xref: /freebsd/sys/netgraph/ng_pipe.c (revision 01ded8b942effbbb4d9225c4227f264e499e9698)
1 /*-
2  * Copyright (c) 2004-2010 University of Zagreb
3  * Copyright (c) 2007-2008 FreeBSD Foundation
4  *
5  * This software was developed by the University of Zagreb and the
6  * FreeBSD Foundation under sponsorship by the Stichting NLnet and the
7  * FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  * $FreeBSD$
31  */
32 
33 /*
34  * This node permits simple traffic shaping by emulating bandwidth
35  * and delay, as well as random packet losses.
36  * The node has two hooks, upper and lower. Traffic flowing from upper to
37  * lower hook is referenced as downstream, and vice versa. Parameters for
38  * both directions can be set separately, except for delay.
39  */
40 
41 
42 #include <sys/param.h>
43 #include <sys/errno.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/time.h>
49 
50 #include <vm/uma.h>
51 
52 #include <net/vnet.h>
53 
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/ip.h>
57 
58 #include <netgraph/ng_message.h>
59 #include <netgraph/netgraph.h>
60 #include <netgraph/ng_parse.h>
61 #include <netgraph/ng_pipe.h>
62 
63 static MALLOC_DEFINE(M_NG_PIPE, "ng_pipe", "ng_pipe");
64 
65 /* Packet header struct */
66 struct ngp_hdr {
67 	TAILQ_ENTRY(ngp_hdr)	ngp_link;	/* next pkt in queue */
68 	struct timeval		when;		/* this packet's due time */
69 	struct mbuf		*m;		/* ptr to the packet data */
70 };
71 TAILQ_HEAD(p_head, ngp_hdr);
72 
73 /* FIFO queue struct */
74 struct ngp_fifo {
75 	TAILQ_ENTRY(ngp_fifo)	fifo_le;	/* list of active queues only */
76 	struct p_head		packet_head;	/* FIFO queue head */
77 	u_int32_t		hash;		/* flow signature */
78 	struct timeval		vtime;		/* virtual time, for WFQ */
79 	u_int32_t		rr_deficit;	/* for DRR */
80 	u_int32_t		packets;	/* # of packets in this queue */
81 };
82 
83 /* Per hook info */
84 struct hookinfo {
85 	hook_p			hook;
86 	int			noqueue;	/* bypass any processing */
87 	TAILQ_HEAD(, ngp_fifo)	fifo_head;	/* FIFO queues */
88 	TAILQ_HEAD(, ngp_hdr)	qout_head;	/* delay queue head */
89 	struct timeval		qin_utime;
90 	struct ng_pipe_hookcfg	cfg;
91 	struct ng_pipe_hookrun	run;
92 	struct ng_pipe_hookstat	stats;
93 	uint64_t		*ber_p;		/* loss_p(BER,psize) map */
94 };
95 
96 /* Per node info */
97 struct node_priv {
98 	u_int64_t		delay;
99 	u_int32_t		overhead;
100 	u_int32_t		header_offset;
101 	struct hookinfo		lower;
102 	struct hookinfo		upper;
103 	struct callout		timer;
104 	int			timer_scheduled;
105 };
106 typedef struct node_priv *priv_p;
107 
108 /* Macro for calculating the virtual time for packet dequeueing in WFQ */
109 #define FIFO_VTIME_SORT(plen)						\
110 	if (hinfo->cfg.wfq && hinfo->cfg.bandwidth) {			\
111 		ngp_f->vtime.tv_usec = now->tv_usec + ((uint64_t) (plen) \
112 			+ priv->overhead ) * hinfo->run.fifo_queues *	\
113 			8000000 / hinfo->cfg.bandwidth;			\
114 		ngp_f->vtime.tv_sec = now->tv_sec +			\
115 			ngp_f->vtime.tv_usec / 1000000;			\
116 		ngp_f->vtime.tv_usec = ngp_f->vtime.tv_usec % 1000000;	\
117 		TAILQ_FOREACH(ngp_f1, &hinfo->fifo_head, fifo_le)	\
118 			if (ngp_f1->vtime.tv_sec > ngp_f->vtime.tv_sec || \
119 			    (ngp_f1->vtime.tv_sec == ngp_f->vtime.tv_sec && \
