xref: /freebsd/sys/netpfil/ipfw/ip_dummynet.c (revision b0d29bc47dba79f6f38e67eabadfb4b32ffd9390)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Codel/FQ_Codel and PIE/FQ-PIE Code:
5  * Copyright (C) 2016 Centre for Advanced Internet Architectures,
6  *  Swinburne University of Technology, Melbourne, Australia.
7  * Portions of this code were made possible in part by a gift from
8  *  The Comcast Innovation Fund.
9  * Implemented by Rasool Al-Saadi <ralsaadi@swin.edu.au>
10  *
11  * Copyright (c) 1998-2002,2010 Luigi Rizzo, Universita` di Pisa
12  * Portions Copyright (c) 2000 Akamba Corp.
13  * All rights reserved
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 /*
41  * Configuration and internal object management for dummynet.
42  */
43 
44 #include "opt_inet6.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/module.h>
53 #include <sys/mutex.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/rwlock.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/time.h>
60 #include <sys/taskqueue.h>
61 #include <net/if.h>	/* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */
62 #include <netinet/in.h>
63 #include <netinet/ip_var.h>	/* ip_output(), IP_FORWARDING */
64 #include <netinet/ip_fw.h>
65 #include <netinet/ip_dummynet.h>
66 
67 #include <netpfil/ipfw/ip_fw_private.h>
68 #include <netpfil/ipfw/dn_heap.h>
69 #include <netpfil/ipfw/ip_dn_private.h>
70 #ifdef NEW_AQM
71 #include <netpfil/ipfw/dn_aqm.h>
72 #endif
73 #include <netpfil/ipfw/dn_sched.h>
74 
75 /* which objects to copy */
76 #define DN_C_LINK 	0x01
77 #define DN_C_SCH	0x02
78 #define DN_C_FLOW	0x04
79 #define DN_C_FS		0x08
80 #define DN_C_QUEUE	0x10
81 
82 /* we use this argument in case of a schk_new */
83 struct schk_new_arg {
84 	struct dn_alg *fp;
85 	struct dn_sch *sch;
86 };
87 
88 /*---- callout hooks. ----*/
89 static struct callout dn_timeout;
90 static int dn_gone;
91 static struct task	dn_task;
92 static struct taskqueue	*dn_tq = NULL;
93 
94 static void
95 dummynet(void *arg)
96 {
97 
98 	(void)arg;	/* UNUSED */
99 	taskqueue_enqueue(dn_tq, &dn_task);
100 }
101 
102 void
103 dn_reschedule(void)
104 {
105 
106 	if (dn_gone != 0)
107 		return;
108 	callout_reset_sbt(&dn_timeout, tick_sbt, 0, dummynet, NULL,
109 	    C_HARDCLOCK | C_DIRECT_EXEC);
110 }
111 /*----- end of callout hooks -----*/
112 
113 #ifdef NEW_AQM
114 /* Return AQM descriptor for given type or name. */
115 static struct dn_aqm *
116 find_aqm_type(int type, char *name)
117 {
118 	struct dn_aqm *d;
119 
120 	SLIST_FOREACH(d, &dn_cfg.aqmlist, next) {
121 		if (d->type == type || (name && !strcasecmp(d->name, name)))
122 			return d;
123 	}
124 	return NULL; /* not found */
125 }
126 #endif
127 
128 /* Return a scheduler descriptor given the type or name. */
129 static struct dn_alg *
130 find_sched_type(int type, char *name)
131 {
132 	struct dn_alg *d;
133 
134 	SLIST_FOREACH(d, &dn_cfg.schedlist, next) {
135 		if (d->type == type || (name && !strcasecmp(d->name, name)))
136 			return d;
137 	}
138 	return NULL; /* not found */
139 }
140 
141 int
142 ipdn_bound_var(int *v, int dflt, int lo, int hi, const char *msg)
143 {
144 	int oldv = *v;
145 	const char *op = NULL;
146 	if (dflt < lo)
147 		dflt = lo;
148 	if (dflt > hi)
149 		dflt = hi;
150 	if (oldv < lo) {
151 		*v = dflt;
152 		op = "Bump";
153 	} else if (oldv > hi) {
154 		*v = hi;
155 		op = "Clamp";
156 	} else
157 		return *v;
158 	if (op && msg)
159 		printf("%s %s to %d (was %d)\n", op, msg, *v, oldv);
160 	return *v;
161 }
162 
163 /*---- flow_id mask, hash and compare functions ---*/
164 /*
165  * The flow_id includes the 5-tuple, the queue/pipe number
166  * which we store in the extra area in host order,
167  * and for ipv6 also the flow_id6.
168  * XXX see if we want the tos byte (can store in 'flags')
169  */
170 static struct ipfw_flow_id *
171 flow_id_mask(struct ipfw_flow_id *mask, struct ipfw_flow_id *id)
172 {
173 	int is_v6 = IS_IP6_FLOW_ID(id);
174 
175 	id->dst_port &= mask->dst_port;
176 	id->src_port &= mask->src_port;
177 	id->proto &= mask->proto;
178 	id->extra &= mask->extra;
179 	if (is_v6) {
180 		APPLY_MASK(&id->dst_ip6, &mask->dst_ip6);
181 		APPLY_MASK(&id->src_ip6, &mask->src_ip6);
182 		id->flow_id6 &= mask->flow_id6;
183 	} else {
184 		id->dst_ip &= mask->dst_ip;
185 		id->src_ip &= mask->src_ip;
186 	}
187 	return id;
188 }
189 
190 /* computes an OR of two masks, result in dst and also returned */
191 static struct ipfw_flow_id *
192 flow_id_or(struct ipfw_flow_id *src, struct ipfw_flow_id *dst)
193 {
194 	int is_v6 = IS_IP6_FLOW_ID(dst);
195 
196 	dst->dst_port |= src->dst_port;
197 	dst->src_port |= src->src_port;
198 	dst->proto |= src->proto;
199 	dst->extra |= src->extra;
200 	if (is_v6) {
201 #define OR_MASK(_d, _s)                          \
202     (_d)->__u6_addr.__u6_addr32[0] |= (_s)->__u6_addr.__u6_addr32[0]; \
203     (_d)->__u6_addr.__u6_addr32[1] |= (_s)->__u6_addr.__u6_addr32[1]; \
204     (_d)->__u6_addr.__u6_addr32[2] |= (_s)->__u6_addr.__u6_addr32[2]; \
205     (_d)->__u6_addr.__u6_addr32[3] |= (_s)->__u6_addr.__u6_addr32[3];
206 		OR_MASK(&dst->dst_ip6, &src->dst_ip6);
207 		OR_MASK(&dst->src_ip6, &src->src_ip6);
208 #undef OR_MASK
209 		dst->flow_id6 |= src->flow_id6;
210 	} else {
211 		dst->dst_ip |= src->dst_ip;
212 		dst->src_ip |= src->src_ip;
213 	}
214 	return dst;
215 }
216 
217 static int
218 nonzero_mask(struct ipfw_flow_id *m)
219 {
220 	if (m->dst_port || m->src_port || m->proto || m->extra)
221 		return 1;
222 	if (IS_IP6_FLOW_ID(m)) {
223 		return
224 			m->dst_ip6.__u6_addr.__u6_addr32[0] ||
225 			m->dst_ip6.__u6_addr.__u6_addr32[1] ||
226 			m->dst_ip6.__u6_addr.__u6_addr32[2] ||
227 			m->dst_ip6.__u6_addr.__u6_addr32[3] ||
228 			m->src_ip6.__u6_addr.__u6_addr32[0] ||
229 			m->src_ip6.__u6_addr.__u6_addr32[1] ||
230 			m->src_ip6.__u6_addr.__u6_addr32[2] ||
231 			m->src_ip6.__u6_addr.__u6_addr32[3] ||
232 			m->flow_id6;
233 	} else {
234 		return m->dst_ip || m->src_ip;
235 	}
236 }
237 
238 /* XXX we may want a better hash function */
239 static uint32_t
240 flow_id_hash(struct ipfw_flow_id *id)
241 {
242     uint32_t i;
243 
244     if (IS_IP6_FLOW_ID(id)) {
245 	uint32_t *d = (uint32_t *)&id->dst_ip6;
246 	uint32_t *s = (uint32_t *)&id->src_ip6;
247         i = (d[0]      ) ^ (d[1])       ^
248             (d[2]      ) ^ (d[3])       ^
249             (d[0] >> 15) ^ (d[1] >> 15) ^
250             (d[2] >> 15) ^ (d[3] >> 15) ^
251             (s[0] <<  1) ^ (s[1] <<  1) ^
252             (s[2] <<  1) ^ (s[3] <<  1) ^
253             (s[0] << 16) ^ (s[1] << 16) ^
254             (s[2] << 16) ^ (s[3] << 16) ^
255             (id->dst_port << 1) ^ (id->src_port) ^
256 	    (id->extra) ^
257             (id->proto ) ^ (id->flow_id6);
258     } else {
259         i = (id->dst_ip)        ^ (id->dst_ip >> 15) ^
260             (id->src_ip << 1)   ^ (id->src_ip >> 16) ^
261 	    (id->extra) ^
262             (id->dst_port << 1) ^ (id->src_port)     ^ (id->proto);
263     }
264     return i;
265 }
266 
267 /* Like bcmp, returns 0 if ids match, 1 otherwise. */
268 static int
269 flow_id_cmp(struct ipfw_flow_id *id1, struct ipfw_flow_id *id2)
270 {
271 	int is_v6 = IS_IP6_FLOW_ID(id1);
272 
273 	if (!is_v6) {
274 	    if (IS_IP6_FLOW_ID(id2))
275 		return 1; /* different address families */
276 
277 	    return (id1->dst_ip == id2->dst_ip &&
278 		    id1->src_ip == id2->src_ip &&
279 		    id1->dst_port == id2->dst_port &&
280 		    id1->src_port == id2->src_port &&
281 		    id1->proto == id2->proto &&
282 		    id1->extra == id2->extra) ? 0 : 1;
283 	}
284 	/* the ipv6 case */
285 	return (
286 	    !bcmp(&id1->dst_ip6,&id2->dst_ip6, sizeof(id1->dst_ip6)) &&
287 	    !bcmp(&id1->src_ip6,&id2->src_ip6, sizeof(id1->src_ip6)) &&
288 	    id1->dst_port == id2->dst_port &&
289 	    id1->src_port == id2->src_port &&
290 	    id1->proto == id2->proto &&
291 	    id1->extra == id2->extra &&
292 	    id1->flow_id6 == id2->flow_id6) ? 0 : 1;
293 }
294 /*--------- end of flow-id mask, hash and compare ---------*/
295 
296 /*--- support functions for the qht hashtable ----
297  * Entries are hashed by flow-id
298  */
299 static uint32_t
300 q_hash(uintptr_t key, int flags, void *arg)
301 {
302 	/* compute the hash slot from the flow id */
303 	struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
304 		&((struct dn_queue *)key)->ni.fid :
305 		(struct ipfw_flow_id *)key;
306 
307 	return flow_id_hash(id);
308 }
309 
310 static int
311 q_match(void *obj, uintptr_t key, int flags, void *arg)
312 {
313 	struct dn_queue *o = (struct dn_queue *)obj;
314 	struct ipfw_flow_id *id2;
315 
316 	if (flags & DNHT_KEY_IS_OBJ) {
317 		/* compare pointers */
318 		id2 = &((struct dn_queue *)key)->ni.fid;
319 	} else {
320 		id2 = (struct ipfw_flow_id *)key;
321 	}
322 	return (0 == flow_id_cmp(&o->ni.fid,  id2));
323 }
324 
325 /*
326  * create a new queue instance for the given 'key'.
327  */
328 static void *
329 q_new(uintptr_t key, int flags, void *arg)
330 {
331 	struct dn_queue *q, *template = arg;
332 	struct dn_fsk *fs = template->fs;
333 	int size = sizeof(*q) + fs->sched->fp->q_datalen;
334 
335 	q = malloc(size, M_DUMMYNET, M_NOWAIT | M_ZERO);
336 	if (q == NULL) {
337 		D("no memory for new queue");
338 		return NULL;
339 	}
340 
341 	set_oid(&q->ni.oid, DN_QUEUE, size);
342 	if (fs->fs.flags & DN_QHT_HASH)
343 		q->ni.fid = *(struct ipfw_flow_id *)key;
344 	q->fs = fs;
345 	q->_si = template->_si;
346 	q->_si->q_count++;
347 
348 	if (fs->sched->fp->new_queue)
349 		fs->sched->fp->new_queue(q);
350 
351 #ifdef NEW_AQM
352 	/* call AQM init function after creating a queue*/
353 	if (fs->aqmfp && fs->aqmfp->init)
354 		if(fs->aqmfp->init(q))
355 			D("unable to init AQM for fs %d", fs->fs.fs_nr);
356 #endif
357 	dn_cfg.queue_count++;
358 
359 	return q;
360 }
361 
362 /*
363  * Notify schedulers that a queue is going away.
364  * If (flags & DN_DESTROY), also free the packets.
365  * The version for callbacks is called q_delete_cb().
