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/ip_dummynet.h>
57 #include <netpfil/ipfw/ip_fw_private.h>
58 #include <netpfil/ipfw/dn_heap.h>
59 #include <netpfil/ipfw/ip_dn_private.h>
60 #include <netpfil/ipfw/dn_sched.h>
62 /* which objects to copy */
63 #define DN_C_LINK 0x01
65 #define DN_C_FLOW 0x04
67 #define DN_C_QUEUE 0x10
69 /* we use this argument in case of a schk_new */
75 /*---- callout hooks. ----*/
76 static struct callout dn_timeout;
77 static struct task dn_task;
78 static struct taskqueue *dn_tq = NULL;
81 dummynet(void * __unused unused)
84 taskqueue_enqueue(dn_tq, &dn_task);
90 callout_reset(&dn_timeout, 1, dummynet, NULL);
92 /*----- end of callout hooks -----*/
94 /* Return a scheduler descriptor given the type or name. */
95 static struct dn_alg *
96 find_sched_type(int type, char *name)
100 SLIST_FOREACH(d, &dn_cfg.schedlist, next) {
101 if (d->type == type || (name && !strcasecmp(d->name, name)))
104 return NULL; /* not found */
108 ipdn_bound_var(int *v, int dflt, int lo, int hi, const char *msg)
111 const char *op = NULL;
119 } else if (oldv > hi) {
125 printf("%s %s to %d (was %d)\n", op, msg, *v, oldv);
129 /*---- flow_id mask, hash and compare functions ---*/
131 * The flow_id includes the 5-tuple, the queue/pipe number
132 * which we store in the extra area in host order,
133 * and for ipv6 also the flow_id6.
134 * XXX see if we want the tos byte (can store in 'flags')
136 static struct ipfw_flow_id *
137 flow_id_mask(struct ipfw_flow_id *mask, struct ipfw_flow_id *id)
139 int is_v6 = IS_IP6_FLOW_ID(id);
141 id->dst_port &= mask->dst_port;
142 id->src_port &= mask->src_port;
143 id->proto &= mask->proto;
144 id->extra &= mask->extra;
146 APPLY_MASK(&id->dst_ip6, &mask->dst_ip6);
147 APPLY_MASK(&id->src_ip6, &mask->src_ip6);
148 id->flow_id6 &= mask->flow_id6;
150 id->dst_ip &= mask->dst_ip;
151 id->src_ip &= mask->src_ip;
156 /* computes an OR of two masks, result in dst and also returned */
157 static struct ipfw_flow_id *
158 flow_id_or(struct ipfw_flow_id *src, struct ipfw_flow_id *dst)
160 int is_v6 = IS_IP6_FLOW_ID(dst);
162 dst->dst_port |= src->dst_port;
163 dst->src_port |= src->src_port;
164 dst->proto |= src->proto;
165 dst->extra |= src->extra;
167 #define OR_MASK(_d, _s) \
168 (_d)->__u6_addr.__u6_addr32[0] |= (_s)->__u6_addr.__u6_addr32[0]; \
169 (_d)->__u6_addr.__u6_addr32[1] |= (_s)->__u6_addr.__u6_addr32[1]; \
170 (_d)->__u6_addr.__u6_addr32[2] |= (_s)->__u6_addr.__u6_addr32[2]; \
171 (_d)->__u6_addr.__u6_addr32[3] |= (_s)->__u6_addr.__u6_addr32[3];
172 OR_MASK(&dst->dst_ip6, &src->dst_ip6);
173 OR_MASK(&dst->src_ip6, &src->src_ip6);
175 dst->flow_id6 |= src->flow_id6;
177 dst->dst_ip |= src->dst_ip;
178 dst->src_ip |= src->src_ip;
184 nonzero_mask(struct ipfw_flow_id *m)
186 if (m->dst_port || m->src_port || m->proto || m->extra)
188 if (IS_IP6_FLOW_ID(m)) {
190 m->dst_ip6.__u6_addr.__u6_addr32[0] ||
191 m->dst_ip6.__u6_addr.__u6_addr32[1] ||
192 m->dst_ip6.__u6_addr.__u6_addr32[2] ||
193 m->dst_ip6.__u6_addr.__u6_addr32[3] ||
194 m->src_ip6.__u6_addr.__u6_addr32[0] ||
195 m->src_ip6.__u6_addr.__u6_addr32[1] ||
196 m->src_ip6.__u6_addr.__u6_addr32[2] ||
197 m->src_ip6.__u6_addr.__u6_addr32[3] ||
200 return m->dst_ip || m->src_ip;
204 /* XXX we may want a better hash function */
206 flow_id_hash(struct ipfw_flow_id *id)
210 if (IS_IP6_FLOW_ID(id)) {
211 uint32_t *d = (uint32_t *)&id->dst_ip6;
212 uint32_t *s = (uint32_t *)&id->src_ip6;
213 i = (d[0] ) ^ (d[1]) ^
215 (d[0] >> 15) ^ (d[1] >> 15) ^
216 (d[2] >> 15) ^ (d[3] >> 15) ^
217 (s[0] << 1) ^ (s[1] << 1) ^
218 (s[2] << 1) ^ (s[3] << 1) ^
219 (s[0] << 16) ^ (s[1] << 16) ^
220 (s[2] << 16) ^ (s[3] << 16) ^
221 (id->dst_port << 1) ^ (id->src_port) ^
223 (id->proto ) ^ (id->flow_id6);
225 i = (id->dst_ip) ^ (id->dst_ip >> 15) ^
226 (id->src_ip << 1) ^ (id->src_ip >> 16) ^
228 (id->dst_port << 1) ^ (id->src_port) ^ (id->proto);
233 /* Like bcmp, returns 0 if ids match, 1 otherwise. */
235 flow_id_cmp(struct ipfw_flow_id *id1, struct ipfw_flow_id *id2)
237 int is_v6 = IS_IP6_FLOW_ID(id1);
240 if (IS_IP6_FLOW_ID(id2))
241 return 1; /* different address families */
243 return (id1->dst_ip == id2->dst_ip &&
244 id1->src_ip == id2->src_ip &&
245 id1->dst_port == id2->dst_port &&
246 id1->src_port == id2->src_port &&
247 id1->proto == id2->proto &&
248 id1->extra == id2->extra) ? 0 : 1;
252 !bcmp(&id1->dst_ip6,&id2->dst_ip6, sizeof(id1->dst_ip6)) &&
253 !bcmp(&id1->src_ip6,&id2->src_ip6, sizeof(id1->src_ip6)) &&
254 id1->dst_port == id2->dst_port &&
255 id1->src_port == id2->src_port &&
256 id1->proto == id2->proto &&
257 id1->extra == id2->extra &&
258 id1->flow_id6 == id2->flow_id6) ? 0 : 1;
260 /*--------- end of flow-id mask, hash and compare ---------*/
262 /*--- support functions for the qht hashtable ----
263 * Entries are hashed by flow-id
266 q_hash(uintptr_t key, int flags, void *arg)
268 /* compute the hash slot from the flow id */
269 struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
270 &((struct dn_queue *)key)->ni.fid :
271 (struct ipfw_flow_id *)key;
273 return flow_id_hash(id);
277 q_match(void *obj, uintptr_t key, int flags, void *arg)
279 struct dn_queue *o = (struct dn_queue *)obj;
280 struct ipfw_flow_id *id2;
282 if (flags & DNHT_KEY_IS_OBJ) {
283 /* compare pointers */
284 id2 = &((struct dn_queue *)key)->ni.fid;
286 id2 = (struct ipfw_flow_id *)key;
288 return (0 == flow_id_cmp(&o->ni.fid, id2));
292 * create a new queue instance for the given 'key'.
295 q_new(uintptr_t key, int flags, void *arg)
297 struct dn_queue *q, *template = arg;
298 struct dn_fsk *fs = template->fs;
299 int size = sizeof(*q) + fs->sched->fp->q_datalen;
301 q = malloc(size, M_DUMMYNET, M_NOWAIT | M_ZERO);
303 D("no memory for new queue");
307 set_oid(&q->ni.oid, DN_QUEUE, size);
308 if (fs->fs.flags & DN_QHT_HASH)
309 q->ni.fid = *(struct ipfw_flow_id *)key;
311 q->_si = template->_si;
314 if (fs->sched->fp->new_queue)
315 fs->sched->fp->new_queue(q);
316 dn_cfg.queue_count++;
321 * Notify schedulers that a queue is going away.
