2 * Copyright (c) 1998-2002,2010 Luigi Rizzo, Universita` di Pisa
3 * Portions Copyright (c) 2000 Akamba Corp.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
32 * Configuration and internal object management for dummynet.
35 #include "opt_inet6.h"
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/malloc.h>
41 #include <sys/kernel.h>
43 #include <sys/module.h>
46 #include <sys/rwlock.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
50 #include <sys/taskqueue.h>
51 #include <net/if.h> /* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */
52 #include <netinet/in.h>
53 #include <netinet/ip_var.h> /* ip_output(), IP_FORWARDING */
54 #include <netinet/ip_fw.h>
55 #include <netinet/ip_dummynet.h>
57 #include <netpfil/ipfw/ip_fw_private.h>
58 #include <netpfil/ipfw/dn_heap.h>
59 #include <netpfil/ipfw/ip_dn_private.h>
60 #include <netpfil/ipfw/dn_sched.h>
62 /* which objects to copy */
63 #define DN_C_LINK 0x01
65 #define DN_C_FLOW 0x04
67 #define DN_C_QUEUE 0x10
69 /* we use this argument in case of a schk_new */
75 /*---- callout hooks. ----*/
76 static struct callout dn_timeout;
77 static struct task dn_task;
78 static struct taskqueue *dn_tq = NULL;
84 (void)arg; /* UNUSED */
85 taskqueue_enqueue_fast(dn_tq, &dn_task);
92 callout_reset_sbt(&dn_timeout, tick_sbt, 0, dummynet, NULL,
93 C_HARDCLOCK | C_DIRECT_EXEC);
95 /*----- end of callout hooks -----*/
97 /* Return a scheduler descriptor given the type or name. */
98 static struct dn_alg *
99 find_sched_type(int type, char *name)
103 SLIST_FOREACH(d, &dn_cfg.schedlist, next) {
104 if (d->type == type || (name && !strcasecmp(d->name, name)))
107 return NULL; /* not found */
111 ipdn_bound_var(int *v, int dflt, int lo, int hi, const char *msg)
114 const char *op = NULL;
122 } else if (oldv > hi) {
128 printf("%s %s to %d (was %d)\n", op, msg, *v, oldv);
132 /*---- flow_id mask, hash and compare functions ---*/
134 * The flow_id includes the 5-tuple, the queue/pipe number
135 * which we store in the extra area in host order,
136 * and for ipv6 also the flow_id6.
137 * XXX see if we want the tos byte (can store in 'flags')
139 static struct ipfw_flow_id *
140 flow_id_mask(struct ipfw_flow_id *mask, struct ipfw_flow_id *id)
142 int is_v6 = IS_IP6_FLOW_ID(id);
144 id->dst_port &= mask->dst_port;
145 id->src_port &= mask->src_port;
146 id->proto &= mask->proto;
147 id->extra &= mask->extra;
149 APPLY_MASK(&id->dst_ip6, &mask->dst_ip6);
150 APPLY_MASK(&id->src_ip6, &mask->src_ip6);
151 id->flow_id6 &= mask->flow_id6;
153 id->dst_ip &= mask->dst_ip;
154 id->src_ip &= mask->src_ip;
159 /* computes an OR of two masks, result in dst and also returned */
160 static struct ipfw_flow_id *
161 flow_id_or(struct ipfw_flow_id *src, struct ipfw_flow_id *dst)
163 int is_v6 = IS_IP6_FLOW_ID(dst);
165 dst->dst_port |= src->dst_port;
166 dst->src_port |= src->src_port;
167 dst->proto |= src->proto;
168 dst->extra |= src->extra;
170 #define OR_MASK(_d, _s) \
171 (_d)->__u6_addr.__u6_addr32[0] |= (_s)->__u6_addr.__u6_addr32[0]; \
172 (_d)->__u6_addr.__u6_addr32[1] |= (_s)->__u6_addr.__u6_addr32[1]; \
173 (_d)->__u6_addr.__u6_addr32[2] |= (_s)->__u6_addr.__u6_addr32[2]; \
174 (_d)->__u6_addr.__u6_addr32[3] |= (_s)->__u6_addr.__u6_addr32[3];
175 OR_MASK(&dst->dst_ip6, &src->dst_ip6);
176 OR_MASK(&dst->src_ip6, &src->src_ip6);
178 dst->flow_id6 |= src->flow_id6;
180 dst->dst_ip |= src->dst_ip;
181 dst->src_ip |= src->src_ip;
187 nonzero_mask(struct ipfw_flow_id *m)
189 if (m->dst_port || m->src_port || m->proto || m->extra)
191 if (IS_IP6_FLOW_ID(m)) {
193 m->dst_ip6.__u6_addr.__u6_addr32[0] ||
194 m->dst_ip6.__u6_addr.__u6_addr32[1] ||
195 m->dst_ip6.__u6_addr.__u6_addr32[2] ||
196 m->dst_ip6.__u6_addr.__u6_addr32[3] ||
197 m->src_ip6.__u6_addr.__u6_addr32[0] ||
198 m->src_ip6.__u6_addr.__u6_addr32[1] ||
199 m->src_ip6.__u6_addr.__u6_addr32[2] ||
200 m->src_ip6.__u6_addr.__u6_addr32[3] ||
203 return m->dst_ip || m->src_ip;
207 /* XXX we may want a better hash function */
209 flow_id_hash(struct ipfw_flow_id *id)
213 if (IS_IP6_FLOW_ID(id)) {
214 uint32_t *d = (uint32_t *)&id->dst_ip6;
215 uint32_t *s = (uint32_t *)&id->src_ip6;
216 i = (d[0] ) ^ (d[1]) ^
218 (d[0] >> 15) ^ (d[1] >> 15) ^
219 (d[2] >> 15) ^ (d[3] >> 15) ^
220 (s[0] << 1) ^ (s[1] << 1) ^
221 (s[2] << 1) ^ (s[3] << 1) ^
222 (s[0] << 16) ^ (s[1] << 16) ^
223 (s[2] << 16) ^ (s[3] << 16) ^
224 (id->dst_port << 1) ^ (id->src_port) ^
226 (id->proto ) ^ (id->flow_id6);
