xref: /freebsd/sys/netgraph/ng_pipe.c (revision ba3c1f5972d7b90feb6e6da47905ff2757e0fe57)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2004-2010 University of Zagreb
5  * Copyright (c) 2007-2008 FreeBSD Foundation
6  *
7  * This software was developed by the University of Zagreb and the
8  * FreeBSD Foundation under sponsorship by the Stichting NLnet and the
9  * FreeBSD Foundation.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  * $FreeBSD$
33  */
34 
35 /*
36  * This node permits simple traffic shaping by emulating bandwidth
37  * and delay, as well as random packet losses.
38  * The node has two hooks, upper and lower. Traffic flowing from upper to
39  * lower hook is referenced as downstream, and vice versa. Parameters for
40  * both directions can be set separately, except for delay.
41  */
42 
43 #include <sys/param.h>
44 #include <sys/errno.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/time.h>
50 
51 #include <vm/uma.h>
52 
53 #include <net/vnet.h>
54 
55 #include <netinet/in.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/ip.h>
58 
59 #include <netgraph/ng_message.h>
60 #include <netgraph/netgraph.h>
61 #include <netgraph/ng_parse.h>
62 #include <netgraph/ng_pipe.h>
63 
64 static MALLOC_DEFINE(M_NG_PIPE, "ng_pipe", "ng_pipe");
65 
66 /* Packet header struct */
67 struct ngp_hdr {
68 	TAILQ_ENTRY(ngp_hdr)	ngp_link;	/* next pkt in queue */
69 	struct timeval		when;		/* this packet's due time */
70 	struct mbuf		*m;		/* ptr to the packet data */
71 };
72 TAILQ_HEAD(p_head, ngp_hdr);
73 
74 /* FIFO queue struct */
75 struct ngp_fifo {
76 	TAILQ_ENTRY(ngp_fifo)	fifo_le;	/* list of active queues only */
77 	struct p_head		packet_head;	/* FIFO queue head */
78 	u_int32_t		hash;		/* flow signature */
79 	struct timeval		vtime;		/* virtual time, for WFQ */
80 	u_int32_t		rr_deficit;	/* for DRR */
81 	u_int32_t		packets;	/* # of packets in this queue */
82 };
83 
84 /* Per hook info */
85 struct hookinfo {
86 	hook_p			hook;
87 	int			noqueue;	/* bypass any processing */
88 	TAILQ_HEAD(, ngp_fifo)	fifo_head;	/* FIFO queues */
89 	TAILQ_HEAD(, ngp_hdr)	qout_head;	/* delay queue head */
90 	struct timeval		qin_utime;
91 	struct ng_pipe_hookcfg	cfg;
92 	struct ng_pipe_hookrun	run;
93 	struct ng_pipe_hookstat	stats;
94 	uint64_t		*ber_p;		/* loss_p(BER,psize) map */
95 };
96 
97 /* Per node info */
98 struct node_priv {
99 	u_int64_t		delay;
100 	u_int32_t		overhead;
101 	u_int32_t		header_offset;
102 	struct hookinfo		lower;
103 	struct hookinfo		upper;
104 	struct callout		timer;
105 	int			timer_scheduled;
106 };
107 typedef struct node_priv *priv_p;
108 
109 /* Macro for calculating the virtual time for packet dequeueing in WFQ */
110 #define FIFO_VTIME_SORT(plen)						\
111 	if (hinfo->cfg.wfq && hinfo->cfg.bandwidth) {			\
112 		ngp_f->vtime.tv_usec = now->tv_usec + ((uint64_t) (plen) \
113 			+ priv->overhead ) * hinfo->run.fifo_queues *	\
114 			8000000 / hinfo->cfg.bandwidth;			\
115 		ngp_f->vtime.tv_sec = now->tv_sec +			\
116 			ngp_f->vtime.tv_usec / 1000000;			\
117 		ngp_f->vtime.tv_usec = ngp_f->vtime.tv_usec % 1000000;	\
118 		TAILQ_FOREACH(ngp_f1, &hinfo->fifo_head, fifo_le)	\
119 			if (ngp_f1->vtime.tv_sec > ngp_f->vtime.tv_sec || \
120 			    (ngp_f1->vtime.tv_sec == ngp_f->vtime.tv_sec && \
121 			    ngp_f1->vtime.tv_usec > ngp_f->vtime.tv_usec)) \
122 				break;					\
123 		if (ngp_f1 == NULL)					\
124 			TAILQ_INSERT_TAIL(&hinfo->fifo_head, ngp_f, fifo_le); \
125 		else							\
126 			TAILQ_INSERT_BEFORE(ngp_f1, ngp_f, fifo_le);	\
127 	} else								\
128 		TAILQ_INSERT_TAIL(&hinfo->fifo_head, ngp_f, fifo_le);	\
129 