366  */
367 static void
368 dn_delete_queue(struct dn_queue *q, int flags)
369 {
370 	struct dn_fsk *fs = q->fs;
371 
372 #ifdef NEW_AQM
373 	/* clean up AQM status for queue 'q'
374 	 * cleanup here is called just with MULTIQUEUE
375 	 */
376 	if (fs && fs->aqmfp && fs->aqmfp->cleanup)
377 		fs->aqmfp->cleanup(q);
378 #endif
379 	// D("fs %p si %p\n", fs, q->_si);
380 	/* notify the parent scheduler that the queue is going away */
381 	if (fs && fs->sched->fp->free_queue)
382 		fs->sched->fp->free_queue(q);
383 	q->_si->q_count--;
384 	q->_si = NULL;
385 	if (flags & DN_DESTROY) {
386 		if (q->mq.head)
387 			dn_free_pkts(q->mq.head);
388 		bzero(q, sizeof(*q));	// safety
389 		free(q, M_DUMMYNET);
390 		dn_cfg.queue_count--;
391 	}
392 }
393 
394 static int
395 q_delete_cb(void *q, void *arg)
396 {
397 	int flags = (int)(uintptr_t)arg;
398 	dn_delete_queue(q, flags);
399 	return (flags & DN_DESTROY) ? DNHT_SCAN_DEL : 0;
400 }
401 
402 /*
403  * calls dn_delete_queue/q_delete_cb on all queues,
404  * which notifies the parent scheduler and possibly drains packets.
405  * flags & DN_DESTROY: drains queues and destroy qht;
406  */
407 static void
408 qht_delete(struct dn_fsk *fs, int flags)
409 {
410 	ND("fs %d start flags %d qht %p",
411 		fs->fs.fs_nr, flags, fs->qht);
412 	if (!fs->qht)
413 		return;
414 	if (fs->fs.flags & DN_QHT_HASH) {
415 		dn_ht_scan(fs->qht, q_delete_cb, (void *)(uintptr_t)flags);
416 		if (flags & DN_DESTROY) {
417 			dn_ht_free(fs->qht, 0);
418 			fs->qht = NULL;
419 		}
420 	} else {
421 		dn_delete_queue((struct dn_queue *)(fs->qht), flags);
422 		if (flags & DN_DESTROY)
423 			fs->qht = NULL;
424 	}
425 }
426 
427 /*
428  * Find and possibly create the queue for a MULTIQUEUE scheduler.
429  * We never call it for !MULTIQUEUE (the queue is in the sch_inst).
430  */
431 struct dn_queue *
432 ipdn_q_find(struct dn_fsk *fs, struct dn_sch_inst *si,
433 	struct ipfw_flow_id *id)
434 {
435 	struct dn_queue template;
436 
437 	template._si = si;
438 	template.fs = fs;
439 
440 	if (fs->fs.flags & DN_QHT_HASH) {
441 		struct ipfw_flow_id masked_id;
442 		if (fs->qht == NULL) {
443 			fs->qht = dn_ht_init(NULL, fs->fs.buckets,
444 				offsetof(struct dn_queue, q_next),
445 				q_hash, q_match, q_new);
446 			if (fs->qht == NULL)
447 				return NULL;
448 		}
449 		masked_id = *id;
450 		flow_id_mask(&fs->fsk_mask, &masked_id);
451 		return dn_ht_find(fs->qht, (uintptr_t)&masked_id,
452 			DNHT_INSERT, &template);
453 	} else {
454 		if (fs->qht == NULL)
455 			fs->qht = q_new(0, 0, &template);
456 		return (struct dn_queue *)fs->qht;
457 	}
458 }
459 /*--- end of queue hash table ---*/
460 
461 /*--- support functions for the sch_inst hashtable ----
462  *
463  * These are hashed by flow-id
464  */
465 static uint32_t
466 si_hash(uintptr_t key, int flags, void *arg)
467 {
468 	/* compute the hash slot from the flow id */
469 	struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
470 		&((struct dn_sch_inst *)key)->ni.fid :
471 		(struct ipfw_flow_id *)key;
472 
473 	return flow_id_hash(id);
474 }
475 
476 static int
477 si_match(void *obj, uintptr_t key, int flags, void *arg)
478 {
479 	struct dn_sch_inst *o = obj;
480 	struct ipfw_flow_id *id2;
481 
482 	id2 = (flags & DNHT_KEY_IS_OBJ) ?
483 		&((struct dn_sch_inst *)key)->ni.fid :
484 		(struct ipfw_flow_id *)key;
485 	return flow_id_cmp(&o->ni.fid,  id2) == 0;
486 }
487 
488 /*
489  * create a new instance for the given 'key'
490  * Allocate memory for instance, delay line and scheduler private data.
491  */
492 static void *
493 si_new(uintptr_t key, int flags, void *arg)
494 {
495 	struct dn_schk *s = arg;
496 	struct dn_sch_inst *si;
497 	int l = sizeof(*si) + s->fp->si_datalen;
498 
499 	si = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
500 	if (si == NULL)
501 		goto error;
502 
503 	/* Set length only for the part passed up to userland. */
504 	set_oid(&si->ni.oid, DN_SCH_I, sizeof(struct dn_flow));
505 	set_oid(&(si->dline.oid), DN_DELAY_LINE,
506 		sizeof(struct delay_line));
507 	/* mark si and dline as outside the event queue */
508 	si->ni.oid.id = si->dline.oid.id = -1;
509 
510 	si->sched = s;
511 	si->dline.si = si;
512 
513 	if (s->fp->new_sched && s->fp->new_sched(si)) {
514 		D("new_sched error");
515 		goto error;
516 	}
517 	if (s->sch.flags & DN_HAVE_MASK)
518 		si->ni.fid = *(struct ipfw_flow_id *)key;
519 
520 #ifdef NEW_AQM
521 	/* init AQM status for !DN_MULTIQUEUE sched*/
522 	if (!(s->fp->flags & DN_MULTIQUEUE))
523 		if (s->fs->aqmfp && s->fs->aqmfp->init)
524 			if(s->fs->aqmfp->init((struct dn_queue *)(si + 1))) {
525 				D("unable to init AQM for fs %d", s->fs->fs.fs_nr);
526 				goto error;
527 			}
528 #endif
529 
530 	dn_cfg.si_count++;
531 	return si;
532 
533 error:
534 	if (si) {
535 		bzero(si, sizeof(*si)); // safety
536 		free(si, M_DUMMYNET);
537 	}
538         return NULL;
539 }
540 
541 /*
542  * Callback from siht to delete all scheduler instances. Remove
543  * si and delay line from the system heap, destroy all queues.
544  * We assume that all flowset have been notified and do not
545  * point to us anymore.
546  */
547 static int
548 si_destroy(void *_si, void *arg)
549 {
550 	struct dn_sch_inst *si = _si;
551 	struct dn_schk *s = si->sched;
552 	struct delay_line *dl = &si->dline;
553 
554 	if (dl->oid.subtype) /* remove delay line from event heap */
555 		heap_extract(&dn_cfg.evheap, dl);
556 	dn_free_pkts(dl->mq.head);	/* drain delay line */
557 	if (si->kflags & DN_ACTIVE) /* remove si from event heap */
558 		heap_extract(&dn_cfg.evheap, si);
559 
560 #ifdef NEW_AQM
561 	/* clean up AQM status for !DN_MULTIQUEUE sched
562 	 * Note that all queues belong to fs were cleaned up in fsk_detach.
563 	 * When drain_scheduler is called s->fs and q->fs are pointing
564 	 * to a correct fs, so we can use fs in this case.
565 	 */
566 	if (!(s->fp->flags & DN_MULTIQUEUE)) {
567 		struct dn_queue *q = (struct dn_queue *)(si + 1);
568 		if (q->aqm_status && q->fs->aqmfp)
569 			if (q->fs->aqmfp->cleanup)
570 				q->fs->aqmfp->cleanup(q);
571 	}
572 #endif
573 	if (s->fp->free_sched)
574 		s->fp->free_sched(si);
575 	bzero(si, sizeof(*si));	/* safety */
576 	free(si, M_DUMMYNET);
577 	dn_cfg.si_count--;
578 	return DNHT_SCAN_DEL;
579 }
580 
581 /*
582  * Find the scheduler instance for this packet. If we need to apply
583  * a mask, do on a local copy of the flow_id to preserve the original.
584  * Assume siht is always initialized if we have a mask.
585  */
586 struct dn_sch_inst *
587 ipdn_si_find(struct dn_schk *s, struct ipfw_flow_id *id)
588 {
589 
590 	if (s->sch.flags & DN_HAVE_MASK) {
591 		struct ipfw_flow_id id_t = *id;
592 		flow_id_mask(&s->sch.sched_mask, &id_t);
593 		return dn_ht_find(s->siht, (uintptr_t)&id_t,
594 			DNHT_INSERT, s);
595 	}
596 	if (!s->siht)
597 		s->siht = si_new(0, 0, s);
598 	return (struct dn_sch_inst *)s->siht;
599 }
600 
601 /* callback to flush credit for the scheduler instance */
602 static int
603 si_reset_credit(void *_si, void *arg)
604 {
605 	struct dn_sch_inst *si = _si;
606 	struct dn_link *p = &si->sched->link;
607 
608 	si->credit = p->burst + (dn_cfg.io_fast ?  p->bandwidth : 0);
609 	return 0;
610 }
611 
612 static void
613 schk_reset_credit(struct dn_schk *s)
614 {
615 	if (s->sch.flags & DN_HAVE_MASK)
616 		dn_ht_scan(s->siht, si_reset_credit, NULL);
617 	else if (s->siht)
618 		si_reset_credit(s->siht, NULL);
619 }
620 /*---- end of sch_inst hashtable ---------------------*/
621 
622 /*-------------------------------------------------------
623  * flowset hash (fshash) support. Entries are hashed by fs_nr.
624  * New allocations are put in the fsunlinked list, from which
625  * they are removed when they point to a specific scheduler.
626  */
627 static uint32_t
628 fsk_hash(uintptr_t key, int flags, void *arg)
629 {
630 	uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
631 		((struct dn_fsk *)key)->fs.fs_nr;
632 
633 	return ( (i>>8)^(i>>4)^i );
634 }
635 
636 static int
637 fsk_match(void *obj, uintptr_t key, int flags, void *arg)
638 {
639 	struct dn_fsk *fs = obj;
640 	int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
641 		((struct dn_fsk *)key)->fs.fs_nr;
642 
643 	return (fs->fs.fs_nr == i);
644 }
645 
646 static void *
647 fsk_new(uintptr_t key, int flags, void *arg)
648 {
649 	struct dn_fsk *fs;
650 
651 	fs = malloc(sizeof(*fs), M_DUMMYNET, M_NOWAIT | M_ZERO);
652 	if (fs) {
653 		set_oid(&fs->fs.oid, DN_FS, sizeof(fs->fs));
654 		dn_cfg.fsk_count++;
655 		fs->drain_bucket = 0;
656 		SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
657 	}
658 	return fs;
659 }
660 
661 #ifdef NEW_AQM
662 /* callback function for cleaning up AQM queue status belongs to a flowset
663  * connected to scheduler instance '_si' (for !DN_MULTIQUEUE only).
664  */
665 static int
666 si_cleanup_q(void *_si, void *arg)
667 {
668 	struct dn_sch_inst *si = _si;
669 
670 	if (!(si->sched->fp->flags & DN_MULTIQUEUE)) {
671 		if (si->sched->fs->aqmfp && si->sched->fs->aqmfp->cleanup)
672 			si->sched->fs->aqmfp->cleanup((struct dn_queue *) (si+1));
673 	}
674 	return 0;
675 }
676 
677 /* callback to clean up queue AQM status.*/
678 static int
679 q_cleanup_q(void *_q, void *arg)
680 {
681 	struct dn_queue *q = _q;
682 	q->fs->aqmfp->cleanup(q);
683 	return 0;
684 }
685 
686 /* Clean up all AQM queues status belongs to flowset 'fs' and then
687  * deconfig AQM for flowset 'fs'
688  */
689 static void
690 aqm_cleanup_deconfig_fs(struct dn_fsk *fs)
691 {
692 	struct dn_sch_inst *si;
693 
694 	/* clean up AQM status for all queues for !DN_MULTIQUEUE sched*/
695 	if (fs->fs.fs_nr > DN_MAX_ID) {
696 		if (fs->sched && !(fs->sched->fp->flags & DN_MULTIQUEUE)) {
697 			if (fs->sched->sch.flags & DN_HAVE_MASK)
698 				dn_ht_scan(fs->sched->siht, si_cleanup_q, NULL);
699 			else {
700 					/* single si i.e. no sched mask */
701 					si = (struct dn_sch_inst *) fs->sched->siht;
702 					if (si && fs->aqmfp && fs->aqmfp->cleanup)
703 						fs->aqmfp->cleanup((struct dn_queue *) (si+1));
704 			}
705 		}
706 	}
707 
708 	/* clean up AQM status for all queues for DN_MULTIQUEUE sched*/
709 	if (fs->sched && fs->sched->fp->flags & DN_MULTIQUEUE && fs->qht) {
710 			if (fs->fs.flags & DN_QHT_HASH)
711 				dn_ht_scan(fs->qht, q_cleanup_q, NULL);
712 			else
713 				fs->aqmfp->cleanup((struct dn_queue *)(fs->qht));
714 	}
715 
716 	/* deconfig AQM */
717 	if(fs->aqmcfg && fs->aqmfp && fs->aqmfp->deconfig)
718 		fs->aqmfp->deconfig(fs);
719 }
720 #endif
721 
722 /*
723  * detach flowset from its current scheduler. Flags as follows:
724  * DN_DETACH removes from the fsk_list
725  * DN_DESTROY deletes individual queues
726  * DN_DELETE_FS destroys the flowset (otherwise goes in unlinked).
727  */
728 static void
729 fsk_detach(struct dn_fsk *fs, int flags)
730 {
731 	if (flags & DN_DELETE_FS)
732 		flags |= DN_DESTROY;
733 	ND("fs %d from sched %d flags %s %s %s",
734 		fs->fs.fs_nr, fs->fs.sched_nr,
735 		(flags & DN_DELETE_FS) ? "DEL_FS":"",
736 		(flags & DN_DESTROY) ? "DEL":"",
737 		(flags & DN_DETACH) ? "DET":"");
738 	if (flags & DN_DETACH) { /* detach from the list */
739 		struct dn_fsk_head *h;
740 		h = fs->sched ? &fs->sched->fsk_list : &dn_cfg.fsu;
741 		SLIST_REMOVE(h, fs, dn_fsk, sch_chain);
742 	}
743 	/* Free the RED parameters, they will be recomputed on
744 	 * subsequent attach if needed.