228 i = (id->dst_ip) ^ (id->dst_ip >> 15) ^
229 (id->src_ip << 1) ^ (id->src_ip >> 16) ^
231 (id->dst_port << 1) ^ (id->src_port) ^ (id->proto);
236 /* Like bcmp, returns 0 if ids match, 1 otherwise. */
238 flow_id_cmp(struct ipfw_flow_id *id1, struct ipfw_flow_id *id2)
240 int is_v6 = IS_IP6_FLOW_ID(id1);
243 if (IS_IP6_FLOW_ID(id2))
244 return 1; /* different address families */
246 return (id1->dst_ip == id2->dst_ip &&
247 id1->src_ip == id2->src_ip &&
248 id1->dst_port == id2->dst_port &&
249 id1->src_port == id2->src_port &&
250 id1->proto == id2->proto &&
251 id1->extra == id2->extra) ? 0 : 1;
255 !bcmp(&id1->dst_ip6,&id2->dst_ip6, sizeof(id1->dst_ip6)) &&
256 !bcmp(&id1->src_ip6,&id2->src_ip6, sizeof(id1->src_ip6)) &&
257 id1->dst_port == id2->dst_port &&
258 id1->src_port == id2->src_port &&
259 id1->proto == id2->proto &&
260 id1->extra == id2->extra &&
261 id1->flow_id6 == id2->flow_id6) ? 0 : 1;
263 /*--------- end of flow-id mask, hash and compare ---------*/
265 /*--- support functions for the qht hashtable ----
266 * Entries are hashed by flow-id
269 q_hash(uintptr_t key, int flags, void *arg)
271 /* compute the hash slot from the flow id */
272 struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
273 &((struct dn_queue *)key)->ni.fid :
274 (struct ipfw_flow_id *)key;
276 return flow_id_hash(id);
280 q_match(void *obj, uintptr_t key, int flags, void *arg)
282 struct dn_queue *o = (struct dn_queue *)obj;
283 struct ipfw_flow_id *id2;
285 if (flags & DNHT_KEY_IS_OBJ) {
286 /* compare pointers */
287 id2 = &((struct dn_queue *)key)->ni.fid;
289 id2 = (struct ipfw_flow_id *)key;
291 return (0 == flow_id_cmp(&o->ni.fid, id2));
295 * create a new queue instance for the given 'key'.
298 q_new(uintptr_t key, int flags, void *arg)
300 struct dn_queue *q, *template = arg;
301 struct dn_fsk *fs = template->fs;
302 int size = sizeof(*q) + fs->sched->fp->q_datalen;
304 q = malloc(size, M_DUMMYNET, M_NOWAIT | M_ZERO);
306 D("no memory for new queue");
310 set_oid(&q->ni.oid, DN_QUEUE, size);
311 if (fs->fs.flags & DN_QHT_HASH)
312 q->ni.fid = *(struct ipfw_flow_id *)key;
314 q->_si = template->_si;
317 if (fs->sched->fp->new_queue)
318 fs->sched->fp->new_queue(q);
319 dn_cfg.queue_count++;
324 * Notify schedulers that a queue is going away.
325 * If (flags & DN_DESTROY), also free the packets.
326 * The version for callbacks is called q_delete_cb().
329 dn_delete_queue(struct dn_queue *q, int flags)
331 struct dn_fsk *fs = q->fs;
333 // D("fs %p si %p\n", fs, q->_si);
334 /* notify the parent scheduler that the queue is going away */
335 if (fs && fs->sched->fp->free_queue)
336 fs->sched->fp->free_queue(q);
339 if (flags & DN_DESTROY) {
341 dn_free_pkts(q->mq.head);
342 bzero(q, sizeof(*q)); // safety
344 dn_cfg.queue_count--;
349 q_delete_cb(void *q, void *arg)
351 int flags = (int)(uintptr_t)arg;
352 dn_delete_queue(q, flags);
353 return (flags & DN_DESTROY) ? DNHT_SCAN_DEL : 0;
357 * calls dn_delete_queue/q_delete_cb on all queues,
358 * which notifies the parent scheduler and possibly drains packets.
359 * flags & DN_DESTROY: drains queues and destroy qht;
362 qht_delete(struct dn_fsk *fs, int flags)
364 ND("fs %d start flags %d qht %p",
365 fs->fs.fs_nr, flags, fs->qht);
368 if (fs->fs.flags & DN_QHT_HASH) {
369 dn_ht_scan(fs->qht, q_delete_cb, (void *)(uintptr_t)flags);
370 if (flags & DN_DESTROY) {
371 dn_ht_free(fs->qht, 0);
375 dn_delete_queue((struct dn_queue *)(fs->qht), flags);
376 if (flags & DN_DESTROY)
382 * Find and possibly create the queue for a MULTIQUEUE scheduler.
383 * We never call it for !MULTIQUEUE (the queue is in the sch_inst).
386 ipdn_q_find(struct dn_fsk *fs, struct dn_sch_inst *si,
387 struct ipfw_flow_id *id)
389 struct dn_queue template;
394 if (fs->fs.flags & DN_QHT_HASH) {
395 struct ipfw_flow_id masked_id;
396 if (fs->qht == NULL) {
397 fs->qht = dn_ht_init(NULL, fs->fs.buckets,
398 offsetof(struct dn_queue, q_next),
399 q_hash, q_match, q_new);
404 flow_id_mask(&fs->fsk_mask, &masked_id);
405 return dn_ht_find(fs->qht, (uintptr_t)&masked_id,
406 DNHT_INSERT, &template);
409 fs->qht = q_new(0, 0, &template);
410 return (struct dn_queue *)fs->qht;
413 /*--- end of queue hash table ---*/
415 /*--- support functions for the sch_inst hashtable ----
417 * These are hashed by flow-id
420 si_hash(uintptr_t key, int flags, void *arg)
422 /* compute the hash slot from the flow id */
423 struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ?
