNET_SET(vnet_iter);
1154 pfh_inet = pfil_head_get(PFIL_TYPE_AF, AF_INET);
1155 #ifdef SIFTR_IPV6
1156 pfh_inet6 = pfil_head_get(PFIL_TYPE_AF, AF_INET6);
1157 #endif
1158
1159 if (action == HOOK) {
1160 pfil_add_hook(siftr_chkpkt, NULL,
1161 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet);
1162 #ifdef SIFTR_IPV6
1163 pfil_add_hook(siftr_chkpkt6, NULL,
1164 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6);
1165 #endif
1166 } else if (action == UNHOOK) {
1167 pfil_remove_hook(siftr_chkpkt, NULL,
1168 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet);
1169 #ifdef SIFTR_IPV6
1170 pfil_remove_hook(siftr_chkpkt6, NULL,
1171 PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6);
1172 #endif
1173 }
1174 CURVNET_RESTORE();
1175 }
1176 VNET_LIST_RUNLOCK();
1177
1178 return (0);
1179 }
1180
1181
1182 static int
siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)1183 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1184 {
1185 struct alq *new_alq;
1186 int error;
1187
1188 error = sysctl_handle_string(oidp, arg1, arg2, req);
1189
1190 /* Check for error or same filename */
1191 if (error != 0 || req->newptr == NULL ||
1192 strncmp(siftr_logfile, arg1, arg2) == 0)
1193 goto done;
1194
1195 /* Filname changed */
1196 error = alq_open(&new_alq, arg1, curthread->td_ucred,
1197 SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1198 if (error != 0)
1199 goto done;
1200
1201 /*
1202 * If disabled, siftr_alq == NULL so we simply close
1203 * the alq as we've proved it can be opened.
1204 * If enabled, close the existing alq and switch the old
1205 * for the new.
1206 */
1207 if (siftr_alq == NULL) {
1208 alq_close(new_alq);
1209 } else {
1210 alq_close(siftr_alq);
1211 siftr_alq = new_alq;
1212 }
1213
1214 /* Update filename upon success */
1215 strlcpy(siftr_logfile, arg1, arg2);
1216 done:
1217 return (error);
1218 }
1219
1220 static int
siftr_manage_ops(uint8_t action)1221 siftr_manage_ops(uint8_t action)
1222 {
1223 struct siftr_stats totalss;
1224 struct timeval tval;
1225 struct flow_hash_node *counter, *tmp_counter;
1226 struct sbuf *s;
1227 int i, key_index, error;
1228 uint32_t bytes_to_write, total_skipped_pkts;
1229 uint16_t lport, fport;
1230 uint8_t *key, ipver __unused;
1231
1232 #ifdef SIFTR_IPV6
1233 uint32_t laddr[4];
1234 uint32_t faddr[4];
1235 #else
1236 uint8_t laddr[4];
1237 uint8_t faddr[4];
1238 #endif
1239
1240 error = 0;
1241 total_skipped_pkts = 0;
1242
1243 /* Init an autosizing sbuf that initially holds 200 chars. */
1244 if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1245 return (-1);
1246
1247 if (action == SIFTR_ENABLE && siftr_pkt_manager_thr == NULL) {
1248 /*
1249 * Create our alq
1250 * XXX: We should abort if alq_open fails!
1251 */
1252 alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1253 SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1254
1255 STAILQ_INIT(&pkt_queue);
1256
1257 DPCPU_ZERO(ss);
1258
1259 siftr_exit_pkt_manager_thread = 0;
1260
1261 kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1262 &siftr_pkt_manager_thr, RFNOWAIT, 0,
1263 "siftr_pkt_manager_thr");
1264
1265 siftr_pfil(HOOK);
1266
1267 microtime(&tval);
1268
1269 sbuf_printf(s,
1270 "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1271 "siftrver=%s\thz=%u\ttcp_rtt_scale=%u\tsysname=%s\t"
1272 "sysver=%u\tipmode=%u\n",
1273 (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, hz,
1274 TCP_RTT_SCALE, SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1275
1276 sbuf_finish(s);
1277 alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1278
1279 } else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1280 /*
1281 * Remove the pfil hook functions. All threads currently in
1282 * the hook functions are allowed to exit before siftr_pfil()
1283 * returns.
1284 */
1285 siftr_pfil(UNHOOK);
1286
1287 /* This will block until the pkt manager thread unlocks it. */
1288 mtx_lock(&siftr_pkt_mgr_mtx);
1289
1290 /* Tell the pkt manager thread that it should exit now. */
1291 siftr_exit_pkt_manager_thread = 1;
1292
1293 /*
1294 * Wake the pkt_manager thread so it realises that
1295 * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1296 * The wakeup won't be delivered until we unlock
1297 * siftr_pkt_mgr_mtx so this isn't racy.
