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>
44 #include <sys/mutex.h>
47 #include <sys/rwlock.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
51 #include <sys/taskqueue.h>
52 #include <net/if.h> /* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */
53 #include <netinet/in.h>
54 #include <netinet/ip_var.h> /* ip_output(), IP_FORWARDING */
55 #include <netinet/ip_fw.h>
56 #include <netinet/ip_dummynet.h>
58 #include <netpfil/ipfw/ip_fw_private.h>
59 #include <netpfil/ipfw/dn_heap.h>
60 #include <netpfil/ipfw/ip_dn_private.h>
61 #include <netpfil/ipfw/dn_sched.h>
63 /* which objects to copy */
64 #define DN_C_LINK 0x01
66 #define DN_C_FLOW 0x04
68 #define DN_C_QUEUE 0x10
70 /* we use this argument in case of a schk_new */
76 /*---- callout hooks. ----*/
77 static struct callout dn_timeout;
78 static struct task dn_task;
79 static struct taskqueue *dn_tq = NULL;
85 (void)arg; /* UNUSED */
86 taskqueue_enqueue_fast(dn_tq, &dn_task);
93 callout_reset_sbt(&dn_timeout, tick_sbt, 0, dummynet, NULL,
94 C_HARDCLOCK | C_DIRECT_EXEC);
96 /*----- end of callout hooks -----*/
98 /* Return a scheduler descriptor given the type or name. */
99 static struct dn_alg *
100 find_sched_type(int type, char *name)
104 SLIST_FOREACH(d, &dn_cfg.schedlist, next) {
105 if (d->type == type || (name && !strcasecmp(d->name, name)))
108 return NULL; /* not found */
112 ipdn_bound_var(int *v, int dflt, int lo, int hi, const char *msg)
115 const char *op = NULL;
123 } else if (oldv > hi) {
129 printf("%s %s to %d (was %d)\n", op, msg, *v, oldv);
133 /*---- flow_id mask, hash and compare functions ---*/
135 * The flow_id includes the 5-tuple, the queue/pipe number
136 * which we store in the extra area in host order,
137 * and for ipv6 also the flow_id6.
138 * XXX see if we want the tos byte (can store in 'flags')
140 static struct ipfw_flow_id *
141 flow_id_mask(struct ipfw_flow_id *mask, struct ipfw_flow_id *id)
143 int is_v6 = IS_IP6_FLOW_ID(id);
145 id->dst_port &= mask->dst_port;
146 id->src_port &= mask->src_port;
147 id->proto &= mask->proto;
148 id->extra &= mask->extra;
150 APPLY_MASK(&id->dst_ip6, &mask->dst_ip6);
151 APPLY_MASK(&id->src_ip6, &mask->src_ip6);
152 id->flow_id6 &= mask->flow_id6;
154 id->dst_ip &= mask->dst_ip;
155 id->src_ip &= mask->src_ip;
160 /* computes an OR of two masks, result in dst and also returned */
161 static struct ipfw_flow_id *
162 flow_id_or(struct ipfw_flow_id *src, struct ipfw_flow_id *dst)
164 int is_v6 = IS_IP6_FLOW_ID(dst);
166 dst->dst_port |= src->dst_port;
167 dst->src_port |= src->src_port;
168 dst->proto |= src->proto;
169 dst->extra |= src->extra;
171 #define OR_MASK(_d, _s) \
172 (_d)->__u6_addr.__u6_addr32[0] |= (_s)->__u6_addr.__u6_addr32[0]; \
173 (_d)->__u6_addr.__u6_addr32[1] |= (_s)->__u6_addr.__u6_addr32[1]; \
174 (_d)->__u6_addr.__u6_addr32[2] |= (_s)->__u6_addr.__u6_addr32[2]; \
175 (_d)->__u6_addr.__u6_addr32[3] |= (_s)->__u6_addr.__u6_addr32[3];
176 OR_MASK(&dst->dst_ip6, &src->dst_ip6);
177 OR_MASK(&dst->src_ip6, &src->src_ip6);
179 dst->flow_id6 |= src->flow_id6;
181 dst->dst_ip |= src->dst_ip;
182 dst->src_ip |= src->src_ip;
188 nonzero_mask(struct ipfw_flow_id *m)
190 if (m->dst_port || m->src_port || m->proto || m->extra)
192 if (IS_IP6_FLOW_ID(m)) {
194 m->dst_ip6.__u6_addr.__u6_addr32[0] ||
195 m->dst_ip6.__u6_addr.__u6_addr32[1] ||
196 m->dst_ip6.__u6_addr.__u6_addr32[2] ||
197 m->dst_ip6.__u6_addr.__u6_addr32[3] ||
198 m->src_ip6.__u6_addr.__u6_addr32[0] ||
199 m->src_ip6.__u6_addr.__u6_addr32[1] ||
200 m->src_ip6.__u6_addr.__u6_addr32[2] ||
201 m->src_ip6.__u6_addr.__u6_addr32[3] ||
204 return m->dst_ip || m->src_ip;
208 /* XXX we may want a better hash function */
210 flow_id_hash(struct ipfw_flow_id *id)
214 if (IS_IP6_FLOW_ID(id)) {
215 uint32_t *d = (uint32_t *)&id->dst_ip6;
216 uint32_t *s = (uint32_t *)&id->src_ip6;
217 i = (d[0] ) ^ (d[1]) ^
219 (d[0] >> 15) ^ (d[1] >> 15) ^
220 (d[2] >> 15) ^ (d[3] >> 15) ^
221 (s[0] << 1) ^ (s[1] << 1) ^
222 (s[2] << 1) ^ (s[3] << 1) ^
223 (s[0] << 16) ^ (s[1] << 16) ^
224 (s[2] << 16) ^ (s[3] << 16) ^
225 (id->dst_port << 1) ^ (id->src_port) ^
227 (id->proto ) ^ (id->flow_id6);
229 i = (id->dst_ip) ^ (id->dst_ip >> 15) ^
230 (id->src_ip << 1) ^ (id->src_ip >> 16) ^
232 (id->dst_port << 1) ^ (id->src_port) ^ (id->proto);
237 /* Like bcmp, returns 0 if ids match, 1 otherwise. */
239 flow_id_cmp(struct ipfw_flow_id *id1, struct ipfw_flow_id *id2)
241 int is_v6 = IS_IP6_FLOW_ID(id1);
244 if (IS_IP6_FLOW_ID(id2))
245 return 1; /* different address families */
247 return (id1->dst_ip == id2->dst_ip &&
248 id1->src_ip == id2->src_ip &&
249 id1->dst_port == id2->dst_port &&
250 id1->src_port == id2->src_port &&
251 id1->proto == id2->proto &&
252 id1->extra == id2->extra) ? 0 : 1;
256 !bcmp(&id1->dst_ip6,&id2->dst_ip6, sizeof(id1->dst_ip6)) &&
257 !bcmp(&id1->src_ip6,&id2->src_ip6, sizeof(id1->src_ip6)) &&
258 id1->dst_port == id2->dst_port &&
259 id1->src_port == id2->src_port &&
260 id1->proto == id2->proto &&
261 id1->extra == id2->extra &&
262 id1->flow_id6 == id2->flow_id6) ? 0 : 1;
264 /*--------- end of flow-id mask, hash and compare ---------*/
266 /*--- support functions for the qht hashtable ----
267 * Entries are hashed by flow-id
270 q_hash(uintptr_t key, int flags, void *arg)
272 /* compute the hash slot from the flow id */
273 struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
274 &((struct dn_queue *)key)->ni.fid :
275 (struct ipfw_flow_id *)key;
277 return flow_id_hash(id);
281 q_match(void *obj, uintptr_t key, int flags, void *arg)
283 struct dn_queue *o = (struct dn_queue *)obj;
284 struct ipfw_flow_id *id2;
286 if (flags & DNHT_KEY_IS_OBJ) {
287 /* compare pointers */
288 id2 = &((struct dn_queue *)key)->ni.fid;
290 id2 = (struct ipfw_flow_id *)key;
292 return (0 == flow_id_cmp(&o->ni.fid, id2));
296 * create a new queue instance for the given 'key'.