424 &((struct dn_sch_inst *)key)->ni.fid :
425 (struct ipfw_flow_id *)key;
427 return flow_id_hash(id);
431 si_match(void *obj, uintptr_t key, int flags, void *arg)
433 struct dn_sch_inst *o = obj;
434 struct ipfw_flow_id *id2;
436 id2 = (flags & DNHT_KEY_IS_OBJ) ?
437 &((struct dn_sch_inst *)key)->ni.fid :
438 (struct ipfw_flow_id *)key;
439 return flow_id_cmp(&o->ni.fid, id2) == 0;
443 * create a new instance for the given 'key'
444 * Allocate memory for instance, delay line and scheduler private data.
447 si_new(uintptr_t key, int flags, void *arg)
449 struct dn_schk *s = arg;
450 struct dn_sch_inst *si;
451 int l = sizeof(*si) + s->fp->si_datalen;
453 si = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
457 /* Set length only for the part passed up to userland. */
458 set_oid(&si->ni.oid, DN_SCH_I, sizeof(struct dn_flow));
459 set_oid(&(si->dline.oid), DN_DELAY_LINE,
460 sizeof(struct delay_line));
461 /* mark si and dline as outside the event queue */
462 si->ni.oid.id = si->dline.oid.id = -1;
467 if (s->fp->new_sched && s->fp->new_sched(si)) {
468 D("new_sched error");
471 if (s->sch.flags & DN_HAVE_MASK)
472 si->ni.fid = *(struct ipfw_flow_id *)key;
479 bzero(si, sizeof(*si)); // safety
480 free(si, M_DUMMYNET);
486 * Callback from siht to delete all scheduler instances. Remove
487 * si and delay line from the system heap, destroy all queues.
488 * We assume that all flowset have been notified and do not
489 * point to us anymore.
492 si_destroy(void *_si, void *arg)
494 struct dn_sch_inst *si = _si;
495 struct dn_schk *s = si->sched;
496 struct delay_line *dl = &si->dline;
498 if (dl->oid.subtype) /* remove delay line from event heap */
499 heap_extract(&dn_cfg.evheap, dl);
500 dn_free_pkts(dl->mq.head); /* drain delay line */
501 if (si->kflags & DN_ACTIVE) /* remove si from event heap */
502 heap_extract(&dn_cfg.evheap, si);
503 if (s->fp->free_sched)
504 s->fp->free_sched(si);
505 bzero(si, sizeof(*si)); /* safety */
506 free(si, M_DUMMYNET);
508 return DNHT_SCAN_DEL;
512 * Find the scheduler instance for this packet. If we need to apply
513 * a mask, do on a local copy of the flow_id to preserve the original.
514 * Assume siht is always initialized if we have a mask.
517 ipdn_si_find(struct dn_schk *s, struct ipfw_flow_id *id)
520 if (s->sch.flags & DN_HAVE_MASK) {
521 struct ipfw_flow_id id_t = *id;
522 flow_id_mask(&s->sch.sched_mask, &id_t);
523 return dn_ht_find(s->siht, (uintptr_t)&id_t,
527 s->siht = si_new(0, 0, s);
528 return (struct dn_sch_inst *)s->siht;
531 /* callback to flush credit for the scheduler instance */
533 si_reset_credit(void *_si, void *arg)
535 struct dn_sch_inst *si = _si;
536 struct dn_link *p = &si->sched->link;
538 si->credit = p->burst + (dn_cfg.io_fast ? p->bandwidth : 0);
543 schk_reset_credit(struct dn_schk *s)
545 if (s->sch.flags & DN_HAVE_MASK)
546 dn_ht_scan(s->siht, si_reset_credit, NULL);
548 si_reset_credit(s->siht, NULL);
550 /*---- end of sch_inst hashtable ---------------------*/
552 /*-------------------------------------------------------
553 * flowset hash (fshash) support. Entries are hashed by fs_nr.
554 * New allocations are put in the fsunlinked list, from which
555 * they are removed when they point to a specific scheduler.
558 fsk_hash(uintptr_t key, int flags, void *arg)
560 uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
561 ((struct dn_fsk *)key)->fs.fs_nr;
563 return ( (i>>8)^(i>>4)^i );
567 fsk_match(void *obj, uintptr_t key, int flags, void *arg)
569 struct dn_fsk *fs = obj;
570 int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
571 ((struct dn_fsk *)key)->fs.fs_nr;
573 return (fs->fs.fs_nr == i);
577 fsk_new(uintptr_t key, int flags, void *arg)
581 fs = malloc(sizeof(*fs), M_DUMMYNET, M_NOWAIT | M_ZERO);
583 set_oid(&fs->fs.oid, DN_FS, sizeof(fs->fs));
585 fs->drain_bucket = 0;
586 SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
592 * detach flowset from its current scheduler. Flags as follows:
593 * DN_DETACH removes from the fsk_list
594 * DN_DESTROY deletes individual queues
595 * DN_DELETE_FS destroys the flowset (otherwise goes in unlinked).
598 fsk_detach(struct dn_fsk *fs, int flags)
600 if (flags & DN_DELETE_FS)
602 ND("fs %d from sched %d flags %s %s %s",
603 fs->fs.fs_nr, fs->fs.sched_nr,
604 (flags & DN_DELETE_FS) ? "DEL_FS":"",
605 (flags & DN_DESTROY) ? "DEL":"",
606 (flags & DN_DETACH) ? "DET":"");
607 if (flags & DN_DETACH) { /* detach from the list */
608 struct dn_fsk_head *h;
609 h = fs->sched ? &fs->sched->fsk_list : &dn_cfg.fsu;
610 SLIST_REMOVE(h, fs, dn_fsk, sch_chain);
612 /* Free the RED parameters, they will be recomputed on
613 * subsequent attach if needed.
