2 * Copyright (c) 1998-2002,2010 Luigi Rizzo, Universita` di Pisa
3 * Portions Copyright (c) 2000 Akamba Corp.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
32 * Configuration and internal object management for dummynet.
35 #include "opt_inet6.h"
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/malloc.h>
41 #include <sys/kernel.h>
43 #include <sys/module.h>
46 #include <sys/rwlock.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
50 #include <sys/taskqueue.h>
51 #include <net/if.h> /* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */
52 #include <netinet/in.h>
53 #include <netinet/ip_var.h> /* ip_output(), IP_FORWARDING */
54 #include <netinet/ip_fw.h>
55 #include <netinet/ipfw/ip_fw_private.h>
56 #include <netinet/ipfw/dn_heap.h>
57 #include <netinet/ip_dummynet.h>
58 #include <netinet/ipfw/ip_dn_private.h>
59 #include <netinet/ipfw/dn_sched.h>
61 /* which objects to copy */
62 #define DN_C_LINK 0x01
64 #define DN_C_FLOW 0x04
66 #define DN_C_QUEUE 0x10
68 /* we use this argument in case of a schk_new */
74 /*---- callout hooks. ----*/
75 static struct callout dn_timeout;
76 static struct task dn_task;
77 static struct taskqueue *dn_tq = NULL;
80 dummynet(void * __unused unused)
83 taskqueue_enqueue(dn_tq, &dn_task);
89 callout_reset(&dn_timeout, 1, dummynet, NULL);
91 /*----- end of callout hooks -----*/
93 /* Return a scheduler descriptor given the type or name. */
94 static struct dn_alg *
95 find_sched_type(int type, char *name)
99 SLIST_FOREACH(d, &dn_cfg.schedlist, next) {
100 if (d->type == type || (name && !strcmp(d->name, name)))
103 return NULL; /* not found */
107 ipdn_bound_var(int *v, int dflt, int lo, int hi, const char *msg)
110 const char *op = NULL;
114 } else if (oldv > hi) {
120 printf("%s %s to %d (was %d)\n", op, msg, *v, oldv);
124 /*---- flow_id mask, hash and compare functions ---*/
126 * The flow_id includes the 5-tuple, the queue/pipe number
127 * which we store in the extra area in host order,
128 * and for ipv6 also the flow_id6.
129 * XXX see if we want the tos byte (can store in 'flags')
131 static struct ipfw_flow_id *
132 flow_id_mask(struct ipfw_flow_id *mask, struct ipfw_flow_id *id)
134 int is_v6 = IS_IP6_FLOW_ID(id);
136 id->dst_port &= mask->dst_port;
137 id->src_port &= mask->src_port;
138 id->proto &= mask->proto;
139 id->extra &= mask->extra;
141 APPLY_MASK(&id->dst_ip6, &mask->dst_ip6);
142 APPLY_MASK(&id->src_ip6, &mask->src_ip6);
143 id->flow_id6 &= mask->flow_id6;
145 id->dst_ip &= mask->dst_ip;
146 id->src_ip &= mask->src_ip;
151 /* computes an OR of two masks, result in dst and also returned */
152 static struct ipfw_flow_id *
153 flow_id_or(struct ipfw_flow_id *src, struct ipfw_flow_id *dst)
155 int is_v6 = IS_IP6_FLOW_ID(dst);
157 dst->dst_port |= src->dst_port;
158 dst->src_port |= src->src_port;
159 dst->proto |= src->proto;
160 dst->extra |= src->extra;
162 #define OR_MASK(_d, _s) \
163 (_d)->__u6_addr.__u6_addr32[0] |= (_s)->__u6_addr.__u6_addr32[0]; \
164 (_d)->__u6_addr.__u6_addr32[1] |= (_s)->__u6_addr.__u6_addr32[1]; \
165 (_d)->__u6_addr.__u6_addr32[2] |= (_s)->__u6_addr.__u6_addr32[2]; \
166 (_d)->__u6_addr.__u6_addr32[3] |= (_s)->__u6_addr.__u6_addr32[3];
167 OR_MASK(&dst->dst_ip6, &src->dst_ip6);
168 OR_MASK(&dst->src_ip6, &src->src_ip6);
170 dst->flow_id6 |= src->flow_id6;
172 dst->dst_ip |= src->dst_ip;
173 dst->src_ip |= src->src_ip;
179 nonzero_mask(struct ipfw_flow_id *m)
181 if (m->dst_port || m->src_port || m->proto || m->extra)
183 if (IS_IP6_FLOW_ID(m)) {
185 m->dst_ip6.__u6_addr.__u6_addr32[0] ||
186 m->dst_ip6.__u6_addr.__u6_addr32[1] ||
187 m->dst_ip6.__u6_addr.__u6_addr32[2] ||
188 m->dst_ip6.__u6_addr.__u6_addr32[3] ||
189 m->src_ip6.__u6_addr.__u6_addr32[0] ||
190 m->src_ip6.__u6_addr.__u6_addr32[1] ||
191 m->src_ip6.__u6_addr.__u6_addr32[2] ||
192 m->src_ip6.__u6_addr.__u6_addr32[3] ||
195 return m->dst_ip || m->src_ip;
199 /* XXX we may want a better hash function */
201 flow_id_hash(struct ipfw_flow_id *id)
205 if (IS_IP6_FLOW_ID(id)) {
206 uint32_t *d = (uint32_t *)&id->dst_ip6;
207 uint32_t *s = (uint32_t *)&id->src_ip6;
208 i = (d[0] ) ^ (d[1]) ^
210 (d[0] >> 15) ^ (d[1] >> 15) ^
211 (d[2] >> 15) ^ (d[3] >> 15) ^
212 (s[0] << 1) ^ (s[1] << 1) ^
213 (s[2] << 1) ^ (s[3] << 1) ^
214 (s[0] << 16) ^ (s[1] << 16) ^
215 (s[2] << 16) ^ (s[3] << 16) ^
216 (id->dst_port << 1) ^ (id->src_port) ^
218 (id->proto ) ^ (id->flow_id6);
220 i = (id->dst_ip) ^ (id->dst_ip >> 15) ^
221 (id->src_ip << 1) ^ (id->src_ip >> 16) ^
223 (id->dst_port << 1) ^ (id->src_port) ^ (id->proto);
228 /* Like bcmp, returns 0 if ids match, 1 otherwise. */
230 flow_id_cmp(struct ipfw_flow_id *id1, struct ipfw_flow_id *id2)
232 int is_v6 = IS_IP6_FLOW_ID(id1);
235 if (IS_IP6_FLOW_ID(id2))
236 return 1; /* different address families */
238 return (id1->dst_ip == id2->dst_ip &&
239 id1->src_ip == id2->src_ip &&
240 id1->dst_port == id2->dst_port &&
241 id1->src_port == id2->src_port &&
242 id1->proto == id2->proto &&
243 id1->extra == id2->extra) ? 0 : 1;
247 !bcmp(&id1->dst_ip6,&id2->dst_ip6, sizeof(id1->dst_ip6)) &&
248 !bcmp(&id1->src_ip6,&id2->src_ip6, sizeof(id1->src_ip6)) &&
249 id1->dst_port == id2->dst_port &&
250 id1->src_port == id2->src_port &&
251 id1->proto == id2->proto &&
252 id1->extra == id2->extra &&
253 id1->flow_id6 == id2->flow_id6) ? 0 : 1;
255 /*--------- end of flow-id mask, hash and compare ---------*/
257 /*--- support functions for the qht hashtable ----
258 * Entries are hashed by flow-id
261 q_hash(uintptr_t key, int flags, void *arg)
263 /* compute the hash slot from the flow id */
264 struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
265 &((struct dn_queue *)key)->ni.fid :
266 (struct ipfw_flow_id *)key;
268 return flow_id_hash(id);
272 q_match(void *obj, uintptr_t key, int flags, void *arg)
274 struct dn_queue *o = (struct dn_queue *)obj;
275 struct ipfw_flow_id *id2;
277 if (flags & DNHT_KEY_IS_OBJ) {
278 /* compare pointers */
279 id2 = &((struct dn_queue *)key)->ni.fid;
281 id2 = (struct ipfw_flow_id *)key;
283 return (0 == flow_id_cmp(&o->ni.fid, id2));
287 * create a new queue instance for the given 'key'.
