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: user/luigi/ipfw3-head/sys/netinet/ipfw/ip_dummynet.c 203340 2010-02-01 12:06:37Z luigi $");
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;
79 /* dummynet and ipfw_tick can't be static in windows */
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 && !strcmp(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;
115 } else if (oldv > hi) {
121 printf("%s %s to %d (was %d)\n", op, msg, *v, oldv);
125 /*---- flow_id mask, hash and compare functions ---*/
127 * The flow_id includes the 5-tuple, the queue/pipe number
128 * which we store in the extra area in host order,
129 * and for ipv6 also the flow_id6.
130 * XXX see if we want the tos byte (can store in 'flags')
132 static struct ipfw_flow_id *
133 flow_id_mask(struct ipfw_flow_id *mask, struct ipfw_flow_id *id)
135 int is_v6 = IS_IP6_FLOW_ID(id);
137 id->dst_port &= mask->dst_port;
138 id->src_port &= mask->src_port;
139 id->proto &= mask->proto;
140 id->extra &= mask->extra;
142 APPLY_MASK(&id->dst_ip6, &mask->dst_ip6);
143 APPLY_MASK(&id->src_ip6, &mask->src_ip6);
144 id->flow_id6 &= mask->flow_id6;
146 id->dst_ip &= mask->dst_ip;
147 id->src_ip &= mask->src_ip;
152 /* computes an OR of two masks, result in dst and also returned */
153 static struct ipfw_flow_id *
154 flow_id_or(struct ipfw_flow_id *src, struct ipfw_flow_id *dst)
156 int is_v6 = IS_IP6_FLOW_ID(dst);
158 dst->dst_port |= src->dst_port;
159 dst->src_port |= src->src_port;
160 dst->proto |= src->proto;
161 dst->extra |= src->extra;
163 #define OR_MASK(_d, _s) \
164 (_d)->__u6_addr.__u6_addr32[0] |= (_s)->__u6_addr.__u6_addr32[0]; \
165 (_d)->__u6_addr.__u6_addr32[1] |= (_s)->__u6_addr.__u6_addr32[1]; \
166 (_d)->__u6_addr.__u6_addr32[2] |= (_s)->__u6_addr.__u6_addr32[2]; \
167 (_d)->__u6_addr.__u6_addr32[3] |= (_s)->__u6_addr.__u6_addr32[3];
168 OR_MASK(&dst->dst_ip6, &src->dst_ip6);
169 OR_MASK(&dst->src_ip6, &src->src_ip6);
171 dst->flow_id6 |= src->flow_id6;
173 dst->dst_ip |= src->dst_ip;
174 dst->src_ip |= src->src_ip;
180 nonzero_mask(struct ipfw_flow_id *m)
182 if (m->dst_port || m->src_port || m->proto || m->extra)
184 if (IS_IP6_FLOW_ID(m)) {
186 m->dst_ip6.__u6_addr.__u6_addr32[0] ||
187 m->dst_ip6.__u6_addr.__u6_addr32[1] ||
188 m->dst_ip6.__u6_addr.__u6_addr32[2] ||
189 m->dst_ip6.__u6_addr.__u6_addr32[3] ||
190 m->src_ip6.__u6_addr.__u6_addr32[0] ||
191 m->src_ip6.__u6_addr.__u6_addr32[1] ||
192 m->src_ip6.__u6_addr.__u6_addr32[2] ||
193 m->src_ip6.__u6_addr.__u6_addr32[3] ||
196 return m->dst_ip || m->src_ip;
200 /* XXX we may want a better hash function */
202 flow_id_hash(struct ipfw_flow_id *id)
206 if (IS_IP6_FLOW_ID(id)) {
207 uint32_t *d = (uint32_t *)&id->dst_ip6;
208 uint32_t *s = (uint32_t *)&id->src_ip6;
209 i = (d[0] ) ^ (d[1]) ^
211 (d[0] >> 15) ^ (d[1] >> 15) ^
212 (d[2] >> 15) ^ (d[3] >> 15) ^
213 (s[0] << 1) ^ (s[1] << 1) ^
214 (s[2] << 1) ^ (s[3] << 1) ^
215 (s[0] << 16) ^ (s[1] << 16) ^
216 (s[2] << 16) ^ (s[3] << 16) ^
217 (id->dst_port << 1) ^ (id->src_port) ^
219 (id->proto ) ^ (id->flow_id6);
221 i = (id->dst_ip) ^ (id->dst_ip >> 15) ^
222 (id->src_ip << 1) ^ (id->src_ip >> 16) ^
224 (id->dst_port << 1) ^ (id->src_port) ^ (id->proto);
229 /* Like bcmp, returns 0 if ids match, 1 otherwise. */
231 flow_id_cmp(struct ipfw_flow_id *id1, struct ipfw_flow_id *id2)
233 int is_v6 = IS_IP6_FLOW_ID(id1);
236 if (IS_IP6_FLOW_ID(id2))
237 return 1; /* different address families */
239 return (id1->dst_ip == id2->dst_ip &&
240 id1->src_ip == id2->src_ip &&
241 id1->dst_port == id2->dst_port &&
242 id1->src_port == id2->src_port &&
243 id1->proto == id2->proto &&
244 id1->extra == id2->extra) ? 0 : 1;
248 !bcmp(&id1->dst_ip6,&id2->dst_ip6, sizeof(id1->dst_ip6)) &&
249 !bcmp(&id1->src_ip6,&id2->src_ip6, sizeof(id1->src_ip6)) &&
250 id1->dst_port == id2->dst_port &&
251 id1->src_port == id2->src_port &&
252 id1->proto == id2->proto &&
253 id1->extra == id2->extra &&
254 id1->flow_id6 == id2->flow_id6) ? 0 : 1;
256 /*--------- end of flow-id mask, hash and compare ---------*/
258 /*--- support functions for the qht hashtable ----
259 * Entries are hashed by flow-id
262 q_hash(uintptr_t key, int flags, void *arg)
264 /* compute the hash slot from the flow id */
265 struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
266 &((struct dn_queue *)key)->ni.fid :
267 (struct ipfw_flow_id *)key;
269 return flow_id_hash(id);
273 q_match(void *obj, uintptr_t key, int flags, void *arg)
275 struct dn_queue *o = (struct dn_queue *)obj;
276 struct ipfw_flow_id *id2;
278 if (flags & DNHT_KEY_IS_OBJ) {
279 /* compare pointers */
280 id2 = &((struct dn_queue *)key)->ni.fid;
282 id2 = (struct ipfw_flow_id *)key;
284 return (0 == flow_id_cmp(&o->ni.fid, id2));
288 * create a new queue instance for the given 'key'.
291 q_new(uintptr_t key, int flags, void *arg)
293 struct dn_queue *q, *template = arg;
294 struct dn_fsk *fs = template->fs;
295 int size = sizeof(*q) + fs->sched->fp->q_datalen;
297 q = malloc(size, M_DUMMYNET, M_NOWAIT | M_ZERO);
299 D("no memory for new queue");
303 set_oid(&q->ni.oid, DN_QUEUE, size);
304 if (fs->fs.flags & DN_QHT_HASH)
305 q->ni.fid = *(struct ipfw_flow_id *)key;
307 q->_si = template->_si;
310 if (fs->sched->fp->new_queue)
311 fs->sched->fp->new_queue(q);
312 dn_cfg.queue_count++;
317 * Notify schedulers that a queue is going away.
318 * If (flags & DN_DESTROY), also free the packets.
319 * The version for callbacks is called q_delete_cb().
322 dn_delete_queue(struct dn_queue *q, int flags)
324 struct dn_fsk *fs = q->fs;
326 // D("fs %p si %p\n", fs, q->_si);
327 /* notify the parent scheduler that the queue is going away */
328 if (fs && fs->sched->fp->free_queue)
329 fs->sched->fp->free_queue(q);
332 if (flags & DN_DESTROY) {
334 dn_free_pkts(q->mq.head);
335 bzero(q, sizeof(*q)); // safety
337 dn_cfg.queue_count--;
342 q_delete_cb(void *q, void *arg)
344 int flags = (int)(uintptr_t)arg;
345 dn_delete_queue(q, flags);
346 return (flags & DN_DESTROY) ? DNHT_SCAN_DEL : 0;
350 * calls dn_delete_queue/q_delete_cb on all queues,
351 * which notifies the parent scheduler and possibly drains packets.