299 q_new(uintptr_t key, int flags, void *arg)
301 struct dn_queue *q, *template = arg;
302 struct dn_fsk *fs = template->fs;
303 int size = sizeof(*q) + fs->sched->fp->q_datalen;
305 q = malloc(size, M_DUMMYNET, M_NOWAIT | M_ZERO);
307 D("no memory for new queue");
311 set_oid(&q->ni.oid, DN_QUEUE, size);
312 if (fs->fs.flags & DN_QHT_HASH)
313 q->ni.fid = *(struct ipfw_flow_id *)key;
315 q->_si = template->_si;
318 if (fs->sched->fp->new_queue)
319 fs->sched->fp->new_queue(q);
320 dn_cfg.queue_count++;
325 * Notify schedulers that a queue is going away.
326 * If (flags & DN_DESTROY), also free the packets.
327 * The version for callbacks is called q_delete_cb().
330 dn_delete_queue(struct dn_queue *q, int flags)
332 struct dn_fsk *fs = q->fs;
334 // D("fs %p si %p\n", fs, q->_si);
335 /* notify the parent scheduler that the queue is going away */
336 if (fs && fs->sched->fp->free_queue)
337 fs->sched->fp->free_queue(q);
340 if (flags & DN_DESTROY) {
342 dn_free_pkts(q->mq.head);
343 bzero(q, sizeof(*q)); // safety
345 dn_cfg.queue_count--;
350 q_delete_cb(void *q, void *arg)
352 int flags = (int)(uintptr_t)arg;
353 dn_delete_queue(q, flags);
354 return (flags & DN_DESTROY) ? DNHT_SCAN_DEL : 0;
358 * calls dn_delete_queue/q_delete_cb on all queues,
359 * which notifies the parent scheduler and possibly drains packets.
360 * flags & DN_DESTROY: drains queues and destroy qht;
363 qht_delete(struct dn_fsk *fs, int flags)
365 ND("fs %d start flags %d qht %p",
366 fs->fs.fs_nr, flags, fs->qht);
369 if (fs->fs.flags & DN_QHT_HASH) {
370 dn_ht_scan(fs->qht, q_delete_cb, (void *)(uintptr_t)flags);
371 if (flags & DN_DESTROY) {
372 dn_ht_free(fs->qht, 0);
376 dn_delete_queue((struct dn_queue *)(fs->qht), flags);
377 if (flags & DN_DESTROY)
383 * Find and possibly create the queue for a MULTIQUEUE scheduler.
384 * We never call it for !MULTIQUEUE (the queue is in the sch_inst).
387 ipdn_q_find(struct dn_fsk *fs, struct dn_sch_inst *si,
388 struct ipfw_flow_id *id)
390 struct dn_queue template;
395 if (fs->fs.flags & DN_QHT_HASH) {
396 struct ipfw_flow_id masked_id;
397 if (fs->qht == NULL) {
398 fs->qht = dn_ht_init(NULL, fs->fs.buckets,
399 offsetof(struct dn_queue, q_next),
400 q_hash, q_match, q_new);
405 flow_id_mask(&fs->fsk_mask, &masked_id);
406 return dn_ht_find(fs->qht, (uintptr_t)&masked_id,
407 DNHT_INSERT, &template);
410 fs->qht = q_new(0, 0, &template);
411 return (struct dn_queue *)fs->qht;
414 /*--- end of queue hash table ---*/
416 /*--- support functions for the sch_inst hashtable ----
418 * These are hashed by flow-id
421 si_hash(uintptr_t key, int flags, void *arg)
423 /* compute the hash slot from the flow id */
424 struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
425 &((struct dn_sch_inst *)key)->ni.fid :
426 (struct ipfw_flow_id *)key;
428 return flow_id_hash(id);
432 si_match(void *obj, uintptr_t key, int flags, void *arg)
434 struct dn_sch_inst *o = obj;
435 struct ipfw_flow_id *id2;
437 id2 = (flags & DNHT_KEY_IS_OBJ) ?
438 &((struct dn_sch_inst *)key)->ni.fid :
439 (struct ipfw_flow_id *)key;
440 return flow_id_cmp(&o->ni.fid, id2) == 0;
444 * create a new instance for the given 'key'
445 * Allocate memory for instance, delay line and scheduler private data.
448 si_new(uintptr_t key, int flags, void *arg)
450 struct dn_schk *s = arg;
451 struct dn_sch_inst *si;
452 int l = sizeof(*si) + s->fp->si_datalen;
454 si = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
458 /* Set length only for the part passed up to userland. */
459 set_oid(&si->ni.oid, DN_SCH_I, sizeof(struct dn_flow));
460 set_oid(&(si->dline.oid), DN_DELAY_LINE,
461 sizeof(struct delay_line));
462 /* mark si and dline as outside the event queue */
463 si->ni.oid.id = si->dline.oid.id = -1;
468 if (s->fp->new_sched && s->fp->new_sched(si)) {
469 D("new_sched error");
472 if (s->sch.flags & DN_HAVE_MASK)
473 si->ni.fid = *(struct ipfw_flow_id *)key;
480 bzero(si, sizeof(*si)); // safety
481 free(si, M_DUMMYNET);
487 * Callback from siht to delete all scheduler instances. Remove
488 * si and delay line from the system heap, destroy all queues.
489 * We assume that all flowset have been notified and do not
490 * point to us anymore.
493 si_destroy(void *_si, void *arg)
495 struct dn_sch_inst *si = _si;
496 struct dn_schk *s = si->sched;
497 struct delay_line *dl = &si->dline;
499 if (dl->oid.subtype) /* remove delay line from event heap */
500 heap_extract(&dn_cfg.evheap, dl);
501 dn_free_pkts(dl->mq.head); /* drain delay line */
502 if (si->kflags & DN_ACTIVE) /* remove si from event heap */
503 heap_extract(&dn_cfg.evheap, si);
504 if (s->fp->free_sched)
505 s->fp->free_sched(si);
506 bzero(si, sizeof(*si)); /* safety */
507 free(si, M_DUMMYNET);
509 return DNHT_SCAN_DEL;
513 * Find the scheduler instance for this packet. If we need to apply
514 * a mask, do on a local copy of the flow_id to preserve the original.
515 * Assume siht is always initialized if we have a mask.
518 ipdn_si_find(struct dn_schk *s, struct ipfw_flow_id *id)
521 if (s->sch.flags & DN_HAVE_MASK) {
522 struct ipfw_flow_id id_t = *id;
523 flow_id_mask(&s->sch.sched_mask, &id_t);
524 return dn_ht_find(s->siht, (uintptr_t)&id_t,
528 s->siht = si_new(0, 0, s);
529 return (struct dn_sch_inst *)s->siht;
532 /* callback to flush credit for the scheduler instance */
534 si_reset_credit(void *_si, void *arg)
536 struct dn_sch_inst *si = _si;
537 struct dn_link *p = &si->sched->link;
539 si->credit = p->burst + (dn_cfg.io_fast ? p->bandwidth : 0);
544 schk_reset_credit(struct dn_schk *s)
546 if (s->sch.flags & DN_HAVE_MASK)
547 dn_ht_scan(s->siht, si_reset_credit, NULL);
549 si_reset_credit(s->siht, NULL);
551 /*---- end of sch_inst hashtable ---------------------*/
553 /*-------------------------------------------------------
554 * flowset hash (fshash) support. Entries are hashed by fs_nr.
555 * New allocations are put in the fsunlinked list, from which
556 * they are removed when they point to a specific scheduler.
559 fsk_hash(uintptr_t key, int flags, void *arg)
561 uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
562 ((struct dn_fsk *)key)->fs.fs_nr;
564 return ( (i>>8)^(i>>4)^i );
568 fsk_match(void *obj, uintptr_t key, int flags, void *arg)
570 struct dn_fsk *fs = obj;
571 int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
572 ((struct dn_fsk *)key)->fs.fs_nr;
574 return (fs->fs.fs_nr == i);
578 fsk_new(uintptr_t key, int flags, void *arg)
582 fs = malloc(sizeof(*fs), M_DUMMYNET, M_NOWAIT | M_ZERO);
584 set_oid(&fs->fs.oid, DN_FS, sizeof(fs->fs));
586 fs->drain_bucket = 0;
587 SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
593 * detach flowset from its current scheduler. Flags as follows:
594 * DN_DETACH removes from the fsk_list
595 * DN_DESTROY deletes individual queues
596 * DN_DELETE_FS destroys the flowset (otherwise goes in unlinked).