616 free(fs->w_q_lookup, M_DUMMYNET);
617 fs->w_q_lookup = NULL;
618 qht_delete(fs, flags);
619 if (fs->sched && fs->sched->fp->free_fsk)
620 fs->sched->fp->free_fsk(fs);
622 if (flags & DN_DELETE_FS) {
623 bzero(fs, sizeof(*fs)); /* safety */
624 free(fs, M_DUMMYNET);
627 SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain);
632 * Detach or destroy all flowsets in a list.
633 * flags specifies what to do:
634 * DN_DESTROY: flush all queues
635 * DN_DELETE_FS: DN_DESTROY + destroy flowset
636 * DN_DELETE_FS implies DN_DESTROY
639 fsk_detach_list(struct dn_fsk_head *h, int flags)
642 int n = 0; /* only for stats */
644 ND("head %p flags %x", h, flags);
645 while ((fs = SLIST_FIRST(h))) {
646 SLIST_REMOVE_HEAD(h, sch_chain);
648 fsk_detach(fs, flags);
650 ND("done %d flowsets", n);
654 * called on 'queue X delete' -- removes the flowset from fshash,
655 * deletes all queues for the flowset, and removes the flowset.
658 delete_fs(int i, int locked)
665 fs = dn_ht_find(dn_cfg.fshash, i, DNHT_REMOVE, NULL);
666 ND("fs %d found %p", i, fs);
668 fsk_detach(fs, DN_DETACH | DN_DELETE_FS);
677 /*----- end of flowset hashtable support -------------*/
679 /*------------------------------------------------------------
680 * Scheduler hash. When searching by index we pass sched_nr,
681 * otherwise we pass struct dn_sch * which is the first field in
682 * struct dn_schk so we can cast between the two. We use this trick
683 * because in the create phase (but it should be fixed).
686 schk_hash(uintptr_t key, int flags, void *_arg)
688 uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key :
689 ((struct dn_schk *)key)->sch.sched_nr;
690 return ( (i>>8)^(i>>4)^i );
694 schk_match(void *obj, uintptr_t key, int flags, void *_arg)
696 struct dn_schk *s = (struct dn_schk *)obj;
697 int i = !(flags & DNHT_KEY_IS_OBJ) ? key :
698 ((struct dn_schk *)key)->sch.sched_nr;
699 return (s->sch.sched_nr == i);
703 * Create the entry and intialize with the sched hash if needed.
704 * Leave s->fp unset so we can tell whether a dn_ht_find() returns
705 * a new object or a previously existing one.
708 schk_new(uintptr_t key, int flags, void *arg)
710 struct schk_new_arg *a = arg;
712 int l = sizeof(*s) +a->fp->schk_datalen;
714 s = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO);
717 set_oid(&s->link.oid, DN_LINK, sizeof(s->link));
718 s->sch = *a->sch; // copy initial values
719 s->link.link_nr = s->sch.sched_nr;
720 SLIST_INIT(&s->fsk_list);
721 /* initialize the hash table or create the single instance */
722 s->fp = a->fp; /* si_new needs this */
724 if (s->sch.flags & DN_HAVE_MASK) {
725 s->siht = dn_ht_init(NULL, s->sch.buckets,
726 offsetof(struct dn_sch_inst, si_next),
727 si_hash, si_match, si_new);
728 if (s->siht == NULL) {
733 s->fp = NULL; /* mark as a new scheduler */
739 * Callback for sched delete. Notify all attached flowsets to
740 * detach from the scheduler, destroy the internal flowset, and
741 * all instances. The scheduler goes away too.
742 * arg is 0 (only detach flowsets and destroy instances)
743 * DN_DESTROY (detach & delete queues, delete schk)
744 * or DN_DELETE_FS (delete queues and flowsets, delete schk)
747 schk_delete_cb(void *obj, void *arg)
749 struct dn_schk *s = obj;
752 ND("sched %d arg %s%s",
754 a&DN_DESTROY ? "DEL ":"",
755 a&DN_DELETE_FS ? "DEL_FS":"");
757 fsk_detach_list(&s->fsk_list, arg ? DN_DESTROY : 0);
758 /* no more flowset pointing to us now */
759 if (s->sch.flags & DN_HAVE_MASK) {
760 dn_ht_scan(s->siht, si_destroy, NULL);
761 dn_ht_free(s->siht, 0);
763 si_destroy(s->siht, NULL);
765 free(s->profile, M_DUMMYNET);
771 bzero(s, sizeof(*s)); // safety
772 free(obj, M_DUMMYNET);
774 return DNHT_SCAN_DEL;
778 * called on a 'sched X delete' command. Deletes a single scheduler.
779 * This is done by removing from the schedhash, unlinking all
780 * flowsets and deleting their traffic.
787 s = dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
791 delete_fs(i + DN_MAX_ID, 1); /* first delete internal fs */
792 /* then detach flowsets, delete traffic */
793 schk_delete_cb(s, (void*)(uintptr_t)DN_DESTROY);
796 /*--- end of schk hashtable support ---*/
799 copy_obj(char **start, char *end, void *_o, const char *msg, int i)
801 struct dn_id *o = _o;
802 int have = end - *start;
804 if (have < o->len || o->len == 0 || o->type == 0) {
805 D("(WARN) type %d %s %d have %d need %d",
806 o->type, msg, i, have, o->len);
809 ND("type %d %s %d len %d", o->type, msg, i, o->len);
810 bcopy(_o, *start, o->len);
811 if (o->type == DN_LINK) {
812 /* Adjust burst parameter for link */
813 struct dn_link *l = (struct dn_link *)*start;
814 l->burst = div64(l->burst, 8 * hz);
815 l->delay = l->delay * 1000 / hz;
816 } else if (o->type == DN_SCH) {
817 /* Set id->id to the number of instances */
818 struct dn_schk *s = _o;
819 struct dn_id *id = (struct dn_id *)(*start);
820 id->id = (s->sch.flags & DN_HAVE_MASK) ?
821 dn_ht_entries(s->siht) : (s->siht ? 1 : 0);
827 /* Specific function to copy a queue.
828 * Copies only the user-visible part of a queue (which is in
829 * a struct dn_flow), and sets len accordingly.