1298 */
1299 wakeup(&wait_for_pkt);
1300
1301 /* Wait for the pkt_manager thread to exit. */
1302 mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1303 "thrwait", 0);
1304
1305 siftr_pkt_manager_thr = NULL;
1306 mtx_unlock(&siftr_pkt_mgr_mtx);
1307
1308 totalss.n_in = DPCPU_VARSUM(ss, n_in);
1309 totalss.n_out = DPCPU_VARSUM(ss, n_out);
1310 totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1311 totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1312 totalss.nskip_in_mtx = DPCPU_VARSUM(ss, nskip_in_mtx);
1313 totalss.nskip_out_mtx = DPCPU_VARSUM(ss, nskip_out_mtx);
1314 totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1315 totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1316 totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1317 totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1318
1319 total_skipped_pkts = totalss.nskip_in_malloc +
1320 totalss.nskip_out_malloc + totalss.nskip_in_mtx +
1321 totalss.nskip_out_mtx + totalss.nskip_in_tcpcb +
1322 totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1323 totalss.nskip_out_inpcb;
1324
1325 microtime(&tval);
1326
1327 sbuf_printf(s,
1328 "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1329 "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1330 "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1331 "num_outbound_skipped_pkts_malloc=%u\t"
1332 "num_inbound_skipped_pkts_mtx=%u\t"
1333 "num_outbound_skipped_pkts_mtx=%u\t"
1334 "num_inbound_skipped_pkts_tcpcb=%u\t"
1335 "num_outbound_skipped_pkts_tcpcb=%u\t"
1336 "num_inbound_skipped_pkts_inpcb=%u\t"
1337 "num_outbound_skipped_pkts_inpcb=%u\t"
1338 "total_skipped_tcp_pkts=%u\tflow_list=",
1339 (intmax_t)tval.tv_sec,
1340 tval.tv_usec,
1341 (uintmax_t)totalss.n_in,
1342 (uintmax_t)totalss.n_out,
1343 (uintmax_t)(totalss.n_in + totalss.n_out),
1344 totalss.nskip_in_malloc,
1345 totalss.nskip_out_malloc,
1346 totalss.nskip_in_mtx,
1347 totalss.nskip_out_mtx,
1348 totalss.nskip_in_tcpcb,
1349 totalss.nskip_out_tcpcb,
1350 totalss.nskip_in_inpcb,
1351 totalss.nskip_out_inpcb,
1352 total_skipped_pkts);
1353
1354 /*
1355 * Iterate over the flow hash, printing a summary of each
1356 * flow seen and freeing any malloc'd memory.
1357 * The hash consists of an array of LISTs (man 3 queue).
1358 */
1359 for (i = 0; i <= siftr_hashmask; i++) {
1360 LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1361 tmp_counter) {
1362 key = counter->key;
1363 key_index = 1;
1364
1365 ipver = key[0];
1366
1367 memcpy(laddr, key + key_index, sizeof(laddr));
1368 key_index += sizeof(laddr);
1369 memcpy(&lport, key + key_index, sizeof(lport));
1370 key_index += sizeof(lport);
1371 memcpy(faddr, key + key_index, sizeof(faddr));
1372 key_index += sizeof(faddr);
1373 memcpy(&fport, key + key_index, sizeof(fport));
1374
1375 #ifdef SIFTR_IPV6
1376 laddr[3] = ntohl(laddr[3]);
1377 faddr[3] = ntohl(faddr[3]);
1378
1379 if (ipver == INP_IPV6) {
1380 laddr[0] = ntohl(laddr[0]);
1381 laddr[1] = ntohl(laddr[1]);
1382 laddr[2] = ntohl(laddr[2]);
1383 faddr[0] = ntohl(faddr[0]);
1384 faddr[1] = ntohl(faddr[1]);
1385 faddr[2] = ntohl(faddr[2]);
1386
1387 sbuf_printf(s,
1388 "%x:%x:%x:%x:%x:%x:%x:%x;%u-"
1389 "%x:%x:%x:%x:%x:%x:%x:%x;%u,",
1390 UPPER_SHORT(laddr[0]),
1391 LOWER_SHORT(laddr[0]),
1392 UPPER_SHORT(laddr[1]),
1393 LOWER_SHORT(laddr[1]),
1394 UPPER_SHORT(laddr[2]),
1395 LOWER_SHORT(laddr[2]),
1396 UPPER_SHORT(laddr[3]),
1397 LOWER_SHORT(laddr[3]),
1398 ntohs(lport),
1399 UPPER_SHORT(faddr[0]),
1400 LOWER_SHORT(faddr[0]),
1401 UPPER_SHORT(faddr[1]),
1402 LOWER_SHORT(faddr[1]),
1403 UPPER_SHORT(faddr[2]),
1404 LOWER_SHORT(faddr[2]),
1405 UPPER_SHORT(faddr[3]),
1406 LOWER_SHORT(faddr[3]),
1407 ntohs(fport));
1408 } else {
1409 laddr[0] = FIRST_OCTET(laddr[3]);
1410 laddr[1] = SECOND_OCTET(laddr[3]);
1411 laddr[2] = THIRD_OCTET(laddr[3]);
1412 laddr[3] = FOURTH_OCTET(laddr[3]);
1413 faddr[0] = FIRST_OCTET(faddr[3]);
1414 faddr[1] = SECOND_OCTET(faddr[3]);
1415 faddr[2] = THIRD_OCTET(faddr[3]);
1416 faddr[3] = FOURTH_OCTET(faddr[3]);
1417 #endif
1418 sbuf_printf(s,
1419 "%u.%u.%u.%u;%u-%u.%u.%u.%u;%u,",
1420 laddr[0],
1421 laddr[1],
1422 laddr[2],
1423 laddr[3],
1424 ntohs(lport),
1425 faddr[0],
1426 faddr[1],
1427 faddr[2],
1428 faddr[3],
1429 ntohs(fport));
1430 #ifdef SIFTR_IPV6
1431 }
1432 #endif
1433
