xref: /freebsd/sys/netpfil/ipfw/ip_dummynet.c (revision 8b959dd6a3921c35395bef4a6d7ad2426a3bd88e)
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 dn_reschedule(void)
114 {
115 
116 	if (dn_gone != 0)
117 		return;
118 	callout_reset_sbt(&dn_timeout, tick_sbt, 0, dummynet, NULL,
119 	    C_HARDCLOCK | C_DIRECT_EXEC);
120 }
121 /*----- end of callout hooks -----*/
122 
123 #ifdef NEW_AQM
124 /* Return AQM descriptor for given type or name. */
125 static struct dn_aqm *
126 find_aqm_type(int type, char *name)
127 {
128 	struct dn_aqm *d;
129 
130 	NET_EPOCH_ASSERT();
131 
132 	CK_LIST_FOREACH(d, &aqmlist, next) {
133 		if (d->type == type || (name && !strcasecmp(d->name, name)))
134 			return d;
135 	}
136 	return NULL; /* not found */
137 }
138 #endif
139 
140 /* Return a scheduler descriptor given the type or name. */
141 static struct dn_alg *
142 find_sched_type(int type, char *name)
143 {
144 	struct dn_alg *d;
145 
146 	NET_EPOCH_ASSERT();
147 
148 	CK_LIST_FOREACH(d, &schedlist, next) {
149 		if (d->type == type || (name && !strcasecmp(d->name, name)))
150 			return d;
151 	}
152 	return NULL; /* not found */
153 }
154 
155 int
156 ipdn_bound_var(int *v, int dflt, int lo, int hi, const char *msg)
157 {
158 	int oldv = *v;
159 	const char *op = NULL;
160 	if (dflt < lo)
161 		dflt = lo;
162 	if (dflt > hi)
163 		dflt = hi;
164 	if (oldv < lo) {
165 		*v = dflt;
166 		op = "Bump";
167 	} else if (oldv > hi) {
168 		*v = hi;
169 		op = "Clamp";
170 	} else
171 		return *v;
172 	if (op && msg && bootverbose)
173 		printf("%s %s to %d (was %d)\n", op, msg, *v, oldv);
174 	return *v;
175 }
176 
177 /*---- flow_id mask, hash and compare functions ---*/
178 /*
179  * The flow_id includes the 5-tuple, the queue/pipe number
180  * which we store in the extra area in host order,
181  * and for ipv6 also the flow_id6.
182  * XXX see if we want the tos byte (can store in 'flags')
183  */
184 static struct ipfw_flow_id *
185 flow_id_mask(struct ipfw_flow_id *mask, struct ipfw_flow_id *id)
186 {
187 	int is_v6 = IS_IP6_FLOW_ID(id);
188 
189 	id->dst_port &= mask->dst_port;
190 	id->src_port &= mask->src_port;
191 	id->proto &= mask->proto;
192 	id->extra &= mask->extra;
193 	if (is_v6) {
194 		APPLY_MASK(&id->dst_ip6, &mask->dst_ip6);
195 		APPLY_MASK(&id->src_ip6, &mask->src_ip6);
196 		id->flow_id6 &= mask->flow_id6;
197 	} else {
198 		id->dst_ip &= mask->dst_ip;
199 		id->src_ip &= mask->src_ip;
200 	}
201 	return id;
202 }
203 
204 /* computes an OR of two masks, result in dst and also returned */
205 static struct ipfw_flow_id *
206 flow_id_or(struct ipfw_flow_id *src, struct ipfw_flow_id *dst)
207 {
208 	int is_v6 = IS_IP6_FLOW_ID(dst);
209 
210 	dst->dst_port |= src->dst_port;
211 	dst->src_port |= src->src_port;
212 	dst->proto |= src->proto;
213 	dst->extra |= src->extra;
214 	if (is_v6) {
215 #define OR_MASK(_d, _s)                          \
216     (_d)->__u6_addr.__u6_addr32[0] |= (_s)->__u6_addr.__u6_addr32[0]; \
217     (_d)->__u6_addr.__u6_addr32[1] |= (_s)->__u6_addr.__u6_addr32[1]; \
218     (_d)->__u6_addr.__u6_addr32[2] |= (_s)->__u6_addr.__u6_addr32[2]; \
219     (_d)->__u6_addr.__u6_addr32[3] |= (_s)->__u6_addr.__u6_addr32[3];
220 		OR_MASK(&dst->dst_ip6, &src->dst_ip6);
221 		OR_MASK(&dst->src_ip6, &src->src_ip6);
222 #undef OR_MASK
223 		dst->flow_id6 |= src->flow_id6;
224 	} else {
225 		dst->dst_ip |= src->dst_ip;
226 		dst->src_ip |= src->src_ip;
227 	}
228 	return dst;
229 }
230 
231 static int
232 nonzero_mask(struct ipfw_flow_id *m)
233 {
234 	if (m->dst_port || m->src_port || m->proto || m->extra)
235 		return 1;
236 	if (IS_IP6_FLOW_ID(m)) {
237 		return
238 			m->dst_ip6.__u6_addr.__u6_addr32[0] ||
239 			m->dst_ip6.__u6_addr.__u6_addr32[1] ||
240 			m->dst_ip6.__u6_addr.__u6_addr32[2] ||
241 			m->dst_ip6.__u6_addr.__u6_addr32[3] ||
242 			m->src_ip6.__u6_addr.__u6_addr32[0] ||
243 			m->src_ip6.__u6_addr.__u6_addr32[1] ||
244 			m->src_ip6.__u6_addr.__u6_addr32[2] ||
245 			m->src_ip6.__u6_addr.__u6_addr32[3] ||
246 			m->flow_id6;
247 	} else {
248 		return m->dst_ip || m->src_ip;
249 	}
250 }
251 
252 /* XXX we may want a better hash function */
253 static uint32_t
254 flow_id_hash(struct ipfw_flow_id *id)
255 {
256     uint32_t i;
257 
258     if (IS_IP6_FLOW_ID(id)) {
259 	uint32_t *d = (uint32_t *)&id->dst_ip6;
260 	uint32_t *s = (uint32_t *)&id->src_ip6;
261         i = (d[0]      ) ^ (d[1])       ^
262             (d[2]      ) ^ (d[3])       ^
263             (d[0] >> 15) ^ (d[1] >> 15) ^
264             (d[2] >> 15) ^ (d[3] >> 15) ^
265             (s[0] <<  1) ^ (s[1] <<  1) ^
266             (s[2] <<  1) ^ (s[3] <<  1) ^
267             (s[0] << 16) ^ (s[1] << 16) ^
268             (s[2] << 16) ^ (s[3] << 16) ^
269             (id->dst_port << 1) ^ (id->src_port) ^
270 	    (id->extra) ^
271             (id->proto ) ^ (id->flow_id6);
272     } else {
273         i = (id->dst_ip)        ^ (id->dst_ip >> 15) ^
274             (id->src_ip << 1)   ^ (id->src_ip >> 16) ^
275 	    (id->extra) ^
276             (id->dst_port << 1) ^ (id->src_port)     ^ (id->proto);
277     }
278     return i;
279 }
280 
281 /* Like bcmp, returns 0 if ids match, 1 otherwise. */
282 static int
283 flow_id_cmp(struct ipfw_flow_id *id1, struct ipfw_flow_id *id2)
284 {
285 	int is_v6 = IS_IP6_FLOW_ID(id1);
286 
287 	if (!is_v6) {
288 	    if (IS_IP6_FLOW_ID(id2))
289 		return 1; /* different address families */
290 
291 	    return (id1->dst_ip == id2->dst_ip &&
292 		    id1->src_ip == id2->src_ip &&
293 		    id1->dst_port == id2->dst_port &&
294 		    id1->src_port == id2->src_port &&
295 		    id1->proto == id2->proto &&
296 		    id1->extra == id2->extra) ? 0 : 1;
297 	}
298 	/* the ipv6 case */
299 	return (
300 	    !bcmp(&id1->dst_ip6,&id2->dst_ip6, sizeof(id1->dst_ip6)) &&
301 	    !bcmp(&id1->src_ip6,&id2->src_ip6, sizeof(id1->src_ip6)) &&
302 	    id1->dst_port == id2->dst_port &&
303 	    id1->src_port == id2->src_port &&
304 	    id1->proto == id2->proto &&
305 	    id1->extra == id2->extra &&
306 	    id1->flow_id6 == id2->flow_id6) ? 0 : 1;
307 }
308 /*--------- end of flow-id mask, hash and compare ---------*/
309 
310 /*--- support functions for the qht hashtable ----
311  * Entries are hashed by flow-id
312  */
313 static uint32_t
314 q_hash(uintptr_t key, int flags, void *arg)
315 {
316 	/* compute the hash slot from the flow id */
317 	struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
318 		&((struct dn_queue *)key)->ni.fid :
319 		(struct ipfw_flow_id *)key;
320 
321 	return flow_id_hash(id);
322 }
323 
324 static int
325 q_match(void *obj, uintptr_t key, int flags, void *arg)
326 {
327 	struct dn_queue *o = (struct dn_queue *)obj;
328 	struct ipfw_flow_id *id2;
329 
330 	if (flags & DNHT_KEY_IS_OBJ) {
331 		/* compare pointers */
332 		id2 = &((struct dn_queue *)key)->ni.fid;
333 	} else {
334 		id2 = (struct ipfw_flow_id *)key;
335 	}
336 	return (0 == flow_id_cmp(&o->ni.fid,  id2));
337 }
338 
339 /*
340  * create a new queue instance for the given 'key'.
341  */
342 static void *
343 q_new(uintptr_t key, int flags, void *arg)
344 {
345 	struct dn_queue *q, *template = arg;
346 	struct dn_fsk *fs = template->fs;
347 	int size = sizeof(*q) + fs->sched->fp->q_datalen;
348 
349 	q = malloc(size, M_DUMMYNET, M_NOWAIT | M_ZERO);
350 	if (q == NULL) {
351 		D("no memory for new queue");
352 		return NULL;
353 	}
354 
355 	set_oid(&q->ni.oid, DN_QUEUE, size);
356 	if (fs->fs.flags & DN_QHT_HASH)
357 		q->ni.fid = *(struct ipfw_flow_id *)key;
358 	q->fs = fs;
359 	q->_si = template->_si;
360 	q->_si->q_count++;
361 
362 	if (fs->sched->fp->new_queue)
363 		fs->sched->fp->new_queue(q);
364 
365 #ifdef NEW_AQM
366 	/* call AQM init function after creating a queue*/
367 	if (fs->aqmfp && fs->aqmfp->init)
368 		if(fs->aqmfp->init(q))
369 			D("unable to init AQM for fs %d", fs->fs.fs_nr);
370 #endif
371 	V_dn_cfg.queue_count++;
372 
373 	return q;
374 }
375 
376 /*
377  * Notify schedulers that a queue is going away.
378  * If (flags & DN_DESTROY), also free the packets.
379  * The version for callbacks is called q_delete_cb().
380  */
381 static void
382 dn_delete_queue(struct dn_queue *q, int flags)
383 {
384 	struct dn_fsk *fs = q->fs;
385 
386 #ifdef NEW_AQM
387 	/* clean up AQM status for queue 'q'
388 	 * cleanup here is called just with MULTIQUEUE
389 	 */
390 	if (fs && fs->aqmfp && fs->aqmfp->cleanup)
391 		fs->aqmfp->cleanup(q);
392 #endif
393 	// D("fs %p si %p\n", fs, q->_si);
394 	/* notify the parent scheduler that the queue is going away */
395 	if (fs && fs->sched->fp->free_queue)
396 		fs->sched->fp->free_queue(q);
397 	q->_si->q_count--;
398 	q->_si = NULL;
399 	if (flags & DN_DESTROY) {
400 		if (q->mq.head)
401 			dn_free_pkts(q->mq.head);
402 		bzero(q, sizeof(*q));	// safety
403 		free(q, M_DUMMYNET);
404 		V_dn_cfg.queue_count--;
405 	}
406 }
407 
408 static int
409 q_delete_cb(void *q, void *arg)
410 {
411 	int flags = (int)(uintptr_t)arg;
412 	dn_delete_queue(q, flags);
413 	return (flags & DN_DESTROY) ? DNHT_SCAN_DEL : 0;
414 }
415 
416 /*
417  * calls dn_delete_queue/q_delete_cb on all queues,
418  * which notifies the parent scheduler and possibly drains packets.
419  * flags & DN_DESTROY: drains queues and destroy qht;
420  */
421 static void
422 qht_delete(struct dn_fsk *fs, int flags)
423 {
424 	ND("fs %d start flags %d qht %p",
425 		fs->fs.fs_nr, flags, fs->qht);
426 	if (!fs->qht)
427 		return;
428 	if (fs->fs.flags & DN_QHT_HASH) {
429 		dn_ht_scan(fs->qht, q_delete_cb, (void *)(uintptr_t)flags);
430 		if (flags & DN_DESTROY) {
431 			dn_ht_free(fs->qht, 0);
432 			fs->qht = NULL;
433 		}
434 	} else {
435 		dn_delete_queue((struct dn_queue *)(fs->qht), flags);
436 		if (flags & DN_DESTROY)
437 			fs->qht = NULL;
438 	}
439 }
440 
441 /*
442  * Find and possibly create the queue for a MULTIQUEUE scheduler.
443  * We never call it for !MULTIQUEUE (the queue is in the sch_inst).