599 fsk_detach(struct dn_fsk *fs, int flags)
601 if (flags & DN_DELETE_FS)
603 ND("fs %d from sched %d flags %s %s %s",
604 fs->fs.fs_nr, fs->fs.sched_nr,
605 (flags & DN_DELETE_FS) ? "DEL_FS":"",
606 (flags & DN_DESTROY) ? "DEL":"",
607 (flags & DN_DETACH) ? "DET":"");
608 if (flags & DN_DETACH) { /* detach from the list */
609 struct dn_fsk_head *h;
610 h = fs->sched ? &fs->sched->fsk_list : &dn_cfg.fsu;
611 SLIST_REMOVE(h, fs, dn_fsk, sch_chain);
613 /* Free the RED parameters, they will be recomputed on
614 * subsequent attach if needed.
617 free(fs->w_q_lookup, M_DUMMYNET);
618 fs->w_q_lookup = NULL;
619 qht_delete(fs, flags);
620 if (fs->sched && fs->sched->fp->free_fsk)
621 fs->sched->fp->free_fsk(fs);
623 if (flags & DN_DELETE_FS) {
624 bzero(fs, sizeof(*fs)); /* safety */
625 free(fs, M_DUMMYNET);
628 SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
633 * Detach or destroy all flowsets in a list.
634 * flags specifies what to do:
635 * DN_DESTROY: flush all queues
636 * DN_DELETE_FS: DN_DESTROY + destroy flowset
637 * DN_DELETE_FS implies DN_DESTROY
640 fsk_detach_list(struct dn_fsk_head *h, int flags)
643 int n = 0; /* only for stats */
645 ND("head %p flags %x", h, flags);
646 while ((fs = SLIST_FIRST(h))) {
647 SLIST_REMOVE_HEAD(h, sch_chain);
649 fsk_detach(fs, flags);
651 ND("done %d flowsets", n);
655 * called on 'queue X delete' -- removes the flowset from fshash,
656 * deletes all queues for the flowset, and removes the flowset.
659 delete_fs(int i, int locked)
666 fs = dn_ht_find(dn_cfg.fshash, i, DNHT_REMOVE, NULL);
667 ND("fs %d found %p", i, fs);
669 fsk_detach(fs, DN_DETACH | DN_DELETE_FS);
678 /*----- end of flowset hashtable support -------------*/
680 /*------------------------------------------------------------
681 * Scheduler hash. When searching by index we pass sched_nr,
682 * otherwise we pass struct dn_sch * which is the first field in
683 * struct dn_schk so we can cast between the two. We use this trick
684 * because in the create phase (but it should be fixed).
687 schk_hash(uintptr_t key, int flags, void *_arg)
689 uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
690 ((struct dn_schk *)key)->sch.sched_nr;
691 return ( (i>>8)^(i>>4)^i );
695 schk_match(void *obj, uintptr_t key, int flags, void *_arg)
697 struct dn_schk *s = (struct dn_schk *)obj;
698 int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
699 ((struct dn_schk *)key)->sch.sched_nr;
700 return (s->sch.sched_nr == i);
704 * Create the entry and intialize with the sched hash if needed.
705 * Leave s->fp unset so we can tell whether a dn_ht_find() returns
706 * a new object or a previously existing one.
709 schk_new(uintptr_t key, int flags, void *arg)
711 struct schk_new_arg *a = arg;
713 int l = sizeof(*s) +a->fp->schk_datalen;
715 s = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
718 set_oid(&s->link.oid, DN_LINK, sizeof(s->link));
719 s->sch = *a->sch; // copy initial values
720 s->link.link_nr = s->sch.sched_nr;
721 SLIST_INIT(&s->fsk_list);
722 /* initialize the hash table or create the single instance */
723 s->fp = a->fp; /* si_new needs this */
725 if (s->sch.flags & DN_HAVE_MASK) {
726 s->siht = dn_ht_init(NULL, s->sch.buckets,
727 offsetof(struct dn_sch_inst, si_next),
728 si_hash, si_match, si_new);
729 if (s->siht == NULL) {
734 s->fp = NULL; /* mark as a new scheduler */
740 * Callback for sched delete. Notify all attached flowsets to
741 * detach from the scheduler, destroy the internal flowset, and
742 * all instances. The scheduler goes away too.
743 * arg is 0 (only detach flowsets and destroy instances)
744 * DN_DESTROY (detach & delete queues, delete schk)
745 * or DN_DELETE_FS (delete queues and flowsets, delete schk)
748 schk_delete_cb(void *obj, void *arg)
750 struct dn_schk *s = obj;
753 ND("sched %d arg %s%s",
755 a&DN_DESTROY ? "DEL ":"",
756 a&DN_DELETE_FS ? "DEL_FS":"");
758 fsk_detach_list(&s->fsk_list, arg ? DN_DESTROY : 0);
759 /* no more flowset pointing to us now */
760 if (s->sch.flags & DN_HAVE_MASK) {
761 dn_ht_scan(s->siht, si_destroy, NULL);
762 dn_ht_free(s->siht, 0);
764 si_destroy(s->siht, NULL);
766 free(s->profile, M_DUMMYNET);
772 bzero(s, sizeof(*s)); // safety
773 free(obj, M_DUMMYNET);
775 return DNHT_SCAN_DEL;
779 * called on a 'sched X delete' command. Deletes a single scheduler.
780 * This is done by removing from the schedhash, unlinking all
781 * flowsets and deleting their traffic.
788 s = dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
792 delete_fs(i + DN_MAX_ID, 1); /* first delete internal fs */
793 /* then detach flowsets, delete traffic */
794 schk_delete_cb(s, (void*)(uintptr_t)DN_DESTROY);
797 /*--- end of schk hashtable support ---*/
800 copy_obj(char **start, char *end, void *_o, const char *msg, int i)
802 struct dn_id *o = _o;
803 int have = end - *start;
805 if (have < o->len || o->len == 0 || o->type == 0) {
806 D("(WARN) type %d %s %d have %d need %d",
807 o->type, msg, i, have, o->len);
810 ND("type %d %s %d len %d", o->type, msg, i, o->len);
811 bcopy(_o, *start, o->len);
812 if (o->type == DN_LINK) {
813 /* Adjust burst parameter for link */
814 struct dn_link *l = (struct dn_link *)*start;
815 l->burst = div64(l->burst, 8 * hz);
816 l->delay = l->delay * 1000 / hz;
817 } else if (o->type == DN_SCH) {
818 /* Set id->id to the number of instances */
819 struct dn_schk *s = _o;
820 struct dn_id *id = (struct dn_id *)(*start);
821 id->id = (s->sch.flags & DN_HAVE_MASK) ?
822 dn_ht_entries(s->siht) : (s->siht ? 1 : 0);
828 /* Specific function to copy a queue.
829 * Copies only the user-visible part of a queue (which is in
830 * a struct dn_flow), and sets len accordingly.
833 copy_obj_q(char **start, char *end, void *_o, const char *msg, int i)
835 struct dn_id *o = _o;
836 int have = end - *start;
837 int len = sizeof(struct dn_flow); /* see above comment */
839 if (have < len || o->len == 0 || o->type != DN_QUEUE) {
840 D("ERROR type %d %s %d have %d need %d",
841 o->type, msg, i, have, len);
844 ND("type %d %s %d len %d", o->type, msg, i, len);
845 bcopy(_o, *start, len);
846 ((struct dn_id*)(*start))->len = len;
852 copy_q_cb(void *obj, void *arg)
854 struct dn_queue *q = obj;
855 struct copy_args *a = arg;
856 struct dn_flow *ni = (struct dn_flow *)(*a->start);
857 if (copy_obj_q(a->start, a->end, &q->ni, "queue", -1))
858 return DNHT_SCAN_END;
859 ni->oid.type = DN_FLOW; /* override the DN_QUEUE */
860 ni->oid.id = si_hash((uintptr_t)&ni->fid, 0, NULL);
865 copy_q(struct copy_args *a, struct dn_fsk *fs, int flags)
869 if (fs->fs.flags & DN_QHT_HASH)
870 dn_ht_scan(fs->qht, copy_q_cb, a);
872 copy_q_cb(fs->qht, a);
877 * This routine only copies the initial part of a profile ? XXX
880 copy_profile(struct copy_args *a, struct dn_profile *p)
882 int have = a->end - *a->start;
883 /* XXX here we check for max length */
884 int profile_len = sizeof(struct dn_profile) -
885 ED_MAX_SAMPLES_NO*sizeof(int);
889 if (have < profile_len) {
890 D("error have %d need %d", have, profile_len);
893 bcopy(p, *a->start, profile_len);
894 ((struct dn_id *)(*a->start))->len = profile_len;
895 *a->start += profile_len;
900 copy_flowset(struct copy_args *a, struct dn_fsk *fs, int flags)
902 struct dn_fs *ufs = (struct dn_fs *)(*a->start);
905 ND("flowset %d", fs->fs.fs_nr);
906 if (copy_obj(a->start, a->end, &fs->fs, "flowset", fs->fs.fs_nr))
907 return DNHT_SCAN_END;
908 ufs->oid.id = (fs->fs.flags & DN_QHT_HASH) ?