745 	 */
746 	if (fs->w_q_lookup)
747 		free(fs->w_q_lookup, M_DUMMYNET);
748 	fs->w_q_lookup = NULL;
749 	qht_delete(fs, flags);
750 #ifdef NEW_AQM
751 	aqm_cleanup_deconfig_fs(fs);
752 #endif
753 
754 	if (fs->sched && fs->sched->fp->free_fsk)
755 		fs->sched->fp->free_fsk(fs);
756 	fs->sched = NULL;
757 	if (flags & DN_DELETE_FS) {
758 		bzero(fs, sizeof(*fs));	/* safety */
759 		free(fs, M_DUMMYNET);
760 		dn_cfg.fsk_count--;
761 	} else {
762 		SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
763 	}
764 }
765 
766 /*
767  * Detach or destroy all flowsets in a list.
768  * flags specifies what to do:
769  * DN_DESTROY:	flush all queues
770  * DN_DELETE_FS:	DN_DESTROY + destroy flowset
771  *	DN_DELETE_FS implies DN_DESTROY
772  */
773 static void
774 fsk_detach_list(struct dn_fsk_head *h, int flags)
775 {
776 	struct dn_fsk *fs;
777 	int n = 0; /* only for stats */
778 
779 	ND("head %p flags %x", h, flags);
780 	while ((fs = SLIST_FIRST(h))) {
781 		SLIST_REMOVE_HEAD(h, sch_chain);
782 		n++;
783 		fsk_detach(fs, flags);
784 	}
785 	ND("done %d flowsets", n);
786 }
787 
788 /*
789  * called on 'queue X delete' -- removes the flowset from fshash,
790  * deletes all queues for the flowset, and removes the flowset.
791  */
792 static int
793 delete_fs(int i, int locked)
794 {
795 	struct dn_fsk *fs;
796 	int err = 0;
797 
798 	if (!locked)
799 		DN_BH_WLOCK();
800 	fs = dn_ht_find(dn_cfg.fshash, i, DNHT_REMOVE, NULL);
801 	ND("fs %d found %p", i, fs);
802 	if (fs) {
803 		fsk_detach(fs, DN_DETACH | DN_DELETE_FS);
804 		err = 0;
805 	} else
806 		err = EINVAL;
807 	if (!locked)
808 		DN_BH_WUNLOCK();
809 	return err;
810 }
811 
812 /*----- end of flowset hashtable support -------------*/
813 
814 /*------------------------------------------------------------
815  * Scheduler hash. When searching by index we pass sched_nr,
816  * otherwise we pass struct dn_sch * which is the first field in
817  * struct dn_schk so we can cast between the two. We use this trick
818  * because in the create phase (but it should be fixed).
819  */
820 static uint32_t
821 schk_hash(uintptr_t key, int flags, void *_arg)
822 {
823 	uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
824 		((struct dn_schk *)key)->sch.sched_nr;
825 	return ( (i>>8)^(i>>4)^i );
826 }
827 
828 static int
829 schk_match(void *obj, uintptr_t key, int flags, void *_arg)
830 {
831 	struct dn_schk *s = (struct dn_schk *)obj;
832 	int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
833 		((struct dn_schk *)key)->sch.sched_nr;
834 	return (s->sch.sched_nr == i);
835 }
836 
837 /*
838  * Create the entry and intialize with the sched hash if needed.
839  * Leave s->fp unset so we can tell whether a dn_ht_find() returns
840  * a new object or a previously existing one.
841  */
842 static void *
843 schk_new(uintptr_t key, int flags, void *arg)
844 {
845 	struct schk_new_arg *a = arg;
846 	struct dn_schk *s;
847 	int l = sizeof(*s) +a->fp->schk_datalen;
848 
849 	s = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
850 	if (s == NULL)
851 		return NULL;
852 	set_oid(&s->link.oid, DN_LINK, sizeof(s->link));
853 	s->sch = *a->sch; // copy initial values
854 	s->link.link_nr = s->sch.sched_nr;
855 	SLIST_INIT(&s->fsk_list);
856 	/* initialize the hash table or create the single instance */
857 	s->fp = a->fp;	/* si_new needs this */
858 	s->drain_bucket = 0;
859 	if (s->sch.flags & DN_HAVE_MASK) {
860 		s->siht = dn_ht_init(NULL, s->sch.buckets,
861 			offsetof(struct dn_sch_inst, si_next),
862 			si_hash, si_match, si_new);
863 		if (s->siht == NULL) {
864 			free(s, M_DUMMYNET);
865 			return NULL;
866 		}
867 	}
868 	s->fp = NULL;	/* mark as a new scheduler */
869 	dn_cfg.schk_count++;
870 	return s;
871 }
872 
873 /*
874  * Callback for sched delete. Notify all attached flowsets to
875  * detach from the scheduler, destroy the internal flowset, and
876  * all instances. The scheduler goes away too.
877  * arg is 0 (only detach flowsets and destroy instances)
878  * DN_DESTROY (detach & delete queues, delete schk)
879  * or DN_DELETE_FS (delete queues and flowsets, delete schk)
880  */
881 static int
882 schk_delete_cb(void *obj, void *arg)
883 {
884 	struct dn_schk *s = obj;
885 #if 0
886 	int a = (int)arg;
887 	ND("sched %d arg %s%s",
888 		s->sch.sched_nr,
889 		a&DN_DESTROY ? "DEL ":"",
890 		a&DN_DELETE_FS ? "DEL_FS":"");
891 #endif
892 	fsk_detach_list(&s->fsk_list, arg ? DN_DESTROY : 0);
893 	/* no more flowset pointing to us now */
894 	if (s->sch.flags & DN_HAVE_MASK) {
895 		dn_ht_scan(s->siht, si_destroy, NULL);
896 		dn_ht_free(s->siht, 0);
897 	} else if (s->siht)
898 		si_destroy(s->siht, NULL);
899 	if (s->profile) {
900 		free(s->profile, M_DUMMYNET);
901 		s->profile = NULL;
902 	}
903 	s->siht = NULL;
904 	if (s->fp->destroy)
905 		s->fp->destroy(s);
906 	bzero(s, sizeof(*s));	// safety
907 	free(obj, M_DUMMYNET);
908 	dn_cfg.schk_count--;
909 	return DNHT_SCAN_DEL;
910 }
911 
912 /*
913  * called on a 'sched X delete' command. Deletes a single scheduler.
914  * This is done by removing from the schedhash, unlinking all
915  * flowsets and deleting their traffic.
916  */
917 static int
918 delete_schk(int i)
919 {
920 	struct dn_schk *s;
921 
922 	s = dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
923 	ND("%d %p", i, s);
924 	if (!s)
925 		return EINVAL;
926 	delete_fs(i + DN_MAX_ID, 1); /* first delete internal fs */
927 	/* then detach flowsets, delete traffic */
928 	schk_delete_cb(s, (void*)(uintptr_t)DN_DESTROY);
929 	return 0;
930 }
931 /*--- end of schk hashtable support ---*/
932 
933 static int
934 copy_obj(char **start, char *end, void *_o, const char *msg, int i)
935 {
936 	struct dn_id o;
937 	union {
938 		struct dn_link l;
939 		struct dn_schk s;
940 	} dn;
941 	int have = end - *start;
942 
943 	memcpy(&o, _o, sizeof(o));
944 	if (have < o.len || o.len == 0 || o.type == 0) {
945 		D("(WARN) type %d %s %d have %d need %d",
946 		    o.type, msg, i, have, o.len);
947 		return 1;
948 	}
949 	ND("type %d %s %d len %d", o.type, msg, i, o.len);
950 	if (o.type == DN_LINK) {
951 		memcpy(&dn.l, _o, sizeof(dn.l));
952 		/* Adjust burst parameter for link */
953 		dn.l.burst = div64(dn.l.burst, 8 * hz);
954 		dn.l.delay = dn.l.delay * 1000 / hz;
955 		memcpy(*start, &dn.l, sizeof(dn.l));
956 	} else if (o.type == DN_SCH) {
957 		/* Set dn.s.sch.oid.id to the number of instances */
958 		memcpy(&dn.s, _o, sizeof(dn.s));
959 		dn.s.sch.oid.id = (dn.s.sch.flags & DN_HAVE_MASK) ?
960 		    dn_ht_entries(dn.s.siht) : (dn.s.siht ? 1 : 0);
961 		memcpy(*start, &dn.s, sizeof(dn.s));
962 	} else
963 		memcpy(*start, _o, o.len);
964 	*start += o.len;
965 	return 0;
966 }
967 
968 /* Specific function to copy a queue.
969  * Copies only the user-visible part of a queue (which is in
970  * a struct dn_flow), and sets len accordingly.
971  */
972 static int
973 copy_obj_q(char **start, char *end, void *_o, const char *msg, int i)
974 {
975 	struct dn_id *o = _o;
976 	int have = end - *start;
977 	int len = sizeof(struct dn_flow); /* see above comment */
978 
979 	if (have < len || o->len == 0 || o->type != DN_QUEUE) {
980 		D("ERROR type %d %s %d have %d need %d",
981 			o->type, msg, i, have, len);
982 		return 1;
983 	}
984 	ND("type %d %s %d len %d", o->type, msg, i, len);
985 	memcpy(*start, _o, len);
986 	((struct dn_id*)(*start))->len = len;
987 	*start += len;
988 	return 0;
989 }
990 
991 static int
992 copy_q_cb(void *obj, void *arg)
993 {
994 	struct dn_queue *q = obj;
995 	struct copy_args *a = arg;
996 	struct dn_flow *ni = (struct dn_flow *)(*a->start);
997         if (copy_obj_q(a->start, a->end, &q->ni, "queue", -1))
998                 return DNHT_SCAN_END;
999         ni->oid.type = DN_FLOW; /* override the DN_QUEUE */
1000         ni->oid.id = si_hash((uintptr_t)&ni->fid, 0, NULL);
1001         return 0;
1002 }
1003 
1004 static int
1005 copy_q(struct copy_args *a, struct dn_fsk *fs, int flags)
1006 {
1007 	if (!fs->qht)
1008 		return 0;
1009 	if (fs->fs.flags & DN_QHT_HASH)
1010 		dn_ht_scan(fs->qht, copy_q_cb, a);
1011 	else
1012 		copy_q_cb(fs->qht, a);
1013 	return 0;
1014 }
1015 
1016 /*
1017  * This routine only copies the initial part of a profile ? XXX
1018  */
1019 static int
1020 copy_profile(struct copy_args *a, struct dn_profile *p)
1021 {
1022 	int have = a->end - *a->start;
1023 	/* XXX here we check for max length */
1024 	int profile_len = sizeof(struct dn_profile) -
1025 		ED_MAX_SAMPLES_NO*sizeof(int);
1026 
1027 	if (p == NULL)
1028 		return 0;
1029 	if (have < profile_len) {
1030 		D("error have %d need %d", have, profile_len);
1031 		return 1;
1032 	}
1033 	memcpy(*a->start, p, profile_len);
1034 	((struct dn_id *)(*a->start))->len = profile_len;
1035 	*a->start += profile_len;
1036 	return 0;
1037 }
1038 
1039 static int
1040 copy_flowset(struct copy_args *a, struct dn_fsk *fs, int flags)
1041 {
1042 	struct dn_fs *ufs = (struct dn_fs *)(*a->start);
1043 	if (!fs)
1044 		return 0;
1045 	ND("flowset %d", fs->fs.fs_nr);
1046 	if (copy_obj(a->start, a->end, &fs->fs, "flowset", fs->fs.fs_nr))
1047 		return DNHT_SCAN_END;
1048 	ufs->oid.id = (fs->fs.flags & DN_QHT_HASH) ?