444  */
445 struct dn_queue *
446 ipdn_q_find(struct dn_fsk *fs, struct dn_sch_inst *si,
447 	struct ipfw_flow_id *id)
448 {
449 	struct dn_queue template;
450 
451 	template._si = si;
452 	template.fs = fs;
453 
454 	if (fs->fs.flags & DN_QHT_HASH) {
455 		struct ipfw_flow_id masked_id;
456 		if (fs->qht == NULL) {
457 			fs->qht = dn_ht_init(NULL, fs->fs.buckets,
458 				offsetof(struct dn_queue, q_next),
459 				q_hash, q_match, q_new);
460 			if (fs->qht == NULL)
461 				return NULL;
462 		}
463 		masked_id = *id;
464 		flow_id_mask(&fs->fsk_mask, &masked_id);
465 		return dn_ht_find(fs->qht, (uintptr_t)&masked_id,
466 			DNHT_INSERT, &template);
467 	} else {
468 		if (fs->qht == NULL)
469 			fs->qht = q_new(0, 0, &template);
470 		return (struct dn_queue *)fs->qht;
471 	}
472 }
473 /*--- end of queue hash table ---*/
474 
475 /*--- support functions for the sch_inst hashtable ----
476  *
477  * These are hashed by flow-id
478  */
479 static uint32_t
480 si_hash(uintptr_t key, int flags, void *arg)
481 {
482 	/* compute the hash slot from the flow id */
483 	struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
484 		&((struct dn_sch_inst *)key)->ni.fid :
485 		(struct ipfw_flow_id *)key;
486 
487 	return flow_id_hash(id);
488 }
489 
490 static int
491 si_match(void *obj, uintptr_t key, int flags, void *arg)
492 {
493 	struct dn_sch_inst *o = obj;
494 	struct ipfw_flow_id *id2;
495 
496 	id2 = (flags & DNHT_KEY_IS_OBJ) ?
497 		&((struct dn_sch_inst *)key)->ni.fid :
498 		(struct ipfw_flow_id *)key;
499 	return flow_id_cmp(&o->ni.fid,  id2) == 0;
500 }
501 
502 /*
503  * create a new instance for the given 'key'
504  * Allocate memory for instance, delay line and scheduler private data.
505  */
506 static void *
507 si_new(uintptr_t key, int flags, void *arg)
508 {
509 	struct dn_schk *s = arg;
510 	struct dn_sch_inst *si;
511 	int l = sizeof(*si) + s->fp->si_datalen;
512 
513 	si = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
514 	if (si == NULL)
515 		goto error;
516 
517 	/* Set length only for the part passed up to userland. */
518 	set_oid(&si->ni.oid, DN_SCH_I, sizeof(struct dn_flow));
519 	set_oid(&(si->dline.oid), DN_DELAY_LINE,
520 		sizeof(struct delay_line));
521 	/* mark si and dline as outside the event queue */
522 	si->ni.oid.id = si->dline.oid.id = -1;
523 
524 	si->sched = s;
525 	si->dline.si = si;
526 
527 	if (s->fp->new_sched && s->fp->new_sched(si)) {
528 		D("new_sched error");
529 		goto error;
530 	}
531 	if (s->sch.flags & DN_HAVE_MASK)
532 		si->ni.fid = *(struct ipfw_flow_id *)key;
533 
534 #ifdef NEW_AQM
535 	/* init AQM status for !DN_MULTIQUEUE sched*/
536 	if (!(s->fp->flags & DN_MULTIQUEUE))
537 		if (s->fs->aqmfp && s->fs->aqmfp->init)
538 			if(s->fs->aqmfp->init((struct dn_queue *)(si + 1))) {
539 				D("unable to init AQM for fs %d", s->fs->fs.fs_nr);
540 				goto error;
541 			}
542 #endif
543 
544 	V_dn_cfg.si_count++;
545 	return si;
546 
547 error:
548 	if (si) {
549 		bzero(si, sizeof(*si)); // safety
550 		free(si, M_DUMMYNET);
551 	}
552         return NULL;
553 }
554 
555 /*
556  * Callback from siht to delete all scheduler instances. Remove
557  * si and delay line from the system heap, destroy all queues.
558  * We assume that all flowset have been notified and do not
559  * point to us anymore.
560  */
561 static int
562 si_destroy(void *_si, void *arg)
563 {
564 	struct dn_sch_inst *si = _si;
565 	struct dn_schk *s = si->sched;
566 	struct delay_line *dl = &si->dline;
567 
568 	if (dl->oid.subtype) /* remove delay line from event heap */
569 		heap_extract(&V_dn_cfg.evheap, dl);
570 	dn_free_pkts(dl->mq.head);	/* drain delay line */
571 	if (si->kflags & DN_ACTIVE) /* remove si from event heap */
572 		heap_extract(&V_dn_cfg.evheap, si);
573 
574 #ifdef NEW_AQM
575 	/* clean up AQM status for !DN_MULTIQUEUE sched
576 	 * Note that all queues belong to fs were cleaned up in fsk_detach.
577 	 * When drain_scheduler is called s->fs and q->fs are pointing
578 	 * to a correct fs, so we can use fs in this case.
579 	 */
580 	if (!(s->fp->flags & DN_MULTIQUEUE)) {
581 		struct dn_queue *q = (struct dn_queue *)(si + 1);
582 		if (q->aqm_status && q->fs->aqmfp)
583 			if (q->fs->aqmfp->cleanup)
584 				q->fs->aqmfp->cleanup(q);
585 	}
586 #endif
587 	if (s->fp->free_sched)
588 		s->fp->free_sched(si);
589 	bzero(si, sizeof(*si));	/* safety */
590 	free(si, M_DUMMYNET);
591 	V_dn_cfg.si_count--;
592 	return DNHT_SCAN_DEL;
593 }
594 
595 /*
596  * Find the scheduler instance for this packet. If we need to apply
597  * a mask, do on a local copy of the flow_id to preserve the original.
598  * Assume siht is always initialized if we have a mask.
599  */
600 struct dn_sch_inst *
601 ipdn_si_find(struct dn_schk *s, struct ipfw_flow_id *id)
602 {
603 
604 	if (s->sch.flags & DN_HAVE_MASK) {
605 		struct ipfw_flow_id id_t = *id;
606 		flow_id_mask(&s->sch.sched_mask, &id_t);
607 		return dn_ht_find(s->siht, (uintptr_t)&id_t,
608 			DNHT_INSERT, s);
609 	}
610 	if (!s->siht)
611 		s->siht = si_new(0, 0, s);
612 	return (struct dn_sch_inst *)s->siht;
613 }
614 
615 /* callback to flush credit for the scheduler instance */
616 static int
617 si_reset_credit(void *_si, void *arg)
618 {
619 	struct dn_sch_inst *si = _si;
620 	struct dn_link *p = &si->sched->link;
621 
622 	si->credit = p->burst + (V_dn_cfg.io_fast ?  p->bandwidth : 0);
623 	return 0;
624 }
625 
626 static void
627 schk_reset_credit(struct dn_schk *s)
628 {
629 	if (s->sch.flags & DN_HAVE_MASK)
630 		dn_ht_scan(s->siht, si_reset_credit, NULL);
631 	else if (s->siht)
632 		si_reset_credit(s->siht, NULL);
633 }
634 /*---- end of sch_inst hashtable ---------------------*/
635 
636 /*-------------------------------------------------------
637  * flowset hash (fshash) support. Entries are hashed by fs_nr.
638  * New allocations are put in the fsunlinked list, from which
639  * they are removed when they point to a specific scheduler.
640  */
641 static uint32_t
642 fsk_hash(uintptr_t key, int flags, void *arg)
643 {
644 	uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
645 		((struct dn_fsk *)key)->fs.fs_nr;
646 
647 	return ( (i>>8)^(i>>4)^i );
648 }
649 
650 static int
651 fsk_match(void *obj, uintptr_t key, int flags, void *arg)
652 {
653 	struct dn_fsk *fs = obj;
654 	int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
655 		((struct dn_fsk *)key)->fs.fs_nr;
656 
657 	return (fs->fs.fs_nr == i);
658 }
659 
660 static void *
661 fsk_new(uintptr_t key, int flags, void *arg)
662 {
663 	struct dn_fsk *fs;
664 
665 	fs = malloc(sizeof(*fs), M_DUMMYNET, M_NOWAIT | M_ZERO);
666 	if (fs) {
667 		set_oid(&fs->fs.oid, DN_FS, sizeof(fs->fs));
668 		V_dn_cfg.fsk_count++;
669 		fs->drain_bucket = 0;
670 		SLIST_INSERT_HEAD(&V_dn_cfg.fsu, fs, sch_chain);
671 	}
672 	return fs;
673 }
674 
675 #ifdef NEW_AQM
676 /* callback function for cleaning up AQM queue status belongs to a flowset
677  * connected to scheduler instance '_si' (for !DN_MULTIQUEUE only).
678  */
679 static int
680 si_cleanup_q(void *_si, void *arg)
681 {
682 	struct dn_sch_inst *si = _si;
683 
684 	if (!(si->sched->fp->flags & DN_MULTIQUEUE)) {
685 		if (si->sched->fs->aqmfp && si->sched->fs->aqmfp->cleanup)
686 			si->sched->fs->aqmfp->cleanup((struct dn_queue *) (si+1));
687 	}
688 	return 0;
689 }
690 
691 /* callback to clean up queue AQM status.*/
692 static int
693 q_cleanup_q(void *_q, void *arg)
694 {
695 	struct dn_queue *q = _q;
696 	q->fs->aqmfp->cleanup(q);
697 	return 0;
698 }
699 
700 /* Clean up all AQM queues status belongs to flowset 'fs' and then
701  * deconfig AQM for flowset 'fs'
702  */
703 static void
704 aqm_cleanup_deconfig_fs(struct dn_fsk *fs)
705 {
706 	struct dn_sch_inst *si;
707 
708 	/* clean up AQM status for all queues for !DN_MULTIQUEUE sched*/
709 	if (fs->fs.fs_nr > DN_MAX_ID) {
710 		if (fs->sched && !(fs->sched->fp->flags & DN_MULTIQUEUE)) {
711 			if (fs->sched->sch.flags & DN_HAVE_MASK)
712 				dn_ht_scan(fs->sched->siht, si_cleanup_q, NULL);
713 			else {
714 					/* single si i.e. no sched mask */
715 					si = (struct dn_sch_inst *) fs->sched->siht;
716 					if (si && fs->aqmfp && fs->aqmfp->cleanup)
717 						fs->aqmfp->cleanup((struct dn_queue *) (si+1));
718 			}
719 		}
720 	}
721 
722 	/* clean up AQM status for all queues for DN_MULTIQUEUE sched*/
723 	if (fs->sched && fs->sched->fp->flags & DN_MULTIQUEUE && fs->qht) {
724 			if (fs->fs.flags & DN_QHT_HASH)
725 				dn_ht_scan(fs->qht, q_cleanup_q, NULL);
726 			else
727 				fs->aqmfp->cleanup((struct dn_queue *)(fs->qht));
728 	}
729 
730 	/* deconfig AQM */
731 	if(fs->aqmcfg && fs->aqmfp && fs->aqmfp->deconfig)
732 		fs->aqmfp->deconfig(fs);
733 }
734 #endif
735 
736 /*
737  * detach flowset from its current scheduler. Flags as follows:
738  * DN_DETACH removes from the fsk_list
739  * DN_DESTROY deletes individual queues
740  * DN_DELETE_FS destroys the flowset (otherwise goes in unlinked).
741  */
742 static void
743 fsk_detach(struct dn_fsk *fs, int flags)
744 {
745 	if (flags & DN_DELETE_FS)
746 		flags |= DN_DESTROY;
747 	ND("fs %d from sched %d flags %s %s %s",
748 		fs->fs.fs_nr, fs->fs.sched_nr,
749 		(flags & DN_DELETE_FS) ? "DEL_FS":"",
750 		(flags & DN_DESTROY) ? "DEL":"",
751 		(flags & DN_DETACH) ? "DET":"");
752 	if (flags & DN_DETACH) { /* detach from the list */
753 		struct dn_fsk_head *h;
754 		h = fs->sched ? &fs->sched->fsk_list : &V_dn_cfg.fsu;
755 		SLIST_REMOVE(h, fs, dn_fsk, sch_chain);
756 	}
757 	/* Free the RED parameters, they will be recomputed on
758 	 * subsequent attach if needed.
759 	 */
760 	free(fs->w_q_lookup, M_DUMMYNET);
761 	fs->w_q_lookup = NULL;
762 	qht_delete(fs, flags);
763 #ifdef NEW_AQM
764 	aqm_cleanup_deconfig_fs(fs);
765 #endif
766 
767 	if (fs->sched && fs->sched->fp->free_fsk)
768 		fs->sched->fp->free_fsk(fs);
769 	fs->sched = NULL;
770 	if (flags & DN_DELETE_FS) {
771 		bzero(fs, sizeof(*fs));	/* safety */
772 		free(fs, M_DUMMYNET);
773 		V_dn_cfg.fsk_count--;
774 	} else {
775 		SLIST_INSERT_HEAD(&V_dn_cfg.fsu, fs, sch_chain);
776 	}
777 }
778 
779 /*
780  * Detach or destroy all flowsets in a list.
781  * flags specifies what to do:
782  * DN_DESTROY:	flush all queues
783  * DN_DELETE_FS:	DN_DESTROY + destroy flowset
784  *	DN_DELETE_FS implies DN_DESTROY
785  */
786 static void
787 fsk_detach_list(struct dn_fsk_head *h, int flags)
788 {
789 	struct dn_fsk *fs;
790 	int n = 0; /* only for stats */
791 
792 	ND("head %p flags %x", h, flags);
793 	while ((fs = SLIST_FIRST(h))) {
794 		SLIST_REMOVE_HEAD(h, sch_chain);
795 		n++;
796 		fsk_detach(fs, flags);
797 	}
798 	ND("done %d flowsets", n);
799 }
800 
801 /*
802  * called on 'queue X delete' -- removes the flowset from fshash,
803  * deletes all queues for the flowset, and removes the flowset.