909 dn_ht_entries(fs->qht) : (fs->qht ? 1 : 0);
910 if (flags) { /* copy queues */
917 copy_si_cb(void *obj, void *arg)
919 struct dn_sch_inst *si = obj;
920 struct copy_args *a = arg;
921 struct dn_flow *ni = (struct dn_flow *)(*a->start);
922 if (copy_obj(a->start, a->end, &si->ni, "inst",
923 si->sched->sch.sched_nr))
924 return DNHT_SCAN_END;
925 ni->oid.type = DN_FLOW; /* override the DN_SCH_I */
926 ni->oid.id = si_hash((uintptr_t)si, DNHT_KEY_IS_OBJ, NULL);
931 copy_si(struct copy_args *a, struct dn_schk *s, int flags)
933 if (s->sch.flags & DN_HAVE_MASK)
934 dn_ht_scan(s->siht, copy_si_cb, a);
936 copy_si_cb(s->siht, a);
941 * compute a list of children of a scheduler and copy up
944 copy_fsk_list(struct copy_args *a, struct dn_schk *s, int flags)
950 int n = 0, space = sizeof(*o);
951 SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
952 if (fs->fs.fs_nr < DN_MAX_ID)
955 space += n * sizeof(uint32_t);
956 DX(3, "sched %d has %d flowsets", s->sch.sched_nr, n);
957 if (a->end - *(a->start) < space)
958 return DNHT_SCAN_END;
959 o = (struct dn_id *)(*(a->start));
963 p = (uint32_t *)(o+1);
964 SLIST_FOREACH(fs, &s->fsk_list, sch_chain)
965 if (fs->fs.fs_nr < DN_MAX_ID)
971 copy_data_helper(void *_o, void *_arg)
973 struct copy_args *a = _arg;
974 uint32_t *r = a->extra->r; /* start of first range */
975 uint32_t *lim; /* first invalid pointer */
978 lim = (uint32_t *)((char *)(a->extra) + a->extra->o.len);
980 if (a->type == DN_LINK || a->type == DN_SCH) {
981 /* pipe|sched show, we receive a dn_schk */
982 struct dn_schk *s = _o;
985 if (a->type == DN_SCH && n >= DN_MAX_ID)
986 return 0; /* not a scheduler */
987 if (a->type == DN_LINK && n <= DN_MAX_ID)
988 return 0; /* not a pipe */
990 /* see if the object is within one of our ranges */
991 for (;r < lim; r += 2) {
992 if (n < r[0] || n > r[1])
994 /* Found a valid entry, copy and we are done */
995 if (a->flags & DN_C_LINK) {
996 if (copy_obj(a->start, a->end,
997 &s->link, "link", n))
998 return DNHT_SCAN_END;
999 if (copy_profile(a, s->profile))
1000 return DNHT_SCAN_END;
1001 if (copy_flowset(a, s->fs, 0))
1002 return DNHT_SCAN_END;
1004 if (a->flags & DN_C_SCH) {
1005 if (copy_obj(a->start, a->end,
1006 &s->sch, "sched", n))
1007 return DNHT_SCAN_END;
1008 /* list all attached flowsets */
1009 if (copy_fsk_list(a, s, 0))
1010 return DNHT_SCAN_END;
1012 if (a->flags & DN_C_FLOW)
1016 } else if (a->type == DN_FS) {
1017 /* queue show, skip internal flowsets */
1018 struct dn_fsk *fs = _o;
1023 /* see if the object is within one of our ranges */
1024 for (;r < lim; r += 2) {
1025 if (n < r[0] || n > r[1])
1027 if (copy_flowset(a, fs, 0))
1028 return DNHT_SCAN_END;
1030 break; /* we are done */
1036 static inline struct dn_schk *
1037 locate_scheduler(int i)
1039 return dn_ht_find(dn_cfg.schedhash, i, 0, NULL);
1043 * red parameters are in fixed point arithmetic.
1046 config_red(struct dn_fsk *fs)
1048 int64_t s, idle, weight, w0;
1051 fs->w_q = fs->fs.w_q;
1052 fs->max_p = fs->fs.max_p;
1054 /* Doing stuff that was in userland */
1055 i = fs->sched->link.bandwidth;
1057 hz * dn_cfg.red_avg_pkt_size * 8 * SCALE(1) / i;
1059 idle = div64((s * 3) , fs->w_q); /* s, fs->w_q scaled; idle not scaled */
1060 fs->lookup_step = div64(idle , dn_cfg.red_lookup_depth);
1061 /* fs->lookup_step not scaled, */
1062 if (!fs->lookup_step)
1063 fs->lookup_step = 1;
1064 w0 = weight = SCALE(1) - fs->w_q; //fs->w_q scaled
1066 for (t = fs->lookup_step; t > 1; --t)
1067 weight = SCALE_MUL(weight, w0);
1068 fs->lookup_weight = (int)(weight); // scaled
1070 /* Now doing stuff that was in kerneland */
1071 fs->min_th = SCALE(fs->fs.min_th);
1072 fs->max_th = SCALE(fs->fs.max_th);
1074 if (fs->fs.max_th == fs->fs.min_th)
1075 fs->c_1 = fs->max_p;
1077 fs->c_1 = SCALE((int64_t)(fs->max_p)) / (fs->fs.max_th - fs->fs.min_th);
1078 fs->c_2 = SCALE_MUL(fs->c_1, SCALE(fs->fs.min_th));
1080 if (fs->fs.flags & DN_IS_GENTLE_RED) {
1081 fs->c_3 = (SCALE(1) - fs->max_p) / fs->fs.max_th;
1082 fs->c_4 = SCALE(1) - 2 * fs->max_p;
1085 /* If the lookup table already exist, free and create it again. */
1086 if (fs->w_q_lookup) {
1087 free(fs->w_q_lookup, M_DUMMYNET);
1088 fs->w_q_lookup = NULL;
1090 if (dn_cfg.red_lookup_depth == 0) {
1091 printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1093 fs->fs.flags &= ~DN_IS_RED;
1094 fs->fs.flags &= ~DN_IS_GENTLE_RED;
1097 fs->lookup_depth = dn_cfg.red_lookup_depth;
1098 fs->w_q_lookup = (u_int *)malloc(fs->lookup_depth * sizeof(int),
1099 M_DUMMYNET, M_NOWAIT);
1100 if (fs->w_q_lookup == NULL) {
1101 printf("dummynet: sorry, cannot allocate red lookup table\n");
1102 fs->fs.flags &= ~DN_IS_RED;
1103 fs->fs.flags &= ~DN_IS_GENTLE_RED;
1107 /* Fill the lookup table with (1 - w_q)^x */
1108 fs->w_q_lookup[0] = SCALE(1) - fs->w_q;
1110 for (i = 1; i < fs->lookup_depth; i++)
1112 SCALE_MUL(fs->w_q_lookup[i - 1], fs->lookup_weight);
1114 if (dn_cfg.red_avg_pkt_size < 1)
1115 dn_cfg.red_avg_pkt_size = 512;
1116 fs->avg_pkt_size = dn_cfg.red_avg_pkt_size;
1117 if (dn_cfg.red_max_pkt_size < 1)
1118 dn_cfg.red_max_pkt_size = 1500;
1119 fs->max_pkt_size = dn_cfg.red_max_pkt_size;
1124 /* Scan all flowset attached to this scheduler and update red */
1126 update_red(struct dn_schk *s)
1129 SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1130 if (fs && (fs->fs.flags & DN_IS_RED))
1135 /* attach flowset to scheduler s, possibly requeue */
1137 fsk_attach(struct dn_fsk *fs, struct dn_schk *s)
1139 ND("remove fs %d from fsunlinked, link to sched %d",
1140 fs->fs.fs_nr, s->sch.sched_nr);
1141 SLIST_REMOVE(&dn_cfg.fsu, fs, dn_fsk, sch_chain);
1143 SLIST_INSERT_HEAD(&s->fsk_list, fs, sch_chain);
1146 /* XXX compute fsk_mask */
1147 fs->fsk_mask = fs->fs.flow_mask;
1148 if (fs->sched->sch.flags & DN_HAVE_MASK)
1149 flow_id_or(&fs->sched->sch.sched_mask, &fs->fsk_mask);
1152 * we must drain qht according to the old
1153 * type, and reinsert according to the new one.