290 q_new(uintptr_t key, int flags, void *arg)
292 struct dn_queue *q, *template = arg;
293 struct dn_fsk *fs = template->fs;
294 int size = sizeof(*q) + fs->sched->fp->q_datalen;
296 q = malloc(size, M_DUMMYNET, M_NOWAIT | M_ZERO);
298 D("no memory for new queue");
302 set_oid(&q->ni.oid, DN_QUEUE, size);
303 if (fs->fs.flags & DN_QHT_HASH)
304 q->ni.fid = *(struct ipfw_flow_id *)key;
306 q->_si = template->_si;
309 if (fs->sched->fp->new_queue)
310 fs->sched->fp->new_queue(q);
311 dn_cfg.queue_count++;
316 * Notify schedulers that a queue is going away.
317 * If (flags & DN_DESTROY), also free the packets.
318 * The version for callbacks is called q_delete_cb().
321 dn_delete_queue(struct dn_queue *q, int flags)
323 struct dn_fsk *fs = q->fs;
325 // D("fs %p si %p\n", fs, q->_si);
326 /* notify the parent scheduler that the queue is going away */
327 if (fs && fs->sched->fp->free_queue)
328 fs->sched->fp->free_queue(q);
331 if (flags & DN_DESTROY) {
333 dn_free_pkts(q->mq.head);
334 bzero(q, sizeof(*q)); // safety
336 dn_cfg.queue_count--;
341 q_delete_cb(void *q, void *arg)
343 int flags = (int)(uintptr_t)arg;
344 dn_delete_queue(q, flags);
345 return (flags & DN_DESTROY) ? DNHT_SCAN_DEL : 0;
349 * calls dn_delete_queue/q_delete_cb on all queues,
350 * which notifies the parent scheduler and possibly drains packets.
351 * flags & DN_DESTROY: drains queues and destroy qht;
354 qht_delete(struct dn_fsk *fs, int flags)
356 ND("fs %d start flags %d qht %p",
357 fs->fs.fs_nr, flags, fs->qht);
360 if (fs->fs.flags & DN_QHT_HASH) {
361 dn_ht_scan(fs->qht, q_delete_cb, (void *)(uintptr_t)flags);
362 if (flags & DN_DESTROY) {
363 dn_ht_free(fs->qht, 0);
367 dn_delete_queue((struct dn_queue *)(fs->qht), flags);
368 if (flags & DN_DESTROY)
374 * Find and possibly create the queue for a MULTIQUEUE scheduler.
375 * We never call it for !MULTIQUEUE (the queue is in the sch_inst).
378 ipdn_q_find(struct dn_fsk *fs, struct dn_sch_inst *si,
379 struct ipfw_flow_id *id)
381 struct dn_queue template;
386 if (fs->fs.flags & DN_QHT_HASH) {
387 struct ipfw_flow_id masked_id;
388 if (fs->qht == NULL) {
389 fs->qht = dn_ht_init(NULL, fs->fs.buckets,
390 offsetof(struct dn_queue, q_next),
391 q_hash, q_match, q_new);
396 flow_id_mask(&fs->fsk_mask, &masked_id);
397 return dn_ht_find(fs->qht, (uintptr_t)&masked_id,
398 DNHT_INSERT, &template);
401 fs->qht = q_new(0, 0, &template);
402 return (struct dn_queue *)fs->qht;
405 /*--- end of queue hash table ---*/
407 /*--- support functions for the sch_inst hashtable ----
409 * These are hashed by flow-id
412 si_hash(uintptr_t key, int flags, void *arg)
414 /* compute the hash slot from the flow id */
415 struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
416 &((struct dn_sch_inst *)key)->ni.fid :
417 (struct ipfw_flow_id *)key;
419 return flow_id_hash(id);
423 si_match(void *obj, uintptr_t key, int flags, void *arg)
425 struct dn_sch_inst *o = obj;
426 struct ipfw_flow_id *id2;
428 id2 = (flags & DNHT_KEY_IS_OBJ) ?
429 &((struct dn_sch_inst *)key)->ni.fid :
430 (struct ipfw_flow_id *)key;
431 return flow_id_cmp(&o->ni.fid, id2) == 0;
435 * create a new instance for the given 'key'
436 * Allocate memory for instance, delay line and scheduler private data.
439 si_new(uintptr_t key, int flags, void *arg)
441 struct dn_schk *s = arg;
442 struct dn_sch_inst *si;
443 int l = sizeof(*si) + s->fp->si_datalen;
445 si = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
448 /* Set length only for the part passed up to userland. */
449 set_oid(&si->ni.oid, DN_SCH_I, sizeof(struct dn_flow));
450 set_oid(&(si->dline.oid), DN_DELAY_LINE,
451 sizeof(struct delay_line));
452 /* mark si and dline as outside the event queue */
453 si->ni.oid.id = si->dline.oid.id = -1;
458 if (s->fp->new_sched && s->fp->new_sched(si)) {
459 D("new_sched error");
462 if (s->sch.flags & DN_HAVE_MASK)
463 si->ni.fid = *(struct ipfw_flow_id *)key;
470 bzero(si, sizeof(*si)); // safety
471 free(si, M_DUMMYNET);
477 * Callback from siht to delete all scheduler instances. Remove
478 * si and delay line from the system heap, destroy all queues.
479 * We assume that all flowset have been notified and do not
480 * point to us anymore.
483 si_destroy(void *_si, void *arg)
485 struct dn_sch_inst *si = _si;
486 struct dn_schk *s = si->sched;
487 struct delay_line *dl = &si->dline;
489 if (dl->oid.subtype) /* remove delay line from event heap */
490 heap_extract(&dn_cfg.evheap, dl);
491 dn_free_pkts(dl->mq.head); /* drain delay line */
492 if (si->kflags & DN_ACTIVE) /* remove si from event heap */
493 heap_extract(&dn_cfg.evheap, si);
494 if (s->fp->free_sched)
495 s->fp->free_sched(si);
496 bzero(si, sizeof(*si)); /* safety */
497 free(si, M_DUMMYNET);
499 return DNHT_SCAN_DEL;
503 * Find the scheduler instance for this packet. If we need to apply
504 * a mask, do on a local copy of the flow_id to preserve the original.
505 * Assume siht is always initialized if we have a mask.
508 ipdn_si_find(struct dn_schk *s, struct ipfw_flow_id *id)
511 if (s->sch.flags & DN_HAVE_MASK) {
512 struct ipfw_flow_id id_t = *id;
513 flow_id_mask(&s->sch.sched_mask, &id_t);
514 return dn_ht_find(s->siht, (uintptr_t)&id_t,
518 s->siht = si_new(0, 0, s);
519 return (struct dn_sch_inst *)s->siht;
522 /* callback to flush credit for the scheduler instance */
524 si_reset_credit(void *_si, void *arg)
526 struct dn_sch_inst *si = _si;
527 struct dn_link *p = &si->sched->link;
529 si->credit = p->burst + (dn_cfg.io_fast ? p->bandwidth : 0);
534 schk_reset_credit(struct dn_schk *s)
536 if (s->sch.flags & DN_HAVE_MASK)
537 dn_ht_scan(s->siht, si_reset_credit, NULL);
539 si_reset_credit(s->siht, NULL);
541 /*---- end of sch_inst hashtable ---------------------*/
543 /*-------------------------------------------------------
544 * flowset hash (fshash) support. Entries are hashed by fs_nr.
545 * New allocations are put in the fsunlinked list, from which
546 * they are removed when they point to a specific scheduler.
549 fsk_hash(uintptr_t key, int flags, void *arg)
551 uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
552 ((struct dn_fsk *)key)->fs.fs_nr;
554 return ( (i>>8)^(i>>4)^i );
558 fsk_match(void *obj, uintptr_t key, int flags, void *arg)
560 struct dn_fsk *fs = obj;
561 int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
562 ((struct dn_fsk *)key)->fs.fs_nr;
564 return (fs->fs.fs_nr == i);
568 fsk_new(uintptr_t key, int flags, void *arg)
572 fs = malloc(sizeof(*fs), M_DUMMYNET, M_NOWAIT | M_ZERO);
574 set_oid(&fs->fs.oid, DN_FS, sizeof(fs->fs));
576 fs->drain_bucket = 0;
577 SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
583 * detach flowset from its current scheduler. Flags as follows:
584 * DN_DETACH removes from the fsk_list
585 * DN_DESTROY deletes individual queues
586 * DN_DELETE_FS destroys the flowset (otherwise goes in unlinked).