322 * If (flags & DN_DESTROY), also free the packets.
323 * The version for callbacks is called q_delete_cb().
326 dn_delete_queue(struct dn_queue *q, int flags)
328 struct dn_fsk *fs = q->fs;
330 // D("fs %p si %p\n", fs, q->_si);
331 /* notify the parent scheduler that the queue is going away */
332 if (fs && fs->sched->fp->free_queue)
333 fs->sched->fp->free_queue(q);
336 if (flags & DN_DESTROY) {
338 dn_free_pkts(q->mq.head);
339 bzero(q, sizeof(*q)); // safety
341 dn_cfg.queue_count--;
346 q_delete_cb(void *q, void *arg)
348 int flags = (int)(uintptr_t)arg;
349 dn_delete_queue(q, flags);
350 return (flags & DN_DESTROY) ? DNHT_SCAN_DEL : 0;
354 * calls dn_delete_queue/q_delete_cb on all queues,
355 * which notifies the parent scheduler and possibly drains packets.
356 * flags & DN_DESTROY: drains queues and destroy qht;
359 qht_delete(struct dn_fsk *fs, int flags)
361 ND("fs %d start flags %d qht %p",
362 fs->fs.fs_nr, flags, fs->qht);
365 if (fs->fs.flags & DN_QHT_HASH) {
366 dn_ht_scan(fs->qht, q_delete_cb, (void *)(uintptr_t)flags);
367 if (flags & DN_DESTROY) {
368 dn_ht_free(fs->qht, 0);
372 dn_delete_queue((struct dn_queue *)(fs->qht), flags);
373 if (flags & DN_DESTROY)
379 * Find and possibly create the queue for a MULTIQUEUE scheduler.
380 * We never call it for !MULTIQUEUE (the queue is in the sch_inst).
383 ipdn_q_find(struct dn_fsk *fs, struct dn_sch_inst *si,
384 struct ipfw_flow_id *id)
386 struct dn_queue template;
391 if (fs->fs.flags & DN_QHT_HASH) {
392 struct ipfw_flow_id masked_id;
393 if (fs->qht == NULL) {
394 fs->qht = dn_ht_init(NULL, fs->fs.buckets,
395 offsetof(struct dn_queue, q_next),
396 q_hash, q_match, q_new);
401 flow_id_mask(&fs->fsk_mask, &masked_id);
402 return dn_ht_find(fs->qht, (uintptr_t)&masked_id,
403 DNHT_INSERT, &template);
406 fs->qht = q_new(0, 0, &template);
407 return (struct dn_queue *)fs->qht;
410 /*--- end of queue hash table ---*/
412 /*--- support functions for the sch_inst hashtable ----
414 * These are hashed by flow-id
417 si_hash(uintptr_t key, int flags, void *arg)
419 /* compute the hash slot from the flow id */
420 struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
421 &((struct dn_sch_inst *)key)->ni.fid :
422 (struct ipfw_flow_id *)key;
424 return flow_id_hash(id);
428 si_match(void *obj, uintptr_t key, int flags, void *arg)
430 struct dn_sch_inst *o = obj;
431 struct ipfw_flow_id *id2;
433 id2 = (flags & DNHT_KEY_IS_OBJ) ?
434 &((struct dn_sch_inst *)key)->ni.fid :
435 (struct ipfw_flow_id *)key;
436 return flow_id_cmp(&o->ni.fid, id2) == 0;
440 * create a new instance for the given 'key'
441 * Allocate memory for instance, delay line and scheduler private data.
444 si_new(uintptr_t key, int flags, void *arg)
446 struct dn_schk *s = arg;
447 struct dn_sch_inst *si;
448 int l = sizeof(*si) + s->fp->si_datalen;
450 si = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
454 /* Set length only for the part passed up to userland. */
455 set_oid(&si->ni.oid, DN_SCH_I, sizeof(struct dn_flow));
456 set_oid(&(si->dline.oid), DN_DELAY_LINE,
457 sizeof(struct delay_line));
458 /* mark si and dline as outside the event queue */
459 si->ni.oid.id = si->dline.oid.id = -1;
464 if (s->fp->new_sched && s->fp->new_sched(si)) {
465 D("new_sched error");
468 if (s->sch.flags & DN_HAVE_MASK)
469 si->ni.fid = *(struct ipfw_flow_id *)key;
476 bzero(si, sizeof(*si)); // safety
477 free(si, M_DUMMYNET);
483 * Callback from siht to delete all scheduler instances. Remove
484 * si and delay line from the system heap, destroy all queues.
485 * We assume that all flowset have been notified and do not
486 * point to us anymore.
489 si_destroy(void *_si, void *arg)
491 struct dn_sch_inst *si = _si;
492 struct dn_schk *s = si->sched;
493 struct delay_line *dl = &si->dline;
495 if (dl->oid.subtype) /* remove delay line from event heap */
496 heap_extract(&dn_cfg.evheap, dl);
497 dn_free_pkts(dl->mq.head); /* drain delay line */
498 if (si->kflags & DN_ACTIVE) /* remove si from event heap */
499 heap_extract(&dn_cfg.evheap, si);
500 if (s->fp->free_sched)
501 s->fp->free_sched(si);
502 bzero(si, sizeof(*si)); /* safety */
503 free(si, M_DUMMYNET);
505 return DNHT_SCAN_DEL;
509 * Find the scheduler instance for this packet. If we need to apply
510 * a mask, do on a local copy of the flow_id to preserve the original.
511 * Assume siht is always initialized if we have a mask.
514 ipdn_si_find(struct dn_schk *s, struct ipfw_flow_id *id)
517 if (s->sch.flags & DN_HAVE_MASK) {
518 struct ipfw_flow_id id_t = *id;
519 flow_id_mask(&s->sch.sched_mask, &id_t);
520 return dn_ht_find(s->siht, (uintptr_t)&id_t,
524 s->siht = si_new(0, 0, s);
525 return (struct dn_sch_inst *)s->siht;
528 /* callback to flush credit for the scheduler instance */
530 si_reset_credit(void *_si, void *arg)
532 struct dn_sch_inst *si = _si;
533 struct dn_link *p = &si->sched->link;
535 si->credit = p->burst + (dn_cfg.io_fast ? p->bandwidth : 0);
540 schk_reset_credit(struct dn_schk *s)
542 if (s->sch.flags & DN_HAVE_MASK)
543 dn_ht_scan(s->siht, si_reset_credit, NULL);
545 si_reset_credit(s->siht, NULL);
547 /*---- end of sch_inst hashtable ---------------------*/
549 /*-------------------------------------------------------
550 * flowset hash (fshash) support. Entries are hashed by fs_nr.
551 * New allocations are put in the fsunlinked list, from which
552 * they are removed when they point to a specific scheduler.
555 fsk_hash(uintptr_t key, int flags, void *arg)
557 uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
558 ((struct dn_fsk *)key)->fs.fs_nr;
560 return ( (i>>8)^(i>>4)^i );
564 fsk_match(void *obj, uintptr_t key, int flags, void *arg)
566 struct dn_fsk *fs = obj;
567 int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
568 ((struct dn_fsk *)key)->fs.fs_nr;
570 return (fs->fs.fs_nr == i);
574 fsk_new(uintptr_t key, int flags, void *arg)
578 fs = malloc(sizeof(*fs), M_DUMMYNET, M_NOWAIT | M_ZERO);
580 set_oid(&fs->fs.oid, DN_FS, sizeof(fs->fs));
582 fs->drain_bucket = 0;
583 SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
589 * detach flowset from its current scheduler. Flags as follows:
590 * DN_DETACH removes from the fsk_list
591 * DN_DESTROY deletes individual queues
592 * DN_DELETE_FS destroys the flowset (otherwise goes in unlinked).
