xref: /freebsd/sys/netpfil/ipfw/ip_dummynet.c (revision 81ad626541db97eb356e2c1d4a20eb2a26a766ab)
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/ck.h>
48 #include <sys/systm.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/module.h>
54 #include <sys/mutex.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/rwlock.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/time.h>
61 #include <sys/taskqueue.h>
62 #include <net/if.h>	/* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */
63 #include <netinet/in.h>
64 #include <netinet/ip_var.h>	/* ip_output(), IP_FORWARDING */
65 #include <netinet/ip_fw.h>
66 #include <netinet/ip_dummynet.h>
67 #include <net/vnet.h>
68 
69 #include <netpfil/ipfw/ip_fw_private.h>
70 #include <netpfil/ipfw/dn_heap.h>
71 #include <netpfil/ipfw/ip_dn_private.h>
72 #ifdef NEW_AQM
73 #include <netpfil/ipfw/dn_aqm.h>
74 #endif
75 #include <netpfil/ipfw/dn_sched.h>
76 
77 /* which objects to copy */
78 #define DN_C_LINK 	0x01
79 #define DN_C_SCH	0x02
80 #define DN_C_FLOW	0x04
81 #define DN_C_FS		0x08
82 #define DN_C_QUEUE	0x10
83 
84 /* we use this argument in case of a schk_new */
85 struct schk_new_arg {
86 	struct dn_alg *fp;
87 	struct dn_sch *sch;
88 };
89 
90 /*---- callout hooks. ----*/
91 static struct callout dn_timeout;
92 static int dn_tasks_started = 0;
93 static int dn_gone;
94 static struct task	dn_task;
95 static struct taskqueue	*dn_tq = NULL;
96 
97 /* global scheduler list */
98 struct mtx		sched_mtx;
99 CK_LIST_HEAD(, dn_alg)	schedlist;
100 #ifdef NEW_AQM
101 CK_LIST_HEAD(, dn_aqm)	aqmlist;	/* list of AQMs */
102 #endif
103 
104 static void
105 dummynet(void *arg)
106 {
107 
108 	(void)arg;	/* UNUSED */
109 	taskqueue_enqueue(dn_tq, &dn_task);
110 }
111 
112 void
113 dummynet_sched_lock(void)
114 {
115 	mtx_lock(&sched_mtx);
116 }
117 
118 void
119 dummynet_sched_unlock(void)
120 {
121 	mtx_unlock(&sched_mtx);
122 }
123 
124 void
125 dn_reschedule(void)
126 {
127 
128 	if (dn_gone != 0)
129 		return;
130 	callout_reset_sbt(&dn_timeout, tick_sbt, 0, dummynet, NULL,
131 	    C_HARDCLOCK | C_DIRECT_EXEC);
132 }
133 /*----- end of callout hooks -----*/
134 
135 #ifdef NEW_AQM
136 /* Return AQM descriptor for given type or name. */
137 static struct dn_aqm *
138 find_aqm_type(int type, char *name)
139 {
140 	struct dn_aqm *d;
141 
142 	NET_EPOCH_ASSERT();
143 
144 	CK_LIST_FOREACH(d, &aqmlist, next) {
145 		if (d->type == type || (name && !strcasecmp(d->name, name)))
146 			return d;
147 	}
148 	return NULL; /* not found */
149 }
150 #endif
151 
152 /* Return a scheduler descriptor given the type or name. */
153 static struct dn_alg *
154 find_sched_type(int type, char *name)
155 {
156 	struct dn_alg *d;
157 
158 	NET_EPOCH_ASSERT();
159 
160 	CK_LIST_FOREACH(d, &schedlist, next) {
161 		if (d->type == type || (name && !strcasecmp(d->name, name)))
162 			return d;
163 	}
164 	return NULL; /* not found */
165 }
166 
167 int
168 ipdn_bound_var(int *v, int dflt, int lo, int hi, const char *msg)
169 {
170 	int oldv = *v;
171 	const char *op = NULL;
172 	if (dflt < lo)
173 		dflt = lo;
174 	if (dflt > hi)
175 		dflt = hi;
176 	if (oldv < lo) {
177 		*v = dflt;
178 		op = "Bump";
179 	} else if (oldv > hi) {
180 		*v = hi;
181 		op = "Clamp";
182 	} else
183 		return *v;
184 	if (op && msg && bootverbose)
185 		printf("%s %s to %d (was %d)\n", op, msg, *v, oldv);
186 	return *v;
187 }
188 
189 /*---- flow_id mask, hash and compare functions ---*/
190 /*
191  * The flow_id includes the 5-tuple, the queue/pipe number
192  * which we store in the extra area in host order,
193  * and for ipv6 also the flow_id6.
194  * XXX see if we want the tos byte (can store in 'flags')
195  */
196 static struct ipfw_flow_id *
197 flow_id_mask(struct ipfw_flow_id *mask, struct ipfw_flow_id *id)
198 {
199 	int is_v6 = IS_IP6_FLOW_ID(id);
200 
201 	id->dst_port &= mask->dst_port;
202 	id->src_port &= mask->src_port;
203 	id->proto &= mask->proto;
204 	id->extra &= mask->extra;
205 	if (is_v6) {
206 		APPLY_MASK(&id->dst_ip6, &mask->dst_ip6);
207 		APPLY_MASK(&id->src_ip6, &mask->src_ip6);
208 		id->flow_id6 &= mask->flow_id6;
209 	} else {
210 		id->dst_ip &= mask->dst_ip;
211 		id->src_ip &= mask->src_ip;
212 	}
213 	return id;
214 }
215 
216 /* computes an OR of two masks, result in dst and also returned */
217 static struct ipfw_flow_id *
218 flow_id_or(struct ipfw_flow_id *src, struct ipfw_flow_id *dst)
219 {
220 	int is_v6 = IS_IP6_FLOW_ID(dst);
221 
222 	dst->dst_port |= src->dst_port;
223 	dst->src_port |= src->src_port;
224 	dst->proto |= src->proto;
225 	dst->extra |= src->extra;
226 	if (is_v6) {
227 #define OR_MASK(_d, _s)                          \
228     (_d)->__u6_addr.__u6_addr32[0] |= (_s)->__u6_addr.__u6_addr32[0]; \
229     (_d)->__u6_addr.__u6_addr32[1] |= (_s)->__u6_addr.__u6_addr32[1]; \
230     (_d)->__u6_addr.__u6_addr32[2] |= (_s)->__u6_addr.__u6_addr32[2]; \
231     (_d)->__u6_addr.__u6_addr32[3] |= (_s)->__u6_addr.__u6_addr32[3];
232 		OR_MASK(&dst->dst_ip6, &src->dst_ip6);
233 		OR_MASK(&dst->src_ip6, &src->src_ip6);
234 #undef OR_MASK
235 		dst->flow_id6 |= src->flow_id6;
236 	} else {
237 		dst->dst_ip |= src->dst_ip;
238 		dst->src_ip |= src->src_ip;
239 	}
240 	return dst;
241 }
242 
243 static int
244 nonzero_mask(struct ipfw_flow_id *m)
245 {
246 	if (m->dst_port || m->src_port || m->proto || m->extra)
247 		return 1;
248 	if (IS_IP6_FLOW_ID(m)) {
249 		return
250 			m->dst_ip6.__u6_addr.__u6_addr32[0] ||
251 			m->dst_ip6.__u6_addr.__u6_addr32[1] ||
252 			m->dst_ip6.__u6_addr.__u6_addr32[2] ||
253 			m->dst_ip6.__u6_addr.__u6_addr32[3] ||
254 			m->src_ip6.__u6_addr.__u6_addr32[0] ||
255 			m->src_ip6.__u6_addr.__u6_addr32[1] ||
256 			m->src_ip6.__u6_addr.__u6_addr32[2] ||
257 			m->src_ip6.__u6_addr.__u6_addr32[3] ||
258 			m->flow_id6;
259 	} else {
260 		return m->dst_ip || m->src_ip;
261 	}
262 }
263 
264 /* XXX we may want a better hash function */
265 static uint32_t
266 flow_id_hash(struct ipfw_flow_id *id)
267 {
268     uint32_t i;
269 
270     if (IS_IP6_FLOW_ID(id)) {
271 	uint32_t *d = (uint32_t *)&id->dst_ip6;
272 	uint32_t *s = (uint32_t *)&id->src_ip6;
273         i = (d[0]      ) ^ (d[1])       ^
274             (d[2]      ) ^ (d[3])       ^
275             (d[0] >> 15) ^ (d[1] >> 15) ^
276             (d[2] >> 15) ^ (d[3] >> 15) ^
277             (s[0] <<  1) ^ (s[1] <<  1) ^
278             (s[2] <<  1) ^ (s[3] <<  1) ^
279             (s[0] << 16) ^ (s[1] << 16) ^
280             (s[2] << 16) ^ (s[3] << 16) ^
281             (id->dst_port << 1) ^ (id->src_port) ^
282 	    (id->extra) ^
283             (id->proto ) ^ (id->flow_id6);
284     } else {
285         i = (id->dst_ip)        ^ (id->dst_ip >> 15) ^
286             (id->src_ip << 1)   ^ (id->src_ip >> 16) ^
287 	    (id->extra) ^
288             (id->dst_port << 1) ^ (id->src_port)     ^ (id->proto);
289     }
290     return i;
291 }
292 
293 /* Like bcmp, returns 0 if ids match, 1 otherwise. */
294 static int
295 flow_id_cmp(struct ipfw_flow_id *id1, struct ipfw_flow_id *id2)
296 {
297 	int is_v6 = IS_IP6_FLOW_ID(id1);
298 
299 	if (!is_v6) {
300 	    if (IS_IP6_FLOW_ID(id2))
301 		return 1; /* different address families */
302 
303 	    return (id1->dst_ip == id2->dst_ip &&
304 		    id1->src_ip == id2->src_ip &&
305 		    id1->dst_port == id2->dst_port &&
306 		    id1->src_port == id2->src_port &&
307 		    id1->proto == id2->proto &&
308 		    id1->extra == id2->extra) ? 0 : 1;
309 	}
310 	/* the ipv6 case */
311 	return (
312 	    !bcmp(&id1->dst_ip6,&id2->dst_ip6, sizeof(id1->dst_ip6)) &&
313 	    !bcmp(&id1->src_ip6,&id2->src_ip6, sizeof(id1->src_ip6)) &&
314 	    id1->dst_port == id2->dst_port &&
315 	    id1->src_port == id2->src_port &&
316 	    id1->proto == id2->proto &&
317 	    id1->extra == id2->extra &&
318 	    id1->flow_id6 == id2->flow_id6) ? 0 : 1;
319 }
320 /*--------- end of flow-id mask, hash and compare ---------*/
321 
322 /*--- support functions for the qht hashtable ----
323  * Entries are hashed by flow-id
324  */
325 static uint32_t
326 q_hash(uintptr_t key, int flags, void *arg)
327 {
328 	/* compute the hash slot from the flow id */
329 	struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
330 		&((struct dn_queue *)key)->ni.fid :
331 		(struct ipfw_flow_id *)key;
332 
333 	return flow_id_hash(id);
334 }
335 
336 static int
337 q_match(void *obj, uintptr_t key, int flags, void *arg)
338 {
339 	struct dn_queue *o = (struct dn_queue *)obj;
340 	struct ipfw_flow_id *id2;
341 
342 	if (flags & DNHT_KEY_IS_OBJ) {
343 		/* compare pointers */
344 		id2 = &((struct dn_queue *)key)->ni.fid;
345 	} else {
346 		id2 = (struct ipfw_flow_id *)key;
347 	}
348 	return (0 == flow_id_cmp(&o->ni.fid,  id2));
349 }
350 
351 /*
352  * create a new queue instance for the given 'key'.
353  */
354 static void *
355 q_new(uintptr_t key, int flags, void *arg)
356 {
357 	struct dn_queue *q, *template = arg;
358 	struct dn_fsk *fs = template->fs;
359 	int size = sizeof(*q) + fs->sched->fp->q_datalen;
360 
361 	q = malloc(size, M_DUMMYNET, M_NOWAIT | M_ZERO);
362 	if (q == NULL) {
363 		D("no memory for new queue");
364 		return NULL;
365 	}
366 
367 	set_oid(&q->ni.oid, DN_QUEUE, size);
368 	if (fs->fs.flags & DN_QHT_HASH)
369 		q->ni.fid = *(struct ipfw_flow_id *)key;
370 	q->fs = fs;
371 	q->_si = template->_si;
372 	q->_si->q_count++;
373 
374 	if (fs->sched->fp->new_queue)
375 		fs->sched->fp->new_queue(q);
376 
377 #ifdef NEW_AQM
378 	/* call AQM init function after creating a queue*/
379 	if (fs->aqmfp && fs->aqmfp->init)
380 		if(fs->aqmfp->init(q))
381 			D("unable to init AQM for fs %d", fs->fs.fs_nr);
382 #endif
383 	V_dn_cfg.queue_count++;
384 
385 	return q;
386 }
387 
388 /*
389  * Notify schedulers that a queue is going away.
390  * If (flags & DN_DESTROY), also free the packets.
391  * The version for callbacks is called q_delete_cb().
392  */
393 static void
394 dn_delete_queue(struct dn_queue *q, int flags)
395 {
396 	struct dn_fsk *fs = q->fs;
397 
398 #ifdef NEW_AQM
399 	/* clean up AQM status for queue 'q'
400 	 * cleanup here is called just with MULTIQUEUE
401 	 */
402 	if (fs && fs->aqmfp && fs->aqmfp->cleanup)
403 		fs->aqmfp->cleanup(q);
404 #endif
405 	// D("fs %p si %p\n", fs, q->_si);
406 	/* notify the parent scheduler that the queue is going away */
407 	if (fs && fs->sched->fp->free_queue)
408 		fs->sched->fp->free_queue(q);
409 	q->_si->q_count--;
410 	q->_si = NULL;
411 	if (flags & DN_DESTROY) {
412 		if (q->mq.head)
413 			dn_free_pkts(q->mq.head);
414 		bzero(q, sizeof(*q));	// safety
415 		free(q, M_DUMMYNET);
416 		V_dn_cfg.queue_count--;
417 	}
418 }
419 
420 static int
421 q_delete_cb(void *q, void *arg)
422 {
423 	int flags = (int)(uintptr_t)arg;
424 	dn_delete_queue(q, flags);
425 	return (flags & DN_DESTROY) ? DNHT_SCAN_DEL : 0;
426 }
427 
428 /*
429  * calls dn_delete_queue/q_delete_cb on all queues,
430  * which notifies the parent scheduler and possibly drains packets.