589 fsk_detach(struct dn_fsk *fs, int flags)
591 if (flags & DN_DELETE_FS)
593 ND("fs %d from sched %d flags %s %s %s",
594 fs->fs.fs_nr, fs->fs.sched_nr,
595 (flags & DN_DELETE_FS) ? "DEL_FS":"",
596 (flags & DN_DESTROY) ? "DEL":"",
597 (flags & DN_DETACH) ? "DET":"");
598 if (flags & DN_DETACH) { /* detach from the list */
599 struct dn_fsk_head *h;
600 h = fs->sched ? &fs->sched->fsk_list : &dn_cfg.fsu;
601 SLIST_REMOVE(h, fs, dn_fsk, sch_chain);
603 /* Free the RED parameters, they will be recomputed on
604 * subsequent attach if needed.
607 free(fs->w_q_lookup, M_DUMMYNET);
608 fs->w_q_lookup = NULL;
609 qht_delete(fs, flags);
610 if (fs->sched && fs->sched->fp->free_fsk)
611 fs->sched->fp->free_fsk(fs);
613 if (flags & DN_DELETE_FS) {
614 bzero(fs, sizeof(fs)); /* safety */
615 free(fs, M_DUMMYNET);
618 SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
623 * Detach or destroy all flowsets in a list.
624 * flags specifies what to do:
625 * DN_DESTROY: flush all queues
626 * DN_DELETE_FS: DN_DESTROY + destroy flowset
627 * DN_DELETE_FS implies DN_DESTROY
630 fsk_detach_list(struct dn_fsk_head *h, int flags)
633 int n = 0; /* only for stats */
635 ND("head %p flags %x", h, flags);
636 while ((fs = SLIST_FIRST(h))) {
637 SLIST_REMOVE_HEAD(h, sch_chain);
639 fsk_detach(fs, flags);
641 ND("done %d flowsets", n);
645 * called on 'queue X delete' -- removes the flowset from fshash,
646 * deletes all queues for the flowset, and removes the flowset.
649 delete_fs(int i, int locked)
656 fs = dn_ht_find(dn_cfg.fshash, i, DNHT_REMOVE, NULL);
657 ND("fs %d found %p", i, fs);
659 fsk_detach(fs, DN_DETACH | DN_DELETE_FS);
668 /*----- end of flowset hashtable support -------------*/
670 /*------------------------------------------------------------
671 * Scheduler hash. When searching by index we pass sched_nr,
672 * otherwise we pass struct dn_sch * which is the first field in
673 * struct dn_schk so we can cast between the two. We use this trick
674 * because in the create phase (but it should be fixed).
677 schk_hash(uintptr_t key, int flags, void *_arg)
679 uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
680 ((struct dn_schk *)key)->sch.sched_nr;
681 return ( (i>>8)^(i>>4)^i );
685 schk_match(void *obj, uintptr_t key, int flags, void *_arg)
687 struct dn_schk *s = (struct dn_schk *)obj;
688 int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
689 ((struct dn_schk *)key)->sch.sched_nr;
690 return (s->sch.sched_nr == i);
694 * Create the entry and intialize with the sched hash if needed.
695 * Leave s->fp unset so we can tell whether a dn_ht_find() returns
696 * a new object or a previously existing one.
699 schk_new(uintptr_t key, int flags, void *arg)
701 struct schk_new_arg *a = arg;
703 int l = sizeof(*s) +a->fp->schk_datalen;
705 s = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
708 set_oid(&s->link.oid, DN_LINK, sizeof(s->link));
709 s->sch = *a->sch; // copy initial values
710 s->link.link_nr = s->sch.sched_nr;
711 SLIST_INIT(&s->fsk_list);
712 /* initialize the hash table or create the single instance */
713 s->fp = a->fp; /* si_new needs this */
715 if (s->sch.flags & DN_HAVE_MASK) {
716 s->siht = dn_ht_init(NULL, s->sch.buckets,
717 offsetof(struct dn_sch_inst, si_next),
718 si_hash, si_match, si_new);
719 if (s->siht == NULL) {
724 s->fp = NULL; /* mark as a new scheduler */
730 * Callback for sched delete. Notify all attached flowsets to
731 * detach from the scheduler, destroy the internal flowset, and
732 * all instances. The scheduler goes away too.
733 * arg is 0 (only detach flowsets and destroy instances)
734 * DN_DESTROY (detach & delete queues, delete schk)
735 * or DN_DELETE_FS (delete queues and flowsets, delete schk)
738 schk_delete_cb(void *obj, void *arg)
740 struct dn_schk *s = obj;
743 ND("sched %d arg %s%s",
745 a&DN_DESTROY ? "DEL ":"",
746 a&DN_DELETE_FS ? "DEL_FS":"");
748 fsk_detach_list(&s->fsk_list, arg ? DN_DESTROY : 0);
749 /* no more flowset pointing to us now */
750 if (s->sch.flags & DN_HAVE_MASK)
751 dn_ht_scan(s->siht, si_destroy, NULL);
753 si_destroy(s->siht, NULL);
755 free(s->profile, M_DUMMYNET);
761 bzero(s, sizeof(*s)); // safety
762 free(obj, M_DUMMYNET);
764 return DNHT_SCAN_DEL;
768 * called on a 'sched X delete' command. Deletes a single scheduler.
769 * This is done by removing from the schedhash, unlinking all
770 * flowsets and deleting their traffic.
777 s = dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
781 delete_fs(i + DN_MAX_ID, 1); /* first delete internal fs */
782 /* then detach flowsets, delete traffic */
783 schk_delete_cb(s, (void*)(uintptr_t)DN_DESTROY);
786 /*--- end of schk hashtable support ---*/
789 copy_obj(char **start, char *end, void *_o, const char *msg, int i)
791 struct dn_id *o = _o;
792 int have = end - *start;
794 if (have < o->len || o->len == 0 || o->type == 0) {
795 D("(WARN) type %d %s %d have %d need %d",
796 o->type, msg, i, have, o->len);
799 ND("type %d %s %d len %d", o->type, msg, i, o->len);
800 bcopy(_o, *start, o->len);
801 if (o->type == DN_LINK) {
802 /* Adjust burst parameter for link */
803 struct dn_link *l = (struct dn_link *)*start;
804 l->burst = div64(l->burst, 8 * hz);
805 } else if (o->type == DN_SCH) {
806 /* Set id->id to the number of instances */
807 struct dn_schk *s = _o;
808 struct dn_id *id = (struct dn_id *)(*start);
809 id->id = (s->sch.flags & DN_HAVE_MASK) ?
810 dn_ht_entries(s->siht) : (s->siht ? 1 : 0);
816 /* Specific function to copy a queue.
817 * Copies only the user-visible part of a queue (which is in
818 * a struct dn_flow), and sets len accordingly.
821 copy_obj_q(char **start, char *end, void *_o, const char *msg, int i)
823 struct dn_id *o = _o;
824 int have = end - *start;
825 int len = sizeof(struct dn_flow); /* see above comment */
827 if (have < len || o->len == 0 || o->type != DN_QUEUE) {
828 D("ERROR type %d %s %d have %d need %d",
829 o->type, msg, i, have, len);
832 ND("type %d %s %d len %d", o->type, msg, i, len);
833 bcopy(_o, *start, len);
834 ((struct dn_id*)(*start))->len = len;
840 copy_q_cb(void *obj, void *arg)
842 struct dn_queue *q = obj;
843 struct copy_args *a = arg;
844 struct dn_flow *ni = (struct dn_flow *)(*a->start);
845 if (copy_obj_q(a->start, a->end, &q->ni, "queue", -1))
846 return DNHT_SCAN_END;
847 ni->oid.type = DN_FLOW; /* override the DN_QUEUE */
848 ni->oid.id = si_hash((uintptr_t)&ni->fid, 0, NULL);
853 copy_q(struct copy_args *a, struct dn_fsk *fs, int flags)
857 if (fs->fs.flags & DN_QHT_HASH)
858 dn_ht_scan(fs->qht, copy_q_cb, a);
860 copy_q_cb(fs->qht, a);
865 * This routine only copies the initial part of a profile ? XXX
868 copy_profile(struct copy_args *a, struct dn_profile *p)
870 int have = a->end - *a->start;
871 /* XXX here we check for max length */
872 int profile_len = sizeof(struct dn_profile) -
873 ED_MAX_SAMPLES_NO*sizeof(int);
877 if (have < profile_len) {
878 D("error have %d need %d", have, profile_len);
881 bcopy(p, *a->start, profile_len);
882 ((struct dn_id *)(*a->start))->len = profile_len;
883 *a->start += profile_len;
888 copy_flowset(struct copy_args *a, struct dn_fsk *fs, int flags)
890 struct dn_fs *ufs = (struct dn_fs *)(*a->start);
893 ND("flowset %d", fs->fs.fs_nr);
894 if (copy_obj(a->start, a->end, &fs->fs, "flowset", fs->fs.fs_nr))
895 return DNHT_SCAN_END;
896 ufs->oid.id = (fs->fs.flags & DN_QHT_HASH) ?