804  */
805 static int
806 delete_fs(int i, int locked)
807 {
808 	struct dn_fsk *fs;
809 	int err = 0;
810 
811 	if (!locked)
812 		DN_BH_WLOCK();
813 	fs = dn_ht_find(V_dn_cfg.fshash, i, DNHT_REMOVE, NULL);
814 	ND("fs %d found %p", i, fs);
815 	if (fs) {
816 		fsk_detach(fs, DN_DETACH | DN_DELETE_FS);
817 		err = 0;
818 	} else
819 		err = EINVAL;
820 	if (!locked)
821 		DN_BH_WUNLOCK();
822 	return err;
823 }
824 
825 /*----- end of flowset hashtable support -------------*/
826 
827 /*------------------------------------------------------------
828  * Scheduler hash. When searching by index we pass sched_nr,
829  * otherwise we pass struct dn_sch * which is the first field in
830  * struct dn_schk so we can cast between the two. We use this trick
831  * because in the create phase (but it should be fixed).
832  */
833 static uint32_t
834 schk_hash(uintptr_t key, int flags, void *_arg)
835 {
836 	uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
837 		((struct dn_schk *)key)->sch.sched_nr;
838 	return ( (i>>8)^(i>>4)^i );
839 }
840 
841 static int
842 schk_match(void *obj, uintptr_t key, int flags, void *_arg)
843 {
844 	struct dn_schk *s = (struct dn_schk *)obj;
845 	int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
846 		((struct dn_schk *)key)->sch.sched_nr;
847 	return (s->sch.sched_nr == i);
848 }
849 
850 /*
851  * Create the entry and intialize with the sched hash if needed.
852  * Leave s->fp unset so we can tell whether a dn_ht_find() returns
853  * a new object or a previously existing one.
854  */
855 static void *
856 schk_new(uintptr_t key, int flags, void *arg)
857 {
858 	struct schk_new_arg *a = arg;
859 	struct dn_schk *s;
860 	int l = sizeof(*s) +a->fp->schk_datalen;
861 
862 	s = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
863 	if (s == NULL)
864 		return NULL;
865 	set_oid(&s->link.oid, DN_LINK, sizeof(s->link));
866 	s->sch = *a->sch; // copy initial values
867 	s->link.link_nr = s->sch.sched_nr;
868 	SLIST_INIT(&s->fsk_list);
869 	/* initialize the hash table or create the single instance */
870 	s->fp = a->fp;	/* si_new needs this */
871 	s->drain_bucket = 0;
872 	if (s->sch.flags & DN_HAVE_MASK) {
873 		s->siht = dn_ht_init(NULL, s->sch.buckets,
874 			offsetof(struct dn_sch_inst, si_next),
875 			si_hash, si_match, si_new);
876 		if (s->siht == NULL) {
877 			free(s, M_DUMMYNET);
878 			return NULL;
879 		}
880 	}
881 	s->fp = NULL;	/* mark as a new scheduler */
882 	V_dn_cfg.schk_count++;
883 	return s;
884 }
885 
886 /*
887  * Callback for sched delete. Notify all attached flowsets to
888  * detach from the scheduler, destroy the internal flowset, and
889  * all instances. The scheduler goes away too.
890  * arg is 0 (only detach flowsets and destroy instances)
891  * DN_DESTROY (detach & delete queues, delete schk)
892  * or DN_DELETE_FS (delete queues and flowsets, delete schk)
893  */
894 static int
895 schk_delete_cb(void *obj, void *arg)
896 {
897 	struct dn_schk *s = obj;
898 #if 0
899 	int a = (int)arg;
900 	ND("sched %d arg %s%s",
901 		s->sch.sched_nr,
902 		a&DN_DESTROY ? "DEL ":"",
903 		a&DN_DELETE_FS ? "DEL_FS":"");
904 #endif
905 	fsk_detach_list(&s->fsk_list, arg ? DN_DESTROY : 0);
906 	/* no more flowset pointing to us now */
907 	if (s->sch.flags & DN_HAVE_MASK) {
908 		dn_ht_scan(s->siht, si_destroy, NULL);
909 		dn_ht_free(s->siht, 0);
910 	} else if (s->siht)
911 		si_destroy(s->siht, NULL);
912 
913 	free(s->profile, M_DUMMYNET);
914 	s->profile = NULL;
915 	s->siht = NULL;
916 	if (s->fp->destroy)
917 		s->fp->destroy(s);
918 	bzero(s, sizeof(*s));	// safety
919 	free(obj, M_DUMMYNET);
920 	V_dn_cfg.schk_count--;
921 	return DNHT_SCAN_DEL;
922 }
923 
924 /*
925  * called on a 'sched X delete' command. Deletes a single scheduler.
926  * This is done by removing from the schedhash, unlinking all
927  * flowsets and deleting their traffic.
928  */
929 static int
930 delete_schk(int i)
931 {
932 	struct dn_schk *s;
933 
934 	s = dn_ht_find(V_dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
935 	ND("%d %p", i, s);
936 	if (!s)
937 		return EINVAL;
938 	delete_fs(i + DN_MAX_ID, 1); /* first delete internal fs */
939 	/* then detach flowsets, delete traffic */
940 	schk_delete_cb(s, (void*)(uintptr_t)DN_DESTROY);
941 	return 0;
942 }
943 /*--- end of schk hashtable support ---*/
944 
945 static int
946 copy_obj(char **start, char *end, void *_o, const char *msg, int i)
947 {
948 	struct dn_id o;
949 	union {
950 		struct dn_link l;
951 		struct dn_schk s;
952 	} dn;
953 	int have = end - *start;
954 
955 	memcpy(&o, _o, sizeof(o));
956 	if (have < o.len || o.len == 0 || o.type == 0) {
957 		D("(WARN) type %d %s %d have %d need %d",
958 		    o.type, msg, i, have, o.len);
959 		return 1;
960 	}
961 	ND("type %d %s %d len %d", o.type, msg, i, o.len);
962 	if (o.type == DN_LINK) {
963 		memcpy(&dn.l, _o, sizeof(dn.l));
964 		/* Adjust burst parameter for link */
965 		dn.l.burst = div64(dn.l.burst, 8 * hz);
966 		dn.l.delay = dn.l.delay * 1000 / hz;
967 		memcpy(*start, &dn.l, sizeof(dn.l));
968 	} else if (o.type == DN_SCH) {
969 		/* Set dn.s.sch.oid.id to the number of instances */
970 		memcpy(&dn.s, _o, sizeof(dn.s));
971 		dn.s.sch.oid.id = (dn.s.sch.flags & DN_HAVE_MASK) ?
972 		    dn_ht_entries(dn.s.siht) : (dn.s.siht ? 1 : 0);
973 		memcpy(*start, &dn.s, sizeof(dn.s));
974 	} else
975 		memcpy(*start, _o, o.len);
976 	*start += o.len;
977 	return 0;
978 }
979 
980 /* Specific function to copy a queue.
981  * Copies only the user-visible part of a queue (which is in
982  * a struct dn_flow), and sets len accordingly.
983  */
984 static int
985 copy_obj_q(char **start, char *end, void *_o, const char *msg, int i)
986 {
987 	struct dn_id *o = _o;
988 	int have = end - *start;
989 	int len = sizeof(struct dn_flow); /* see above comment */
990 
991 	if (have < len || o->len == 0 || o->type != DN_QUEUE) {
992 		D("ERROR type %d %s %d have %d need %d",
993 			o->type, msg, i, have, len);
994 		return 1;
995 	}
996 	ND("type %d %s %d len %d", o->type, msg, i, len);
997 	memcpy(*start, _o, len);
998 	((struct dn_id*)(*start))->len = len;
999 	*start += len;
1000 	return 0;
1001 }
1002 
1003 static int
1004 copy_q_cb(void *obj, void *arg)
1005 {
1006 	struct dn_queue *q = obj;
1007 	struct copy_args *a = arg;
1008 	struct dn_flow *ni = (struct dn_flow *)(*a->start);
1009         if (copy_obj_q(a->start, a->end, &q->ni, "queue", -1))
1010                 return DNHT_SCAN_END;
1011         ni->oid.type = DN_FLOW; /* override the DN_QUEUE */
1012         ni->oid.id = si_hash((uintptr_t)&ni->fid, 0, NULL);
1013         return 0;
1014 }
1015 
1016 static int
1017 copy_q(struct copy_args *a, struct dn_fsk *fs, int flags)
1018 {
1019 	if (!fs->qht)
1020 		return 0;
1021 	if (fs->fs.flags & DN_QHT_HASH)
1022 		dn_ht_scan(fs->qht, copy_q_cb, a);
1023 	else
1024 		copy_q_cb(fs->qht, a);
1025 	return 0;
1026 }
1027 
1028 /*
1029  * This routine only copies the initial part of a profile ? XXX
1030  */
1031 static int
1032 copy_profile(struct copy_args *a, struct dn_profile *p)
1033 {
1034 	int have = a->end - *a->start;
1035 	/* XXX here we check for max length */
1036 	int profile_len = sizeof(struct dn_profile) -
1037 		ED_MAX_SAMPLES_NO*sizeof(int);
1038 
1039 	if (p == NULL)
1040 		return 0;
1041 	if (have < profile_len) {
1042 		D("error have %d need %d", have, profile_len);
1043 		return 1;
1044 	}
1045 	memcpy(*a->start, p, profile_len);
1046 	((struct dn_id *)(*a->start))->len = profile_len;
1047 	*a->start += profile_len;
1048 	return 0;
1049 }
1050 
1051 static int
1052 copy_flowset(struct copy_args *a, struct dn_fsk *fs, int flags)
1053 {
1054 	struct dn_fs *ufs = (struct dn_fs *)(*a->start);
1055 	if (!fs)
1056 		return 0;
1057 	ND("flowset %d", fs->fs.fs_nr);
1058 	if (copy_obj(a->start, a->end, &fs->fs, "flowset", fs->fs.fs_nr))
1059 		return DNHT_SCAN_END;
1060 	ufs->oid.id = (fs->fs.flags & DN_QHT_HASH) ?
1061 		dn_ht_entries(fs->qht) : (fs->qht ? 1 : 0);
1062 	if (flags) {	/* copy queues */
1063 		copy_q(a, fs, 0);
1064 	}
1065 	return 0;
1066 }
1067 
1068 static int
1069 copy_si_cb(void *obj, void *arg)
1070 {
1071 	struct dn_sch_inst *si = obj;
1072 	struct copy_args *a = arg;
1073 	struct dn_flow *ni = (struct dn_flow *)(*a->start);
1074 	if (copy_obj(a->start, a->end, &si->ni, "inst",
1075 			si->sched->sch.sched_nr))
1076 		return DNHT_SCAN_END;
1077 	ni->oid.type = DN_FLOW; /* override the DN_SCH_I */
1078 	ni->oid.id = si_hash((uintptr_t)si, DNHT_KEY_IS_OBJ, NULL);
1079 	return 0;
1080 }
1081 
1082 static int
1083 copy_si(struct copy_args *a, struct dn_schk *s, int flags)
1084 {
1085 	if (s->sch.flags & DN_HAVE_MASK)
1086 		dn_ht_scan(s->siht, copy_si_cb, a);
1087 	else if (s->siht)
1088 		copy_si_cb(s->siht, a);
1089 	return 0;
1090 }
1091 
1092 /*
1093  * compute a list of children of a scheduler and copy up
1094  */
1095 static int
1096 copy_fsk_list(struct copy_args *a, struct dn_schk *s, int flags)
1097 {
1098 	struct dn_fsk *fs;
1099 	struct dn_id *o;
1100 	uint32_t *p;
1101 
1102 	int n = 0, space = sizeof(*o);
1103 	SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1104 		if (fs->fs.fs_nr < DN_MAX_ID)
1105 			n++;
1106 	}
1107 	space += n * sizeof(uint32_t);
1108 	DX(3, "sched %d has %d flowsets", s->sch.sched_nr, n);
1109 	if (a->end - *(a->start) < space)
1110 		return DNHT_SCAN_END;
1111 	o = (struct dn_id *)(*(a->start));
1112 	o->len = space;
1113 	*a->start += o->len;
1114 	o->type = DN_TEXT;
1115 	p = (uint32_t *)(o+1);
1116 	SLIST_FOREACH(fs, &s->fsk_list, sch_chain)
1117 		if (fs->fs.fs_nr < DN_MAX_ID)
1118 			*p++ = fs->fs.fs_nr;
1119 	return 0;
1120 }
1121 
1122 static int
1123 copy_data_helper(void *_o, void *_arg)
1124 {
1125 	struct copy_args *a = _arg;
1126 	uint32_t *r = a->extra->r; /* start of first range */
1127 	uint32_t *lim;	/* first invalid pointer */
1128 	int n;
1129 
1130 	lim = (uint32_t *)((char *)(a->extra) + a->extra->o.len);
1131 
1132 	if (a->type == DN_LINK || a->type == DN_SCH) {
1133 		/* pipe|sched show, we receive a dn_schk */
1134 		struct dn_schk *s = _o;
1135 
1136 		n = s->sch.sched_nr;
1137 		if (a->type == DN_SCH && n >= DN_MAX_ID)
1138 			return 0;	/* not a scheduler */
1139 		if (a->type == DN_LINK && n <= DN_MAX_ID)
1140 		    return 0;	/* not a pipe */
1141 
1142 		/* see if the object is within one of our ranges */
1143 		for (;r < lim; r += 2) {
1144 			if (n < r[0] || n > r[1])
1145 				continue;
1146 			/* Found a valid entry, copy and we are done */
1147 			if (a->flags & DN_C_LINK) {
1148 				if (copy_obj(a->start, a->end,
1149 				    &s->link, "link", n))
1150 					return DNHT_SCAN_END;
1151 				if (copy_profile(a, s->profile))
1152 					return DNHT_SCAN_END;
1153 				if (copy_flowset(a, s->fs, 0))
1154 					return DNHT_SCAN_END;
1155 			}
1156 			if (a->flags & DN_C_SCH) {
1157 				if (copy_obj(a->start, a->end,
1158 				    &s->sch, "sched", n))
1159 					return DNHT_SCAN_END;
1160 				/* list all attached flowsets */
1161 				if (copy_fsk_list(a, s, 0))
1162 					return DNHT_SCAN_END;
1163 			}
1164 			if (a->flags & DN_C_FLOW)
1165 				copy_si(a, s, 0);
1166 			break;
1167 		}
1168 	} else if (a->type == DN_FS) {
1169 		/* queue show, skip internal flowsets */
1170 		struct dn_fsk *fs = _o;
1171 
1172 		n = fs->fs.fs_nr;
1173 		if (n >= DN_MAX_ID)
1174 			return 0;
1175 		/* see if the object is within one of our ranges */
1176 		for (;r < lim; r += 2) {
1177 			if (n < r[0] || n > r[1])
1178 				continue;
1179 			if (copy_flowset(a, fs, 0))
1180 				return DNHT_SCAN_END;
1181 			copy_q(a, fs, 0);
1182 			break; /* we are done */
1183 		}
1184 	}
1185 	return 0;
1186 }
1187 
1188 static inline struct dn_schk *
1189 locate_scheduler(int i)
1190 {
1191 	return dn_ht_find(V_dn_cfg.schedhash, i, 0, NULL);
1192 }
1193 
1194 /*
1195  * red parameters are in fixed point arithmetic.
