]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/netinet/siftr.c
Prevent cwnd to collapse down to 1 MSS after exiting recovery.
[FreeBSD/FreeBSD.git] / sys / netinet / siftr.c
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: lastewart@swin.edu.au
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/ip_var.h>
98 #include <netinet/tcp_var.h>
99
100 #ifdef SIFTR_IPV6
101 #include <netinet/ip6.h>
102 #include <netinet/ip6_var.h>
103 #include <netinet6/in6_pcb.h>
104 #endif /* SIFTR_IPV6 */
105
106 #include <machine/in_cksum.h>
107
108 /*
109  * Three digit version number refers to X.Y.Z where:
110  * X is the major version number
111  * Y is bumped to mark backwards incompatible changes
112  * Z is bumped to mark backwards compatible changes
113  */
114 #define V_MAJOR         1
115 #define V_BACKBREAK     2
116 #define V_BACKCOMPAT    4
117 #define MODVERSION      __CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT))
118 #define MODVERSION_STR  __XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \
119     __XSTRING(V_BACKCOMPAT)
120
121 #define HOOK 0
122 #define UNHOOK 1
123 #define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536
124 #define SYS_NAME "FreeBSD"
125 #define PACKET_TAG_SIFTR 100
126 #define PACKET_COOKIE_SIFTR 21749576
127 #define SIFTR_LOG_FILE_MODE 0644
128 #define SIFTR_DISABLE 0
129 #define SIFTR_ENABLE 1
130
131 /*
132  * Hard upper limit on the length of log messages. Bump this up if you add new
133  * data fields such that the line length could exceed the below value.
134  */
135 #define MAX_LOG_MSG_LEN 200
136 /* XXX: Make this a sysctl tunable. */
137 #define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN)
138
139 /*
140  * 1 byte for IP version
141  * IPv4: src/dst IP (4+4) + src/dst port (2+2) = 12 bytes
142  * IPv6: src/dst IP (16+16) + src/dst port (2+2) = 36 bytes
143  */
144 #ifdef SIFTR_IPV6
145 #define FLOW_KEY_LEN 37
146 #else
147 #define FLOW_KEY_LEN 13
148 #endif
149
150 #ifdef SIFTR_IPV6
151 #define SIFTR_IPMODE 6
152 #else
153 #define SIFTR_IPMODE 4
154 #endif
155
156 /* useful macros */
157 #define UPPER_SHORT(X)  (((X) & 0xFFFF0000) >> 16)
158 #define LOWER_SHORT(X)  ((X) & 0x0000FFFF)
159
160 #define FIRST_OCTET(X)  (((X) & 0xFF000000) >> 24)
161 #define SECOND_OCTET(X) (((X) & 0x00FF0000) >> 16)
162 #define THIRD_OCTET(X)  (((X) & 0x0000FF00) >> 8)
163 #define FOURTH_OCTET(X) ((X) & 0x000000FF)
164
165 static MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR");
166 static MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode",
167     "SIFTR pkt_node struct");
168 static MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode",
169     "SIFTR flow_hash_node struct");
170
171 /* Used as links in the pkt manager queue. */
172 struct pkt_node {
173         /* Timestamp of pkt as noted in the pfil hook. */
174         struct timeval          tval;
175         /* Direction pkt is travelling. */
176         enum {
177                 DIR_IN = 0,
178                 DIR_OUT = 1,
179         }                       direction;
180         /* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */
181         uint8_t                 ipver;
182         /* Hash of the pkt which triggered the log message. */
183         uint32_t                hash;
184         /* Local/foreign IP address. */
185 #ifdef SIFTR_IPV6
186         uint32_t                ip_laddr[4];
187         uint32_t                ip_faddr[4];
188 #else
189         uint8_t                 ip_laddr[4];
190         uint8_t                 ip_faddr[4];
191 #endif
192         /* Local TCP port. */
193         uint16_t                tcp_localport;
194         /* Foreign TCP port. */
195         uint16_t                tcp_foreignport;
196         /* Congestion Window (bytes). */
197         u_long                  snd_cwnd;
198         /* Sending Window (bytes). */
199         u_long                  snd_wnd;
200         /* Receive Window (bytes). */
201         u_long                  rcv_wnd;
202         /* Unused (was: Bandwidth Controlled Window (bytes)). */
203         u_long                  snd_bwnd;
204         /* Slow Start Threshold (bytes). */
205         u_long                  snd_ssthresh;
206         /* Current state of the TCP FSM. */
207         int                     conn_state;
208         /* Max Segment Size (bytes). */
209         u_int                   max_seg_size;
210         /*
211          * Smoothed RTT stored as found in the TCP control block
212          * in units of (TCP_RTT_SCALE*hz).
213          */
214         int                     smoothed_rtt;
215         /* Is SACK enabled? */
216         u_char                  sack_enabled;
217         /* Window scaling for snd window. */
218         u_char                  snd_scale;
219         /* Window scaling for recv window. */
220         u_char                  rcv_scale;
221         /* TCP control block flags. */
222         u_int                   flags;
223         /* Retransmit timeout length. */
224         int                     rxt_length;
225         /* Size of the TCP send buffer in bytes. */
226         u_int                   snd_buf_hiwater;
227         /* Current num bytes in the send socket buffer. */
228         u_int                   snd_buf_cc;
229         /* Size of the TCP receive buffer in bytes. */
230         u_int                   rcv_buf_hiwater;
231         /* Current num bytes in the receive socket buffer. */
232         u_int                   rcv_buf_cc;
233         /* Number of bytes inflight that we are waiting on ACKs for. */
234         u_int                   sent_inflight_bytes;
235         /* Number of segments currently in the reassembly queue. */
236         int                     t_segqlen;
237         /* Flowid for the connection. */
238         u_int                   flowid; 
239         /* Flow type for the connection. */
240         u_int                   flowtype;       
241         /* Link to next pkt_node in the list. */
242         STAILQ_ENTRY(pkt_node)  nodes;
243 };
244
245 struct flow_hash_node
246 {
247         uint16_t counter;
248         uint8_t key[FLOW_KEY_LEN];
249         LIST_ENTRY(flow_hash_node) nodes;
250 };
251
252 struct siftr_stats
253 {
254         /* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */
255         uint64_t n_in;
256         uint64_t n_out;
257         /* # pkts skipped due to failed malloc calls. */
258         uint32_t nskip_in_malloc;
259         uint32_t nskip_out_malloc;
260         /* # pkts skipped due to failed mtx acquisition. */
261         uint32_t nskip_in_mtx;
262         uint32_t nskip_out_mtx;
263         /* # pkts skipped due to failed inpcb lookups. */
264         uint32_t nskip_in_inpcb;
265         uint32_t nskip_out_inpcb;
266         /* # pkts skipped due to failed tcpcb lookups. */
267         uint32_t nskip_in_tcpcb;
268         uint32_t nskip_out_tcpcb;
269         /* # pkts skipped due to stack reinjection. */