897 dn_ht_entries(fs->qht) : (fs->qht ? 1 : 0);
898 if (flags) { /* copy queues */
905 copy_si_cb(void *obj, void *arg)
907 struct dn_sch_inst *si = obj;
908 struct copy_args *a = arg;
909 struct dn_flow *ni = (struct dn_flow *)(*a->start);
910 if (copy_obj(a->start, a->end, &si->ni, "inst",
911 si->sched->sch.sched_nr))
912 return DNHT_SCAN_END;
913 ni->oid.type = DN_FLOW; /* override the DN_SCH_I */
914 ni->oid.id = si_hash((uintptr_t)si, DNHT_KEY_IS_OBJ, NULL);
919 copy_si(struct copy_args *a, struct dn_schk *s, int flags)
921 if (s->sch.flags & DN_HAVE_MASK)
922 dn_ht_scan(s->siht, copy_si_cb, a);
924 copy_si_cb(s->siht, a);
929 * compute a list of children of a scheduler and copy up
932 copy_fsk_list(struct copy_args *a, struct dn_schk *s, int flags)
938 int n = 0, space = sizeof(*o);
939 SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
940 if (fs->fs.fs_nr < DN_MAX_ID)
943 space += n * sizeof(uint32_t);
944 DX(3, "sched %d has %d flowsets", s->sch.sched_nr, n);
945 if (a->end - *(a->start) < space)
946 return DNHT_SCAN_END;
947 o = (struct dn_id *)(*(a->start));
951 p = (uint32_t *)(o+1);
952 SLIST_FOREACH(fs, &s->fsk_list, sch_chain)
953 if (fs->fs.fs_nr < DN_MAX_ID)
959 copy_data_helper(void *_o, void *_arg)
961 struct copy_args *a = _arg;
962 uint32_t *r = a->extra->r; /* start of first range */
963 uint32_t *lim; /* first invalid pointer */
966 lim = (uint32_t *)((char *)(a->extra) + a->extra->o.len);
968 if (a->type == DN_LINK || a->type == DN_SCH) {
969 /* pipe|sched show, we receive a dn_schk */
970 struct dn_schk *s = _o;
973 if (a->type == DN_SCH && n >= DN_MAX_ID)
974 return 0; /* not a scheduler */
975 if (a->type == DN_LINK && n <= DN_MAX_ID)
976 return 0; /* not a pipe */
978 /* see if the object is within one of our ranges */
979 for (;r < lim; r += 2) {
980 if (n < r[0] || n > r[1])
982 /* Found a valid entry, copy and we are done */
983 if (a->flags & DN_C_LINK) {
984 if (copy_obj(a->start, a->end,
985 &s->link, "link", n))
986 return DNHT_SCAN_END;
987 if (copy_profile(a, s->profile))
988 return DNHT_SCAN_END;
989 if (copy_flowset(a, s->fs, 0))
990 return DNHT_SCAN_END;
992 if (a->flags & DN_C_SCH) {
993 if (copy_obj(a->start, a->end,
994 &s->sch, "sched", n))
995 return DNHT_SCAN_END;
996 /* list all attached flowsets */
997 if (copy_fsk_list(a, s, 0))
998 return DNHT_SCAN_END;
1000 if (a->flags & DN_C_FLOW)
1004 } else if (a->type == DN_FS) {
1005 /* queue show, skip internal flowsets */
1006 struct dn_fsk *fs = _o;
1011 /* see if the object is within one of our ranges */
1012 for (;r < lim; r += 2) {
1013 if (n < r[0] || n > r[1])
1015 if (copy_flowset(a, fs, 0))
1016 return DNHT_SCAN_END;
1018 break; /* we are done */
1024 static inline struct dn_schk *
1025 locate_scheduler(int i)
1027 return dn_ht_find(dn_cfg.schedhash, i, 0, NULL);
1031 * red parameters are in fixed point arithmetic.
1034 config_red(struct dn_fsk *fs)
1036 int64_t s, idle, weight, w0;
1039 fs->w_q = fs->fs.w_q;
1040 fs->max_p = fs->fs.max_p;
1042 /* Doing stuff that was in userland */
1043 i = fs->sched->link.bandwidth;
1045 hz * dn_cfg.red_avg_pkt_size * 8 * SCALE(1) / i;
1047 idle = div64((s * 3) , fs->w_q); /* s, fs->w_q scaled; idle not scaled */
1048 fs->lookup_step = div64(idle , dn_cfg.red_lookup_depth);
1049 /* fs->lookup_step not scaled, */
1050 if (!fs->lookup_step)
1051 fs->lookup_step = 1;
1052 w0 = weight = SCALE(1) - fs->w_q; //fs->w_q scaled
1054 for (t = fs->lookup_step; t > 1; --t)
1055 weight = SCALE_MUL(weight, w0);
1056 fs->lookup_weight = (int)(weight); // scaled
1058 /* Now doing stuff that was in kerneland */
1059 fs->min_th = SCALE(fs->fs.min_th);
1060 fs->max_th = SCALE(fs->fs.max_th);
1062 fs->c_1 = fs->max_p / (fs->fs.max_th - fs->fs.min_th);
1063 fs->c_2 = SCALE_MUL(fs->c_1, SCALE(fs->fs.min_th));
1065 if (fs->fs.flags & DN_IS_GENTLE_RED) {
1066 fs->c_3 = (SCALE(1) - fs->max_p) / fs->fs.max_th;
1067 fs->c_4 = SCALE(1) - 2 * fs->max_p;
1070 /* If the lookup table already exist, free and create it again. */
1071 if (fs->w_q_lookup) {
1072 free(fs->w_q_lookup, M_DUMMYNET);
1073 fs->w_q_lookup = NULL;
1075 if (dn_cfg.red_lookup_depth == 0) {
1076 printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1078 fs->fs.flags &= ~DN_IS_RED;
1079 fs->fs.flags &= ~DN_IS_GENTLE_RED;
1082 fs->lookup_depth = dn_cfg.red_lookup_depth;
1083 fs->w_q_lookup = (u_int *)malloc(fs->lookup_depth * sizeof(int),
1084 M_DUMMYNET, M_NOWAIT);
1085 if (fs->w_q_lookup == NULL) {
1086 printf("dummynet: sorry, cannot allocate red lookup table\n");
1087 fs->fs.flags &= ~DN_IS_RED;
1088 fs->fs.flags &= ~DN_IS_GENTLE_RED;
1092 /* Fill the lookup table with (1 - w_q)^x */
1093 fs->w_q_lookup[0] = SCALE(1) - fs->w_q;
1095 for (i = 1; i < fs->lookup_depth; i++)
1097 SCALE_MUL(fs->w_q_lookup[i - 1], fs->lookup_weight);
1099 if (dn_cfg.red_avg_pkt_size < 1)
1100 dn_cfg.red_avg_pkt_size = 512;
1101 fs->avg_pkt_size = dn_cfg.red_avg_pkt_size;
1102 if (dn_cfg.red_max_pkt_size < 1)
1103 dn_cfg.red_max_pkt_size = 1500;
1104 fs->max_pkt_size = dn_cfg.red_max_pkt_size;
1109 /* Scan all flowset attached to this scheduler and update red */
1111 update_red(struct dn_schk *s)
1114 SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1115 if (fs && (fs->fs.flags & DN_IS_RED))
1120 /* attach flowset to scheduler s, possibly requeue */
1122 fsk_attach(struct dn_fsk *fs, struct dn_schk *s)
1124 ND("remove fs %d from fsunlinked, link to sched %d",
1125 fs->fs.fs_nr, s->sch.sched_nr);
1126 SLIST_REMOVE(&dn_cfg.fsu, fs, dn_fsk, sch_chain);
1128 SLIST_INSERT_HEAD(&s->fsk_list, fs, sch_chain);
1131 /* XXX compute fsk_mask */
1132 fs->fsk_mask = fs->fs.flow_mask;
1133 if (fs->sched->sch.flags & DN_HAVE_MASK)
1134 flow_id_or(&fs->sched->sch.sched_mask, &fs->fsk_mask);
1137 * we must drain qht according to the old
1138 * type, and reinsert according to the new one.