1049 		dn_ht_entries(fs->qht) : (fs->qht ? 1 : 0);
1050 	if (flags) {	/* copy queues */
1051 		copy_q(a, fs, 0);
1052 	}
1053 	return 0;
1054 }
1055 
1056 static int
1057 copy_si_cb(void *obj, void *arg)
1058 {
1059 	struct dn_sch_inst *si = obj;
1060 	struct copy_args *a = arg;
1061 	struct dn_flow *ni = (struct dn_flow *)(*a->start);
1062 	if (copy_obj(a->start, a->end, &si->ni, "inst",
1063 			si->sched->sch.sched_nr))
1064 		return DNHT_SCAN_END;
1065 	ni->oid.type = DN_FLOW; /* override the DN_SCH_I */
1066 	ni->oid.id = si_hash((uintptr_t)si, DNHT_KEY_IS_OBJ, NULL);
1067 	return 0;
1068 }
1069 
1070 static int
1071 copy_si(struct copy_args *a, struct dn_schk *s, int flags)
1072 {
1073 	if (s->sch.flags & DN_HAVE_MASK)
1074 		dn_ht_scan(s->siht, copy_si_cb, a);
1075 	else if (s->siht)
1076 		copy_si_cb(s->siht, a);
1077 	return 0;
1078 }
1079 
1080 /*
1081  * compute a list of children of a scheduler and copy up
1082  */
1083 static int
1084 copy_fsk_list(struct copy_args *a, struct dn_schk *s, int flags)
1085 {
1086 	struct dn_fsk *fs;
1087 	struct dn_id *o;
1088 	uint32_t *p;
1089 
1090 	int n = 0, space = sizeof(*o);
1091 	SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1092 		if (fs->fs.fs_nr < DN_MAX_ID)
1093 			n++;
1094 	}
1095 	space += n * sizeof(uint32_t);
1096 	DX(3, "sched %d has %d flowsets", s->sch.sched_nr, n);
1097 	if (a->end - *(a->start) < space)
1098 		return DNHT_SCAN_END;
1099 	o = (struct dn_id *)(*(a->start));
1100 	o->len = space;
1101 	*a->start += o->len;
1102 	o->type = DN_TEXT;
1103 	p = (uint32_t *)(o+1);
1104 	SLIST_FOREACH(fs, &s->fsk_list, sch_chain)
1105 		if (fs->fs.fs_nr < DN_MAX_ID)
1106 			*p++ = fs->fs.fs_nr;
1107 	return 0;
1108 }
1109 
1110 static int
1111 copy_data_helper(void *_o, void *_arg)
1112 {
1113 	struct copy_args *a = _arg;
1114 	uint32_t *r = a->extra->r; /* start of first range */
1115 	uint32_t *lim;	/* first invalid pointer */
1116 	int n;
1117 
1118 	lim = (uint32_t *)((char *)(a->extra) + a->extra->o.len);
1119 
1120 	if (a->type == DN_LINK || a->type == DN_SCH) {
1121 		/* pipe|sched show, we receive a dn_schk */
1122 		struct dn_schk *s = _o;
1123 
1124 		n = s->sch.sched_nr;
1125 		if (a->type == DN_SCH && n >= DN_MAX_ID)
1126 			return 0;	/* not a scheduler */
1127 		if (a->type == DN_LINK && n <= DN_MAX_ID)
1128 		    return 0;	/* not a pipe */
1129 
1130 		/* see if the object is within one of our ranges */
1131 		for (;r < lim; r += 2) {
1132 			if (n < r[0] || n > r[1])
1133 				continue;
1134 			/* Found a valid entry, copy and we are done */
1135 			if (a->flags & DN_C_LINK) {
1136 				if (copy_obj(a->start, a->end,
1137 				    &s->link, "link", n))
1138 					return DNHT_SCAN_END;
1139 				if (copy_profile(a, s->profile))
1140 					return DNHT_SCAN_END;
1141 				if (copy_flowset(a, s->fs, 0))
1142 					return DNHT_SCAN_END;
1143 			}
1144 			if (a->flags & DN_C_SCH) {
1145 				if (copy_obj(a->start, a->end,
1146 				    &s->sch, "sched", n))
1147 					return DNHT_SCAN_END;
1148 				/* list all attached flowsets */
1149 				if (copy_fsk_list(a, s, 0))
1150 					return DNHT_SCAN_END;
1151 			}
1152 			if (a->flags & DN_C_FLOW)
1153 				copy_si(a, s, 0);
1154 			break;
1155 		}
1156 	} else if (a->type == DN_FS) {
1157 		/* queue show, skip internal flowsets */
1158 		struct dn_fsk *fs = _o;
1159 
1160 		n = fs->fs.fs_nr;
1161 		if (n >= DN_MAX_ID)
1162 			return 0;
1163 		/* see if the object is within one of our ranges */
1164 		for (;r < lim; r += 2) {
1165 			if (n < r[0] || n > r[1])
1166 				continue;
1167 			if (copy_flowset(a, fs, 0))
1168 				return DNHT_SCAN_END;
1169 			copy_q(a, fs, 0);
1170 			break; /* we are done */
1171 		}
1172 	}
1173 	return 0;
1174 }
1175 
1176 static inline struct dn_schk *
1177 locate_scheduler(int i)
1178 {
1179 	return dn_ht_find(dn_cfg.schedhash, i, 0, NULL);
1180 }
1181 
1182 /*
1183  * red parameters are in fixed point arithmetic.
1184  */
1185 static int
1186 config_red(struct dn_fsk *fs)
1187 {
1188 	int64_t s, idle, weight, w0;
1189 	int t, i;
1190 
1191 	fs->w_q = fs->fs.w_q;
1192 	fs->max_p = fs->fs.max_p;
1193 	ND("called");
1194 	/* Doing stuff that was in userland */
1195 	i = fs->sched->link.bandwidth;
1196 	s = (i <= 0) ? 0 :
1197 		hz * dn_cfg.red_avg_pkt_size * 8 * SCALE(1) / i;
1198 
1199 	idle = div64((s * 3) , fs->w_q); /* s, fs->w_q scaled; idle not scaled */
1200 	fs->lookup_step = div64(idle , dn_cfg.red_lookup_depth);
1201 	/* fs->lookup_step not scaled, */
1202 	if (!fs->lookup_step)
1203 		fs->lookup_step = 1;
1204 	w0 = weight = SCALE(1) - fs->w_q; //fs->w_q scaled
1205 
1206 	for (t = fs->lookup_step; t > 1; --t)
1207 		weight = SCALE_MUL(weight, w0);
1208 	fs->lookup_weight = (int)(weight); // scaled
1209 
1210 	/* Now doing stuff that was in kerneland */
1211 	fs->min_th = SCALE(fs->fs.min_th);
1212 	fs->max_th = SCALE(fs->fs.max_th);
1213 
1214 	if (fs->fs.max_th == fs->fs.min_th)
1215 		fs->c_1 = fs->max_p;
1216 	else
1217 		fs->c_1 = SCALE((int64_t)(fs->max_p)) / (fs->fs.max_th - fs->fs.min_th);
1218 	fs->c_2 = SCALE_MUL(fs->c_1, SCALE(fs->fs.min_th));
1219 
1220 	if (fs->fs.flags & DN_IS_GENTLE_RED) {
1221 		fs->c_3 = (SCALE(1) - fs->max_p) / fs->fs.max_th;
1222 		fs->c_4 = SCALE(1) - 2 * fs->max_p;
1223 	}
1224 
1225 	/* If the lookup table already exist, free and create it again. */
1226 	if (fs->w_q_lookup) {
1227 		free(fs->w_q_lookup, M_DUMMYNET);
1228 		fs->w_q_lookup = NULL;
1229 	}
1230 	if (dn_cfg.red_lookup_depth == 0) {
1231 		printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1232 		    "must be > 0\n");
1233 		fs->fs.flags &= ~DN_IS_RED;
1234 		fs->fs.flags &= ~DN_IS_GENTLE_RED;
1235 		return (EINVAL);
1236 	}
1237 	fs->lookup_depth = dn_cfg.red_lookup_depth;
1238 	fs->w_q_lookup = (u_int *)malloc(fs->lookup_depth * sizeof(int),
1239 	    M_DUMMYNET, M_NOWAIT);
1240 	if (fs->w_q_lookup == NULL) {
1241 		printf("dummynet: sorry, cannot allocate red lookup table\n");
1242 		fs->fs.flags &= ~DN_IS_RED;
1243 		fs->fs.flags &= ~DN_IS_GENTLE_RED;
1244 		return(ENOSPC);
1245 	}
1246 
1247 	/* Fill the lookup table with (1 - w_q)^x */
1248 	fs->w_q_lookup[0] = SCALE(1) - fs->w_q;
1249 
1250 	for (i = 1; i < fs->lookup_depth; i++)
1251 		fs->w_q_lookup[i] =
1252 		    SCALE_MUL(fs->w_q_lookup[i - 1], fs->lookup_weight);
1253 
1254 	if (dn_cfg.red_avg_pkt_size < 1)
1255 		dn_cfg.red_avg_pkt_size = 512;
1256 	fs->avg_pkt_size = dn_cfg.red_avg_pkt_size;
1257 	if (dn_cfg.red_max_pkt_size < 1)
1258 		dn_cfg.red_max_pkt_size = 1500;
1259 	fs->max_pkt_size = dn_cfg.red_max_pkt_size;
1260 	ND("exit");
1261 	return 0;
1262 }
1263 
1264 /* Scan all flowset attached to this scheduler and update red */
1265 static void
1266 update_red(struct dn_schk *s)
1267 {
1268 	struct dn_fsk *fs;
1269 	SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1270 		if (fs && (fs->fs.flags & DN_IS_RED))
1271 			config_red(fs);
1272 	}
1273 }
1274 
1275 /* attach flowset to scheduler s, possibly requeue */
1276 static void
1277 fsk_attach(struct dn_fsk *fs, struct dn_schk *s)
1278 {
1279 	ND("remove fs %d from fsunlinked, link to sched %d",
1280 		fs->fs.fs_nr, s->sch.sched_nr);
1281 	SLIST_REMOVE(&dn_cfg.fsu, fs, dn_fsk, sch_chain);
1282 	fs->sched = s;
1283 	SLIST_INSERT_HEAD(&s->fsk_list, fs, sch_chain);
1284 	if (s->fp->new_fsk)
1285 		s->fp->new_fsk(fs);
1286 	/* XXX compute fsk_mask */
1287 	fs->fsk_mask = fs->fs.flow_mask;
1288 	if (fs->sched->sch.flags & DN_HAVE_MASK)
1289 		flow_id_or(&fs->sched->sch.sched_mask, &fs->fsk_mask);
1290 	if (fs->qht) {
1291 		/*
1292 		 * we must drain qht according to the old
1293 		 * type, and reinsert according to the new one.
1294 		 * The requeue is complex -- in general we need to
1295 		 * reclassify every single packet.
1296 		 * For the time being, let's hope qht is never set
1297 		 * when we reach this point.
1298 		 */
1299 		D("XXX TODO requeue from fs %d to sch %d",
1300 			fs->fs.fs_nr, s->sch.sched_nr);
1301 		fs->qht = NULL;
1302 	}
1303 	/* set the new type for qht */
1304 	if (nonzero_mask(&fs->fsk_mask))
1305 		fs->fs.flags |= DN_QHT_HASH;
1306 	else
1307 		fs->fs.flags &= ~DN_QHT_HASH;
1308 
1309 	/* XXX config_red() can fail... */
1310 	if (fs->fs.flags & DN_IS_RED)
1311 		config_red(fs);
1312 }
1313 
1314 /* update all flowsets which may refer to this scheduler */
1315 static void
1316 update_fs(struct dn_schk *s)
1317 {
1318 	struct dn_fsk *fs, *tmp;
1319 
1320 	SLIST_FOREACH_SAFE(fs, &dn_cfg.fsu, sch_chain, tmp) {
1321 		if (s->sch.sched_nr != fs->fs.sched_nr) {
1322 			D("fs %d for sch %d not %d still unlinked",
1323 				fs->fs.fs_nr, fs->fs.sched_nr,
1324 				s->sch.sched_nr);
1325 			continue;
1326 		}
1327 		fsk_attach(fs, s);
1328 	}
1329 }
1330 
1331 #ifdef NEW_AQM
1332 /* Retrieve AQM configurations to ipfw userland
1333  */
1334 static int
1335 get_aqm_parms(struct sockopt *sopt)
1336 {
1337 	struct dn_extra_parms  *ep;
1338 	struct dn_fsk *fs;
1339 	size_t sopt_valsize;
1340 	int l, err = 0;
1341 
1342 	sopt_valsize = sopt->sopt_valsize;
1343 	l = sizeof(*ep);
1344 	if (sopt->sopt_valsize < l) {
1345 		D("bad len sopt->sopt_valsize %d len %d",
1346 			(int) sopt->sopt_valsize , l);
1347 		err = EINVAL;
1348 		return err;
1349 	}
1350 	ep = malloc(l, M_DUMMYNET, M_WAITOK);
1351 	if(!ep) {
1352 		err = ENOMEM ;
1353 		return err;
1354 	}
1355 	do {
1356 		err = sooptcopyin(sopt, ep, l, l);
1357 		if(err)
1358 			break;
1359 		sopt->sopt_valsize = sopt_valsize;
1360 		if (ep->oid.len < l) {
1361 			err = EINVAL;
1362 			break;
1363 		}
1364 
1365 		fs = dn_ht_find(dn_cfg.fshash, ep->nr, 0, NULL);
1366 		if (!fs) {
1367 			D("fs %d not found", ep->nr);
1368 			err = EINVAL;
1369 			break;
1370 		}
1371 
1372 		if (fs->aqmfp && fs->aqmfp->getconfig) {
1373 			if(fs->aqmfp->getconfig(fs, ep)) {
1374 				D("Error while trying to get AQM params");
1375 				err = EINVAL;
1376 				break;
1377 			}
1378 			ep->oid.len = l;
1379 			err = sooptcopyout(sopt, ep, l);
1380 		}
1381 	}while(0);
1382 
1383 	free(ep, M_DUMMYNET);
1384 	return err;
1385 }
1386 
1387 /* Retrieve AQM configurations to ipfw userland
1388  */
1389 static int
1390 get_sched_parms(struct sockopt *sopt)
1391 {
1392 	struct dn_extra_parms  *ep;
1393 	struct dn_schk *schk;
1394 	size_t sopt_valsize;
1395 	int l, err = 0;
1396 
1397 	sopt_valsize = sopt->sopt_valsize;
1398 	l = sizeof(*ep);
1399 	if (sopt->sopt_valsize < l) {
1400 		D("bad len sopt->sopt_valsize %d len %d",
1401 			(int) sopt->sopt_valsize , l);
1402 		err = EINVAL;
1403 		return err;
1404 	}
1405 	ep = malloc(l, M_DUMMYNET, M_WAITOK);
1406 	if(!ep) {
1407 		err = ENOMEM ;
1408 		return err;
1409 	}
1410 	do {
1411 		err = sooptcopyin(sopt, ep, l, l);
1412 		if(err)
1413 			break;
1414 		sopt->sopt_valsize = sopt_valsize;
1415 		if (ep->oid.len < l) {
1416 			err = EINVAL;
1417 			break;
1418 		}
1419 
1420 		schk = locate_scheduler(ep->nr);
1421 		if (!schk) {
1422 			D("sched %d not found", ep->nr);
1423 			err = EINVAL;
1424 			break;
1425 		}
1426 
1427 		if (schk->fp && schk->fp->getconfig) {
1428 			if(schk->fp->getconfig(schk, ep)) {
1429 				D("Error while trying to get sched params");
1430 				err = EINVAL;
1431 				break;
1432 			}
1433 			ep->oid.len = l;
1434 			err = sooptcopyout(sopt, ep, l);
1435 		}
1436 	}while(0);
1437 	free(ep, M_DUMMYNET);
1438 
1439 	return err;
1440 }
1441 
1442 /* Configure AQM for flowset 'fs'.