270         uint32_t nskip_in_dejavu;
271         uint32_t nskip_out_dejavu;
272 };
273
274 DPCPU_DEFINE_STATIC(struct siftr_stats, ss);
275
276 static volatile unsigned int siftr_exit_pkt_manager_thread = 0;
277 static unsigned int siftr_enabled = 0;
278 static unsigned int siftr_pkts_per_log = 1;
279 static unsigned int siftr_generate_hashes = 0;
280 static uint16_t     siftr_port_filter = 0;
281 /* static unsigned int siftr_binary_log = 0; */
282 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log";
283 static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log";
284 static u_long siftr_hashmask;
285 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue);
286 LIST_HEAD(listhead, flow_hash_node) *counter_hash;
287 static int wait_for_pkt;
288 static struct alq *siftr_alq = NULL;
289 static struct mtx siftr_pkt_queue_mtx;
290 static struct mtx siftr_pkt_mgr_mtx;
291 static struct thread *siftr_pkt_manager_thr = NULL;
292 static char direction[2] = {'i','o'};
293
294 /* Required function prototypes. */
295 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS);
296 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS);
297
298
299 /* Declare the net.inet.siftr sysctl tree and populate it. */
300
301 SYSCTL_DECL(_net_inet_siftr);
302
303 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW, NULL,
304     "siftr related settings");
305
306 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, CTLTYPE_UINT|CTLFLAG_RW,
307     &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU",
308     "switch siftr module operations on/off");
309
310 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, CTLTYPE_STRING|CTLFLAG_RW,
311     &siftr_logfile_shadow, sizeof(siftr_logfile_shadow), &siftr_sysctl_logfile_name_handler,
312     "A", "file to save siftr log messages to");
313
314 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW,
315     &siftr_pkts_per_log, 1,
316     "number of packets between generating a log message");
317
318 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, genhashes, CTLFLAG_RW,
319     &siftr_generate_hashes, 0,
320     "enable packet hash generation");
321
322 SYSCTL_U16(_net_inet_siftr, OID_AUTO, port_filter, CTLFLAG_RW,
323     &siftr_port_filter, 0,
324     "enable packet filter on a TCP port");
325
326 /* XXX: TODO
327 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW,
328     &siftr_binary_log, 0,
329     "write log files in binary instead of ascii");
330 */
331
332
333 /* Begin functions. */
334
335 static void
336 siftr_process_pkt(struct pkt_node * pkt_node)
337 {
338         struct flow_hash_node *hash_node;
339         struct listhead *counter_list;
340         struct siftr_stats *ss;
341         struct ale *log_buf;
342         uint8_t key[FLOW_KEY_LEN];
343         uint8_t found_match, key_offset;
344
345         hash_node = NULL;
346         ss = DPCPU_PTR(ss);
347         found_match = 0;
348         key_offset = 1;
349
350         /*
351          * Create the key that will be used to create a hash index
352          * into our hash table. Our key consists of:
353          * ipversion, localip, localport, foreignip, foreignport
354          */
355         key[0] = pkt_node->ipver;
356         memcpy(key + key_offset, &pkt_node->ip_laddr,
357             sizeof(pkt_node->ip_laddr));
358         key_offset += sizeof(pkt_node->ip_laddr);
359         memcpy(key + key_offset, &pkt_node->tcp_localport,
360             sizeof(pkt_node->tcp_localport));
361         key_offset += sizeof(pkt_node->tcp_localport);
362         memcpy(key + key_offset, &pkt_node->ip_faddr,
363             sizeof(pkt_node->ip_faddr));
364         key_offset += sizeof(pkt_node->ip_faddr);
365         memcpy(key + key_offset, &pkt_node->tcp_foreignport,
366             sizeof(pkt_node->tcp_foreignport));
367
368         counter_list = counter_hash +
369             (hash32_buf(key, sizeof(key), 0) & siftr_hashmask);
370
371         /*
372          * If the list is not empty i.e. the hash index has
373          * been used by another flow previously.
374          */
375         if (LIST_FIRST(counter_list) != NULL) {
376                 /*
377                  * Loop through the hash nodes in the list.
378                  * There should normally only be 1 hash node in the list,
379                  * except if there have been collisions at the hash index
380                  * computed by hash32_buf().
381                  */
382                 LIST_FOREACH(hash_node, counter_list, nodes) {
383                         /*
384                          * Check if the key for the pkt we are currently
385                          * processing is the same as the key stored in the
386                          * hash node we are currently processing.
387                          * If they are the same, then we've found the
388                          * hash node that stores the counter for the flow
389                          * the pkt belongs to.
390                          */
391                         if (memcmp(hash_node->key, key, sizeof(key)) == 0) {
392                                 found_match = 1;
393                                 break;
394                         }
395                 }
396         }
397
398         /* If this flow hash hasn't been seen before or we have a collision. */
399         if (hash_node == NULL || !found_match) {
400                 /* Create a new hash node to store the flow's counter. */
401                 hash_node = malloc(sizeof(struct flow_hash_node),
402                     M_SIFTR_HASHNODE, M_WAITOK);
403
404                 if (hash_node != NULL) {
405                         /* Initialise our new hash node list entry. */
406                         hash_node->counter = 0;
407                         memcpy(hash_node->key, key, sizeof(key));
408                         LIST_INSERT_HEAD(counter_list, hash_node, nodes);
409                 } else {
410                         /* Malloc failed. */
411                         if (pkt_node->direction == DIR_IN)
412                                 ss->nskip_in_malloc++;
413                         else
414                                 ss->nskip_out_malloc++;
415
416                         return;
417                 }
418         } else if (siftr_pkts_per_log > 1) {
419                 /*
420                  * Taking the remainder of the counter divided
421                  * by the current value of siftr_pkts_per_log
422                  * and storing that in counter provides a neat
423                  * way to modulate the frequency of log
424                  * messages being written to the log file.
425                  */
426                 hash_node->counter = (hash_node->counter + 1) %
427                     siftr_pkts_per_log;
428
429                 /*
430                  * If we have not seen enough packets since the last time
431                  * we wrote a log message for this connection, return.