130 static void	parse_cfg(struct ng_pipe_hookcfg *, struct ng_pipe_hookcfg *,
131 			struct hookinfo *, priv_p);
132 static void	pipe_dequeue(struct hookinfo *, struct timeval *);
133 static void	ngp_callout(node_p, hook_p, void *, int);
134 static int	ngp_modevent(module_t, int, void *);
135 
136 /* zone for storing ngp_hdr-s */
137 static uma_zone_t ngp_zone;
138 
139 /* Netgraph methods */
140 static ng_constructor_t	ngp_constructor;
141 static ng_rcvmsg_t	ngp_rcvmsg;
142 static ng_shutdown_t	ngp_shutdown;
143 static ng_newhook_t	ngp_newhook;
144 static ng_rcvdata_t	ngp_rcvdata;
145 static ng_disconnect_t	ngp_disconnect;
146 
147 /* Parse type for struct ng_pipe_hookstat */
148 static const struct ng_parse_struct_field
149 	ng_pipe_hookstat_type_fields[] = NG_PIPE_HOOKSTAT_INFO;
150 static const struct ng_parse_type ng_pipe_hookstat_type = {
151 	&ng_parse_struct_type,
152 	&ng_pipe_hookstat_type_fields
153 };
154 
155 /* Parse type for struct ng_pipe_stats */
156 static const struct ng_parse_struct_field ng_pipe_stats_type_fields[] =
157 	NG_PIPE_STATS_INFO(&ng_pipe_hookstat_type);
158 static const struct ng_parse_type ng_pipe_stats_type = {
159 	&ng_parse_struct_type,
160 	&ng_pipe_stats_type_fields
161 };
162 
163 /* Parse type for struct ng_pipe_hookrun */
164 static const struct ng_parse_struct_field
165 	ng_pipe_hookrun_type_fields[] = NG_PIPE_HOOKRUN_INFO;
166 static const struct ng_parse_type ng_pipe_hookrun_type = {
167 	&ng_parse_struct_type,
168 	&ng_pipe_hookrun_type_fields
169 };
170 
171 /* Parse type for struct ng_pipe_run */
172 static const struct ng_parse_struct_field
173 	ng_pipe_run_type_fields[] = NG_PIPE_RUN_INFO(&ng_pipe_hookrun_type);
174 static const struct ng_parse_type ng_pipe_run_type = {
175 	&ng_parse_struct_type,
176 	&ng_pipe_run_type_fields
177 };
178 
179 /* Parse type for struct ng_pipe_hookcfg */
180 static const struct ng_parse_struct_field
181 	ng_pipe_hookcfg_type_fields[] = NG_PIPE_HOOKCFG_INFO;
182 static const struct ng_parse_type ng_pipe_hookcfg_type = {
183 	&ng_parse_struct_type,
184 	&ng_pipe_hookcfg_type_fields
185 };
186 
187 /* Parse type for struct ng_pipe_cfg */
188 static const struct ng_parse_struct_field
189 	ng_pipe_cfg_type_fields[] = NG_PIPE_CFG_INFO(&ng_pipe_hookcfg_type);
190 static const struct ng_parse_type ng_pipe_cfg_type = {
191 	&ng_parse_struct_type,
192 	&ng_pipe_cfg_type_fields
193 };
194 
195 /* List of commands and how to convert arguments to/from ASCII */
196 static const struct ng_cmdlist ngp_cmds[] = {
197 	{
198 		.cookie =	NGM_PIPE_COOKIE,
199 		.cmd =		NGM_PIPE_GET_STATS,
200 		.name = 	"getstats",
201 		.respType =	 &ng_pipe_stats_type
202 	},
203 	{
204 		.cookie =	NGM_PIPE_COOKIE,
205 		.cmd =		NGM_PIPE_CLR_STATS,
206 		.name =		"clrstats"
207 	},
208 	{
209 		.cookie =	NGM_PIPE_COOKIE,
210 		.cmd =		NGM_PIPE_GETCLR_STATS,
211 		.name =		"getclrstats",
212 		.respType =	&ng_pipe_stats_type
213 	},
214 	{
215 		.cookie =	NGM_PIPE_COOKIE,
216 		.cmd =		NGM_PIPE_GET_RUN,
217 		.name =		"getrun",
218 		.respType =	&ng_pipe_run_type
219 	},
220 	{
221 		.cookie =	NGM_PIPE_COOKIE,
222 		.cmd =		NGM_PIPE_GET_CFG,
223 		.name =		"getcfg",
224 		.respType =	&ng_pipe_cfg_type
225 	},
226 	{
227 		.cookie =	NGM_PIPE_COOKIE,
228 		.cmd =		NGM_PIPE_SET_CFG,
229 		.name =		"setcfg",
230 		.mesgType =	&ng_pipe_cfg_type,
231 	},
232 	{ 0 }
233 };
234 
235 /* Netgraph type descriptor */
236 static struct ng_type ng_pipe_typestruct = {
237 	.version =	NG_ABI_VERSION,
238 	.name =		NG_PIPE_NODE_TYPE,
239 	.mod_event =	ngp_modevent,
240 	.constructor =	ngp_constructor,
241 	.shutdown =	ngp_shutdown,