1139 * The requeue is complex -- in general we need to
1140 * reclassify every single packet.
1141 * For the time being, let's hope qht is never set
1142 * when we reach this point.
1144 D("XXX TODO requeue from fs %d to sch %d",
1145 fs->fs.fs_nr, s->sch.sched_nr);
1148 /* set the new type for qht */
1149 if (nonzero_mask(&fs->fsk_mask))
1150 fs->fs.flags |= DN_QHT_HASH;
1152 fs->fs.flags &= ~DN_QHT_HASH;
1154 /* XXX config_red() can fail... */
1155 if (fs->fs.flags & DN_IS_RED)
1159 /* update all flowsets which may refer to this scheduler */
1161 update_fs(struct dn_schk *s)
1163 struct dn_fsk *fs, *tmp;
1165 SLIST_FOREACH_SAFE(fs, &dn_cfg.fsu, sch_chain, tmp) {
1166 if (s->sch.sched_nr != fs->fs.sched_nr) {
1167 D("fs %d for sch %d not %d still unlinked",
1168 fs->fs.fs_nr, fs->fs.sched_nr,
1177 * Configuration -- to preserve backward compatibility we use
1178 * the following scheme (N is 65536)
1179 * NUMBER SCHED LINK FLOWSET
1180 * 1 .. N-1 (1)WFQ (2)WFQ (3)queue
1181 * N+1 .. 2N-1 (4)FIFO (5)FIFO (6)FIFO for sched 1..N-1
1182 * 2N+1 .. 3N-1 -- -- (7)FIFO for sched N+1..2N-1
1184 * "pipe i config" configures #1, #2 and #3
1185 * "sched i config" configures #1 and possibly #6
1186 * "queue i config" configures #3
1187 * #1 is configured with 'pipe i config' or 'sched i config'
1188 * #2 is configured with 'pipe i config', and created if not
1189 * existing with 'sched i config'
1190 * #3 is configured with 'queue i config'
1191 * #4 is automatically configured after #1, can only be FIFO
1192 * #5 is automatically configured after #2
1193 * #6 is automatically created when #1 is !MULTIQUEUE,
1194 * and can be updated.
1195 * #7 is automatically configured after #2
1199 * configure a link (and its FIFO instance)
1202 config_link(struct dn_link *p, struct dn_id *arg)
1206 if (p->oid.len != sizeof(*p)) {
1207 D("invalid pipe len %d", p->oid.len);
1211 if (i <= 0 || i >= DN_MAX_ID)
1214 * The config program passes parameters as follows:
1215 * bw = bits/second (0 means no limits),
1216 * delay = ms, must be translated into ticks.
1217 * qsize = slots/bytes
1220 p->delay = (p->delay * hz) / 1000;
1221 /* Scale burst size: bytes -> bits * hz */
1225 /* do it twice, base link and FIFO link */
1226 for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1227 struct dn_schk *s = locate_scheduler(i);
1230 D("sched %d not found", i);
1233 /* remove profile if exists */
1235 free(s->profile, M_DUMMYNET);
1238 /* copy all parameters */
1239 s->link.oid = p->oid;
1240 s->link.link_nr = i;
1241 s->link.delay = p->delay;
1242 if (s->link.bandwidth != p->bandwidth) {
1243 /* XXX bandwidth changes, need to update red params */
1244 s->link.bandwidth = p->bandwidth;
1247 s->link.burst = p->burst;
1248 schk_reset_credit(s);
1256 * configure a flowset. Can be called from inside with locked=1,
1258 static struct dn_fsk *
1259 config_fs(struct dn_fs *nfs, struct dn_id *arg, int locked)
1264 if (nfs->oid.len != sizeof(*nfs)) {
1265 D("invalid flowset len %d", nfs->oid.len);
1269 if (i <= 0 || i >= 3*DN_MAX_ID)
1271 ND("flowset %d", i);
1272 /* XXX other sanity checks */
1273 if (nfs->flags & DN_QSIZE_BYTES) {
1274 ipdn_bound_var(&nfs->qsize, 16384,
1275 1500, dn_cfg.byte_limit, NULL); // "queue byte size");
1277 ipdn_bound_var(&nfs->qsize, 50,
1278 1, dn_cfg.slot_limit, NULL); // "queue slot size");
1280 if (nfs->flags & DN_HAVE_MASK) {
1281 /* make sure we have some buckets */
1282 ipdn_bound_var(&nfs->buckets, dn_cfg.hash_size,
1283 1, dn_cfg.max_hash_size, "flowset buckets");
1285 nfs->buckets = 1; /* we only need 1 */
1289 do { /* exit with break when done */
1291 int flags = nfs->sched_nr ? DNHT_INSERT : 0;
1293 int oldc = dn_cfg.fsk_count;
1294 fs = dn_ht_find(dn_cfg.fshash, i, flags, NULL);
1296 D("missing sched for flowset %d", i);
1299 /* grab some defaults from the existing one */
1300 if (nfs->sched_nr == 0) /* reuse */
1301 nfs->sched_nr = fs->fs.sched_nr;
1302 for (j = 0; j < sizeof(nfs->par)/sizeof(nfs->par[0]); j++) {
1303 if (nfs->par[j] == -1) /* reuse */
1304 nfs->par[j] = fs->fs.par[j];
1306 if (bcmp(&fs->fs, nfs, sizeof(*nfs)) == 0) {
1307 ND("flowset %d unchanged", i);
1308 break; /* no change, nothing to do */
1310 if (oldc != dn_cfg.fsk_count) /* new item */
1312 s = locate_scheduler(nfs->sched_nr);
1313 /* detach from old scheduler if needed, preserving
1314 * queues if we need to reattach. Then update the
1315 * configuration, and possibly attach to the new sched.
1317 DX(2, "fs %d changed sched %d@%p to %d@%p",
1319 fs->fs.sched_nr, fs->sched, nfs->sched_nr, s);
1321 int flags = s ? DN_DETACH : (DN_DETACH | DN_DESTROY);
1322 flags |= DN_DESTROY; /* XXX temporary */
1323 fsk_detach(fs, flags);
1325 fs->fs = *nfs; /* copy configuration */
1335 * config/reconfig a scheduler and its FIFO variant.
1336 * For !MULTIQUEUE schedulers, also set up the flowset.
1338 * On reconfigurations (detected because s->fp is set),
1339 * detach existing flowsets preserving traffic, preserve link,
1340 * and delete the old scheduler creating a new one.
1343 config_sched(struct dn_sch *_nsch, struct dn_id *arg)
1346 struct schk_new_arg a; /* argument for schk_new */
1348 struct dn_link p; /* copy of oldlink */
1349 struct dn_profile *pf = NULL; /* copy of old link profile */
1350 /* Used to preserv mask parameter */
1351 struct ipfw_flow_id new_mask;
1352 int new_buckets = 0;
1358 if (a.sch->oid.len != sizeof(*a.sch)) {
1359 D("bad sched len %d", a.sch->oid.len);
1362 i = a.sch->sched_nr;
1363 if (i <= 0 || i >= DN_MAX_ID)
1365 /* make sure we have some buckets */
1366 if (a.sch->flags & DN_HAVE_MASK)
1367 ipdn_bound_var(&a.sch->buckets, dn_cfg.hash_size,
1368 1, dn_cfg.max_hash_size, "sched buckets");
1369 /* XXX other sanity checks */
1370 bzero(&p, sizeof(p));
1372 pipe_cmd = a.sch->flags & DN_PIPE_CMD;
1373 a.sch->flags &= ~DN_PIPE_CMD; //XXX do it even if is not set?