432                  */
433                 if (hash_node->counter > 0)
434                         return;
435         }
436
437         log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK);
438
439         if (log_buf == NULL)
440                 return; /* Should only happen if the ALQ is shutting down. */
441
442 #ifdef SIFTR_IPV6
443         pkt_node->ip_laddr[3] = ntohl(pkt_node->ip_laddr[3]);
444         pkt_node->ip_faddr[3] = ntohl(pkt_node->ip_faddr[3]);
445
446         if (pkt_node->ipver == INP_IPV6) { /* IPv6 packet */
447                 pkt_node->ip_laddr[0] = ntohl(pkt_node->ip_laddr[0]);
448                 pkt_node->ip_laddr[1] = ntohl(pkt_node->ip_laddr[1]);
449                 pkt_node->ip_laddr[2] = ntohl(pkt_node->ip_laddr[2]);
450                 pkt_node->ip_faddr[0] = ntohl(pkt_node->ip_faddr[0]);
451                 pkt_node->ip_faddr[1] = ntohl(pkt_node->ip_faddr[1]);
452                 pkt_node->ip_faddr[2] = ntohl(pkt_node->ip_faddr[2]);
453
454                 /* Construct an IPv6 log message. */
455                 log_buf->ae_bytesused = snprintf(log_buf->ae_data,
456                     MAX_LOG_MSG_LEN,
457                     "%c,0x%08x,%zd.%06ld,%x:%x:%x:%x:%x:%x:%x:%x,%u,%x:%x:%x:"
458                     "%x:%x:%x:%x:%x,%u,%ld,%ld,%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,"
459                     "%u,%d,%u,%u,%u,%u,%u,%u,%u,%u\n",
460                     direction[pkt_node->direction],
461                     pkt_node->hash,
462                     pkt_node->tval.tv_sec,
463                     pkt_node->tval.tv_usec,
464                     UPPER_SHORT(pkt_node->ip_laddr[0]),
465                     LOWER_SHORT(pkt_node->ip_laddr[0]),
466                     UPPER_SHORT(pkt_node->ip_laddr[1]),
467                     LOWER_SHORT(pkt_node->ip_laddr[1]),
468                     UPPER_SHORT(pkt_node->ip_laddr[2]),
469                     LOWER_SHORT(pkt_node->ip_laddr[2]),
470                     UPPER_SHORT(pkt_node->ip_laddr[3]),
471                     LOWER_SHORT(pkt_node->ip_laddr[3]),
472                     ntohs(pkt_node->tcp_localport),
473                     UPPER_SHORT(pkt_node->ip_faddr[0]),
474                     LOWER_SHORT(pkt_node->ip_faddr[0]),
475                     UPPER_SHORT(pkt_node->ip_faddr[1]),
476                     LOWER_SHORT(pkt_node->ip_faddr[1]),
477                     UPPER_SHORT(pkt_node->ip_faddr[2]),
478                     LOWER_SHORT(pkt_node->ip_faddr[2]),
479                     UPPER_SHORT(pkt_node->ip_faddr[3]),
480                     LOWER_SHORT(pkt_node->ip_faddr[3]),
481                     ntohs(pkt_node->tcp_foreignport),
482                     pkt_node->snd_ssthresh,
483                     pkt_node->snd_cwnd,
484                     pkt_node->snd_bwnd,
485                     pkt_node->snd_wnd,
486                     pkt_node->rcv_wnd,
487                     pkt_node->snd_scale,
488                     pkt_node->rcv_scale,
489                     pkt_node->conn_state,
490                     pkt_node->max_seg_size,
491                     pkt_node->smoothed_rtt,
492                     pkt_node->sack_enabled,
493                     pkt_node->flags,
494                     pkt_node->rxt_length,
495                     pkt_node->snd_buf_hiwater,
496                     pkt_node->snd_buf_cc,
497                     pkt_node->rcv_buf_hiwater,
498                     pkt_node->rcv_buf_cc,
499                     pkt_node->sent_inflight_bytes,
500                     pkt_node->t_segqlen,
501                     pkt_node->flowid,
502                     pkt_node->flowtype);
503         } else { /* IPv4 packet */
504                 pkt_node->ip_laddr[0] = FIRST_OCTET(pkt_node->ip_laddr[3]);
505                 pkt_node->ip_laddr[1] = SECOND_OCTET(pkt_node->ip_laddr[3]);
506                 pkt_node->ip_laddr[2] = THIRD_OCTET(pkt_node->ip_laddr[3]);
507                 pkt_node->ip_laddr[3] = FOURTH_OCTET(pkt_node->ip_laddr[3]);
508                 pkt_node->ip_faddr[0] = FIRST_OCTET(pkt_node->ip_faddr[3]);
509                 pkt_node->ip_faddr[1] = SECOND_OCTET(pkt_node->ip_faddr[3]);
510                 pkt_node->ip_faddr[2] = THIRD_OCTET(pkt_node->ip_faddr[3]);
511                 pkt_node->ip_faddr[3] = FOURTH_OCTET(pkt_node->ip_faddr[3]);
512 #endif /* SIFTR_IPV6 */
513
514                 /* Construct an IPv4 log message. */
515                 log_buf->ae_bytesused = snprintf(log_buf->ae_data,
516                     MAX_LOG_MSG_LEN,
517                     "%c,0x%08x,%jd.%06ld,%u.%u.%u.%u,%u,%u.%u.%u.%u,%u,%ld,%ld,"
518                     "%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,%u,%d,%u,%u,%u,%u,%u,%u,%u,%u\n",
519                     direction[pkt_node->direction],
520                     pkt_node->hash,
521                     (intmax_t)pkt_node->tval.tv_sec,
522                     pkt_node->tval.tv_usec,
523                     pkt_node->ip_laddr[0],
524                     pkt_node->ip_laddr[1],
525                     pkt_node->ip_laddr[2],
526                     pkt_node->ip_laddr[3],
527                     ntohs(pkt_node->tcp_localport),
528                     pkt_node->ip_faddr[0],
529                     pkt_node->ip_faddr[1],
530                     pkt_node->ip_faddr[2],
531                     pkt_node->ip_faddr[3],
532                     ntohs(pkt_node->tcp_foreignport),
533                     pkt_node->snd_ssthresh,
534                     pkt_node->snd_cwnd,
535                     pkt_node->snd_bwnd,
536                     pkt_node->snd_wnd,
537                     pkt_node->rcv_wnd,
538                     pkt_node->snd_scale,
539                     pkt_node->rcv_scale,
540                     pkt_node->conn_state,
541                     pkt_node->max_seg_size,
542                     pkt_node->smoothed_rtt,
543                     pkt_node->sack_enabled,
544                     pkt_node->flags,
545                     pkt_node->rxt_length,
546                     pkt_node->snd_buf_hiwater,
547                     pkt_node->snd_buf_cc,
548                     pkt_node->rcv_buf_hiwater,
549                     pkt_node->rcv_buf_cc,
550                     pkt_node->sent_inflight_bytes,
551                     pkt_node->t_segqlen,
552                     pkt_node->flowid,
553                     pkt_node->flowtype);
554 #ifdef SIFTR_IPV6
555         }
556 #endif
557
558         alq_post_flags(siftr_alq, log_buf, 0);
559 }
560
561
562 static void
563 siftr_pkt_manager_thread(void *arg)
564 {
565         STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
566             STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
567         struct pkt_node *pkt_node, *pkt_node_temp;
568         uint8_t draining;
569
570         draining = 2;
571
572         mtx_lock(&siftr_pkt_mgr_mtx);
573
574         /* draining == 0 when queue has been flushed and it's safe to exit. */
575         while (draining) {
576                 /*
577                  * Sleep until we are signalled to wake because thread has
578                  * been told to exit or until 1 tick has passed.
579                  */
580                 mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
581                     1);
582
583                 /* Gain exclusive access to the pkt_node queue. */
584                 mtx_lock(&siftr_pkt_queue_mtx);
585
586                 /*
587                  * Move pkt_queue to tmp_pkt_queue, which leaves
588                  * pkt_queue empty and ready to receive more pkt_nodes.
589                  */
590                 STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
591
592                 /*
593                  * We've finished making changes to the list. Unlock it
594                  * so the pfil hooks can continue queuing pkt_nodes.