242 	.rcvmsg =	ngp_rcvmsg,
243 	.newhook =	ngp_newhook,
244 	.rcvdata =	ngp_rcvdata,
245 	.disconnect =	ngp_disconnect,
246 	.cmdlist =	ngp_cmds
247 };
248 NETGRAPH_INIT(pipe, &ng_pipe_typestruct);
249 
250 /* Node constructor */
251 static int
252 ngp_constructor(node_p node)
253 {
254 	priv_p priv;
255 
256 	priv = malloc(sizeof(*priv), M_NG_PIPE, M_ZERO | M_WAITOK);
257 	NG_NODE_SET_PRIVATE(node, priv);
258 
259 	/* Mark node as single-threaded */
260 	NG_NODE_FORCE_WRITER(node);
261 
262 	ng_callout_init(&priv->timer);
263 
264 	return (0);
265 }
266 
267 /* Add a hook */
268 static int
269 ngp_newhook(node_p node, hook_p hook, const char *name)
270 {
271 	const priv_p priv = NG_NODE_PRIVATE(node);
272 	struct hookinfo *hinfo;
273 
274 	if (strcmp(name, NG_PIPE_HOOK_UPPER) == 0) {
275 		bzero(&priv->upper, sizeof(priv->upper));
276 		priv->upper.hook = hook;
277 		NG_HOOK_SET_PRIVATE(hook, &priv->upper);
278 	} else if (strcmp(name, NG_PIPE_HOOK_LOWER) == 0) {
279 		bzero(&priv->lower, sizeof(priv->lower));
280 		priv->lower.hook = hook;
281 		NG_HOOK_SET_PRIVATE(hook, &priv->lower);
282 	} else
283 		return (EINVAL);
284 
285 	/* Load non-zero initial cfg values */
286 	hinfo = NG_HOOK_PRIVATE(hook);
287 	hinfo->cfg.qin_size_limit = 50;
288 	hinfo->cfg.fifo = 1;
289 	hinfo->cfg.droptail = 1;
290 	TAILQ_INIT(&hinfo->fifo_head);
291 	TAILQ_INIT(&hinfo->qout_head);
292 	return (0);
293 }
294 
295 /* Receive a control message */
296 static int
297 ngp_rcvmsg(node_p node, item_p item, hook_p lasthook)
298 {
299 	const priv_p priv = NG_NODE_PRIVATE(node);
300 	struct ng_mesg *resp = NULL;
301 	struct ng_mesg *msg, *flow_msg;
302 	struct ng_pipe_stats *stats;
303 	struct ng_pipe_run *run;
304 	struct ng_pipe_cfg *cfg;
305 	int error = 0;
306 	int prev_down, now_down, cmd;
307 
308 	NGI_GET_MSG(item, msg);
309 	switch (msg->header.typecookie) {
310 	case NGM_PIPE_COOKIE:
311 		switch (msg->header.cmd) {
312 		case NGM_PIPE_GET_STATS:
313 		case NGM_PIPE_CLR_STATS:
314 		case NGM_PIPE_GETCLR_STATS:
315 			if (msg->header.cmd != NGM_PIPE_CLR_STATS) {
316 				NG_MKRESPONSE(resp, msg,
317 				    sizeof(*stats), M_NOWAIT);
318 				if (resp == NULL) {
319 					error = ENOMEM;
320 					break;
321 				}
322 				stats = (struct ng_pipe_stats *) resp->data;
323 				bcopy(&priv->upper.stats, &stats->downstream,
324 				    sizeof(stats->downstream));
325 				bcopy(&priv->lower.stats, &stats->upstream,
326 				    sizeof(stats->upstream));
327 			}
328 			if (msg->header.cmd != NGM_PIPE_GET_STATS) {
329 				bzero(&priv->upper.stats,
330 				    sizeof(priv->upper.stats));
331 				bzero(&priv->lower.stats,
332 				    sizeof(priv->lower.stats));
333 			}
334 			break;
335 		case NGM_PIPE_GET_RUN:
336 			NG_MKRESPONSE(resp, msg, sizeof(*run), M_NOWAIT);
337 			if (resp == NULL) {
338 				error = ENOMEM;
339 				break;
340 			}
341 			run = (struct ng_pipe_run *) resp->data;
342 			bcopy(&priv->upper.run, &run->downstream,
343 				sizeof(run->downstream));
344 			bcopy(&priv->lower.run, &run->upstream,
345 				sizeof(run->upstream));
346 			break;
347 		case NGM_PIPE_GET_CFG:
348 			NG_MKRESPONSE(resp, msg, sizeof(*cfg), M_NOWAIT);
349 			if (resp == NULL) {
350 				error = ENOMEM;
351 				break;
352 			}
353 			cfg = (struct ng_pipe_cfg *) resp->data;
354 			bcopy(&priv->upper.cfg, &cfg->downstream,
355 				sizeof(cfg->downstream));
356 			bcopy(&priv->lower.cfg, &cfg->upstream,
357 				sizeof(cfg->upstream));
358 			cfg->delay = priv->delay;
359 			cfg->overhead = priv->overhead;
360 			cfg->header_offset = priv->header_offset;
361 			if (cfg->upstream.bandwidth ==