431  * flags & DN_DESTROY: drains queues and destroy qht;
432  */
433 static void
434 qht_delete(struct dn_fsk *fs, int flags)
435 {
436 	ND("fs %d start flags %d qht %p",
437 		fs->fs.fs_nr, flags, fs->qht);
438 	if (!fs->qht)
439 		return;
440 	if (fs->fs.flags & DN_QHT_HASH) {
441 		dn_ht_scan(fs->qht, q_delete_cb, (void *)(uintptr_t)flags);
442 		if (flags & DN_DESTROY) {
443 			dn_ht_free(fs->qht, 0);
444 			fs->qht = NULL;
445 		}
446 	} else {
447 		dn_delete_queue((struct dn_queue *)(fs->qht), flags);
448 		if (flags & DN_DESTROY)
449 			fs->qht = NULL;
450 	}
451 }
452 
453 /*
454  * Find and possibly create the queue for a MULTIQUEUE scheduler.
455  * We never call it for !MULTIQUEUE (the queue is in the sch_inst).
456  */
457 struct dn_queue *
458 ipdn_q_find(struct dn_fsk *fs, struct dn_sch_inst *si,
459 	struct ipfw_flow_id *id)
460 {
461 	struct dn_queue template;
462 
463 	template._si = si;
464 	template.fs = fs;
465 
466 	if (fs->fs.flags & DN_QHT_HASH) {
467 		struct ipfw_flow_id masked_id;
468 		if (fs->qht == NULL) {
469 			fs->qht = dn_ht_init(NULL, fs->fs.buckets,
470 				offsetof(struct dn_queue, q_next),
471 				q_hash, q_match, q_new);
472 			if (fs->qht == NULL)
473 				return NULL;
474 		}
475 		masked_id = *id;
476 		flow_id_mask(&fs->fsk_mask, &masked_id);
477 		return dn_ht_find(fs->qht, (uintptr_t)&masked_id,
478 			DNHT_INSERT, &template);
479 	} else {
480 		if (fs->qht == NULL)
481 			fs->qht = q_new(0, 0, &template);
482 		return (struct dn_queue *)fs->qht;
483 	}
484 }
485 /*--- end of queue hash table ---*/
486 
487 /*--- support functions for the sch_inst hashtable ----
488  *
489  * These are hashed by flow-id
490  */
491 static uint32_t
492 si_hash(uintptr_t key, int flags, void *arg)
493 {
494 	/* compute the hash slot from the flow id */
495 	struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
496 		&((struct dn_sch_inst *)key)->ni.fid :
497 		(struct ipfw_flow_id *)key;
498 
499 	return flow_id_hash(id);
500 }
501 
502 static int
503 si_match(void *obj, uintptr_t key, int flags, void *arg)
504 {
505 	struct dn_sch_inst *o = obj;
506 	struct ipfw_flow_id *id2;
507 
508 	id2 = (flags & DNHT_KEY_IS_OBJ) ?
509 		&((struct dn_sch_inst *)key)->ni.fid :
510 		(struct ipfw_flow_id *)key;
511 	return flow_id_cmp(&o->ni.fid,  id2) == 0;
512 }
513 
514 /*
515  * create a new instance for the given 'key'
516  * Allocate memory for instance, delay line and scheduler private data.
517  */
518 static void *
519 si_new(uintptr_t key, int flags, void *arg)
520 {
521 	struct dn_schk *s = arg;
522 	struct dn_sch_inst *si;
523 	int l = sizeof(*si) + s->fp->si_datalen;
524 
525 	si = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
526 	if (si == NULL)
527 		goto error;
528 
529 	/* Set length only for the part passed up to userland. */
530 	set_oid(&si->ni.oid, DN_SCH_I, sizeof(struct dn_flow));
531 	set_oid(&(si->dline.oid), DN_DELAY_LINE,
532 		sizeof(struct delay_line));
533 	/* mark si and dline as outside the event queue */
534 	si->ni.oid.id = si->dline.oid.id = -1;
535 
536 	si->sched = s;
537 	si->dline.si = si;
538 
539 	if (s->fp->new_sched && s->fp->new_sched(si)) {
540 		D("new_sched error");
541 		goto error;
542 	}
543 	if (s->sch.flags & DN_HAVE_MASK)
544 		si->ni.fid = *(struct ipfw_flow_id *)key;
545 
546 #ifdef NEW_AQM
547 	/* init AQM status for !DN_MULTIQUEUE sched*/
548 	if (!(s->fp->flags & DN_MULTIQUEUE))
549 		if (s->fs->aqmfp && s->fs->aqmfp->init)
550 			if(s->fs->aqmfp->init((struct dn_queue *)(si + 1))) {
551 				D("unable to init AQM for fs %d", s->fs->fs.fs_nr);
552 				goto error;
553 			}
554 #endif
555 
556 	V_dn_cfg.si_count++;
557 	return si;
558 
559 error:
560 	if (si) {
561 		bzero(si, sizeof(*si)); // safety
562 		free(si, M_DUMMYNET);
563 	}
564         return NULL;
565 }
566 
567 /*
568  * Callback from siht to delete all scheduler instances. Remove
569  * si and delay line from the system heap, destroy all queues.
570  * We assume that all flowset have been notified and do not
571  * point to us anymore.
572  */
573 static int
574 si_destroy(void *_si, void *arg)
575 {
576 	struct dn_sch_inst *si = _si;
577 	struct dn_schk *s = si->sched;
578 	struct delay_line *dl = &si->dline;
579 
580 	if (dl->oid.subtype) /* remove delay line from event heap */
581 		heap_extract(&V_dn_cfg.evheap, dl);
582 	dn_free_pkts(dl->mq.head);	/* drain delay line */
583 	if (si->kflags & DN_ACTIVE) /* remove si from event heap */
584 		heap_extract(&V_dn_cfg.evheap, si);
585 
586 #ifdef NEW_AQM
587 	/* clean up AQM status for !DN_MULTIQUEUE sched
588 	 * Note that all queues belong to fs were cleaned up in fsk_detach.
589 	 * When drain_scheduler is called s->fs and q->fs are pointing
590 	 * to a correct fs, so we can use fs in this case.
591 	 */
592 	if (!(s->fp->flags & DN_MULTIQUEUE)) {
593 		struct dn_queue *q = (struct dn_queue *)(si + 1);
594 		if (q->aqm_status && q->fs->aqmfp)
595 			if (q->fs->aqmfp->cleanup)
596 				q->fs->aqmfp->cleanup(q);
597 	}
598 #endif
599 	if (s->fp->free_sched)
600 		s->fp->free_sched(si);
601 	bzero(si, sizeof(*si));	/* safety */
602 	free(si, M_DUMMYNET);
603 	V_dn_cfg.si_count--;
604 	return DNHT_SCAN_DEL;
605 }
606 
607 /*
608  * Find the scheduler instance for this packet. If we need to apply
609  * a mask, do on a local copy of the flow_id to preserve the original.
610  * Assume siht is always initialized if we have a mask.
611  */
612 struct dn_sch_inst *
613 ipdn_si_find(struct dn_schk *s, struct ipfw_flow_id *id)
614 {
615 
616 	if (s->sch.flags & DN_HAVE_MASK) {
617 		struct ipfw_flow_id id_t = *id;
618 		flow_id_mask(&s->sch.sched_mask, &id_t);
619 		return dn_ht_find(s->siht, (uintptr_t)&id_t,
620 			DNHT_INSERT, s);
621 	}
622 	if (!s->siht)
623 		s->siht = si_new(0, 0, s);
624 	return (struct dn_sch_inst *)s->siht;
625 }
626 
627 /* callback to flush credit for the scheduler instance */
628 static int
629 si_reset_credit(void *_si, void *arg)
630 {
631 	struct dn_sch_inst *si = _si;
632 	struct dn_link *p = &si->sched->link;
633 
634 	si->credit = p->burst + (V_dn_cfg.io_fast ?  p->bandwidth : 0);
635 	return 0;
636 }
637 
638 static void
639 schk_reset_credit(struct dn_schk *s)
640 {
641 	if (s->sch.flags & DN_HAVE_MASK)
642 		dn_ht_scan(s->siht, si_reset_credit, NULL);
643 	else if (s->siht)
644 		si_reset_credit(s->siht, NULL);
645 }
646 /*---- end of sch_inst hashtable ---------------------*/
647 
648 /*-------------------------------------------------------
649  * flowset hash (fshash) support. Entries are hashed by fs_nr.
650  * New allocations are put in the fsunlinked list, from which
651  * they are removed when they point to a specific scheduler.
652  */
653 static uint32_t
654 fsk_hash(uintptr_t key, int flags, void *arg)
655 {
656 	uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
657 		((struct dn_fsk *)key)->fs.fs_nr;
658 
659 	return ( (i>>8)^(i>>4)^i );
660 }
661 
662 static int
663 fsk_match(void *obj, uintptr_t key, int flags, void *arg)
664 {
665 	struct dn_fsk *fs = obj;
666 	int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
667 		((struct dn_fsk *)key)->fs.fs_nr;
668 
669 	return (fs->fs.fs_nr == i);
670 }
671 
672 static void *
673 fsk_new(uintptr_t key, int flags, void *arg)
674 {
675 	struct dn_fsk *fs;
676 
677 	fs = malloc(sizeof(*fs), M_DUMMYNET, M_NOWAIT | M_ZERO);
678 	if (fs) {
679 		set_oid(&fs->fs.oid, DN_FS, sizeof(fs->fs));
680 		V_dn_cfg.fsk_count++;
681 		fs->drain_bucket = 0;
682 		SLIST_INSERT_HEAD(&V_dn_cfg.fsu, fs, sch_chain);
683 	}
684 	return fs;
685 }
686 
687 #ifdef NEW_AQM
688 /* callback function for cleaning up AQM queue status belongs to a flowset
689  * connected to scheduler instance '_si' (for !DN_MULTIQUEUE only).
690  */
691 static int
692 si_cleanup_q(void *_si, void *arg)
693 {
694 	struct dn_sch_inst *si = _si;
695 
696 	if (!(si->sched->fp->flags & DN_MULTIQUEUE)) {
697 		if (si->sched->fs->aqmfp && si->sched->fs->aqmfp->cleanup)
698 			si->sched->fs->aqmfp->cleanup((struct dn_queue *) (si+1));
699 	}
700 	return 0;
701 }
702 
703 /* callback to clean up queue AQM status.*/
704 static int
705 q_cleanup_q(void *_q, void *arg)
706 {
707 	struct dn_queue *q = _q;
708 	q->fs->aqmfp->cleanup(q);
709 	return 0;
710 }
711 
712 /* Clean up all AQM queues status belongs to flowset 'fs' and then
713  * deconfig AQM for flowset 'fs'
714  */
715 static void
716 aqm_cleanup_deconfig_fs(struct dn_fsk *fs)
717 {
718 	struct dn_sch_inst *si;
719 
720 	/* clean up AQM status for all queues for !DN_MULTIQUEUE sched*/
721 	if (fs->fs.fs_nr > DN_MAX_ID) {
722 		if (fs->sched && !(fs->sched->fp->flags & DN_MULTIQUEUE)) {
723 			if (fs->sched->sch.flags & DN_HAVE_MASK)
724 				dn_ht_scan(fs->sched->siht, si_cleanup_q, NULL);
725 			else {
726 					/* single si i.e. no sched mask */
727 					si = (struct dn_sch_inst *) fs->sched->siht;
728 					if (si && fs->aqmfp && fs->aqmfp->cleanup)
729 						fs->aqmfp->cleanup((struct dn_queue *) (si+1));
730 			}
731 		}
732 	}
733 
734 	/* clean up AQM status for all queues for DN_MULTIQUEUE sched*/
735 	if (fs->sched && fs->sched->fp->flags & DN_MULTIQUEUE && fs->qht) {
736 			if (fs->fs.flags & DN_QHT_HASH)
737 				dn_ht_scan(fs->qht, q_cleanup_q, NULL);
738 			else
739 				fs->aqmfp->cleanup((struct dn_queue *)(fs->qht));
740 	}
741 
742 	/* deconfig AQM */
743 	if(fs->aqmcfg && fs->aqmfp && fs->aqmfp->deconfig)
744 		fs->aqmfp->deconfig(fs);
745 }
746 #endif
747 
748 /*
749  * detach flowset from its current scheduler. Flags as follows:
750  * DN_DETACH removes from the fsk_list
751  * DN_DESTROY deletes individual queues
752  * DN_DELETE_FS destroys the flowset (otherwise goes in unlinked).
753  */
754 static void
755 fsk_detach(struct dn_fsk *fs, int flags)
756 {
757 	if (flags & DN_DELETE_FS)
758 		flags |= DN_DESTROY;
759 	ND("fs %d from sched %d flags %s %s %s",
760 		fs->fs.fs_nr, fs->fs.sched_nr,
761 		(flags & DN_DELETE_FS) ? "DEL_FS":"",
762 		(flags & DN_DESTROY) ? "DEL":"",
763 		(flags & DN_DETACH) ? "DET":"");
764 	if (flags & DN_DETACH) { /* detach from the list */
765 		struct dn_fsk_head *h;
766 		h = fs->sched ? &fs->sched->fsk_list : &V_dn_cfg.fsu;
767 		SLIST_REMOVE(h, fs, dn_fsk, sch_chain);
768 	}
769 	/* Free the RED parameters, they will be recomputed on
770 	 * subsequent attach if needed.