595                  */
596                 mtx_unlock(&siftr_pkt_queue_mtx);
597
598                 /*
599                  * We can't hold a mutex whilst calling siftr_process_pkt
600                  * because ALQ might sleep waiting for buffer space.
601                  */
602                 mtx_unlock(&siftr_pkt_mgr_mtx);
603
604                 /* Flush all pkt_nodes to the log file. */
605                 STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes,
606                     pkt_node_temp) {
607                         siftr_process_pkt(pkt_node);
608                         STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
609                         free(pkt_node, M_SIFTR_PKTNODE);
610                 }
611
612                 KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
613                     ("SIFTR tmp_pkt_queue not empty after flush"));
614
615                 mtx_lock(&siftr_pkt_mgr_mtx);
616
617                 /*
618                  * If siftr_exit_pkt_manager_thread gets set during the window
619                  * where we are draining the tmp_pkt_queue above, there might
620                  * still be pkts in pkt_queue that need to be drained.
621                  * Allow one further iteration to occur after
622                  * siftr_exit_pkt_manager_thread has been set to ensure
623                  * pkt_queue is completely empty before we kill the thread.
624                  *
625                  * siftr_exit_pkt_manager_thread is set only after the pfil
626                  * hooks have been removed, so only 1 extra iteration
627                  * is needed to drain the queue.
628                  */
629                 if (siftr_exit_pkt_manager_thread)
630                         draining--;
631         }
632
633         mtx_unlock(&siftr_pkt_mgr_mtx);
634
635         /* Calls wakeup on this thread's struct thread ptr. */
636         kthread_exit();
637 }
638
639
640 static uint32_t
641 hash_pkt(struct mbuf *m, uint32_t offset)
642 {
643         uint32_t hash;
644
645         hash = 0;
646
647         while (m != NULL && offset > m->m_len) {
648                 /*
649                  * The IP packet payload does not start in this mbuf, so
650                  * need to figure out which mbuf it starts in and what offset
651                  * into the mbuf's data region the payload starts at.
652                  */
653                 offset -= m->m_len;
654                 m = m->m_next;
655         }
656
657         while (m != NULL) {
658                 /* Ensure there is data in the mbuf */
659                 if ((m->m_len - offset) > 0)
660                         hash = hash32_buf(m->m_data + offset,
661                             m->m_len - offset, hash);
662
663                 m = m->m_next;
664                 offset = 0;
665         }
666
667         return (hash);
668 }
669
670
671 /*
672  * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
673  * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
674  * Return value >0 means the caller should skip processing this mbuf.
675  */
676 static inline int
677 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
678 {
679         if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
680             != NULL) {
681                 if (dir == PFIL_IN)
682                         ss->nskip_in_dejavu++;
683                 else
684                         ss->nskip_out_dejavu++;
685
686                 return (1);
687         } else {
688                 struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
689                     PACKET_TAG_SIFTR, 0, M_NOWAIT);
690                 if (tag == NULL) {
691                         if (dir == PFIL_IN)
692                                 ss->nskip_in_malloc++;
693                         else
694                                 ss->nskip_out_malloc++;
695
696                         return (1);
697                 }
698
699                 m_tag_prepend(m, tag);
700         }
701
702         return (0);
703 }
704
705
706 /*
707  * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
708  * otherwise.
709  */
710 static inline struct inpcb *
711 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
712     uint16_t dport, int dir, struct siftr_stats *ss)
713 {
714         struct inpcb *inp;
715
716         /* We need the tcbinfo lock. */
717         INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
718
719         if (dir == PFIL_IN)
720                 inp = (ipver == INP_IPV4 ?
721                     in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
722                     dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
723                     :
724 #ifdef SIFTR_IPV6
725                     in6_pcblookup(&V_tcbinfo,
726                     &((struct ip6_hdr *)ip)->ip6_src, sport,
727                     &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB,
728                     m->m_pkthdr.rcvif)
729 #else
730                     NULL
731 #endif
732                     );
733
734         else
735                 inp = (ipver == INP_IPV4 ?
736                     in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
737                     sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
738                     :
739 #ifdef SIFTR_IPV6
740                     in6_pcblookup(&V_tcbinfo,
741                     &((struct ip6_hdr *)ip)->ip6_dst, dport,
742                     &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB,
743                     m->m_pkthdr.rcvif)
744 #else
745                     NULL
746 #endif
747                     );
748
749         /* If we can't find the inpcb, bail. */
750         if (inp == NULL) {
751                 if (dir == PFIL_IN)
752                         ss->nskip_in_inpcb++;
753                 else
754                         ss->nskip_out_inpcb++;
755         }
756
757         return (inp);
758 }
759
760
761 static inline void
762 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
763     int ipver, int dir, int inp_locally_locked)
764 {
765 #ifdef SIFTR_IPV6
766         if (ipver == INP_IPV4) {
767                 pn->ip_laddr[3] = inp->inp_laddr.s_addr;
768                 pn->ip_faddr[3] = inp->inp_faddr.s_addr;
769 #else
770                 *((uint32_t *)pn->ip_laddr) = inp->inp_laddr.s_addr;
771                 *((uint32_t *)pn->ip_faddr) = inp->inp_faddr.s_addr;
772 #endif
773 #ifdef SIFTR_IPV6
774         } else {
775                 pn->ip_laddr[0] = inp->in6p_laddr.s6_addr32[0];
776                 pn->ip_laddr[1] = inp->in6p_laddr.s6_addr32[1];
777                 pn->ip_laddr[2] = inp->in6p_laddr.s6_addr32[2];
778                 pn->ip_laddr[3] = inp->in6p_laddr.s6_addr32[3];
779                 pn->ip_faddr[0] = inp->in6p_faddr.s6_addr32[0];
780                 pn->ip_faddr[1] = inp->in6p_faddr.s6_addr32[1];
781                 pn->ip_faddr[2] = inp->in6p_faddr.s6_addr32[2];
782                 pn->ip_faddr[3] = inp->in6p_faddr.s6_addr32[3];
783         }
784 #endif
785         pn->tcp_localport = inp->inp_lport;
786         pn->tcp_foreignport = inp->inp_fport;
787         pn->snd_cwnd = tp->snd_cwnd;
788         pn->snd_wnd = tp->snd_wnd;
789         pn->rcv_wnd = tp->rcv_wnd;
790         pn->snd_bwnd = 0;               /* Unused, kept for compat. */
791         pn->snd_ssthresh = tp->snd_ssthresh;
792         pn->snd_scale = tp->snd_scale;
793         pn->rcv_scale = tp->rcv_scale;
794         pn->conn_state = tp->t_state;
795         pn->max_seg_size = tp->t_maxseg;
796         pn->smoothed_rtt = tp->t_srtt;
797         pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
798         pn->flags = tp->t_flags;
799         pn->rxt_length = tp->t_rxtcur;
800         pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
801         pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd);
802         pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
803         pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv);
804         pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
805         pn->t_segqlen = tp->t_segqlen;
806         pn->flowid = inp->inp_flowid;
807         pn->flowtype = inp->inp_flowtype;
808
809         /* We've finished accessing the tcb so release the lock. */
810         if (inp_locally_locked)
811                 INP_RUNLOCK(inp);
812
813         pn->ipver = ipver;
814         pn->direction = (dir == PFIL_IN ? DIR_IN : DIR_OUT);
815
816         /*
817          * Significantly more accurate than using getmicrotime(), but slower!