352 * flags & DN_DESTROY: drains queues and destroy qht;
355 qht_delete(struct dn_fsk *fs, int flags)
357 ND("fs %d start flags %d qht %p",
358 fs->fs.fs_nr, flags, fs->qht);
361 if (fs->fs.flags & DN_QHT_HASH) {
362 dn_ht_scan(fs->qht, q_delete_cb, (void *)(uintptr_t)flags);
363 if (flags & DN_DESTROY) {
364 dn_ht_free(fs->qht, 0);
368 dn_delete_queue((struct dn_queue *)(fs->qht), flags);
369 if (flags & DN_DESTROY)
375 * Find and possibly create the queue for a MULTIQUEUE scheduler.
376 * We never call it for !MULTIQUEUE (the queue is in the sch_inst).
379 ipdn_q_find(struct dn_fsk *fs, struct dn_sch_inst *si,
380 struct ipfw_flow_id *id)
382 struct dn_queue template;
387 if (fs->fs.flags & DN_QHT_HASH) {
388 struct ipfw_flow_id masked_id;
389 if (fs->qht == NULL) {
390 fs->qht = dn_ht_init(NULL, fs->fs.buckets,
391 offsetof(struct dn_queue, q_next),
392 q_hash, q_match, q_new);
397 flow_id_mask(&fs->fsk_mask, &masked_id);
398 return dn_ht_find(fs->qht, (uintptr_t)&masked_id,
399 DNHT_INSERT, &template);
402 fs->qht = q_new(0, 0, &template);
403 return (struct dn_queue *)fs->qht;
406 /*--- end of queue hash table ---*/
408 /*--- support functions for the sch_inst hashtable ----
410 * These are hashed by flow-id
413 si_hash(uintptr_t key, int flags, void *arg)
415 /* compute the hash slot from the flow id */
416 struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
417 &((struct dn_sch_inst *)key)->ni.fid :
418 (struct ipfw_flow_id *)key;
420 return flow_id_hash(id);
424 si_match(void *obj, uintptr_t key, int flags, void *arg)
426 struct dn_sch_inst *o = obj;
427 struct ipfw_flow_id *id2;
429 id2 = (flags & DNHT_KEY_IS_OBJ) ?
430 &((struct dn_sch_inst *)key)->ni.fid :
431 (struct ipfw_flow_id *)key;
432 return flow_id_cmp(&o->ni.fid, id2) == 0;
436 * create a new instance for the given 'key'
437 * Allocate memory for instance, delay line and scheduler private data.
440 si_new(uintptr_t key, int flags, void *arg)
442 struct dn_schk *s = arg;
443 struct dn_sch_inst *si;
444 int l = sizeof(*si) + s->fp->si_datalen;
446 si = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
449 /* Set length only for the part passed up to userland. */
450 set_oid(&si->ni.oid, DN_SCH_I, sizeof(struct dn_flow));
451 set_oid(&(si->dline.oid), DN_DELAY_LINE,
452 sizeof(struct delay_line));
453 /* mark si and dline as outside the event queue */
454 si->ni.oid.id = si->dline.oid.id = -1;
459 if (s->fp->new_sched && s->fp->new_sched(si)) {
460 D("new_sched error");
463 if (s->sch.flags & DN_HAVE_MASK)
464 si->ni.fid = *(struct ipfw_flow_id *)key;
471 bzero(si, sizeof(*si)); // safety
472 free(si, M_DUMMYNET);
478 * Callback from siht to delete all scheduler instances. Remove
479 * si and delay line from the system heap, destroy all queues.
480 * We assume that all flowset have been notified and do not
481 * point to us anymore.
484 si_destroy(void *_si, void *arg)
486 struct dn_sch_inst *si = _si;
487 struct dn_schk *s = si->sched;
488 struct delay_line *dl = &si->dline;
490 if (dl->oid.subtype) /* remove delay line from event heap */
491 heap_extract(&dn_cfg.evheap, dl);
492 dn_free_pkts(dl->mq.head); /* drain delay line */
493 if (si->kflags & DN_ACTIVE) /* remove si from event heap */
494 heap_extract(&dn_cfg.evheap, si);
495 if (s->fp->free_sched)
496 s->fp->free_sched(si);
497 bzero(si, sizeof(*si)); /* safety */
498 free(si, M_DUMMYNET);
500 return DNHT_SCAN_DEL;
504 * Find the scheduler instance for this packet. If we need to apply
505 * a mask, do on a local copy of the flow_id to preserve the original.
506 * Assume siht is always initialized if we have a mask.
509 ipdn_si_find(struct dn_schk *s, struct ipfw_flow_id *id)
512 if (s->sch.flags & DN_HAVE_MASK) {
513 struct ipfw_flow_id id_t = *id;
514 flow_id_mask(&s->sch.sched_mask, &id_t);
515 return dn_ht_find(s->siht, (uintptr_t)&id_t,
519 s->siht = si_new(0, 0, s);
520 return (struct dn_sch_inst *)s->siht;
523 /* callback to flush credit for the scheduler instance */
525 si_reset_credit(void *_si, void *arg)
527 struct dn_sch_inst *si = _si;
528 struct dn_link *p = &si->sched->link;
530 si->credit = p->burst + (dn_cfg.io_fast ? p->bandwidth : 0);
535 schk_reset_credit(struct dn_schk *s)
537 if (s->sch.flags & DN_HAVE_MASK)
538 dn_ht_scan(s->siht, si_reset_credit, NULL);
540 si_reset_credit(s->siht, NULL);
542 /*---- end of sch_inst hashtable ---------------------*/
544 /*-------------------------------------------------------
545 * flowset hash (fshash) support. Entries are hashed by fs_nr.
546 * New allocations are put in the fsunlinked list, from which
547 * they are removed when they point to a specific scheduler.
550 fsk_hash(uintptr_t key, int flags, void *arg)
552 uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
553 ((struct dn_fsk *)key)->fs.fs_nr;
555 return ( (i>>8)^(i>>4)^i );
559 fsk_match(void *obj, uintptr_t key, int flags, void *arg)
561 struct dn_fsk *fs = obj;
562 int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
563 ((struct dn_fsk *)key)->fs.fs_nr;
565 return (fs->fs.fs_nr == i);
569 fsk_new(uintptr_t key, int flags, void *arg)
573 fs = malloc(sizeof(*fs), M_DUMMYNET, M_NOWAIT | M_ZERO);
575 set_oid(&fs->fs.oid, DN_FS, sizeof(fs->fs));
577 fs->drain_bucket = 0;
578 SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
584 * detach flowset from its current scheduler. Flags as follows:
585 * DN_DETACH removes from the fsk_list
586 * DN_DESTROY deletes individual queues
587 * DN_DELETE_FS destroys the flowset (otherwise goes in unlinked).
590 fsk_detach(struct dn_fsk *fs, int flags)
592 if (flags & DN_DELETE_FS)
594 ND("fs %d from sched %d flags %s %s %s",
595 fs->fs.fs_nr, fs->fs.sched_nr,
596 (flags & DN_DELETE_FS) ? "DEL_FS":"",
597 (flags & DN_DESTROY) ? "DEL":"",
598 (flags & DN_DETACH) ? "DET":"");
599 if (flags & DN_DETACH) { /* detach from the list */
600 struct dn_fsk_head *h;
601 h = fs->sched ? &fs->sched->fsk_list : &dn_cfg.fsu;
602 SLIST_REMOVE(h, fs, dn_fsk, sch_chain);
604 /* Free the RED parameters, they will be recomputed on
605 * subsequent attach if needed.
608 free(fs->w_q_lookup, M_DUMMYNET);
609 fs->w_q_lookup = NULL;
610 qht_delete(fs, flags);
611 if (fs->sched && fs->sched->fp->free_fsk)
612 fs->sched->fp->free_fsk(fs);
614 if (flags & DN_DELETE_FS) {
615 bzero(fs, sizeof(fs)); /* safety */
616 free(fs, M_DUMMYNET);
619 SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
624 * Detach or destroy all flowsets in a list.