1196  */
1197 static int
1198 config_red(struct dn_fsk *fs)
1199 {
1200 	int64_t s, idle, weight, w0;
1201 	int t, i;
1202 
1203 	fs->w_q = fs->fs.w_q;
1204 	fs->max_p = fs->fs.max_p;
1205 	ND("called");
1206 	/* Doing stuff that was in userland */
1207 	i = fs->sched->link.bandwidth;
1208 	s = (i <= 0) ? 0 :
1209 		hz * V_dn_cfg.red_avg_pkt_size * 8 * SCALE(1) / i;
1210 
1211 	idle = div64((s * 3) , fs->w_q); /* s, fs->w_q scaled; idle not scaled */
1212 	fs->lookup_step = div64(idle , V_dn_cfg.red_lookup_depth);
1213 	/* fs->lookup_step not scaled, */
1214 	if (!fs->lookup_step)
1215 		fs->lookup_step = 1;
1216 	w0 = weight = SCALE(1) - fs->w_q; //fs->w_q scaled
1217 
1218 	for (t = fs->lookup_step; t > 1; --t)
1219 		weight = SCALE_MUL(weight, w0);
1220 	fs->lookup_weight = (int)(weight); // scaled
1221 
1222 	/* Now doing stuff that was in kerneland */
1223 	fs->min_th = SCALE(fs->fs.min_th);
1224 	fs->max_th = SCALE(fs->fs.max_th);
1225 
1226 	if (fs->fs.max_th == fs->fs.min_th)
1227 		fs->c_1 = fs->max_p;
1228 	else
1229 		fs->c_1 = SCALE((int64_t)(fs->max_p)) / (fs->fs.max_th - fs->fs.min_th);
1230 	fs->c_2 = SCALE_MUL(fs->c_1, SCALE(fs->fs.min_th));
1231 
1232 	if (fs->fs.flags & DN_IS_GENTLE_RED) {
1233 		fs->c_3 = (SCALE(1) - fs->max_p) / fs->fs.max_th;
1234 		fs->c_4 = SCALE(1) - 2 * fs->max_p;
1235 	}
1236 
1237 	/* If the lookup table already exist, free and create it again. */
1238 	free(fs->w_q_lookup, M_DUMMYNET);
1239 	fs->w_q_lookup = NULL;
1240 	if (V_dn_cfg.red_lookup_depth == 0) {
1241 		printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1242 		    "must be > 0\n");
1243 		fs->fs.flags &= ~DN_IS_RED;
1244 		fs->fs.flags &= ~DN_IS_GENTLE_RED;
1245 		return (EINVAL);
1246 	}
1247 	fs->lookup_depth = V_dn_cfg.red_lookup_depth;
1248 	fs->w_q_lookup = (u_int *)malloc(fs->lookup_depth * sizeof(int),
1249 	    M_DUMMYNET, M_NOWAIT);
1250 	if (fs->w_q_lookup == NULL) {
1251 		printf("dummynet: sorry, cannot allocate red lookup table\n");
1252 		fs->fs.flags &= ~DN_IS_RED;
1253 		fs->fs.flags &= ~DN_IS_GENTLE_RED;
1254 		return(ENOSPC);
1255 	}
1256 
1257 	/* Fill the lookup table with (1 - w_q)^x */
1258 	fs->w_q_lookup[0] = SCALE(1) - fs->w_q;
1259 
1260 	for (i = 1; i < fs->lookup_depth; i++)
1261 		fs->w_q_lookup[i] =
1262 		    SCALE_MUL(fs->w_q_lookup[i - 1], fs->lookup_weight);
1263 
1264 	if (V_dn_cfg.red_avg_pkt_size < 1)
1265 		V_dn_cfg.red_avg_pkt_size = 512;
1266 	fs->avg_pkt_size = V_dn_cfg.red_avg_pkt_size;
1267 	if (V_dn_cfg.red_max_pkt_size < 1)
1268 		V_dn_cfg.red_max_pkt_size = 1500;
1269 	fs->max_pkt_size = V_dn_cfg.red_max_pkt_size;
1270 	ND("exit");
1271 	return 0;
1272 }
1273 
1274 /* Scan all flowset attached to this scheduler and update red */
1275 static void
1276 update_red(struct dn_schk *s)
1277 {
1278 	struct dn_fsk *fs;
1279 	SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1280 		if (fs && (fs->fs.flags & DN_IS_RED))
1281 			config_red(fs);
1282 	}
1283 }
1284 
1285 /* attach flowset to scheduler s, possibly requeue */
1286 static void
1287 fsk_attach(struct dn_fsk *fs, struct dn_schk *s)
1288 {
1289 	ND("remove fs %d from fsunlinked, link to sched %d",
1290 		fs->fs.fs_nr, s->sch.sched_nr);
1291 	SLIST_REMOVE(&V_dn_cfg.fsu, fs, dn_fsk, sch_chain);
1292 	fs->sched = s;
1293 	SLIST_INSERT_HEAD(&s->fsk_list, fs, sch_chain);
1294 	if (s->fp->new_fsk)
1295 		s->fp->new_fsk(fs);
1296 	/* XXX compute fsk_mask */
1297 	fs->fsk_mask = fs->fs.flow_mask;
1298 	if (fs->sched->sch.flags & DN_HAVE_MASK)
1299 		flow_id_or(&fs->sched->sch.sched_mask, &fs->fsk_mask);
1300 	if (fs->qht) {
1301 		/*
1302 		 * we must drain qht according to the old
1303 		 * type, and reinsert according to the new one.
1304 		 * The requeue is complex -- in general we need to
1305 		 * reclassify every single packet.
1306 		 * For the time being, let's hope qht is never set
1307 		 * when we reach this point.
1308 		 */
1309 		D("XXX TODO requeue from fs %d to sch %d",
1310 			fs->fs.fs_nr, s->sch.sched_nr);
1311 		fs->qht = NULL;
1312 	}
1313 	/* set the new type for qht */
1314 	if (nonzero_mask(&fs->fsk_mask))
1315 		fs->fs.flags |= DN_QHT_HASH;
1316 	else
1317 		fs->fs.flags &= ~DN_QHT_HASH;
1318 
1319 	/* XXX config_red() can fail... */
1320 	if (fs->fs.flags & DN_IS_RED)
1321 		config_red(fs);
1322 }
1323 
1324 /* update all flowsets which may refer to this scheduler */
1325 static void
1326 update_fs(struct dn_schk *s)
1327 {
1328 	struct dn_fsk *fs, *tmp;
1329 
1330 	SLIST_FOREACH_SAFE(fs, &V_dn_cfg.fsu, sch_chain, tmp) {
1331 		if (s->sch.sched_nr != fs->fs.sched_nr) {
1332 			D("fs %d for sch %d not %d still unlinked",
1333 				fs->fs.fs_nr, fs->fs.sched_nr,
1334 				s->sch.sched_nr);
1335 			continue;
1336 		}
1337 		fsk_attach(fs, s);
1338 	}
1339 }
1340 
1341 #ifdef NEW_AQM
1342 /* Retrieve AQM configurations to ipfw userland
1343  */
1344 static int
1345 get_aqm_parms(struct sockopt *sopt)
1346 {
1347 	struct dn_extra_parms  *ep;
1348 	struct dn_fsk *fs;
1349 	size_t sopt_valsize;
1350 	int l, err = 0;
1351 
1352 	sopt_valsize = sopt->sopt_valsize;
1353 	l = sizeof(*ep);
1354 	if (sopt->sopt_valsize < l) {
1355 		D("bad len sopt->sopt_valsize %d len %d",
1356 			(int) sopt->sopt_valsize , l);
1357 		err = EINVAL;
1358 		return err;
1359 	}
1360 	ep = malloc(l, M_DUMMYNET, M_NOWAIT);
1361 	if(!ep) {
1362 		err = ENOMEM ;
1363 		return err;
1364 	}
1365 	do {
1366 		err = sooptcopyin(sopt, ep, l, l);
1367 		if(err)
1368 			break;
1369 		sopt->sopt_valsize = sopt_valsize;
1370 		if (ep->oid.len < l) {
1371 			err = EINVAL;
1372 			break;
1373 		}
1374 
1375 		fs = dn_ht_find(V_dn_cfg.fshash, ep->nr, 0, NULL);
1376 		if (!fs) {
1377 			D("fs %d not found", ep->nr);
1378 			err = EINVAL;
1379 			break;
1380 		}
1381 
1382 		if (fs->aqmfp && fs->aqmfp->getconfig) {
1383 			if(fs->aqmfp->getconfig(fs, ep)) {
1384 				D("Error while trying to get AQM params");
1385 				err = EINVAL;
1386 				break;
1387 			}
1388 			ep->oid.len = l;
1389 			err = sooptcopyout(sopt, ep, l);
1390 		}
1391 	}while(0);
1392 
1393 	free(ep, M_DUMMYNET);
1394 	return err;
1395 }
1396 
1397 /* Retrieve AQM configurations to ipfw userland
1398  */
1399 static int
1400 get_sched_parms(struct sockopt *sopt)
1401 {
1402 	struct dn_extra_parms  *ep;
1403 	struct dn_schk *schk;
1404 	size_t sopt_valsize;
1405 	int l, err = 0;
1406 
1407 	sopt_valsize = sopt->sopt_valsize;
1408 	l = sizeof(*ep);
1409 	if (sopt->sopt_valsize < l) {
1410 		D("bad len sopt->sopt_valsize %d len %d",
1411 			(int) sopt->sopt_valsize , l);
1412 		err = EINVAL;
1413 		return err;
1414 	}
1415 	ep = malloc(l, M_DUMMYNET, M_NOWAIT);
1416 	if(!ep) {
1417 		err = ENOMEM ;
1418 		return err;
1419 	}
1420 	do {
1421 		err = sooptcopyin(sopt, ep, l, l);
1422 		if(err)
1423 			break;
1424 		sopt->sopt_valsize = sopt_valsize;
1425 		if (ep->oid.len < l) {
1426 			err = EINVAL;
1427 			break;
1428 		}
1429 
1430 		schk = locate_scheduler(ep->nr);
1431 		if (!schk) {
1432 			D("sched %d not found", ep->nr);
1433 			err = EINVAL;
1434 			break;
1435 		}
1436 
1437 		if (schk->fp && schk->fp->getconfig) {
1438 			if(schk->fp->getconfig(schk, ep)) {
1439 				D("Error while trying to get sched params");
1440 				err = EINVAL;
1441 				break;
1442 			}
1443 			ep->oid.len = l;
1444 			err = sooptcopyout(sopt, ep, l);
1445 		}
1446 	}while(0);
1447 	free(ep, M_DUMMYNET);
1448 
1449 	return err;
1450 }
1451 
1452 /* Configure AQM for flowset 'fs'.
1453  * extra parameters are passed from userland.