1443  * extra parameters are passed from userland.
1444  */
1445 static int
1446 config_aqm(struct dn_fsk *fs, struct  dn_extra_parms *ep, int busy)
1447 {
1448 	int err = 0;
1449 
1450 	do {
1451 		/* no configurations */
1452 		if (!ep) {
1453 			err = 0;
1454 			break;
1455 		}
1456 
1457 		/* no AQM for this flowset*/
1458 		if (!strcmp(ep->name,"")) {
1459 			err = 0;
1460 			break;
1461 		}
1462 		if (ep->oid.len < sizeof(*ep)) {
1463 			D("short aqm len %d", ep->oid.len);
1464 				err = EINVAL;
1465 				break;
1466 		}
1467 
1468 		if (busy) {
1469 			D("Unable to configure flowset, flowset busy!");
1470 			err = EINVAL;
1471 			break;
1472 		}
1473 
1474 		/* deconfigure old aqm if exist */
1475 		if (fs->aqmcfg && fs->aqmfp && fs->aqmfp->deconfig) {
1476 			aqm_cleanup_deconfig_fs(fs);
1477 		}
1478 
1479 		if (!(fs->aqmfp = find_aqm_type(0, ep->name))) {
1480 			D("AQM functions not found for type %s!", ep->name);
1481 			fs->fs.flags &= ~DN_IS_AQM;
1482 			err = EINVAL;
1483 			break;
1484 		} else
1485 			fs->fs.flags |= DN_IS_AQM;
1486 
1487 		if (ep->oid.subtype != DN_AQM_PARAMS) {
1488 				D("Wrong subtype");
1489 				err = EINVAL;
1490 				break;
1491 		}
1492 
1493 		if (fs->aqmfp->config) {
1494 			err = fs->aqmfp->config(fs, ep, ep->oid.len);
1495 			if (err) {
1496 					D("Unable to configure AQM for FS %d", fs->fs.fs_nr );
1497 					fs->fs.flags &= ~DN_IS_AQM;
1498 					fs->aqmfp = NULL;
1499 					break;
1500 			}
1501 		}
1502 	} while(0);
1503 
1504 	return err;
1505 }
1506 #endif
1507 
1508 /*
1509  * Configuration -- to preserve backward compatibility we use
1510  * the following scheme (N is 65536)
1511  *	NUMBER		SCHED	LINK	FLOWSET
1512  *	   1 ..  N-1	(1)WFQ	(2)WFQ	(3)queue
1513  *	 N+1 .. 2N-1	(4)FIFO (5)FIFO	(6)FIFO for sched 1..N-1
1514  *	2N+1 .. 3N-1	--	--	(7)FIFO for sched N+1..2N-1
1515  *
1516  * "pipe i config" configures #1, #2 and #3
1517  * "sched i config" configures #1 and possibly #6
1518  * "queue i config" configures #3
1519  * #1 is configured with 'pipe i config' or 'sched i config'
1520  * #2 is configured with 'pipe i config', and created if not
1521  *	existing with 'sched i config'
1522  * #3 is configured with 'queue i config'
1523  * #4 is automatically configured after #1, can only be FIFO
1524  * #5 is automatically configured after #2
1525  * #6 is automatically created when #1 is !MULTIQUEUE,
1526  *	and can be updated.
1527  * #7 is automatically configured after #2
1528  */
1529 
1530 /*
1531  * configure a link (and its FIFO instance)
1532  */
1533 static int
1534 config_link(struct dn_link *p, struct dn_id *arg)
1535 {
1536 	int i;
1537 
1538 	if (p->oid.len != sizeof(*p)) {
1539 		D("invalid pipe len %d", p->oid.len);
1540 		return EINVAL;
1541 	}
1542 	i = p->link_nr;
1543 	if (i <= 0 || i >= DN_MAX_ID)
1544 		return EINVAL;
1545 	/*
1546 	 * The config program passes parameters as follows:
1547 	 * bw = bits/second (0 means no limits),
1548 	 * delay = ms, must be translated into ticks.
1549 	 * qsize = slots/bytes
1550 	 * burst ???
1551 	 */
1552 	p->delay = (p->delay * hz) / 1000;
1553 	/* Scale burst size: bytes -> bits * hz */
1554 	p->burst *= 8 * hz;
1555 
1556 	DN_BH_WLOCK();
1557 	/* do it twice, base link and FIFO link */
1558 	for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1559 	    struct dn_schk *s = locate_scheduler(i);
1560 	    if (s == NULL) {
1561 		DN_BH_WUNLOCK();
1562 		D("sched %d not found", i);
1563 		return EINVAL;
1564 	    }
1565 	    /* remove profile if exists */
1566 	    if (s->profile) {
1567 		free(s->profile, M_DUMMYNET);
1568 		s->profile = NULL;
1569 	    }
1570 	    /* copy all parameters */
1571 	    s->link.oid = p->oid;
1572 	    s->link.link_nr = i;
1573 	    s->link.delay = p->delay;
1574 	    if (s->link.bandwidth != p->bandwidth) {
1575 		/* XXX bandwidth changes, need to update red params */
1576 	    s->link.bandwidth = p->bandwidth;
1577 		update_red(s);
1578 	    }
1579 	    s->link.burst = p->burst;
1580 	    schk_reset_credit(s);
1581 	}
1582 	dn_cfg.id++;
1583 	DN_BH_WUNLOCK();
1584 	return 0;
1585 }
1586 
1587 /*
1588  * configure a flowset. Can be called from inside with locked=1,
1589  */
1590 static struct dn_fsk *
1591 config_fs(struct dn_fs *nfs, struct dn_id *arg, int locked)
1592 {
1593 	int i;
1594 	struct dn_fsk *fs;
1595 #ifdef NEW_AQM
1596 	struct dn_extra_parms *ep;
1597 #endif
1598 
1599 	if (nfs->oid.len != sizeof(*nfs)) {
1600 		D("invalid flowset len %d", nfs->oid.len);
1601 		return NULL;
1602 	}
1603 	i = nfs->fs_nr;
1604 	if (i <= 0 || i >= 3*DN_MAX_ID)
1605 		return NULL;
1606 #ifdef NEW_AQM
1607 	ep = NULL;
1608 	if (arg != NULL) {
1609 		ep = malloc(sizeof(*ep), M_TEMP, locked ? M_NOWAIT : M_WAITOK);
1610 		if (ep == NULL)
1611 			return (NULL);
1612 		memcpy(ep, arg, sizeof(*ep));
1613 	}
1614 #endif
1615 	ND("flowset %d", i);
1616 	/* XXX other sanity checks */
1617         if (nfs->flags & DN_QSIZE_BYTES) {
1618 		ipdn_bound_var(&nfs->qsize, 16384,
1619 		    1500, dn_cfg.byte_limit, NULL); // "queue byte size");
1620         } else {
1621 		ipdn_bound_var(&nfs->qsize, 50,
1622 		    1, dn_cfg.slot_limit, NULL); // "queue slot size");
1623         }
1624 	if (nfs->flags & DN_HAVE_MASK) {
1625 		/* make sure we have some buckets */
1626 		ipdn_bound_var((int *)&nfs->buckets, dn_cfg.hash_size,
1627 			1, dn_cfg.max_hash_size, "flowset buckets");
1628 	} else {
1629 		nfs->buckets = 1;	/* we only need 1 */
1630 	}
1631 	if (!locked)
1632 		DN_BH_WLOCK();
1633 	do { /* exit with break when done */
1634 	    struct dn_schk *s;
1635 	    int flags = nfs->sched_nr ? DNHT_INSERT : 0;
1636 	    int j;
1637 	    int oldc = dn_cfg.fsk_count;
1638 	    fs = dn_ht_find(dn_cfg.fshash, i, flags, NULL);
1639 	    if (fs == NULL) {
1640 		D("missing sched for flowset %d", i);
1641 	        break;
1642 	    }
1643 	    /* grab some defaults from the existing one */
1644 	    if (nfs->sched_nr == 0) /* reuse */
1645 		nfs->sched_nr = fs->fs.sched_nr;
1646 	    for (j = 0; j < sizeof(nfs->par)/sizeof(nfs->par[0]); j++) {
1647 		if (nfs->par[j] == -1) /* reuse */
1648 		    nfs->par[j] = fs->fs.par[j];
1649 	    }
1650 	    if (bcmp(&fs->fs, nfs, sizeof(*nfs)) == 0) {
1651 		ND("flowset %d unchanged", i);
1652 #ifdef NEW_AQM
1653 		if (ep != NULL) {
1654 			/*
1655 			 * Reconfigure AQM as the parameters can be changed.
1656 			 * We consider the flowset as busy if it has scheduler
1657 			 * instance(s).
1658 			 */
1659 			s = locate_scheduler(nfs->sched_nr);
1660 			config_aqm(fs, ep, s != NULL && s->siht != NULL);
1661 		}
1662 #endif
1663 		break; /* no change, nothing to do */
1664 	    }
1665 	    if (oldc != dn_cfg.fsk_count)	/* new item */
1666 		dn_cfg.id++;
1667 	    s = locate_scheduler(nfs->sched_nr);
1668 	    /* detach from old scheduler if needed, preserving
1669 	     * queues if we need to reattach. Then update the
1670 	     * configuration, and possibly attach to the new sched.
1671 	     */
1672 	    DX(2, "fs %d changed sched %d@%p to %d@%p",
1673 		fs->fs.fs_nr,
1674 		fs->fs.sched_nr, fs->sched, nfs->sched_nr, s);
1675 	    if (fs->sched) {
1676 		int flags = s ? DN_DETACH : (DN_DETACH | DN_DESTROY);
1677 		flags |= DN_DESTROY; /* XXX temporary */
1678 		fsk_detach(fs, flags);
1679 	    }
1680 	    fs->fs = *nfs; /* copy configuration */
1681 #ifdef NEW_AQM
1682 			fs->aqmfp = NULL;
1683 			if (ep != NULL)
1684 				config_aqm(fs, ep, s != NULL &&
1685 				    s->siht != NULL);
1686 #endif
1687 	    if (s != NULL)
1688 		fsk_attach(fs, s);
1689 	} while (0);
1690 	if (!locked)
1691 		DN_BH_WUNLOCK();
1692 #ifdef NEW_AQM
1693 	if (ep != NULL)
1694 		free(ep, M_TEMP);
1695 #endif
1696 	return fs;
1697 }
1698 
1699 /*
1700  * config/reconfig a scheduler and its FIFO variant.
1701  * For !MULTIQUEUE schedulers, also set up the flowset.
1702  *
1703  * On reconfigurations (detected because s->fp is set),
1704  * detach existing flowsets preserving traffic, preserve link,
1705  * and delete the old scheduler creating a new one.
1706  */
1707 static int
1708 config_sched(struct dn_sch *_nsch, struct dn_id *arg)
1709 {
1710 	struct dn_schk *s;
1711 	struct schk_new_arg a; /* argument for schk_new */
1712 	int i;
1713 	struct dn_link p;	/* copy of oldlink */
1714 	struct dn_profile *pf = NULL;	/* copy of old link profile */
1715 	/* Used to preserv mask parameter */
1716 	struct ipfw_flow_id new_mask;
1717 	int new_buckets = 0;
1718 	int new_flags = 0;
1719 	int pipe_cmd;
1720 	int err = ENOMEM;
1721 
1722 	a.sch = _nsch;
1723 	if (a.sch->oid.len != sizeof(*a.sch)) {
1724 		D("bad sched len %d", a.sch->oid.len);
1725 		return EINVAL;
1726 	}
1727 	i = a.sch->sched_nr;
1728 	if (i <= 0 || i >= DN_MAX_ID)
1729 		return EINVAL;
1730 	/* make sure we have some buckets */
1731 	if (a.sch->flags & DN_HAVE_MASK)
1732 		ipdn_bound_var((int *)&a.sch->buckets, dn_cfg.hash_size,
1733 			1, dn_cfg.max_hash_size, "sched buckets");
1734 	/* XXX other sanity checks */
1735 	bzero(&p, sizeof(p));
1736 
1737 	pipe_cmd = a.sch->flags & DN_PIPE_CMD;
1738 	a.sch->flags &= ~DN_PIPE_CMD; //XXX do it even if is not set?
1739 	if (pipe_cmd) {
1740 		/* Copy mask parameter */
1741 		new_mask = a.sch->sched_mask;
1742 		new_buckets = a.sch->buckets;
1743 		new_flags = a.sch->flags;
1744 	}
1745 	DN_BH_WLOCK();
1746 again: /* run twice, for wfq and fifo */
1747 	/*
1748 	 * lookup the type. If not supplied, use the previous one
1749 	 * or default to WF2Q+. Otherwise, return an error.
1750 	 */
1751 	dn_cfg.id++;
1752 	a.fp = find_sched_type(a.sch->oid.subtype, a.sch->name);
1753 	if (a.fp != NULL) {
1754 		/* found. Lookup or create entry */
1755 		s = dn_ht_find(dn_cfg.schedhash, i, DNHT_INSERT, &a);
1756 	} else if (a.sch->oid.subtype == 0 && !a.sch->name[0]) {
1757 		/* No type. search existing s* or retry with WF2Q+ */
1758 		s = dn_ht_find(dn_cfg.schedhash, i, 0, &a);
1759 		if (s != NULL) {
1760 			a.fp = s->fp;
1761 			/* Scheduler exists, skip to FIFO scheduler
1762 			 * if command was pipe config...