625 * flags specifies what to do:
626 * DN_DESTROY: flush all queues
627 * DN_DELETE_FS: DN_DESTROY + destroy flowset
628 * DN_DELETE_FS implies DN_DESTROY
631 fsk_detach_list(struct dn_fsk_head *h, int flags)
634 int n = 0; /* only for stats */
636 ND("head %p flags %x", h, flags);
637 while ((fs = SLIST_FIRST(h))) {
638 SLIST_REMOVE_HEAD(h, sch_chain);
640 fsk_detach(fs, flags);
642 ND("done %d flowsets", n);
646 * called on 'queue X delete' -- removes the flowset from fshash,
647 * deletes all queues for the flowset, and removes the flowset.
650 delete_fs(int i, int locked)
657 fs = dn_ht_find(dn_cfg.fshash, i, DNHT_REMOVE, NULL);
658 ND("fs %d found %p", i, fs);
660 fsk_detach(fs, DN_DETACH | DN_DELETE_FS);
669 /*----- end of flowset hashtable support -------------*/
671 /*------------------------------------------------------------
672 * Scheduler hash. When searching by index we pass sched_nr,
673 * otherwise we pass struct dn_sch * which is the first field in
674 * struct dn_schk so we can cast between the two. We use this trick
675 * because in the create phase (but it should be fixed).
678 schk_hash(uintptr_t key, int flags, void *_arg)
680 uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
681 ((struct dn_schk *)key)->sch.sched_nr;
682 return ( (i>>8)^(i>>4)^i );
686 schk_match(void *obj, uintptr_t key, int flags, void *_arg)
688 struct dn_schk *s = (struct dn_schk *)obj;
689 int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
690 ((struct dn_schk *)key)->sch.sched_nr;
691 return (s->sch.sched_nr == i);
695 * Create the entry and intialize with the sched hash if needed.
696 * Leave s->fp unset so we can tell whether a dn_ht_find() returns
697 * a new object or a previously existing one.
700 schk_new(uintptr_t key, int flags, void *arg)
702 struct schk_new_arg *a = arg;
704 int l = sizeof(*s) +a->fp->schk_datalen;
706 s = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
709 set_oid(&s->link.oid, DN_LINK, sizeof(s->link));
710 s->sch = *a->sch; // copy initial values
711 s->link.link_nr = s->sch.sched_nr;
712 SLIST_INIT(&s->fsk_list);
713 /* initialize the hash table or create the single instance */
714 s->fp = a->fp; /* si_new needs this */
716 if (s->sch.flags & DN_HAVE_MASK) {
717 s->siht = dn_ht_init(NULL, s->sch.buckets,
718 offsetof(struct dn_sch_inst, si_next),
719 si_hash, si_match, si_new);
720 if (s->siht == NULL) {
725 s->fp = NULL; /* mark as a new scheduler */
731 * Callback for sched delete. Notify all attached flowsets to
732 * detach from the scheduler, destroy the internal flowset, and
733 * all instances. The scheduler goes away too.
734 * arg is 0 (only detach flowsets and destroy instances)
735 * DN_DESTROY (detach & delete queues, delete schk)
736 * or DN_DELETE_FS (delete queues and flowsets, delete schk)
739 schk_delete_cb(void *obj, void *arg)
741 struct dn_schk *s = obj;
744 ND("sched %d arg %s%s",
746 a&DN_DESTROY ? "DEL ":"",
747 a&DN_DELETE_FS ? "DEL_FS":"");
749 fsk_detach_list(&s->fsk_list, arg ? DN_DESTROY : 0);
750 /* no more flowset pointing to us now */
751 if (s->sch.flags & DN_HAVE_MASK)
752 dn_ht_scan(s->siht, si_destroy, NULL);
754 si_destroy(s->siht, NULL);
756 free(s->profile, M_DUMMYNET);
762 bzero(s, sizeof(*s)); // safety
763 free(obj, M_DUMMYNET);
765 return DNHT_SCAN_DEL;
769 * called on a 'sched X delete' command. Deletes a single scheduler.
770 * This is done by removing from the schedhash, unlinking all
771 * flowsets and deleting their traffic.
778 s = dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
782 delete_fs(i + DN_MAX_ID, 1); /* first delete internal fs */
783 /* then detach flowsets, delete traffic */
784 schk_delete_cb(s, (void*)(uintptr_t)DN_DESTROY);
787 /*--- end of schk hashtable support ---*/
790 copy_obj(char **start, char *end, void *_o, const char *msg, int i)
792 struct dn_id *o = _o;
793 int have = end - *start;
795 if (have < o->len || o->len == 0 || o->type == 0) {
796 D("(WARN) type %d %s %d have %d need %d",
797 o->type, msg, i, have, o->len);
800 ND("type %d %s %d len %d", o->type, msg, i, o->len);
801 bcopy(_o, *start, o->len);
802 if (o->type == DN_LINK) {
803 /* Adjust burst parameter for link */
804 struct dn_link *l = (struct dn_link *)*start;
805 l->burst = div64(l->burst, 8 * hz);
806 } else if (o->type == DN_SCH) {
807 /* Set id->id to the number of instances */
808 struct dn_schk *s = _o;
809 struct dn_id *id = (struct dn_id *)(*start);
810 id->id = (s->sch.flags & DN_HAVE_MASK) ?
811 dn_ht_entries(s->siht) : (s->siht ? 1 : 0);
817 /* Specific function to copy a queue.
818 * Copies only the user-visible part of a queue (which is in
819 * a struct dn_flow), and sets len accordingly.
822 copy_obj_q(char **start, char *end, void *_o, const char *msg, int i)
824 struct dn_id *o = _o;
825 int have = end - *start;
826 int len = sizeof(struct dn_flow); /* see above comment */
828 if (have < len || o->len == 0 || o->type != DN_QUEUE) {
829 D("ERROR type %d %s %d have %d need %d",
830 o->type, msg, i, have, len);
833 ND("type %d %s %d len %d", o->type, msg, i, len);
834 bcopy(_o, *start, len);
835 ((struct dn_id*)(*start))->len = len;
841 copy_q_cb(void *obj, void *arg)
843 struct dn_queue *q = obj;
844 struct copy_args *a = arg;
845 struct dn_flow *ni = (struct dn_flow *)(*a->start);
846 if (copy_obj_q(a->start, a->end, &q->ni, "queue", -1))
847 return DNHT_SCAN_END;
848 ni->oid.type = DN_FLOW; /* override the DN_QUEUE */
849 ni->oid.id = si_hash((uintptr_t)&ni->fid, 0, NULL);
854 copy_q(struct copy_args *a, struct dn_fsk *fs, int flags)
858 if (fs->fs.flags & DN_QHT_HASH)
859 dn_ht_scan(fs->qht, copy_q_cb, a);
861 copy_q_cb(fs->qht, a);
866 * This routine only copies the initial part of a profile ? XXX
869 copy_profile(struct copy_args *a, struct dn_profile *p)
871 int have = a->end - *a->start;
872 /* XXX here we check for max length */
873 int profile_len = sizeof(struct dn_profile) -
874 ED_MAX_SAMPLES_NO*sizeof(int);
878 if (have < profile_len) {
879 D("error have %d need %d", have, profile_len);
882 bcopy(p, *a->start, profile_len);
883 ((struct dn_id *)(*a->start))->len = profile_len;
884 *a->start += profile_len;
889 copy_flowset(struct copy_args *a, struct dn_fsk *fs, int flags)
891 struct dn_fs *ufs = (struct dn_fs *)(*a->start);
894 ND("flowset %d", fs->fs.fs_nr);
895 if (copy_obj(a->start, a->end, &fs->fs, "flowset", fs->fs.fs_nr))
896 return DNHT_SCAN_END;
897 ufs->oid.id = (fs->fs.flags & DN_QHT_HASH) ?