362 			    cfg->downstream.bandwidth) {
363 				cfg->bandwidth = cfg->upstream.bandwidth;
364 				cfg->upstream.bandwidth = 0;
365 				cfg->downstream.bandwidth = 0;
366 			} else
367 				cfg->bandwidth = 0;
368 			break;
369 		case NGM_PIPE_SET_CFG:
370 			cfg = (struct ng_pipe_cfg *) msg->data;
371 			if (msg->header.arglen != sizeof(*cfg)) {
372 				error = EINVAL;
373 				break;
374 			}
375 
376 			if (cfg->delay == -1)
377 				priv->delay = 0;
378 			else if (cfg->delay > 0 && cfg->delay < 10000000)
379 				priv->delay = cfg->delay;
380 
381 			if (cfg->bandwidth == -1) {
382 				priv->upper.cfg.bandwidth = 0;
383 				priv->lower.cfg.bandwidth = 0;
384 				priv->overhead = 0;
385 			} else if (cfg->bandwidth >= 100 &&
386 			    cfg->bandwidth <= 1000000000) {
387 				priv->upper.cfg.bandwidth = cfg->bandwidth;
388 				priv->lower.cfg.bandwidth = cfg->bandwidth;
389 				if (cfg->bandwidth >= 10000000)
390 					priv->overhead = 8+4+12; /* Ethernet */
391 				else
392 					priv->overhead = 10; /* HDLC */
393 			}
394 
395 			if (cfg->overhead == -1)
396 				priv->overhead = 0;
397 			else if (cfg->overhead > 0 &&
398 			    cfg->overhead < MAX_OHSIZE)
399 				priv->overhead = cfg->overhead;
400 
401 			if (cfg->header_offset == -1)
402 				priv->header_offset = 0;
403 			else if (cfg->header_offset > 0 &&
404 			    cfg->header_offset < 64)
405 				priv->header_offset = cfg->header_offset;
406 
407 			prev_down = priv->upper.cfg.ber == 1 ||
408 			    priv->lower.cfg.ber == 1;
409 			parse_cfg(&priv->upper.cfg, &cfg->downstream,
410 			    &priv->upper, priv);
411 			parse_cfg(&priv->lower.cfg, &cfg->upstream,
412 			    &priv->lower, priv);
413 			now_down = priv->upper.cfg.ber == 1 ||
414 			    priv->lower.cfg.ber == 1;
415 
416 			if (prev_down != now_down) {
417 				if (now_down)
418 					cmd = NGM_LINK_IS_DOWN;
419 				else
420 					cmd = NGM_LINK_IS_UP;
421 
422 				if (priv->lower.hook != NULL) {
423 					NG_MKMESSAGE(flow_msg, NGM_FLOW_COOKIE,
424 					    cmd, 0, M_NOWAIT);
425 					if (flow_msg != NULL)
426 						NG_SEND_MSG_HOOK(error, node,
427 						    flow_msg, priv->lower.hook,
428 						    0);
429 				}
430 				if (priv->upper.hook != NULL) {
431 					NG_MKMESSAGE(flow_msg, NGM_FLOW_COOKIE,
432 					    cmd, 0, M_NOWAIT);
433 					if (flow_msg != NULL)
434 						NG_SEND_MSG_HOOK(error, node,
435 						    flow_msg, priv->upper.hook,
436 						    0);
437 				}
438 			}
439 			break;
440 		default:
441 			error = EINVAL;
442 			break;
443 		}
444 		break;
445 	default:
446 		error = EINVAL;
447 		break;
448 	}
449 	NG_RESPOND_MSG(error, node, item, resp);
450 	NG_FREE_MSG(msg);
451 
452 	return (error);
453 }
454 
455 static void
456 parse_cfg(struct ng_pipe_hookcfg *current, struct ng_pipe_hookcfg *new,
457 	struct hookinfo *hinfo, priv_p priv)
458 {
459 
460 	if (new->ber == -1) {
461 		current->ber = 0;
462 		if (hinfo->ber_p) {
463 			free(hinfo->ber_p, M_NG_PIPE);
464 			hinfo->ber_p = NULL;
465 		}
466 	} else if (new->ber >= 1 && new->ber <= 1000000000000) {
467 		static const uint64_t one = 0x1000000000000; /* = 2^48 */
468 		uint64_t p0, p;
469 		uint32_t fsize, i;
470 
471 		if (hinfo->ber_p == NULL)
472 			hinfo->ber_p =
473 			    malloc((MAX_FSIZE + MAX_OHSIZE) * sizeof(uint64_t),
474 			    M_NG_PIPE, M_WAITOK);
475 		current->ber = new->ber;
476 
477 		/*
478 		 * For given BER and each frame size N (in bytes) calculate
479 		 * the probability P_OK that the frame is clean:
480 		 *
481 		 * P_OK(BER,N) = (1 - 1/BER)^(N*8)
482 		 *
483 		 * We use a 64-bit fixed-point format with decimal point
484 		 * positioned between bits 47 and 48.