1763 			 */
1764 			if (pipe_cmd)
1765 				goto next;
1766 		} else {
1767 			/* New scheduler, create a wf2q+ with no mask
1768 			 * if command was pipe config...
1769 			 */
1770 			if (pipe_cmd) {
1771 				/* clear mask parameter */
1772 				bzero(&a.sch->sched_mask, sizeof(new_mask));
1773 				a.sch->buckets = 0;
1774 				a.sch->flags &= ~DN_HAVE_MASK;
1775 			}
1776 			a.sch->oid.subtype = DN_SCHED_WF2QP;
1777 			goto again;
1778 		}
1779 	} else {
1780 		D("invalid scheduler type %d %s",
1781 			a.sch->oid.subtype, a.sch->name);
1782 		err = EINVAL;
1783 		goto error;
1784 	}
1785 	/* normalize name and subtype */
1786 	a.sch->oid.subtype = a.fp->type;
1787 	bzero(a.sch->name, sizeof(a.sch->name));
1788 	strlcpy(a.sch->name, a.fp->name, sizeof(a.sch->name));
1789 	if (s == NULL) {
1790 		D("cannot allocate scheduler %d", i);
1791 		goto error;
1792 	}
1793 	/* restore existing link if any */
1794 	if (p.link_nr) {
1795 		s->link = p;
1796 		if (!pf || pf->link_nr != p.link_nr) { /* no saved value */
1797 			s->profile = NULL; /* XXX maybe not needed */
1798 		} else {
1799 			s->profile = malloc(sizeof(struct dn_profile),
1800 					     M_DUMMYNET, M_NOWAIT | M_ZERO);
1801 			if (s->profile == NULL) {
1802 				D("cannot allocate profile");
1803 				goto error; //XXX
1804 			}
1805 			memcpy(s->profile, pf, sizeof(*pf));
1806 		}
1807 	}
1808 	p.link_nr = 0;
1809 	if (s->fp == NULL) {
1810 		DX(2, "sched %d new type %s", i, a.fp->name);
1811 	} else if (s->fp != a.fp ||
1812 			bcmp(a.sch, &s->sch, sizeof(*a.sch)) ) {
1813 		/* already existing. */
1814 		DX(2, "sched %d type changed from %s to %s",
1815 			i, s->fp->name, a.fp->name);
1816 		DX(4, "   type/sub %d/%d -> %d/%d",
1817 			s->sch.oid.type, s->sch.oid.subtype,
1818 			a.sch->oid.type, a.sch->oid.subtype);
1819 		if (s->link.link_nr == 0)
1820 			D("XXX WARNING link 0 for sched %d", i);
1821 		p = s->link;	/* preserve link */
1822 		if (s->profile) {/* preserve profile */
1823 			if (!pf)
1824 				pf = malloc(sizeof(*pf),
1825 				    M_DUMMYNET, M_NOWAIT | M_ZERO);
1826 			if (pf)	/* XXX should issue a warning otherwise */
1827 				memcpy(pf, s->profile, sizeof(*pf));
1828 		}
1829 		/* remove from the hash */
1830 		dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
1831 		/* Detach flowsets, preserve queues. */
1832 		// schk_delete_cb(s, NULL);
1833 		// XXX temporarily, kill queues
1834 		schk_delete_cb(s, (void *)DN_DESTROY);
1835 		goto again;
1836 	} else {
1837 		DX(4, "sched %d unchanged type %s", i, a.fp->name);
1838 	}
1839 	/* complete initialization */
1840 	s->sch = *a.sch;
1841 	s->fp = a.fp;
1842 	s->cfg = arg;
1843 	// XXX schk_reset_credit(s);
1844 	/* create the internal flowset if needed,
1845 	 * trying to reuse existing ones if available
1846 	 */
1847 	if (!(s->fp->flags & DN_MULTIQUEUE) && !s->fs) {
1848 	        s->fs = dn_ht_find(dn_cfg.fshash, i, 0, NULL);
1849 		if (!s->fs) {
1850 			struct dn_fs fs;
1851 			bzero(&fs, sizeof(fs));
1852 			set_oid(&fs.oid, DN_FS, sizeof(fs));
1853 			fs.fs_nr = i + DN_MAX_ID;
1854 			fs.sched_nr = i;
1855 			s->fs = config_fs(&fs, NULL, 1 /* locked */);
1856 		}
1857 		if (!s->fs) {
1858 			schk_delete_cb(s, (void *)DN_DESTROY);
1859 			D("error creating internal fs for %d", i);
1860 			goto error;
1861 		}
1862 	}
1863 	/* call init function after the flowset is created */
1864 	if (s->fp->config)
1865 		s->fp->config(s);
1866 	update_fs(s);
1867 next:
1868 	if (i < DN_MAX_ID) { /* now configure the FIFO instance */
1869 		i += DN_MAX_ID;
1870 		if (pipe_cmd) {
1871 			/* Restore mask parameter for FIFO */
1872 			a.sch->sched_mask = new_mask;
1873 			a.sch->buckets = new_buckets;
1874 			a.sch->flags = new_flags;
1875 		} else {
1876 			/* sched config shouldn't modify the FIFO scheduler */
1877 			if (dn_ht_find(dn_cfg.schedhash, i, 0, &a) != NULL) {
1878 				/* FIFO already exist, don't touch it */
1879 				err = 0; /* and this is not an error */
1880 				goto error;
1881 			}
1882 		}
1883 		a.sch->sched_nr = i;
1884 		a.sch->oid.subtype = DN_SCHED_FIFO;
1885 		bzero(a.sch->name, sizeof(a.sch->name));
1886 		goto again;
1887 	}
1888 	err = 0;
1889 error:
1890 	DN_BH_WUNLOCK();
1891 	if (pf)
1892 		free(pf, M_DUMMYNET);
1893 	return err;
1894 }
1895 
1896 /*
1897  * attach a profile to a link
1898  */
1899 static int
1900 config_profile(struct dn_profile *pf, struct dn_id *arg)
1901 {
1902 	struct dn_schk *s;
1903 	int i, olen, err = 0;
1904 
1905 	if (pf->oid.len < sizeof(*pf)) {
1906 		D("short profile len %d", pf->oid.len);
1907 		return EINVAL;
1908 	}
1909 	i = pf->link_nr;
1910 	if (i <= 0 || i >= DN_MAX_ID)
1911 		return EINVAL;
1912 	/* XXX other sanity checks */
1913 	DN_BH_WLOCK();
1914 	for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1915 		s = locate_scheduler(i);
1916 
1917 		if (s == NULL) {
1918 			err = EINVAL;
1919 			break;
1920 		}
1921 		dn_cfg.id++;
1922 		/*
1923 		 * If we had a profile and the new one does not fit,
1924 		 * or it is deleted, then we need to free memory.
1925 		 */
1926 		if (s->profile && (pf->samples_no == 0 ||
1927 		    s->profile->oid.len < pf->oid.len)) {
1928 			free(s->profile, M_DUMMYNET);
1929 			s->profile = NULL;
1930 		}
1931 		if (pf->samples_no == 0)
1932 			continue;
1933 		/*
1934 		 * new profile, possibly allocate memory
1935 		 * and copy data.
1936 		 */
1937 		if (s->profile == NULL)
1938 			s->profile = malloc(pf->oid.len,
1939 				    M_DUMMYNET, M_NOWAIT | M_ZERO);
1940 		if (s->profile == NULL) {
1941 			D("no memory for profile %d", i);
1942 			err = ENOMEM;
1943 			break;
1944 		}
1945 		/* preserve larger length XXX double check */
1946 		olen = s->profile->oid.len;
1947 		if (olen < pf->oid.len)
1948 			olen = pf->oid.len;
1949 		memcpy(s->profile, pf, pf->oid.len);
1950 		s->profile->oid.len = olen;
1951 	}
1952 	DN_BH_WUNLOCK();
1953 	return err;
1954 }
1955 
1956 /*
1957  * Delete all objects:
1958  */
1959 static void
1960 dummynet_flush(void)
1961 {
1962 
1963 	/* delete all schedulers and related links/queues/flowsets */
1964 	dn_ht_scan(dn_cfg.schedhash, schk_delete_cb,
1965 		(void *)(uintptr_t)DN_DELETE_FS);
1966 	/* delete all remaining (unlinked) flowsets */
1967 	DX(4, "still %d unlinked fs", dn_cfg.fsk_count);
1968 	dn_ht_free(dn_cfg.fshash, DNHT_REMOVE);
1969 	fsk_detach_list(&dn_cfg.fsu, DN_DELETE_FS);
1970 	/* Reinitialize system heap... */
1971 	heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
1972 }
1973 
1974 /*
1975  * Main handler for configuration. We are guaranteed to be called
1976  * with an oid which is at least a dn_id.
1977  * - the first object is the command (config, delete, flush, ...)
1978  * - config_link must be issued after the corresponding config_sched
1979  * - parameters (DN_TXT) for an object must precede the object
1980  *   processed on a config_sched.
1981  */
1982 int
1983 do_config(void *p, int l)
1984 {
1985 	struct dn_id o;
1986 	union {
1987 		struct dn_profile profile;
1988 		struct dn_fs fs;
1989 		struct dn_link link;
1990 		struct dn_sch sched;
1991 	} *dn;
1992 	struct dn_id *arg;
1993 	uintptr_t a;
1994 	int err, err2, off;
1995 
1996 	memcpy(&o, p, sizeof(o));
1997 	if (o.id != DN_API_VERSION) {
1998 		D("invalid api version got %d need %d", o.id, DN_API_VERSION);
1999 		return EINVAL;
2000 	}
2001 	arg = NULL;
2002 	dn = NULL;
2003 	for (off = 0; l >= sizeof(o); memcpy(&o, (char *)p + off, sizeof(o))) {
2004 		if (o.len < sizeof(o) || l < o.len) {
2005 			D("bad len o.len %d len %d", o.len, l);
2006 			err = EINVAL;
2007 			break;
2008 		}
2009 		l -= o.len;
2010 		err = 0;
2011 		switch (o.type) {
2012 		default:
2013 			D("cmd %d not implemented", o.type);
2014 			break;
2015 
2016 #ifdef EMULATE_SYSCTL
2017 		/* sysctl emulation.
2018 		 * if we recognize the command, jump to the correct
2019 		 * handler and return
2020 		 */
2021 		case DN_SYSCTL_SET:
2022 			err = kesysctl_emu_set(p, l);
2023 			return err;
2024 #endif
2025 
2026 		case DN_CMD_CONFIG: /* simply a header */
2027 			break;
2028 
2029 		case DN_CMD_DELETE:
2030 			/* the argument is in the first uintptr_t after o */
2031 			if (o.len < sizeof(o) + sizeof(a)) {
2032 				err = EINVAL;
2033 				break;
2034 			}
2035 			memcpy(&a, (char *)p + off + sizeof(o), sizeof(a));
2036 			switch (o.subtype) {
2037 			case DN_LINK:
2038 				/* delete base and derived schedulers */
2039 				DN_BH_WLOCK();
2040 				err = delete_schk(a);
2041 				err2 = delete_schk(a + DN_MAX_ID);
2042 				DN_BH_WUNLOCK();
2043 				if (!err)
2044 					err = err2;
2045 				break;
2046 
2047 			default:
2048 				D("invalid delete type %d", o.subtype);
2049 				err = EINVAL;
2050 				break;
2051 
2052 			case DN_FS:
2053 				err = (a < 1 || a >= DN_MAX_ID) ?