898 dn_ht_entries(fs->qht) : (fs->qht ? 1 : 0);
899 if (flags) { /* copy queues */
906 copy_si_cb(void *obj, void *arg)
908 struct dn_sch_inst *si = obj;
909 struct copy_args *a = arg;
910 struct dn_flow *ni = (struct dn_flow *)(*a->start);
911 if (copy_obj(a->start, a->end, &si->ni, "inst",
912 si->sched->sch.sched_nr))
913 return DNHT_SCAN_END;
914 ni->oid.type = DN_FLOW; /* override the DN_SCH_I */
915 ni->oid.id = si_hash((uintptr_t)si, DNHT_KEY_IS_OBJ, NULL);
920 copy_si(struct copy_args *a, struct dn_schk *s, int flags)
922 if (s->sch.flags & DN_HAVE_MASK)
923 dn_ht_scan(s->siht, copy_si_cb, a);
925 copy_si_cb(s->siht, a);
930 * compute a list of children of a scheduler and copy up
933 copy_fsk_list(struct copy_args *a, struct dn_schk *s, int flags)
939 int n = 0, space = sizeof(*o);
940 SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
941 if (fs->fs.fs_nr < DN_MAX_ID)
944 space += n * sizeof(uint32_t);
945 DX(3, "sched %d has %d flowsets", s->sch.sched_nr, n);
946 if (a->end - *(a->start) < space)
947 return DNHT_SCAN_END;
948 o = (struct dn_id *)(*(a->start));
952 p = (uint32_t *)(o+1);
953 SLIST_FOREACH(fs, &s->fsk_list, sch_chain)
954 if (fs->fs.fs_nr < DN_MAX_ID)
960 copy_data_helper(void *_o, void *_arg)
962 struct copy_args *a = _arg;
963 uint32_t *r = a->extra->r; /* start of first range */
964 uint32_t *lim; /* first invalid pointer */
967 lim = (uint32_t *)((char *)(a->extra) + a->extra->o.len);
969 if (a->type == DN_LINK || a->type == DN_SCH) {
970 /* pipe|sched show, we receive a dn_schk */
971 struct dn_schk *s = _o;
974 if (a->type == DN_SCH && n >= DN_MAX_ID)
975 return 0; /* not a scheduler */
976 if (a->type == DN_LINK && n <= DN_MAX_ID)
977 return 0; /* not a pipe */
979 /* see if the object is within one of our ranges */
980 for (;r < lim; r+=2) {
981 if (n < r[0] || n > r[1])
983 /* Found a valid entry, copy and we are done */
984 if (a->flags & DN_C_LINK) {
985 if (copy_obj(a->start, a->end,
986 &s->link, "link", n))
987 return DNHT_SCAN_END;
988 if (copy_profile(a, s->profile))
989 return DNHT_SCAN_END;
990 if (copy_flowset(a, s->fs, 0))
991 return DNHT_SCAN_END;
993 if (a->flags & DN_C_SCH) {
994 if (copy_obj(a->start, a->end,
995 &s->sch, "sched", n))
996 return DNHT_SCAN_END;
997 /* list all attached flowsets */
998 if (copy_fsk_list(a, s, 0))
999 return DNHT_SCAN_END;
1001 if (a->flags & DN_C_FLOW)
1005 } else if (a->type == DN_FS) {
1006 /* queue show, skip internal flowsets */
1007 struct dn_fsk *fs = _o;
1012 /* see if the object is within one of our ranges */
1013 for (;r < lim; r+=2) {
1014 if (n < r[0] || n > r[1])
1016 if (copy_flowset(a, fs, 0))
1017 return DNHT_SCAN_END;
1019 break; /* we are done */
1025 static inline struct dn_schk *
1026 locate_scheduler(int i)
1028 return dn_ht_find(dn_cfg.schedhash, i, 0, NULL);
1032 * red parameters are in fixed point arithmetic.
1035 config_red(struct dn_fsk *fs)
1037 int64_t s, idle, weight, w0;
1040 fs->w_q = fs->fs.w_q;
1041 fs->max_p = fs->fs.max_p;
1043 /* Doing stuff that was in userland */
1044 i = fs->sched->link.bandwidth;
1046 hz * dn_cfg.red_avg_pkt_size * 8 * SCALE(1) / i;
1048 idle = div64((s * 3) , fs->w_q); /* s, fs->w_q scaled; idle not scaled */
1049 fs->lookup_step = div64(idle , dn_cfg.red_lookup_depth);
1050 /* fs->lookup_step not scaled, */
1051 if (!fs->lookup_step)
1052 fs->lookup_step = 1;
1053 w0 = weight = SCALE(1) - fs->w_q; //fs->w_q scaled
1055 for (t = fs->lookup_step; t > 1; --t)
1056 weight = SCALE_MUL(weight, w0);
1057 fs->lookup_weight = (int)(weight); // scaled
1059 /* Now doing stuff that was in kerneland */
1060 fs->min_th = SCALE(fs->fs.min_th);
1061 fs->max_th = SCALE(fs->fs.max_th);
1063 fs->c_1 = fs->max_p / (fs->fs.max_th - fs->fs.min_th);
1064 fs->c_2 = SCALE_MUL(fs->c_1, SCALE(fs->fs.min_th));
1066 if (fs->fs.flags & DN_IS_GENTLE_RED) {
1067 fs->c_3 = (SCALE(1) - fs->max_p) / fs->fs.max_th;
1068 fs->c_4 = SCALE(1) - 2 * fs->max_p;
1071 /* If the lookup table already exist, free and create it again. */
1072 if (fs->w_q_lookup) {
1073 free(fs->w_q_lookup, M_DUMMYNET);
1074 fs->w_q_lookup = NULL;
1076 if (dn_cfg.red_lookup_depth == 0) {
1077 printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1079 fs->fs.flags &= ~DN_IS_RED;
1080 fs->fs.flags &= ~DN_IS_GENTLE_RED;
1083 fs->lookup_depth = dn_cfg.red_lookup_depth;
1084 fs->w_q_lookup = (u_int *)malloc(fs->lookup_depth * sizeof(int),
1085 M_DUMMYNET, M_NOWAIT);
1086 if (fs->w_q_lookup == NULL) {
1087 printf("dummynet: sorry, cannot allocate red lookup table\n");
1088 fs->fs.flags &= ~DN_IS_RED;
1089 fs->fs.flags &= ~DN_IS_GENTLE_RED;
1093 /* Fill the lookup table with (1 - w_q)^x */
1094 fs->w_q_lookup[0] = SCALE(1) - fs->w_q;
1096 for (i = 1; i < fs->lookup_depth; i++)
1098 SCALE_MUL(fs->w_q_lookup[i - 1], fs->lookup_weight);
1100 if (dn_cfg.red_avg_pkt_size < 1)
1101 dn_cfg.red_avg_pkt_size = 512;
1102 fs->avg_pkt_size = dn_cfg.red_avg_pkt_size;
1103 if (dn_cfg.red_max_pkt_size < 1)
1104 dn_cfg.red_max_pkt_size = 1500;
1105 fs->max_pkt_size = dn_cfg.red_max_pkt_size;
1110 /* Scan all flowset attached to this scheduler and update red */
1112 update_red(struct dn_schk *s)
1115 SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1116 if (fs && (fs->fs.flags & DN_IS_RED))
1121 /* attach flowset to scheduler s, possibly requeue */
1123 fsk_attach(struct dn_fsk *fs, struct dn_schk *s)
1125 ND("remove fs %d from fsunlinked, link to sched %d",
1126 fs->fs.fs_nr, s->sch.sched_nr);
1127 SLIST_REMOVE(&dn_cfg.fsu, fs, dn_fsk, sch_chain);
1129 SLIST_INSERT_HEAD(&s->fsk_list, fs, sch_chain);
1132 /* XXX compute fsk_mask */
1133 fs->fsk_mask = fs->fs.flow_mask;
1134 if (fs->sched->sch.flags & DN_HAVE_MASK)
1135 flow_id_or(&fs->sched->sch.sched_mask, &fs->fsk_mask);
1138 * we must drain qht according to the old
1139 * type, and reinsert according to the new one.
1140 * The requeue is complex -- in general we need to
1141 * reclassify every single packet.
1142 * For the time being, let's hope qht is never set
1143 * when we reach this point.