1154 * The requeue is complex -- in general we need to
1155 * reclassify every single packet.
1156 * For the time being, let's hope qht is never set
1157 * when we reach this point.
1159 D("XXX TODO requeue from fs %d to sch %d",
1160 fs->fs.fs_nr, s->sch.sched_nr);
1163 /* set the new type for qht */
1164 if (nonzero_mask(&fs->fsk_mask))
1165 fs->fs.flags |= DN_QHT_HASH;
1167 fs->fs.flags &= ~DN_QHT_HASH;
1169 /* XXX config_red() can fail... */
1170 if (fs->fs.flags & DN_IS_RED)
1174 /* update all flowsets which may refer to this scheduler */
1176 update_fs(struct dn_schk *s)
1178 struct dn_fsk *fs, *tmp;
1180 SLIST_FOREACH_SAFE(fs, &dn_cfg.fsu, sch_chain, tmp) {
1181 if (s->sch.sched_nr != fs->fs.sched_nr) {
1182 D("fs %d for sch %d not %d still unlinked",
1183 fs->fs.fs_nr, fs->fs.sched_nr,
1192 * Configuration -- to preserve backward compatibility we use
1193 * the following scheme (N is 65536)
1194 * NUMBER SCHED LINK FLOWSET
1195 * 1 .. N-1 (1)WFQ (2)WFQ (3)queue
1196 * N+1 .. 2N-1 (4)FIFO (5)FIFO (6)FIFO for sched 1..N-1
1197 * 2N+1 .. 3N-1 -- -- (7)FIFO for sched N+1..2N-1
1199 * "pipe i config" configures #1, #2 and #3
1200 * "sched i config" configures #1 and possibly #6
1201 * "queue i config" configures #3
1202 * #1 is configured with 'pipe i config' or 'sched i config'
1203 * #2 is configured with 'pipe i config', and created if not
1204 * existing with 'sched i config'
1205 * #3 is configured with 'queue i config'
1206 * #4 is automatically configured after #1, can only be FIFO
1207 * #5 is automatically configured after #2
1208 * #6 is automatically created when #1 is !MULTIQUEUE,
1209 * and can be updated.
1210 * #7 is automatically configured after #2
1214 * configure a link (and its FIFO instance)
1217 config_link(struct dn_link *p, struct dn_id *arg)
1221 if (p->oid.len != sizeof(*p)) {
1222 D("invalid pipe len %d", p->oid.len);
1226 if (i <= 0 || i >= DN_MAX_ID)
1229 * The config program passes parameters as follows:
1230 * bw = bits/second (0 means no limits),
1231 * delay = ms, must be translated into ticks.
1232 * qsize = slots/bytes
1235 p->delay = (p->delay * hz) / 1000;
1236 /* Scale burst size: bytes -> bits * hz */
1240 /* do it twice, base link and FIFO link */
1241 for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1242 struct dn_schk *s = locate_scheduler(i);
1245 D("sched %d not found", i);
1248 /* remove profile if exists */
1250 free(s->profile, M_DUMMYNET);
1253 /* copy all parameters */
1254 s->link.oid = p->oid;
1255 s->link.link_nr = i;
1256 s->link.delay = p->delay;
1257 if (s->link.bandwidth != p->bandwidth) {
1258 /* XXX bandwidth changes, need to update red params */
1259 s->link.bandwidth = p->bandwidth;
1262 s->link.burst = p->burst;
1263 schk_reset_credit(s);
1271 * configure a flowset. Can be called from inside with locked=1,
1273 static struct dn_fsk *
1274 config_fs(struct dn_fs *nfs, struct dn_id *arg, int locked)
1279 if (nfs->oid.len != sizeof(*nfs)) {
1280 D("invalid flowset len %d", nfs->oid.len);
1284 if (i <= 0 || i >= 3*DN_MAX_ID)
1286 ND("flowset %d", i);
1287 /* XXX other sanity checks */
1288 if (nfs->flags & DN_QSIZE_BYTES) {
1289 ipdn_bound_var(&nfs->qsize, 16384,
1290 1500, dn_cfg.byte_limit, NULL); // "queue byte size");
1292 ipdn_bound_var(&nfs->qsize, 50,
1293 1, dn_cfg.slot_limit, NULL); // "queue slot size");
1295 if (nfs->flags & DN_HAVE_MASK) {
1296 /* make sure we have some buckets */
1297 ipdn_bound_var((int *)&nfs->buckets, dn_cfg.hash_size,
1298 1, dn_cfg.max_hash_size, "flowset buckets");
1300 nfs->buckets = 1; /* we only need 1 */
1304 do { /* exit with break when done */
1306 int flags = nfs->sched_nr ? DNHT_INSERT : 0;
1308 int oldc = dn_cfg.fsk_count;
1309 fs = dn_ht_find(dn_cfg.fshash, i, flags, NULL);
1311 D("missing sched for flowset %d", i);
1314 /* grab some defaults from the existing one */
1315 if (nfs->sched_nr == 0) /* reuse */
1316 nfs->sched_nr = fs->fs.sched_nr;
1317 for (j = 0; j < sizeof(nfs->par)/sizeof(nfs->par[0]); j++) {
1318 if (nfs->par[j] == -1) /* reuse */
1319 nfs->par[j] = fs->fs.par[j];
1321 if (bcmp(&fs->fs, nfs, sizeof(*nfs)) == 0) {
1322 ND("flowset %d unchanged", i);
1323 break; /* no change, nothing to do */
1325 if (oldc != dn_cfg.fsk_count) /* new item */
1327 s = locate_scheduler(nfs->sched_nr);
1328 /* detach from old scheduler if needed, preserving
1329 * queues if we need to reattach. Then update the
1330 * configuration, and possibly attach to the new sched.
1332 DX(2, "fs %d changed sched %d@%p to %d@%p",
1334 fs->fs.sched_nr, fs->sched, nfs->sched_nr, s);
1336 int flags = s ? DN_DETACH : (DN_DETACH | DN_DESTROY);
1337 flags |= DN_DESTROY; /* XXX temporary */
1338 fsk_detach(fs, flags);
1340 fs->fs = *nfs; /* copy configuration */
1350 * config/reconfig a scheduler and its FIFO variant.
1351 * For !MULTIQUEUE schedulers, also set up the flowset.
1353 * On reconfigurations (detected because s->fp is set),
1354 * detach existing flowsets preserving traffic, preserve link,
1355 * and delete the old scheduler creating a new one.
1358 config_sched(struct dn_sch *_nsch, struct dn_id *arg)
1361 struct schk_new_arg a; /* argument for schk_new */
1363 struct dn_link p; /* copy of oldlink */
1364 struct dn_profile *pf = NULL; /* copy of old link profile */
1365 /* Used to preserv mask parameter */
1366 struct ipfw_flow_id new_mask;
1367 int new_buckets = 0;
1373 if (a.sch->oid.len != sizeof(*a.sch)) {
1374 D("bad sched len %d", a.sch->oid.len);
1377 i = a.sch->sched_nr;
1378 if (i <= 0 || i >= DN_MAX_ID)
1380 /* make sure we have some buckets */
1381 if (a.sch->flags & DN_HAVE_MASK)
1382 ipdn_bound_var((int *)&a.sch->buckets, dn_cfg.hash_size,
1383 1, dn_cfg.max_hash_size, "sched buckets");
1384 /* XXX other sanity checks */
1385 bzero(&p, sizeof(p));
1387 pipe_cmd = a.sch->flags & DN_PIPE_CMD;
1388 a.sch->flags &= ~DN_PIPE_CMD; //XXX do it even if is not set?
1390 /* Copy mask parameter */
1391 new_mask = a.sch->sched_mask;
1392 new_buckets = a.sch->buckets;
1393 new_flags = a.sch->flags;
1396 again: /* run twice, for wfq and fifo */
1398 * lookup the type. If not supplied, use the previous one
1399 * or default to WF2Q+. Otherwise, return an error.