771 	 */
772 	free(fs->w_q_lookup, M_DUMMYNET);
773 	fs->w_q_lookup = NULL;
774 	qht_delete(fs, flags);
775 #ifdef NEW_AQM
776 	aqm_cleanup_deconfig_fs(fs);
777 #endif
778 
779 	if (fs->sched && fs->sched->fp->free_fsk)
780 		fs->sched->fp->free_fsk(fs);
781 	fs->sched = NULL;
782 	if (flags & DN_DELETE_FS) {
783 		bzero(fs, sizeof(*fs));	/* safety */
784 		free(fs, M_DUMMYNET);
785 		V_dn_cfg.fsk_count--;
786 	} else {
787 		SLIST_INSERT_HEAD(&V_dn_cfg.fsu, fs, sch_chain);
788 	}
789 }
790 
791 /*
792  * Detach or destroy all flowsets in a list.
793  * flags specifies what to do:
794  * DN_DESTROY:	flush all queues
795  * DN_DELETE_FS:	DN_DESTROY + destroy flowset
796  *	DN_DELETE_FS implies DN_DESTROY
797  */
798 static void
799 fsk_detach_list(struct dn_fsk_head *h, int flags)
800 {
801 	struct dn_fsk *fs;
802 	int n __unused = 0; /* only for stats */
803 
804 	ND("head %p flags %x", h, flags);
805 	while ((fs = SLIST_FIRST(h))) {
806 		SLIST_REMOVE_HEAD(h, sch_chain);
807 		n++;
808 		fsk_detach(fs, flags);
809 	}
810 	ND("done %d flowsets", n);
811 }
812 
813 /*
814  * called on 'queue X delete' -- removes the flowset from fshash,
815  * deletes all queues for the flowset, and removes the flowset.
816  */
817 static int
818 delete_fs(int i, int locked)
819 {
820 	struct dn_fsk *fs;
821 	int err = 0;
822 
823 	if (!locked)
824 		DN_BH_WLOCK();
825 	fs = dn_ht_find(V_dn_cfg.fshash, i, DNHT_REMOVE, NULL);
826 	ND("fs %d found %p", i, fs);
827 	if (fs) {
828 		fsk_detach(fs, DN_DETACH | DN_DELETE_FS);
829 		err = 0;
830 	} else
831 		err = EINVAL;
832 	if (!locked)
833 		DN_BH_WUNLOCK();
834 	return err;
835 }
836 
837 /*----- end of flowset hashtable support -------------*/
838 
839 /*------------------------------------------------------------
840  * Scheduler hash. When searching by index we pass sched_nr,
841  * otherwise we pass struct dn_sch * which is the first field in
842  * struct dn_schk so we can cast between the two. We use this trick
843  * because in the create phase (but it should be fixed).
844  */
845 static uint32_t
846 schk_hash(uintptr_t key, int flags, void *_arg)
847 {
848 	uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
849 		((struct dn_schk *)key)->sch.sched_nr;
850 	return ( (i>>8)^(i>>4)^i );
851 }
852 
853 static int
854 schk_match(void *obj, uintptr_t key, int flags, void *_arg)
855 {
856 	struct dn_schk *s = (struct dn_schk *)obj;
857 	int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
858 		((struct dn_schk *)key)->sch.sched_nr;
859 	return (s->sch.sched_nr == i);
860 }
861 
862 /*
863  * Create the entry and intialize with the sched hash if needed.
864  * Leave s->fp unset so we can tell whether a dn_ht_find() returns
865  * a new object or a previously existing one.
866  */
867 static void *
868 schk_new(uintptr_t key, int flags, void *arg)
869 {
870 	struct schk_new_arg *a = arg;
871 	struct dn_schk *s;
872 	int l = sizeof(*s) +a->fp->schk_datalen;
873 
874 	s = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
875 	if (s == NULL)
876 		return NULL;
877 	set_oid(&s->link.oid, DN_LINK, sizeof(s->link));
878 	s->sch = *a->sch; // copy initial values
879 	s->link.link_nr = s->sch.sched_nr;
880 	SLIST_INIT(&s->fsk_list);
881 	/* initialize the hash table or create the single instance */
882 	s->fp = a->fp;	/* si_new needs this */
883 	s->drain_bucket = 0;
884 	if (s->sch.flags & DN_HAVE_MASK) {
885 		s->siht = dn_ht_init(NULL, s->sch.buckets,
886 			offsetof(struct dn_sch_inst, si_next),
887 			si_hash, si_match, si_new);
888 		if (s->siht == NULL) {
889 			free(s, M_DUMMYNET);
890 			return NULL;
891 		}
892 	}
893 	s->fp = NULL;	/* mark as a new scheduler */
894 	V_dn_cfg.schk_count++;
895 	return s;
896 }
897 
898 /*
899  * Callback for sched delete. Notify all attached flowsets to
900  * detach from the scheduler, destroy the internal flowset, and
901  * all instances. The scheduler goes away too.
902  * arg is 0 (only detach flowsets and destroy instances)
903  * DN_DESTROY (detach & delete queues, delete schk)
904  * or DN_DELETE_FS (delete queues and flowsets, delete schk)
905  */
906 static int
907 schk_delete_cb(void *obj, void *arg)
908 {
909 	struct dn_schk *s = obj;
910 #if 0
911 	int a = (int)arg;
912 	ND("sched %d arg %s%s",
913 		s->sch.sched_nr,
914 		a&DN_DESTROY ? "DEL ":"",
915 		a&DN_DELETE_FS ? "DEL_FS":"");
916 #endif
917 	fsk_detach_list(&s->fsk_list, arg ? DN_DESTROY : 0);
918 	/* no more flowset pointing to us now */
919 	if (s->sch.flags & DN_HAVE_MASK) {
920 		dn_ht_scan(s->siht, si_destroy, NULL);
921 		dn_ht_free(s->siht, 0);
922 	} else if (s->siht)
923 		si_destroy(s->siht, NULL);
924 
925 	free(s->profile, M_DUMMYNET);
926 	s->profile = NULL;
927 	s->siht = NULL;
928 	if (s->fp->destroy)
929 		s->fp->destroy(s);
930 	bzero(s, sizeof(*s));	// safety
931 	free(obj, M_DUMMYNET);
932 	V_dn_cfg.schk_count--;
933 	return DNHT_SCAN_DEL;
934 }
935 
936 /*
937  * called on a 'sched X delete' command. Deletes a single scheduler.
938  * This is done by removing from the schedhash, unlinking all
939  * flowsets and deleting their traffic.
940  */
941 static int
942 delete_schk(int i)
943 {
944 	struct dn_schk *s;
945 
946 	s = dn_ht_find(V_dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
947 	ND("%d %p", i, s);
948 	if (!s)
949 		return EINVAL;
950 	delete_fs(i + DN_MAX_ID, 1); /* first delete internal fs */
951 	/* then detach flowsets, delete traffic */
952 	schk_delete_cb(s, (void*)(uintptr_t)DN_DESTROY);
953 	return 0;
954 }
955 /*--- end of schk hashtable support ---*/
956 
957 static int
958 copy_obj(char **start, char *end, void *_o, const char *msg, int i)
959 {
960 	struct dn_id o;
961 	union {
962 		struct dn_link l;
963 		struct dn_schk s;
964 	} dn;
965 	int have = end - *start;
966 
967 	memcpy(&o, _o, sizeof(o));
968 	if (have < o.len || o.len == 0 || o.type == 0) {
969 		D("(WARN) type %d %s %d have %d need %d",
970 		    o.type, msg, i, have, o.len);
971 		return 1;
972 	}
973 	ND("type %d %s %d len %d", o.type, msg, i, o.len);
974 	if (o.type == DN_LINK) {
975 		memcpy(&dn.l, _o, sizeof(dn.l));
976 		/* Adjust burst parameter for link */
977 		dn.l.burst = div64(dn.l.burst, 8 * hz);
978 		dn.l.delay = dn.l.delay * 1000 / hz;
979 		memcpy(*start, &dn.l, sizeof(dn.l));
980 	} else if (o.type == DN_SCH) {
981 		/* Set dn.s.sch.oid.id to the number of instances */
982 		memcpy(&dn.s, _o, sizeof(dn.s));
983 		dn.s.sch.oid.id = (dn.s.sch.flags & DN_HAVE_MASK) ?
984 		    dn_ht_entries(dn.s.siht) : (dn.s.siht ? 1 : 0);
985 		memcpy(*start, &dn.s, sizeof(dn.s));
986 	} else
987 		memcpy(*start, _o, o.len);
988 	*start += o.len;
989 	return 0;
990 }
991 
992 /* Specific function to copy a queue.
993  * Copies only the user-visible part of a queue (which is in
994  * a struct dn_flow), and sets len accordingly.
995  */
996 static int
997 copy_obj_q(char **start, char *end, void *_o, const char *msg, int i)
998 {
999 	struct dn_id *o = _o;
1000 	int have = end - *start;
1001 	int len = sizeof(struct dn_flow); /* see above comment */
1002 
1003 	if (have < len || o->len == 0 || o->type != DN_QUEUE) {
1004 		D("ERROR type %d %s %d have %d need %d",
1005 			o->type, msg, i, have, len);
1006 		return 1;
1007 	}
1008 	ND("type %d %s %d len %d", o->type, msg, i, len);
1009 	memcpy(*start, _o, len);
1010 	((struct dn_id*)(*start))->len = len;
1011 	*start += len;
1012 	return 0;
1013 }
1014 
1015 static int
1016 copy_q_cb(void *obj, void *arg)
1017 {
1018 	struct dn_queue *q = obj;
1019 	struct copy_args *a = arg;
1020 	struct dn_flow *ni = (struct dn_flow *)(*a->start);
1021         if (copy_obj_q(a->start, a->end, &q->ni, "queue", -1))
1022                 return DNHT_SCAN_END;
1023         ni->oid.type = DN_FLOW; /* override the DN_QUEUE */
1024         ni->oid.id = si_hash((uintptr_t)&ni->fid, 0, NULL);
1025         return 0;
1026 }
1027 
1028 static int
1029 copy_q(struct copy_args *a, struct dn_fsk *fs, int flags)
1030 {
1031 	if (!fs->qht)
1032 		return 0;
1033 	if (fs->fs.flags & DN_QHT_HASH)
1034 		dn_ht_scan(fs->qht, copy_q_cb, a);
1035 	else
1036 		copy_q_cb(fs->qht, a);
1037 	return 0;
1038 }
1039 
1040 /*
1041  * This routine only copies the initial part of a profile ? XXX
1042  */
1043 static int
1044 copy_profile(struct copy_args *a, struct dn_profile *p)
1045 {
1046 	int have = a->end - *a->start;
1047 	/* XXX here we check for max length */
1048 	int profile_len = sizeof(struct dn_profile) -
1049 		ED_MAX_SAMPLES_NO*sizeof(int);
1050 
1051 	if (p == NULL)
1052 		return 0;
1053 	if (have < profile_len) {
1054 		D("error have %d need %d", have, profile_len);
1055 		return 1;
1056 	}
1057 	memcpy(*a->start, p, profile_len);
1058 	((struct dn_id *)(*a->start))->len = profile_len;
1059 	*a->start += profile_len;
1060 	return 0;
1061 }
1062 
1063 static int
1064 copy_flowset(struct copy_args *a, struct dn_fsk *fs, int flags)
1065 {
1066 	struct dn_fs *ufs = (struct dn_fs *)(*a->start);
1067 	if (!fs)
1068 		return 0;
1069 	ND("flowset %d", fs->fs.fs_nr);
1070 	if (copy_obj(a->start, a->end, &fs->fs, "flowset", fs->fs.fs_nr))
1071 		return DNHT_SCAN_END;
1072 	ufs->oid.id = (fs->fs.flags & DN_QHT_HASH) ?