832 copy_obj_q(char **start, char *end, void *_o, const char *msg, int i)
834 struct dn_id *o = _o;
835 int have = end - *start;
836 int len = sizeof(struct dn_flow); /* see above comment */
838 if (have < len || o->len == 0 || o->type != DN_QUEUE) {
839 D("ERROR type %d %s %d have %d need %d",
840 o->type, msg, i, have, len);
843 ND("type %d %s %d len %d", o->type, msg, i, len);
844 bcopy(_o, *start, len);
845 ((struct dn_id*)(*start))->len = len;
851 copy_q_cb(void *obj, void *arg)
853 struct dn_queue *q = obj;
854 struct copy_args *a = arg;
855 struct dn_flow *ni = (struct dn_flow *)(*a->start);
856 if (copy_obj_q(a->start, a->end, &q->ni, "queue", -1))
857 return DNHT_SCAN_END;
858 ni->oid.type = DN_FLOW; /* override the DN_QUEUE */
859 ni->oid.id = si_hash((uintptr_t)&ni->fid, 0, NULL);
864 copy_q(struct copy_args *a, struct dn_fsk *fs, int flags)
868 if (fs->fs.flags & DN_QHT_HASH)
869 dn_ht_scan(fs->qht, copy_q_cb, a);
871 copy_q_cb(fs->qht, a);
876 * This routine only copies the initial part of a profile ? XXX
879 copy_profile(struct copy_args *a, struct dn_profile *p)
881 int have = a->end - *a->start;
882 /* XXX here we check for max length */
883 int profile_len = sizeof(struct dn_profile) -
884 ED_MAX_SAMPLES_NO*sizeof(int);
888 if (have < profile_len) {
889 D("error have %d need %d", have, profile_len);
892 bcopy(p, *a->start, profile_len);
893 ((struct dn_id *)(*a->start))->len = profile_len;
894 *a->start += profile_len;
899 copy_flowset(struct copy_args *a, struct dn_fsk *fs, int flags)
901 struct dn_fs *ufs = (struct dn_fs *)(*a->start);
904 ND("flowset %d", fs->fs.fs_nr);
905 if (copy_obj(a->start, a->end, &fs->fs, "flowset", fs->fs.fs_nr))
906 return DNHT_SCAN_END;
907 ufs->oid.id = (fs->fs.flags & DN_QHT_HASH) ?
908 dn_ht_entries(fs->qht) : (fs->qht ? 1 : 0);
909 if (flags) { /* copy queues */
916 copy_si_cb(void *obj, void *arg)
918 struct dn_sch_inst *si = obj;
919 struct copy_args *a = arg;
920 struct dn_flow *ni = (struct dn_flow *)(*a->start);
921 if (copy_obj(a->start, a->end, &si->ni, "inst",
922 si->sched->sch.sched_nr))
923 return DNHT_SCAN_END;
924 ni->oid.type = DN_FLOW; /* override the DN_SCH_I */
925 ni->oid.id = si_hash((uintptr_t)si, DNHT_KEY_IS_OBJ, NULL);
930 copy_si(struct copy_args *a, struct dn_schk *s, int flags)
932 if (s->sch.flags & DN_HAVE_MASK)
933 dn_ht_scan(s->siht, copy_si_cb, a);
935 copy_si_cb(s->siht, a);
940 * compute a list of children of a scheduler and copy up
943 copy_fsk_list(struct copy_args *a, struct dn_schk *s, int flags)
949 int n = 0, space = sizeof(*o);
950 SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
951 if (fs->fs.fs_nr < DN_MAX_ID)
954 space += n * sizeof(uint32_t);
955 DX(3, "sched %d has %d flowsets", s->sch.sched_nr, n);
956 if (a->end - *(a->start) < space)
957 return DNHT_SCAN_END;
958 o = (struct dn_id *)(*(a->start));
962 p = (uint32_t *)(o+1);
963 SLIST_FOREACH(fs, &s->fsk_list, sch_chain)
964 if (fs->fs.fs_nr < DN_MAX_ID)
970 copy_data_helper(void *_o, void *_arg)
972 struct copy_args *a = _arg;
973 uint32_t *r = a->extra->r; /* start of first range */
974 uint32_t *lim; /* first invalid pointer */
977 lim = (uint32_t *)((char *)(a->extra) + a->extra->o.len);
979 if (a->type == DN_LINK || a->type == DN_SCH) {
980 /* pipe|sched show, we receive a dn_schk */
981 struct dn_schk *s = _o;
984 if (a->type == DN_SCH && n >= DN_MAX_ID)
985 return 0; /* not a scheduler */
986 if (a->type == DN_LINK && n <= DN_MAX_ID)
987 return 0; /* not a pipe */
989 /* see if the object is within one of our ranges */
990 for (;r < lim; r += 2) {
991 if (n < r[0] || n > r[1])
993 /* Found a valid entry, copy and we are done */
994 if (a->flags & DN_C_LINK) {
995 if (copy_obj(a->start, a->end,
996 &s->link, "link", n))
997 return DNHT_SCAN_END;
998 if (copy_profile(a, s->profile))
999 return DNHT_SCAN_END;
1000 if (copy_flowset(a, s->fs, 0))
1001 return DNHT_SCAN_END;
1003 if (a->flags & DN_C_SCH) {
1004 if (copy_obj(a->start, a->end,
1005 &s->sch, "sched", n))
1006 return DNHT_SCAN_END;
1007 /* list all attached flowsets */
1008 if (copy_fsk_list(a, s, 0))
1009 return DNHT_SCAN_END;
1011 if (a->flags & DN_C_FLOW)
1015 } else if (a->type == DN_FS) {
1016 /* queue show, skip internal flowsets */
1017 struct dn_fsk *fs = _o;
1022 /* see if the object is within one of our ranges */
1023 for (;r < lim; r += 2) {
1024 if (n < r[0] || n > r[1])
1026 if (copy_flowset(a, fs, 0))
1027 return DNHT_SCAN_END;
1029 break; /* we are done */
1035 static inline struct dn_schk *
1036 locate_scheduler(int i)
1038 return dn_ht_find(dn_cfg.schedhash, i, 0, NULL);
1042 * red parameters are in fixed point arithmetic.