818          * Gives true microsecond resolution at the expense of a hit to
819          * maximum pps throughput processing when SIFTR is loaded and enabled.
820          */
821         microtime(&pn->tval);
822         TCP_PROBE1(siftr, &pn);
823
824 }
825
826
827 /*
828  * pfil hook that is called for each IPv4 packet making its way through the
829  * stack in either direction.
830  * The pfil subsystem holds a non-sleepable mutex somewhere when
831  * calling our hook function, so we can't sleep at all.
832  * It's very important to use the M_NOWAIT flag with all function calls
833  * that support it so that they won't sleep, otherwise you get a panic.
834  */
835 static pfil_return_t
836 siftr_chkpkt(struct mbuf **m, struct ifnet *ifp, int flags,
837     void *ruleset __unused, struct inpcb *inp)
838 {
839         struct pkt_node *pn;
840         struct ip *ip;
841         struct tcphdr *th;
842         struct tcpcb *tp;
843         struct siftr_stats *ss;
844         unsigned int ip_hl;
845         int inp_locally_locked, dir;
846
847         inp_locally_locked = 0;
848         dir = PFIL_DIR(flags);
849         ss = DPCPU_PTR(ss);
850
851         /*
852          * m_pullup is not required here because ip_{input|output}
853          * already do the heavy lifting for us.
854          */
855
856         ip = mtod(*m, struct ip *);
857
858         /* Only continue processing if the packet is TCP. */
859         if (ip->ip_p != IPPROTO_TCP)
860                 goto ret;
861
862         /*
863          * If a kernel subsystem reinjects packets into the stack, our pfil
864          * hook will be called multiple times for the same packet.
865          * Make sure we only process unique packets.
866          */
867         if (siftr_chkreinject(*m, dir, ss))
868                 goto ret;
869
870         if (dir == PFIL_IN)
871                 ss->n_in++;
872         else
873                 ss->n_out++;
874
875         /*
876          * Create a tcphdr struct starting at the correct offset
877          * in the IP packet. ip->ip_hl gives the ip header length
878          * in 4-byte words, so multiply it to get the size in bytes.
879          */
880         ip_hl = (ip->ip_hl << 2);
881         th = (struct tcphdr *)((caddr_t)ip + ip_hl);
882
883         /*
884          * If the pfil hooks don't provide a pointer to the
885          * inpcb, we need to find it ourselves and lock it.
886          */
887         if (!inp) {
888                 /* Find the corresponding inpcb for this pkt. */
889                 inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
890                     th->th_dport, dir, ss);
891
892                 if (inp == NULL)
893                         goto ret;
894                 else
895                         inp_locally_locked = 1;
896         }
897
898         INP_LOCK_ASSERT(inp);
899
900         /* Find the TCP control block that corresponds with this packet */
901         tp = intotcpcb(inp);
902
903         /*
904          * If we can't find the TCP control block (happens occasionaly for a
905          * packet sent during the shutdown phase of a TCP connection),
906          * or we're in the timewait state, bail
907          */
908         if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
909                 if (dir == PFIL_IN)
910                         ss->nskip_in_tcpcb++;
911                 else
912                         ss->nskip_out_tcpcb++;
913
914                 goto inp_unlock;
915         }
916
917         /*
918          * Only pkts selected by the tcp port filter
919          * can be inserted into the pkt_queue
920          */
921         if ((siftr_port_filter != 0) && 
922             (siftr_port_filter != ntohs(inp->inp_lport)) &&
923             (siftr_port_filter != ntohs(inp->inp_fport))) {
924                 goto inp_unlock;
925         }
926
927         pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
928
929         if (pn == NULL) {
930                 if (dir == PFIL_IN)
931                         ss->nskip_in_malloc++;
932                 else
933                         ss->nskip_out_malloc++;
934
935                 goto inp_unlock;
936         }
937
938         siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
939
940         if (siftr_generate_hashes) {
941                 if ((*m)->m_pkthdr.csum_flags & CSUM_TCP) {
942                         /*
943                          * For outbound packets, the TCP checksum isn't
944                          * calculated yet. This is a problem for our packet
945                          * hashing as the receiver will calc a different hash
946                          * to ours if we don't include the correct TCP checksum
947                          * in the bytes being hashed. To work around this
948                          * problem, we manually calc the TCP checksum here in
949                          * software. We unset the CSUM_TCP flag so the lower
950                          * layers don't recalc it.
951                          */
952                         (*m)->m_pkthdr.csum_flags &= ~CSUM_TCP;
953
954                         /*
955                          * Calculate the TCP checksum in software and assign
956                          * to correct TCP header field, which will follow the
957                          * packet mbuf down the stack. The trick here is that
958                          * tcp_output() sets th->th_sum to the checksum of the
959                          * pseudo header for us already. Because of the nature
960                          * of the checksumming algorithm, we can sum over the
961                          * entire IP payload (i.e. TCP header and data), which
962                          * will include the already calculated pseduo header
963                          * checksum, thus giving us the complete TCP checksum.
964                          *
965                          * To put it in simple terms, if checksum(1,2,3,4)=10,
966                          * then checksum(1,2,3,4,5) == checksum(10,5).
967                          * This property is what allows us to "cheat" and
968                          * checksum only the IP payload which has the TCP
969                          * th_sum field populated with the pseudo header's
970                          * checksum, and not need to futz around checksumming
971                          * pseudo header bytes and TCP header/data in one hit.
972                          * Refer to RFC 1071 for more info.
973                          *
974                          * NB: in_cksum_skip(struct mbuf *m, int len, int skip)
975                          * in_cksum_skip 2nd argument is NOT the number of
976                          * bytes to read from the mbuf at "skip" bytes offset
977                          * from the start of the mbuf (very counter intuitive!).
978                          * The number of bytes to read is calculated internally
979                          * by the function as len-skip i.e. to sum over the IP
980                          * payload (TCP header + data) bytes, it is INCORRECT
981                          * to call the function like this:
982                          * in_cksum_skip(at, ip->ip_len - offset, offset)
983                          * Rather, it should be called like this:
984                          * in_cksum_skip(at, ip->ip_len, offset)
985                          * which means read "ip->ip_len - offset" bytes from
986                          * the mbuf cluster "at" at offset "offset" bytes from
987                          * the beginning of the "at" mbuf's data pointer.
988                          */
989                         th->th_sum = in_cksum_skip(*m, ntohs(ip->ip_len),
990                             ip_hl);
991                 }
992
993                 /*
994                  * XXX: Having to calculate the checksum in software and then
995                  * hash over all bytes is really inefficient. Would be nice to
996                  * find a way to create the hash and checksum in the same pass
997                  * over the bytes.