485 		 */
486 		p0 = one - one / new->ber;
487 		p = one;
488 		for (fsize = 0; fsize < MAX_FSIZE + MAX_OHSIZE; fsize++) {
489 			hinfo->ber_p[fsize] = p;
490 			for (i = 0; i < 8; i++)
491 				p = (p * (p0 & 0xffff) >> 48) +
492 				    (p * ((p0 >> 16) & 0xffff) >> 32) +
493 				    (p * (p0 >> 32) >> 16);
494 		}
495 	}
496 
497 	if (new->qin_size_limit == -1)
498 		current->qin_size_limit = 0;
499 	else if (new->qin_size_limit >= 5)
500 		current->qin_size_limit = new->qin_size_limit;
501 
502 	if (new->qout_size_limit == -1)
503 		current->qout_size_limit = 0;
504 	else if (new->qout_size_limit >= 5)
505 		current->qout_size_limit = new->qout_size_limit;
506 
507 	if (new->duplicate == -1)
508 		current->duplicate = 0;
509 	else if (new->duplicate > 0 && new->duplicate <= 50)
510 		current->duplicate = new->duplicate;
511 
512 	if (new->fifo) {
513 		current->fifo = 1;
514 		current->wfq = 0;
515 		current->drr = 0;
516 	}
517 
518 	if (new->wfq) {
519 		current->fifo = 0;
520 		current->wfq = 1;
521 		current->drr = 0;
522 	}
523 
524 	if (new->drr) {
525 		current->fifo = 0;
526 		current->wfq = 0;
527 		/* DRR quantum */
528 		if (new->drr >= 32)
529 			current->drr = new->drr;
530 		else
531 			current->drr = 2048;		/* default quantum */
532 	}
533 
534 	if (new->droptail) {
535 		current->droptail = 1;
536 		current->drophead = 0;
537 	}
538 
539 	if (new->drophead) {
540 		current->droptail = 0;
541 		current->drophead = 1;
542 	}
543 
544 	if (new->bandwidth == -1) {
545 		current->bandwidth = 0;
546 		current->fifo = 1;
547 		current->wfq = 0;
548 		current->drr = 0;
549 	} else if (new->bandwidth >= 100 && new->bandwidth <= 1000000000)
550 		current->bandwidth = new->bandwidth;
551 
552 	if (current->bandwidth | priv->delay |
553 	    current->duplicate | current->ber)
554 		hinfo->noqueue = 0;
555 	else
556 		hinfo->noqueue = 1;
557 }
558 
559 /*
560  * Compute a hash signature for a packet. This function suffers from the
561  * NIH sindrome, so probably it would be wise to look around what other
562  * folks have found out to be a good and efficient IP hash function...
563  */
564 static int
565 ip_hash(struct mbuf *m, int offset)
566 {
567 	u_int64_t i;
568 	struct ip *ip = (struct ip *)(mtod(m, u_char *) + offset);
569 
570 	if (m->m_len < sizeof(struct ip) + offset ||
571 	    ip->ip_v != 4 || ip->ip_hl << 2 != sizeof(struct ip))
572 		return 0;
573 
574 	i = ((u_int64_t) ip->ip_src.s_addr ^
575 	    ((u_int64_t) ip->ip_src.s_addr << 13) ^
576 	    ((u_int64_t) ip->ip_dst.s_addr << 7) ^
577 	    ((u_int64_t) ip->ip_dst.s_addr << 19));
578 	return (i ^ (i >> 32));
579 }
580 
581 /*
582  * Receive data on a hook - both in upstream and downstream direction.
583  * We put the frame on the inbound queue, and try to initiate dequeuing
584  * sequence immediately. If inbound queue is full, discard one frame
585  * depending on dropping policy (from the head or from the tail of the
586  * queue).
587  */
588 static int
589 ngp_rcvdata(hook_p hook, item_p item)
590 {
591 	struct hookinfo *const hinfo = NG_HOOK_PRIVATE(hook);
592 	const priv_p priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook));
593 	struct timeval uuptime;
594 	struct timeval *now = &uuptime;
595 	struct ngp_fifo *ngp_f = NULL, *ngp_f1;
596 	struct ngp_hdr *ngp_h = NULL;
597 	struct mbuf *m;
598 	int hash, plen;
599 	int error = 0;
600 
601 	/*
602 	 * Shortcut from inbound to outbound hook when neither of
603 	 * bandwidth, delay, BER or duplication probability is
604 	 * configured, nor we have queued frames to drain.
605 	 */
606 	if (hinfo->run.qin_frames == 0 && hinfo->run.qout_frames == 0 &&
607 	    hinfo->noqueue) {
608 		struct hookinfo *dest;
609 		if (hinfo == &priv->lower)
610 			dest = &priv->upper;
611 		else
612 			dest = &priv->lower;
613 
614 		/* Send the frame. */
615 		plen = NGI_M(item)->m_pkthdr.len;
616 		NG_FWD_ITEM_HOOK(error, item, dest->hook);
617 
618 		/* Update stats. */
619 		if (error) {
620 			hinfo->stats.out_disc_frames++;
621 			hinfo->stats.out_disc_octets += plen;
622 		} else {
623 			hinfo->stats.fwd_frames++;
624 			hinfo->stats.fwd_octets += plen;
625 		}
626 
627 		return (error);
628 	}
629 
630 	microuptime(now);
631 
632 	/*
633 	 * If this was an empty queue, update service deadline time.