1045 config_red(struct dn_fsk *fs)
1047 int64_t s, idle, weight, w0;
1050 fs->w_q = fs->fs.w_q;
1051 fs->max_p = fs->fs.max_p;
1053 /* Doing stuff that was in userland */
1054 i = fs->sched->link.bandwidth;
1056 hz * dn_cfg.red_avg_pkt_size * 8 * SCALE(1) / i;
1058 idle = div64((s * 3) , fs->w_q); /* s, fs->w_q scaled; idle not scaled */
1059 fs->lookup_step = div64(idle , dn_cfg.red_lookup_depth);
1060 /* fs->lookup_step not scaled, */
1061 if (!fs->lookup_step)
1062 fs->lookup_step = 1;
1063 w0 = weight = SCALE(1) - fs->w_q; //fs->w_q scaled
1065 for (t = fs->lookup_step; t > 1; --t)
1066 weight = SCALE_MUL(weight, w0);
1067 fs->lookup_weight = (int)(weight); // scaled
1069 /* Now doing stuff that was in kerneland */
1070 fs->min_th = SCALE(fs->fs.min_th);
1071 fs->max_th = SCALE(fs->fs.max_th);
1073 fs->c_1 = fs->max_p / (fs->fs.max_th - fs->fs.min_th);
1074 fs->c_2 = SCALE_MUL(fs->c_1, SCALE(fs->fs.min_th));
1076 if (fs->fs.flags & DN_IS_GENTLE_RED) {
1077 fs->c_3 = (SCALE(1) - fs->max_p) / fs->fs.max_th;
1078 fs->c_4 = SCALE(1) - 2 * fs->max_p;
1081 /* If the lookup table already exist, free and create it again. */
1082 if (fs->w_q_lookup) {
1083 free(fs->w_q_lookup, M_DUMMYNET);
1084 fs->w_q_lookup = NULL;
1086 if (dn_cfg.red_lookup_depth == 0) {
1087 printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1089 fs->fs.flags &= ~DN_IS_RED;
1090 fs->fs.flags &= ~DN_IS_GENTLE_RED;
1093 fs->lookup_depth = dn_cfg.red_lookup_depth;
1094 fs->w_q_lookup = (u_int *)malloc(fs->lookup_depth * sizeof(int),
1095 M_DUMMYNET, M_NOWAIT);
1096 if (fs->w_q_lookup == NULL) {
1097 printf("dummynet: sorry, cannot allocate red lookup table\n");
1098 fs->fs.flags &= ~DN_IS_RED;
1099 fs->fs.flags &= ~DN_IS_GENTLE_RED;
1103 /* Fill the lookup table with (1 - w_q)^x */
1104 fs->w_q_lookup[0] = SCALE(1) - fs->w_q;
1106 for (i = 1; i < fs->lookup_depth; i++)
1108 SCALE_MUL(fs->w_q_lookup[i - 1], fs->lookup_weight);
1110 if (dn_cfg.red_avg_pkt_size < 1)
1111 dn_cfg.red_avg_pkt_size = 512;
1112 fs->avg_pkt_size = dn_cfg.red_avg_pkt_size;
1113 if (dn_cfg.red_max_pkt_size < 1)
1114 dn_cfg.red_max_pkt_size = 1500;
1115 fs->max_pkt_size = dn_cfg.red_max_pkt_size;
1120 /* Scan all flowset attached to this scheduler and update red */
1122 update_red(struct dn_schk *s)
1125 SLIST_FOREACH(fs, &s->fsk_list, sch_chain) {
1126 if (fs && (fs->fs.flags & DN_IS_RED))
1131 /* attach flowset to scheduler s, possibly requeue */
1133 fsk_attach(struct dn_fsk *fs, struct dn_schk *s)
1135 ND("remove fs %d from fsunlinked, link to sched %d",
1136 fs->fs.fs_nr, s->sch.sched_nr);
1137 SLIST_REMOVE(&dn_cfg.fsu, fs, dn_fsk, sch_chain);
1139 SLIST_INSERT_HEAD(&s->fsk_list, fs, sch_chain);
1142 /* XXX compute fsk_mask */
1143 fs->fsk_mask = fs->fs.flow_mask;
1144 if (fs->sched->sch.flags & DN_HAVE_MASK)
1145 flow_id_or(&fs->sched->sch.sched_mask, &fs->fsk_mask);
1148 * we must drain qht according to the old
1149 * type, and reinsert according to the new one.
1150 * The requeue is complex -- in general we need to
1151 * reclassify every single packet.
1152 * For the time being, let's hope qht is never set
1153 * when we reach this point.
1155 D("XXX TODO requeue from fs %d to sch %d",
1156 fs->fs.fs_nr, s->sch.sched_nr);
1159 /* set the new type for qht */
1160 if (nonzero_mask(&fs->fsk_mask))
1161 fs->fs.flags |= DN_QHT_HASH;
1163 fs->fs.flags &= ~DN_QHT_HASH;
1165 /* XXX config_red() can fail... */
1166 if (fs->fs.flags & DN_IS_RED)
1170 /* update all flowsets which may refer to this scheduler */
1172 update_fs(struct dn_schk *s)
1174 struct dn_fsk *fs, *tmp;
1176 SLIST_FOREACH_SAFE(fs, &dn_cfg.fsu, sch_chain, tmp) {
1177 if (s->sch.sched_nr != fs->fs.sched_nr) {
1178 D("fs %d for sch %d not %d still unlinked",
1179 fs->fs.fs_nr, fs->fs.sched_nr,
1188 * Configuration -- to preserve backward compatibility we use
1189 * the following scheme (N is 65536)
1190 * NUMBER SCHED LINK FLOWSET
1191 * 1 .. N-1 (1)WFQ (2)WFQ (3)queue
1192 * N+1 .. 2N-1 (4)FIFO (5)FIFO (6)FIFO for sched 1..N-1
1193 * 2N+1 .. 3N-1 -- -- (7)FIFO for sched N+1..2N-1
1195 * "pipe i config" configures #1, #2 and #3
1196 * "sched i config" configures #1 and possibly #6
1197 * "queue i config" configures #3
1198 * #1 is configured with 'pipe i config' or 'sched i config'
1199 * #2 is configured with 'pipe i config', and created if not
1200 * existing with 'sched i config'
1201 * #3 is configured with 'queue i config'
1202 * #4 is automatically configured after #1, can only be FIFO
1203 * #5 is automatically configured after #2
1204 * #6 is automatically created when #1 is !MULTIQUEUE,
1205 * and can be updated.