1145 D("XXX TODO requeue from fs %d to sch %d",
1146 fs->fs.fs_nr, s->sch.sched_nr);
1149 /* set the new type for qht */
1150 if (nonzero_mask(&fs->fsk_mask))
1151 fs->fs.flags |= DN_QHT_HASH;
1153 fs->fs.flags &= ~DN_QHT_HASH;
1155 /* XXX config_red() can fail... */
1156 if (fs->fs.flags & DN_IS_RED)
1160 /* update all flowsets which may refer to this scheduler */
1162 update_fs(struct dn_schk *s)
1164 struct dn_fsk *fs, *tmp;
1166 SLIST_FOREACH_SAFE(fs, &dn_cfg.fsu, sch_chain, tmp) {
1167 if (s->sch.sched_nr != fs->fs.sched_nr) {
1168 D("fs %d for sch %d not %d still unlinked",
1169 fs->fs.fs_nr, fs->fs.sched_nr,
1178 * Configuration -- to preserve backward compatibility we use
1179 * the following scheme (N is 65536)
1180 * NUMBER SCHED LINK FLOWSET
1181 * 1 .. N-1 (1)WFQ (2)WFQ (3)queue
1182 * N+1 .. 2N-1 (4)FIFO (5)FIFO (6)FIFO for sched 1..N-1
1183 * 2N+1 .. 3N-1 -- -- (7)FIFO for sched N+1..2N-1
1185 * "pipe i config" configures #1, #2 and #3
1186 * "sched i config" configures #1 and possibly #6
1187 * "queue i config" configures #3
1188 * #1 is configured with 'pipe i config' or 'sched i config'
1189 * #2 is configured with 'pipe i config', and created if not
1190 * existing with 'sched i config'
1191 * #3 is configured with 'queue i config'
1192 * #4 is automatically configured after #1, can only be FIFO
1193 * #5 is automatically configured after #2
1194 * #6 is automatically created when #1 is !MULTIQUEUE,
1195 * and can be updated.
1196 * #7 is automatically configured after #2
1200 * configure a link (and its FIFO instance)
1203 config_link(struct dn_link *p, struct dn_id *arg)
1207 if (p->oid.len != sizeof(*p)) {
1208 D("invalid pipe len %d", p->oid.len);
1212 if (i <= 0 || i >= DN_MAX_ID)
1215 * The config program passes parameters as follows:
1216 * bw = bits/second (0 means no limits),
1217 * delay = ms, must be translated into ticks.
1218 * qsize = slots/bytes
1221 p->delay = (p->delay * hz) / 1000;
1222 /* Scale burst size: bytes -> bits * hz */
1226 /* do it twice, base link and FIFO link */
1227 for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1228 struct dn_schk *s = locate_scheduler(i);
1231 D("sched %d not found", i);
1234 /* remove profile if exists */
1236 free(s->profile, M_DUMMYNET);
1239 /* copy all parameters */
1240 s->link.oid = p->oid;
1241 s->link.link_nr = i;
1242 s->link.delay = p->delay;
1243 if (s->link.bandwidth != p->bandwidth) {
1244 /* XXX bandwidth changes, need to update red params */
1245 s->link.bandwidth = p->bandwidth;
1248 s->link.burst = p->burst;
1249 schk_reset_credit(s);
1257 * configure a flowset. Can be called from inside with locked=1,
1259 static struct dn_fsk *
1260 config_fs(struct dn_fs *nfs, struct dn_id *arg, int locked)
1265 if (nfs->oid.len != sizeof(*nfs)) {
1266 D("invalid flowset len %d", nfs->oid.len);
1270 if (i <= 0 || i >= 3*DN_MAX_ID)
1272 ND("flowset %d", i);
1273 /* XXX other sanity checks */
1274 if (nfs->flags & DN_QSIZE_BYTES) {
1275 ipdn_bound_var(&nfs->qsize, 16384,
1276 1500, dn_cfg.byte_limit, NULL); // "queue byte size");
1278 ipdn_bound_var(&nfs->qsize, 50,
1279 1, dn_cfg.slot_limit, NULL); // "queue slot size");
1281 if (nfs->flags & DN_HAVE_MASK) {
1282 /* make sure we have some buckets */
1283 ipdn_bound_var(&nfs->buckets, dn_cfg.hash_size,
1284 1, dn_cfg.max_hash_size, "flowset buckets");
1286 nfs->buckets = 1; /* we only need 1 */
1290 do { /* exit with break when done */
1292 int flags = nfs->sched_nr ? DNHT_INSERT : 0;
1294 int oldc = dn_cfg.fsk_count;
1295 fs = dn_ht_find(dn_cfg.fshash, i, flags, NULL);
1297 D("missing sched for flowset %d", i);
1300 /* grab some defaults from the existing one */
1301 if (nfs->sched_nr == 0) /* reuse */
1302 nfs->sched_nr = fs->fs.sched_nr;
1303 for (j = 0; j < sizeof(nfs->par)/sizeof(nfs->par[0]); j++) {
1304 if (nfs->par[j] == -1) /* reuse */
1305 nfs->par[j] = fs->fs.par[j];
1307 if (bcmp(&fs->fs, nfs, sizeof(*nfs)) == 0) {
1308 ND("flowset %d unchanged", i);
1309 break; /* no change, nothing to do */
1311 if (oldc != dn_cfg.fsk_count) /* new item */
1313 s = locate_scheduler(nfs->sched_nr);
1314 /* detach from old scheduler if needed, preserving
1315 * queues if we need to reattach. Then update the
1316 * configuration, and possibly attach to the new sched.
1318 DX(2, "fs %d changed sched %d@%p to %d@%p",
1320 fs->fs.sched_nr, fs->sched, nfs->sched_nr, s);
1322 int flags = s ? DN_DETACH : (DN_DETACH | DN_DESTROY);
1323 flags |= DN_DESTROY; /* XXX temporary */
1324 fsk_detach(fs, flags);
1326 fs->fs = *nfs; /* copy configuration */
1336 * config/reconfig a scheduler and its FIFO variant.
1337 * For !MULTIQUEUE schedulers, also set up the flowset.
1339 * On reconfigurations (detected because s->fp is set),
1340 * detach existing flowsets preserving traffic, preserve link,
1341 * and delete the old scheduler creating a new one.
1344 config_sched(struct dn_sch *_nsch, struct dn_id *arg)
1347 struct schk_new_arg a; /* argument for schk_new */
1349 struct dn_link p; /* copy of oldlink */
1350 struct dn_profile *pf = NULL; /* copy of old link profile */
1351 /* Used to preserv mask parameter */
1352 struct ipfw_flow_id new_mask;
1353 int new_buckets = 0;
1359 if (a.sch->oid.len != sizeof(*a.sch)) {
1360 D("bad sched len %d", a.sch->oid.len);
1363 i = a.sch->sched_nr;
1364 if (i <= 0 || i >= DN_MAX_ID)
1366 /* make sure we have some buckets */
1367 if (a.sch->flags & DN_HAVE_MASK)
1368 ipdn_bound_var(&a.sch->buckets, dn_cfg.hash_size,
1369 1, dn_cfg.max_hash_size, "sched buckets");
1370 /* XXX other sanity checks */
1371 bzero(&p, sizeof(p));
1373 pipe_cmd = a.sch->flags & DN_PIPE_CMD;
1374 a.sch->flags &= ~DN_PIPE_CMD; //XXX do it even if is not set?
1376 /* Copy mask parameter */
1377 new_mask = a.sch->sched_mask;
1378 new_buckets = a.sch->buckets;
1379 new_flags = a.sch->flags;
1382 again: /* run twice, for wfq and fifo */
1384 * lookup the type. If not supplied, use the previous one
1385 * or default to WF2Q+. Otherwise, return an error.
1388 a.fp = find_sched_type(a.sch->oid.subtype, a.sch->name);
1390 /* found. Lookup or create entry */
1391 s = dn_ht_find(dn_cfg.schedhash, i, DNHT_INSERT, &a);
1392 } else if (a.sch->oid.subtype == 0 && !a.sch->name[0]) {
1393 /* No type. search existing s* or retry with WF2Q+ */
1394 s = dn_ht_find(dn_cfg.schedhash, i, 0, &a);
1397 /* Scheduler exists, skip to FIFO scheduler
1398 * if command was pipe config...