1073 		dn_ht_entries(fs->qht) : (fs->qht ? 1 : 0);
1074 	if (flags) {	/* copy queues */
1075 		copy_q(a, fs, 0);
1076 	}
1077 	return 0;
1078 }
1079 
1080 static int
1081 copy_si_cb(void *obj, void *arg)
1082 {
1083 	struct dn_sch_inst *si = obj;
1084 	struct copy_args *a = arg;
1085 	struct dn_flow *ni = (struct dn_flow *)(*a->start);
1086 	if (copy_obj(a->start, a->end, &si->ni, "inst",
1087 			si->sched->sch.sched_nr))
1088 		return DNHT_SCAN_END;
1089 	ni->oid.type = DN_FLOW; /* override the DN_SCH_I */
1090 	ni->oid.id = si_hash((uintptr_t)si, DNHT_KEY_IS_OBJ, NULL);
1091 	return 0;
1092 }
1093 
1094 static int
1095 copy_si(struct copy_args *a, struct dn_schk *s, int flags)
1096 {
1097 	if (s->sch.flags & DN_HAVE_MASK)
1098 		dn_ht_scan(s->siht, copy_si_cb, a);
1099 	else if (s->siht)
1100 		copy_si_cb(s->siht, a);
1101 	return 0;
1102 }
1103 
1104 /*
1105  * compute a list of children of a scheduler and copy up
1106  */
1107 static int
1108 copy_fsk_list(struct copy_args *a, struct dn_schk *s, int flags)
1109 {
1110 	struct dn_fsk *fs;
1111 	struct dn_id *o;
1112 	uint32_t *p;
1113 
1114 	int n = 0, space = sizeof(*o);
1115 	SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1116 		if (fs->fs.fs_nr < DN_MAX_ID)
1117 			n++;
1118 	}
1119 	space += n * sizeof(uint32_t);
1120 	DX(3, "sched %d has %d flowsets", s->sch.sched_nr, n);
1121 	if (a->end - *(a->start) < space)
1122 		return DNHT_SCAN_END;
1123 	o = (struct dn_id *)(*(a->start));
1124 	o->len = space;
1125 	*a->start += o->len;
1126 	o->type = DN_TEXT;
1127 	p = (uint32_t *)(o+1);
1128 	SLIST_FOREACH(fs, &s->fsk_list, sch_chain)
1129 		if (fs->fs.fs_nr < DN_MAX_ID)
1130 			*p++ = fs->fs.fs_nr;
1131 	return 0;
1132 }
1133 
1134 static int
1135 copy_data_helper(void *_o, void *_arg)
1136 {
1137 	struct copy_args *a = _arg;
1138 	uint32_t *r = a->extra->r; /* start of first range */
1139 	uint32_t *lim;	/* first invalid pointer */
1140 	int n;
1141 
1142 	lim = (uint32_t *)((char *)(a->extra) + a->extra->o.len);
1143 
1144 	if (a->type == DN_LINK || a->type == DN_SCH) {
1145 		/* pipe|sched show, we receive a dn_schk */
1146 		struct dn_schk *s = _o;
1147 
1148 		n = s->sch.sched_nr;
1149 		if (a->type == DN_SCH && n >= DN_MAX_ID)
1150 			return 0;	/* not a scheduler */
1151 		if (a->type == DN_LINK && n <= DN_MAX_ID)
1152 		    return 0;	/* not a pipe */
1153 
1154 		/* see if the object is within one of our ranges */
1155 		for (;r < lim; r += 2) {
1156 			if (n < r[0] || n > r[1])
1157 				continue;
1158 			/* Found a valid entry, copy and we are done */
1159 			if (a->flags & DN_C_LINK) {
1160 				if (copy_obj(a->start, a->end,
1161 				    &s->link, "link", n))
1162 					return DNHT_SCAN_END;
1163 				if (copy_profile(a, s->profile))
1164 					return DNHT_SCAN_END;
1165 				if (copy_flowset(a, s->fs, 0))
1166 					return DNHT_SCAN_END;
1167 			}
1168 			if (a->flags & DN_C_SCH) {
1169 				if (copy_obj(a->start, a->end,
1170 				    &s->sch, "sched", n))
1171 					return DNHT_SCAN_END;
1172 				/* list all attached flowsets */
1173 				if (copy_fsk_list(a, s, 0))
1174 					return DNHT_SCAN_END;
1175 			}
1176 			if (a->flags & DN_C_FLOW)
1177 				copy_si(a, s, 0);
1178 			break;
1179 		}
1180 	} else if (a->type == DN_FS) {
1181 		/* queue show, skip internal flowsets */
1182 		struct dn_fsk *fs = _o;
1183 
1184 		n = fs->fs.fs_nr;
1185 		if (n >= DN_MAX_ID)
1186 			return 0;
1187 		/* see if the object is within one of our ranges */
1188 		for (;r < lim; r += 2) {
1189 			if (n < r[0] || n > r[1])
1190 				continue;
1191 			if (copy_flowset(a, fs, 0))
1192 				return DNHT_SCAN_END;
1193 			copy_q(a, fs, 0);
1194 			break; /* we are done */
1195 		}
1196 	}
1197 	return 0;
1198 }
1199 
1200 static inline struct dn_schk *
1201 locate_scheduler(int i)
1202 {
1203 	return dn_ht_find(V_dn_cfg.schedhash, i, 0, NULL);
1204 }
1205 
1206 /*
1207  * red parameters are in fixed point arithmetic.
1208  */
1209 static int
1210 config_red(struct dn_fsk *fs)
1211 {
1212 	int64_t s, idle, weight, w0;
1213 	int t, i;
1214 
1215 	fs->w_q = fs->fs.w_q;
1216 	fs->max_p = fs->fs.max_p;
1217 	ND("called");
1218 	/* Doing stuff that was in userland */
1219 	i = fs->sched->link.bandwidth;
1220 	s = (i <= 0) ? 0 :
1221 		hz * V_dn_cfg.red_avg_pkt_size * 8 * SCALE(1) / i;
1222 
1223 	idle = div64((s * 3) , fs->w_q); /* s, fs->w_q scaled; idle not scaled */
1224 	fs->lookup_step = div64(idle , V_dn_cfg.red_lookup_depth);
1225 	/* fs->lookup_step not scaled, */
1226 	if (!fs->lookup_step)
1227 		fs->lookup_step = 1;
1228 	w0 = weight = SCALE(1) - fs->w_q; //fs->w_q scaled
1229 
1230 	for (t = fs->lookup_step; t > 1; --t)
1231 		weight = SCALE_MUL(weight, w0);
1232 	fs->lookup_weight = (int)(weight); // scaled
1233 
1234 	/* Now doing stuff that was in kerneland */
1235 	fs->min_th = SCALE(fs->fs.min_th);
1236 	fs->max_th = SCALE(fs->fs.max_th);
1237 
1238 	if (fs->fs.max_th == fs->fs.min_th)
1239 		fs->c_1 = fs->max_p;
1240 	else
1241 		fs->c_1 = SCALE((int64_t)(fs->max_p)) / (fs->fs.max_th - fs->fs.min_th);
1242 	fs->c_2 = SCALE_MUL(fs->c_1, SCALE(fs->fs.min_th));
1243 
1244 	if (fs->fs.flags & DN_IS_GENTLE_RED) {
1245 		fs->c_3 = (SCALE(1) - fs->max_p) / fs->fs.max_th;
1246 		fs->c_4 = SCALE(1) - 2 * fs->max_p;
1247 	}
1248 
1249 	/* If the lookup table already exist, free and create it again. */
1250 	free(fs->w_q_lookup, M_DUMMYNET);
1251 	fs->w_q_lookup = NULL;
1252 	if (V_dn_cfg.red_lookup_depth == 0) {
1253 		printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1254 		    "must be > 0\n");
1255 		fs->fs.flags &= ~DN_IS_RED;
1256 		fs->fs.flags &= ~DN_IS_GENTLE_RED;
1257 		return (EINVAL);
1258 	}
1259 	fs->lookup_depth = V_dn_cfg.red_lookup_depth;
1260 	fs->w_q_lookup = (u_int *)malloc(fs->lookup_depth * sizeof(int),
1261 	    M_DUMMYNET, M_NOWAIT);
1262 	if (fs->w_q_lookup == NULL) {
1263 		printf("dummynet: sorry, cannot allocate red lookup table\n");
1264 		fs->fs.flags &= ~DN_IS_RED;
1265 		fs->fs.flags &= ~DN_IS_GENTLE_RED;
1266 		return(ENOSPC);
1267 	}
1268 
1269 	/* Fill the lookup table with (1 - w_q)^x */
1270 	fs->w_q_lookup[0] = SCALE(1) - fs->w_q;
1271 
1272 	for (i = 1; i < fs->lookup_depth; i++)
1273 		fs->w_q_lookup[i] =
1274 		    SCALE_MUL(fs->w_q_lookup[i - 1], fs->lookup_weight);
1275 
1276 	if (V_dn_cfg.red_avg_pkt_size < 1)
1277 		V_dn_cfg.red_avg_pkt_size = 512;
1278 	fs->avg_pkt_size = V_dn_cfg.red_avg_pkt_size;
1279 	if (V_dn_cfg.red_max_pkt_size < 1)
1280 		V_dn_cfg.red_max_pkt_size = 1500;
1281 	fs->max_pkt_size = V_dn_cfg.red_max_pkt_size;
1282 	ND("exit");
1283 	return 0;
1284 }
1285 
1286 /* Scan all flowset attached to this scheduler and update red */
1287 static void
1288 update_red(struct dn_schk *s)
1289 {
1290 	struct dn_fsk *fs;
1291 	SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1292 		if (fs && (fs->fs.flags & DN_IS_RED))
1293 			config_red(fs);
1294 	}
1295 }
1296 
1297 /* attach flowset to scheduler s, possibly requeue */
1298 static void
1299 fsk_attach(struct dn_fsk *fs, struct dn_schk *s)
1300 {
1301 	ND("remove fs %d from fsunlinked, link to sched %d",
1302 		fs->fs.fs_nr, s->sch.sched_nr);
1303 	SLIST_REMOVE(&V_dn_cfg.fsu, fs, dn_fsk, sch_chain);
1304 	fs->sched = s;
1305 	SLIST_INSERT_HEAD(&s->fsk_list, fs, sch_chain);
1306 	if (s->fp->new_fsk)
1307 		s->fp->new_fsk(fs);
1308 	/* XXX compute fsk_mask */
1309 	fs->fsk_mask = fs->fs.flow_mask;
1310 	if (fs->sched->sch.flags & DN_HAVE_MASK)
1311 		flow_id_or(&fs->sched->sch.sched_mask, &fs->fsk_mask);
1312 	if (fs->qht) {
1313 		/*
1314 		 * we must drain qht according to the old
1315 		 * type, and reinsert according to the new one.
1316 		 * The requeue is complex -- in general we need to
1317 		 * reclassify every single packet.
1318 		 * For the time being, let's hope qht is never set
1319 		 * when we reach this point.
1320 		 */
1321 		D("XXX TODO requeue from fs %d to sch %d",
1322 			fs->fs.fs_nr, s->sch.sched_nr);
1323 		fs->qht = NULL;
1324 	}
1325 	/* set the new type for qht */
1326 	if (nonzero_mask(&fs->fsk_mask))
1327 		fs->fs.flags |= DN_QHT_HASH;
1328 	else
1329 		fs->fs.flags &= ~DN_QHT_HASH;
1330 
1331 	/* XXX config_red() can fail... */
1332 	if (fs->fs.flags & DN_IS_RED)
1333 		config_red(fs);
1334 }
1335 
1336 /* update all flowsets which may refer to this scheduler */
1337 static void
1338 update_fs(struct dn_schk *s)
1339 {
1340 	struct dn_fsk *fs, *tmp;
1341 
1342 	SLIST_FOREACH_SAFE(fs, &V_dn_cfg.fsu, sch_chain, tmp) {
1343 		if (s->sch.sched_nr != fs->fs.sched_nr) {
1344 			D("fs %d for sch %d not %d still unlinked",
1345 				fs->fs.fs_nr, fs->fs.sched_nr,
1346 				s->sch.sched_nr);
1347 			continue;
1348 		}
1349 		fsk_attach(fs, s);
1350 	}
1351 }
1352 
1353 #ifdef NEW_AQM
1354 /* Retrieve AQM configurations to ipfw userland
1355  */
1356 static int
1357 get_aqm_parms(struct sockopt *sopt)
1358 {
1359 	struct dn_extra_parms  *ep;
1360 	struct dn_fsk *fs;
1361 	size_t sopt_valsize;
1362 	int l, err = 0;
1363 
1364 	sopt_valsize = sopt->sopt_valsize;
1365 	l = sizeof(*ep);
1366 	if (sopt->sopt_valsize < l) {
1367 		D("bad len sopt->sopt_valsize %d len %d",
1368 			(int) sopt->sopt_valsize , l);
1369 		err = EINVAL;
1370 		return err;
1371 	}
1372 	ep = malloc(l, M_DUMMYNET, M_NOWAIT);
1373 	if(!ep) {
1374 		err = ENOMEM ;
1375 		return err;
1376 	}
1377 	do {
1378 		err = sooptcopyin(sopt, ep, l, l);
1379 		if(err)
1380 			break;
1381 		sopt->sopt_valsize = sopt_valsize;
1382 		if (ep->oid.len < l) {
1383 			err = EINVAL;
1384 			break;
1385 		}
1386 
1387 		fs = dn_ht_find(V_dn_cfg.fshash, ep->nr, 0, NULL);
1388 		if (!fs) {
1389 			D("fs %d not found", ep->nr);
1390 			err = EINVAL;
1391 			break;
1392 		}
1393 
1394 		if (fs->aqmfp && fs->aqmfp->getconfig) {
1395 			if(fs->aqmfp->getconfig(fs, ep)) {
1396 				D("Error while trying to get AQM params");
1397 				err = EINVAL;
1398 				break;
1399 			}
1400 			ep->oid.len = l;
1401 			err = sooptcopyout(sopt, ep, l);
1402 		}
1403 	}while(0);
1404 
1405 	free(ep, M_DUMMYNET);
1406 	return err;
1407 }
1408 
1409 /* Retrieve AQM configurations to ipfw userland
1410  */
1411 static int
1412 get_sched_parms(struct sockopt *sopt)
1413 {
1414 	struct dn_extra_parms  *ep;
1415 	struct dn_schk *schk;
1416 	size_t sopt_valsize;
1417 	int l, err = 0;
1418 
1419 	sopt_valsize = sopt->sopt_valsize;
1420 	l = sizeof(*ep);
1421 	if (sopt->sopt_valsize < l) {
1422 		D("bad len sopt->sopt_valsize %d len %d",
1423 			(int) sopt->sopt_valsize , l);
1424 		err = EINVAL;
1425 		return err;
1426 	}
1427 	ep = malloc(l, M_DUMMYNET, M_NOWAIT);
1428 	if(!ep) {
1429 		err = ENOMEM ;
1430 		return err;
1431 	}
1432 	do {
1433 		err = sooptcopyin(sopt, ep, l, l);
1434 		if(err)
1435 			break;
1436 		sopt->sopt_valsize = sopt_valsize;
1437 		if (ep->oid.len < l) {
1438 			err = EINVAL;
1439 			break;
1440 		}
1441 
1442 		schk = locate_scheduler(ep->nr);
1443 		if (!schk) {
1444 			D("sched %d not found", ep->nr);
1445 			err = EINVAL;
1446 			break;
1447 		}
1448 
1449 		if (schk->fp && schk->fp->getconfig) {
1450 			if(schk->fp->getconfig(schk, ep)) {
1451 				D("Error while trying to get sched params");
1452 				err = EINVAL;
1453 				break;
1454 			}
1455 			ep->oid.len = l;
1456 			err = sooptcopyout(sopt, ep, l);
1457 		}
1458 	}while(0);
1459 	free(ep, M_DUMMYNET);
1460 
1461 	return err;
1462 }
1463 
1464 /* Configure AQM for flowset 'fs'.
1465  * extra parameters are passed from userland.