1434 free(counter, M_SIFTR_HASHNODE);
1435 }
1436
1437 LIST_INIT(counter_hash + i);
1438 }
1439
1440 sbuf_printf(s, "\n");
1441 sbuf_finish(s);
1442
1443 i = 0;
1444 do {
1445 bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1446 alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1447 i += bytes_to_write;
1448 } while (i < sbuf_len(s));
1449
1450 alq_close(siftr_alq);
1451 siftr_alq = NULL;
1452 } else
1453 error = EINVAL;
1454
1455 sbuf_delete(s);
1456
1457 /*
1458 * XXX: Should be using ret to check if any functions fail
1459 * and set error appropriately
1460 */
1461
1462 return (error);
1463 }
1464
1465
1466 static int
siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)1467 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1468 {
1469 int error;
1470 uint32_t new;
1471
1472 new = siftr_enabled;
1473 error = sysctl_handle_int(oidp, &new, 0, req);
1474 if (error == 0 && req->newptr != NULL) {
1475 if (new > 1)
1476 return (EINVAL);
1477 else if (new != siftr_enabled) {
1478 if ((error = siftr_manage_ops(new)) == 0) {
1479 siftr_enabled = new;
1480 } else {
1481 siftr_manage_ops(SIFTR_DISABLE);
1482 }
1483 }
1484 }
1485
1486 return (error);
1487 }
1488
1489
1490 static void
siftr_shutdown_handler(void * arg)1491 siftr_shutdown_handler(void *arg)
1492 {
1493 if (siftr_enabled == 1) {
1494 siftr_manage_ops(SIFTR_DISABLE);
1495 }
1496 }
1497
1498
1499 /*
1500 * Module is being unloaded or machine is shutting down. Take care of cleanup.
1501 */
1502 static int
deinit_siftr(void)1503 deinit_siftr(void)
1504 {
1505 /* Cleanup. */
1506 siftr_manage_ops(SIFTR_DISABLE);
1507 hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1508 mtx_destroy(&siftr_pkt_queue_mtx);
1509 mtx_destroy(&siftr_pkt_mgr_mtx);
1510
1511 return (0);
1512 }
1513
1514
1515 /*
1516 * Module has just been loaded into the kernel.
1517 */
1518 static int
init_siftr(void)1519 init_siftr(void)
1520 {
1521 EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL,
1522 SHUTDOWN_PRI_FIRST);
1523
1524 /* Initialise our flow counter hash table. */
1525 counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1526 &siftr_hashmask);
1527
1528 mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1529 mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1530
1531 /* Print message to the user's current terminal. */
1532 uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1533 " http://caia.swin.edu.au/urp/newtcp\n\n",
1534 MODVERSION_STR);
1535
1536 return (0);
1537 }
1538
1539
1540 /*
1541 * This is the function that is called to load and unload the module.
1542 * When the module is loaded, this function is called once with
1543 * "what" == MOD_LOAD
1544 * When the module is unloaded, this function is called twice with
1545 * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1546 * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1547 * this function is called once with "what" = MOD_SHUTDOWN
1548 * When the system is shut down, the handler isn't called until the very end
1549 * of the shutdown sequence i.e. after the disks have been synced.
1550 */
1551 static int
siftr_load_handler(module_t mod,int what,void * arg)1552 siftr_load_handler(module_t mod, int what, void *arg)
1553 {
1554 int ret;
1555
1556 switch (what) {
1557 case MOD_LOAD:
1558 ret = init_siftr();
1559 break;
1560
1561 case MOD_QUIESCE:
1562 case MOD_SHUTDOWN:
1563 ret = deinit_siftr();
1564 break;
1565
1566 case MOD_UNLOAD:
1567 ret = 0;
1568 break;
1569
1570 default:
1571 ret = EINVAL;
1572 break;
1573 }
1574
1575 return (ret);
1576 }
1577
1578
1579 static moduledata_t siftr_mod = {
1580 .name = "siftr",
1581 .evhand = siftr_load_handler,
1582 };
1583
1584 /*
1585 * Param 1: name of the kernel module
1586 * Param 2: moduledata_t struct containing info about the kernel module
1587 * and the execution entry point for the module
1588 * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1589 * Defines the module initialisation order
1590 * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1591 * Defines the initialisation order of this kld relative to others
1592 * within the same subsystem as defined by param 3
1593 */
1594 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_LAST, SI_ORDER_ANY);
1595 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1596 MODULE_VERSION(siftr, MODVERSION);
1597