998                  */
999                 pn->hash = hash_pkt(*m, ip_hl);
1000         }
1001
1002         mtx_lock(&siftr_pkt_queue_mtx);
1003         STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
1004         mtx_unlock(&siftr_pkt_queue_mtx);
1005         goto ret;
1006
1007 inp_unlock:
1008         if (inp_locally_locked)
1009                 INP_RUNLOCK(inp);
1010
1011 ret:
1012         return (PFIL_PASS);
1013 }
1014
1015
1016 #ifdef SIFTR_IPV6
1017 static int
1018 siftr_chkpkt6(struct mbuf **m, struct ifnet *ifp, int flags, 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, dir;
1027
1028         inp_locally_locked = 0;
1029         dir = PFIL_DIR(flags);
1030         ss = DPCPU_PTR(ss);
1031
1032         /*
1033          * m_pullup is not required here because ip6_{input|output}
1034          * already do the heavy lifting for us.
1035          */
1036
1037         ip6 = mtod(*m, struct ip6_hdr *);
1038
1039         /*
1040          * Only continue processing if the packet is TCP
1041          * XXX: We should follow the next header fields
1042          * as shown on Pg 6 RFC 2460, but right now we'll
1043          * only check pkts that have no extension headers.
1044          */
1045         if (ip6->ip6_nxt != IPPROTO_TCP)
1046                 goto ret6;
1047
1048         /*
1049          * If a kernel subsystem reinjects packets into the stack, our pfil
1050          * hook will be called multiple times for the same packet.
1051          * Make sure we only process unique packets.
1052          */
1053         if (siftr_chkreinject(*m, dir, ss))
1054                 goto ret6;
1055
1056         if (dir == PFIL_IN)
1057                 ss->n_in++;
1058         else
1059                 ss->n_out++;
1060
1061         ip6_hl = sizeof(struct ip6_hdr);
1062
1063         /*
1064          * Create a tcphdr struct starting at the correct offset
1065          * in the ipv6 packet. ip->ip_hl gives the ip header length
1066          * in 4-byte words, so multiply it to get the size in bytes.
1067          */
1068         th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
1069
1070         /*
1071          * For inbound packets, the pfil hooks don't provide a pointer to the
1072          * inpcb, so we need to find it ourselves and lock it.
1073          */
1074         if (!inp) {
1075                 /* Find the corresponding inpcb for this pkt. */
1076                 inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
1077                     th->th_sport, th->th_dport, dir, ss);
1078
1079                 if (inp == NULL)
1080                         goto ret6;
1081                 else
1082                         inp_locally_locked = 1;
1083         }
1084
1085         /* Find the TCP control block that corresponds with this packet. */
1086         tp = intotcpcb(inp);
1087
1088         /*
1089          * If we can't find the TCP control block (happens occasionaly for a
1090          * packet sent during the shutdown phase of a TCP connection),
1091          * or we're in the timewait state, bail.
1092          */
1093         if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
1094                 if (dir == PFIL_IN)
1095                         ss->nskip_in_tcpcb++;
1096                 else
1097                         ss->nskip_out_tcpcb++;
1098
1099                 goto inp_unlock6;
1100         }
1101
1102         /*
1103          * Only pkts selected by the tcp port filter
1104          * can be inserted into the pkt_queue
1105          */
1106         if ((siftr_port_filter != 0) && 
1107             (siftr_port_filter != ntohs(inp->inp_lport)) &&
1108             (siftr_port_filter != ntohs(inp->inp_fport))) {
1109                 goto inp_unlock6;
1110         }
1111
1112         pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
1113
1114         if (pn == NULL) {
1115                 if (dir == PFIL_IN)
1116                         ss->nskip_in_malloc++;
1117                 else
1118                         ss->nskip_out_malloc++;
1119
1120                 goto inp_unlock6;
1121         }
1122
1123         siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
1124
1125         /* XXX: Figure out how to generate hashes for IPv6 packets. */
1126
1127         mtx_lock(&siftr_pkt_queue_mtx);
1128         STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
1129         mtx_unlock(&siftr_pkt_queue_mtx);
1130         goto ret6;
1131
1132 inp_unlock6:
1133         if (inp_locally_locked)
1134                 INP_RUNLOCK(inp);
1135
1136 ret6:
1137         /* Returning 0 ensures pfil will not discard the pkt. */
1138         return (0);
1139 }
1140 #endif /* #ifdef SIFTR_IPV6 */
1141
1142 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet_hook);
1143 #define V_siftr_inet_hook       VNET(siftr_inet_hook)
1144 #ifdef INET6
1145 VNET_DEFINE_STATIC(pfil_hook_t, siftr_inet6_hook);
1146 #define V_siftr_inet6_hook      VNET(siftr_inet6_hook)
1147 #endif
1148 static int
1149 siftr_pfil(int action)
1150 {
1151         struct pfil_hook_args pha;
1152         struct pfil_link_args pla;
1153
1154         pha.pa_version = PFIL_VERSION;
1155         pha.pa_flags = PFIL_IN | PFIL_OUT;
1156         pha.pa_modname = "siftr";
1157         pha.pa_ruleset = NULL;
1158         pha.pa_rulname = "default";
1159
1160         pla.pa_version = PFIL_VERSION;
1161         pla.pa_flags = PFIL_IN | PFIL_OUT |
1162             PFIL_HEADPTR | PFIL_HOOKPTR;
1163
1164         VNET_ITERATOR_DECL(vnet_iter);
1165
1166         VNET_LIST_RLOCK();
1167         VNET_FOREACH(vnet_iter) {
1168                 CURVNET_SET(vnet_iter);
1169
1170                 if (action == HOOK) {
1171                         pha.pa_func = siftr_chkpkt;
1172                         pha.pa_type = PFIL_TYPE_IP4;
1173                         V_siftr_inet_hook = pfil_add_hook(&pha);
1174                         pla.pa_hook = V_siftr_inet_hook;
1175                         pla.pa_head = V_inet_pfil_head;
1176                         (void)pfil_link(&pla);
1177 #ifdef SIFTR_IPV6
1178                         pha.pa_func = siftr_chkpkt6;
1179                         pha.pa_type = PFIL_TYPE_IP6;
1180                         V_siftr_inet6_hook = pfil_add_hook(&pha);
1181                         pla.pa_hook = V_siftr_inet6_hook;
1182                         pla.pa_head = V_inet6_pfil_head;
1183                         (void)pfil_link(&pla);
1184 #endif
1185                 } else if (action == UNHOOK) {
1186                         pfil_remove_hook(V_siftr_inet_hook);
1187 #ifdef SIFTR_IPV6
1188                         pfil_remove_hook(V_siftr_inet6_hook);
1189 #endif
1190                 }
1191                 CURVNET_RESTORE();
1192         }
1193         VNET_LIST_RUNLOCK();
1194
1195         return (0);
1196 }
1197
1198
1199 static int
1200 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1201 {
1202         struct alq *new_alq;
1203         int error;
1204
1205         error = sysctl_handle_string(oidp, arg1, arg2, req);
1206
1207         /* Check for error or same filename */
1208         if (error != 0 || req->newptr == NULL ||
1209             strncmp(siftr_logfile, arg1, arg2) == 0)
1210                 goto done;
1211
1212         /* Filname changed */
1213         error = alq_open(&new_alq, arg1, curthread->td_ucred,
1214             SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1215         if (error != 0)
1216                 goto done;
1217
1218         /*
1219          * If disabled, siftr_alq == NULL so we simply close
1220          * the alq as we've proved it can be opened.