1375 /* Copy mask parameter */
1376 new_mask = a.sch->sched_mask;
1377 new_buckets = a.sch->buckets;
1378 new_flags = a.sch->flags;
1381 again: /* run twice, for wfq and fifo */
1383 * lookup the type. If not supplied, use the previous one
1384 * or default to WF2Q+. Otherwise, return an error.
1387 a.fp = find_sched_type(a.sch->oid.subtype, a.sch->name);
1389 /* found. Lookup or create entry */
1390 s = dn_ht_find(dn_cfg.schedhash, i, DNHT_INSERT, &a);
1391 } else if (a.sch->oid.subtype == 0 && !a.sch->name[0]) {
1392 /* No type. search existing s* or retry with WF2Q+ */
1393 s = dn_ht_find(dn_cfg.schedhash, i, 0, &a);
1396 /* Scheduler exists, skip to FIFO scheduler
1397 * if command was pipe config...
1402 /* New scheduler, create a wf2q+ with no mask
1403 * if command was pipe config...
1406 /* clear mask parameter */
1407 bzero(&a.sch->sched_mask, sizeof(new_mask));
1409 a.sch->flags &= ~DN_HAVE_MASK;
1411 a.sch->oid.subtype = DN_SCHED_WF2QP;
1415 D("invalid scheduler type %d %s",
1416 a.sch->oid.subtype, a.sch->name);
1420 /* normalize name and subtype */
1421 a.sch->oid.subtype = a.fp->type;
1422 bzero(a.sch->name, sizeof(a.sch->name));
1423 strlcpy(a.sch->name, a.fp->name, sizeof(a.sch->name));
1425 D("cannot allocate scheduler %d", i);
1428 /* restore existing link if any */
1431 if (!pf || pf->link_nr != p.link_nr) { /* no saved value */
1432 s->profile = NULL; /* XXX maybe not needed */
1434 s->profile = malloc(sizeof(struct dn_profile),
1435 M_DUMMYNET, M_NOWAIT | M_ZERO);
1436 if (s->profile == NULL) {
1437 D("cannot allocate profile");
1440 bcopy(pf, s->profile, sizeof(*pf));
1444 if (s->fp == NULL) {
1445 DX(2, "sched %d new type %s", i, a.fp->name);
1446 } else if (s->fp != a.fp ||
1447 bcmp(a.sch, &s->sch, sizeof(*a.sch)) ) {
1448 /* already existing. */
1449 DX(2, "sched %d type changed from %s to %s",
1450 i, s->fp->name, a.fp->name);
1451 DX(4, " type/sub %d/%d -> %d/%d",
1452 s->sch.oid.type, s->sch.oid.subtype,
1453 a.sch->oid.type, a.sch->oid.subtype);
1454 if (s->link.link_nr == 0)
1455 D("XXX WARNING link 0 for sched %d", i);
1456 p = s->link; /* preserve link */
1457 if (s->profile) {/* preserve profile */
1459 pf = malloc(sizeof(*pf),
1460 M_DUMMYNET, M_NOWAIT | M_ZERO);
1461 if (pf) /* XXX should issue a warning otherwise */
1462 bcopy(s->profile, pf, sizeof(*pf));
1464 /* remove from the hash */
1465 dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
1466 /* Detach flowsets, preserve queues. */
1467 // schk_delete_cb(s, NULL);
1468 // XXX temporarily, kill queues
1469 schk_delete_cb(s, (void *)DN_DESTROY);
1472 DX(4, "sched %d unchanged type %s", i, a.fp->name);
1474 /* complete initialization */
1478 // XXX schk_reset_credit(s);
1479 /* create the internal flowset if needed,
1480 * trying to reuse existing ones if available
1482 if (!(s->fp->flags & DN_MULTIQUEUE) && !s->fs) {
1483 s->fs = dn_ht_find(dn_cfg.fshash, i, 0, NULL);
1486 bzero(&fs, sizeof(fs));
1487 set_oid(&fs.oid, DN_FS, sizeof(fs));
1488 fs.fs_nr = i + DN_MAX_ID;
1490 s->fs = config_fs(&fs, NULL, 1 /* locked */);
1493 schk_delete_cb(s, (void *)DN_DESTROY);
1494 D("error creating internal fs for %d", i);
1498 /* call init function after the flowset is created */
1503 if (i < DN_MAX_ID) { /* now configure the FIFO instance */
1506 /* Restore mask parameter for FIFO */
1507 a.sch->sched_mask = new_mask;
1508 a.sch->buckets = new_buckets;
1509 a.sch->flags = new_flags;
1511 /* sched config shouldn't modify the FIFO scheduler */
1512 if (dn_ht_find(dn_cfg.schedhash, i, 0, &a) != NULL) {
1513 /* FIFO already exist, don't touch it */
1514 err = 0; /* and this is not an error */
1518 a.sch->sched_nr = i;
1519 a.sch->oid.subtype = DN_SCHED_FIFO;
1520 bzero(a.sch->name, sizeof(a.sch->name));
1527 free(pf, M_DUMMYNET);
1532 * attach a profile to a link
1535 config_profile(struct dn_profile *pf, struct dn_id *arg)
1538 int i, olen, err = 0;
1540 if (pf->oid.len < sizeof(*pf)) {
1541 D("short profile len %d", pf->oid.len);
1545 if (i <= 0 || i >= DN_MAX_ID)
1547 /* XXX other sanity checks */
1549 for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1550 s = locate_scheduler(i);
1558 * If we had a profile and the new one does not fit,
1559 * or it is deleted, then we need to free memory.
1561 if (s->profile && (pf->samples_no == 0 ||
1562 s->profile->oid.len < pf->oid.len)) {
1563 free(s->profile, M_DUMMYNET);
1566 if (pf->samples_no == 0)
1569 * new profile, possibly allocate memory
1572 if (s->profile == NULL)
1573 s->profile = malloc(pf->oid.len,
1574 M_DUMMYNET, M_NOWAIT | M_ZERO);
1575 if (s->profile == NULL) {
1576 D("no memory for profile %d", i);
1580 /* preserve larger length XXX double check */
1581 olen = s->profile->oid.len;
1582 if (olen < pf->oid.len)
1584 bcopy(pf, s->profile, pf->oid.len);
1585 s->profile->oid.len = olen;
1592 * Delete all objects:
1595 dummynet_flush(void)
1598 /* delete all schedulers and related links/queues/flowsets */
1599 dn_ht_scan(dn_cfg.schedhash, schk_delete_cb,
1600 (void *)(uintptr_t)DN_DELETE_FS);
1601 /* delete all remaining (unlinked) flowsets */
1602 DX(4, "still %d unlinked fs", dn_cfg.fsk_count);
1603 dn_ht_free(dn_cfg.fshash, DNHT_REMOVE);
1604 fsk_detach_list(&dn_cfg.fsu, DN_DELETE_FS);
1605 /* Reinitialize system heap... */
1606 heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
1610 * Main handler for configuration. We are guaranteed to be called
1611 * with an oid which is at least a dn_id.
1612 * - the first object is the command (config, delete, flush, ...)
1613 * - config_link must be issued after the corresponding config_sched
1614 * - parameters (DN_TXT) for an object must preceed the object
1615 * processed on a config_sched.
1618 do_config(void *p, int l)
1620 struct dn_id *next, *o;
1621 int err = 0, err2 = 0;
1622 struct dn_id *arg = NULL;
1626 if (o->id != DN_API_VERSION) {
1627 D("invalid api version got %d need %d",
1628 o->id, DN_API_VERSION);
1631 for (; l >= sizeof(*o); o = next) {
1632 struct dn_id *prev = arg;
1633 if (o->len < sizeof(*o) || l < o->len) {
1634 D("bad len o->len %d len %d", o->len, l);
1639 next = (struct dn_id *)((char *)o + o->len);
1643 D("cmd %d not implemented", o->type);
1645 #ifdef EMULATE_SYSCTL
1646 /* sysctl emulation.
1647 * if we recognize the command, jump to the correct
1648 * handler and return
1651 err = kesysctl_emu_set(p, l);
1654 case DN_CMD_CONFIG: /* simply a header */
1658 /* the argument is in the first uintptr_t after o */
1659 a = (uintptr_t *)(o+1);
1660 if (o->len < sizeof(*o) + sizeof(*a)) {
1664 switch (o->subtype) {
1666 /* delete base and derived schedulers */
1668 err = delete_schk(*a);
1669 err2 = delete_schk(*a + DN_MAX_ID);
1676 D("invalid delete type %d",
1682 err = (*a <1 || *a >= DN_MAX_ID) ?