120 			    ngp_f1->vtime.tv_usec > ngp_f->vtime.tv_usec)) \
121 				break;					\
122 		if (ngp_f1 == NULL)					\
123 			TAILQ_INSERT_TAIL(&hinfo->fifo_head, ngp_f, fifo_le); \
124 		else							\
125 			TAILQ_INSERT_BEFORE(ngp_f1, ngp_f, fifo_le);	\
126 	} else								\
127 		TAILQ_INSERT_TAIL(&hinfo->fifo_head, ngp_f, fifo_le);	\
128 
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_NOWAIT);
257 	if (priv == NULL)
258 		return (ENOMEM);
259 	NG_NODE_SET_PRIVATE(node, priv);
260 
261 	/* Mark node as single-threaded */
262 	NG_NODE_FORCE_WRITER(node);
263 
264 	ng_callout_init(&priv->timer);
265 
266 	return (0);
267 }
268 
269 /* Add a hook */
270 static int
271 ngp_newhook(node_p node, hook_p hook, const char *name)
272 {
273 	const priv_p priv = NG_NODE_PRIVATE(node);
274 	struct hookinfo *hinfo;
275 
276 	if (strcmp(name, NG_PIPE_HOOK_UPPER) == 0) {
277 		bzero(&priv->upper, sizeof(priv->upper));
278 		priv->upper.hook = hook;
279 		NG_HOOK_SET_PRIVATE(hook, &priv->upper);
280 	} else if (strcmp(name, NG_PIPE_HOOK_LOWER) == 0) {
281 		bzero(&priv->lower, sizeof(priv->lower));
282 		priv->lower.hook = hook;
283 		NG_HOOK_SET_PRIVATE(hook, &priv->lower);
284 	} else
285 		return (EINVAL);
286 
287 	/* Load non-zero initial cfg values */
288 	hinfo = NG_HOOK_PRIVATE(hook);
289 	hinfo->cfg.qin_size_limit = 50;
290 	hinfo->cfg.fifo = 1;
291 	hinfo->cfg.droptail = 1;
292 	TAILQ_INIT(&hinfo->fifo_head);
293 	TAILQ_INIT(&hinfo->qout_head);
294 	return (0);
295 }
296 
297 /* Receive a control message */
298 static int
299 ngp_rcvmsg(node_p node, item_p item, hook_p lasthook)
300 {
301 	const priv_p priv = NG_NODE_PRIVATE(node);
302 	struct ng_mesg *resp = NULL;
303 	struct ng_mesg *msg;
304 	struct ng_pipe_stats *stats;
305 	struct ng_pipe_run *run;
306 	struct ng_pipe_cfg *cfg;
307 	int error = 0;
308 
309 	NGI_GET_MSG(item, msg);
310 	switch (msg->header.typecookie) {
311 	case NGM_PIPE_COOKIE:
312 		switch (msg->header.cmd) {
313 		case NGM_PIPE_GET_STATS:
314 		case NGM_PIPE_CLR_STATS:
315 		case NGM_PIPE_GETCLR_STATS:
316 			if (msg->header.cmd != NGM_PIPE_CLR_STATS) {
317 				NG_MKRESPONSE(resp, msg,
318 				    sizeof(*stats), M_NOWAIT);
319 				if (resp == NULL) {
320 					error = ENOMEM;
321 					break;
322 				}
323 				stats = (struct ng_pipe_stats *) resp->data;
324 				bcopy(&priv->upper.stats, &stats->downstream,
325 				    sizeof(stats->downstream));
326 				bcopy(&priv->lower.stats, &stats->upstream,
327 				    sizeof(stats->upstream));
328 			}
329 			if (msg->header.cmd != NGM_PIPE_GET_STATS) {
330 				bzero(&priv->upper.stats,
331 				    sizeof(priv->upper.stats));
332 				bzero(&priv->lower.stats,
333 				    sizeof(priv->lower.stats));
334 			}
335 			break;
336 		case NGM_PIPE_GET_RUN:
337 			NG_MKRESPONSE(resp, msg, sizeof(*run), M_NOWAIT);
338 			if (resp == NULL) {
339 				error = ENOMEM;
340 				break;
341 			}
342 			run = (struct ng_pipe_run *) resp->data;
343 			bcopy(&priv->upper.run, &run->downstream,
344 				sizeof(run->downstream));
345 			bcopy(&priv->lower.run, &run->upstream,
346 				sizeof(run->upstream));
347 			break;
348 		case NGM_PIPE_GET_CFG:
349 			NG_MKRESPONSE(resp, msg, sizeof(*cfg), M_NOWAIT);
350 			if (resp == NULL) {