1454  */
1455 static int
1456 config_aqm(struct dn_fsk *fs, struct  dn_extra_parms *ep, int busy)
1457 {
1458 	int err = 0;
1459 
1460 	NET_EPOCH_ASSERT();
1461 
1462 	do {
1463 		/* no configurations */
1464 		if (!ep) {
1465 			err = 0;
1466 			break;
1467 		}
1468 
1469 		/* no AQM for this flowset*/
1470 		if (!strcmp(ep->name,"")) {
1471 			err = 0;
1472 			break;
1473 		}
1474 		if (ep->oid.len < sizeof(*ep)) {
1475 			D("short aqm len %d", ep->oid.len);
1476 				err = EINVAL;
1477 				break;
1478 		}
1479 
1480 		if (busy) {
1481 			D("Unable to configure flowset, flowset busy!");
1482 			err = EINVAL;
1483 			break;
1484 		}
1485 
1486 		/* deconfigure old aqm if exist */
1487 		if (fs->aqmcfg && fs->aqmfp && fs->aqmfp->deconfig) {
1488 			aqm_cleanup_deconfig_fs(fs);
1489 		}
1490 
1491 		if (!(fs->aqmfp = find_aqm_type(0, ep->name))) {
1492 			D("AQM functions not found for type %s!", ep->name);
1493 			fs->fs.flags &= ~DN_IS_AQM;
1494 			err = EINVAL;
1495 			break;
1496 		} else
1497 			fs->fs.flags |= DN_IS_AQM;
1498 
1499 		if (ep->oid.subtype != DN_AQM_PARAMS) {
1500 				D("Wrong subtype");
1501 				err = EINVAL;
1502 				break;
1503 		}
1504 
1505 		if (fs->aqmfp->config) {
1506 			err = fs->aqmfp->config(fs, ep, ep->oid.len);
1507 			if (err) {
1508 					D("Unable to configure AQM for FS %d", fs->fs.fs_nr );
1509 					fs->fs.flags &= ~DN_IS_AQM;
1510 					fs->aqmfp = NULL;
1511 					break;
1512 			}
1513 		}
1514 	} while(0);
1515 
1516 	return err;
1517 }
1518 #endif
1519 
1520 /*
1521  * Configuration -- to preserve backward compatibility we use
1522  * the following scheme (N is 65536)
1523  *	NUMBER		SCHED	LINK	FLOWSET
1524  *	   1 ..  N-1	(1)WFQ	(2)WFQ	(3)queue
1525  *	 N+1 .. 2N-1	(4)FIFO (5)FIFO	(6)FIFO for sched 1..N-1
1526  *	2N+1 .. 3N-1	--	--	(7)FIFO for sched N+1..2N-1
1527  *
1528  * "pipe i config" configures #1, #2 and #3
1529  * "sched i config" configures #1 and possibly #6
1530  * "queue i config" configures #3
1531  * #1 is configured with 'pipe i config' or 'sched i config'
1532  * #2 is configured with 'pipe i config', and created if not
1533  *	existing with 'sched i config'
1534  * #3 is configured with 'queue i config'
1535  * #4 is automatically configured after #1, can only be FIFO
1536  * #5 is automatically configured after #2
1537  * #6 is automatically created when #1 is !MULTIQUEUE,
1538  *	and can be updated.
1539  * #7 is automatically configured after #2
1540  */
1541 
1542 /*
1543  * configure a link (and its FIFO instance)
1544  */
1545 static int
1546 config_link(struct dn_link *p, struct dn_id *arg)
1547 {
1548 	int i;
1549 
1550 	if (p->oid.len != sizeof(*p)) {
1551 		D("invalid pipe len %d", p->oid.len);
1552 		return EINVAL;
1553 	}
1554 	i = p->link_nr;
1555 	if (i <= 0 || i >= DN_MAX_ID)
1556 		return EINVAL;
1557 	/*
1558 	 * The config program passes parameters as follows:
1559 	 * bw = bits/second (0 means no limits),
1560 	 * delay = ms, must be translated into ticks.
1561 	 * qsize = slots/bytes
1562 	 * burst ???
1563 	 */
1564 	p->delay = (p->delay * hz) / 1000;
1565 	/* Scale burst size: bytes -> bits * hz */
1566 	p->burst *= 8 * hz;
1567 
1568 	DN_BH_WLOCK();
1569 	/* do it twice, base link and FIFO link */
1570 	for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1571 	    struct dn_schk *s = locate_scheduler(i);
1572 	    if (s == NULL) {
1573 		DN_BH_WUNLOCK();
1574 		D("sched %d not found", i);
1575 		return EINVAL;
1576 	    }
1577 	    /* remove profile if exists */
1578 	    free(s->profile, M_DUMMYNET);
1579 	    s->profile = NULL;
1580 
1581 	    /* copy all parameters */
1582 	    s->link.oid = p->oid;
1583 	    s->link.link_nr = i;
1584 	    s->link.delay = p->delay;
1585 	    if (s->link.bandwidth != p->bandwidth) {
1586 		/* XXX bandwidth changes, need to update red params */
1587 	    s->link.bandwidth = p->bandwidth;
1588 		update_red(s);
1589 	    }
1590 	    s->link.burst = p->burst;
1591 	    schk_reset_credit(s);
1592 	}
1593 	V_dn_cfg.id++;
1594 	DN_BH_WUNLOCK();
1595 	return 0;
1596 }
1597 
1598 /*
1599  * configure a flowset. Can be called from inside with locked=1,
1600  */
1601 static struct dn_fsk *
1602 config_fs(struct dn_fs *nfs, struct dn_id *arg, int locked)
1603 {
1604 	int i;
1605 	struct dn_fsk *fs;
1606 #ifdef NEW_AQM
1607 	struct dn_extra_parms *ep;
1608 #endif
1609 
1610 	if (nfs->oid.len != sizeof(*nfs)) {
1611 		D("invalid flowset len %d", nfs->oid.len);
1612 		return NULL;
1613 	}
1614 	i = nfs->fs_nr;
1615 	if (i <= 0 || i >= 3*DN_MAX_ID)
1616 		return NULL;
1617 #ifdef NEW_AQM
1618 	ep = NULL;
1619 	if (arg != NULL) {
1620 		ep = malloc(sizeof(*ep), M_TEMP, M_NOWAIT);
1621 		if (ep == NULL)
1622 			return (NULL);
1623 		memcpy(ep, arg, sizeof(*ep));
1624 	}
1625 #endif
1626 	ND("flowset %d", i);
1627 	/* XXX other sanity checks */
1628         if (nfs->flags & DN_QSIZE_BYTES) {
1629 		ipdn_bound_var(&nfs->qsize, 16384,
1630 		    1500, V_dn_cfg.byte_limit, NULL); // "queue byte size");
1631         } else {
1632 		ipdn_bound_var(&nfs->qsize, 50,
1633 		    1, V_dn_cfg.slot_limit, NULL); // "queue slot size");
1634         }
1635 	if (nfs->flags & DN_HAVE_MASK) {
1636 		/* make sure we have some buckets */
1637 		ipdn_bound_var((int *)&nfs->buckets, V_dn_cfg.hash_size,
1638 			1, V_dn_cfg.max_hash_size, "flowset buckets");
1639 	} else {
1640 		nfs->buckets = 1;	/* we only need 1 */
1641 	}
1642 	if (!locked)
1643 		DN_BH_WLOCK();
1644 	do { /* exit with break when done */
1645 	    struct dn_schk *s;
1646 	    int flags = nfs->sched_nr ? DNHT_INSERT : 0;
1647 	    int j;
1648 	    int oldc = V_dn_cfg.fsk_count;
1649 	    fs = dn_ht_find(V_dn_cfg.fshash, i, flags, NULL);
1650 	    if (fs == NULL) {
1651 		D("missing sched for flowset %d", i);
1652 	        break;
1653 	    }
1654 	    /* grab some defaults from the existing one */
1655 	    if (nfs->sched_nr == 0) /* reuse */
1656 		nfs->sched_nr = fs->fs.sched_nr;
1657 	    for (j = 0; j < sizeof(nfs->par)/sizeof(nfs->par[0]); j++) {
1658 		if (nfs->par[j] == -1) /* reuse */
1659 		    nfs->par[j] = fs->fs.par[j];
1660 	    }
1661 	    if (bcmp(&fs->fs, nfs, sizeof(*nfs)) == 0) {
1662 		ND("flowset %d unchanged", i);
1663 #ifdef NEW_AQM
1664 		if (ep != NULL) {
1665 			/*
1666 			 * Reconfigure AQM as the parameters can be changed.
1667 			 * We consider the flowset as busy if it has scheduler
1668 			 * instance(s).
1669 			 */
1670 			s = locate_scheduler(nfs->sched_nr);
1671 			config_aqm(fs, ep, s != NULL && s->siht != NULL);
1672 		}
1673 #endif
1674 		break; /* no change, nothing to do */
1675 	    }
1676 	    if (oldc != V_dn_cfg.fsk_count)	/* new item */
1677 		V_dn_cfg.id++;
1678 	    s = locate_scheduler(nfs->sched_nr);
1679 	    /* detach from old scheduler if needed, preserving
1680 	     * queues if we need to reattach. Then update the
1681 	     * configuration, and possibly attach to the new sched.
1682 	     */
1683 	    DX(2, "fs %d changed sched %d@%p to %d@%p",
1684 		fs->fs.fs_nr,
1685 		fs->fs.sched_nr, fs->sched, nfs->sched_nr, s);
1686 	    if (fs->sched) {
1687 		int flags = s ? DN_DETACH : (DN_DETACH | DN_DESTROY);
1688 		flags |= DN_DESTROY; /* XXX temporary */
1689 		fsk_detach(fs, flags);
1690 	    }
1691 	    fs->fs = *nfs; /* copy configuration */
1692 #ifdef NEW_AQM
1693 			fs->aqmfp = NULL;
1694 			if (ep != NULL)
1695 				config_aqm(fs, ep, s != NULL &&
1696 				    s->siht != NULL);
1697 #endif
1698 	    if (s != NULL)
1699 		fsk_attach(fs, s);
1700 	} while (0);
1701 	if (!locked)
1702 		DN_BH_WUNLOCK();
1703 #ifdef NEW_AQM
1704 	free(ep, M_TEMP);
1705 #endif
1706 	return fs;
1707 }
1708 
1709 /*
1710  * config/reconfig a scheduler and its FIFO variant.
1711  * For !MULTIQUEUE schedulers, also set up the flowset.
1712  *
1713  * On reconfigurations (detected because s->fp is set),
1714  * detach existing flowsets preserving traffic, preserve link,
1715  * and delete the old scheduler creating a new one.
1716  */
1717 static int
1718 config_sched(struct dn_sch *_nsch, struct dn_id *arg)
1719 {
1720 	struct dn_schk *s;
1721 	struct schk_new_arg a; /* argument for schk_new */
1722 	int i;
1723 	struct dn_link p;	/* copy of oldlink */
1724 	struct dn_profile *pf = NULL;	/* copy of old link profile */
1725 	/* Used to preserv mask parameter */
1726 	struct ipfw_flow_id new_mask;
1727 	int new_buckets = 0;
1728 	int new_flags = 0;
1729 	int pipe_cmd;
1730 	int err = ENOMEM;
1731 
1732 	NET_EPOCH_ASSERT();
1733 
1734 	a.sch = _nsch;
1735 	if (a.sch->oid.len != sizeof(*a.sch)) {
1736 		D("bad sched len %d", a.sch->oid.len);
1737 		return EINVAL;
1738 	}
1739 	i = a.sch->sched_nr;
1740 	if (i <= 0 || i >= DN_MAX_ID)
1741 		return EINVAL;
1742 	/* make sure we have some buckets */
1743 	if (a.sch->flags & DN_HAVE_MASK)
1744 		ipdn_bound_var((int *)&a.sch->buckets, V_dn_cfg.hash_size,
1745 			1, V_dn_cfg.max_hash_size, "sched buckets");
1746 	/* XXX other sanity checks */
1747 	bzero(&p, sizeof(p));
1748 
1749 	pipe_cmd = a.sch->flags & DN_PIPE_CMD;
1750 	a.sch->flags &= ~DN_PIPE_CMD; //XXX do it even if is not set?
1751 	if (pipe_cmd) {
1752 		/* Copy mask parameter */
1753 		new_mask = a.sch->sched_mask;
1754 		new_buckets = a.sch->buckets;
1755 		new_flags = a.sch->flags;
1756 	}
1757 	DN_BH_WLOCK();
1758 again: /* run twice, for wfq and fifo */
1759 	/*
1760 	 * lookup the type. If not supplied, use the previous one
1761 	 * or default to WF2Q+. Otherwise, return an error.
1762 	 */
1763 	V_dn_cfg.id++;
1764 	a.fp = find_sched_type(a.sch->oid.subtype, a.sch->name);
1765 	if (a.fp != NULL) {
1766 		/* found. Lookup or create entry */
1767 		s = dn_ht_find(V_dn_cfg.schedhash, i, DNHT_INSERT, &a);
1768 	} else if (a.sch->oid.subtype == 0 && !a.sch->name[0]) {
1769 		/* No type. search existing s* or retry with WF2Q+ */
1770 		s = dn_ht_find(V_dn_cfg.schedhash, i, 0, &a);
1771 		if (s != NULL) {
1772 			a.fp = s->fp;
1773 			/* Scheduler exists, skip to FIFO scheduler
1774 			 * if command was pipe config...
1775 			 */
1776 			if (pipe_cmd)
1777 				goto next;
1778 		} else {
1779 			/* New scheduler, create a wf2q+ with no mask
1780 			 * if command was pipe config...