1403 /* New scheduler, create a wf2q+ with no mask
1404 * if command was pipe config...
1407 /* clear mask parameter */
1408 bzero(&a.sch->sched_mask, sizeof(new_mask));
1410 a.sch->flags &= ~DN_HAVE_MASK;
1412 a.sch->oid.subtype = DN_SCHED_WF2QP;
1416 D("invalid scheduler type %d %s",
1417 a.sch->oid.subtype, a.sch->name);
1421 /* normalize name and subtype */
1422 a.sch->oid.subtype = a.fp->type;
1423 bzero(a.sch->name, sizeof(a.sch->name));
1424 strlcpy(a.sch->name, a.fp->name, sizeof(a.sch->name));
1426 D("cannot allocate scheduler %d", i);
1429 /* restore existing link if any */
1432 if (!pf || pf->link_nr != p.link_nr) { /* no saved value */
1433 s->profile = NULL; /* XXX maybe not needed */
1435 s->profile = malloc(sizeof(struct dn_profile),
1436 M_DUMMYNET, M_NOWAIT | M_ZERO);
1437 if (s->profile == NULL) {
1438 D("cannot allocate profile");
1441 bcopy(pf, s->profile, sizeof(*pf));
1445 if (s->fp == NULL) {
1446 DX(2, "sched %d new type %s", i, a.fp->name);
1447 } else if (s->fp != a.fp ||
1448 bcmp(a.sch, &s->sch, sizeof(*a.sch)) ) {
1449 /* already existing. */
1450 DX(2, "sched %d type changed from %s to %s",
1451 i, s->fp->name, a.fp->name);
1452 DX(4, " type/sub %d/%d -> %d/%d",
1453 s->sch.oid.type, s->sch.oid.subtype,
1454 a.sch->oid.type, a.sch->oid.subtype);
1455 if (s->link.link_nr == 0)
1456 D("XXX WARNING link 0 for sched %d", i);
1457 p = s->link; /* preserve link */
1458 if (s->profile) {/* preserve profile */
1460 pf = malloc(sizeof(*pf),
1461 M_DUMMYNET, M_NOWAIT | M_ZERO);
1462 if (pf) /* XXX should issue a warning otherwise */
1463 bcopy(s->profile, pf, sizeof(*pf));
1465 /* remove from the hash */
1466 dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
1467 /* Detach flowsets, preserve queues. */
1468 // schk_delete_cb(s, NULL);
1469 // XXX temporarily, kill queues
1470 schk_delete_cb(s, (void *)DN_DESTROY);
1473 DX(4, "sched %d unchanged type %s", i, a.fp->name);
1475 /* complete initialization */
1479 // XXX schk_reset_credit(s);
1480 /* create the internal flowset if needed,
1481 * trying to reuse existing ones if available
1483 if (!(s->fp->flags & DN_MULTIQUEUE) && !s->fs) {
1484 s->fs = dn_ht_find(dn_cfg.fshash, i, 0, NULL);
1487 bzero(&fs, sizeof(fs));
1488 set_oid(&fs.oid, DN_FS, sizeof(fs));
1489 fs.fs_nr = i + DN_MAX_ID;
1491 s->fs = config_fs(&fs, NULL, 1 /* locked */);
1494 schk_delete_cb(s, (void *)DN_DESTROY);
1495 D("error creating internal fs for %d", i);
1499 /* call init function after the flowset is created */
1504 if (i < DN_MAX_ID) { /* now configure the FIFO instance */
1507 /* Restore mask parameter for FIFO */
1508 a.sch->sched_mask = new_mask;
1509 a.sch->buckets = new_buckets;
1510 a.sch->flags = new_flags;
1512 /* sched config shouldn't modify the FIFO scheduler */
1513 if (dn_ht_find(dn_cfg.schedhash, i, 0, &a) != NULL) {
1514 /* FIFO already exist, don't touch it */
1515 err = 0; /* and this is not an error */
1519 a.sch->sched_nr = i;
1520 a.sch->oid.subtype = DN_SCHED_FIFO;
1521 bzero(a.sch->name, sizeof(a.sch->name));
1528 free(pf, M_DUMMYNET);
1533 * attach a profile to a link
1536 config_profile(struct dn_profile *pf, struct dn_id *arg)
1539 int i, olen, err = 0;
1541 if (pf->oid.len < sizeof(*pf)) {
1542 D("short profile len %d", pf->oid.len);
1546 if (i <= 0 || i >= DN_MAX_ID)
1548 /* XXX other sanity checks */
1550 for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1551 s = locate_scheduler(i);
1559 * If we had a profile and the new one does not fit,
1560 * or it is deleted, then we need to free memory.
1562 if (s->profile && (pf->samples_no == 0 ||
1563 s->profile->oid.len < pf->oid.len)) {
1564 free(s->profile, M_DUMMYNET);
1567 if (pf->samples_no == 0)
1570 * new profile, possibly allocate memory
1573 if (s->profile == NULL)
1574 s->profile = malloc(pf->oid.len,
1575 M_DUMMYNET, M_NOWAIT | M_ZERO);
1576 if (s->profile == NULL) {
1577 D("no memory for profile %d", i);
1581 /* preserve larger length XXX double check */
1582 olen = s->profile->oid.len;
1583 if (olen < pf->oid.len)
1585 bcopy(pf, s->profile, pf->oid.len);
1586 s->profile->oid.len = olen;
1593 * Delete all objects:
1596 dummynet_flush(void)
1599 /* delete all schedulers and related links/queues/flowsets */
1600 dn_ht_scan(dn_cfg.schedhash, schk_delete_cb,
1601 (void *)(uintptr_t)DN_DELETE_FS);
1602 /* delete all remaining (unlinked) flowsets */
1603 DX(4, "still %d unlinked fs", dn_cfg.fsk_count);
1604 dn_ht_free(dn_cfg.fshash, DNHT_REMOVE);
1605 fsk_detach_list(&dn_cfg.fsu, DN_DELETE_FS);
1606 /* Reinitialize system heap... */
1607 heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
1611 * Main handler for configuration. We are guaranteed to be called
1612 * with an oid which is at least a dn_id.
1613 * - the first object is the command (config, delete, flush, ...)
1614 * - config_link must be issued after the corresponding config_sched
1615 * - parameters (DN_TXT) for an object must preceed the object
1616 * processed on a config_sched.
1619 do_config(void *p, int l)
1621 struct dn_id *next, *o;
1622 int err = 0, err2 = 0;
1623 struct dn_id *arg = NULL;
1627 if (o->id != DN_API_VERSION) {
1628 D("invalid api version got %d need %d",
1629 o->id, DN_API_VERSION);
1632 for (; l >= sizeof(*o); o = next) {
1633 struct dn_id *prev = arg;
1634 if (o->len < sizeof(*o) || l < o->len) {
1635 D("bad len o->len %d len %d", o->len, l);
1640 next = (struct dn_id *)((char *)o + o->len);
1644 D("cmd %d not implemented", o->type);
1646 #ifdef EMULATE_SYSCTL
1647 /* sysctl emulation.
1648 * if we recognize the command, jump to the correct
1649 * handler and return
1652 err = kesysctl_emu_set(p,l);
1655 case DN_CMD_CONFIG: /* simply a header */
1659 /* the argument is in the first uintptr_t after o */
1660 a = (uintptr_t *)(o+1);
1661 if (o->len < sizeof(*o) + sizeof(*a)) {
1665 switch (o->subtype) {
1667 /* delete base and derived schedulers */
1669 err = delete_schk(*a);
1670 err2 = delete_schk(*a + DN_MAX_ID);
1677 D("invalid delete type %d",
1683 err = (*a <1 || *a >= DN_MAX_ID) ?