351 				error = ENOMEM;
352 				break;
353 			}
354 			cfg = (struct ng_pipe_cfg *) resp->data;
355 			bcopy(&priv->upper.cfg, &cfg->downstream,
356 				sizeof(cfg->downstream));
357 			bcopy(&priv->lower.cfg, &cfg->upstream,
358 				sizeof(cfg->upstream));
359 			cfg->delay = priv->delay;
360 			cfg->overhead = priv->overhead;
361 			cfg->header_offset = priv->header_offset;
362 			if (cfg->upstream.bandwidth ==
363 			    cfg->downstream.bandwidth) {
364 				cfg->bandwidth = cfg->upstream.bandwidth;
365 				cfg->upstream.bandwidth = 0;
366 				cfg->downstream.bandwidth = 0;
367 			} else
368 				cfg->bandwidth = 0;
369 			break;
370 		case NGM_PIPE_SET_CFG:
371 			cfg = (struct ng_pipe_cfg *) msg->data;
372 			if (msg->header.arglen != sizeof(*cfg)) {
373 				error = EINVAL;
374 				break;
375 			}
376 
377 			if (cfg->delay == -1)
378 				priv->delay = 0;
379 			else if (cfg->delay > 0 && cfg->delay < 10000000)
380 				priv->delay = cfg->delay;
381 
382 			if (cfg->bandwidth == -1) {
383 				priv->upper.cfg.bandwidth = 0;
384 				priv->lower.cfg.bandwidth = 0;
385 				priv->overhead = 0;
386 			} else if (cfg->bandwidth >= 100 &&
387 			    cfg->bandwidth <= 1000000000) {
388 				priv->upper.cfg.bandwidth = cfg->bandwidth;
389 				priv->lower.cfg.bandwidth = cfg->bandwidth;
390 				if (cfg->bandwidth >= 10000000)
391 					priv->overhead = 8+4+12; /* Ethernet */
392 				else
393 					priv->overhead = 10; /* HDLC */
394 			}
395 
396 			if (cfg->overhead == -1)
397 				priv->overhead = 0;
398 			else if (cfg->overhead > 0 &&
399 			    cfg->overhead < MAX_OHSIZE)
400 				priv->overhead = cfg->overhead;
401 
402 			if (cfg->header_offset == -1)
403 				priv->header_offset = 0;
404 			else if (cfg->header_offset > 0 &&
405 			    cfg->header_offset < 64)
406 				priv->header_offset = cfg->header_offset;
407 
408 			parse_cfg(&priv->upper.cfg, &cfg->downstream,
409 			    &priv->upper, priv);
410 			parse_cfg(&priv->lower.cfg, &cfg->upstream,
411 			    &priv->lower, priv);
412 			break;
413 		default:
414 			error = EINVAL;
415 			break;
416 		}
417 		break;
418 	default:
419 		error = EINVAL;
420 		break;
421 	}
422 	NG_RESPOND_MSG(error, node, item, resp);
423 	NG_FREE_MSG(msg);
424 
425 	return (error);
426 }
427 
428 static void
429 parse_cfg(struct ng_pipe_hookcfg *current, struct ng_pipe_hookcfg *new,
430 	struct hookinfo *hinfo, priv_p priv)
431 {
432 
433 	if (new->ber == -1) {
434 		current->ber = 0;
435 		if (hinfo->ber_p) {
436 			free(hinfo->ber_p, M_NG_PIPE);
437 			hinfo->ber_p = NULL;
438 		}
439 	} else if (new->ber >= 1 && new->ber <= 1000000000000) {
440 		static const uint64_t one = 0x1000000000000; /* = 2^48 */
441 		uint64_t p0, p;
442 		uint32_t fsize, i;
443 
444 		if (hinfo->ber_p == NULL)
445 			hinfo->ber_p =
446 			    malloc((MAX_FSIZE + MAX_OHSIZE) * sizeof(uint64_t),
447 			    M_NG_PIPE, M_NOWAIT);
448 		current->ber = new->ber;
449 
450 		/*
451 		 * For given BER and each frame size N (in bytes) calculate
452 		 * the probability P_OK that the frame is clean:
453 		 *
454 		 * P_OK(BER,N) = (1 - 1/BER)^(N*8)
455 		 *
456 		 * We use a 64-bit fixed-point format with decimal point
457 		 * positioned between bits 47 and 48.