1781 			 */
1782 			if (pipe_cmd) {
1783 				/* clear mask parameter */
1784 				bzero(&a.sch->sched_mask, sizeof(new_mask));
1785 				a.sch->buckets = 0;
1786 				a.sch->flags &= ~DN_HAVE_MASK;
1787 			}
1788 			a.sch->oid.subtype = DN_SCHED_WF2QP;
1789 			goto again;
1790 		}
1791 	} else {
1792 		D("invalid scheduler type %d %s",
1793 			a.sch->oid.subtype, a.sch->name);
1794 		err = EINVAL;
1795 		goto error;
1796 	}
1797 	/* normalize name and subtype */
1798 	a.sch->oid.subtype = a.fp->type;
1799 	bzero(a.sch->name, sizeof(a.sch->name));
1800 	strlcpy(a.sch->name, a.fp->name, sizeof(a.sch->name));
1801 	if (s == NULL) {
1802 		D("cannot allocate scheduler %d", i);
1803 		goto error;
1804 	}
1805 	/* restore existing link if any */
1806 	if (p.link_nr) {
1807 		s->link = p;
1808 		if (!pf || pf->link_nr != p.link_nr) { /* no saved value */
1809 			s->profile = NULL; /* XXX maybe not needed */
1810 		} else {
1811 			s->profile = malloc(sizeof(struct dn_profile),
1812 					     M_DUMMYNET, M_NOWAIT | M_ZERO);
1813 			if (s->profile == NULL) {
1814 				D("cannot allocate profile");
1815 				goto error; //XXX
1816 			}
1817 			memcpy(s->profile, pf, sizeof(*pf));
1818 		}
1819 	}
1820 	p.link_nr = 0;
1821 	if (s->fp == NULL) {
1822 		DX(2, "sched %d new type %s", i, a.fp->name);
1823 	} else if (s->fp != a.fp ||
1824 			bcmp(a.sch, &s->sch, sizeof(*a.sch)) ) {
1825 		/* already existing. */
1826 		DX(2, "sched %d type changed from %s to %s",
1827 			i, s->fp->name, a.fp->name);
1828 		DX(4, "   type/sub %d/%d -> %d/%d",
1829 			s->sch.oid.type, s->sch.oid.subtype,
1830 			a.sch->oid.type, a.sch->oid.subtype);
1831 		if (s->link.link_nr == 0)
1832 			D("XXX WARNING link 0 for sched %d", i);
1833 		p = s->link;	/* preserve link */
1834 		if (s->profile) {/* preserve profile */
1835 			if (!pf)
1836 				pf = malloc(sizeof(*pf),
1837 				    M_DUMMYNET, M_NOWAIT | M_ZERO);
1838 			if (pf)	/* XXX should issue a warning otherwise */
1839 				memcpy(pf, s->profile, sizeof(*pf));
1840 		}
1841 		/* remove from the hash */
1842 		dn_ht_find(V_dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
1843 		/* Detach flowsets, preserve queues. */
1844 		// schk_delete_cb(s, NULL);
1845 		// XXX temporarily, kill queues
1846 		schk_delete_cb(s, (void *)DN_DESTROY);
1847 		goto again;
1848 	} else {
1849 		DX(4, "sched %d unchanged type %s", i, a.fp->name);
1850 	}
1851 	/* complete initialization */
1852 	s->sch = *a.sch;
1853 	s->fp = a.fp;
1854 	s->cfg = arg;
1855 	// XXX schk_reset_credit(s);
1856 	/* create the internal flowset if needed,
1857 	 * trying to reuse existing ones if available
1858 	 */
1859 	if (!(s->fp->flags & DN_MULTIQUEUE) && !s->fs) {
1860 	        s->fs = dn_ht_find(V_dn_cfg.fshash, i, 0, NULL);
1861 		if (!s->fs) {
1862 			struct dn_fs fs;
1863 			bzero(&fs, sizeof(fs));
1864 			set_oid(&fs.oid, DN_FS, sizeof(fs));
1865 			fs.fs_nr = i + DN_MAX_ID;
1866 			fs.sched_nr = i;
1867 			s->fs = config_fs(&fs, NULL, 1 /* locked */);
1868 		}
1869 		if (!s->fs) {
1870 			schk_delete_cb(s, (void *)DN_DESTROY);
1871 			D("error creating internal fs for %d", i);
1872 			goto error;
1873 		}
1874 	}
1875 	/* call init function after the flowset is created */
1876 	if (s->fp->config)
1877 		s->fp->config(s);
1878 	update_fs(s);
1879 next:
1880 	if (i < DN_MAX_ID) { /* now configure the FIFO instance */
1881 		i += DN_MAX_ID;
1882 		if (pipe_cmd) {
1883 			/* Restore mask parameter for FIFO */
1884 			a.sch->sched_mask = new_mask;
1885 			a.sch->buckets = new_buckets;
1886 			a.sch->flags = new_flags;
1887 		} else {
1888 			/* sched config shouldn't modify the FIFO scheduler */
1889 			if (dn_ht_find(V_dn_cfg.schedhash, i, 0, &a) != NULL) {
1890 				/* FIFO already exist, don't touch it */
1891 				err = 0; /* and this is not an error */
1892 				goto error;
1893 			}
1894 		}
1895 		a.sch->sched_nr = i;
1896 		a.sch->oid.subtype = DN_SCHED_FIFO;
1897 		bzero(a.sch->name, sizeof(a.sch->name));
1898 		goto again;
1899 	}
1900 	err = 0;
1901 error:
1902 	DN_BH_WUNLOCK();
1903 	free(pf, M_DUMMYNET);
1904 	return err;
1905 }
1906 
1907 /*
1908  * attach a profile to a link
1909  */
1910 static int
1911 config_profile(struct dn_profile *pf, struct dn_id *arg)
1912 {
1913 	struct dn_schk *s;
1914 	int i, olen, err = 0;
1915 
1916 	if (pf->oid.len < sizeof(*pf)) {
1917 		D("short profile len %d", pf->oid.len);
1918 		return EINVAL;
1919 	}
1920 	i = pf->link_nr;
1921 	if (i <= 0 || i >= DN_MAX_ID)
1922 		return EINVAL;
1923 	/* XXX other sanity checks */
1924 	DN_BH_WLOCK();
1925 	for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1926 		s = locate_scheduler(i);
1927 
1928 		if (s == NULL) {
1929 			err = EINVAL;
1930 			break;
1931 		}
1932 		V_dn_cfg.id++;
1933 		/*
1934 		 * If we had a profile and the new one does not fit,
1935 		 * or it is deleted, then we need to free memory.
1936 		 */
1937 		if (s->profile && (pf->samples_no == 0 ||
1938 		    s->profile->oid.len < pf->oid.len)) {
1939 			free(s->profile, M_DUMMYNET);
1940 			s->profile = NULL;
1941 		}
1942 		if (pf->samples_no == 0)
1943 			continue;
1944 		/*
1945 		 * new profile, possibly allocate memory
1946 		 * and copy data.
1947 		 */
1948 		if (s->profile == NULL)
1949 			s->profile = malloc(pf->oid.len,
1950 				    M_DUMMYNET, M_NOWAIT | M_ZERO);
1951 		if (s->profile == NULL) {
1952 			D("no memory for profile %d", i);
1953 			err = ENOMEM;
1954 			break;
1955 		}
1956 		/* preserve larger length XXX double check */
1957 		olen = s->profile->oid.len;
1958 		if (olen < pf->oid.len)
1959 			olen = pf->oid.len;
1960 		memcpy(s->profile, pf, pf->oid.len);
1961 		s->profile->oid.len = olen;
1962 	}
1963 	DN_BH_WUNLOCK();
1964 	return err;
1965 }
1966 
1967 /*
1968  * Delete all objects:
1969  */
1970 static void
1971 dummynet_flush(void)
1972 {
1973 
1974 	/* delete all schedulers and related links/queues/flowsets */
1975 	dn_ht_scan(V_dn_cfg.schedhash, schk_delete_cb,
1976 		(void *)(uintptr_t)DN_DELETE_FS);
1977 	/* delete all remaining (unlinked) flowsets */
1978 	DX(4, "still %d unlinked fs", V_dn_cfg.fsk_count);
1979 	dn_ht_free(V_dn_cfg.fshash, DNHT_REMOVE);
1980 	fsk_detach_list(&V_dn_cfg.fsu, DN_DELETE_FS);
1981 	/* Reinitialize system heap... */
1982 	heap_init(&V_dn_cfg.evheap, 16, offsetof(struct dn_id, id));
1983 }
1984 
1985 /*
1986  * Main handler for configuration. We are guaranteed to be called
1987  * with an oid which is at least a dn_id.
1988  * - the first object is the command (config, delete, flush, ...)
1989  * - config_link must be issued after the corresponding config_sched
1990  * - parameters (DN_TXT) for an object must precede the object
1991  *   processed on a config_sched.
1992  */
1993 int
1994 do_config(void *p, int l)
1995 {
1996 	struct dn_id o;
1997 	union {
1998 		struct dn_profile profile;
1999 		struct dn_fs fs;
2000 		struct dn_link link;
2001 		struct dn_sch sched;
2002 	} *dn;
2003 	struct dn_id *arg;
2004 	uintptr_t a;
2005 	int err, err2, off;
2006 
2007 	memcpy(&o, p, sizeof(o));
2008 	if (o.id != DN_API_VERSION) {
2009 		D("invalid api version got %d need %d", o.id, DN_API_VERSION);
2010 		return EINVAL;
2011 	}
2012 	arg = NULL;
2013 	dn = NULL;
2014 	for (off = 0; l >= sizeof(o); memcpy(&o, (char *)p + off, sizeof(o))) {
2015 		if (o.len < sizeof(o) || l < o.len) {
2016 			D("bad len o.len %d len %d", o.len, l);
2017 			err = EINVAL;
2018 			break;
2019 		}
2020 		l -= o.len;
2021 		err = 0;
2022 		switch (o.type) {
2023 		default:
2024 			D("cmd %d not implemented", o.type);
2025 			break;
2026 
2027 #ifdef EMULATE_SYSCTL
2028 		/* sysctl emulation.
2029 		 * if we recognize the command, jump to the correct
2030 		 * handler and return
2031 		 */
2032 		case DN_SYSCTL_SET:
2033 			err = kesysctl_emu_set(p, l);
2034 			return err;
2035 #endif
2036 
2037 		case DN_CMD_CONFIG: /* simply a header */
2038 			break;
2039 
2040 		case DN_CMD_DELETE:
2041 			/* the argument is in the first uintptr_t after o */
2042 			if (o.len < sizeof(o) + sizeof(a)) {
2043 				err = EINVAL;
2044 				break;
2045 			}
2046 			memcpy(&a, (char *)p + off + sizeof(o), sizeof(a));
2047 			switch (o.subtype) {
2048 			case DN_LINK:
2049 				/* delete base and derived schedulers */
2050 				DN_BH_WLOCK();
2051 				err = delete_schk(a);
2052 				err2 = delete_schk(a + DN_MAX_ID);
2053 				DN_BH_WUNLOCK();
2054 				if (!err)
2055 					err = err2;
2056 				break;
2057 
2058 			default:
2059 				D("invalid delete type %d", o.subtype);
2060 				err = EINVAL;
2061 				break;
2062 
2063 			case DN_FS:
2064 				err = (a < 1 || a >= DN_MAX_ID) ?
2065 				    EINVAL : delete_fs(a, 0) ;
2066 				break;
2067 			}
2068 			break;
2069 
2070 		case DN_CMD_FLUSH:
2071 			DN_BH_WLOCK();
2072 			dummynet_flush();
2073 			DN_BH_WUNLOCK();
2074 			break;
2075 		case DN_TEXT:	/* store argument of next block */
2076 			free(arg, M_TEMP);
2077 			arg = malloc(o.len, M_TEMP, M_NOWAIT);
2078 			if (arg == NULL) {
2079 				err = ENOMEM;
2080 				break;
2081 			}
2082 			memcpy(arg, (char *)p + off, o.len);
2083 			break;
2084 		case DN_LINK:
2085 			if (dn == NULL)
2086 				dn = malloc(sizeof(*dn), M_TEMP, M_NOWAIT);
2087 			if (dn == NULL) {
2088 				err = ENOMEM;
2089 				break;
2090 			}
2091 			memcpy(&dn->link, (char *)p + off, sizeof(dn->link));
2092 			err = config_link(&dn->link, arg);
2093 			break;
2094 		case DN_PROFILE:
2095 			if (dn == NULL)
2096 				dn = malloc(sizeof(*dn), M_TEMP, M_NOWAIT);
2097 			if (dn == NULL) {
2098 				err = ENOMEM;
2099 				break;
2100 			}
2101 			memcpy(&dn->profile, (char *)p + off,
2102 			    sizeof(dn->profile));
2103 			err = config_profile(&dn->profile, arg);
2104 			break;
2105 		case DN_SCH:
2106 			if (dn == NULL)
2107 				dn = malloc(sizeof(*dn), M_TEMP, M_NOWAIT);
2108 			if (dn == NULL) {
2109 				err = ENOMEM;
2110 				break;
2111 			}
2112 			memcpy(&dn->sched, (char *)p + off,
2113 			    sizeof(dn->sched));
2114 			err = config_sched(&dn->sched, arg);
2115 			break;
2116 		case DN_FS:
2117 			if (dn == NULL)
2118 				dn = malloc(sizeof(*dn), M_TEMP, M_NOWAIT);
2119 			if (dn == NULL) {
2120 				err = ENOMEM;
2121 				break;
2122 			}
2123 			memcpy(&dn->fs, (char *)p + off, sizeof(dn->fs));
2124 			err = (NULL == config_fs(&dn->fs, arg, 0));
2125 			break;
2126 		}
2127 		if (err != 0)
2128 			break;
2129 		off += o.len;
2130 	}
2131 	free(arg, M_TEMP);
2132 	free(dn, M_TEMP);
2133 	return err;
2134 }
2135 
2136 static int
2137 compute_space(struct dn_id *cmd, struct copy_args *a)
2138 {
2139 	int x = 0, need = 0;
2140 	int profile_size = sizeof(struct dn_profile) -
2141 		ED_MAX_SAMPLES_NO*sizeof(int);
2142 
2143 	/* NOTE about compute space:
2144 	 * NP 	= V_dn_cfg.schk_count
2145 	 * NSI 	= V_dn_cfg.si_count
2146 	 * NF 	= V_dn_cfg.fsk_count
2147 	 * NQ 	= V_dn_cfg.queue_count
2148 	 * - ipfw pipe show
2149 	 *   (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
2150 	 *                             link, scheduler template, flowset
2151 	 *                             integrated in scheduler and header
2152 	 *                             for flowset list
2153 	 *   (NSI)*(dn_flow) all scheduler instance (includes
2154 	 *                              the queue instance)
2155 	 * - ipfw sched show
2156 	 *   (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
2157 	 *                             link, scheduler template, flowset
2158 	 *                             integrated in scheduler and header
2159 	 *                             for flowset list
2160 	 *   (NSI * dn_flow) all scheduler instances
2161 	 *   (NF * sizeof(uint_32)) space for flowset list linked to scheduler
2162 	 *   (NQ * dn_queue) all queue [XXXfor now not listed]
2163 	 * - ipfw queue show
2164 	 *   (NF * dn_fs) all flowset
2165 	 *   (NQ * dn_queue) all queues
2166 	 */
2167 	switch (cmd->subtype) {
2168 	default:
2169 		return -1;
2170 	/* XXX where do LINK and SCH differ ? */
2171 	/* 'ipfw sched show' could list all queues associated to
2172 	 * a scheduler. This feature for now is disabled
2173 	 */
2174 	case DN_LINK:	/* pipe show */
2175 		x = DN_C_LINK | DN_C_SCH | DN_C_FLOW;
2176 		need += V_dn_cfg.schk_count *
2177 			(sizeof(struct dn_fs) + profile_size) / 2;
2178 		need += V_dn_cfg.fsk_count * sizeof(uint32_t);
2179 		break;
2180 	case DN_SCH:	/* sched show */
2181 		need += V_dn_cfg.schk_count *
2182 			(sizeof(struct dn_fs) + profile_size) / 2;
2183 		need += V_dn_cfg.fsk_count * sizeof(uint32_t);
2184 		x = DN_C_SCH | DN_C_LINK | DN_C_FLOW;
2185 		break;
2186 	case DN_FS:	/* queue show */
2187 		x = DN_C_FS | DN_C_QUEUE;
2188 		break;
2189 	case DN_GET_COMPAT:	/* compatibility mode */
2190 		need =  dn_compat_calc_size();
2191 		break;
2192 	}
2193 	a->flags = x;
2194 	if (x & DN_C_SCH) {
2195 		need += V_dn_cfg.schk_count * sizeof(struct dn_sch) / 2;
2196 		/* NOT also, each fs might be attached to a sched */
2197 		need += V_dn_cfg.schk_count * sizeof(struct dn_id) / 2;
2198 	}
2199 	if (x & DN_C_FS)
2200 		need += V_dn_cfg.fsk_count * sizeof(struct dn_fs);
2201 	if (x & DN_C_LINK) {
2202 		need += V_dn_cfg.schk_count * sizeof(struct dn_link) / 2;
2203 	}
2204 	/*
2205 	 * When exporting a queue to userland, only pass up the
2206 	 * struct dn_flow, which is the only visible part.