1206 * #7 is automatically configured after #2
1210 * configure a link (and its FIFO instance)
1213 config_link(struct dn_link *p, struct dn_id *arg)
1217 if (p->oid.len != sizeof(*p)) {
1218 D("invalid pipe len %d", p->oid.len);
1222 if (i <= 0 || i >= DN_MAX_ID)
1225 * The config program passes parameters as follows:
1226 * bw = bits/second (0 means no limits),
1227 * delay = ms, must be translated into ticks.
1228 * qsize = slots/bytes
1231 p->delay = (p->delay * hz) / 1000;
1232 /* Scale burst size: bytes -> bits * hz */
1236 /* do it twice, base link and FIFO link */
1237 for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1238 struct dn_schk *s = locate_scheduler(i);
1241 D("sched %d not found", i);
1244 /* remove profile if exists */
1246 free(s->profile, M_DUMMYNET);
1249 /* copy all parameters */
1250 s->link.oid = p->oid;
1251 s->link.link_nr = i;
1252 s->link.delay = p->delay;
1253 if (s->link.bandwidth != p->bandwidth) {
1254 /* XXX bandwidth changes, need to update red params */
1255 s->link.bandwidth = p->bandwidth;
1258 s->link.burst = p->burst;
1259 schk_reset_credit(s);
1267 * configure a flowset. Can be called from inside with locked=1,
1269 static struct dn_fsk *
1270 config_fs(struct dn_fs *nfs, struct dn_id *arg, int locked)
1275 if (nfs->oid.len != sizeof(*nfs)) {
1276 D("invalid flowset len %d", nfs->oid.len);
1280 if (i <= 0 || i >= 3*DN_MAX_ID)
1282 ND("flowset %d", i);
1283 /* XXX other sanity checks */
1284 if (nfs->flags & DN_QSIZE_BYTES) {
1285 ipdn_bound_var(&nfs->qsize, 16384,
1286 1500, dn_cfg.byte_limit, NULL); // "queue byte size");
1288 ipdn_bound_var(&nfs->qsize, 50,
1289 1, dn_cfg.slot_limit, NULL); // "queue slot size");
1291 if (nfs->flags & DN_HAVE_MASK) {
1292 /* make sure we have some buckets */
1293 ipdn_bound_var((int *)&nfs->buckets, dn_cfg.hash_size,
1294 1, dn_cfg.max_hash_size, "flowset buckets");
1296 nfs->buckets = 1; /* we only need 1 */
1300 do { /* exit with break when done */
1302 int flags = nfs->sched_nr ? DNHT_INSERT : 0;
1304 int oldc = dn_cfg.fsk_count;
1305 fs = dn_ht_find(dn_cfg.fshash, i, flags, NULL);
1307 D("missing sched for flowset %d", i);
1310 /* grab some defaults from the existing one */
1311 if (nfs->sched_nr == 0) /* reuse */
1312 nfs->sched_nr = fs->fs.sched_nr;
1313 for (j = 0; j < sizeof(nfs->par)/sizeof(nfs->par[0]); j++) {
1314 if (nfs->par[j] == -1) /* reuse */
1315 nfs->par[j] = fs->fs.par[j];
1317 if (bcmp(&fs->fs, nfs, sizeof(*nfs)) == 0) {
1318 ND("flowset %d unchanged", i);
1319 break; /* no change, nothing to do */
1321 if (oldc != dn_cfg.fsk_count) /* new item */
1323 s = locate_scheduler(nfs->sched_nr);
1324 /* detach from old scheduler if needed, preserving
1325 * queues if we need to reattach. Then update the
1326 * configuration, and possibly attach to the new sched.
1328 DX(2, "fs %d changed sched %d@%p to %d@%p",
1330 fs->fs.sched_nr, fs->sched, nfs->sched_nr, s);
1332 int flags = s ? DN_DETACH : (DN_DETACH | DN_DESTROY);
1333 flags |= DN_DESTROY; /* XXX temporary */
1334 fsk_detach(fs, flags);
1336 fs->fs = *nfs; /* copy configuration */
1346 * config/reconfig a scheduler and its FIFO variant.
1347 * For !MULTIQUEUE schedulers, also set up the flowset.
1349 * On reconfigurations (detected because s->fp is set),
1350 * detach existing flowsets preserving traffic, preserve link,
1351 * and delete the old scheduler creating a new one.
1354 config_sched(struct dn_sch *_nsch, struct dn_id *arg)
1357 struct schk_new_arg a; /* argument for schk_new */
1359 struct dn_link p; /* copy of oldlink */
1360 struct dn_profile *pf = NULL; /* copy of old link profile */
1361 /* Used to preserv mask parameter */
1362 struct ipfw_flow_id new_mask;
1363 int new_buckets = 0;
1369 if (a.sch->oid.len != sizeof(*a.sch)) {
1370 D("bad sched len %d", a.sch->oid.len);
1373 i = a.sch->sched_nr;
1374 if (i <= 0 || i >= DN_MAX_ID)
1376 /* make sure we have some buckets */
1377 if (a.sch->flags & DN_HAVE_MASK)
1378 ipdn_bound_var((int *)&a.sch->buckets, dn_cfg.hash_size,
1379 1, dn_cfg.max_hash_size, "sched buckets");
1380 /* XXX other sanity checks */
1381 bzero(&p, sizeof(p));
1383 pipe_cmd = a.sch->flags & DN_PIPE_CMD;
1384 a.sch->flags &= ~DN_PIPE_CMD; //XXX do it even if is not set?