458 		 */
459 		p0 = one - one / new->ber;
460 		p = one;
461 		for (fsize = 0; fsize < MAX_FSIZE + MAX_OHSIZE; fsize++) {
462 			hinfo->ber_p[fsize] = p;
463 			for (i = 0; i < 8; i++)
464 				p = (p * (p0 & 0xffff) >> 48) +
465 				    (p * ((p0 >> 16) & 0xffff) >> 32) +
466 				    (p * (p0 >> 32) >> 16);
467 		}
468 	}
469 
470 	if (new->qin_size_limit == -1)
471 		current->qin_size_limit = 0;
472 	else if (new->qin_size_limit >= 5)
473 		current->qin_size_limit = new->qin_size_limit;
474 
475 	if (new->qout_size_limit == -1)
476 		current->qout_size_limit = 0;
477 	else if (new->qout_size_limit >= 5)
478 		current->qout_size_limit = new->qout_size_limit;
479 
480 	if (new->duplicate == -1)
481 		current->duplicate = 0;
482 	else if (new->duplicate > 0 && new->duplicate <= 50)
483 		current->duplicate = new->duplicate;
484 
485 	if (new->fifo) {
486 		current->fifo = 1;
487 		current->wfq = 0;
488 		current->drr = 0;
489 	}
490 
491 	if (new->wfq) {
492 		current->fifo = 0;
493 		current->wfq = 1;
494 		current->drr = 0;
495 	}
496 
497 	if (new->drr) {
498 		current->fifo = 0;
499 		current->wfq = 0;
500 		/* DRR quantum */
501 		if (new->drr >= 32)
502 			current->drr = new->drr;
503 		else
504 			current->drr = 2048;		/* default quantum */
505 	}
506 
507 	if (new->droptail) {
508 		current->droptail = 1;
509 		current->drophead = 0;
510 	}
511 
512 	if (new->drophead) {
513 		current->droptail = 0;
514 		current->drophead = 1;
515 	}
516 
517 	if (new->bandwidth == -1) {
518 		current->bandwidth = 0;
519 		current->fifo = 1;
520 		current->wfq = 0;
521 		current->drr = 0;
522 	} else if (new->bandwidth >= 100 && new->bandwidth <= 1000000000)
523 		current->bandwidth = new->bandwidth;
524 
525 	if (current->bandwidth | priv->delay |
526 	    current->duplicate | current->ber)
527 		hinfo->noqueue = 0;
528 	else
529 		hinfo->noqueue = 1;
530 }
531 
532 /*
533  * Compute a hash signature for a packet. This function suffers from the
534  * NIH sindrome, so probably it would be wise to look around what other
535  * folks have found out to be a good and efficient IP hash function...
536  */
537 static int
538 ip_hash(struct mbuf *m, int offset)
539 {
540 	u_int64_t i;
541 	struct ip *ip = (struct ip *)(mtod(m, u_char *) + offset);
542 
543 	if (m->m_len < sizeof(struct ip) + offset ||
544 	    ip->ip_v != 4 || ip->ip_hl << 2 != sizeof(struct ip))
545 		return 0;
546 
547 	i = ((u_int64_t) ip->ip_src.s_addr ^
548 	    ((u_int64_t) ip->ip_src.s_addr << 13) ^
549 	    ((u_int64_t) ip->ip_dst.s_addr << 7) ^
550 	    ((u_int64_t) ip->ip_dst.s_addr << 19));
551 	return (i ^ (i >> 32));
552 }
553 
554 /*
555  * Receive data on a hook - both in upstream and downstream direction.
556  * We put the frame on the inbound queue, and try to initiate dequeuing
557  * sequence immediately. If inbound queue is full, discard one frame
558  * depending on dropping policy (from the head or from the tail of the
559  * queue).
560  */
561 static int
562 ngp_rcvdata(hook_p hook, item_p item)
563 {
564 	struct hookinfo *const hinfo = NG_HOOK_PRIVATE(hook);
565 	const priv_p priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook));
566 	struct timeval uuptime;
567 	struct timeval *now = &uuptime;
568 	struct ngp_fifo *ngp_f = NULL, *ngp_f1;
569 	struct ngp_hdr *ngp_h = NULL;
570 	struct mbuf *m;
571 	int hash, plen;
572 	int error = 0;
573 
574 	/*
575 	 * Shortcut from inbound to outbound hook when neither of
576 	 * bandwidth, delay, BER or duplication probability is
577 	 * configured, nor we have queued frames to drain.