1402 a.fp = find_sched_type(a.sch->oid.subtype, a.sch->name);
1404 /* found. Lookup or create entry */
1405 s = dn_ht_find(dn_cfg.schedhash, i, DNHT_INSERT, &a);
1406 } else if (a.sch->oid.subtype == 0 && !a.sch->name[0]) {
1407 /* No type. search existing s* or retry with WF2Q+ */
1408 s = dn_ht_find(dn_cfg.schedhash, i, 0, &a);
1411 /* Scheduler exists, skip to FIFO scheduler
1412 * if command was pipe config...
1417 /* New scheduler, create a wf2q+ with no mask
1418 * if command was pipe config...
1421 /* clear mask parameter */
1422 bzero(&a.sch->sched_mask, sizeof(new_mask));
1424 a.sch->flags &= ~DN_HAVE_MASK;
1426 a.sch->oid.subtype = DN_SCHED_WF2QP;
1430 D("invalid scheduler type %d %s",
1431 a.sch->oid.subtype, a.sch->name);
1435 /* normalize name and subtype */
1436 a.sch->oid.subtype = a.fp->type;
1437 bzero(a.sch->name, sizeof(a.sch->name));
1438 strlcpy(a.sch->name, a.fp->name, sizeof(a.sch->name));
1440 D("cannot allocate scheduler %d", i);
1443 /* restore existing link if any */
1446 if (!pf || pf->link_nr != p.link_nr) { /* no saved value */
1447 s->profile = NULL; /* XXX maybe not needed */
1449 s->profile = malloc(sizeof(struct dn_profile),
1450 M_DUMMYNET, M_NOWAIT | M_ZERO);
1451 if (s->profile == NULL) {
1452 D("cannot allocate profile");
1455 bcopy(pf, s->profile, sizeof(*pf));
1459 if (s->fp == NULL) {
1460 DX(2, "sched %d new type %s", i, a.fp->name);
1461 } else if (s->fp != a.fp ||
1462 bcmp(a.sch, &s->sch, sizeof(*a.sch)) ) {
1463 /* already existing. */
1464 DX(2, "sched %d type changed from %s to %s",
1465 i, s->fp->name, a.fp->name);
1466 DX(4, " type/sub %d/%d -> %d/%d",
1467 s->sch.oid.type, s->sch.oid.subtype,
1468 a.sch->oid.type, a.sch->oid.subtype);
1469 if (s->link.link_nr == 0)
1470 D("XXX WARNING link 0 for sched %d", i);
1471 p = s->link; /* preserve link */
1472 if (s->profile) {/* preserve profile */
1474 pf = malloc(sizeof(*pf),
1475 M_DUMMYNET, M_NOWAIT | M_ZERO);
1476 if (pf) /* XXX should issue a warning otherwise */
1477 bcopy(s->profile, pf, sizeof(*pf));
1479 /* remove from the hash */
1480 dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
1481 /* Detach flowsets, preserve queues. */
1482 // schk_delete_cb(s, NULL);
1483 // XXX temporarily, kill queues
1484 schk_delete_cb(s, (void *)DN_DESTROY);
1487 DX(4, "sched %d unchanged type %s", i, a.fp->name);
1489 /* complete initialization */
1493 // XXX schk_reset_credit(s);
1494 /* create the internal flowset if needed,
1495 * trying to reuse existing ones if available
1497 if (!(s->fp->flags & DN_MULTIQUEUE) && !s->fs) {
1498 s->fs = dn_ht_find(dn_cfg.fshash, i, 0, NULL);
1501 bzero(&fs, sizeof(fs));
1502 set_oid(&fs.oid, DN_FS, sizeof(fs));
1503 fs.fs_nr = i + DN_MAX_ID;
1505 s->fs = config_fs(&fs, NULL, 1 /* locked */);
1508 schk_delete_cb(s, (void *)DN_DESTROY);
1509 D("error creating internal fs for %d", i);
1513 /* call init function after the flowset is created */
1518 if (i < DN_MAX_ID) { /* now configure the FIFO instance */
1521 /* Restore mask parameter for FIFO */
1522 a.sch->sched_mask = new_mask;
1523 a.sch->buckets = new_buckets;
1524 a.sch->flags = new_flags;
1526 /* sched config shouldn't modify the FIFO scheduler */
1527 if (dn_ht_find(dn_cfg.schedhash, i, 0, &a) != NULL) {
1528 /* FIFO already exist, don't touch it */
1529 err = 0; /* and this is not an error */
1533 a.sch->sched_nr = i;
1534 a.sch->oid.subtype = DN_SCHED_FIFO;
1535 bzero(a.sch->name, sizeof(a.sch->name));
1542 free(pf, M_DUMMYNET);
1547 * attach a profile to a link
1550 config_profile(struct dn_profile *pf, struct dn_id *arg)
1553 int i, olen, err = 0;
1555 if (pf->oid.len < sizeof(*pf)) {
1556 D("short profile len %d", pf->oid.len);
1560 if (i <= 0 || i >= DN_MAX_ID)
1562 /* XXX other sanity checks */
1564 for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1565 s = locate_scheduler(i);
1573 * If we had a profile and the new one does not fit,
1574 * or it is deleted, then we need to free memory.
1576 if (s->profile && (pf->samples_no == 0 ||
1577 s->profile->oid.len < pf->oid.len)) {
1578 free(s->profile, M_DUMMYNET);
1581 if (pf->samples_no == 0)
1584 * new profile, possibly allocate memory
1587 if (s->profile == NULL)
1588 s->profile = malloc(pf->oid.len,
1589 M_DUMMYNET, M_NOWAIT | M_ZERO);
1590 if (s->profile == NULL) {
1591 D("no memory for profile %d", i);
1595 /* preserve larger length XXX double check */
1596 olen = s->profile->oid.len;
1597 if (olen < pf->oid.len)
1599 bcopy(pf, s->profile, pf->oid.len);
1600 s->profile->oid.len = olen;
1607 * Delete all objects:
1610 dummynet_flush(void)
1613 /* delete all schedulers and related links/queues/flowsets */
1614 dn_ht_scan(dn_cfg.schedhash, schk_delete_cb,
1615 (void *)(uintptr_t)DN_DELETE_FS);
1616 /* delete all remaining (unlinked) flowsets */
1617 DX(4, "still %d unlinked fs", dn_cfg.fsk_count);
1618 dn_ht_free(dn_cfg.fshash, DNHT_REMOVE);
1619 fsk_detach_list(&dn_cfg.fsu, DN_DELETE_FS);
1620 /* Reinitialize system heap... */
1621 heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
1625 * Main handler for configuration. We are guaranteed to be called
1626 * with an oid which is at least a dn_id.
1627 * - the first object is the command (config, delete, flush, ...)
1628 * - config_link must be issued after the corresponding config_sched
1629 * - parameters (DN_TXT) for an object must preceed the object
1630 * processed on a config_sched.
1633 do_config(void *p, int l)
1635 struct dn_id *next, *o;
1636 int err = 0, err2 = 0;
1637 struct dn_id *arg = NULL;
1641 if (o->id != DN_API_VERSION) {
1642 D("invalid api version got %d need %d",
1643 o->id, DN_API_VERSION);
1646 for (; l >= sizeof(*o); o = next) {
1647 struct dn_id *prev = arg;
1648 if (o->len < sizeof(*o) || l < o->len) {
1649 D("bad len o->len %d len %d", o->len, l);
1654 next = (struct dn_id *)((char *)o + o->len);
1658 D("cmd %d not implemented", o->type);
1661 #ifdef EMULATE_SYSCTL
1662 /* sysctl emulation.
1663 * if we recognize the command, jump to the correct
1664 * handler and return
1667 err = kesysctl_emu_set(p, l);
1671 case DN_CMD_CONFIG: /* simply a header */
1675 /* the argument is in the first uintptr_t after o */
1676 a = (uintptr_t *)(o+1);
1677 if (o->len < sizeof(*o) + sizeof(*a)) {
1681 switch (o->subtype) {
1683 /* delete base and derived schedulers */
1685 err = delete_schk(*a);
1686 err2 = delete_schk(*a + DN_MAX_ID);
1693 D("invalid delete type %d",
1699 err = (*a <1 || *a >= DN_MAX_ID) ?