2054 				    EINVAL : delete_fs(a, 0) ;
2055 				break;
2056 			}
2057 			break;
2058 
2059 		case DN_CMD_FLUSH:
2060 			DN_BH_WLOCK();
2061 			dummynet_flush();
2062 			DN_BH_WUNLOCK();
2063 			break;
2064 		case DN_TEXT:	/* store argument of next block */
2065 			if (arg != NULL)
2066 				free(arg, M_TEMP);
2067 			arg = malloc(o.len, M_TEMP, M_WAITOK);
2068 			memcpy(arg, (char *)p + off, o.len);
2069 			break;
2070 		case DN_LINK:
2071 			if (dn == NULL)
2072 				dn = malloc(sizeof(*dn), M_TEMP, M_WAITOK);
2073 			memcpy(&dn->link, (char *)p + off, sizeof(dn->link));
2074 			err = config_link(&dn->link, arg);
2075 			break;
2076 		case DN_PROFILE:
2077 			if (dn == NULL)
2078 				dn = malloc(sizeof(*dn), M_TEMP, M_WAITOK);
2079 			memcpy(&dn->profile, (char *)p + off,
2080 			    sizeof(dn->profile));
2081 			err = config_profile(&dn->profile, arg);
2082 			break;
2083 		case DN_SCH:
2084 			if (dn == NULL)
2085 				dn = malloc(sizeof(*dn), M_TEMP, M_WAITOK);
2086 			memcpy(&dn->sched, (char *)p + off,
2087 			    sizeof(dn->sched));
2088 			err = config_sched(&dn->sched, arg);
2089 			break;
2090 		case DN_FS:
2091 			if (dn == NULL)
2092 				dn = malloc(sizeof(*dn), M_TEMP, M_WAITOK);
2093 			memcpy(&dn->fs, (char *)p + off, sizeof(dn->fs));
2094 			err = (NULL == config_fs(&dn->fs, arg, 0));
2095 			break;
2096 		}
2097 		if (err != 0)
2098 			break;
2099 		off += o.len;
2100 	}
2101 	if (arg != NULL)
2102 		free(arg, M_TEMP);
2103 	if (dn != NULL)
2104 		free(dn, M_TEMP);
2105 	return err;
2106 }
2107 
2108 static int
2109 compute_space(struct dn_id *cmd, struct copy_args *a)
2110 {
2111 	int x = 0, need = 0;
2112 	int profile_size = sizeof(struct dn_profile) -
2113 		ED_MAX_SAMPLES_NO*sizeof(int);
2114 
2115 	/* NOTE about compute space:
2116 	 * NP 	= dn_cfg.schk_count
2117 	 * NSI 	= dn_cfg.si_count
2118 	 * NF 	= dn_cfg.fsk_count
2119 	 * NQ 	= dn_cfg.queue_count
2120 	 * - ipfw pipe show
2121 	 *   (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
2122 	 *                             link, scheduler template, flowset
2123 	 *                             integrated in scheduler and header
2124 	 *                             for flowset list
2125 	 *   (NSI)*(dn_flow) all scheduler instance (includes
2126 	 *                              the queue instance)
2127 	 * - ipfw sched show
2128 	 *   (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
2129 	 *                             link, scheduler template, flowset
2130 	 *                             integrated in scheduler and header
2131 	 *                             for flowset list
2132 	 *   (NSI * dn_flow) all scheduler instances
2133 	 *   (NF * sizeof(uint_32)) space for flowset list linked to scheduler
2134 	 *   (NQ * dn_queue) all queue [XXXfor now not listed]
2135 	 * - ipfw queue show
2136 	 *   (NF * dn_fs) all flowset
2137 	 *   (NQ * dn_queue) all queues
2138 	 */
2139 	switch (cmd->subtype) {
2140 	default:
2141 		return -1;
2142 	/* XXX where do LINK and SCH differ ? */
2143 	/* 'ipfw sched show' could list all queues associated to
2144 	 * a scheduler. This feature for now is disabled
2145 	 */
2146 	case DN_LINK:	/* pipe show */
2147 		x = DN_C_LINK | DN_C_SCH | DN_C_FLOW;
2148 		need += dn_cfg.schk_count *
2149 			(sizeof(struct dn_fs) + profile_size) / 2;
2150 		need += dn_cfg.fsk_count * sizeof(uint32_t);
2151 		break;
2152 	case DN_SCH:	/* sched show */
2153 		need += dn_cfg.schk_count *
2154 			(sizeof(struct dn_fs) + profile_size) / 2;
2155 		need += dn_cfg.fsk_count * sizeof(uint32_t);
2156 		x = DN_C_SCH | DN_C_LINK | DN_C_FLOW;
2157 		break;
2158 	case DN_FS:	/* queue show */
2159 		x = DN_C_FS | DN_C_QUEUE;
2160 		break;
2161 	case DN_GET_COMPAT:	/* compatibility mode */
2162 		need =  dn_compat_calc_size();
2163 		break;
2164 	}
2165 	a->flags = x;
2166 	if (x & DN_C_SCH) {
2167 		need += dn_cfg.schk_count * sizeof(struct dn_sch) / 2;
2168 		/* NOT also, each fs might be attached to a sched */
2169 		need += dn_cfg.schk_count * sizeof(struct dn_id) / 2;
2170 	}
2171 	if (x & DN_C_FS)
2172 		need += dn_cfg.fsk_count * sizeof(struct dn_fs);
2173 	if (x & DN_C_LINK) {
2174 		need += dn_cfg.schk_count * sizeof(struct dn_link) / 2;
2175 	}
2176 	/*
2177 	 * When exporting a queue to userland, only pass up the
2178 	 * struct dn_flow, which is the only visible part.
2179 	 */
2180 
2181 	if (x & DN_C_QUEUE)
2182 		need += dn_cfg.queue_count * sizeof(struct dn_flow);
2183 	if (x & DN_C_FLOW)
2184 		need += dn_cfg.si_count * (sizeof(struct dn_flow));
2185 	return need;
2186 }
2187 
2188 /*
2189  * If compat != NULL dummynet_get is called in compatibility mode.
2190  * *compat will be the pointer to the buffer to pass to ipfw
2191  */
2192 int
2193 dummynet_get(struct sockopt *sopt, void **compat)
2194 {
2195 	int have, i, need, error;
2196 	char *start = NULL, *buf;
2197 	size_t sopt_valsize;
2198 	struct dn_id *cmd;
2199 	struct copy_args a;
2200 	struct copy_range r;
2201 	int l = sizeof(struct dn_id);
2202 
2203 	bzero(&a, sizeof(a));
2204 	bzero(&r, sizeof(r));
2205 
2206 	/* save and restore original sopt_valsize around copyin */
2207 	sopt_valsize = sopt->sopt_valsize;
2208 
2209 	cmd = &r.o;
2210 
2211 	if (!compat) {
2212 		/* copy at least an oid, and possibly a full object */
2213 		error = sooptcopyin(sopt, cmd, sizeof(r), sizeof(*cmd));
2214 		sopt->sopt_valsize = sopt_valsize;
2215 		if (error)
2216 			goto done;
2217 		l = cmd->len;
2218 #ifdef EMULATE_SYSCTL
2219 		/* sysctl emulation. */
2220 		if (cmd->type == DN_SYSCTL_GET)
2221 			return kesysctl_emu_get(sopt);
2222 #endif
2223 		if (l > sizeof(r)) {
2224 			/* request larger than default, allocate buffer */
2225 			cmd = malloc(l,  M_DUMMYNET, M_WAITOK);
2226 			error = sooptcopyin(sopt, cmd, l, l);
2227 			sopt->sopt_valsize = sopt_valsize;
2228 			if (error)
2229 				goto done;
2230 		}
2231 	} else { /* compatibility */
2232 		error = 0;
2233 		cmd->type = DN_CMD_GET;
2234 		cmd->len = sizeof(struct dn_id);
2235 		cmd->subtype = DN_GET_COMPAT;
2236 		// cmd->id = sopt_valsize;
2237 		D("compatibility mode");
2238 	}
2239 
2240 #ifdef NEW_AQM
2241 	/* get AQM params */
2242 	if(cmd->subtype == DN_AQM_PARAMS) {
2243 		error = get_aqm_parms(sopt);
2244 		goto done;
2245 	/* get Scheduler params */
2246 	} else if (cmd->subtype == DN_SCH_PARAMS) {
2247 		error = get_sched_parms(sopt);
2248 		goto done;
2249 	}
2250 #endif
2251 
2252 	a.extra = (struct copy_range *)cmd;
2253 	if (cmd->len == sizeof(*cmd)) { /* no range, create a default */
2254 		uint32_t *rp = (uint32_t *)(cmd + 1);
2255 		cmd->len += 2* sizeof(uint32_t);
2256 		rp[0] = 1;
2257 		rp[1] = DN_MAX_ID - 1;
2258 		if (cmd->subtype == DN_LINK) {
2259 			rp[0] += DN_MAX_ID;
2260 			rp[1] += DN_MAX_ID;
2261 		}
2262 	}
2263 	/* Count space (under lock) and allocate (outside lock).
2264 	 * Exit with lock held if we manage to get enough buffer.
2265 	 * Try a few times then give up.
2266 	 */
2267 	for (have = 0, i = 0; i < 10; i++) {
2268 		DN_BH_WLOCK();
2269 		need = compute_space(cmd, &a);
2270 
2271 		/* if there is a range, ignore value from compute_space() */
2272 		if (l > sizeof(*cmd))
2273 			need = sopt_valsize - sizeof(*cmd);
2274 
2275 		if (need < 0) {
2276 			DN_BH_WUNLOCK();
2277 			error = EINVAL;
2278 			goto done;
2279 		}
2280 		need += sizeof(*cmd);
2281 		cmd->id = need;
2282 		if (have >= need)
2283 			break;
2284 
2285 		DN_BH_WUNLOCK();
2286 		if (start)
2287 			free(start, M_DUMMYNET);
2288 		start = NULL;
2289 		if (need > sopt_valsize)
2290 			break;
2291 
2292 		have = need;
2293 		start = malloc(have, M_DUMMYNET, M_WAITOK | M_ZERO);
2294 	}
2295 
2296 	if (start == NULL) {
2297 		if (compat) {
2298 			*compat = NULL;
2299 			error =  1; // XXX
2300 		} else {
2301 			error = sooptcopyout(sopt, cmd, sizeof(*cmd));
2302 		}
2303 		goto done;
2304 	}
2305 	ND("have %d:%d sched %d, %d:%d links %d, %d:%d flowsets %d, "
2306 		"%d:%d si %d, %d:%d queues %d",
2307 		dn_cfg.schk_count, sizeof(struct dn_sch), DN_SCH,
2308 		dn_cfg.schk_count, sizeof(struct dn_link), DN_LINK,
2309 		dn_cfg.fsk_count, sizeof(struct dn_fs), DN_FS,
2310 		dn_cfg.si_count, sizeof(struct dn_flow), DN_SCH_I,
2311 		dn_cfg.queue_count, sizeof(struct dn_queue), DN_QUEUE);
2312 	sopt->sopt_valsize = sopt_valsize;
2313 	a.type = cmd->subtype;
2314 
2315 	if (compat == NULL) {
2316 		memcpy(start, cmd, sizeof(*cmd));
2317 		((struct dn_id*)(start))->len = sizeof(struct dn_id);
2318 		buf = start + sizeof(*cmd);
2319 	} else
2320 		buf = start;
2321 	a.start = &buf;
2322 	a.end = start + have;
2323 	/* start copying other objects */
2324 	if (compat) {
2325 		a.type = DN_COMPAT_PIPE;
2326 		dn_ht_scan(dn_cfg.schedhash, copy_data_helper_compat, &a);
2327 		a.type = DN_COMPAT_QUEUE;
2328 		dn_ht_scan(dn_cfg.fshash, copy_data_helper_compat, &a);
2329 	} else if (a.type == DN_FS) {
2330 		dn_ht_scan(dn_cfg.fshash, copy_data_helper, &a);
2331 	} else {
2332 		dn_ht_scan(dn_cfg.schedhash, copy_data_helper, &a);
2333 	}
2334 	DN_BH_WUNLOCK();
2335 
2336 	if (compat) {
2337 		*compat = start;
2338 		sopt->sopt_valsize = buf - start;
2339 		/* free() is done by ip_dummynet_compat() */
2340 		start = NULL; //XXX hack
2341 	} else {
2342 		error = sooptcopyout(sopt, start, buf - start);
2343 	}
2344 done:
2345 	if (cmd && cmd != &r.o)
2346 		free(cmd, M_DUMMYNET);
2347 	if (start)
2348 		free(start, M_DUMMYNET);
2349 	return error;
2350 }
2351 
2352 /* Callback called on scheduler instance to delete it if idle */
2353 static int
2354 drain_scheduler_cb(void *_si, void *arg)
2355 {
2356 	struct dn_sch_inst *si = _si;
2357 
2358 	if ((si->kflags & DN_ACTIVE) || si->dline.mq.head != NULL)
2359 		return 0;
2360 
2361 	if (si->sched->fp->flags & DN_MULTIQUEUE) {
2362 		if (si->q_count == 0)
2363 			return si_destroy(si, NULL);
2364 		else
2365 			return 0;
2366 	} else { /* !DN_MULTIQUEUE */
2367 		if ((si+1)->ni.length == 0)
2368 			return si_destroy(si, NULL);
2369 		else
2370 			return 0;
2371 	}
2372 	return 0; /* unreachable */
2373 }
2374 
2375 /* Callback called on scheduler to check if it has instances */
2376 static int
2377 drain_scheduler_sch_cb(void *_s, void *arg)
2378 {
2379 	struct dn_schk *s = _s;
2380 
2381 	if (s->sch.flags & DN_HAVE_MASK) {
2382 		dn_ht_scan_bucket(s->siht, &s->drain_bucket,
2383 				drain_scheduler_cb, NULL);
2384 		s->drain_bucket++;
2385 	} else {
2386 		if (s->siht) {
2387 			if (drain_scheduler_cb(s->siht, NULL) == DNHT_SCAN_DEL)
2388 				s->siht = NULL;
2389 		}
2390 	}
2391 	return 0;
2392 }
2393 
2394 /* Called every tick, try to delete a 'bucket' of scheduler */
2395 void
2396 dn_drain_scheduler(void)
2397 {
2398 	dn_ht_scan_bucket(dn_cfg.schedhash, &dn_cfg.drain_sch,
2399 			   drain_scheduler_sch_cb, NULL);
2400 	dn_cfg.drain_sch++;
2401 }
2402 
2403 /* Callback called on queue to delete if it is idle */
2404 static int
2405 drain_queue_cb(void *_q, void *arg)
2406 {
2407 	struct dn_queue *q = _q;
2408 
2409 	if (q->ni.length == 0) {
2410 		dn_delete_queue(q, DN_DESTROY);
2411 		return DNHT_SCAN_DEL; /* queue is deleted */
2412 	}
2413 
2414 	return 0; /* queue isn't deleted */
2415 }
2416 
2417 /* Callback called on flowset used to check if it has queues */
2418 static int
2419 drain_queue_fs_cb(void *_fs, void *arg)
2420 {
2421 	struct dn_fsk *fs = _fs;
2422 
2423 	if (fs->fs.flags & DN_QHT_HASH) {
2424 		/* Flowset has a hash table for queues */
2425 		dn_ht_scan_bucket(fs->qht, &fs->drain_bucket,
2426 				drain_queue_cb, NULL);
2427 		fs->drain_bucket++;
2428 	} else {
2429 		/* No hash table for this flowset, null the pointer
2430 		 * if the queue is deleted
2431 		 */
2432 		if (fs->qht) {
2433 			if (drain_queue_cb(fs->qht, NULL) == DNHT_SCAN_DEL)
2434 				fs->qht = NULL;
2435 		}
2436 	}
2437 	return 0;
2438 }
2439 
2440 /* Called every tick, try to delete a 'bucket' of queue */
2441 void
2442 dn_drain_queue(void)
2443 {
2444 	/* scan a bucket of flowset */
2445 	dn_ht_scan_bucket(dn_cfg.fshash, &dn_cfg.drain_fs,
2446                                drain_queue_fs_cb, NULL);
2447 	dn_cfg.drain_fs++;
2448 }
2449 
2450 /*
2451  * Handler for the various dummynet socket options
2452  */
2453 static int
2454 ip_dn_ctl(struct sockopt *sopt)
2455 {
2456 	void *p = NULL;
2457 	int error, l;
2458 
2459 	error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2460 	if (error)
2461 		return (error);
2462 
2463 	/* Disallow sets in really-really secure mode. */
2464 	if (sopt->sopt_dir == SOPT_SET) {
2465 		error =  securelevel_ge(sopt->sopt_td->td_ucred, 3);
2466 		if (error)
2467 			return (error);
2468 	}
2469 
2470 	switch (sopt->sopt_name) {
2471 	default :
2472 		D("dummynet: unknown option %d", sopt->sopt_name);
2473 		error = EINVAL;
2474 		break;
2475 
2476 	case IP_DUMMYNET_FLUSH:
2477 	case IP_DUMMYNET_CONFIGURE:
2478 	case IP_DUMMYNET_DEL:	/* remove a pipe or queue */
2479 	case IP_DUMMYNET_GET:
2480 		D("dummynet: compat option %d", sopt->sopt_name);
2481 		error = ip_dummynet_compat(sopt);
2482 		break;
2483 
2484 	case IP_DUMMYNET3 :
2485 		if (sopt->sopt_dir == SOPT_GET) {
2486 			error = dummynet_get(sopt, NULL);
2487 			break;
2488 		}
2489 		l = sopt->sopt_valsize;
2490 		if (l < sizeof(struct dn_id) || l > 12000) {
2491 			D("argument len %d invalid", l);
2492 			break;
2493 		}
2494 		p = malloc(l, M_TEMP, M_WAITOK); // XXX can it fail ?