578 	 */
579 	if (hinfo->run.qin_frames == 0 && hinfo->run.qout_frames == 0 &&
580 	    hinfo->noqueue) {
581 		struct hookinfo *dest;
582 		if (hinfo == &priv->lower)
583 			dest = &priv->upper;
584 		else
585 			dest = &priv->lower;
586 
587 		/* Send the frame. */
588 		plen = NGI_M(item)->m_pkthdr.len;
589 		NG_FWD_ITEM_HOOK(error, item, dest->hook);
590 
591 		/* Update stats. */
592 		if (error) {
593 			hinfo->stats.out_disc_frames++;
594 			hinfo->stats.out_disc_octets += plen;
595 		} else {
596 			hinfo->stats.fwd_frames++;
597 			hinfo->stats.fwd_octets += plen;
598 		}
599 
600 		return (error);
601 	}
602 
603 	microuptime(now);
604 
605 	/*
606 	 * If this was an empty queue, update service deadline time.
607 	 */
608 	if (hinfo->run.qin_frames == 0) {
609 		struct timeval *when = &hinfo->qin_utime;
610 		if (when->tv_sec < now->tv_sec || (when->tv_sec == now->tv_sec
611 		    && when->tv_usec < now->tv_usec)) {
612 			when->tv_sec = now->tv_sec;
613 			when->tv_usec = now->tv_usec;
614 		}
615 	}
616 
617 	/* Populate the packet header */
618 	ngp_h = uma_zalloc(ngp_zone, M_NOWAIT);
619 	KASSERT((ngp_h != NULL), ("ngp_h zalloc failed (1)"));
620 	NGI_GET_M(item, m);
621 	KASSERT(m != NULL, ("NGI_GET_M failed"));
622 	ngp_h->m = m;
623 	NG_FREE_ITEM(item);
624 
625 	if (hinfo->cfg.fifo)
626 		hash = 0;	/* all packets go into a single FIFO queue */
627 	else
628 		hash = ip_hash(m, priv->header_offset);
629 
630 	/* Find the appropriate FIFO queue for the packet and enqueue it*/
631 	TAILQ_FOREACH(ngp_f, &hinfo->fifo_head, fifo_le)
632 		if (hash == ngp_f->hash)
633 			break;
634 	if (ngp_f == NULL) {
635 		ngp_f = uma_zalloc(ngp_zone, M_NOWAIT);
636 		KASSERT(ngp_h != NULL, ("ngp_h zalloc failed (2)"));
637 		TAILQ_INIT(&ngp_f->packet_head);
638 		ngp_f->hash = hash;
639 		ngp_f->packets = 1;
640 		ngp_f->rr_deficit = hinfo->cfg.drr;	/* DRR quantum */
641 		hinfo->run.fifo_queues++;
642 		TAILQ_INSERT_TAIL(&ngp_f->packet_head, ngp_h, ngp_link);
643 		FIFO_VTIME_SORT(m->m_pkthdr.len);
644 	} else {
645 		TAILQ_INSERT_TAIL(&ngp_f->packet_head, ngp_h, ngp_link);
646 		ngp_f->packets++;
647 	}
648 	hinfo->run.qin_frames++;
649 	hinfo->run.qin_octets += m->m_pkthdr.len;
650 
651 	/* Discard a frame if inbound queue limit has been reached */
652 	if (hinfo->run.qin_frames > hinfo->cfg.qin_size_limit) {
653 		struct mbuf *m1;
654 		int longest = 0;
655 
656 		/* Find the longest queue */
657 		TAILQ_FOREACH(ngp_f1, &hinfo->fifo_head, fifo_le)
658 			if (ngp_f1->packets > longest) {
659 				longest = ngp_f1->packets;
660 				ngp_f = ngp_f1;
661 			}
662 
663 		/* Drop a frame from the queue head/tail, depending on cfg */
664 		if (hinfo->cfg.drophead)
665 			ngp_h = TAILQ_FIRST(&ngp_f->packet_head);
666 		else
667 			ngp_h = TAILQ_LAST(&ngp_f->packet_head, p_head);
668 		TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link);
669 		m1 = ngp_h->m;
670 		uma_zfree(ngp_zone, ngp_h);
671 		hinfo->run.qin_octets -= m1->m_pkthdr.len;
672 		hinfo->stats.in_disc_octets += m1->m_pkthdr.len;
673 		m_freem(m1);
674 		if (--(ngp_f->packets) == 0) {
675 			TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
676 			uma_zfree(ngp_zone, ngp_f);
677 			hinfo->run.fifo_queues--;
678 		}
679 		hinfo->run.qin_frames--;