1700 EINVAL : delete_fs(*a, 0) ;
1710 case DN_TEXT: /* store argument the next block */
1715 err = config_link((struct dn_link *)o, arg);
1718 err = config_profile((struct dn_profile *)o, arg);
1721 err = config_sched((struct dn_sch *)o, arg);
1724 err = (NULL==config_fs((struct dn_fs *)o, arg, 0));
1736 compute_space(struct dn_id *cmd, struct copy_args *a)
1738 int x = 0, need = 0;
1739 int profile_size = sizeof(struct dn_profile) -
1740 ED_MAX_SAMPLES_NO*sizeof(int);
1742 /* NOTE about compute space:
1743 * NP = dn_cfg.schk_count
1744 * NSI = dn_cfg.si_count
1745 * NF = dn_cfg.fsk_count
1746 * NQ = dn_cfg.queue_count
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 instance (includes
1753 * the queue instance)
1755 * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
1756 * link, scheduler template, flowset
1757 * integrated in scheduler and header
1759 * (NSI * dn_flow) all scheduler instances
1760 * (NF * sizeof(uint_32)) space for flowset list linked to scheduler
1761 * (NQ * dn_queue) all queue [XXXfor now not listed]
1763 * (NF * dn_fs) all flowset
1764 * (NQ * dn_queue) all queues
1766 switch (cmd->subtype) {
1769 /* XXX where do LINK and SCH differ ? */
1770 /* 'ipfw sched show' could list all queues associated to
1771 * a scheduler. This feature for now is disabled
1773 case DN_LINK: /* pipe show */
1774 x = DN_C_LINK | DN_C_SCH | DN_C_FLOW;
1775 need += dn_cfg.schk_count *
1776 (sizeof(struct dn_fs) + profile_size) / 2;
1777 need += dn_cfg.fsk_count * sizeof(uint32_t);
1779 case DN_SCH: /* sched show */
1780 need += dn_cfg.schk_count *
1781 (sizeof(struct dn_fs) + profile_size) / 2;
1782 need += dn_cfg.fsk_count * sizeof(uint32_t);
1783 x = DN_C_SCH | DN_C_LINK | DN_C_FLOW;
1785 case DN_FS: /* queue show */
1786 x = DN_C_FS | DN_C_QUEUE;
1788 case DN_GET_COMPAT: /* compatibility mode */
1789 need = dn_compat_calc_size();
1794 need += dn_cfg.schk_count * sizeof(struct dn_sch) / 2;
1795 /* NOT also, each fs might be attached to a sched */
1796 need += dn_cfg.schk_count * sizeof(struct dn_id) / 2;
1799 need += dn_cfg.fsk_count * sizeof(struct dn_fs);
1800 if (x & DN_C_LINK) {
1801 need += dn_cfg.schk_count * sizeof(struct dn_link) / 2;
1804 * When exporting a queue to userland, only pass up the
1805 * struct dn_flow, which is the only visible part.
1809 need += dn_cfg.queue_count * sizeof(struct dn_flow);
1811 need += dn_cfg.si_count * (sizeof(struct dn_flow));
1816 * If compat != NULL dummynet_get is called in compatibility mode.
1817 * *compat will be the pointer to the buffer to pass to ipfw
1820 dummynet_get(struct sockopt *sopt, void **compat)
1822 int have, i, need, error;
1823 char *start = NULL, *buf;
1824 size_t sopt_valsize;
1827 struct copy_range r;
1828 int l = sizeof(struct dn_id);
1830 bzero(&a, sizeof(a));
1831 bzero(&r, sizeof(r));
1833 /* save and restore original sopt_valsize around copyin */
1834 sopt_valsize = sopt->sopt_valsize;
1839 /* copy at least an oid, and possibly a full object */
1840 error = sooptcopyin(sopt, cmd, sizeof(r), sizeof(*cmd));
1841 sopt->sopt_valsize = sopt_valsize;
1845 #ifdef EMULATE_SYSCTL
1846 /* sysctl emulation. */
1847 if (cmd->type == DN_SYSCTL_GET)
1848 return kesysctl_emu_get(sopt);
1850 if (l > sizeof(r)) {
1851 /* request larger than default, allocate buffer */
1852 cmd = malloc(l, M_DUMMYNET, M_WAITOK);
1853 error = sooptcopyin(sopt, cmd, l, l);
1854 sopt->sopt_valsize = sopt_valsize;
1858 } else { /* compatibility */
1860 cmd->type = DN_CMD_GET;
1861 cmd->len = sizeof(struct dn_id);
1862 cmd->subtype = DN_GET_COMPAT;
1863 // cmd->id = sopt_valsize;
1864 D("compatibility mode");
1866 a.extra = (struct copy_range *)cmd;
1867 if (cmd->len == sizeof(*cmd)) { /* no range, create a default */
1868 uint32_t *rp = (uint32_t *)(cmd + 1);
1869 cmd->len += 2* sizeof(uint32_t);
1871 rp[1] = DN_MAX_ID - 1;
1872 if (cmd->subtype == DN_LINK) {
1877 /* Count space (under lock) and allocate (outside lock).
1878 * Exit with lock held if we manage to get enough buffer.
1879 * Try a few times then give up.
1881 for (have = 0, i = 0; i < 10; i++) {
1883 need = compute_space(cmd, &a);
1885 /* if there is a range, ignore value from compute_space() */
1886 if (l > sizeof(*cmd))
1887 need = sopt_valsize - sizeof(*cmd);
1894 need += sizeof(*cmd);
1901 free(start, M_DUMMYNET);
1903 if (need > sopt_valsize)
1907 start = malloc(have, M_DUMMYNET, M_WAITOK | M_ZERO);
1910 if (start == NULL) {
1915 error = sooptcopyout(sopt, cmd, sizeof(*cmd));
1919 ND("have %d:%d sched %d, %d:%d links %d, %d:%d flowsets %d, "
1920 "%d:%d si %d, %d:%d queues %d",
1921 dn_cfg.schk_count, sizeof(struct dn_sch), DN_SCH,
1922 dn_cfg.schk_count, sizeof(struct dn_link), DN_LINK,
1923 dn_cfg.fsk_count, sizeof(struct dn_fs), DN_FS,
1924 dn_cfg.si_count, sizeof(struct dn_flow), DN_SCH_I,
1925 dn_cfg.queue_count, sizeof(struct dn_queue), DN_QUEUE);
1926 sopt->sopt_valsize = sopt_valsize;
1927 a.type = cmd->subtype;
1929 if (compat == NULL) {
1930 bcopy(cmd, start, sizeof(*cmd));
1931 ((struct dn_id*)(start))->len = sizeof(struct dn_id);
1932 buf = start + sizeof(*cmd);
1936 a.end = start + have;
1937 /* start copying other objects */
1939 a.type = DN_COMPAT_PIPE;
1940 dn_ht_scan(dn_cfg.schedhash, copy_data_helper_compat, &a);
1941 a.type = DN_COMPAT_QUEUE;
1942 dn_ht_scan(dn_cfg.fshash, copy_data_helper_compat, &a);
1943 } else if (a.type == DN_FS) {
1944 dn_ht_scan(dn_cfg.fshash, copy_data_helper, &a);
1946 dn_ht_scan(dn_cfg.schedhash, copy_data_helper, &a);
1952 sopt->sopt_valsize = buf - start;
1953 /* free() is done by ip_dummynet_compat() */
1954 start = NULL; //XXX hack
1956 error = sooptcopyout(sopt, start, buf - start);
1959 if (cmd && cmd != &r.o)
1960 free(cmd, M_DUMMYNET);
1962 free(start, M_DUMMYNET);
1966 /* Callback called on scheduler instance to delete it if idle */
1968 drain_scheduler_cb(void *_si, void *arg)
1970 struct dn_sch_inst *si = _si;
1972 if ((si->kflags & DN_ACTIVE) || si->dline.mq.head != NULL)
1975 if (si->sched->fp->flags & DN_MULTIQUEUE) {
1976 if (si->q_count == 0)
1977 return si_destroy(si, NULL);
1980 } else { /* !DN_MULTIQUEUE */
1981 if ((si+1)->ni.length == 0)
1982 return si_destroy(si, NULL);
1986 return 0; /* unreachable */
1989 /* Callback called on scheduler to check if it has instances */
1991 drain_scheduler_sch_cb(void *_s, void *arg)
1993 struct dn_schk *s = _s;
1995 if (s->sch.flags & DN_HAVE_MASK) {
1996 dn_ht_scan_bucket(s->siht, &s->drain_bucket,
1997 drain_scheduler_cb, NULL);
2001 if (drain_scheduler_cb(s->siht, NULL) == DNHT_SCAN_DEL)
2008 /* Called every tick, try to delete a 'bucket' of scheduler */
2010 dn_drain_scheduler(void)
2012 dn_ht_scan_bucket(dn_cfg.schedhash, &dn_cfg.drain_sch,
2013 drain_scheduler_sch_cb, NULL);
2017 /* Callback called on queue to delete if it is idle */
2019 drain_queue_cb(void *_q, void *arg)
2021 struct dn_queue *q = _q;
2023 if (q->ni.length == 0) {
2024 dn_delete_queue(q, DN_DESTROY);
2025 return DNHT_SCAN_DEL; /* queue is deleted */
2028 return 0; /* queue isn't deleted */
2031 /* Callback called on flowset used to check if it has queues */
2033 drain_queue_fs_cb(void *_fs, void *arg)
2035 struct dn_fsk *fs = _fs;
2037 if (fs->fs.flags & DN_QHT_HASH) {
2038 /* Flowset has a hash table for queues */
2039 dn_ht_scan_bucket(fs->qht, &fs->drain_bucket,
2040 drain_queue_cb, NULL);
2043 /* No hash table for this flowset, null the pointer
2044 * if the queue is deleted
2047 if (drain_queue_cb(fs->qht, NULL) == DNHT_SCAN_DEL)
2054 /* Called every tick, try to delete a 'bucket' of queue */
2056 dn_drain_queue(void)
2058 /* scan a bucket of flowset */
2059 dn_ht_scan_bucket(dn_cfg.fshash, &dn_cfg.drain_fs,
2060 drain_queue_fs_cb, NULL);
2065 * Handler for the various dummynet socket options
2068 ip_dn_ctl(struct sockopt *sopt)
2073 error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2077 /* Disallow sets in really-really secure mode. */
2078 if (sopt->sopt_dir == SOPT_SET) {
2079 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2084 switch (sopt->sopt_name) {
2086 D("dummynet: unknown option %d", sopt->sopt_name);
2090 case IP_DUMMYNET_FLUSH:
2091 case IP_DUMMYNET_CONFIGURE:
2092 case IP_DUMMYNET_DEL: /* remove a pipe or queue */
2093 case IP_DUMMYNET_GET:
2094 D("dummynet: compat option %d", sopt->sopt_name);
2095 error = ip_dummynet_compat(sopt);
2099 if (sopt->sopt_dir == SOPT_GET) {
2100 error = dummynet_get(sopt, NULL);
2103 l = sopt->sopt_valsize;
2104 if (l < sizeof(struct dn_id) || l > 12000) {
2105 D("argument len %d invalid", l);
2108 p = malloc(l, M_TEMP, M_WAITOK); // XXX can it fail ?