1466  */
1467 static int
1468 config_aqm(struct dn_fsk *fs, struct  dn_extra_parms *ep, int busy)
1469 {
1470 	int err = 0;
1471 
1472 	NET_EPOCH_ASSERT();
1473 
1474 	do {
1475 		/* no configurations */
1476 		if (!ep) {
1477 			err = 0;
1478 			break;
1479 		}
1480 
1481 		/* no AQM for this flowset*/
1482 		if (!strcmp(ep->name,"")) {
1483 			err = 0;
1484 			break;
1485 		}
1486 		if (ep->oid.len < sizeof(*ep)) {
1487 			D("short aqm len %d", ep->oid.len);
1488 				err = EINVAL;
1489 				break;
1490 		}
1491 
1492 		if (busy) {
1493 			D("Unable to configure flowset, flowset busy!");
1494 			err = EINVAL;
1495 			break;
1496 		}
1497 
1498 		/* deconfigure old aqm if exist */
1499 		if (fs->aqmcfg && fs->aqmfp && fs->aqmfp->deconfig) {
1500 			aqm_cleanup_deconfig_fs(fs);
1501 		}
1502 
1503 		if (!(fs->aqmfp = find_aqm_type(0, ep->name))) {
1504 			D("AQM functions not found for type %s!", ep->name);
1505 			fs->fs.flags &= ~DN_IS_AQM;
1506 			err = EINVAL;
1507 			break;
1508 		} else
1509 			fs->fs.flags |= DN_IS_AQM;
1510 
1511 		if (ep->oid.subtype != DN_AQM_PARAMS) {
1512 				D("Wrong subtype");
1513 				err = EINVAL;
1514 				break;
1515 		}
1516 
1517 		if (fs->aqmfp->config) {
1518 			err = fs->aqmfp->config(fs, ep, ep->oid.len);
1519 			if (err) {
1520 					D("Unable to configure AQM for FS %d", fs->fs.fs_nr );
1521 					fs->fs.flags &= ~DN_IS_AQM;
1522 					fs->aqmfp = NULL;
1523 					break;
1524 			}
1525 		}
1526 	} while(0);
1527 
1528 	return err;
1529 }
1530 #endif
1531 
1532 /*
1533  * Configuration -- to preserve backward compatibility we use
1534  * the following scheme (N is 65536)
1535  *	NUMBER		SCHED	LINK	FLOWSET
1536  *	   1 ..  N-1	(1)WFQ	(2)WFQ	(3)queue
1537  *	 N+1 .. 2N-1	(4)FIFO (5)FIFO	(6)FIFO for sched 1..N-1
1538  *	2N+1 .. 3N-1	--	--	(7)FIFO for sched N+1..2N-1
1539  *
1540  * "pipe i config" configures #1, #2 and #3
1541  * "sched i config" configures #1 and possibly #6
1542  * "queue i config" configures #3
1543  * #1 is configured with 'pipe i config' or 'sched i config'
1544  * #2 is configured with 'pipe i config', and created if not
1545  *	existing with 'sched i config'
1546  * #3 is configured with 'queue i config'
1547  * #4 is automatically configured after #1, can only be FIFO
1548  * #5 is automatically configured after #2
1549  * #6 is automatically created when #1 is !MULTIQUEUE,
1550  *	and can be updated.
1551  * #7 is automatically configured after #2
1552  */
1553 
1554 /*
1555  * configure a link (and its FIFO instance)
1556  */
1557 static int
1558 config_link(struct dn_link *p, struct dn_id *arg)
1559 {
1560 	int i;
1561 
1562 	if (p->oid.len != sizeof(*p)) {
1563 		D("invalid pipe len %d", p->oid.len);
1564 		return EINVAL;
1565 	}
1566 	i = p->link_nr;
1567 	if (i <= 0 || i >= DN_MAX_ID)
1568 		return EINVAL;
1569 	/*
1570 	 * The config program passes parameters as follows:
1571 	 * bw = bits/second (0 means no limits),
1572 	 * delay = ms, must be translated into ticks.
1573 	 * qsize = slots/bytes
1574 	 * burst ???
1575 	 */
1576 	p->delay = (p->delay * hz) / 1000;
1577 	/* Scale burst size: bytes -> bits * hz */
1578 	p->burst *= 8 * hz;
1579 
1580 	DN_BH_WLOCK();
1581 	/* do it twice, base link and FIFO link */
1582 	for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1583 	    struct dn_schk *s = locate_scheduler(i);
1584 	    if (s == NULL) {
1585 		DN_BH_WUNLOCK();
1586 		D("sched %d not found", i);
1587 		return EINVAL;
1588 	    }
1589 	    /* remove profile if exists */
1590 	    free(s->profile, M_DUMMYNET);
1591 	    s->profile = NULL;
1592 
1593 	    /* copy all parameters */
1594 	    s->link.oid = p->oid;
1595 	    s->link.link_nr = i;
1596 	    s->link.delay = p->delay;
1597 	    if (s->link.bandwidth != p->bandwidth) {
1598 		/* XXX bandwidth changes, need to update red params */
1599 	    s->link.bandwidth = p->bandwidth;
1600 		update_red(s);
1601 	    }
1602 	    s->link.burst = p->burst;
1603 	    schk_reset_credit(s);
1604 	}
1605 	V_dn_cfg.id++;
1606 	DN_BH_WUNLOCK();
1607 	return 0;
1608 }
1609 
1610 /*
1611  * configure a flowset. Can be called from inside with locked=1,
1612  */
1613 static struct dn_fsk *
1614 config_fs(struct dn_fs *nfs, struct dn_id *arg, int locked)
1615 {
1616 	int i;
1617 	struct dn_fsk *fs;
1618 #ifdef NEW_AQM
1619 	struct dn_extra_parms *ep;
1620 #endif
1621 
1622 	if (nfs->oid.len != sizeof(*nfs)) {
1623 		D("invalid flowset len %d", nfs->oid.len);
1624 		return NULL;
1625 	}
1626 	i = nfs->fs_nr;
1627 	if (i <= 0 || i >= 3*DN_MAX_ID)
1628 		return NULL;
1629 #ifdef NEW_AQM
1630 	ep = NULL;
1631 	if (arg != NULL) {
1632 		ep = malloc(sizeof(*ep), M_TEMP, M_NOWAIT);
1633 		if (ep == NULL)
1634 			return (NULL);
1635 		memcpy(ep, arg, sizeof(*ep));
1636 	}
1637 #endif
1638 	ND("flowset %d", i);
1639 	/* XXX other sanity checks */
1640         if (nfs->flags & DN_QSIZE_BYTES) {
1641 		ipdn_bound_var(&nfs->qsize, 16384,
1642 		    1500, V_dn_cfg.byte_limit, NULL); // "queue byte size");
1643         } else {
1644 		ipdn_bound_var(&nfs->qsize, 50,
1645 		    1, V_dn_cfg.slot_limit, NULL); // "queue slot size");
1646         }
1647 	if (nfs->flags & DN_HAVE_MASK) {
1648 		/* make sure we have some buckets */
1649 		ipdn_bound_var((int *)&nfs->buckets, V_dn_cfg.hash_size,
1650 			1, V_dn_cfg.max_hash_size, "flowset buckets");
1651 	} else {
1652 		nfs->buckets = 1;	/* we only need 1 */
1653 	}
1654 	if (!locked)
1655 		DN_BH_WLOCK();
1656 	do { /* exit with break when done */
1657 	    struct dn_schk *s;
1658 	    int flags = nfs->sched_nr ? DNHT_INSERT : 0;
1659 	    int j;
1660 	    int oldc = V_dn_cfg.fsk_count;
1661 	    fs = dn_ht_find(V_dn_cfg.fshash, i, flags, NULL);
1662 	    if (fs == NULL) {
1663 		D("missing sched for flowset %d", i);
1664 	        break;
1665 	    }
1666 	    /* grab some defaults from the existing one */
1667 	    if (nfs->sched_nr == 0) /* reuse */
1668 		nfs->sched_nr = fs->fs.sched_nr;
1669 	    for (j = 0; j < sizeof(nfs->par)/sizeof(nfs->par[0]); j++) {
1670 		if (nfs->par[j] == -1) /* reuse */
1671 		    nfs->par[j] = fs->fs.par[j];
1672 	    }
1673 	    if (bcmp(&fs->fs, nfs, sizeof(*nfs)) == 0) {
1674 		ND("flowset %d unchanged", i);
1675 #ifdef NEW_AQM
1676 		if (ep != NULL) {
1677 			/*
1678 			 * Reconfigure AQM as the parameters can be changed.
1679 			 * We consider the flowset as busy if it has scheduler
1680 			 * instance(s).
1681 			 */
1682 			s = locate_scheduler(nfs->sched_nr);
1683 			config_aqm(fs, ep, s != NULL && s->siht != NULL);
1684 		}
1685 #endif
1686 		break; /* no change, nothing to do */
1687 	    }
1688 	    if (oldc != V_dn_cfg.fsk_count)	/* new item */
1689 		V_dn_cfg.id++;
1690 	    s = locate_scheduler(nfs->sched_nr);
1691 	    /* detach from old scheduler if needed, preserving
1692 	     * queues if we need to reattach. Then update the
1693 	     * configuration, and possibly attach to the new sched.
1694 	     */
1695 	    DX(2, "fs %d changed sched %d@%p to %d@%p",
1696 		fs->fs.fs_nr,
1697 		fs->fs.sched_nr, fs->sched, nfs->sched_nr, s);
1698 	    if (fs->sched) {
1699 		int flags = s ? DN_DETACH : (DN_DETACH | DN_DESTROY);
1700 		flags |= DN_DESTROY; /* XXX temporary */
1701 		fsk_detach(fs, flags);
1702 	    }
1703 	    fs->fs = *nfs; /* copy configuration */
1704 #ifdef NEW_AQM
1705 			fs->aqmfp = NULL;
1706 			if (ep != NULL)
1707 				config_aqm(fs, ep, s != NULL &&
1708 				    s->siht != NULL);
1709 #endif
1710 	    if (s != NULL)
1711 		fsk_attach(fs, s);
1712 	} while (0);
1713 	if (!locked)
1714 		DN_BH_WUNLOCK();
1715 #ifdef NEW_AQM
1716 	free(ep, M_TEMP);
1717 #endif
1718 	return fs;
1719 }
1720 
1721 /*
1722  * config/reconfig a scheduler and its FIFO variant.
1723  * For !MULTIQUEUE schedulers, also set up the flowset.
1724  *
1725  * On reconfigurations (detected because s->fp is set),
1726  * detach existing flowsets preserving traffic, preserve link,
1727  * and delete the old scheduler creating a new one.
1728  */
1729 static int
1730 config_sched(struct dn_sch *_nsch, struct dn_id *arg)
1731 {
1732 	struct dn_schk *s;
1733 	struct schk_new_arg a; /* argument for schk_new */
1734 	int i;
1735 	struct dn_link p;	/* copy of oldlink */
1736 	struct dn_profile *pf = NULL;	/* copy of old link profile */
1737 	/* Used to preserv mask parameter */
1738 	struct ipfw_flow_id new_mask;
1739 	int new_buckets = 0;
1740 	int new_flags = 0;
1741 	int pipe_cmd;
1742 	int err = ENOMEM;
1743 
1744 	NET_EPOCH_ASSERT();
1745 
1746 	a.sch = _nsch;
1747 	if (a.sch->oid.len != sizeof(*a.sch)) {
1748 		D("bad sched len %d", a.sch->oid.len);
1749 		return EINVAL;
1750 	}
1751 	i = a.sch->sched_nr;
1752 	if (i <= 0 || i >= DN_MAX_ID)
1753 		return EINVAL;
1754 	/* make sure we have some buckets */
1755 	if (a.sch->flags & DN_HAVE_MASK)
1756 		ipdn_bound_var((int *)&a.sch->buckets, V_dn_cfg.hash_size,
1757 			1, V_dn_cfg.max_hash_size, "sched buckets");
1758 	/* XXX other sanity checks */
1759 	bzero(&p, sizeof(p));
1760 
1761 	pipe_cmd = a.sch->flags & DN_PIPE_CMD;
1762 	a.sch->flags &= ~DN_PIPE_CMD; //XXX do it even if is not set?
1763 	if (pipe_cmd) {
1764 		/* Copy mask parameter */
1765 		new_mask = a.sch->sched_mask;
1766 		new_buckets = a.sch->buckets;
1767 		new_flags = a.sch->flags;
1768 	}
1769 	DN_BH_WLOCK();
1770 again: /* run twice, for wfq and fifo */
1771 	/*
1772 	 * lookup the type. If not supplied, use the previous one
1773 	 * or default to WF2Q+. Otherwise, return an error.
1774 	 */
1775 	V_dn_cfg.id++;
1776 	a.fp = find_sched_type(a.sch->oid.subtype, a.sch->name);
1777 	if (a.fp != NULL) {
1778 		/* found. Lookup or create entry */
1779 		s = dn_ht_find(V_dn_cfg.schedhash, i, DNHT_INSERT, &a);
1780 	} else if (a.sch->oid.subtype == 0 && !a.sch->name[0]) {
1781 		/* No type. search existing s* or retry with WF2Q+ */
1782 		s = dn_ht_find(V_dn_cfg.schedhash, i, 0, &a);
1783 		if (s != NULL) {
1784 			a.fp = s->fp;
1785 			/* Scheduler exists, skip to FIFO scheduler
1786 			 * if command was pipe config...
1787 			 */
1788 			if (pipe_cmd)
1789 				goto next;
1790 		} else {
1791 			/* New scheduler, create a wf2q+ with no mask
1792 			 * if command was pipe config...