680 		hinfo->stats.in_disc_frames++;
681 	} else if (hinfo->run.qin_frames > hinfo->cfg.qin_size_limit) {
682 		struct mbuf *m1;
683 		int longest = 0;
684 
685 		/* Find the longest queue */
686 		TAILQ_FOREACH(ngp_f1, &hinfo->fifo_head, fifo_le)
687 			if (ngp_f1->packets > longest) {
688 				longest = ngp_f1->packets;
689 				ngp_f = ngp_f1;
690 			}
691 
692 		/* Drop a frame from the queue head/tail, depending on cfg */
693 		if (hinfo->cfg.drophead)
694 			ngp_h = TAILQ_FIRST(&ngp_f->packet_head);
695 		else
696 			ngp_h = TAILQ_LAST(&ngp_f->packet_head, p_head);
697 		TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link);
698 		m1 = ngp_h->m;
699 		uma_zfree(ngp_zone, ngp_h);
700 		hinfo->run.qin_octets -= m1->m_pkthdr.len;
701 		hinfo->stats.in_disc_octets += m1->m_pkthdr.len;
702 		m_freem(m1);
703 		if (--(ngp_f->packets) == 0) {
704 			TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
705 			uma_zfree(ngp_zone, ngp_f);
706 			hinfo->run.fifo_queues--;
707 		}
708 		hinfo->run.qin_frames--;
709 		hinfo->stats.in_disc_frames++;
710 	}
711 
712 	/*
713 	 * Try to start the dequeuing process immediately.
714 	 */
715 	pipe_dequeue(hinfo, now);
716 
717 	return (0);
718 }
719 
720 
721 /*
722  * Dequeueing sequence - we basically do the following:
723  *  1) Try to extract the frame from the inbound (bandwidth) queue;
724  *  2) In accordance to BER specified, discard the frame randomly;
725  *  3) If the frame survives BER, prepend it with delay info and move it
726  *     to outbound (delay) queue;
727  *  4) Loop to 2) until bandwidth quota for this timeslice is reached, or
728  *     inbound queue is flushed completely;
729  *  5) Dequeue frames from the outbound queue and send them downstream until
730  *     outbound queue is flushed completely, or the next frame in the queue
731  *     is not due to be dequeued yet
732  */
733 static void
734 pipe_dequeue(struct hookinfo *hinfo, struct timeval *now) {
735 	static uint64_t rand, oldrand;
736 	const node_p node = NG_HOOK_NODE(hinfo->hook);
737 	const priv_p priv = NG_NODE_PRIVATE(node);
738 	struct hookinfo *dest;
739 	struct ngp_fifo *ngp_f, *ngp_f1;
740 	struct ngp_hdr *ngp_h;
741 	struct timeval *when;
742 	struct mbuf *m;
743 	int plen, error = 0;
744 
745 	/* Which one is the destination hook? */
746 	if (hinfo == &priv->lower)
747 		dest = &priv->upper;
748 	else
749 		dest = &priv->lower;
750 
751 	/* Bandwidth queue processing */
752 	while ((ngp_f = TAILQ_FIRST(&hinfo->fifo_head))) {
753 		when = &hinfo->qin_utime;
754 		if (when->tv_sec > now->tv_sec || (when->tv_sec == now->tv_sec
755 		    && when->tv_usec > now->tv_usec))
756 			break;
757 
758 		ngp_h = TAILQ_FIRST(&ngp_f->packet_head);
759 		m = ngp_h->m;
760 
761 		/* Deficit Round Robin (DRR) processing */
762 		if (hinfo->cfg.drr) {
763 			if (ngp_f->rr_deficit >= m->m_pkthdr.len) {
764 				ngp_f->rr_deficit -= m->m_pkthdr.len;
765 			} else {
766 				ngp_f->rr_deficit += hinfo->cfg.drr;
767 				TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
768 				TAILQ_INSERT_TAIL(&hinfo->fifo_head,
769 				    ngp_f, fifo_le);
770 				continue;
771 			}
772 		}
773 
774 		/*
775 		 * Either create a duplicate and pass it on, or dequeue
776 		 * the original packet...