595 fsk_detach(struct dn_fsk *fs, int flags)
597 if (flags & DN_DELETE_FS)
599 ND("fs %d from sched %d flags %s %s %s",
600 fs->fs.fs_nr, fs->fs.sched_nr,
601 (flags & DN_DELETE_FS) ? "DEL_FS":"",
602 (flags & DN_DESTROY) ? "DEL":"",
603 (flags & DN_DETACH) ? "DET":"");
604 if (flags & DN_DETACH) { /* detach from the list */
605 struct dn_fsk_head *h;
606 h = fs->sched ? &fs->sched->fsk_list : &dn_cfg.fsu;
607 SLIST_REMOVE(h, fs, dn_fsk, sch_chain);
609 /* Free the RED parameters, they will be recomputed on
610 * subsequent attach if needed.
613 free(fs->w_q_lookup, M_DUMMYNET);
614 fs->w_q_lookup = NULL;
615 qht_delete(fs, flags);
616 if (fs->sched && fs->sched->fp->free_fsk)
617 fs->sched->fp->free_fsk(fs);
619 if (flags & DN_DELETE_FS) {
620 bzero(fs, sizeof(fs)); /* safety */
621 free(fs, M_DUMMYNET);
624 SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
629 * Detach or destroy all flowsets in a list.
630 * flags specifies what to do:
631 * DN_DESTROY: flush all queues
632 * DN_DELETE_FS: DN_DESTROY + destroy flowset
633 * DN_DELETE_FS implies DN_DESTROY
636 fsk_detach_list(struct dn_fsk_head *h, int flags)
639 int n = 0; /* only for stats */
641 ND("head %p flags %x", h, flags);
642 while ((fs = SLIST_FIRST(h))) {
643 SLIST_REMOVE_HEAD(h, sch_chain);
645 fsk_detach(fs, flags);
647 ND("done %d flowsets", n);
651 * called on 'queue X delete' -- removes the flowset from fshash,
652 * deletes all queues for the flowset, and removes the flowset.
655 delete_fs(int i, int locked)
662 fs = dn_ht_find(dn_cfg.fshash, i, DNHT_REMOVE, NULL);
663 ND("fs %d found %p", i, fs);
665 fsk_detach(fs, DN_DETACH | DN_DELETE_FS);
674 /*----- end of flowset hashtable support -------------*/
676 /*------------------------------------------------------------
677 * Scheduler hash. When searching by index we pass sched_nr,
678 * otherwise we pass struct dn_sch * which is the first field in
679 * struct dn_schk so we can cast between the two. We use this trick
680 * because in the create phase (but it should be fixed).
683 schk_hash(uintptr_t key, int flags, void *_arg)
685 uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
686 ((struct dn_schk *)key)->sch.sched_nr;
687 return ( (i>>8)^(i>>4)^i );
691 schk_match(void *obj, uintptr_t key, int flags, void *_arg)
693 struct dn_schk *s = (struct dn_schk *)obj;
694 int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
695 ((struct dn_schk *)key)->sch.sched_nr;
696 return (s->sch.sched_nr == i);
700 * Create the entry and intialize with the sched hash if needed.
701 * Leave s->fp unset so we can tell whether a dn_ht_find() returns
702 * a new object or a previously existing one.
705 schk_new(uintptr_t key, int flags, void *arg)
707 struct schk_new_arg *a = arg;
709 int l = sizeof(*s) +a->fp->schk_datalen;
711 s = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
714 set_oid(&s->link.oid, DN_LINK, sizeof(s->link));
715 s->sch = *a->sch; // copy initial values
716 s->link.link_nr = s->sch.sched_nr;
717 SLIST_INIT(&s->fsk_list);
718 /* initialize the hash table or create the single instance */
719 s->fp = a->fp; /* si_new needs this */
721 if (s->sch.flags & DN_HAVE_MASK) {
722 s->siht = dn_ht_init(NULL, s->sch.buckets,
723 offsetof(struct dn_sch_inst, si_next),
724 si_hash, si_match, si_new);
725 if (s->siht == NULL) {
730 s->fp = NULL; /* mark as a new scheduler */
736 * Callback for sched delete. Notify all attached flowsets to
737 * detach from the scheduler, destroy the internal flowset, and
738 * all instances. The scheduler goes away too.
739 * arg is 0 (only detach flowsets and destroy instances)
740 * DN_DESTROY (detach & delete queues, delete schk)
741 * or DN_DELETE_FS (delete queues and flowsets, delete schk)
744 schk_delete_cb(void *obj, void *arg)
746 struct dn_schk *s = obj;
749 ND("sched %d arg %s%s",
751 a&DN_DESTROY ? "DEL ":"",
752 a&DN_DELETE_FS ? "DEL_FS":"");
754 fsk_detach_list(&s->fsk_list, arg ? DN_DESTROY : 0);
755 /* no more flowset pointing to us now */
756 if (s->sch.flags & DN_HAVE_MASK) {
757 dn_ht_scan(s->siht, si_destroy, NULL);
758 dn_ht_free(s->siht, 0);
760 si_destroy(s->siht, NULL);
762 free(s->profile, M_DUMMYNET);
768 bzero(s, sizeof(*s)); // safety
769 free(obj, M_DUMMYNET);
771 return DNHT_SCAN_DEL;
775 * called on a 'sched X delete' command. Deletes a single scheduler.
776 * This is done by removing from the schedhash, unlinking all
777 * flowsets and deleting their traffic.
784 s = dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
788 delete_fs(i + DN_MAX_ID, 1); /* first delete internal fs */
789 /* then detach flowsets, delete traffic */
790 schk_delete_cb(s, (void*)(uintptr_t)DN_DESTROY);
793 /*--- end of schk hashtable support ---*/
796 copy_obj(char **start, char *end, void *_o, const char *msg, int i)
798 struct dn_id *o = _o;
799 int have = end - *start;
801 if (have < o->len || o->len == 0 || o->type == 0) {
802 D("(WARN) type %d %s %d have %d need %d",
803 o->type, msg, i, have, o->len);
806 ND("type %d %s %d len %d", o->type, msg, i, o->len);
807 bcopy(_o, *start, o->len);
808 if (o->type == DN_LINK) {
809 /* Adjust burst parameter for link */
810 struct dn_link *l = (struct dn_link *)*start;
811 l->burst = div64(l->burst, 8 * hz);
812 l->delay = l->delay * 1000 / hz;
813 } else if (o->type == DN_SCH) {
814 /* Set id->id to the number of instances */
815 struct dn_schk *s = _o;
816 struct dn_id *id = (struct dn_id *)(*start);
817 id->id = (s->sch.flags & DN_HAVE_MASK) ?
818 dn_ht_entries(s->siht) : (s->siht ? 1 : 0);
824 /* Specific function to copy a queue.
825 * Copies only the user-visible part of a queue (which is in
826 * a struct dn_flow), and sets len accordingly.
829 copy_obj_q(char **start, char *end, void *_o, const char *msg, int i)
831 struct dn_id *o = _o;
832 int have = end - *start;
833 int len = sizeof(struct dn_flow); /* see above comment */
835 if (have < len || o->len == 0 || o->type != DN_QUEUE) {
836 D("ERROR type %d %s %d have %d need %d",
837 o->type, msg, i, have, len);
840 ND("type %d %s %d len %d", o->type, msg, i, len);
841 bcopy(_o, *start, len);
842 ((struct dn_id*)(*start))->len = len;
848 copy_q_cb(void *obj, void *arg)
850 struct dn_queue *q = obj;
851 struct copy_args *a = arg;
852 struct dn_flow *ni = (struct dn_flow *)(*a->start);
853 if (copy_obj_q(a->start, a->end, &q->ni, "queue", -1))
854 return DNHT_SCAN_END;
855 ni->oid.type = DN_FLOW; /* override the DN_QUEUE */
856 ni->oid.id = si_hash((uintptr_t)&ni->fid, 0, NULL);
861 copy_q(struct copy_args *a, struct dn_fsk *fs, int flags)
865 if (fs->fs.flags & DN_QHT_HASH)
866 dn_ht_scan(fs->qht, copy_q_cb, a);
868 copy_q_cb(fs->qht, a);
873 * This routine only copies the initial part of a profile ? XXX
876 copy_profile(struct copy_args *a, struct dn_profile *p)
878 int have = a->end - *a->start;
879 /* XXX here we check for max length */
880 int profile_len = sizeof(struct dn_profile) -
881 ED_MAX_SAMPLES_NO*sizeof(int);
885 if (have < profile_len) {
886 D("error have %d need %d", have, profile_len);
889 bcopy(p, *a->start, profile_len);
890 ((struct dn_id *)(*a->start))->len = profile_len;
891 *a->start += profile_len;
896 copy_flowset(struct copy_args *a, struct dn_fsk *fs, int flags)
898 struct dn_fs *ufs = (struct dn_fs *)(*a->start);
901 ND("flowset %d", fs->fs.fs_nr);
902 if (copy_obj(a->start, a->end, &fs->fs, "flowset", fs->fs.fs_nr))
903 return DNHT_SCAN_END;
904 ufs->oid.id = (fs->fs.flags & DN_QHT_HASH) ?