2207 	 */
2208 
2209 	if (x & DN_C_QUEUE)
2210 		need += V_dn_cfg.queue_count * sizeof(struct dn_flow);
2211 	if (x & DN_C_FLOW)
2212 		need += V_dn_cfg.si_count * (sizeof(struct dn_flow));
2213 	return need;
2214 }
2215 
2216 /*
2217  * If compat != NULL dummynet_get is called in compatibility mode.
2218  * *compat will be the pointer to the buffer to pass to ipfw
2219  */
2220 int
2221 dummynet_get(struct sockopt *sopt, void **compat)
2222 {
2223 	int have, i, need, error;
2224 	char *start = NULL, *buf;
2225 	size_t sopt_valsize;
2226 	struct dn_id *cmd;
2227 	struct copy_args a;
2228 	struct copy_range r;
2229 	int l = sizeof(struct dn_id);
2230 
2231 	bzero(&a, sizeof(a));
2232 	bzero(&r, sizeof(r));
2233 
2234 	/* save and restore original sopt_valsize around copyin */
2235 	sopt_valsize = sopt->sopt_valsize;
2236 
2237 	cmd = &r.o;
2238 
2239 	if (!compat) {
2240 		/* copy at least an oid, and possibly a full object */
2241 		error = sooptcopyin(sopt, cmd, sizeof(r), sizeof(*cmd));
2242 		sopt->sopt_valsize = sopt_valsize;
2243 		if (error)
2244 			goto done;
2245 		l = cmd->len;
2246 #ifdef EMULATE_SYSCTL
2247 		/* sysctl emulation. */
2248 		if (cmd->type == DN_SYSCTL_GET)
2249 			return kesysctl_emu_get(sopt);
2250 #endif
2251 		if (l > sizeof(r)) {
2252 			/* request larger than default, allocate buffer */
2253 			cmd = malloc(l,  M_DUMMYNET, M_NOWAIT);
2254 			if (cmd == NULL) {
2255 				error = ENOMEM;
2256 				goto done;
2257 			}
2258 			error = sooptcopyin(sopt, cmd, l, l);
2259 			sopt->sopt_valsize = sopt_valsize;
2260 			if (error)
2261 				goto done;
2262 		}
2263 	} else { /* compatibility */
2264 		error = 0;
2265 		cmd->type = DN_CMD_GET;
2266 		cmd->len = sizeof(struct dn_id);
2267 		cmd->subtype = DN_GET_COMPAT;
2268 		// cmd->id = sopt_valsize;
2269 		D("compatibility mode");
2270 	}
2271 
2272 #ifdef NEW_AQM
2273 	/* get AQM params */
2274 	if(cmd->subtype == DN_AQM_PARAMS) {
2275 		error = get_aqm_parms(sopt);
2276 		goto done;
2277 	/* get Scheduler params */
2278 	} else if (cmd->subtype == DN_SCH_PARAMS) {
2279 		error = get_sched_parms(sopt);
2280 		goto done;
2281 	}
2282 #endif
2283 
2284 	a.extra = (struct copy_range *)cmd;
2285 	if (cmd->len == sizeof(*cmd)) { /* no range, create a default */
2286 		uint32_t *rp = (uint32_t *)(cmd + 1);
2287 		cmd->len += 2* sizeof(uint32_t);
2288 		rp[0] = 1;
2289 		rp[1] = DN_MAX_ID - 1;
2290 		if (cmd->subtype == DN_LINK) {
2291 			rp[0] += DN_MAX_ID;
2292 			rp[1] += DN_MAX_ID;
2293 		}
2294 	}
2295 	/* Count space (under lock) and allocate (outside lock).
2296 	 * Exit with lock held if we manage to get enough buffer.
2297 	 * Try a few times then give up.
2298 	 */
2299 	for (have = 0, i = 0; i < 10; i++) {
2300 		DN_BH_WLOCK();
2301 		need = compute_space(cmd, &a);
2302 
2303 		/* if there is a range, ignore value from compute_space() */
2304 		if (l > sizeof(*cmd))
2305 			need = sopt_valsize - sizeof(*cmd);
2306 
2307 		if (need < 0) {
2308 			DN_BH_WUNLOCK();
2309 			error = EINVAL;
2310 			goto done;
2311 		}
2312 		need += sizeof(*cmd);
2313 		cmd->id = need;
2314 		if (have >= need)
2315 			break;
2316 
2317 		DN_BH_WUNLOCK();
2318 		free(start, M_DUMMYNET);
2319 		start = NULL;
2320 		if (need > sopt_valsize)
2321 			break;
2322 
2323 		have = need;
2324 		start = malloc(have, M_DUMMYNET, M_NOWAIT | M_ZERO);
2325 	}
2326 
2327 	if (start == NULL) {
2328 		if (compat) {
2329 			*compat = NULL;
2330 			error =  1; // XXX
2331 		} else {
2332 			error = sooptcopyout(sopt, cmd, sizeof(*cmd));
2333 		}
2334 		goto done;
2335 	}
2336 	ND("have %d:%d sched %d, %d:%d links %d, %d:%d flowsets %d, "
2337 		"%d:%d si %d, %d:%d queues %d",
2338 		V_dn_cfg.schk_count, sizeof(struct dn_sch), DN_SCH,
2339 		V_dn_cfg.schk_count, sizeof(struct dn_link), DN_LINK,
2340 		V_dn_cfg.fsk_count, sizeof(struct dn_fs), DN_FS,
2341 		V_dn_cfg.si_count, sizeof(struct dn_flow), DN_SCH_I,
2342 		V_dn_cfg.queue_count, sizeof(struct dn_queue), DN_QUEUE);
2343 	sopt->sopt_valsize = sopt_valsize;
2344 	a.type = cmd->subtype;
2345 
2346 	if (compat == NULL) {
2347 		memcpy(start, cmd, sizeof(*cmd));
2348 		((struct dn_id*)(start))->len = sizeof(struct dn_id);
2349 		buf = start + sizeof(*cmd);
2350 	} else
2351 		buf = start;
2352 	a.start = &buf;
2353 	a.end = start + have;
2354 	/* start copying other objects */
2355 	if (compat) {
2356 		a.type = DN_COMPAT_PIPE;
2357 		dn_ht_scan(V_dn_cfg.schedhash, copy_data_helper_compat, &a);
2358 		a.type = DN_COMPAT_QUEUE;
2359 		dn_ht_scan(V_dn_cfg.fshash, copy_data_helper_compat, &a);
2360 	} else if (a.type == DN_FS) {
2361 		dn_ht_scan(V_dn_cfg.fshash, copy_data_helper, &a);
2362 	} else {
2363 		dn_ht_scan(V_dn_cfg.schedhash, copy_data_helper, &a);
2364 	}
2365 	DN_BH_WUNLOCK();
2366 
2367 	if (compat) {
2368 		*compat = start;
2369 		sopt->sopt_valsize = buf - start;
2370 		/* free() is done by ip_dummynet_compat() */
2371 		start = NULL; //XXX hack
2372 	} else {
2373 		error = sooptcopyout(sopt, start, buf - start);
2374 	}
2375 done:
2376 	if (cmd != &r.o)
2377 		free(cmd, M_DUMMYNET);
2378 	free(start, M_DUMMYNET);
2379 	return error;
2380 }
2381 
2382 /* Callback called on scheduler instance to delete it if idle */
2383 static int
2384 drain_scheduler_cb(void *_si, void *arg)
2385 {
2386 	struct dn_sch_inst *si = _si;
2387 
2388 	if ((si->kflags & DN_ACTIVE) || si->dline.mq.head != NULL)
2389 		return 0;
2390 
2391 	if (si->sched->fp->flags & DN_MULTIQUEUE) {
2392 		if (si->q_count == 0)
2393 			return si_destroy(si, NULL);
2394 		else
2395 			return 0;
2396 	} else { /* !DN_MULTIQUEUE */
2397 		if ((si+1)->ni.length == 0)
2398 			return si_destroy(si, NULL);
2399 		else
2400 			return 0;
2401 	}
2402 	return 0; /* unreachable */
2403 }
2404 
2405 /* Callback called on scheduler to check if it has instances */
2406 static int
2407 drain_scheduler_sch_cb(void *_s, void *arg)
2408 {
2409 	struct dn_schk *s = _s;
2410 
2411 	if (s->sch.flags & DN_HAVE_MASK) {
2412 		dn_ht_scan_bucket(s->siht, &s->drain_bucket,
2413 				drain_scheduler_cb, NULL);
2414 		s->drain_bucket++;
2415 	} else {
2416 		if (s->siht) {
2417 			if (drain_scheduler_cb(s->siht, NULL) == DNHT_SCAN_DEL)
2418 				s->siht = NULL;
2419 		}
2420 	}
2421 	return 0;
2422 }
2423 
2424 /* Called every tick, try to delete a 'bucket' of scheduler */
2425 void
2426 dn_drain_scheduler(void)
2427 {
2428 	dn_ht_scan_bucket(V_dn_cfg.schedhash, &V_dn_cfg.drain_sch,
2429 			   drain_scheduler_sch_cb, NULL);
2430 	V_dn_cfg.drain_sch++;
2431 }
2432 
2433 /* Callback called on queue to delete if it is idle */
2434 static int
2435 drain_queue_cb(void *_q, void *arg)
2436 {
2437 	struct dn_queue *q = _q;
2438 
2439 	if (q->ni.length == 0) {
2440 		dn_delete_queue(q, DN_DESTROY);
2441 		return DNHT_SCAN_DEL; /* queue is deleted */
2442 	}
2443 
2444 	return 0; /* queue isn't deleted */
2445 }
2446 
2447 /* Callback called on flowset used to check if it has queues */
2448 static int
2449 drain_queue_fs_cb(void *_fs, void *arg)
2450 {
2451 	struct dn_fsk *fs = _fs;
2452 
2453 	if (fs->fs.flags & DN_QHT_HASH) {
2454 		/* Flowset has a hash table for queues */
2455 		dn_ht_scan_bucket(fs->qht, &fs->drain_bucket,
2456 				drain_queue_cb, NULL);
2457 		fs->drain_bucket++;
2458 	} else {
2459 		/* No hash table for this flowset, null the pointer
2460 		 * if the queue is deleted
2461 		 */
2462 		if (fs->qht) {
2463 			if (drain_queue_cb(fs->qht, NULL) == DNHT_SCAN_DEL)
2464 				fs->qht = NULL;
2465 		}
2466 	}
2467 	return 0;
2468 }
2469 
2470 /* Called every tick, try to delete a 'bucket' of queue */
2471 void
2472 dn_drain_queue(void)
2473 {
2474 	/* scan a bucket of flowset */
2475 	dn_ht_scan_bucket(V_dn_cfg.fshash, &V_dn_cfg.drain_fs,
2476                                drain_queue_fs_cb, NULL);
2477 	V_dn_cfg.drain_fs++;
2478 }
2479 
2480 /*
2481  * Handler for the various dummynet socket options
2482  */
2483 static int
2484 ip_dn_ctl(struct sockopt *sopt)
2485 {
2486 	struct epoch_tracker et;
2487 	void *p = NULL;
2488 	int error, l;
2489 
2490 	error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2491 	if (error)
2492 		return (error);
2493 
2494 	/* Disallow sets in really-really secure mode. */
2495 	if (sopt->sopt_dir == SOPT_SET) {
2496 		error =  securelevel_ge(sopt->sopt_td->td_ucred, 3);
2497 		if (error)
2498 			return (error);
2499 	}
2500 
2501 	NET_EPOCH_ENTER(et);
2502 
2503 	switch (sopt->sopt_name) {
2504 	default :
2505 		D("dummynet: unknown option %d", sopt->sopt_name);
2506 		error = EINVAL;
2507 		break;
2508 
2509 	case IP_DUMMYNET_FLUSH:
2510 	case IP_DUMMYNET_CONFIGURE:
2511 	case IP_DUMMYNET_DEL:	/* remove a pipe or queue */
2512 	case IP_DUMMYNET_GET:
2513 		D("dummynet: compat option %d", sopt->sopt_name);
2514 		error = ip_dummynet_compat(sopt);
2515 		break;
2516 
2517 	case IP_DUMMYNET3 :
2518 		if (sopt->sopt_dir == SOPT_GET) {
2519 			error = dummynet_get(sopt, NULL);
2520 			break;
2521 		}
2522 		l = sopt->sopt_valsize;
2523 		if (l < sizeof(struct dn_id) || l > 12000) {
2524 			D("argument len %d invalid", l);
2525 			break;
2526 		}
2527 		p = malloc(l, M_TEMP, M_NOWAIT);
2528 		if (p == NULL) {
2529 			error = ENOMEM;
2530 			break;
2531 		}
2532 		error = sooptcopyin(sopt, p, l, l);
2533 		if (error)
2534 			break ;
2535 		error = do_config(p, l);
2536 		break;
2537 	}
2538 
2539 	free(p, M_TEMP);
2540 
2541 	NET_EPOCH_EXIT(et);
2542 
2543 	return error ;
2544 }
2545 
2546 static void
2547 ip_dn_vnet_init(void)
2548 {
2549 	if (V_dn_cfg.init_done)
2550 		return;
2551 	V_dn_cfg.init_done = 1;
2552 	/* Set defaults here. MSVC does not accept initializers,
2553 	 * and this is also useful for vimages
2554 	 */
2555 	/* queue limits */
2556 	V_dn_cfg.slot_limit = 100; /* Foot shooting limit for queues. */
2557 	V_dn_cfg.byte_limit = 1024 * 1024;
2558 	V_dn_cfg.expire = 1;
2559 
2560 	/* RED parameters */
2561 	V_dn_cfg.red_lookup_depth = 256;	/* default lookup table depth */
2562 	V_dn_cfg.red_avg_pkt_size = 512;	/* default medium packet size */
2563 	V_dn_cfg.red_max_pkt_size = 1500;	/* default max packet size */
2564 
2565 	/* hash tables */
2566 	V_dn_cfg.max_hash_size = 65536;	/* max in the hash tables */
2567 	V_dn_cfg.hash_size = 64;		/* default hash size */
2568 
2569 	/* create hash tables for schedulers and flowsets.