777 		 */
778 		if (hinfo->cfg.duplicate &&
779 		    random() % 100 <= hinfo->cfg.duplicate) {
780 			ngp_h = uma_zalloc(ngp_zone, M_NOWAIT);
781 			KASSERT(ngp_h != NULL, ("ngp_h zalloc failed (3)"));
782 			m = m_dup(m, M_NOWAIT);
783 			KASSERT(m != NULL, ("m_dup failed"));
784 			ngp_h->m = m;
785 		} else {
786 			TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link);
787 			hinfo->run.qin_frames--;
788 			hinfo->run.qin_octets -= m->m_pkthdr.len;
789 			ngp_f->packets--;
790 		}
791 
792 		/* Calculate the serialization delay */
793 		if (hinfo->cfg.bandwidth) {
794 			hinfo->qin_utime.tv_usec +=
795 			    ((uint64_t) m->m_pkthdr.len + priv->overhead ) *
796 			    8000000 / hinfo->cfg.bandwidth;
797 			hinfo->qin_utime.tv_sec +=
798 			    hinfo->qin_utime.tv_usec / 1000000;
799 			hinfo->qin_utime.tv_usec =
800 			    hinfo->qin_utime.tv_usec % 1000000;
801 		}
802 		when = &ngp_h->when;
803 		when->tv_sec = hinfo->qin_utime.tv_sec;
804 		when->tv_usec = hinfo->qin_utime.tv_usec;
805 
806 		/* Sort / rearrange inbound queues */
807 		if (ngp_f->packets) {
808 			if (hinfo->cfg.wfq) {
809 				TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
810 				FIFO_VTIME_SORT(TAILQ_FIRST(
811 				    &ngp_f->packet_head)->m->m_pkthdr.len)
812 			}
813 		} else {
814 			TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
815 			uma_zfree(ngp_zone, ngp_f);
816 			hinfo->run.fifo_queues--;
817 		}
818 
819 		/* Randomly discard the frame, according to BER setting */
820 		if (hinfo->cfg.ber) {
821 			oldrand = rand;
822 			rand = random();
823 			if (((oldrand ^ rand) << 17) >=
824 			    hinfo->ber_p[priv->overhead + m->m_pkthdr.len]) {
825 				hinfo->stats.out_disc_frames++;
826 				hinfo->stats.out_disc_octets += m->m_pkthdr.len;
827 				uma_zfree(ngp_zone, ngp_h);
828 				m_freem(m);
829 				continue;
830 			}
831 		}
832 
833 		/* Discard frame if outbound queue size limit exceeded */
834 		if (hinfo->cfg.qout_size_limit &&
835 		    hinfo->run.qout_frames>=hinfo->cfg.qout_size_limit) {
836 			hinfo->stats.out_disc_frames++;
837 			hinfo->stats.out_disc_octets += m->m_pkthdr.len;
838 			uma_zfree(ngp_zone, ngp_h);
839 			m_freem(m);
840 			continue;
841 		}
842 
843 		/* Calculate the propagation delay */
844 		when->tv_usec += priv->delay;
845 		when->tv_sec += when->tv_usec / 1000000;
846 		when->tv_usec = when->tv_usec % 1000000;
847 
848 		/* Put the frame into the delay queue */
849 		TAILQ_INSERT_TAIL(&hinfo->qout_head, ngp_h, ngp_link);
850 		hinfo->run.qout_frames++;
851 		hinfo->run.qout_octets += m->m_pkthdr.len;
852 	}
853 
854 	/* Delay queue processing */
855 	while ((ngp_h = TAILQ_FIRST(&hinfo->qout_head))) {
856 		when = &ngp_h->when;
857 		m = ngp_h->m;
858 		if (when->tv_sec > now->tv_sec ||
859 		    (when->tv_sec == now->tv_sec &&
860 		    when->tv_usec > now->tv_usec))
861 			break;
862 
863 		/* Update outbound queue stats */
864 		plen = m->m_pkthdr.len;
865 		hinfo->run.qout_frames--;
866 		hinfo->run.qout_octets -= plen;
867 
868 		/* Dequeue the packet from qout */
869 		TAILQ_REMOVE(&hinfo->qout_head, ngp_h, ngp_link);
870 		uma_zfree(ngp_zone, ngp_h);
871 
872 		NG_SEND_DATA(error, dest->hook, m, meta);
873 		if (error) {
874 			hinfo->stats.out_disc_frames++;
875 			hinfo->stats.out_disc_octets += plen;
876 		} else {
877 			hinfo->stats.fwd_frames++;
878 			hinfo->stats.fwd_octets += plen;
879 		}
880 	}
881 
882 	if ((hinfo->run.qin_frames != 0 || hinfo->run.qout_frames != 0) &&
883 	    !priv->timer_scheduled) {
884 		ng_callout(&priv->timer, node, NULL, 1, ngp_callout, NULL, 0);
885 		priv->timer_scheduled = 1;
886 	}
887 }
888 
889 /*
890  * This routine is called on every clock tick.  We poll connected hooks
891  * for queued frames by calling pipe_dequeue().