905 dn_ht_entries(fs->qht) : (fs->qht ? 1 : 0);
906 if (flags) { /* copy queues */
913 copy_si_cb(void *obj, void *arg)
915 struct dn_sch_inst *si = obj;
916 struct copy_args *a = arg;
917 struct dn_flow *ni = (struct dn_flow *)(*a->start);
918 if (copy_obj(a->start, a->end, &si->ni, "inst",
919 si->sched->sch.sched_nr))
920 return DNHT_SCAN_END;
921 ni->oid.type = DN_FLOW; /* override the DN_SCH_I */
922 ni->oid.id = si_hash((uintptr_t)si, DNHT_KEY_IS_OBJ, NULL);
927 copy_si(struct copy_args *a, struct dn_schk *s, int flags)
929 if (s->sch.flags & DN_HAVE_MASK)
930 dn_ht_scan(s->siht, copy_si_cb, a);
932 copy_si_cb(s->siht, a);
937 * compute a list of children of a scheduler and copy up
940 copy_fsk_list(struct copy_args *a, struct dn_schk *s, int flags)
946 int n = 0, space = sizeof(*o);
947 SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
948 if (fs->fs.fs_nr < DN_MAX_ID)
951 space += n * sizeof(uint32_t);
952 DX(3, "sched %d has %d flowsets", s->sch.sched_nr, n);
953 if (a->end - *(a->start) < space)
954 return DNHT_SCAN_END;
955 o = (struct dn_id *)(*(a->start));
959 p = (uint32_t *)(o+1);
960 SLIST_FOREACH(fs, &s->fsk_list, sch_chain)
961 if (fs->fs.fs_nr < DN_MAX_ID)
967 copy_data_helper(void *_o, void *_arg)
969 struct copy_args *a = _arg;
970 uint32_t *r = a->extra->r; /* start of first range */
971 uint32_t *lim; /* first invalid pointer */
974 lim = (uint32_t *)((char *)(a->extra) + a->extra->o.len);
976 if (a->type == DN_LINK || a->type == DN_SCH) {
977 /* pipe|sched show, we receive a dn_schk */
978 struct dn_schk *s = _o;
981 if (a->type == DN_SCH && n >= DN_MAX_ID)
982 return 0; /* not a scheduler */
983 if (a->type == DN_LINK && n <= DN_MAX_ID)
984 return 0; /* not a pipe */
986 /* see if the object is within one of our ranges */
987 for (;r < lim; r += 2) {
988 if (n < r[0] || n > r[1])
990 /* Found a valid entry, copy and we are done */
991 if (a->flags & DN_C_LINK) {
992 if (copy_obj(a->start, a->end,
993 &s->link, "link", n))
994 return DNHT_SCAN_END;
995 if (copy_profile(a, s->profile))
996 return DNHT_SCAN_END;
997 if (copy_flowset(a, s->fs, 0))
998 return DNHT_SCAN_END;
1000 if (a->flags & DN_C_SCH) {
1001 if (copy_obj(a->start, a->end,
1002 &s->sch, "sched", n))
1003 return DNHT_SCAN_END;
1004 /* list all attached flowsets */
1005 if (copy_fsk_list(a, s, 0))
1006 return DNHT_SCAN_END;
1008 if (a->flags & DN_C_FLOW)
1012 } else if (a->type == DN_FS) {
1013 /* queue show, skip internal flowsets */
1014 struct dn_fsk *fs = _o;
1019 /* see if the object is within one of our ranges */
1020 for (;r < lim; r += 2) {
1021 if (n < r[0] || n > r[1])
1023 if (copy_flowset(a, fs, 0))
1024 return DNHT_SCAN_END;
1026 break; /* we are done */
1032 static inline struct dn_schk *
1033 locate_scheduler(int i)
1035 return dn_ht_find(dn_cfg.schedhash, i, 0, NULL);
1039 * red parameters are in fixed point arithmetic.
1042 config_red(struct dn_fsk *fs)
1044 int64_t s, idle, weight, w0;
1047 fs->w_q = fs->fs.w_q;
1048 fs->max_p = fs->fs.max_p;
1050 /* Doing stuff that was in userland */
1051 i = fs->sched->link.bandwidth;
1053 hz * dn_cfg.red_avg_pkt_size * 8 * SCALE(1) / i;
1055 idle = div64((s * 3) , fs->w_q); /* s, fs->w_q scaled; idle not scaled */
1056 fs->lookup_step = div64(idle , dn_cfg.red_lookup_depth);
1057 /* fs->lookup_step not scaled, */
1058 if (!fs->lookup_step)
1059 fs->lookup_step = 1;
1060 w0 = weight = SCALE(1) - fs->w_q; //fs->w_q scaled
1062 for (t = fs->lookup_step; t > 1; --t)
1063 weight = SCALE_MUL(weight, w0);
1064 fs->lookup_weight = (int)(weight); // scaled
1066 /* Now doing stuff that was in kerneland */
1067 fs->min_th = SCALE(fs->fs.min_th);
1068 fs->max_th = SCALE(fs->fs.max_th);
1070 fs->c_1 = fs->max_p / (fs->fs.max_th - fs->fs.min_th);
1071 fs->c_2 = SCALE_MUL(fs->c_1, SCALE(fs->fs.min_th));
1073 if (fs->fs.flags & DN_IS_GENTLE_RED) {
1074 fs->c_3 = (SCALE(1) - fs->max_p) / fs->fs.max_th;
1075 fs->c_4 = SCALE(1) - 2 * fs->max_p;
1078 /* If the lookup table already exist, free and create it again. */
1079 if (fs->w_q_lookup) {
1080 free(fs->w_q_lookup, M_DUMMYNET);
1081 fs->w_q_lookup = NULL;
1083 if (dn_cfg.red_lookup_depth == 0) {
1084 printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1086 fs->fs.flags &= ~DN_IS_RED;
1087 fs->fs.flags &= ~DN_IS_GENTLE_RED;
1090 fs->lookup_depth = dn_cfg.red_lookup_depth;
1091 fs->w_q_lookup = (u_int *)malloc(fs->lookup_depth * sizeof(int),
1092 M_DUMMYNET, M_NOWAIT);
1093 if (fs->w_q_lookup == NULL) {
1094 printf("dummynet: sorry, cannot allocate red lookup table\n");
1095 fs->fs.flags &= ~DN_IS_RED;
1096 fs->fs.flags &= ~DN_IS_GENTLE_RED;
1100 /* Fill the lookup table with (1 - w_q)^x */
1101 fs->w_q_lookup[0] = SCALE(1) - fs->w_q;
1103 for (i = 1; i < fs->lookup_depth; i++)
1105 SCALE_MUL(fs->w_q_lookup[i - 1], fs->lookup_weight);
1107 if (dn_cfg.red_avg_pkt_size < 1)
1108 dn_cfg.red_avg_pkt_size = 512;
1109 fs->avg_pkt_size = dn_cfg.red_avg_pkt_size;
1110 if (dn_cfg.red_max_pkt_size < 1)
1111 dn_cfg.red_max_pkt_size = 1500;
1112 fs->max_pkt_size = dn_cfg.red_max_pkt_size;
1117 /* Scan all flowset attached to this scheduler and update red */
1119 update_red(struct dn_schk *s)
1122 SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1123 if (fs && (fs->fs.flags & DN_IS_RED))
1128 /* attach flowset to scheduler s, possibly requeue */
1130 fsk_attach(struct dn_fsk *fs, struct dn_schk *s)
1132 ND("remove fs %d from fsunlinked, link to sched %d",
1133 fs->fs.fs_nr, s->sch.sched_nr);
1134 SLIST_REMOVE(&dn_cfg.fsu, fs, dn_fsk, sch_chain);
1136 SLIST_INSERT_HEAD(&s->fsk_list, fs, sch_chain);
1139 /* XXX compute fsk_mask */
1140 fs->fsk_mask = fs->fs.flow_mask;
1141 if (fs->sched->sch.flags & DN_HAVE_MASK)
1142 flow_id_or(&fs->sched->sch.sched_mask, &fs->fsk_mask);
1145 * we must drain qht according to the old
1146 * type, and reinsert according to the new one.