1221          * If enabled, close the existing alq and switch the old
1222          * for the new.
1223          */
1224         if (siftr_alq == NULL) {
1225                 alq_close(new_alq);
1226         } else {
1227                 alq_close(siftr_alq);
1228                 siftr_alq = new_alq;
1229         }
1230
1231         /* Update filename upon success */
1232         strlcpy(siftr_logfile, arg1, arg2);
1233 done:
1234         return (error);
1235 }
1236
1237 static int
1238 siftr_manage_ops(uint8_t action)
1239 {
1240         struct siftr_stats totalss;
1241         struct timeval tval;
1242         struct flow_hash_node *counter, *tmp_counter;
1243         struct sbuf *s;
1244         int i, key_index, error;
1245         uint32_t bytes_to_write, total_skipped_pkts;
1246         uint16_t lport, fport;
1247         uint8_t *key, ipver __unused;
1248
1249 #ifdef SIFTR_IPV6
1250         uint32_t laddr[4];
1251         uint32_t faddr[4];
1252 #else
1253         uint8_t laddr[4];
1254         uint8_t faddr[4];
1255 #endif
1256
1257         error = 0;
1258         total_skipped_pkts = 0;
1259
1260         /* Init an autosizing sbuf that initially holds 200 chars. */
1261         if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1262                 return (-1);
1263
1264         if (action == SIFTR_ENABLE && siftr_pkt_manager_thr == NULL) {
1265                 /*
1266                  * Create our alq
1267                  * XXX: We should abort if alq_open fails!
1268                  */
1269                 alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1270                     SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1271
1272                 STAILQ_INIT(&pkt_queue);
1273
1274                 DPCPU_ZERO(ss);
1275
1276                 siftr_exit_pkt_manager_thread = 0;
1277
1278                 kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1279                     &siftr_pkt_manager_thr, RFNOWAIT, 0,
1280                     "siftr_pkt_manager_thr");
1281
1282                 siftr_pfil(HOOK);
1283
1284                 microtime(&tval);
1285
1286                 sbuf_printf(s,
1287                     "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1288                     "siftrver=%s\thz=%u\ttcp_rtt_scale=%u\tsysname=%s\t"
1289                     "sysver=%u\tipmode=%u\n",
1290                     (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, hz,
1291                     TCP_RTT_SCALE, SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1292
1293                 sbuf_finish(s);
1294                 alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1295
1296         } else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1297                 /*
1298                  * Remove the pfil hook functions. All threads currently in
1299                  * the hook functions are allowed to exit before siftr_pfil()
1300                  * returns.
1301                  */
1302                 siftr_pfil(UNHOOK);
1303
1304                 /* This will block until the pkt manager thread unlocks it. */
1305                 mtx_lock(&siftr_pkt_mgr_mtx);
1306
1307                 /* Tell the pkt manager thread that it should exit now. */
1308                 siftr_exit_pkt_manager_thread = 1;
1309
1310                 /*
1311                  * Wake the pkt_manager thread so it realises that
1312                  * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1313                  * The wakeup won't be delivered until we unlock
1314                  * siftr_pkt_mgr_mtx so this isn't racy.
1315                  */
1316                 wakeup(&wait_for_pkt);
1317
1318                 /* Wait for the pkt_manager thread to exit. */
1319                 mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1320                     "thrwait", 0);
1321
1322                 siftr_pkt_manager_thr = NULL;
1323                 mtx_unlock(&siftr_pkt_mgr_mtx);
1324
1325                 totalss.n_in = DPCPU_VARSUM(ss, n_in);
1326                 totalss.n_out = DPCPU_VARSUM(ss, n_out);
1327                 totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1328                 totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1329                 totalss.nskip_in_mtx = DPCPU_VARSUM(ss, nskip_in_mtx);
1330                 totalss.nskip_out_mtx = DPCPU_VARSUM(ss, nskip_out_mtx);
1331                 totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1332                 totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1333                 totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1334                 totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1335
1336                 total_skipped_pkts = totalss.nskip_in_malloc +
1337                     totalss.nskip_out_malloc + totalss.nskip_in_mtx +
1338                     totalss.nskip_out_mtx + totalss.nskip_in_tcpcb +
1339                     totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1340                     totalss.nskip_out_inpcb;
1341
1342                 microtime(&tval);
1343
1344                 sbuf_printf(s,
1345                     "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1346                     "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1347                     "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1348                     "num_outbound_skipped_pkts_malloc=%u\t"
1349                     "num_inbound_skipped_pkts_mtx=%u\t"
1350                     "num_outbound_skipped_pkts_mtx=%u\t"
1351                     "num_inbound_skipped_pkts_tcpcb=%u\t"
1352                     "num_outbound_skipped_pkts_tcpcb=%u\t"
1353                     "num_inbound_skipped_pkts_inpcb=%u\t"
1354                     "num_outbound_skipped_pkts_inpcb=%u\t"
1355                     "total_skipped_tcp_pkts=%u\tflow_list=",
1356                     (intmax_t)tval.tv_sec,
1357                     tval.tv_usec,
1358                     (uintmax_t)totalss.n_in,
1359                     (uintmax_t)totalss.n_out,
1360                     (uintmax_t)(totalss.n_in + totalss.n_out),
1361                     totalss.nskip_in_malloc,
1362                     totalss.nskip_out_malloc,
1363                     totalss.nskip_in_mtx,
1364                     totalss.nskip_out_mtx,
1365                     totalss.nskip_in_tcpcb,
1366                     totalss.nskip_out_tcpcb,
1367                     totalss.nskip_in_inpcb,
1368                     totalss.nskip_out_inpcb,
1369                     total_skipped_pkts);
1370
1371                 /*
1372                  * Iterate over the flow hash, printing a summary of each
1373                  * flow seen and freeing any malloc'd memory.
1374                  * The hash consists of an array of LISTs (man 3 queue).