892  */
893 static void
894 ngp_callout(node_p node, hook_p hook, void *arg1, int arg2)
895 {
896 	const priv_p priv = NG_NODE_PRIVATE(node);
897 	struct timeval now;
898 
899 	priv->timer_scheduled = 0;
900 	microuptime(&now);
901 	if (priv->upper.hook != NULL)
902 		pipe_dequeue(&priv->upper, &now);
903 	if (priv->lower.hook != NULL)
904 		pipe_dequeue(&priv->lower, &now);
905 }
906 
907 /*
908  * Shutdown processing
909  *
910  * This is tricky. If we have both a lower and upper hook, then we
911  * probably want to extricate ourselves and leave the two peers
912  * still linked to each other. Otherwise we should just shut down as
913  * a normal node would.
914  */
915 static int
916 ngp_shutdown(node_p node)
917 {
918 	const priv_p priv = NG_NODE_PRIVATE(node);
919 
920 	if (priv->timer_scheduled)
921 		ng_uncallout(&priv->timer, node);
922 	if (priv->lower.hook && priv->upper.hook)
923 		ng_bypass(priv->lower.hook, priv->upper.hook);
924 	else {
925 		if (priv->upper.hook != NULL)
926 			ng_rmhook_self(priv->upper.hook);
927 		if (priv->lower.hook != NULL)
928 			ng_rmhook_self(priv->lower.hook);
929 	}
930 	NG_NODE_UNREF(node);
931 	free(priv, M_NG_PIPE);
932 	return (0);
933 }
934 
935 
936 /*
937  * Hook disconnection
938  */
939 static int
940 ngp_disconnect(hook_p hook)
941 {
942 	struct hookinfo *const hinfo = NG_HOOK_PRIVATE(hook);
943 	struct ngp_fifo *ngp_f;
944 	struct ngp_hdr *ngp_h;
945 
946 	KASSERT(hinfo != NULL, ("%s: null info", __FUNCTION__));
947 	hinfo->hook = NULL;
948 
949 	/* Flush all fifo queues associated with the hook */
950 	while ((ngp_f = TAILQ_FIRST(&hinfo->fifo_head))) {
951 		while ((ngp_h = TAILQ_FIRST(&ngp_f->packet_head))) {
952 			TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link);
953 			m_freem(ngp_h->m);
954 			uma_zfree(ngp_zone, ngp_h);
955 		}
956 		TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
957 		uma_zfree(ngp_zone, ngp_f);
958 	}
959 
960 	/* Flush the delay queue */
961 	while ((ngp_h = TAILQ_FIRST(&hinfo->qout_head))) {
962 		TAILQ_REMOVE(&hinfo->qout_head, ngp_h, ngp_link);
963 		m_freem(ngp_h->m);
964 		uma_zfree(ngp_zone, ngp_h);
965 	}
966 
967 	/* Release the packet loss probability table (BER) */
968 	if (hinfo->ber_p)
969 		free(hinfo->ber_p, M_NG_PIPE);
970 
971 	return (0);
972 }
973 
974 static int
975 ngp_modevent(module_t mod, int type, void *unused)
976 {
977 	int error = 0;
978 
979 	switch (type) {
980 	case MOD_LOAD:
981 		ngp_zone = uma_zcreate("ng_pipe", max(sizeof(struct ngp_hdr),
982 		    sizeof (struct ngp_fifo)), NULL, NULL, NULL, NULL,
983 		    UMA_ALIGN_PTR, 0);
984 		if (ngp_zone == NULL)
985 			panic("ng_pipe: couldn't allocate descriptor zone");
986 		break;
987 	case MOD_UNLOAD:
988 		uma_zdestroy(ngp_zone);
989 		break;
990 	default:
991 		error = EOPNOTSUPP;
992 		break;
993 	}
994 
995 	return (error);
996 }
997