1147 * The requeue is complex -- in general we need to
1148 * reclassify every single packet.
1149 * For the time being, let's hope qht is never set
1150 * when we reach this point.
1152 D("XXX TODO requeue from fs %d to sch %d",
1153 fs->fs.fs_nr, s->sch.sched_nr);
1156 /* set the new type for qht */
1157 if (nonzero_mask(&fs->fsk_mask))
1158 fs->fs.flags |= DN_QHT_HASH;
1160 fs->fs.flags &= ~DN_QHT_HASH;
1162 /* XXX config_red() can fail... */
1163 if (fs->fs.flags & DN_IS_RED)
1167 /* update all flowsets which may refer to this scheduler */
1169 update_fs(struct dn_schk *s)
1171 struct dn_fsk *fs, *tmp;
1173 SLIST_FOREACH_SAFE(fs, &dn_cfg.fsu, sch_chain, tmp) {
1174 if (s->sch.sched_nr != fs->fs.sched_nr) {
1175 D("fs %d for sch %d not %d still unlinked",
1176 fs->fs.fs_nr, fs->fs.sched_nr,
1185 * Configuration -- to preserve backward compatibility we use
1186 * the following scheme (N is 65536)
1187 * NUMBER SCHED LINK FLOWSET
1188 * 1 .. N-1 (1)WFQ (2)WFQ (3)queue
1189 * N+1 .. 2N-1 (4)FIFO (5)FIFO (6)FIFO for sched 1..N-1
1190 * 2N+1 .. 3N-1 -- -- (7)FIFO for sched N+1..2N-1
1192 * "pipe i config" configures #1, #2 and #3
1193 * "sched i config" configures #1 and possibly #6
1194 * "queue i config" configures #3
1195 * #1 is configured with 'pipe i config' or 'sched i config'
1196 * #2 is configured with 'pipe i config', and created if not
1197 * existing with 'sched i config'
1198 * #3 is configured with 'queue i config'
1199 * #4 is automatically configured after #1, can only be FIFO
1200 * #5 is automatically configured after #2
1201 * #6 is automatically created when #1 is !MULTIQUEUE,
1202 * and can be updated.
1203 * #7 is automatically configured after #2
1207 * configure a link (and its FIFO instance)
1210 config_link(struct dn_link *p, struct dn_id *arg)
1214 if (p->oid.len != sizeof(*p)) {
1215 D("invalid pipe len %d", p->oid.len);
1219 if (i <= 0 || i >= DN_MAX_ID)
1222 * The config program passes parameters as follows:
1223 * bw = bits/second (0 means no limits),
1224 * delay = ms, must be translated into ticks.
1225 * qsize = slots/bytes
1228 p->delay = (p->delay * hz) / 1000;
1229 /* Scale burst size: bytes -> bits * hz */
1233 /* do it twice, base link and FIFO link */
1234 for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1235 struct dn_schk *s = locate_scheduler(i);
1238 D("sched %d not found", i);
1241 /* remove profile if exists */
1243 free(s->profile, M_DUMMYNET);
1246 /* copy all parameters */
1247 s->link.oid = p->oid;
1248 s->link.link_nr = i;
1249 s->link.delay = p->delay;
1250 if (s->link.bandwidth != p->bandwidth) {
1251 /* XXX bandwidth changes, need to update red params */
1252 s->link.bandwidth = p->bandwidth;
1255 s->link.burst = p->burst;
1256 schk_reset_credit(s);
1264 * configure a flowset. Can be called from inside with locked=1,
1266 static struct dn_fsk *
1267 config_fs(struct dn_fs *nfs, struct dn_id *arg, int locked)
1272 if (nfs->oid.len != sizeof(*nfs)) {
1273 D("invalid flowset len %d", nfs->oid.len);
1277 if (i <= 0 || i >= 3*DN_MAX_ID)
1279 ND("flowset %d", i);
1280 /* XXX other sanity checks */
1281 if (nfs->flags & DN_QSIZE_BYTES) {
1282 ipdn_bound_var(&nfs->qsize, 16384,
1283 1500, dn_cfg.byte_limit, NULL); // "queue byte size");
1285 ipdn_bound_var(&nfs->qsize, 50,
1286 1, dn_cfg.slot_limit, NULL); // "queue slot size");
1288 if (nfs->flags & DN_HAVE_MASK) {
1289 /* make sure we have some buckets */
1290 ipdn_bound_var(&nfs->buckets, dn_cfg.hash_size,
1291 1, dn_cfg.max_hash_size, "flowset buckets");
1293 nfs->buckets = 1; /* we only need 1 */
1297 do { /* exit with break when done */
1299 int flags = nfs->sched_nr ? DNHT_INSERT : 0;
1301 int oldc = dn_cfg.fsk_count;
1302 fs = dn_ht_find(dn_cfg.fshash, i, flags, NULL);
1304 D("missing sched for flowset %d", i);
1307 /* grab some defaults from the existing one */
1308 if (nfs->sched_nr == 0) /* reuse */
1309 nfs->sched_nr = fs->fs.sched_nr;
1310 for (j = 0; j < sizeof(nfs->par)/sizeof(nfs->par[0]); j++) {
1311 if (nfs->par[j] == -1) /* reuse */
1312 nfs->par[j] = fs->fs.par[j];
1314 if (bcmp(&fs->fs, nfs, sizeof(*nfs)) == 0) {
1315 ND("flowset %d unchanged", i);
1316 break; /* no change, nothing to do */
1318 if (oldc != dn_cfg.fsk_count) /* new item */
1320 s = locate_scheduler(nfs->sched_nr);
1321 /* detach from old scheduler if needed, preserving
1322 * queues if we need to reattach. Then update the
1323 * configuration, and possibly attach to the new sched.
1325 DX(2, "fs %d changed sched %d@%p to %d@%p",
1327 fs->fs.sched_nr, fs->sched, nfs->sched_nr, s);
1329 int flags = s ? DN_DETACH : (DN_DETACH | DN_DESTROY);
1330 flags |= DN_DESTROY; /* XXX temporary */
1331 fsk_detach(fs, flags);
1333 fs->fs = *nfs; /* copy configuration */
1343 * config/reconfig a scheduler and its FIFO variant.
1344 * For !MULTIQUEUE schedulers, also set up the flowset.
1346 * On reconfigurations (detected because s->fp is set),
1347 * detach existing flowsets preserving traffic, preserve link,
1348 * and delete the old scheduler creating a new one.
1351 config_sched(struct dn_sch *_nsch, struct dn_id *arg)
1354 struct schk_new_arg a; /* argument for schk_new */
1356 struct dn_link p; /* copy of oldlink */
1357 struct dn_profile *pf = NULL; /* copy of old link profile */
1358 /* Used to preserv mask parameter */
1359 struct ipfw_flow_id new_mask;
1360 int new_buckets = 0;
1366 if (a.sch->oid.len != sizeof(*a.sch)) {
1367 D("bad sched len %d", a.sch->oid.len);
1370 i = a.sch->sched_nr;
1371 if (i <= 0 || i >= DN_MAX_ID)
1373 /* make sure we have some buckets */
1374 if (a.sch->flags & DN_HAVE_MASK)
1375 ipdn_bound_var(&a.sch->buckets, dn_cfg.hash_size,
1376 1, dn_cfg.max_hash_size, "sched buckets");
1377 /* XXX other sanity checks */
1378 bzero(&p, sizeof(p));
1380 pipe_cmd = a.sch->flags & DN_PIPE_CMD;
1381 a.sch->flags &= ~DN_PIPE_CMD; //XXX do it even if is not set?