634 	 */
635 	if (hinfo->run.qin_frames == 0) {
636 		struct timeval *when = &hinfo->qin_utime;
637 		if (when->tv_sec < now->tv_sec || (when->tv_sec == now->tv_sec
638 		    && when->tv_usec < now->tv_usec)) {
639 			when->tv_sec = now->tv_sec;
640 			when->tv_usec = now->tv_usec;
641 		}
642 	}
643 
644 	/* Populate the packet header */
645 	ngp_h = uma_zalloc(ngp_zone, M_NOWAIT);
646 	KASSERT((ngp_h != NULL), ("ngp_h zalloc failed (1)"));
647 	NGI_GET_M(item, m);
648 	KASSERT(m != NULL, ("NGI_GET_M failed"));
649 	ngp_h->m = m;
650 	NG_FREE_ITEM(item);
651 
652 	if (hinfo->cfg.fifo)
653 		hash = 0;	/* all packets go into a single FIFO queue */
654 	else
655 		hash = ip_hash(m, priv->header_offset);
656 
657 	/* Find the appropriate FIFO queue for the packet and enqueue it*/
658 	TAILQ_FOREACH(ngp_f, &hinfo->fifo_head, fifo_le)
659 		if (hash == ngp_f->hash)
660 			break;
661 	if (ngp_f == NULL) {
662 		ngp_f = uma_zalloc(ngp_zone, M_NOWAIT);
663 		KASSERT(ngp_h != NULL, ("ngp_h zalloc failed (2)"));
664 		TAILQ_INIT(&ngp_f->packet_head);
665 		ngp_f->hash = hash;
666 		ngp_f->packets = 1;
667 		ngp_f->rr_deficit = hinfo->cfg.drr;	/* DRR quantum */
668 		hinfo->run.fifo_queues++;
669 		TAILQ_INSERT_TAIL(&ngp_f->packet_head, ngp_h, ngp_link);
670 		FIFO_VTIME_SORT(m->m_pkthdr.len);
671 	} else {
672 		TAILQ_INSERT_TAIL(&ngp_f->packet_head, ngp_h, ngp_link);
673 		ngp_f->packets++;
674 	}
675 	hinfo->run.qin_frames++;
676 	hinfo->run.qin_octets += m->m_pkthdr.len;
677 
678 	/* Discard a frame if inbound queue limit has been reached */
679 	if (hinfo->run.qin_frames > hinfo->cfg.qin_size_limit) {
680 		struct mbuf *m1;
681 		int longest = 0;
682 
683 		/* Find the longest queue */
684 		TAILQ_FOREACH(ngp_f1, &hinfo->fifo_head, fifo_le)
685 			if (ngp_f1->packets > longest) {
686 				longest = ngp_f1->packets;
687 				ngp_f = ngp_f1;
688 			}
689 
690 		/* Drop a frame from the queue head/tail, depending on cfg */
691 		if (hinfo->cfg.drophead)
692 			ngp_h = TAILQ_FIRST(&ngp_f->packet_head);
693 		else
694 			ngp_h = TAILQ_LAST(&ngp_f->packet_head, p_head);
695 		TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link);
696 		m1 = ngp_h->m;
697 		uma_zfree(ngp_zone, ngp_h);
698 		hinfo->run.qin_octets -= m1->m_pkthdr.len;
699 		hinfo->stats.in_disc_octets += m1->m_pkthdr.len;
700 		m_freem(m1);
701 		if (--(ngp_f->packets) == 0) {
702 			TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
703 			uma_zfree(ngp_zone, ngp_f);
704 			hinfo->run.fifo_queues--;
705 		}
706 		hinfo->run.qin_frames--;
707 		hinfo->stats.in_disc_frames++;
708 	}
709 
710 	/*
711 	 * Try to start the dequeuing process immediately.