1793 			 */
1794 			if (pipe_cmd) {
1795 				/* clear mask parameter */
1796 				bzero(&a.sch->sched_mask, sizeof(new_mask));
1797 				a.sch->buckets = 0;
1798 				a.sch->flags &= ~DN_HAVE_MASK;
1799 			}
1800 			a.sch->oid.subtype = DN_SCHED_WF2QP;
1801 			goto again;
1802 		}
1803 	} else {
1804 		D("invalid scheduler type %d %s",
1805 			a.sch->oid.subtype, a.sch->name);
1806 		err = EINVAL;
1807 		goto error;
1808 	}
1809 	/* normalize name and subtype */
1810 	a.sch->oid.subtype = a.fp->type;
1811 	bzero(a.sch->name, sizeof(a.sch->name));
1812 	strlcpy(a.sch->name, a.fp->name, sizeof(a.sch->name));
1813 	if (s == NULL) {
1814 		D("cannot allocate scheduler %d", i);
1815 		goto error;
1816 	}
1817 	/* restore existing link if any */
1818 	if (p.link_nr) {
1819 		s->link = p;
1820 		if (!pf || pf->link_nr != p.link_nr) { /* no saved value */
1821 			s->profile = NULL; /* XXX maybe not needed */
1822 		} else {
1823 			s->profile = malloc(sizeof(struct dn_profile),
1824 					     M_DUMMYNET, M_NOWAIT | M_ZERO);
1825 			if (s->profile == NULL) {
1826 				D("cannot allocate profile");
1827 				goto error; //XXX
1828 			}
1829 			memcpy(s->profile, pf, sizeof(*pf));
1830 		}
1831 	}
1832 	p.link_nr = 0;
1833 	if (s->fp == NULL) {
1834 		DX(2, "sched %d new type %s", i, a.fp->name);
1835 	} else if (s->fp != a.fp ||
1836 			bcmp(a.sch, &s->sch, sizeof(*a.sch)) ) {
1837 		/* already existing. */
1838 		DX(2, "sched %d type changed from %s to %s",
1839 			i, s->fp->name, a.fp->name);
1840 		DX(4, "   type/sub %d/%d -> %d/%d",
1841 			s->sch.oid.type, s->sch.oid.subtype,
1842 			a.sch->oid.type, a.sch->oid.subtype);
1843 		if (s->link.link_nr == 0)
1844 			D("XXX WARNING link 0 for sched %d", i);
1845 		p = s->link;	/* preserve link */
1846 		if (s->profile) {/* preserve profile */
1847 			if (!pf)
1848 				pf = malloc(sizeof(*pf),
1849 				    M_DUMMYNET, M_NOWAIT | M_ZERO);
1850 			if (pf)	/* XXX should issue a warning otherwise */
1851 				memcpy(pf, s->profile, sizeof(*pf));
1852 		}
1853 		/* remove from the hash */
1854 		dn_ht_find(V_dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
1855 		/* Detach flowsets, preserve queues. */
1856 		// schk_delete_cb(s, NULL);
1857 		// XXX temporarily, kill queues
1858 		schk_delete_cb(s, (void *)DN_DESTROY);
1859 		goto again;
1860 	} else {
1861 		DX(4, "sched %d unchanged type %s", i, a.fp->name);
1862 	}
1863 	/* complete initialization */
1864 	s->sch = *a.sch;
1865 	s->fp = a.fp;
1866 	s->cfg = arg;
1867 	// XXX schk_reset_credit(s);
1868 	/* create the internal flowset if needed,
1869 	 * trying to reuse existing ones if available
1870 	 */
1871 	if (!(s->fp->flags & DN_MULTIQUEUE) && !s->fs) {
1872 	        s->fs = dn_ht_find(V_dn_cfg.fshash, i, 0, NULL);
1873 		if (!s->fs) {
1874 			struct dn_fs fs;
1875 			bzero(&fs, sizeof(fs));
1876 			set_oid(&fs.oid, DN_FS, sizeof(fs));
1877 			fs.fs_nr = i + DN_MAX_ID;
1878 			fs.sched_nr = i;
1879 			s->fs = config_fs(&fs, NULL, 1 /* locked */);
1880 		}
1881 		if (!s->fs) {
1882 			schk_delete_cb(s, (void *)DN_DESTROY);
1883 			D("error creating internal fs for %d", i);
1884 			goto error;
1885 		}
1886 	}
1887 	/* call init function after the flowset is created */
1888 	if (s->fp->config)
1889 		s->fp->config(s);
1890 	update_fs(s);
1891 next:
1892 	if (i < DN_MAX_ID) { /* now configure the FIFO instance */
1893 		i += DN_MAX_ID;
1894 		if (pipe_cmd) {
1895 			/* Restore mask parameter for FIFO */
1896 			a.sch->sched_mask = new_mask;
1897 			a.sch->buckets = new_buckets;
1898 			a.sch->flags = new_flags;
1899 		} else {
1900 			/* sched config shouldn't modify the FIFO scheduler */
1901 			if (dn_ht_find(V_dn_cfg.schedhash, i, 0, &a) != NULL) {
1902 				/* FIFO already exist, don't touch it */
1903 				err = 0; /* and this is not an error */
1904 				goto error;
1905 			}
1906 		}
1907 		a.sch->sched_nr = i;
1908 		a.sch->oid.subtype = DN_SCHED_FIFO;
1909 		bzero(a.sch->name, sizeof(a.sch->name));
1910 		goto again;
1911 	}
1912 	err = 0;
1913 error:
1914 	DN_BH_WUNLOCK();
1915 	free(pf, M_DUMMYNET);
1916 	return err;
1917 }
1918 
1919 /*
1920  * attach a profile to a link
1921  */
1922 static int
1923 config_profile(struct dn_profile *pf, struct dn_id *arg)
1924 {
1925 	struct dn_schk *s;
1926 	int i, olen, err = 0;
1927 
1928 	if (pf->oid.len < sizeof(*pf)) {
1929 		D("short profile len %d", pf->oid.len);
1930 		return EINVAL;
1931 	}
1932 	i = pf->link_nr;
1933 	if (i <= 0 || i >= DN_MAX_ID)
1934 		return EINVAL;
1935 	/* XXX other sanity checks */
1936 	DN_BH_WLOCK();
1937 	for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1938 		s = locate_scheduler(i);
1939 
1940 		if (s == NULL) {
1941 			err = EINVAL;
1942 			break;
1943 		}
1944 		V_dn_cfg.id++;
1945 		/*
1946 		 * If we had a profile and the new one does not fit,
1947 		 * or it is deleted, then we need to free memory.
1948 		 */
1949 		if (s->profile && (pf->samples_no == 0 ||
1950 		    s->profile->oid.len < pf->oid.len)) {
1951 			free(s->profile, M_DUMMYNET);
1952 			s->profile = NULL;
1953 		}
1954 		if (pf->samples_no == 0)
1955 			continue;
1956 		/*
1957 		 * new profile, possibly allocate memory
1958 		 * and copy data.
1959 		 */
1960 		if (s->profile == NULL)
1961 			s->profile = malloc(pf->oid.len,
1962 				    M_DUMMYNET, M_NOWAIT | M_ZERO);
1963 		if (s->profile == NULL) {
1964 			D("no memory for profile %d", i);
1965 			err = ENOMEM;
1966 			break;
1967 		}
1968 		/* preserve larger length XXX double check */
1969 		olen = s->profile->oid.len;
1970 		if (olen < pf->oid.len)
1971 			olen = pf->oid.len;
1972 		memcpy(s->profile, pf, pf->oid.len);
1973 		s->profile->oid.len = olen;
1974 	}
1975 	DN_BH_WUNLOCK();
1976 	return err;
1977 }
1978 
1979 /*
1980  * Delete all objects:
1981  */
1982 static void
1983 dummynet_flush(void)
1984 {
1985 
1986 	/* delete all schedulers and related links/queues/flowsets */
1987 	dn_ht_scan(V_dn_cfg.schedhash, schk_delete_cb,
1988 		(void *)(uintptr_t)DN_DELETE_FS);
1989 	/* delete all remaining (unlinked) flowsets */
1990 	DX(4, "still %d unlinked fs", V_dn_cfg.fsk_count);
1991 	dn_ht_free(V_dn_cfg.fshash, DNHT_REMOVE);
1992 	fsk_detach_list(&V_dn_cfg.fsu, DN_DELETE_FS);
1993 	/* Reinitialize system heap... */
1994 	heap_init(&V_dn_cfg.evheap, 16, offsetof(struct dn_id, id));
1995 }
1996 
1997 /*
1998  * Main handler for configuration. We are guaranteed to be called
1999  * with an oid which is at least a dn_id.
2000  * - the first object is the command (config, delete, flush, ...)
2001  * - config_link must be issued after the corresponding config_sched
2002  * - parameters (DN_TXT) for an object must precede the object
2003  *   processed on a config_sched.
2004  */
2005 int
2006 do_config(void *p, size_t l)
2007 {
2008 	struct dn_id o;
2009 	union {
2010 		struct dn_profile profile;
2011 		struct dn_fs fs;
2012 		struct dn_link link;
2013 		struct dn_sch sched;
2014 	} *dn;
2015 	struct dn_id *arg;
2016 	uintptr_t a;
2017 	int err, err2, off;
2018 
2019 	memcpy(&o, p, sizeof(o));
2020 	if (o.id != DN_API_VERSION) {
2021 		D("invalid api version got %d need %d", o.id, DN_API_VERSION);
2022 		return EINVAL;
2023 	}
2024 	arg = NULL;
2025 	dn = NULL;
2026 	off = 0;
2027 	while (l >= sizeof(o)) {
2028 		memcpy(&o, (char *)p + off, sizeof(o));
2029 		if (o.len < sizeof(o) || l < o.len) {
2030 			D("bad len o.len %d len %zu", o.len, l);
2031 			err = EINVAL;
2032 			break;
2033 		}
2034 		l -= o.len;
2035 		err = 0;
2036 		switch (o.type) {
2037 		default:
2038 			D("cmd %d not implemented", o.type);
2039 			break;
2040 
2041 #ifdef EMULATE_SYSCTL
2042 		/* sysctl emulation.
2043 		 * if we recognize the command, jump to the correct
2044 		 * handler and return
2045 		 */
2046 		case DN_SYSCTL_SET:
2047 			err = kesysctl_emu_set(p, l);
2048 			return err;
2049 #endif
2050 
2051 		case DN_CMD_CONFIG: /* simply a header */
2052 			break;
2053 
2054 		case DN_CMD_DELETE:
2055 			/* the argument is in the first uintptr_t after o */
2056 			if (o.len < sizeof(o) + sizeof(a)) {
2057 				err = EINVAL;
2058 				break;
2059 			}
2060 			memcpy(&a, (char *)p + off + sizeof(o), sizeof(a));
2061 			switch (o.subtype) {
2062 			case DN_LINK:
2063 				/* delete base and derived schedulers */
2064 				DN_BH_WLOCK();
2065 				err = delete_schk(a);
2066 				err2 = delete_schk(a + DN_MAX_ID);
2067 				DN_BH_WUNLOCK();
2068 				if (!err)
2069 					err = err2;
2070 				break;
2071 
2072 			default:
2073 				D("invalid delete type %d", o.subtype);
2074 				err = EINVAL;
2075 				break;
2076 
2077 			case DN_FS:
2078 				err = (a < 1 || a >= DN_MAX_ID) ?
2079 				    EINVAL : delete_fs(a, 0) ;
2080 				break;
2081 			}
2082 			break;
2083 
2084 		case DN_CMD_FLUSH:
2085 			DN_BH_WLOCK();
2086 			dummynet_flush();
2087 			DN_BH_WUNLOCK();
2088 			break;
2089 		case DN_TEXT:	/* store argument of next block */
2090 			free(arg, M_TEMP);
2091 			arg = malloc(o.len, M_TEMP, M_NOWAIT);
2092 			if (arg == NULL) {
2093 				err = ENOMEM;
2094 				break;
2095 			}
2096 			memcpy(arg, (char *)p + off, o.len);
2097 			break;
2098 		case DN_LINK:
2099 			if (dn == NULL)
2100 				dn = malloc(sizeof(*dn), M_TEMP, M_NOWAIT);
2101 			if (dn == NULL) {
2102 				err = ENOMEM;
2103 				break;
2104 			}
2105 			memcpy(&dn->link, (char *)p + off, sizeof(dn->link));
2106 			err = config_link(&dn->link, arg);
2107 			break;
2108 		case DN_PROFILE:
2109 			if (dn == NULL)
2110 				dn = malloc(sizeof(*dn), M_TEMP, M_NOWAIT);
2111 			if (dn == NULL) {
2112 				err = ENOMEM;
2113 				break;
2114 			}
2115 			memcpy(&dn->profile, (char *)p + off,
2116 			    sizeof(dn->profile));
2117 			err = config_profile(&dn->profile, arg);
2118 			break;
2119 		case DN_SCH:
2120 			if (dn == NULL)
2121 				dn = malloc(sizeof(*dn), M_TEMP, M_NOWAIT);
2122 			if (dn == NULL) {
2123 				err = ENOMEM;
2124 				break;
2125 			}
2126 			memcpy(&dn->sched, (char *)p + off,
2127 			    sizeof(dn->sched));
2128 			err = config_sched(&dn->sched, arg);
2129 			break;
2130 		case DN_FS:
2131 			if (dn == NULL)
2132 				dn = malloc(sizeof(*dn), M_TEMP, M_NOWAIT);
2133 			if (dn == NULL) {
2134 				err = ENOMEM;
2135 				break;
2136 			}
2137 			memcpy(&dn->fs, (char *)p + off, sizeof(dn->fs));
2138 			err = (NULL == config_fs(&dn->fs, arg, 0));
2139 			break;
2140 		}
2141 		if (err != 0)
2142 			break;
2143 		off += o.len;
2144 	}
2145 	free(arg, M_TEMP);
2146 	free(dn, M_TEMP);
2147 	return err;
2148 }
2149 
2150 static int
2151 compute_space(struct dn_id *cmd, struct copy_args *a)
2152 {
2153 	int x = 0, need = 0;
2154 	int profile_size = sizeof(struct dn_profile) -
2155 		ED_MAX_SAMPLES_NO*sizeof(int);
2156 
2157 	/* NOTE about compute space:
2158 	 * NP 	= V_dn_cfg.schk_count
2159 	 * NSI 	= V_dn_cfg.si_count
2160 	 * NF 	= V_dn_cfg.fsk_count
2161 	 * NQ 	= V_dn_cfg.queue_count
2162 	 * - ipfw pipe show
2163 	 *   (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
2164 	 *                             link, scheduler template, flowset
2165 	 *                             integrated in scheduler and header
2166 	 *                             for flowset list
2167 	 *   (NSI)*(dn_flow) all scheduler instance (includes
2168 	 *                              the queue instance)
2169 	 * - ipfw sched show
2170 	 *   (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
2171 	 *                             link, scheduler template, flowset
2172 	 *                             integrated in scheduler and header
2173 	 *                             for flowset list
2174 	 *   (NSI * dn_flow) all scheduler instances
2175 	 *   (NF * sizeof(uint_32)) space for flowset list linked to scheduler
2176 	 *   (NQ * dn_queue) all queue [XXXfor now not listed]
2177 	 * - ipfw queue show
2178 	 *   (NF * dn_fs) all flowset
2179 	 *   (NQ * dn_queue) all queues
2180 	 */
2181 	switch (cmd->subtype) {
2182 	default:
2183 		return -1;
2184 	/* XXX where do LINK and SCH differ ? */
2185 	/* 'ipfw sched show' could list all queues associated to
2186 	 * a scheduler. This feature for now is disabled
2187 	 */
2188 	case DN_LINK:	/* pipe show */
2189 		x = DN_C_LINK | DN_C_SCH | DN_C_FLOW;
2190 		need += V_dn_cfg.schk_count *
2191 			(sizeof(struct dn_fs) + profile_size) / 2;
2192 		need += V_dn_cfg.fsk_count * sizeof(uint32_t);
2193 		break;
2194 	case DN_SCH:	/* sched show */
2195 		need += V_dn_cfg.schk_count *
2196 			(sizeof(struct dn_fs) + profile_size) / 2;
2197 		need += V_dn_cfg.fsk_count * sizeof(uint32_t);
2198 		x = DN_C_SCH | DN_C_LINK | DN_C_FLOW;
2199 		break;
2200 	case DN_FS:	/* queue show */
2201 		x = DN_C_FS | DN_C_QUEUE;
2202 		break;
2203 	case DN_GET_COMPAT:	/* compatibility mode */
2204 		need =  dn_compat_calc_size();
2205 		break;
2206 	}
2207 	a->flags = x;
2208 	if (x & DN_C_SCH) {
2209 		need += V_dn_cfg.schk_count * sizeof(struct dn_sch) / 2;
2210 		/* NOT also, each fs might be attached to a sched */
2211 		need += V_dn_cfg.schk_count * sizeof(struct dn_id) / 2;
2212 	}
2213 	if (x & DN_C_FS)
2214 		need += V_dn_cfg.fsk_count * sizeof(struct dn_fs);
2215 	if (x & DN_C_LINK) {
2216 		need += V_dn_cfg.schk_count * sizeof(struct dn_link) / 2;
2217 	}
2218 	/*
2219 	 * When exporting a queue to userland, only pass up the
2220 	 * struct dn_flow, which is the only visible part.