1383 /* Copy mask parameter */
1384 new_mask = a.sch->sched_mask;
1385 new_buckets = a.sch->buckets;
1386 new_flags = a.sch->flags;
1389 again: /* run twice, for wfq and fifo */
1391 * lookup the type. If not supplied, use the previous one
1392 * or default to WF2Q+. Otherwise, return an error.
1395 a.fp = find_sched_type(a.sch->oid.subtype, a.sch->name);
1397 /* found. Lookup or create entry */
1398 s = dn_ht_find(dn_cfg.schedhash, i, DNHT_INSERT, &a);
1399 } else if (a.sch->oid.subtype == 0 && !a.sch->name[0]) {
1400 /* No type. search existing s* or retry with WF2Q+ */
1401 s = dn_ht_find(dn_cfg.schedhash, i, 0, &a);
1404 /* Scheduler exists, skip to FIFO scheduler
1405 * if command was pipe config...
1410 /* New scheduler, create a wf2q+ with no mask
1411 * if command was pipe config...
1414 /* clear mask parameter */
1415 bzero(&a.sch->sched_mask, sizeof(new_mask));
1417 a.sch->flags &= ~DN_HAVE_MASK;
1419 a.sch->oid.subtype = DN_SCHED_WF2QP;
1423 D("invalid scheduler type %d %s",
1424 a.sch->oid.subtype, a.sch->name);
1428 /* normalize name and subtype */
1429 a.sch->oid.subtype = a.fp->type;
1430 bzero(a.sch->name, sizeof(a.sch->name));
1431 strlcpy(a.sch->name, a.fp->name, sizeof(a.sch->name));
1433 D("cannot allocate scheduler %d", i);
1436 /* restore existing link if any */
1439 if (!pf || pf->link_nr != p.link_nr) { /* no saved value */
1440 s->profile = NULL; /* XXX maybe not needed */
1442 s->profile = malloc(sizeof(struct dn_profile),
1443 M_DUMMYNET, M_NOWAIT | M_ZERO);
1444 if (s->profile == NULL) {
1445 D("cannot allocate profile");
1448 bcopy(pf, s->profile, sizeof(*pf));
1452 if (s->fp == NULL) {
1453 DX(2, "sched %d new type %s", i, a.fp->name);
1454 } else if (s->fp != a.fp ||
1455 bcmp(a.sch, &s->sch, sizeof(*a.sch)) ) {
1456 /* already existing. */
1457 DX(2, "sched %d type changed from %s to %s",
1458 i, s->fp->name, a.fp->name);
1459 DX(4, " type/sub %d/%d -> %d/%d",
1460 s->sch.oid.type, s->sch.oid.subtype,
1461 a.sch->oid.type, a.sch->oid.subtype);
1462 if (s->link.link_nr == 0)
1463 D("XXX WARNING link 0 for sched %d", i);
1464 p = s->link; /* preserve link */
1465 if (s->profile) {/* preserve profile */
1467 pf = malloc(sizeof(*pf),
1468 M_DUMMYNET, M_NOWAIT | M_ZERO);
1469 if (pf) /* XXX should issue a warning otherwise */
1470 bcopy(s->profile, pf, sizeof(*pf));
1472 /* remove from the hash */
1473 dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
1474 /* Detach flowsets, preserve queues. */
1475 // schk_delete_cb(s, NULL);
1476 // XXX temporarily, kill queues
1477 schk_delete_cb(s, (void *)DN_DESTROY);
1480 DX(4, "sched %d unchanged type %s", i, a.fp->name);
1482 /* complete initialization */
1486 // XXX schk_reset_credit(s);
1487 /* create the internal flowset if needed,
1488 * trying to reuse existing ones if available
1490 if (!(s->fp->flags & DN_MULTIQUEUE) && !s->fs) {
1491 s->fs = dn_ht_find(dn_cfg.fshash, i, 0, NULL);
1494 bzero(&fs, sizeof(fs));
1495 set_oid(&fs.oid, DN_FS, sizeof(fs));
1496 fs.fs_nr = i + DN_MAX_ID;
1498 s->fs = config_fs(&fs, NULL, 1 /* locked */);
1501 schk_delete_cb(s, (void *)DN_DESTROY);
1502 D("error creating internal fs for %d", i);
1506 /* call init function after the flowset is created */
1511 if (i < DN_MAX_ID) { /* now configure the FIFO instance */
1514 /* Restore mask parameter for FIFO */
1515 a.sch->sched_mask = new_mask;
1516 a.sch->buckets = new_buckets;
1517 a.sch->flags = new_flags;
1519 /* sched config shouldn't modify the FIFO scheduler */
1520 if (dn_ht_find(dn_cfg.schedhash, i, 0, &a) != NULL) {
1521 /* FIFO already exist, don't touch it */
1522 err = 0; /* and this is not an error */
1526 a.sch->sched_nr = i;
1527 a.sch->oid.subtype = DN_SCHED_FIFO;
1528 bzero(a.sch->name, sizeof(a.sch->name));
1535 free(pf, M_DUMMYNET);
1540 * attach a profile to a link
1543 config_profile(struct dn_profile *pf, struct dn_id *arg)
1546 int i, olen, err = 0;
1548 if (pf->oid.len < sizeof(*pf)) {
1549 D("short profile len %d", pf->oid.len);
1553 if (i <= 0 || i >= DN_MAX_ID)
1555 /* XXX other sanity checks */
1557 for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1558 s = locate_scheduler(i);
1566 * If we had a profile and the new one does not fit,
1567 * or it is deleted, then we need to free memory.
1569 if (s->profile && (pf->samples_no == 0 ||
1570 s->profile->oid.len < pf->oid.len)) {
1571 free(s->profile, M_DUMMYNET);
1574 if (pf->samples_no == 0)
1577 * new profile, possibly allocate memory
1580 if (s->profile == NULL)
1581 s->profile = malloc(pf->oid.len,
1582 M_DUMMYNET, M_NOWAIT | M_ZERO);
1583 if (s->profile == NULL) {
1584 D("no memory for profile %d", i);
1588 /* preserve larger length XXX double check */
1589 olen = s->profile->oid.len;
1590 if (olen < pf->oid.len)
1592 bcopy(pf, s->profile, pf->oid.len);
1593 s->profile->oid.len = olen;
1600 * Delete all objects:
1603 dummynet_flush(void)
1606 /* delete all schedulers and related links/queues/flowsets */
1607 dn_ht_scan(dn_cfg.schedhash, schk_delete_cb,
1608 (void *)(uintptr_t)DN_DELETE_FS);
1609 /* delete all remaining (unlinked) flowsets */
1610 DX(4, "still %d unlinked fs", dn_cfg.fsk_count);
1611 dn_ht_free(dn_cfg.fshash, DNHT_REMOVE);
1612 fsk_detach_list(&dn_cfg.fsu, DN_DELETE_FS);
1613 /* Reinitialize system heap... */
1614 heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
1618 * Main handler for configuration. We are guaranteed to be called
1619 * with an oid which is at least a dn_id.
1620 * - the first object is the command (config, delete, flush, ...)
1621 * - config_link must be issued after the corresponding config_sched
1622 * - parameters (DN_TXT) for an object must preceed the object
1623 * processed on a config_sched.
1626 do_config(void *p, int l)
1628 struct dn_id *next, *o;
1629 int err = 0, err2 = 0;
1630 struct dn_id *arg = NULL;
1634 if (o->id != DN_API_VERSION) {
1635 D("invalid api version got %d need %d",
1636 o->id, DN_API_VERSION);
1639 for (; l >= sizeof(*o); o = next) {
1640 struct dn_id *prev = arg;
1641 if (o->len < sizeof(*o) || l < o->len) {
1642 D("bad len o->len %d len %d", o->len, l);
1647 next = (struct dn_id *)((char *)o + o->len);
1651 D("cmd %d not implemented", o->type);
1654 #ifdef EMULATE_SYSCTL
1655 /* sysctl emulation.