1683 EINVAL : delete_fs(*a, 0) ;
1693 case DN_TEXT: /* store argument the next block */
1698 err = config_link((struct dn_link *)o, arg);
1701 err = config_profile((struct dn_profile *)o, arg);
1704 err = config_sched((struct dn_sch *)o, arg);
1707 err = (NULL==config_fs((struct dn_fs *)o, arg, 0));
1719 compute_space(struct dn_id *cmd, struct copy_args *a)
1721 int x = 0, need = 0;
1722 int profile_size = sizeof(struct dn_profile) -
1723 ED_MAX_SAMPLES_NO*sizeof(int);
1725 /* NOTE about compute space:
1726 * NP = dn_cfg.schk_count
1727 * NSI = dn_cfg.si_count
1728 * NF = dn_cfg.fsk_count
1729 * NQ = dn_cfg.queue_count
1731 * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
1732 * link, scheduler template, flowset
1733 * integrated in scheduler and header
1735 * (NSI)*(dn_flow) all scheduler instance (includes
1736 * the queue instance)
1738 * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
1739 * link, scheduler template, flowset
1740 * integrated in scheduler and header
1742 * (NSI * dn_flow) all scheduler instances
1743 * (NF * sizeof(uint_32)) space for flowset list linked to scheduler
1744 * (NQ * dn_queue) all queue [XXXfor now not listed]
1746 * (NF * dn_fs) all flowset
1747 * (NQ * dn_queue) all queues
1749 switch (cmd->subtype) {
1752 /* XXX where do LINK and SCH differ ? */
1753 /* 'ipfw sched show' could list all queues associated to
1754 * a scheduler. This feature for now is disabled
1756 case DN_LINK: /* pipe show */
1757 x = DN_C_LINK | DN_C_SCH | DN_C_FLOW;
1758 need += dn_cfg.schk_count *
1759 (sizeof(struct dn_fs) + profile_size) / 2;
1760 need += dn_cfg.fsk_count * sizeof(uint32_t);
1762 case DN_SCH: /* sched show */
1763 need += dn_cfg.schk_count *
1764 (sizeof(struct dn_fs) + profile_size) / 2;
1765 need += dn_cfg.fsk_count * sizeof(uint32_t);
1766 x = DN_C_SCH | DN_C_LINK | DN_C_FLOW;
1768 case DN_FS: /* queue show */
1769 x = DN_C_FS | DN_C_QUEUE;
1771 case DN_GET_COMPAT: /* compatibility mode */
1772 need = dn_compat_calc_size(dn_cfg);
1777 need += dn_cfg.schk_count * sizeof(struct dn_sch) / 2;
1778 /* NOT also, each fs might be attached to a sched */
1779 need += dn_cfg.schk_count * sizeof(struct dn_id) / 2;
1782 need += dn_cfg.fsk_count * sizeof(struct dn_fs);
1783 if (x & DN_C_LINK) {
1784 need += dn_cfg.schk_count * sizeof(struct dn_link) / 2;
1787 * When exporting a queue to userland, only pass up the
1788 * struct dn_flow, which is the only visible part.
1792 need += dn_cfg.queue_count * sizeof(struct dn_flow);
1794 need += dn_cfg.si_count * (sizeof(struct dn_flow));
1799 * If compat != NULL dummynet_get is called in compatibility mode.
1800 * *compat will be the pointer to the buffer to pass to ipfw
1803 dummynet_get(struct sockopt *sopt, void **compat)
1805 int have, i, need, error;
1806 char *start = NULL, *buf;
1807 size_t sopt_valsize;
1810 struct copy_range r;
1811 int l = sizeof(struct dn_id);
1813 bzero(&a, sizeof(a));
1814 bzero(&r, sizeof(r));
1816 /* save and restore original sopt_valsize around copyin */
1817 sopt_valsize = sopt->sopt_valsize;
1822 /* copy at least an oid, and possibly a full object */
1823 error = sooptcopyin(sopt, cmd, sizeof(r), sizeof(*cmd));
1824 sopt->sopt_valsize = sopt_valsize;
1828 #ifdef EMULATE_SYSCTL
1829 /* sysctl emulation. */
1830 if (cmd->type == DN_SYSCTL_GET)
1831 return kesysctl_emu_get(sopt);
1833 if (l > sizeof(r)) {
1834 /* request larger than default, allocate buffer */
1835 cmd = malloc(l, M_DUMMYNET, M_WAIT);
1837 return ENOMEM; //XXX
1838 error = sooptcopyin(sopt, cmd, l, l);
1839 sopt->sopt_valsize = sopt_valsize;
1843 } else { /* compatibility */
1845 cmd->type = DN_CMD_GET;
1846 cmd->len = sizeof(struct dn_id);
1847 cmd->subtype = DN_GET_COMPAT;
1848 // cmd->id = sopt_valsize;
1849 D("compatibility mode");
1851 a.extra = (struct copy_range *)cmd;
1852 if (cmd->len == sizeof(*cmd)) { /* no range, create a default */
1853 uint32_t *rp = (uint32_t *)(cmd + 1);
1854 cmd->len += 2* sizeof(uint32_t);
1856 rp[1] = DN_MAX_ID - 1;
1857 if (cmd->subtype == DN_LINK) {
1862 /* Count space (under lock) and allocate (outside lock).
1863 * Exit with lock held if we manage to get enough buffer.
1864 * Try a few times then give up.
1866 for (have = 0, i = 0; i < 10; i++) {
1868 need = compute_space(cmd, &a);
1870 /* if there is a range, ignore value from compute_space() */
1871 if (l > sizeof(*cmd))
1872 need = sopt_valsize - sizeof(*cmd);
1879 need += sizeof(*cmd);
1886 free(start, M_DUMMYNET);
1888 if (need > sopt_valsize)
1892 start = malloc(have, M_DUMMYNET, M_WAITOK | M_ZERO);
1893 if (start == NULL) {
1899 if (start == NULL) {
1904 error = sooptcopyout(sopt, cmd, sizeof(*cmd));
1908 ND("have %d:%d sched %d, %d:%d links %d, %d:%d flowsets %d, "
1909 "%d:%d si %d, %d:%d queues %d",
1910 dn_cfg.schk_count, sizeof(struct dn_sch), DN_SCH,
1911 dn_cfg.schk_count, sizeof(struct dn_link), DN_LINK,
1912 dn_cfg.fsk_count, sizeof(struct dn_fs), DN_FS,
1913 dn_cfg.si_count, sizeof(struct dn_flow), DN_SCH_I,
1914 dn_cfg.queue_count, sizeof(struct dn_queue), DN_QUEUE);
1915 sopt->sopt_valsize = sopt_valsize;
1916 a.type = cmd->subtype;
1918 if (compat == NULL) {
1919 bcopy(cmd, start, sizeof(*cmd));
1920 ((struct dn_id*)(start))->len = sizeof(struct dn_id);
1921 buf = start + sizeof(*cmd);
1925 a.end = start + have;
1926 /* start copying other objects */
1928 a.type = DN_COMPAT_PIPE;
1929 dn_ht_scan(dn_cfg.schedhash, copy_data_helper_compat, &a);
1930 a.type = DN_COMPAT_QUEUE;
1931 dn_ht_scan(dn_cfg.fshash, copy_data_helper_compat, &a);
1932 } else if (a.type == DN_FS) {
1933 dn_ht_scan(dn_cfg.fshash, copy_data_helper, &a);
1935 dn_ht_scan(dn_cfg.schedhash, copy_data_helper, &a);
1941 sopt->sopt_valsize = buf - start;
1942 /* free() is done by ip_dummynet_compat() */
1943 start = NULL; //XXX hack
1945 error = sooptcopyout(sopt, start, buf - start);
1948 if (cmd && cmd != &r.o)
1949 free(cmd, M_DUMMYNET);
1951 free(start, M_DUMMYNET);
1955 /* Callback called on scheduler instance to delete it if idle */
1957 drain_scheduler_cb(void *_si, void *arg)
1959 struct dn_sch_inst *si = _si;
1961 if ((si->kflags & DN_ACTIVE) || si->dline.mq.head != NULL)
1964 if (si->sched->fp->flags & DN_MULTIQUEUE) {
1965 if (si->q_count == 0)
1966 return si_destroy(si, NULL);
1969 } else { /* !DN_MULTIQUEUE */
1970 if ((si+1)->ni.length == 0)