712 	 */
713 	pipe_dequeue(hinfo, now);
714 
715 	return (0);
716 }
717 
718 /*
719  * Dequeueing sequence - we basically do the following:
720  *  1) Try to extract the frame from the inbound (bandwidth) queue;
721  *  2) In accordance to BER specified, discard the frame randomly;
722  *  3) If the frame survives BER, prepend it with delay info and move it
723  *     to outbound (delay) queue;
724  *  4) Loop to 2) until bandwidth quota for this timeslice is reached, or
725  *     inbound queue is flushed completely;
726  *  5) Dequeue frames from the outbound queue and send them downstream until
727  *     outbound queue is flushed completely, or the next frame in the queue
728  *     is not due to be dequeued yet
729  */
730 static void
731 pipe_dequeue(struct hookinfo *hinfo, struct timeval *now) {
732 	static uint64_t rand, oldrand;
733 	const node_p node = NG_HOOK_NODE(hinfo->hook);
734 	const priv_p priv = NG_NODE_PRIVATE(node);
735 	struct hookinfo *dest;
736 	struct ngp_fifo *ngp_f, *ngp_f1;
737 	struct ngp_hdr *ngp_h;
738 	struct timeval *when;
739 	struct mbuf *m;
740 	int plen, error = 0;
741 
742 	/* Which one is the destination hook? */
743 	if (hinfo == &priv->lower)
744 		dest = &priv->upper;
745 	else
746 		dest = &priv->lower;
747 
748 	/* Bandwidth queue processing */
749 	while ((ngp_f = TAILQ_FIRST(&hinfo->fifo_head))) {
750 		when = &hinfo->qin_utime;
751 		if (when->tv_sec > now->tv_sec || (when->tv_sec == now->tv_sec
752 		    && when->tv_usec > now->tv_usec))
753 			break;
754 
755 		ngp_h = TAILQ_FIRST(&ngp_f->packet_head);
756 		m = ngp_h->m;
757 
758 		/* Deficit Round Robin (DRR) processing */
759 		if (hinfo->cfg.drr) {
760 			if (ngp_f->rr_deficit >= m->m_pkthdr.len) {
761 				ngp_f->rr_deficit -= m->m_pkthdr.len;
762 			} else {
763 				ngp_f->rr_deficit += hinfo->cfg.drr;
764 				TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
765 				TAILQ_INSERT_TAIL(&hinfo->fifo_head,
766 				    ngp_f, fifo_le);
767 				continue;
768 			}
769 		}
770 
771 		/*
772 		 * Either create a duplicate and pass it on, or dequeue
773 		 * the original packet...
774 		 */
775 		if (hinfo->cfg.duplicate &&
776 		    random() % 100 <= hinfo->cfg.duplicate) {
777 			ngp_h = uma_zalloc(ngp_zone, M_NOWAIT);
778 			KASSERT(ngp_h != NULL, ("ngp_h zalloc failed (3)"));
779 			m = m_dup(m, M_NOWAIT);
780 			KASSERT(m != NULL, ("m_dup failed"));
781 			ngp_h->m = m;
782 		} else {
783 			TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link);
784 			hinfo->run.qin_frames--;
785 			hinfo->run.qin_octets -= m->m_pkthdr.len;
786 			ngp_f->packets--;
787 		}
788 
789 		/* Calculate the serialization delay */
790 		if (hinfo->cfg.bandwidth) {
791 			hinfo->qin_utime.tv_usec +=
792 			    ((uint64_t) m->m_pkthdr.len + priv->overhead ) *
793 			    8000000 / hinfo->cfg.bandwidth;
794 			hinfo->qin_utime.tv_sec +=
795 			    hinfo->qin_utime.tv_usec / 1000000;
796 			hinfo->qin_utime.tv_usec =
797 			    hinfo->qin_utime.tv_usec % 1000000;
798 		}
799 		when = &ngp_h->when;
800 		when->tv_sec = hinfo->qin_utime.tv_sec;
801 		when->tv_usec = hinfo->qin_utime.tv_usec;
802 
803 		/* Sort / rearrange inbound queues */
804 		if (ngp_f->packets) {
805 			if (hinfo->cfg.wfq) {
806 				TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
807 				FIFO_VTIME_SORT(TAILQ_FIRST(
808 				    &ngp_f->packet_head)->m->m_pkthdr.len)
809 			}
810 		} else {
811 			TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
812 			uma_zfree(ngp_zone, ngp_f);
813 			hinfo->run.fifo_queues--;
814 		}
815 
816 		/* Randomly discard the frame, according to BER setting */
817 		if (hinfo->cfg.ber) {
818 			oldrand = rand;
819 			rand = random();
820 			if (((oldrand ^ rand) << 17) >=
821 			    hinfo->ber_p[priv->overhead + m->m_pkthdr.len]) {
822 				hinfo->stats.out_disc_frames++;
823 				hinfo->stats.out_disc_octets += m->m_pkthdr.len;
824 				uma_zfree(ngp_zone, ngp_h);
825 				m_freem(m);
826 				continue;
827 			}
828 		}
829 
830 		/* Discard frame if outbound queue size limit exceeded */
831 		if (hinfo->cfg.qout_size_limit &&
832 		    hinfo->run.qout_frames>=hinfo->cfg.qout_size_limit) {
833 			hinfo->stats.out_disc_frames++;
834 			hinfo->stats.out_disc_octets += m->m_pkthdr.len;
835 			uma_zfree(ngp_zone, ngp_h);
836 			m_freem(m);
837 			continue;
838 		}
839 
840 		/* Calculate the propagation delay */
841 		when->tv_usec += priv->delay;
842 		when->tv_sec += when->tv_usec / 1000000;
843 		when->tv_usec = when->tv_usec % 1000000;
844 
845 		/* Put the frame into the delay queue */
846 		TAILQ_INSERT_TAIL(&hinfo->qout_head, ngp_h, ngp_link);
847 		hinfo->run.qout_frames++;
848 		hinfo->run.qout_octets += m->m_pkthdr.len;
849 	}
850 
851 	/* Delay queue processing */
852 	while ((ngp_h = TAILQ_FIRST(&hinfo->qout_head))) {
853 		when = &ngp_h->when;
854 		m = ngp_h->m;
855 		if (when->tv_sec > now->tv_sec ||
856 		    (when->tv_sec == now->tv_sec &&
857 		    when->tv_usec > now->tv_usec))
858 			break;
859 
860 		/* Update outbound queue stats */
861 		plen = m->m_pkthdr.len;
862 		hinfo->run.qout_frames--;
863 		hinfo->run.qout_octets -= plen;
864 
865 		/* Dequeue the packet from qout */
866 		TAILQ_REMOVE(&hinfo->qout_head, ngp_h, ngp_link);
867 		uma_zfree(ngp_zone, ngp_h);
868 
869 		NG_SEND_DATA(error, dest->hook, m, meta);
870 		if (error) {
871 			hinfo->stats.out_disc_frames++;
872 			hinfo->stats.out_disc_octets += plen;
873 		} else {
874 			hinfo->stats.fwd_frames++;
875 			hinfo->stats.fwd_octets += plen;
876 		}
877 	}
878 
879 	if ((hinfo->run.qin_frames != 0 || hinfo->run.qout_frames != 0) &&
880 	    !priv->timer_scheduled) {
881 		ng_callout(&priv->timer, node, NULL, 1, ngp_callout, NULL, 0);
882 		priv->timer_scheduled = 1;
883 	}
884 }
885 
886 /*
887  * This routine is called on every clock tick.  We poll connected hooks
888  * for queued frames by calling pipe_dequeue().