1684 EINVAL : delete_fs(*a, 0) ;
1694 case DN_TEXT: /* store argument the next block */
1699 err = config_link((struct dn_link *)o, arg);
1702 err = config_profile((struct dn_profile *)o, arg);
1705 err = config_sched((struct dn_sch *)o, arg);
1708 err = (NULL==config_fs((struct dn_fs *)o, arg, 0));
1720 compute_space(struct dn_id *cmd, struct copy_args *a)
1722 int x = 0, need = 0;
1723 int profile_size = sizeof(struct dn_profile) -
1724 ED_MAX_SAMPLES_NO*sizeof(int);
1726 /* NOTE about compute space:
1727 * NP = dn_cfg.schk_count
1728 * NSI = dn_cfg.si_count
1729 * NF = dn_cfg.fsk_count
1730 * NQ = dn_cfg.queue_count
1732 * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
1733 * link, scheduler template, flowset
1734 * integrated in scheduler and header
1736 * (NSI)*(dn_flow) all scheduler instance (includes
1737 * the queue instance)
1739 * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
1740 * link, scheduler template, flowset
1741 * integrated in scheduler and header
1743 * (NSI * dn_flow) all scheduler instances
1744 * (NF * sizeof(uint_32)) space for flowset list linked to scheduler
1745 * (NQ * dn_queue) all queue [XXXfor now not listed]
1747 * (NF * dn_fs) all flowset
1748 * (NQ * dn_queue) all queues
1750 switch (cmd->subtype) {
1753 /* XXX where do LINK and SCH differ ? */
1754 /* 'ipfw sched show' could list all queues associated to
1755 * a scheduler. This feature for now is disabled
1757 case DN_LINK: /* pipe show */
1758 x = DN_C_LINK | DN_C_SCH | DN_C_FLOW;
1759 need += dn_cfg.schk_count *
1760 (sizeof(struct dn_fs) + profile_size) / 2;
1761 need += dn_cfg.fsk_count * sizeof(uint32_t);
1763 case DN_SCH: /* sched show */
1764 need += dn_cfg.schk_count *
1765 (sizeof(struct dn_fs) + profile_size) / 2;
1766 need += dn_cfg.fsk_count * sizeof(uint32_t);
1767 x = DN_C_SCH | DN_C_LINK | DN_C_FLOW;
1769 case DN_FS: /* queue show */
1770 x = DN_C_FS | DN_C_QUEUE;
1772 case DN_GET_COMPAT: /* compatibility mode */
1773 need = dn_compat_calc_size();
1778 need += dn_cfg.schk_count * sizeof(struct dn_sch) / 2;
1779 /* NOT also, each fs might be attached to a sched */
1780 need += dn_cfg.schk_count * sizeof(struct dn_id) / 2;
1783 need += dn_cfg.fsk_count * sizeof(struct dn_fs);
1784 if (x & DN_C_LINK) {
1785 need += dn_cfg.schk_count * sizeof(struct dn_link) / 2;
1788 * When exporting a queue to userland, only pass up the
1789 * struct dn_flow, which is the only visible part.
1793 need += dn_cfg.queue_count * sizeof(struct dn_flow);
1795 need += dn_cfg.si_count * (sizeof(struct dn_flow));
1800 * If compat != NULL dummynet_get is called in compatibility mode.
1801 * *compat will be the pointer to the buffer to pass to ipfw
1804 dummynet_get(struct sockopt *sopt, void **compat)
1806 int have, i, need, error;
1807 char *start = NULL, *buf;
1808 size_t sopt_valsize;
1811 struct copy_range r;
1812 int l = sizeof(struct dn_id);
1814 bzero(&a, sizeof(a));
1815 bzero(&r, sizeof(r));
1817 /* save and restore original sopt_valsize around copyin */
1818 sopt_valsize = sopt->sopt_valsize;
1823 /* copy at least an oid, and possibly a full object */
1824 error = sooptcopyin(sopt, cmd, sizeof(r), sizeof(*cmd));
1825 sopt->sopt_valsize = sopt_valsize;
1829 #ifdef EMULATE_SYSCTL
1830 /* sysctl emulation. */
1831 if (cmd->type == DN_SYSCTL_GET)
1832 return kesysctl_emu_get(sopt);
1834 if (l > sizeof(r)) {
1835 /* request larger than default, allocate buffer */
1836 cmd = malloc(l, M_DUMMYNET, M_WAIT);
1838 return ENOMEM; //XXX
1839 error = sooptcopyin(sopt, cmd, l, l);
1840 sopt->sopt_valsize = sopt_valsize;
1844 } else { /* compatibility */
1846 cmd->type = DN_CMD_GET;
1847 cmd->len = sizeof(struct dn_id);
1848 cmd->subtype = DN_GET_COMPAT;
1849 // cmd->id = sopt_valsize;
1850 D("compatibility mode");
1852 a.extra = (struct copy_range *)cmd;
1853 if (cmd->len == sizeof(*cmd)) { /* no range, create a default */
1854 uint32_t *rp = (uint32_t *)(cmd + 1);
1855 cmd->len += 2* sizeof(uint32_t);
1857 rp[1] = DN_MAX_ID - 1;
1858 if (cmd->subtype == DN_LINK) {
1863 /* Count space (under lock) and allocate (outside lock).
1864 * Exit with lock held if we manage to get enough buffer.
1865 * Try a few times then give up.
1867 for (have = 0, i = 0; i < 10; i++) {
1869 need = compute_space(cmd, &a);
1871 /* if there is a range, ignore value from compute_space() */
1872 if (l > sizeof(*cmd))
1873 need = sopt_valsize - sizeof(*cmd);
1880 need += sizeof(*cmd);
1887 free(start, M_DUMMYNET);
1889 if (need > sopt_valsize)
1893 start = malloc(have, M_DUMMYNET, M_WAITOK | M_ZERO);
1894 if (start == NULL) {
1900 if (start == NULL) {
1905 error = sooptcopyout(sopt, cmd, sizeof(*cmd));
1909 ND("have %d:%d sched %d, %d:%d links %d, %d:%d flowsets %d, "
1910 "%d:%d si %d, %d:%d queues %d",
1911 dn_cfg.schk_count, sizeof(struct dn_sch), DN_SCH,
1912 dn_cfg.schk_count, sizeof(struct dn_link), DN_LINK,
1913 dn_cfg.fsk_count, sizeof(struct dn_fs), DN_FS,
1914 dn_cfg.si_count, sizeof(struct dn_flow), DN_SCH_I,
1915 dn_cfg.queue_count, sizeof(struct dn_queue), DN_QUEUE);
1916 sopt->sopt_valsize = sopt_valsize;
1917 a.type = cmd->subtype;
1919 if (compat == NULL) {
1920 bcopy(cmd, start, sizeof(*cmd));
1921 ((struct dn_id*)(start))->len = sizeof(struct dn_id);
1922 buf = start + sizeof(*cmd);
1926 a.end = start + have;
1927 /* start copying other objects */
1929 a.type = DN_COMPAT_PIPE;
1930 dn_ht_scan(dn_cfg.schedhash, copy_data_helper_compat, &a);
1931 a.type = DN_COMPAT_QUEUE;
1932 dn_ht_scan(dn_cfg.fshash, copy_data_helper_compat, &a);
1933 } else if (a.type == DN_FS) {
1934 dn_ht_scan(dn_cfg.fshash, copy_data_helper, &a);
1936 dn_ht_scan(dn_cfg.schedhash, copy_data_helper, &a);
1942 sopt->sopt_valsize = buf - start;
1943 /* free() is done by ip_dummynet_compat() */
1944 start = NULL; //XXX hack
1946 error = sooptcopyout(sopt, start, buf - start);
1949 if (cmd && cmd != &r.o)
1950 free(cmd, M_DUMMYNET);
1952 free(start, M_DUMMYNET);
1956 /* Callback called on scheduler instance to delete it if idle */
1958 drain_scheduler_cb(void *_si, void *arg)
1960 struct dn_sch_inst *si = _si;
1962 if ((si->kflags & DN_ACTIVE) || si->dline.mq.head != NULL)
1965 if (si->sched->fp->flags & DN_MULTIQUEUE) {
1966 if (si->q_count == 0)
1967 return si_destroy(si, NULL);
1970 } else { /* !DN_MULTIQUEUE */
1971 if ((si+1)->ni.length == 0)