1971 return si_destroy(si, NULL);
1975 return 0; /* unreachable */
1978 /* Callback called on scheduler to check if it has instances */
1980 drain_scheduler_sch_cb(void *_s, void *arg)
1982 struct dn_schk *s = _s;
1984 if (s->sch.flags & DN_HAVE_MASK) {
1985 dn_ht_scan_bucket(s->siht, &s->drain_bucket,
1986 drain_scheduler_cb, NULL);
1990 if (drain_scheduler_cb(s->siht, NULL) == DNHT_SCAN_DEL)
1997 /* Called every tick, try to delete a 'bucket' of scheduler */
1999 dn_drain_scheduler(void)
2001 dn_ht_scan_bucket(dn_cfg.schedhash, &dn_cfg.drain_sch,
2002 drain_scheduler_sch_cb, NULL);
2006 /* Callback called on queue to delete if it is idle */
2008 drain_queue_cb(void *_q, void *arg)
2010 struct dn_queue *q = _q;
2012 if (q->ni.length == 0) {
2013 dn_delete_queue(q, DN_DESTROY);
2014 return DNHT_SCAN_DEL; /* queue is deleted */
2017 return 0; /* queue isn't deleted */
2020 /* Callback called on flowset used to check if it has queues */
2022 drain_queue_fs_cb(void *_fs, void *arg)
2024 struct dn_fsk *fs = _fs;
2026 if (fs->fs.flags & DN_QHT_HASH) {
2027 /* Flowset has a hash table for queues */
2028 dn_ht_scan_bucket(fs->qht, &fs->drain_bucket,
2029 drain_queue_cb, NULL);
2032 /* No hash table for this flowset, null the pointer
2033 * if the queue is deleted
2036 if (drain_queue_cb(fs->qht, NULL) == DNHT_SCAN_DEL)
2043 /* Called every tick, try to delete a 'bucket' of queue */
2045 dn_drain_queue(void)
2047 /* scan a bucket of flowset */
2048 dn_ht_scan_bucket(dn_cfg.fshash, &dn_cfg.drain_fs,
2049 drain_queue_fs_cb, NULL);
2054 * Handler for the various dummynet socket options
2057 ip_dn_ctl(struct sockopt *sopt)
2062 error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2066 /* Disallow sets in really-really secure mode. */
2067 if (sopt->sopt_dir == SOPT_SET) {
2068 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2073 switch (sopt->sopt_name) {
2075 D("dummynet: unknown option %d", sopt->sopt_name);
2079 case IP_DUMMYNET_FLUSH:
2080 case IP_DUMMYNET_CONFIGURE:
2081 case IP_DUMMYNET_DEL: /* remove a pipe or queue */
2082 case IP_DUMMYNET_GET:
2083 D("dummynet: compat option %d", sopt->sopt_name);
2084 error = ip_dummynet_compat(sopt);
2088 if (sopt->sopt_dir == SOPT_GET) {
2089 error = dummynet_get(sopt, NULL);
2092 l = sopt->sopt_valsize;
2093 if (l < sizeof(struct dn_id) || l > 12000) {
2094 D("argument len %d invalid", l);
2097 p = malloc(l, M_TEMP, M_WAITOK); // XXX can it fail ?
2098 error = sooptcopyin(sopt, p, l, l);
2101 error = do_config(p, l);
2115 static int init_done = 0;
2121 printf("DUMMYNET with IPv6 initialized (100131)\n");
2123 /* Set defaults here. MSVC does not accept initializers,
2124 * and this is also useful for vimages
2127 dn_cfg.slot_limit = 100; /* Foot shooting limit for queues. */
2128 dn_cfg.byte_limit = 1024 * 1024;
2131 /* RED parameters */
2132 dn_cfg.red_lookup_depth = 256; /* default lookup table depth */
2133 dn_cfg.red_avg_pkt_size = 512; /* default medium packet size */
2134 dn_cfg.red_max_pkt_size = 1500; /* default max packet size */
2137 dn_cfg.max_hash_size = 1024; /* max in the hash tables */
2138 dn_cfg.hash_size = 64; /* default hash size */
2140 /* create hash tables for schedulers and flowsets.
2141 * In both we search by key and by pointer.
2143 dn_cfg.schedhash = dn_ht_init(NULL, dn_cfg.hash_size,
2144 offsetof(struct dn_schk, schk_next),
2145 schk_hash, schk_match, schk_new);
2146 dn_cfg.fshash = dn_ht_init(NULL, dn_cfg.hash_size,
2147 offsetof(struct dn_fsk, fsk_next),
2148 fsk_hash, fsk_match, fsk_new);
2150 /* bucket index to drain object */
2151 dn_cfg.drain_fs = 0;
2152 dn_cfg.drain_sch = 0;
2154 heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
2155 SLIST_INIT(&dn_cfg.fsu);
2156 SLIST_INIT(&dn_cfg.schedlist);
2159 ip_dn_ctl_ptr = ip_dn_ctl;
2160 ip_dn_io_ptr = dummynet_io;
2162 TASK_INIT(&dn_task, 0, dummynet_task, NULL);
2163 dn_tq = taskqueue_create_fast("dummynet", M_NOWAIT,
2164 taskqueue_thread_enqueue, &dn_tq);
2165 taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2167 callout_init(&dn_timeout, CALLOUT_MPSAFE);
2168 callout_reset(&dn_timeout, 1, dummynet, NULL);
2170 /* Initialize curr_time adjustment mechanics. */
2171 getmicrouptime(&dn_cfg.prev_t);
2178 callout_drain(&dn_timeout);
2181 ip_dn_ctl_ptr = NULL;
2182 ip_dn_io_ptr = NULL;
2186 taskqueue_drain(dn_tq, &dn_task);
2187 taskqueue_free(dn_tq);
2189 dn_ht_free(dn_cfg.schedhash, 0);
2190 dn_ht_free(dn_cfg.fshash, 0);
2191 heap_free(&dn_cfg.evheap);
2195 #endif /* KLD_MODULE */
2198 dummynet_modevent(module_t mod, int type, void *data)
2201 if (type == MOD_LOAD) {
2203 printf("DUMMYNET already loaded\n");
2208 } else if (type == MOD_UNLOAD) {
2209 #if !defined(KLD_MODULE)
2210 printf("dummynet statically compiled, cannot unload\n");
2220 /* modevent helpers for the modules */
2222 load_dn_sched(struct dn_alg *d)
2227 return 1; /* error */
2228 ip_dn_init(); /* just in case, we need the lock */
2230 /* Check that mandatory funcs exists */
2231 if (d->enqueue == NULL || d->dequeue == NULL) {
2232 D("missing enqueue or dequeue for %s", d->name);
2236 /* Search if scheduler already exists */
2238 SLIST_FOREACH(s, &dn_cfg.schedlist, next) {
2239 if (strcmp(s->name, d->name) == 0) {
2240 D("%s already loaded", d->name);
2241 break; /* scheduler already exists */
2245 SLIST_INSERT_HEAD(&dn_cfg.schedlist, d, next);
2247 D("dn_sched %s %sloaded", d->name, s ? "not ":"");
2252 unload_dn_sched(struct dn_alg *s)
2254 struct dn_alg *tmp, *r;
2257 D("called for %s", s->name);
2260 SLIST_FOREACH_SAFE(r, &dn_cfg.schedlist, next, tmp) {
2261 if (strcmp(s->name, r->name) != 0)
2263 D("ref_count = %d", r->ref_count);
2264 err = (r->ref_count != 0) ? EBUSY : 0;
2266 SLIST_REMOVE(&dn_cfg.schedlist, r, dn_alg, next);
2270 D("dn_sched %s %sunloaded", s->name, err ? "not ":"");
2275 dn_sched_modevent(module_t mod, int cmd, void *arg)
2277 struct dn_alg *sch = arg;
2279 if (cmd == MOD_LOAD)
2280 return load_dn_sched(sch);
2281 else if (cmd == MOD_UNLOAD)
2282 return unload_dn_sched(sch);
2287 static moduledata_t dummynet_mod = {
2288 "dummynet", dummynet_modevent, NULL
2291 DECLARE_MODULE(dummynet, dummynet_mod,
2292 SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY-1);
2293 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
2294 MODULE_VERSION(dummynet, 1);