2221 	 */
2222 
2223 	if (x & DN_C_QUEUE)
2224 		need += V_dn_cfg.queue_count * sizeof(struct dn_flow);
2225 	if (x & DN_C_FLOW)
2226 		need += V_dn_cfg.si_count * (sizeof(struct dn_flow));
2227 	return need;
2228 }
2229 
2230 /*
2231  * If compat != NULL dummynet_get is called in compatibility mode.
2232  * *compat will be the pointer to the buffer to pass to ipfw
2233  */
2234 int
2235 dummynet_get(struct sockopt *sopt, void **compat)
2236 {
2237 	int have, i, need, error;
2238 	char *start = NULL, *buf;
2239 	size_t sopt_valsize;
2240 	struct dn_id *cmd;
2241 	struct copy_args a;
2242 	struct copy_range r;
2243 	int l = sizeof(struct dn_id);
2244 
2245 	bzero(&a, sizeof(a));
2246 	bzero(&r, sizeof(r));
2247 
2248 	/* save and restore original sopt_valsize around copyin */
2249 	sopt_valsize = sopt->sopt_valsize;
2250 
2251 	cmd = &r.o;
2252 
2253 	if (!compat) {
2254 		/* copy at least an oid, and possibly a full object */
2255 		error = sooptcopyin(sopt, cmd, sizeof(r), sizeof(*cmd));
2256 		sopt->sopt_valsize = sopt_valsize;
2257 		if (error)
2258 			goto done;
2259 		l = cmd->len;
2260 #ifdef EMULATE_SYSCTL
2261 		/* sysctl emulation. */
2262 		if (cmd->type == DN_SYSCTL_GET)
2263 			return kesysctl_emu_get(sopt);
2264 #endif
2265 		if (l > sizeof(r)) {
2266 			/* request larger than default, allocate buffer */
2267 			cmd = malloc(l,  M_DUMMYNET, M_NOWAIT);
2268 			if (cmd == NULL) {
2269 				error = ENOMEM;
2270 				goto done;
2271 			}
2272 			error = sooptcopyin(sopt, cmd, l, l);
2273 			sopt->sopt_valsize = sopt_valsize;
2274 			if (error)
2275 				goto done;
2276 		}
2277 	} else { /* compatibility */
2278 		error = 0;
2279 		cmd->type = DN_CMD_GET;
2280 		cmd->len = sizeof(struct dn_id);
2281 		cmd->subtype = DN_GET_COMPAT;
2282 		// cmd->id = sopt_valsize;
2283 		D("compatibility mode");
2284 	}
2285 
2286 #ifdef NEW_AQM
2287 	/* get AQM params */
2288 	if(cmd->subtype == DN_AQM_PARAMS) {
2289 		error = get_aqm_parms(sopt);
2290 		goto done;
2291 	/* get Scheduler params */
2292 	} else if (cmd->subtype == DN_SCH_PARAMS) {
2293 		error = get_sched_parms(sopt);
2294 		goto done;
2295 	}
2296 #endif
2297 
2298 	a.extra = (struct copy_range *)cmd;
2299 	if (cmd->len == sizeof(*cmd)) { /* no range, create a default */
2300 		uint32_t *rp = (uint32_t *)(cmd + 1);
2301 		cmd->len += 2* sizeof(uint32_t);
2302 		rp[0] = 1;
2303 		rp[1] = DN_MAX_ID - 1;
2304 		if (cmd->subtype == DN_LINK) {
2305 			rp[0] += DN_MAX_ID;
2306 			rp[1] += DN_MAX_ID;
2307 		}
2308 	}
2309 	/* Count space (under lock) and allocate (outside lock).
2310 	 * Exit with lock held if we manage to get enough buffer.
2311 	 * Try a few times then give up.
2312 	 */
2313 	for (have = 0, i = 0; i < 10; i++) {
2314 		DN_BH_WLOCK();
2315 		need = compute_space(cmd, &a);
2316 
2317 		/* if there is a range, ignore value from compute_space() */
2318 		if (l > sizeof(*cmd))
2319 			need = sopt_valsize - sizeof(*cmd);
2320 
2321 		if (need < 0) {
2322 			DN_BH_WUNLOCK();
2323 			error = EINVAL;
2324 			goto done;
2325 		}
2326 		need += sizeof(*cmd);
2327 		cmd->id = need;
2328 		if (have >= need)
2329 			break;
2330 
2331 		DN_BH_WUNLOCK();
2332 		free(start, M_DUMMYNET);
2333 		start = NULL;
2334 		if (need > sopt_valsize)
2335 			break;
2336 
2337 		have = need;
2338 		start = malloc(have, M_DUMMYNET, M_NOWAIT | M_ZERO);
2339 	}
2340 
2341 	if (start == NULL) {
2342 		if (compat) {
2343 			*compat = NULL;
2344 			error =  1; // XXX
2345 		} else {
2346 			error = sooptcopyout(sopt, cmd, sizeof(*cmd));
2347 		}
2348 		goto done;
2349 	}
2350 	ND("have %d:%d sched %d, %d:%d links %d, %d:%d flowsets %d, "
2351 		"%d:%d si %d, %d:%d queues %d",
2352 		V_dn_cfg.schk_count, sizeof(struct dn_sch), DN_SCH,
2353 		V_dn_cfg.schk_count, sizeof(struct dn_link), DN_LINK,
2354 		V_dn_cfg.fsk_count, sizeof(struct dn_fs), DN_FS,
2355 		V_dn_cfg.si_count, sizeof(struct dn_flow), DN_SCH_I,
2356 		V_dn_cfg.queue_count, sizeof(struct dn_queue), DN_QUEUE);
2357 	sopt->sopt_valsize = sopt_valsize;
2358 	a.type = cmd->subtype;
2359 
2360 	if (compat == NULL) {
2361 		memcpy(start, cmd, sizeof(*cmd));
2362 		((struct dn_id*)(start))->len = sizeof(struct dn_id);
2363 		buf = start + sizeof(*cmd);
2364 	} else
2365 		buf = start;
2366 	a.start = &buf;
2367 	a.end = start + have;
2368 	/* start copying other objects */
2369 	if (compat) {
2370 		a.type = DN_COMPAT_PIPE;
2371 		dn_ht_scan(V_dn_cfg.schedhash, copy_data_helper_compat, &a);
2372 		a.type = DN_COMPAT_QUEUE;
2373 		dn_ht_scan(V_dn_cfg.fshash, copy_data_helper_compat, &a);
2374 	} else if (a.type == DN_FS) {
2375 		dn_ht_scan(V_dn_cfg.fshash, copy_data_helper, &a);
2376 	} else {
2377 		dn_ht_scan(V_dn_cfg.schedhash, copy_data_helper, &a);
2378 	}
2379 	DN_BH_WUNLOCK();
2380 
2381 	if (compat) {
2382 		*compat = start;
2383 		sopt->sopt_valsize = buf - start;
2384 		/* free() is done by ip_dummynet_compat() */
2385 		start = NULL; //XXX hack
2386 	} else {
2387 		error = sooptcopyout(sopt, start, buf - start);
2388 	}
2389 done:
2390 	if (cmd != &r.o)
2391 		free(cmd, M_DUMMYNET);
2392 	free(start, M_DUMMYNET);
2393 	return error;
2394 }
2395 
2396 /* Callback called on scheduler instance to delete it if idle */
2397 static int
2398 drain_scheduler_cb(void *_si, void *arg)
2399 {
2400 	struct dn_sch_inst *si = _si;
2401 
2402 	if ((si->kflags & DN_ACTIVE) || si->dline.mq.head != NULL)
2403 		return 0;
2404 
2405 	if (si->sched->fp->flags & DN_MULTIQUEUE) {
2406 		if (si->q_count == 0)
2407 			return si_destroy(si, NULL);
2408 		else
2409 			return 0;
2410 	} else { /* !DN_MULTIQUEUE */
2411 		if ((si+1)->ni.length == 0)
2412 			return si_destroy(si, NULL);
2413 		else
2414 			return 0;
2415 	}
2416 	return 0; /* unreachable */
2417 }
2418 
2419 /* Callback called on scheduler to check if it has instances */
2420 static int
2421 drain_scheduler_sch_cb(void *_s, void *arg)
2422 {
2423 	struct dn_schk *s = _s;
2424 
2425 	if (s->sch.flags & DN_HAVE_MASK) {
2426 		dn_ht_scan_bucket(s->siht, &s->drain_bucket,
2427 				drain_scheduler_cb, NULL);
2428 		s->drain_bucket++;
2429 	} else {
2430 		if (s->siht) {
2431 			if (drain_scheduler_cb(s->siht, NULL) == DNHT_SCAN_DEL)
2432 				s->siht = NULL;
2433 		}
2434 	}
2435 	return 0;
2436 }
2437 
2438 /* Called every tick, try to delete a 'bucket' of scheduler */
2439 void
2440 dn_drain_scheduler(void)
2441 {
2442 	dn_ht_scan_bucket(V_dn_cfg.schedhash, &V_dn_cfg.drain_sch,
2443 			   drain_scheduler_sch_cb, NULL);
2444 	V_dn_cfg.drain_sch++;
2445 }
2446 
2447 /* Callback called on queue to delete if it is idle */
2448 static int
2449 drain_queue_cb(void *_q, void *arg)
2450 {
2451 	struct dn_queue *q = _q;
2452 
2453 	if (q->ni.length == 0) {
2454 		dn_delete_queue(q, DN_DESTROY);
2455 		return DNHT_SCAN_DEL; /* queue is deleted */
2456 	}
2457 
2458 	return 0; /* queue isn't deleted */
2459 }
2460 
2461 /* Callback called on flowset used to check if it has queues */
2462 static int
2463 drain_queue_fs_cb(void *_fs, void *arg)
2464 {
2465 	struct dn_fsk *fs = _fs;
2466 
2467 	if (fs->fs.flags & DN_QHT_HASH) {
2468 		/* Flowset has a hash table for queues */
2469 		dn_ht_scan_bucket(fs->qht, &fs->drain_bucket,
2470 				drain_queue_cb, NULL);
2471 		fs->drain_bucket++;
2472 	} else {
2473 		/* No hash table for this flowset, null the pointer
2474 		 * if the queue is deleted
2475 		 */
2476 		if (fs->qht) {
2477 			if (drain_queue_cb(fs->qht, NULL) == DNHT_SCAN_DEL)
2478 				fs->qht = NULL;
2479 		}
2480 	}
2481 	return 0;
2482 }
2483 
2484 /* Called every tick, try to delete a 'bucket' of queue */
2485 void
2486 dn_drain_queue(void)
2487 {
2488 	/* scan a bucket of flowset */
2489 	dn_ht_scan_bucket(V_dn_cfg.fshash, &V_dn_cfg.drain_fs,
2490                                drain_queue_fs_cb, NULL);
2491 	V_dn_cfg.drain_fs++;
2492 }
2493 
2494 /*
2495  * Handler for the various dummynet socket options
2496  */
2497 static int
2498 ip_dn_ctl(struct sockopt *sopt)
2499 {
2500 	struct epoch_tracker et;
2501 	void *p = NULL;
2502 	size_t l;
2503 	int error;
2504 
2505 	error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2506 	if (error)
2507 		return (error);
2508 
2509 	/* Disallow sets in really-really secure mode. */
2510 	if (sopt->sopt_dir == SOPT_SET) {
2511 		error =  securelevel_ge(sopt->sopt_td->td_ucred, 3);
2512 		if (error)
2513 			return (error);
2514 	}
2515 
2516 	NET_EPOCH_ENTER(et);
2517 
2518 	switch (sopt->sopt_name) {
2519 	default :
2520 		D("dummynet: unknown option %d", sopt->sopt_name);
2521 		error = EINVAL;
2522 		break;
2523 
2524 	case IP_DUMMYNET_FLUSH:
2525 	case IP_DUMMYNET_CONFIGURE:
2526 	case IP_DUMMYNET_DEL:	/* remove a pipe or queue */
2527 	case IP_DUMMYNET_GET:
2528 		D("dummynet: compat option %d", sopt->sopt_name);
2529 		error = ip_dummynet_compat(sopt);
2530 		break;
2531 
2532 	case IP_DUMMYNET3:
2533 		if (sopt->sopt_dir == SOPT_GET) {
2534 			error = dummynet_get(sopt, NULL);
2535 			break;
2536 		}
2537 		l = sopt->sopt_valsize;
2538 		if (l < sizeof(struct dn_id) || l > 12000) {
2539 			D("argument len %zu invalid", l);
2540 			break;
2541 		}
2542 		p = malloc(l, M_TEMP, M_NOWAIT);
2543 		if (p == NULL) {
2544 			error = ENOMEM;
2545 			break;
2546 		}
2547 		error = sooptcopyin(sopt, p, l, l);
2548 		if (error == 0)
2549 			error = do_config(p, l);
2550 		break;
2551 	}
2552 
2553 	free(p, M_TEMP);
2554 
2555 	NET_EPOCH_EXIT(et);
2556 
2557 	return error ;
2558 }
2559 
2560 static void
2561 ip_dn_vnet_init(void)
2562 {
2563 	if (V_dn_cfg.init_done)
2564 		return;
2565 
2566 	/* Set defaults here. MSVC does not accept initializers,
2567 	 * and this is also useful for vimages
2568 	 */
2569 	/* queue limits */
2570 	V_dn_cfg.slot_limit = 100; /* Foot shooting limit for queues. */
2571 	V_dn_cfg.byte_limit = 1024 * 1024;
2572 	V_dn_cfg.expire = 1;
2573 
2574 	/* RED parameters */
2575 	V_dn_cfg.red_lookup_depth = 256;	/* default lookup table depth */
2576 	V_dn_cfg.red_avg_pkt_size = 512;	/* default medium packet size */
2577 	V_dn_cfg.red_max_pkt_size = 1500;	/* default max packet size */
2578 
2579 	/* hash tables */
2580 	V_dn_cfg.max_hash_size = 65536;	/* max in the hash tables */
2581 	V_dn_cfg.hash_size = 64;		/* default hash size */
2582 
2583 	/* create hash tables for schedulers and flowsets.