2109 error = sooptcopyin(sopt, p, l, l);
2112 error = do_config(p, l);
2126 if (dn_cfg.init_done)
2128 printf("DUMMYNET %p with IPv6 initialized (100409)\n", curvnet);
2129 dn_cfg.init_done = 1;
2130 /* Set defaults here. MSVC does not accept initializers,
2131 * and this is also useful for vimages
2134 dn_cfg.slot_limit = 100; /* Foot shooting limit for queues. */
2135 dn_cfg.byte_limit = 1024 * 1024;
2138 /* RED parameters */
2139 dn_cfg.red_lookup_depth = 256; /* default lookup table depth */
2140 dn_cfg.red_avg_pkt_size = 512; /* default medium packet size */
2141 dn_cfg.red_max_pkt_size = 1500; /* default max packet size */
2144 dn_cfg.max_hash_size = 65536; /* max in the hash tables */
2145 dn_cfg.hash_size = 64; /* default hash size */
2147 /* create hash tables for schedulers and flowsets.
2148 * In both we search by key and by pointer.
2150 dn_cfg.schedhash = dn_ht_init(NULL, dn_cfg.hash_size,
2151 offsetof(struct dn_schk, schk_next),
2152 schk_hash, schk_match, schk_new);
2153 dn_cfg.fshash = dn_ht_init(NULL, dn_cfg.hash_size,
2154 offsetof(struct dn_fsk, fsk_next),
2155 fsk_hash, fsk_match, fsk_new);
2157 /* bucket index to drain object */
2158 dn_cfg.drain_fs = 0;
2159 dn_cfg.drain_sch = 0;
2161 heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
2162 SLIST_INIT(&dn_cfg.fsu);
2163 SLIST_INIT(&dn_cfg.schedlist);
2167 TASK_INIT(&dn_task, 0, dummynet_task, curvnet);
2168 dn_tq = taskqueue_create_fast("dummynet", M_WAITOK,
2169 taskqueue_thread_enqueue, &dn_tq);
2170 taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2172 callout_init(&dn_timeout, 1);
2175 /* Initialize curr_time adjustment mechanics. */
2176 getmicrouptime(&dn_cfg.prev_t);
2180 ip_dn_destroy(int last)
2182 callout_drain(&dn_timeout);
2186 ND("removing last instance\n");
2187 ip_dn_ctl_ptr = NULL;
2188 ip_dn_io_ptr = NULL;
2193 taskqueue_drain(dn_tq, &dn_task);
2194 taskqueue_free(dn_tq);
2196 dn_ht_free(dn_cfg.schedhash, 0);
2197 dn_ht_free(dn_cfg.fshash, 0);
2198 heap_free(&dn_cfg.evheap);
2204 dummynet_modevent(module_t mod, int type, void *data)
2207 if (type == MOD_LOAD) {
2209 printf("DUMMYNET already loaded\n");
2213 ip_dn_ctl_ptr = ip_dn_ctl;
2214 ip_dn_io_ptr = dummynet_io;
2216 } else if (type == MOD_UNLOAD) {
2217 ip_dn_destroy(1 /* last */);
2223 /* modevent helpers for the modules */
2225 load_dn_sched(struct dn_alg *d)
2230 return 1; /* error */
2231 ip_dn_init(); /* just in case, we need the lock */
2233 /* Check that mandatory funcs exists */
2234 if (d->enqueue == NULL || d->dequeue == NULL) {
2235 D("missing enqueue or dequeue for %s", d->name);
2239 /* Search if scheduler already exists */
2241 SLIST_FOREACH(s, &dn_cfg.schedlist, next) {
2242 if (strcmp(s->name, d->name) == 0) {
2243 D("%s already loaded", d->name);
2244 break; /* scheduler already exists */
2248 SLIST_INSERT_HEAD(&dn_cfg.schedlist, d, next);
2250 D("dn_sched %s %sloaded", d->name, s ? "not ":"");
2255 unload_dn_sched(struct dn_alg *s)
2257 struct dn_alg *tmp, *r;
2260 ND("called for %s", s->name);
2263 SLIST_FOREACH_SAFE(r, &dn_cfg.schedlist, next, tmp) {
2264 if (strcmp(s->name, r->name) != 0)
2266 ND("ref_count = %d", r->ref_count);
2267 err = (r->ref_count != 0) ? EBUSY : 0;
2269 SLIST_REMOVE(&dn_cfg.schedlist, r, dn_alg, next);
2273 D("dn_sched %s %sunloaded", s->name, err ? "not ":"");
2278 dn_sched_modevent(module_t mod, int cmd, void *arg)
2280 struct dn_alg *sch = arg;
2282 if (cmd == MOD_LOAD)
2283 return load_dn_sched(sch);
2284 else if (cmd == MOD_UNLOAD)
2285 return unload_dn_sched(sch);
2290 static moduledata_t dummynet_mod = {
2291 "dummynet", dummynet_modevent, NULL
2294 #define DN_SI_SUB SI_SUB_PROTO_IFATTACHDOMAIN
2295 #define DN_MODEV_ORD (SI_ORDER_ANY - 128) /* after ipfw */
2296 DECLARE_MODULE(dummynet, dummynet_mod, DN_SI_SUB, DN_MODEV_ORD);
2297 MODULE_DEPEND(dummynet, ipfw, 3, 3, 3);
2298 MODULE_VERSION(dummynet, 3);
2301 * Starting up. Done in order after dummynet_modevent() has been called.
2302 * VNET_SYSINIT is also called for each existing vnet and each new vnet.
2304 //VNET_SYSINIT(vnet_dn_init, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_init, NULL);
2307 * Shutdown handlers up shop. These are done in REVERSE ORDER, but still
2308 * after dummynet_modevent() has been called. Not called on reboot.
2309 * VNET_SYSUNINIT is also called for each exiting vnet as it exits.
2310 * or when the module is unloaded.
2312 //VNET_SYSUNINIT(vnet_dn_uninit, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_destroy, NULL);