1375                  */
1376                 for (i = 0; i <= siftr_hashmask; i++) {
1377                         LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1378                             tmp_counter) {
1379                                 key = counter->key;
1380                                 key_index = 1;
1381
1382                                 ipver = key[0];
1383
1384                                 memcpy(laddr, key + key_index, sizeof(laddr));
1385                                 key_index += sizeof(laddr);
1386                                 memcpy(&lport, key + key_index, sizeof(lport));
1387                                 key_index += sizeof(lport);
1388                                 memcpy(faddr, key + key_index, sizeof(faddr));
1389                                 key_index += sizeof(faddr);
1390                                 memcpy(&fport, key + key_index, sizeof(fport));
1391
1392 #ifdef SIFTR_IPV6
1393                                 laddr[3] = ntohl(laddr[3]);
1394                                 faddr[3] = ntohl(faddr[3]);
1395
1396                                 if (ipver == INP_IPV6) {
1397                                         laddr[0] = ntohl(laddr[0]);
1398                                         laddr[1] = ntohl(laddr[1]);
1399                                         laddr[2] = ntohl(laddr[2]);
1400                                         faddr[0] = ntohl(faddr[0]);
1401                                         faddr[1] = ntohl(faddr[1]);
1402                                         faddr[2] = ntohl(faddr[2]);
1403
1404                                         sbuf_printf(s,
1405                                             "%x:%x:%x:%x:%x:%x:%x:%x;%u-"
1406                                             "%x:%x:%x:%x:%x:%x:%x:%x;%u,",
1407                                             UPPER_SHORT(laddr[0]),
1408                                             LOWER_SHORT(laddr[0]),
1409                                             UPPER_SHORT(laddr[1]),
1410                                             LOWER_SHORT(laddr[1]),
1411                                             UPPER_SHORT(laddr[2]),
1412                                             LOWER_SHORT(laddr[2]),
1413                                             UPPER_SHORT(laddr[3]),
1414                                             LOWER_SHORT(laddr[3]),
1415                                             ntohs(lport),
1416                                             UPPER_SHORT(faddr[0]),
1417                                             LOWER_SHORT(faddr[0]),
1418                                             UPPER_SHORT(faddr[1]),
1419                                             LOWER_SHORT(faddr[1]),
1420                                             UPPER_SHORT(faddr[2]),
1421                                             LOWER_SHORT(faddr[2]),
1422                                             UPPER_SHORT(faddr[3]),
1423                                             LOWER_SHORT(faddr[3]),
1424                                             ntohs(fport));
1425                                 } else {
1426                                         laddr[0] = FIRST_OCTET(laddr[3]);
1427                                         laddr[1] = SECOND_OCTET(laddr[3]);
1428                                         laddr[2] = THIRD_OCTET(laddr[3]);
1429                                         laddr[3] = FOURTH_OCTET(laddr[3]);
1430                                         faddr[0] = FIRST_OCTET(faddr[3]);
1431                                         faddr[1] = SECOND_OCTET(faddr[3]);
1432                                         faddr[2] = THIRD_OCTET(faddr[3]);
1433                                         faddr[3] = FOURTH_OCTET(faddr[3]);
1434 #endif
1435                                         sbuf_printf(s,
1436                                             "%u.%u.%u.%u;%u-%u.%u.%u.%u;%u,",
1437                                             laddr[0],
1438                                             laddr[1],
1439                                             laddr[2],
1440                                             laddr[3],
1441                                             ntohs(lport),
1442                                             faddr[0],
1443                                             faddr[1],
1444                                             faddr[2],
1445                                             faddr[3],
1446                                             ntohs(fport));
1447 #ifdef SIFTR_IPV6
1448                                 }
1449 #endif
1450
1451                                 free(counter, M_SIFTR_HASHNODE);
1452                         }
1453
1454                         LIST_INIT(counter_hash + i);
1455                 }
1456
1457                 sbuf_printf(s, "\n");
1458                 sbuf_finish(s);
1459
1460                 i = 0;
1461                 do {
1462                         bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1463                         alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1464                         i += bytes_to_write;
1465                 } while (i < sbuf_len(s));
1466
1467                 alq_close(siftr_alq);
1468                 siftr_alq = NULL;
1469         } else
1470                 error = EINVAL;
1471
1472         sbuf_delete(s);
1473
1474         /*
1475          * XXX: Should be using ret to check if any functions fail
1476          * and set error appropriately
1477          */
1478
1479         return (error);
1480 }
1481
1482
1483 static int
1484 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1485 {
1486         int error;
1487         uint32_t new;
1488
1489         new = siftr_enabled;
1490         error = sysctl_handle_int(oidp, &new, 0, req);
1491         if (error == 0 && req->newptr != NULL) {
1492                 if (new > 1)
1493                         return (EINVAL);
1494                 else if (new != siftr_enabled) {
1495                         if ((error = siftr_manage_ops(new)) == 0) {
1496                                 siftr_enabled = new;
1497                         } else {
1498                                 siftr_manage_ops(SIFTR_DISABLE);
1499                         }
1500                 }
1501         }
1502
1503         return (error);
1504 }
1505
1506
1507 static void
1508 siftr_shutdown_handler(void *arg)
1509 {
1510         if (siftr_enabled == 1) {
1511                 siftr_manage_ops(SIFTR_DISABLE);
1512         }
1513 }
1514
1515
1516 /*
1517  * Module is being unloaded or machine is shutting down. Take care of cleanup.
1518  */
1519 static int
1520 deinit_siftr(void)
1521 {
1522         /* Cleanup. */
1523         siftr_manage_ops(SIFTR_DISABLE);
1524         hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1525         mtx_destroy(&siftr_pkt_queue_mtx);
1526         mtx_destroy(&siftr_pkt_mgr_mtx);
1527
1528         return (0);
1529 }
1530
1531
1532 /*
1533  * Module has just been loaded into the kernel.
1534  */
1535 static int
1536 init_siftr(void)
1537 {
1538         EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL,
1539             SHUTDOWN_PRI_FIRST);
1540
1541         /* Initialise our flow counter hash table. */
1542         counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1543             &siftr_hashmask);
1544
1545         mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1546         mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1547
1548         /* Print message to the user's current terminal. */
1549         uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1550             "          http://caia.swin.edu.au/urp/newtcp\n\n",
1551             MODVERSION_STR);
1552
1553         return (0);
1554 }
1555
1556
1557 /*
1558  * This is the function that is called to load and unload the module.
1559  * When the module is loaded, this function is called once with
1560  * "what" == MOD_LOAD
1561  * When the module is unloaded, this function is called twice with
1562  * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1563  * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1564  * this function is called once with "what" = MOD_SHUTDOWN
1565  * When the system is shut down, the handler isn't called until the very end
1566  * of the shutdown sequence i.e. after the disks have been synced.
1567  */
1568 static int
1569 siftr_load_handler(module_t mod, int what, void *arg)
1570 {
1571         int ret;
1572
1573         switch (what) {
1574         case MOD_LOAD:
1575                 ret = init_siftr();
1576                 break;
1577
1578         case MOD_QUIESCE:
1579         case MOD_SHUTDOWN:
1580                 ret = deinit_siftr();
1581                 break;
1582
1583         case MOD_UNLOAD:
1584                 ret = 0;
1585                 break;
1586
1587         default:
1588                 ret = EINVAL;
1589                 break;
1590         }
1591
1592         return (ret);
1593 }
1594
1595
1596 static moduledata_t siftr_mod = {
1597         .name = "siftr",
1598         .evhand = siftr_load_handler,
1599 };
1600
1601 /*
1602  * Param 1: name of the kernel module
1603  * Param 2: moduledata_t struct containing info about the kernel module
1604  *          and the execution entry point for the module
1605  * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1606  *          Defines the module initialisation order
1607  * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1608  *          Defines the initialisation order of this kld relative to others
1609  *          within the same subsystem as defined by param 3
1610  */
1611 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_LAST, SI_ORDER_ANY);
1612 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1613 MODULE_VERSION(siftr, MODVERSION);