2570 	 * In both we search by key and by pointer.
2571 	 */
2572 	V_dn_cfg.schedhash = dn_ht_init(NULL, V_dn_cfg.hash_size,
2573 		offsetof(struct dn_schk, schk_next),
2574 		schk_hash, schk_match, schk_new);
2575 	V_dn_cfg.fshash = dn_ht_init(NULL, V_dn_cfg.hash_size,
2576 		offsetof(struct dn_fsk, fsk_next),
2577 		fsk_hash, fsk_match, fsk_new);
2578 
2579 	/* bucket index to drain object */
2580 	V_dn_cfg.drain_fs = 0;
2581 	V_dn_cfg.drain_sch = 0;
2582 
2583 	heap_init(&V_dn_cfg.evheap, 16, offsetof(struct dn_id, id));
2584 	SLIST_INIT(&V_dn_cfg.fsu);
2585 
2586 	DN_LOCK_INIT();
2587 
2588 	/* Initialize curr_time adjustment mechanics. */
2589 	getmicrouptime(&V_dn_cfg.prev_t);
2590 }
2591 
2592 static void
2593 ip_dn_vnet_destroy(void)
2594 {
2595 	DN_BH_WLOCK();
2596 	dummynet_flush();
2597 	DN_BH_WUNLOCK();
2598 
2599 	dn_ht_free(V_dn_cfg.schedhash, 0);
2600 	dn_ht_free(V_dn_cfg.fshash, 0);
2601 	heap_free(&V_dn_cfg.evheap);
2602 
2603 	DN_LOCK_DESTROY();
2604 }
2605 
2606 static void
2607 ip_dn_init(void)
2608 {
2609 	if (dn_tasks_started)
2610 		return;
2611 
2612 	mtx_init(&sched_mtx, "dn_sched", NULL, MTX_DEF);
2613 
2614 	dn_tasks_started = 1;
2615 	TASK_INIT(&dn_task, 0, dummynet_task, NULL);
2616 	dn_tq = taskqueue_create_fast("dummynet", M_WAITOK,
2617 	    taskqueue_thread_enqueue, &dn_tq);
2618 	taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2619 
2620 	CK_LIST_INIT(&schedlist);
2621 	callout_init(&dn_timeout, 1);
2622 	dn_reschedule();
2623 }
2624 
2625 static void
2626 ip_dn_destroy(int last)
2627 {
2628 	/* ensure no more callouts are started */
2629 	dn_gone = 1;
2630 
2631 	/* check for last */
2632 	if (last) {
2633 		ND("removing last instance\n");
2634 		ip_dn_ctl_ptr = NULL;
2635 		ip_dn_io_ptr = NULL;
2636 	}
2637 
2638 	callout_drain(&dn_timeout);
2639 	taskqueue_drain(dn_tq, &dn_task);
2640 	taskqueue_free(dn_tq);
2641 }
2642 
2643 static int
2644 dummynet_modevent(module_t mod, int type, void *data)
2645 {
2646 
2647 	if (type == MOD_LOAD) {
2648 		if (ip_dn_io_ptr) {
2649 			printf("DUMMYNET already loaded\n");
2650 			return EEXIST ;
2651 		}
2652 		ip_dn_init();
2653 		ip_dn_ctl_ptr = ip_dn_ctl;
2654 		ip_dn_io_ptr = dummynet_io;
2655 		return 0;
2656 	} else if (type == MOD_UNLOAD) {
2657 		ip_dn_destroy(1 /* last */);
2658 		return 0;
2659 	} else
2660 		return EOPNOTSUPP;
2661 }
2662 
2663 /* modevent helpers for the modules */
2664 static int
2665 load_dn_sched(struct dn_alg *d)
2666 {
2667 	struct dn_alg *s;
2668 
2669 	if (d == NULL)
2670 		return 1; /* error */
2671 	ip_dn_init();	/* just in case, we need the lock */
2672 
2673 	/* Check that mandatory funcs exists */
2674 	if (d->enqueue == NULL || d->dequeue == NULL) {
2675 		D("missing enqueue or dequeue for %s", d->name);
2676 		return 1;
2677 	}
2678 
2679 	/* Search if scheduler already exists */
2680 	mtx_lock(&sched_mtx);
2681 	CK_LIST_FOREACH(s, &schedlist, next) {
2682 		if (strcmp(s->name, d->name) == 0) {
2683 			D("%s already loaded", d->name);
2684 			break; /* scheduler already exists */
2685 		}
2686 	}
2687 	if (s == NULL)
2688 		CK_LIST_INSERT_HEAD(&schedlist, d, next);
2689 	mtx_unlock(&sched_mtx);
2690 	D("dn_sched %s %sloaded", d->name, s ? "not ":"");
2691 	return s ? 1 : 0;
2692 }
2693 
2694 static int
2695 unload_dn_sched(struct dn_alg *s)
2696 {
2697 	struct dn_alg *tmp, *r;
2698 	int err = EINVAL;
2699 
2700 	ND("called for %s", s->name);
2701 
2702 	mtx_lock(&sched_mtx);
2703 	CK_LIST_FOREACH_SAFE(r, &schedlist, next, tmp) {
2704 		if (strcmp(s->name, r->name) != 0)
2705 			continue;
2706 		ND("ref_count = %d", r->ref_count);
2707 		err = (r->ref_count != 0) ? EBUSY : 0;
2708 		if (err == 0)
2709 			CK_LIST_REMOVE(r, next);
2710 		break;
2711 	}
2712 	mtx_unlock(&sched_mtx);
2713 	NET_EPOCH_WAIT();
2714 	D("dn_sched %s %sunloaded", s->name, err ? "not ":"");
2715 	return err;
2716 }
2717 
2718 int
2719 dn_sched_modevent(module_t mod, int cmd, void *arg)
2720 {
2721 	struct dn_alg *sch = arg;
2722 
2723 	if (cmd == MOD_LOAD)
2724 		return load_dn_sched(sch);
2725 	else if (cmd == MOD_UNLOAD)
2726 		return unload_dn_sched(sch);
2727 	else
2728 		return EINVAL;
2729 }
2730 
2731 static moduledata_t dummynet_mod = {
2732 	"dummynet", dummynet_modevent, NULL
2733 };
2734 
2735 #define	DN_SI_SUB	SI_SUB_PROTO_FIREWALL
2736 #define	DN_MODEV_ORD	(SI_ORDER_ANY - 128) /* after ipfw */
2737 DECLARE_MODULE(dummynet, dummynet_mod, DN_SI_SUB, DN_MODEV_ORD);
2738 MODULE_VERSION(dummynet, 3);
2739 
2740 /*
2741  * Starting up. Done in order after dummynet_modevent() has been called.
2742  * VNET_SYSINIT is also called for each existing vnet and each new vnet.
2743  */
2744 VNET_SYSINIT(vnet_dn_init, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_vnet_init, NULL);
2745 
2746 /*
2747  * Shutdown handlers up shop. These are done in REVERSE ORDER, but still
2748  * after dummynet_modevent() has been called. Not called on reboot.
2749  * VNET_SYSUNINIT is also called for each exiting vnet as it exits.
2750  * or when the module is unloaded.
2751  */
2752 VNET_SYSUNINIT(vnet_dn_uninit, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_vnet_destroy, NULL);
2753 
2754 #ifdef NEW_AQM
2755 
2756 /* modevent helpers for the AQM modules */
2757 static int
2758 load_dn_aqm(struct dn_aqm *d)
2759 {
2760 	struct dn_aqm *aqm=NULL;
2761 
2762 	if (d == NULL)
2763 		return 1; /* error */
2764 	ip_dn_init();	/* just in case, we need the lock */
2765 
2766 	/* Check that mandatory funcs exists */
2767 	if (d->enqueue == NULL || d->dequeue == NULL) {
2768 		D("missing enqueue or dequeue for %s", d->name);
2769 		return 1;
2770 	}
2771 
2772 	mtx_lock(&sched_mtx);
2773 
2774 	/* Search if AQM already exists */
2775 	CK_LIST_FOREACH(aqm, &aqmlist, next) {
2776 		if (strcmp(aqm->name, d->name) == 0) {
2777 			D("%s already loaded", d->name);
2778 			break; /* AQM already exists */
2779 		}
2780 	}
2781 	if (aqm == NULL)
2782 		CK_LIST_INSERT_HEAD(&aqmlist, d, next);
2783 
2784 	mtx_unlock(&sched_mtx);
2785 
2786 	D("dn_aqm %s %sloaded", d->name, aqm ? "not ":"");
2787 	return aqm ? 1 : 0;
2788 }
2789 
2790 /* Callback to clean up AQM status for queues connected to a flowset
2791  * and then deconfigure the flowset.
2792  * This function is called before an AQM module is unloaded
2793  */
2794 static int
2795 fs_cleanup(void *_fs, void *arg)
2796 {
2797 	struct dn_fsk *fs = _fs;
2798 	uint32_t type = *(uint32_t *)arg;
2799 
2800 	if (fs->aqmfp && fs->aqmfp->type == type)
2801 		aqm_cleanup_deconfig_fs(fs);
2802 
2803 	return 0;
2804 }
2805 
2806 static int
2807 unload_dn_aqm(struct dn_aqm *aqm)
2808 {
2809 	struct dn_aqm *tmp, *r;
2810 	int err = EINVAL;
2811 	err = 0;
2812 	ND("called for %s", aqm->name);
2813 
2814 	/* clean up AQM status and deconfig flowset */
2815 	dn_ht_scan(V_dn_cfg.fshash, fs_cleanup, &aqm->type);
2816 
2817 	mtx_lock(&sched_mtx);
2818 
2819 	CK_LIST_FOREACH_SAFE(r, &aqmlist, next, tmp) {
2820 		if (strcmp(aqm->name, r->name) != 0)
2821 			continue;
2822 		ND("ref_count = %d", r->ref_count);
2823 		err = (r->ref_count != 0 || r->cfg_ref_count != 0) ? EBUSY : 0;
2824 		if (err == 0)
2825 			CK_LIST_REMOVE(r, next);
2826 		break;
2827 	}
2828 
2829 	mtx_unlock(&sched_mtx);
2830 	NET_EPOCH_WAIT();
2831 
2832 	D("%s %sunloaded", aqm->name, err ? "not ":"");
2833 	if (err)
2834 		D("ref_count=%d, cfg_ref_count=%d", r->ref_count, r->cfg_ref_count);
2835 	return err;
2836 }
2837 
2838 int
2839 dn_aqm_modevent(module_t mod, int cmd, void *arg)
2840 {
2841 	struct dn_aqm *aqm = arg;
2842 
2843 	if (cmd == MOD_LOAD)
2844 		return load_dn_aqm(aqm);
2845 	else if (cmd == MOD_UNLOAD)
2846 		return unload_dn_aqm(aqm);
2847 	else
2848 		return EINVAL;
2849 }
2850 #endif
2851 
2852 /* end of file */
2853