889  */
890 static void
891 ngp_callout(node_p node, hook_p hook, void *arg1, int arg2)
892 {
893 	const priv_p priv = NG_NODE_PRIVATE(node);
894 	struct timeval now;
895 
896 	priv->timer_scheduled = 0;
897 	microuptime(&now);
898 	if (priv->upper.hook != NULL)
899 		pipe_dequeue(&priv->upper, &now);
900 	if (priv->lower.hook != NULL)
901 		pipe_dequeue(&priv->lower, &now);
902 }
903 
904 /*
905  * Shutdown processing
906  *
907  * This is tricky. If we have both a lower and upper hook, then we
908  * probably want to extricate ourselves and leave the two peers
909  * still linked to each other. Otherwise we should just shut down as
910  * a normal node would.
911  */
912 static int
913 ngp_shutdown(node_p node)
914 {
915 	const priv_p priv = NG_NODE_PRIVATE(node);
916 
917 	if (priv->timer_scheduled)
918 		ng_uncallout(&priv->timer, node);
919 	if (priv->lower.hook && priv->upper.hook)
920 		ng_bypass(priv->lower.hook, priv->upper.hook);
921 	else {
922 		if (priv->upper.hook != NULL)
923 			ng_rmhook_self(priv->upper.hook);
924 		if (priv->lower.hook != NULL)
925 			ng_rmhook_self(priv->lower.hook);
926 	}
927 	NG_NODE_UNREF(node);
928 	free(priv, M_NG_PIPE);
929 	return (0);
930 }
931 
932 /*
933  * Hook disconnection
934  */
935 static int
936 ngp_disconnect(hook_p hook)
937 {
938 	struct hookinfo *const hinfo = NG_HOOK_PRIVATE(hook);
939 	struct ngp_fifo *ngp_f;
940 	struct ngp_hdr *ngp_h;
941 
942 	KASSERT(hinfo != NULL, ("%s: null info", __FUNCTION__));
943 	hinfo->hook = NULL;
944 
945 	/* Flush all fifo queues associated with the hook */
946 	while ((ngp_f = TAILQ_FIRST(&hinfo->fifo_head))) {
947 		while ((ngp_h = TAILQ_FIRST(&ngp_f->packet_head))) {
948 			TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link);
949 			m_freem(ngp_h->m);
950 			uma_zfree(ngp_zone, ngp_h);
951 		}
952 		TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
953 		uma_zfree(ngp_zone, ngp_f);
954 	}
955 
956 	/* Flush the delay queue */
957 	while ((ngp_h = TAILQ_FIRST(&hinfo->qout_head))) {
958 		TAILQ_REMOVE(&hinfo->qout_head, ngp_h, ngp_link);
959 		m_freem(ngp_h->m);
960 		uma_zfree(ngp_zone, ngp_h);
961 	}
962 
963 	/* Release the packet loss probability table (BER) */
964 	if (hinfo->ber_p)
965 		free(hinfo->ber_p, M_NG_PIPE);
966 
967 	return (0);
968 }
969 
970 static int
971 ngp_modevent(module_t mod, int type, void *unused)
972 {
973 	int error = 0;
974 
975 	switch (type) {
976 	case MOD_LOAD:
977 		ngp_zone = uma_zcreate("ng_pipe", max(sizeof(struct ngp_hdr),
978 		    sizeof (struct ngp_fifo)), NULL, NULL, NULL, NULL,
979 		    UMA_ALIGN_PTR, 0);
980 		if (ngp_zone == NULL)
981 			panic("ng_pipe: couldn't allocate descriptor zone");
982 		break;
983 	case MOD_UNLOAD:
984 		uma_zdestroy(ngp_zone);
985 		break;
986 	default:
987 		error = EOPNOTSUPP;
988 		break;
989 	}
990 
991 	return (error);
992 }
993