1656 * if we recognize the command, jump to the correct
1657 * handler and return
1660 err = kesysctl_emu_set(p, l);
1664 case DN_CMD_CONFIG: /* simply a header */
1668 /* the argument is in the first uintptr_t after o */
1669 a = (uintptr_t *)(o+1);
1670 if (o->len < sizeof(*o) + sizeof(*a)) {
1674 switch (o->subtype) {
1676 /* delete base and derived schedulers */
1678 err = delete_schk(*a);
1679 err2 = delete_schk(*a + DN_MAX_ID);
1686 D("invalid delete type %d",
1692 err = (*a <1 || *a >= DN_MAX_ID) ?
1693 EINVAL : delete_fs(*a, 0) ;
1703 case DN_TEXT: /* store argument the next block */
1708 err = config_link((struct dn_link *)o, arg);
1711 err = config_profile((struct dn_profile *)o, arg);
1714 err = config_sched((struct dn_sch *)o, arg);
1717 err = (NULL==config_fs((struct dn_fs *)o, arg, 0));
1729 compute_space(struct dn_id *cmd, struct copy_args *a)
1731 int x = 0, need = 0;
1732 int profile_size = sizeof(struct dn_profile) -
1733 ED_MAX_SAMPLES_NO*sizeof(int);
1735 /* NOTE about compute space:
1736 * NP = dn_cfg.schk_count
1737 * NSI = dn_cfg.si_count
1738 * NF = dn_cfg.fsk_count
1739 * NQ = dn_cfg.queue_count
1741 * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
1742 * link, scheduler template, flowset
1743 * integrated in scheduler and header
1745 * (NSI)*(dn_flow) all scheduler instance (includes
1746 * the queue instance)
1748 * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
1749 * link, scheduler template, flowset
1750 * integrated in scheduler and header
1752 * (NSI * dn_flow) all scheduler instances
1753 * (NF * sizeof(uint_32)) space for flowset list linked to scheduler
1754 * (NQ * dn_queue) all queue [XXXfor now not listed]
1756 * (NF * dn_fs) all flowset
1757 * (NQ * dn_queue) all queues
1759 switch (cmd->subtype) {
1762 /* XXX where do LINK and SCH differ ? */
1763 /* 'ipfw sched show' could list all queues associated to
1764 * a scheduler. This feature for now is disabled
1766 case DN_LINK: /* pipe show */
1767 x = DN_C_LINK | DN_C_SCH | DN_C_FLOW;
1768 need += dn_cfg.schk_count *
1769 (sizeof(struct dn_fs) + profile_size) / 2;
1770 need += dn_cfg.fsk_count * sizeof(uint32_t);
1772 case DN_SCH: /* sched show */
1773 need += dn_cfg.schk_count *
1774 (sizeof(struct dn_fs) + profile_size) / 2;
1775 need += dn_cfg.fsk_count * sizeof(uint32_t);
1776 x = DN_C_SCH | DN_C_LINK | DN_C_FLOW;
1778 case DN_FS: /* queue show */
1779 x = DN_C_FS | DN_C_QUEUE;
1781 case DN_GET_COMPAT: /* compatibility mode */
1782 need = dn_compat_calc_size();
1787 need += dn_cfg.schk_count * sizeof(struct dn_sch) / 2;
1788 /* NOT also, each fs might be attached to a sched */
1789 need += dn_cfg.schk_count * sizeof(struct dn_id) / 2;
1792 need += dn_cfg.fsk_count * sizeof(struct dn_fs);
1793 if (x & DN_C_LINK) {
1794 need += dn_cfg.schk_count * sizeof(struct dn_link) / 2;
1797 * When exporting a queue to userland, only pass up the
1798 * struct dn_flow, which is the only visible part.
1802 need += dn_cfg.queue_count * sizeof(struct dn_flow);
1804 need += dn_cfg.si_count * (sizeof(struct dn_flow));
1809 * If compat != NULL dummynet_get is called in compatibility mode.
1810 * *compat will be the pointer to the buffer to pass to ipfw
1813 dummynet_get(struct sockopt *sopt, void **compat)
1815 int have, i, need, error;
1816 char *start = NULL, *buf;
1817 size_t sopt_valsize;
1820 struct copy_range r;
1821 int l = sizeof(struct dn_id);
1823 bzero(&a, sizeof(a));
1824 bzero(&r, sizeof(r));
1826 /* save and restore original sopt_valsize around copyin */
1827 sopt_valsize = sopt->sopt_valsize;
1832 /* copy at least an oid, and possibly a full object */
1833 error = sooptcopyin(sopt, cmd, sizeof(r), sizeof(*cmd));
1834 sopt->sopt_valsize = sopt_valsize;
1838 #ifdef EMULATE_SYSCTL
1839 /* sysctl emulation. */
1840 if (cmd->type == DN_SYSCTL_GET)
1841 return kesysctl_emu_get(sopt);
1843 if (l > sizeof(r)) {
1844 /* request larger than default, allocate buffer */
1845 cmd = malloc(l, M_DUMMYNET, M_WAITOK);
1846 error = sooptcopyin(sopt, cmd, l, l);
1847 sopt->sopt_valsize = sopt_valsize;
1851 } else { /* compatibility */
1853 cmd->type = DN_CMD_GET;
1854 cmd->len = sizeof(struct dn_id);
1855 cmd->subtype = DN_GET_COMPAT;
1856 // cmd->id = sopt_valsize;
1857 D("compatibility mode");
1859 a.extra = (struct copy_range *)cmd;
1860 if (cmd->len == sizeof(*cmd)) { /* no range, create a default */
1861 uint32_t *rp = (uint32_t *)(cmd + 1);
1862 cmd->len += 2* sizeof(uint32_t);
1864 rp[1] = DN_MAX_ID - 1;
1865 if (cmd->subtype == DN_LINK) {
1870 /* Count space (under lock) and allocate (outside lock).
1871 * Exit with lock held if we manage to get enough buffer.
1872 * Try a few times then give up.