2584 	 * In both we search by key and by pointer.
2585 	 */
2586 	V_dn_cfg.schedhash = dn_ht_init(NULL, V_dn_cfg.hash_size,
2587 		offsetof(struct dn_schk, schk_next),
2588 		schk_hash, schk_match, schk_new);
2589 	V_dn_cfg.fshash = dn_ht_init(NULL, V_dn_cfg.hash_size,
2590 		offsetof(struct dn_fsk, fsk_next),
2591 		fsk_hash, fsk_match, fsk_new);
2592 
2593 	/* bucket index to drain object */
2594 	V_dn_cfg.drain_fs = 0;
2595 	V_dn_cfg.drain_sch = 0;
2596 
2597 	heap_init(&V_dn_cfg.evheap, 16, offsetof(struct dn_id, id));
2598 	SLIST_INIT(&V_dn_cfg.fsu);
2599 
2600 	DN_LOCK_INIT();
2601 
2602 	/* Initialize curr_time adjustment mechanics. */
2603 	getmicrouptime(&V_dn_cfg.prev_t);
2604 
2605 	V_dn_cfg.init_done = 1;
2606 }
2607 
2608 static void
2609 ip_dn_vnet_destroy(void)
2610 {
2611 	DN_BH_WLOCK();
2612 	dummynet_flush();
2613 	DN_BH_WUNLOCK();
2614 
2615 	dn_ht_free(V_dn_cfg.schedhash, 0);
2616 	dn_ht_free(V_dn_cfg.fshash, 0);
2617 	heap_free(&V_dn_cfg.evheap);
2618 
2619 	DN_LOCK_DESTROY();
2620 }
2621 
2622 static void
2623 ip_dn_init(void)
2624 {
2625 	if (dn_tasks_started)
2626 		return;
2627 
2628 	mtx_init(&sched_mtx, "dn_sched", NULL, MTX_DEF);
2629 
2630 	dn_tasks_started = 1;
2631 	TASK_INIT(&dn_task, 0, dummynet_task, NULL);
2632 	dn_tq = taskqueue_create_fast("dummynet", M_WAITOK,
2633 	    taskqueue_thread_enqueue, &dn_tq);
2634 	taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2635 
2636 	CK_LIST_INIT(&schedlist);
2637 	callout_init(&dn_timeout, 1);
2638 	dn_reschedule();
2639 }
2640 
2641 static void
2642 ip_dn_destroy(int last)
2643 {
2644 	/* ensure no more callouts are started */
2645 	dn_gone = 1;
2646 
2647 	/* check for last */
2648 	if (last) {
2649 		ND("removing last instance\n");
2650 		ip_dn_ctl_ptr = NULL;
2651 		ip_dn_io_ptr = NULL;
2652 	}
2653 
2654 	callout_drain(&dn_timeout);
2655 	taskqueue_drain(dn_tq, &dn_task);
2656 	taskqueue_free(dn_tq);
2657 }
2658 
2659 static int
2660 dummynet_modevent(module_t mod, int type, void *data)
2661 {
2662 
2663 	if (type == MOD_LOAD) {
2664 		if (ip_dn_io_ptr) {
2665 			printf("DUMMYNET already loaded\n");
2666 			return EEXIST ;
2667 		}
2668 		ip_dn_init();
2669 		ip_dn_ctl_ptr = ip_dn_ctl;
2670 		ip_dn_io_ptr = dummynet_io;
2671 		return 0;
2672 	} else if (type == MOD_UNLOAD) {
2673 		ip_dn_destroy(1 /* last */);
2674 		return 0;
2675 	} else
2676 		return EOPNOTSUPP;
2677 }
2678 
2679 /* modevent helpers for the modules */
2680 static int
2681 load_dn_sched(struct dn_alg *d)
2682 {
2683 	struct dn_alg *s;
2684 
2685 	if (d == NULL)
2686 		return 1; /* error */
2687 	ip_dn_init();	/* just in case, we need the lock */
2688 
2689 	/* Check that mandatory funcs exists */
2690 	if (d->enqueue == NULL || d->dequeue == NULL) {
2691 		D("missing enqueue or dequeue for %s", d->name);
2692 		return 1;
2693 	}
2694 
2695 	/* Search if scheduler already exists */
2696 	mtx_lock(&sched_mtx);
2697 	CK_LIST_FOREACH(s, &schedlist, next) {
2698 		if (strcmp(s->name, d->name) == 0) {
2699 			D("%s already loaded", d->name);
2700 			break; /* scheduler already exists */
2701 		}
2702 	}
2703 	if (s == NULL)
2704 		CK_LIST_INSERT_HEAD(&schedlist, d, next);
2705 	mtx_unlock(&sched_mtx);
2706 	D("dn_sched %s %sloaded", d->name, s ? "not ":"");
2707 	return s ? 1 : 0;
2708 }
2709 
2710 static int
2711 unload_dn_sched(struct dn_alg *s)
2712 {
2713 	struct dn_alg *tmp, *r;
2714 	int err = EINVAL;
2715 
2716 	ND("called for %s", s->name);
2717 
2718 	mtx_lock(&sched_mtx);
2719 	CK_LIST_FOREACH_SAFE(r, &schedlist, next, tmp) {
2720 		if (strcmp(s->name, r->name) != 0)
2721 			continue;
2722 		ND("ref_count = %d", r->ref_count);
2723 		err = (r->ref_count != 0) ? EBUSY : 0;
2724 		if (err == 0)
2725 			CK_LIST_REMOVE(r, next);
2726 		break;
2727 	}
2728 	mtx_unlock(&sched_mtx);
2729 	NET_EPOCH_WAIT();
2730 	D("dn_sched %s %sunloaded", s->name, err ? "not ":"");
2731 	return err;
2732 }
2733 
2734 int
2735 dn_sched_modevent(module_t mod, int cmd, void *arg)
2736 {
2737 	struct dn_alg *sch = arg;
2738 
2739 	if (cmd == MOD_LOAD)
2740 		return load_dn_sched(sch);
2741 	else if (cmd == MOD_UNLOAD)
2742 		return unload_dn_sched(sch);
2743 	else
2744 		return EINVAL;
2745 }
2746 
2747 static moduledata_t dummynet_mod = {
2748 	"dummynet", dummynet_modevent, NULL
2749 };
2750 
2751 #define	DN_SI_SUB	SI_SUB_PROTO_FIREWALL
2752 #define	DN_MODEV_ORD	(SI_ORDER_ANY - 128) /* after ipfw */
2753 DECLARE_MODULE(dummynet, dummynet_mod, DN_SI_SUB, DN_MODEV_ORD);
2754 MODULE_VERSION(dummynet, 3);
2755 
2756 /*
2757  * Starting up. Done in order after dummynet_modevent() has been called.
2758  * VNET_SYSINIT is also called for each existing vnet and each new vnet.
2759  */
2760 VNET_SYSINIT(vnet_dn_init, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_vnet_init, NULL);
2761 
2762 /*
2763  * Shutdown handlers up shop. These are done in REVERSE ORDER, but still
2764  * after dummynet_modevent() has been called. Not called on reboot.
2765  * VNET_SYSUNINIT is also called for each exiting vnet as it exits.
2766  * or when the module is unloaded.
2767  */
2768 VNET_SYSUNINIT(vnet_dn_uninit, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_vnet_destroy, NULL);
2769 
2770 #ifdef NEW_AQM
2771 
2772 /* modevent helpers for the AQM modules */
2773 static int
2774 load_dn_aqm(struct dn_aqm *d)
2775 {
2776 	struct dn_aqm *aqm=NULL;
2777 
2778 	if (d == NULL)
2779 		return 1; /* error */
2780 	ip_dn_init();	/* just in case, we need the lock */
2781 
2782 	/* Check that mandatory funcs exists */
2783 	if (d->enqueue == NULL || d->dequeue == NULL) {
2784 		D("missing enqueue or dequeue for %s", d->name);
2785 		return 1;
2786 	}
2787 
2788 	mtx_lock(&sched_mtx);
2789 
2790 	/* Search if AQM already exists */
2791 	CK_LIST_FOREACH(aqm, &aqmlist, next) {
2792 		if (strcmp(aqm->name, d->name) == 0) {
2793 			D("%s already loaded", d->name);
2794 			break; /* AQM already exists */
2795 		}
2796 	}
2797 	if (aqm == NULL)
2798 		CK_LIST_INSERT_HEAD(&aqmlist, d, next);
2799 
2800 	mtx_unlock(&sched_mtx);
2801 
2802 	D("dn_aqm %s %sloaded", d->name, aqm ? "not ":"");
2803 	return aqm ? 1 : 0;
2804 }
2805 
2806 /* Callback to clean up AQM status for queues connected to a flowset
2807  * and then deconfigure the flowset.
2808  * This function is called before an AQM module is unloaded
2809  */
2810 static int
2811 fs_cleanup(void *_fs, void *arg)
2812 {
2813 	struct dn_fsk *fs = _fs;
2814 	uint32_t type = *(uint32_t *)arg;
2815 
2816 	if (fs->aqmfp && fs->aqmfp->type == type)
2817 		aqm_cleanup_deconfig_fs(fs);
2818 
2819 	return 0;
2820 }
2821 
2822 static int
2823 unload_dn_aqm(struct dn_aqm *aqm)
2824 {
2825 	struct dn_aqm *tmp, *r;
2826 	int err = EINVAL;
2827 	err = 0;
2828 	ND("called for %s", aqm->name);
2829 
2830 	/* clean up AQM status and deconfig flowset */
2831 	dn_ht_scan(V_dn_cfg.fshash, fs_cleanup, &aqm->type);
2832 
2833 	mtx_lock(&sched_mtx);
2834 
2835 	CK_LIST_FOREACH_SAFE(r, &aqmlist, next, tmp) {
2836 		if (strcmp(aqm->name, r->name) != 0)
2837 			continue;
2838 		ND("ref_count = %d", r->ref_count);
2839 		err = (r->ref_count != 0 || r->cfg_ref_count != 0) ? EBUSY : 0;
2840 		if (err == 0)
2841 			CK_LIST_REMOVE(r, next);
2842 		break;
2843 	}
2844 
2845 	mtx_unlock(&sched_mtx);
2846 	NET_EPOCH_WAIT();
2847 
2848 	D("%s %sunloaded", aqm->name, err ? "not ":"");
2849 	if (err)
2850 		D("ref_count=%d, cfg_ref_count=%d", r->ref_count, r->cfg_ref_count);
2851 	return err;
2852 }
2853 
2854 int
2855 dn_aqm_modevent(module_t mod, int cmd, void *arg)
2856 {
2857 	struct dn_aqm *aqm = arg;
2858 
2859 	if (cmd == MOD_LOAD)
2860 		return load_dn_aqm(aqm);
2861 	else if (cmd == MOD_UNLOAD)
2862 		return unload_dn_aqm(aqm);
2863 	else
2864 		return EINVAL;
2865 }
2866 #endif
2867 
2868 /* end of file */
2869