2495 		error = sooptcopyin(sopt, p, l, l);
2496 		if (error)
2497 			break ;
2498 		error = do_config(p, l);
2499 		break;
2500 	}
2501 
2502 	if (p != NULL)
2503 		free(p, M_TEMP);
2504 
2505 	return error ;
2506 }
2507 
2508 
2509 static void
2510 ip_dn_init(void)
2511 {
2512 	if (dn_cfg.init_done)
2513 		return;
2514 	printf("DUMMYNET %p with IPv6 initialized (100409)\n", curvnet);
2515 	dn_cfg.init_done = 1;
2516 	/* Set defaults here. MSVC does not accept initializers,
2517 	 * and this is also useful for vimages
2518 	 */
2519 	/* queue limits */
2520 	dn_cfg.slot_limit = 100; /* Foot shooting limit for queues. */
2521 	dn_cfg.byte_limit = 1024 * 1024;
2522 	dn_cfg.expire = 1;
2523 
2524 	/* RED parameters */
2525 	dn_cfg.red_lookup_depth = 256;	/* default lookup table depth */
2526 	dn_cfg.red_avg_pkt_size = 512;	/* default medium packet size */
2527 	dn_cfg.red_max_pkt_size = 1500;	/* default max packet size */
2528 
2529 	/* hash tables */
2530 	dn_cfg.max_hash_size = 65536;	/* max in the hash tables */
2531 	dn_cfg.hash_size = 64;		/* default hash size */
2532 
2533 	/* create hash tables for schedulers and flowsets.
2534 	 * In both we search by key and by pointer.
2535 	 */
2536 	dn_cfg.schedhash = dn_ht_init(NULL, dn_cfg.hash_size,
2537 		offsetof(struct dn_schk, schk_next),
2538 		schk_hash, schk_match, schk_new);
2539 	dn_cfg.fshash = dn_ht_init(NULL, dn_cfg.hash_size,
2540 		offsetof(struct dn_fsk, fsk_next),
2541 		fsk_hash, fsk_match, fsk_new);
2542 
2543 	/* bucket index to drain object */
2544 	dn_cfg.drain_fs = 0;
2545 	dn_cfg.drain_sch = 0;
2546 
2547 	heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
2548 	SLIST_INIT(&dn_cfg.fsu);
2549 	SLIST_INIT(&dn_cfg.schedlist);
2550 
2551 	DN_LOCK_INIT();
2552 
2553 	NET_TASK_INIT(&dn_task, 0, dummynet_task, curvnet);
2554 	dn_tq = taskqueue_create_fast("dummynet", M_WAITOK,
2555 	    taskqueue_thread_enqueue, &dn_tq);
2556 	taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2557 
2558 	callout_init(&dn_timeout, 1);
2559 	dn_reschedule();
2560 
2561 	/* Initialize curr_time adjustment mechanics. */
2562 	getmicrouptime(&dn_cfg.prev_t);
2563 }
2564 
2565 static void
2566 ip_dn_destroy(int last)
2567 {
2568 	DN_BH_WLOCK();
2569 	/* ensure no more callouts are started */
2570 	dn_gone = 1;
2571 
2572 	/* check for last */
2573 	if (last) {
2574 		ND("removing last instance\n");
2575 		ip_dn_ctl_ptr = NULL;
2576 		ip_dn_io_ptr = NULL;
2577 	}
2578 
2579 	dummynet_flush();
2580 	DN_BH_WUNLOCK();
2581 
2582 	callout_drain(&dn_timeout);
2583 	taskqueue_drain(dn_tq, &dn_task);
2584 	taskqueue_free(dn_tq);
2585 
2586 	dn_ht_free(dn_cfg.schedhash, 0);
2587 	dn_ht_free(dn_cfg.fshash, 0);
2588 	heap_free(&dn_cfg.evheap);
2589 
2590 	DN_LOCK_DESTROY();
2591 }
2592 
2593 static int
2594 dummynet_modevent(module_t mod, int type, void *data)
2595 {
2596 
2597 	if (type == MOD_LOAD) {
2598 		if (ip_dn_io_ptr) {
2599 			printf("DUMMYNET already loaded\n");
2600 			return EEXIST ;
2601 		}
2602 		ip_dn_init();
2603 		ip_dn_ctl_ptr = ip_dn_ctl;
2604 		ip_dn_io_ptr = dummynet_io;
2605 		return 0;
2606 	} else if (type == MOD_UNLOAD) {
2607 		ip_dn_destroy(1 /* last */);
2608 		return 0;
2609 	} else
2610 		return EOPNOTSUPP;
2611 }
2612 
2613 /* modevent helpers for the modules */
2614 static int
2615 load_dn_sched(struct dn_alg *d)
2616 {
2617 	struct dn_alg *s;
2618 
2619 	if (d == NULL)
2620 		return 1; /* error */
2621 	ip_dn_init();	/* just in case, we need the lock */
2622 
2623 	/* Check that mandatory funcs exists */
2624 	if (d->enqueue == NULL || d->dequeue == NULL) {
2625 		D("missing enqueue or dequeue for %s", d->name);
2626 		return 1;
2627 	}
2628 
2629 	/* Search if scheduler already exists */
2630 	DN_BH_WLOCK();
2631 	SLIST_FOREACH(s, &dn_cfg.schedlist, next) {
2632 		if (strcmp(s->name, d->name) == 0) {
2633 			D("%s already loaded", d->name);
2634 			break; /* scheduler already exists */
2635 		}
2636 	}
2637 	if (s == NULL)
2638 		SLIST_INSERT_HEAD(&dn_cfg.schedlist, d, next);
2639 	DN_BH_WUNLOCK();
2640 	D("dn_sched %s %sloaded", d->name, s ? "not ":"");
2641 	return s ? 1 : 0;
2642 }
2643 
2644 static int
2645 unload_dn_sched(struct dn_alg *s)
2646 {
2647 	struct dn_alg *tmp, *r;
2648 	int err = EINVAL;
2649 
2650 	ND("called for %s", s->name);
2651 
2652 	DN_BH_WLOCK();
2653 	SLIST_FOREACH_SAFE(r, &dn_cfg.schedlist, next, tmp) {
2654 		if (strcmp(s->name, r->name) != 0)
2655 			continue;
2656 		ND("ref_count = %d", r->ref_count);
2657 		err = (r->ref_count != 0) ? EBUSY : 0;
2658 		if (err == 0)
2659 			SLIST_REMOVE(&dn_cfg.schedlist, r, dn_alg, next);
2660 		break;
2661 	}
2662 	DN_BH_WUNLOCK();
2663 	D("dn_sched %s %sunloaded", s->name, err ? "not ":"");
2664 	return err;
2665 }
2666 
2667 int
2668 dn_sched_modevent(module_t mod, int cmd, void *arg)
2669 {
2670 	struct dn_alg *sch = arg;
2671 
2672 	if (cmd == MOD_LOAD)
2673 		return load_dn_sched(sch);
2674 	else if (cmd == MOD_UNLOAD)
2675 		return unload_dn_sched(sch);
2676 	else
2677 		return EINVAL;
2678 }
2679 
2680 static moduledata_t dummynet_mod = {
2681 	"dummynet", dummynet_modevent, NULL
2682 };
2683 
2684 #define	DN_SI_SUB	SI_SUB_PROTO_FIREWALL
2685 #define	DN_MODEV_ORD	(SI_ORDER_ANY - 128) /* after ipfw */
2686 DECLARE_MODULE(dummynet, dummynet_mod, DN_SI_SUB, DN_MODEV_ORD);
2687 MODULE_DEPEND(dummynet, ipfw, 3, 3, 3);
2688 MODULE_VERSION(dummynet, 3);
2689 
2690 /*
2691  * Starting up. Done in order after dummynet_modevent() has been called.
2692  * VNET_SYSINIT is also called for each existing vnet and each new vnet.
2693  */
2694 //VNET_SYSINIT(vnet_dn_init, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_init, NULL);
2695 
2696 /*
2697  * Shutdown handlers up shop. These are done in REVERSE ORDER, but still
2698  * after dummynet_modevent() has been called. Not called on reboot.
2699  * VNET_SYSUNINIT is also called for each exiting vnet as it exits.
2700  * or when the module is unloaded.
2701  */
2702 //VNET_SYSUNINIT(vnet_dn_uninit, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_destroy, NULL);
2703 
2704 #ifdef NEW_AQM
2705 
2706 /* modevent helpers for the AQM modules */
2707 static int
2708 load_dn_aqm(struct dn_aqm *d)
2709 {
2710 	struct dn_aqm *aqm=NULL;
2711 
2712 	if (d == NULL)
2713 		return 1; /* error */
2714 	ip_dn_init();	/* just in case, we need the lock */
2715 
2716 	/* Check that mandatory funcs exists */
2717 	if (d->enqueue == NULL || d->dequeue == NULL) {
2718 		D("missing enqueue or dequeue for %s", d->name);
2719 		return 1;
2720 	}
2721 
2722 	/* Search if AQM already exists */
2723 	DN_BH_WLOCK();
2724 	SLIST_FOREACH(aqm, &dn_cfg.aqmlist, next) {
2725 		if (strcmp(aqm->name, d->name) == 0) {
2726 			D("%s already loaded", d->name);
2727 			break; /* AQM already exists */
2728 		}
2729 	}
2730 	if (aqm == NULL)
2731 		SLIST_INSERT_HEAD(&dn_cfg.aqmlist, d, next);
2732 	DN_BH_WUNLOCK();
2733 	D("dn_aqm %s %sloaded", d->name, aqm ? "not ":"");
2734 	return aqm ? 1 : 0;
2735 }
2736 
2737 
2738 /* Callback to clean up AQM status for queues connected to a flowset
2739  * and then deconfigure the flowset.
2740  * This function is called before an AQM module is unloaded
2741  */
2742 static int
2743 fs_cleanup(void *_fs, void *arg)
2744 {
2745 	struct dn_fsk *fs = _fs;
2746 	uint32_t type = *(uint32_t *)arg;
2747 
2748 	if (fs->aqmfp && fs->aqmfp->type == type)
2749 		aqm_cleanup_deconfig_fs(fs);
2750 
2751 	return 0;
2752 }
2753 
2754 static int
2755 unload_dn_aqm(struct dn_aqm *aqm)
2756 {
2757 	struct dn_aqm *tmp, *r;
2758 	int err = EINVAL;
2759 	err = 0;
2760 	ND("called for %s", aqm->name);
2761 
2762 	DN_BH_WLOCK();
2763 
2764 	/* clean up AQM status and deconfig flowset */
2765 	dn_ht_scan(dn_cfg.fshash, fs_cleanup, &aqm->type);
2766 
2767 	SLIST_FOREACH_SAFE(r, &dn_cfg.aqmlist, next, tmp) {
2768 		if (strcmp(aqm->name, r->name) != 0)
2769 			continue;
2770 		ND("ref_count = %d", r->ref_count);
2771 		err = (r->ref_count != 0 || r->cfg_ref_count != 0) ? EBUSY : 0;
2772 		if (err == 0)
2773 			SLIST_REMOVE(&dn_cfg.aqmlist, r, dn_aqm, next);
2774 		break;
2775 	}
2776 	DN_BH_WUNLOCK();
2777 	D("%s %sunloaded", aqm->name, err ? "not ":"");
2778 	if (err)
2779 		D("ref_count=%d, cfg_ref_count=%d", r->ref_count, r->cfg_ref_count);
2780 	return err;
2781 }
2782 
2783 int
2784 dn_aqm_modevent(module_t mod, int cmd, void *arg)
2785 {
2786 	struct dn_aqm *aqm = arg;
2787 
2788 	if (cmd == MOD_LOAD)
2789 		return load_dn_aqm(aqm);
2790 	else if (cmd == MOD_UNLOAD)
2791 		return unload_dn_aqm(aqm);
2792 	else
2793 		return EINVAL;
2794 }
2795 #endif
2796 
2797 /* end of file */
2798 
2799