1972 return si_destroy(si, NULL);
1976 return 0; /* unreachable */
1979 /* Callback called on scheduler to check if it has instances */
1981 drain_scheduler_sch_cb(void *_s, void *arg)
1983 struct dn_schk *s = _s;
1985 if (s->sch.flags & DN_HAVE_MASK) {
1986 dn_ht_scan_bucket(s->siht, &s->drain_bucket,
1987 drain_scheduler_cb, NULL);
1991 if (drain_scheduler_cb(s->siht, NULL) == DNHT_SCAN_DEL)
1998 /* Called every tick, try to delete a 'bucket' of scheduler */
2000 dn_drain_scheduler(void)
2002 dn_ht_scan_bucket(dn_cfg.schedhash, &dn_cfg.drain_sch,
2003 drain_scheduler_sch_cb, NULL);
2007 /* Callback called on queue to delete if it is idle */
2009 drain_queue_cb(void *_q, void *arg)
2011 struct dn_queue *q = _q;
2013 if (q->ni.length == 0) {
2014 dn_delete_queue(q, DN_DESTROY);
2015 return DNHT_SCAN_DEL; /* queue is deleted */
2018 return 0; /* queue isn't deleted */
2021 /* Callback called on flowset used to check if it has queues */
2023 drain_queue_fs_cb(void *_fs, void *arg)
2025 struct dn_fsk *fs = _fs;
2027 if (fs->fs.flags & DN_QHT_HASH) {
2028 /* Flowset has a hash table for queues */
2029 dn_ht_scan_bucket(fs->qht, &fs->drain_bucket,
2030 drain_queue_cb, NULL);
2033 /* No hash table for this flowset, null the pointer
2034 * if the queue is deleted
2037 if (drain_queue_cb(fs->qht, NULL) == DNHT_SCAN_DEL)
2044 /* Called every tick, try to delete a 'bucket' of queue */
2046 dn_drain_queue(void)
2048 /* scan a bucket of flowset */
2049 dn_ht_scan_bucket(dn_cfg.fshash, &dn_cfg.drain_fs,
2050 drain_queue_fs_cb, NULL);
2055 * Handler for the various dummynet socket options
2058 ip_dn_ctl(struct sockopt *sopt)
2063 error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2067 /* Disallow sets in really-really secure mode. */
2068 if (sopt->sopt_dir == SOPT_SET) {
2069 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2074 switch (sopt->sopt_name) {
2076 D("dummynet: unknown option %d", sopt->sopt_name);
2080 case IP_DUMMYNET_FLUSH:
2081 case IP_DUMMYNET_CONFIGURE:
2082 case IP_DUMMYNET_DEL: /* remove a pipe or queue */
2083 case IP_DUMMYNET_GET:
2084 D("dummynet: compat option %d", sopt->sopt_name);
2085 error = ip_dummynet_compat(sopt);
2089 if (sopt->sopt_dir == SOPT_GET) {
2090 error = dummynet_get(sopt, NULL);
2093 l = sopt->sopt_valsize;
2094 if (l < sizeof(struct dn_id) || l > 12000) {
2095 D("argument len %d invalid", l);
2098 p = malloc(l, M_TEMP, M_WAITOK); // XXX can it fail ?
2099 error = sooptcopyin(sopt, p, l, l);
2102 error = do_config(p, l);
2116 static int init_done = 0;
2122 printf("DUMMYNET with IPv6 initialized (100131)\n");
2124 /* Set defaults here. MSVC does not accept initializers,
2125 * and this is also useful for vimages
2128 dn_cfg.slot_limit = 100; /* Foot shooting limit for queues. */
2129 dn_cfg.byte_limit = 1024 * 1024;
2132 /* RED parameters */
2133 dn_cfg.red_lookup_depth = 256; /* default lookup table depth */
2134 dn_cfg.red_avg_pkt_size = 512; /* default medium packet size */
2135 dn_cfg.red_max_pkt_size = 1500; /* default max packet size */
2138 dn_cfg.max_hash_size = 1024; /* max in the hash tables */
2139 dn_cfg.hash_size = 64; /* default hash size */
2141 /* create hash tables for schedulers and flowsets.
2142 * In both we search by key and by pointer.
2144 dn_cfg.schedhash = dn_ht_init(NULL, dn_cfg.hash_size,
2145 offsetof(struct dn_schk, schk_next),
2146 schk_hash, schk_match, schk_new);
2147 dn_cfg.fshash = dn_ht_init(NULL, dn_cfg.hash_size,
2148 offsetof(struct dn_fsk, fsk_next),
2149 fsk_hash, fsk_match, fsk_new);
2151 /* bucket index to drain object */
2152 dn_cfg.drain_fs = 0;
2153 dn_cfg.drain_sch = 0;
2155 heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
2156 SLIST_INIT(&dn_cfg.fsu);
2157 SLIST_INIT(&dn_cfg.schedlist);
2160 ip_dn_ctl_ptr = ip_dn_ctl;
2161 ip_dn_io_ptr = dummynet_io;
2163 TASK_INIT(&dn_task, 0, dummynet_task, NULL);
2164 dn_tq = taskqueue_create_fast("dummynet", M_NOWAIT,
2165 taskqueue_thread_enqueue, &dn_tq);
2166 taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2168 callout_init(&dn_timeout, CALLOUT_MPSAFE);
2169 callout_reset(&dn_timeout, 1, dummynet, NULL);
2171 /* Initialize curr_time adjustment mechanics. */
2172 getmicrouptime(&dn_cfg.prev_t);
2179 callout_drain(&dn_timeout);
2182 ip_dn_ctl_ptr = NULL;
2183 ip_dn_io_ptr = NULL;
2187 taskqueue_drain(dn_tq, &dn_task);
2188 taskqueue_free(dn_tq);
2190 dn_ht_free(dn_cfg.schedhash, 0);
2191 dn_ht_free(dn_cfg.fshash, 0);
2192 heap_free(&dn_cfg.evheap);
2196 #endif /* KLD_MODULE */
2199 dummynet_modevent(module_t mod, int type, void *data)
2202 if (type == MOD_LOAD) {
2204 printf("DUMMYNET already loaded\n");
2209 } else if (type == MOD_UNLOAD) {
2210 #if !defined(KLD_MODULE)
2211 printf("dummynet statically compiled, cannot unload\n");
2221 /* modevent helpers for the modules */
2223 load_dn_sched(struct dn_alg *d)
2228 return 1; /* error */
2229 ip_dn_init(); /* just in case, we need the lock */
2231 /* Check that mandatory funcs exists */
2232 if (d->enqueue == NULL || d->dequeue == NULL) {
2233 D("missing enqueue or dequeue for %s", d->name);
2237 /* Search if scheduler already exists */
2239 SLIST_FOREACH(s, &dn_cfg.schedlist, next) {
2240 if (strcmp(s->name, d->name) == 0) {
2241 D("%s already loaded", d->name);
2242 break; /* scheduler already exists */
2246 SLIST_INSERT_HEAD(&dn_cfg.schedlist, d, next);
2248 D("dn_sched %s %sloaded", d->name, s ? "not ":"");
2253 unload_dn_sched(struct dn_alg *s)
2255 struct dn_alg *tmp, *r;
2258 D("called for %s", s->name);
2261 SLIST_FOREACH_SAFE(r, &dn_cfg.schedlist, next, tmp) {
2262 if (strcmp(s->name, r->name) != 0)
2264 D("ref_count = %d", r->ref_count);
2265 err = (r->ref_count != 0) ? EBUSY : 0;
2267 SLIST_REMOVE(&dn_cfg.schedlist, r, dn_alg, next);
2271 D("dn_sched %s %sunloaded", s->name, err ? "not ":"");
2276 dn_sched_modevent(module_t mod, int cmd, void *arg)
2278 struct dn_alg *sch = arg;
2280 if (cmd == MOD_LOAD)
2281 return load_dn_sched(sch);
2282 else if (cmd == MOD_UNLOAD)
2283 return unload_dn_sched(sch);
2288 static moduledata_t dummynet_mod = {
2289 "dummynet", dummynet_modevent, NULL
2292 DECLARE_MODULE(dummynet, dummynet_mod,
2293 SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY-1);
2294 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
2295 MODULE_VERSION(dummynet, 1);