1874 for (have = 0, i = 0; i < 10; i++) {
1876 need = compute_space(cmd, &a);
1878 /* if there is a range, ignore value from compute_space() */
1879 if (l > sizeof(*cmd))
1880 need = sopt_valsize - sizeof(*cmd);
1887 need += sizeof(*cmd);
1894 free(start, M_DUMMYNET);
1896 if (need > sopt_valsize)
1900 start = malloc(have, M_DUMMYNET, M_WAITOK | M_ZERO);
1903 if (start == NULL) {
1908 error = sooptcopyout(sopt, cmd, sizeof(*cmd));
1912 ND("have %d:%d sched %d, %d:%d links %d, %d:%d flowsets %d, "
1913 "%d:%d si %d, %d:%d queues %d",
1914 dn_cfg.schk_count, sizeof(struct dn_sch), DN_SCH,
1915 dn_cfg.schk_count, sizeof(struct dn_link), DN_LINK,
1916 dn_cfg.fsk_count, sizeof(struct dn_fs), DN_FS,
1917 dn_cfg.si_count, sizeof(struct dn_flow), DN_SCH_I,
1918 dn_cfg.queue_count, sizeof(struct dn_queue), DN_QUEUE);
1919 sopt->sopt_valsize = sopt_valsize;
1920 a.type = cmd->subtype;
1922 if (compat == NULL) {
1923 bcopy(cmd, start, sizeof(*cmd));
1924 ((struct dn_id*)(start))->len = sizeof(struct dn_id);
1925 buf = start + sizeof(*cmd);
1929 a.end = start + have;
1930 /* start copying other objects */
1932 a.type = DN_COMPAT_PIPE;
1933 dn_ht_scan(dn_cfg.schedhash, copy_data_helper_compat, &a);
1934 a.type = DN_COMPAT_QUEUE;
1935 dn_ht_scan(dn_cfg.fshash, copy_data_helper_compat, &a);
1936 } else if (a.type == DN_FS) {
1937 dn_ht_scan(dn_cfg.fshash, copy_data_helper, &a);
1939 dn_ht_scan(dn_cfg.schedhash, copy_data_helper, &a);
1945 sopt->sopt_valsize = buf - start;
1946 /* free() is done by ip_dummynet_compat() */
1947 start = NULL; //XXX hack
1949 error = sooptcopyout(sopt, start, buf - start);
1952 if (cmd && cmd != &r.o)
1953 free(cmd, M_DUMMYNET);
1955 free(start, M_DUMMYNET);
1959 /* Callback called on scheduler instance to delete it if idle */
1961 drain_scheduler_cb(void *_si, void *arg)
1963 struct dn_sch_inst *si = _si;
1965 if ((si->kflags & DN_ACTIVE) || si->dline.mq.head != NULL)
1968 if (si->sched->fp->flags & DN_MULTIQUEUE) {
1969 if (si->q_count == 0)
1970 return si_destroy(si, NULL);
1973 } else { /* !DN_MULTIQUEUE */
1974 if ((si+1)->ni.length == 0)
1975 return si_destroy(si, NULL);
1979 return 0; /* unreachable */
1982 /* Callback called on scheduler to check if it has instances */
1984 drain_scheduler_sch_cb(void *_s, void *arg)
1986 struct dn_schk *s = _s;
1988 if (s->sch.flags & DN_HAVE_MASK) {
1989 dn_ht_scan_bucket(s->siht, &s->drain_bucket,
1990 drain_scheduler_cb, NULL);
1994 if (drain_scheduler_cb(s->siht, NULL) == DNHT_SCAN_DEL)
2001 /* Called every tick, try to delete a 'bucket' of scheduler */
2003 dn_drain_scheduler(void)
2005 dn_ht_scan_bucket(dn_cfg.schedhash, &dn_cfg.drain_sch,
2006 drain_scheduler_sch_cb, NULL);
2010 /* Callback called on queue to delete if it is idle */
2012 drain_queue_cb(void *_q, void *arg)
2014 struct dn_queue *q = _q;
2016 if (q->ni.length == 0) {
2017 dn_delete_queue(q, DN_DESTROY);
2018 return DNHT_SCAN_DEL; /* queue is deleted */
2021 return 0; /* queue isn't deleted */
2024 /* Callback called on flowset used to check if it has queues */
2026 drain_queue_fs_cb(void *_fs, void *arg)
2028 struct dn_fsk *fs = _fs;
2030 if (fs->fs.flags & DN_QHT_HASH) {
2031 /* Flowset has a hash table for queues */
2032 dn_ht_scan_bucket(fs->qht, &fs->drain_bucket,
2033 drain_queue_cb, NULL);
2036 /* No hash table for this flowset, null the pointer
2037 * if the queue is deleted
2040 if (drain_queue_cb(fs->qht, NULL) == DNHT_SCAN_DEL)
2047 /* Called every tick, try to delete a 'bucket' of queue */
2049 dn_drain_queue(void)
2051 /* scan a bucket of flowset */
2052 dn_ht_scan_bucket(dn_cfg.fshash, &dn_cfg.drain_fs,
2053 drain_queue_fs_cb, NULL);
2058 * Handler for the various dummynet socket options
2061 ip_dn_ctl(struct sockopt *sopt)
2066 error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2070 /* Disallow sets in really-really secure mode. */
2071 if (sopt->sopt_dir == SOPT_SET) {
2072 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2077 switch (sopt->sopt_name) {
2079 D("dummynet: unknown option %d", sopt->sopt_name);
2083 case IP_DUMMYNET_FLUSH:
2084 case IP_DUMMYNET_CONFIGURE:
2085 case IP_DUMMYNET_DEL: /* remove a pipe or queue */
2086 case IP_DUMMYNET_GET:
2087 D("dummynet: compat option %d", sopt->sopt_name);
2088 error = ip_dummynet_compat(sopt);
2092 if (sopt->sopt_dir == SOPT_GET) {
2093 error = dummynet_get(sopt, NULL);
2096 l = sopt->sopt_valsize;
2097 if (l < sizeof(struct dn_id) || l > 12000) {
2098 D("argument len %d invalid", l);
2101 p = malloc(l, M_TEMP, M_WAITOK); // XXX can it fail ?
2102 error = sooptcopyin(sopt, p, l, l);
2105 error = do_config(p, l);
2119 if (dn_cfg.init_done)
2121 printf("DUMMYNET %p with IPv6 initialized (100409)\n", curvnet);
2122 dn_cfg.init_done = 1;
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 = 65536; /* 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);
2160 TASK_INIT(&dn_task, 0, dummynet_task, curvnet);
2161 dn_tq = taskqueue_create("dummynet", M_WAITOK,
2162 taskqueue_thread_enqueue, &dn_tq);
2163 taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2165 callout_init(&dn_timeout, CALLOUT_MPSAFE);
2166 callout_reset(&dn_timeout, 1, dummynet, NULL);
2168 /* Initialize curr_time adjustment mechanics. */
2169 getmicrouptime(&dn_cfg.prev_t);
2173 ip_dn_destroy(int last)
2175 callout_drain(&dn_timeout);
2179 ND("removing last instance\n");
2180 ip_dn_ctl_ptr = NULL;
2181 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);
2197 dummynet_modevent(module_t mod, int type, void *data)
2200 if (type == MOD_LOAD) {
2202 printf("DUMMYNET already loaded\n");
2206 ip_dn_ctl_ptr = ip_dn_ctl;
2207 ip_dn_io_ptr = dummynet_io;
2209 } else if (type == MOD_UNLOAD) {
2210 ip_dn_destroy(1 /* last */);
2216 /* modevent helpers for the modules */
2218 load_dn_sched(struct dn_alg *d)
2223 return 1; /* error */
2224 ip_dn_init(); /* just in case, we need the lock */
2226 /* Check that mandatory funcs exists */
2227 if (d->enqueue == NULL || d->dequeue == NULL) {
2228 D("missing enqueue or dequeue for %s", d->name);
2232 /* Search if scheduler already exists */
2234 SLIST_FOREACH(s, &dn_cfg.schedlist, next) {
2235 if (strcmp(s->name, d->name) == 0) {
2236 D("%s already loaded", d->name);
2237 break; /* scheduler already exists */
2241 SLIST_INSERT_HEAD(&dn_cfg.schedlist, d, next);
2243 D("dn_sched %s %sloaded", d->name, s ? "not ":"");
2248 unload_dn_sched(struct dn_alg *s)
2250 struct dn_alg *tmp, *r;
2253 ND("called for %s", s->name);
2256 SLIST_FOREACH_SAFE(r, &dn_cfg.schedlist, next, tmp) {
2257 if (strcmp(s->name, r->name) != 0)
2259 ND("ref_count = %d", r->ref_count);
2260 err = (r->ref_count != 0) ? EBUSY : 0;
2262 SLIST_REMOVE(&dn_cfg.schedlist, r, dn_alg, next);
2266 D("dn_sched %s %sunloaded", s->name, err ? "not ":"");
2271 dn_sched_modevent(module_t mod, int cmd, void *arg)
2273 struct dn_alg *sch = arg;
2275 if (cmd == MOD_LOAD)
2276 return load_dn_sched(sch);
2277 else if (cmd == MOD_UNLOAD)
2278 return unload_dn_sched(sch);
2283 static moduledata_t dummynet_mod = {
2284 "dummynet", dummynet_modevent, NULL
2287 #define DN_SI_SUB SI_SUB_PROTO_IFATTACHDOMAIN
2288 #define DN_MODEV_ORD (SI_ORDER_ANY - 128) /* after ipfw */
2289 DECLARE_MODULE(dummynet, dummynet_mod, DN_SI_SUB, DN_MODEV_ORD);
2290 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
2291 MODULE_VERSION(dummynet, 3);
2294 * Starting up. Done in order after dummynet_modevent() has been called.
2295 * VNET_SYSINIT is also called for each existing vnet and each new vnet.
2297 //VNET_SYSINIT(vnet_dn_init, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_init, NULL);
2300 * Shutdown handlers up shop. These are done in REVERSE ORDER, but still
2301 * after dummynet_modevent() has been called. Not called on reboot.
2302 * VNET_SYSUNINIT is also called for each exiting vnet as it exits.
2303 * or when the module is unloaded.
2305 //VNET_SYSUNINIT(vnet_dn_uninit, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_destroy, NULL);