1386 /* Copy mask parameter */
1387 new_mask = a.sch->sched_mask;
1388 new_buckets = a.sch->buckets;
1389 new_flags = a.sch->flags;
1392 again: /* run twice, for wfq and fifo */
1394 * lookup the type. If not supplied, use the previous one
1395 * or default to WF2Q+. Otherwise, return an error.
1398 a.fp = find_sched_type(a.sch->oid.subtype, a.sch->name);
1400 /* found. Lookup or create entry */
1401 s = dn_ht_find(dn_cfg.schedhash, i, DNHT_INSERT, &a);
1402 } else if (a.sch->oid.subtype == 0 && !a.sch->name[0]) {
1403 /* No type. search existing s* or retry with WF2Q+ */
1404 s = dn_ht_find(dn_cfg.schedhash, i, 0, &a);
1407 /* Scheduler exists, skip to FIFO scheduler
1408 * if command was pipe config...
1413 /* New scheduler, create a wf2q+ with no mask
1414 * if command was pipe config...
1417 /* clear mask parameter */
1418 bzero(&a.sch->sched_mask, sizeof(new_mask));
1420 a.sch->flags &= ~DN_HAVE_MASK;
1422 a.sch->oid.subtype = DN_SCHED_WF2QP;
1426 D("invalid scheduler type %d %s",
1427 a.sch->oid.subtype, a.sch->name);
1431 /* normalize name and subtype */
1432 a.sch->oid.subtype = a.fp->type;
1433 bzero(a.sch->name, sizeof(a.sch->name));
1434 strlcpy(a.sch->name, a.fp->name, sizeof(a.sch->name));
1436 D("cannot allocate scheduler %d", i);
1439 /* restore existing link if any */
1442 if (!pf || pf->link_nr != p.link_nr) { /* no saved value */
1443 s->profile = NULL; /* XXX maybe not needed */
1445 s->profile = malloc(sizeof(struct dn_profile),
1446 M_DUMMYNET, M_NOWAIT | M_ZERO);
1447 if (s->profile == NULL) {
1448 D("cannot allocate profile");
1451 bcopy(pf, s->profile, sizeof(*pf));
1455 if (s->fp == NULL) {
1456 DX(2, "sched %d new type %s", i, a.fp->name);
1457 } else if (s->fp != a.fp ||
1458 bcmp(a.sch, &s->sch, sizeof(*a.sch)) ) {
1459 /* already existing. */
1460 DX(2, "sched %d type changed from %s to %s",
1461 i, s->fp->name, a.fp->name);
1462 DX(4, " type/sub %d/%d -> %d/%d",
1463 s->sch.oid.type, s->sch.oid.subtype,
1464 a.sch->oid.type, a.sch->oid.subtype);
1465 if (s->link.link_nr == 0)
1466 D("XXX WARNING link 0 for sched %d", i);
1467 p = s->link; /* preserve link */
1468 if (s->profile) {/* preserve profile */
1470 pf = malloc(sizeof(*pf),
1471 M_DUMMYNET, M_NOWAIT | M_ZERO);
1472 if (pf) /* XXX should issue a warning otherwise */
1473 bcopy(s->profile, pf, sizeof(*pf));
1475 /* remove from the hash */
1476 dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL);
1477 /* Detach flowsets, preserve queues. */
1478 // schk_delete_cb(s, NULL);
1479 // XXX temporarily, kill queues
1480 schk_delete_cb(s, (void *)DN_DESTROY);
1483 DX(4, "sched %d unchanged type %s", i, a.fp->name);
1485 /* complete initialization */
1489 // XXX schk_reset_credit(s);
1490 /* create the internal flowset if needed,
1491 * trying to reuse existing ones if available
1493 if (!(s->fp->flags & DN_MULTIQUEUE) && !s->fs) {
1494 s->fs = dn_ht_find(dn_cfg.fshash, i, 0, NULL);
1497 bzero(&fs, sizeof(fs));
1498 set_oid(&fs.oid, DN_FS, sizeof(fs));
1499 fs.fs_nr = i + DN_MAX_ID;
1501 s->fs = config_fs(&fs, NULL, 1 /* locked */);
1504 schk_delete_cb(s, (void *)DN_DESTROY);
1505 D("error creating internal fs for %d", i);
1509 /* call init function after the flowset is created */
1514 if (i < DN_MAX_ID) { /* now configure the FIFO instance */
1517 /* Restore mask parameter for FIFO */
1518 a.sch->sched_mask = new_mask;
1519 a.sch->buckets = new_buckets;
1520 a.sch->flags = new_flags;
1522 /* sched config shouldn't modify the FIFO scheduler */
1523 if (dn_ht_find(dn_cfg.schedhash, i, 0, &a) != NULL) {
1524 /* FIFO already exist, don't touch it */
1525 err = 0; /* and this is not an error */
1529 a.sch->sched_nr = i;
1530 a.sch->oid.subtype = DN_SCHED_FIFO;
1531 bzero(a.sch->name, sizeof(a.sch->name));
1538 free(pf, M_DUMMYNET);
1543 * attach a profile to a link
1546 config_profile(struct dn_profile *pf, struct dn_id *arg)
1549 int i, olen, err = 0;
1551 if (pf->oid.len < sizeof(*pf)) {
1552 D("short profile len %d", pf->oid.len);
1556 if (i <= 0 || i >= DN_MAX_ID)
1558 /* XXX other sanity checks */
1560 for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) {
1561 s = locate_scheduler(i);
1569 * If we had a profile and the new one does not fit,
1570 * or it is deleted, then we need to free memory.
1572 if (s->profile && (pf->samples_no == 0 ||
1573 s->profile->oid.len < pf->oid.len)) {
1574 free(s->profile, M_DUMMYNET);
1577 if (pf->samples_no == 0)
1580 * new profile, possibly allocate memory
1583 if (s->profile == NULL)
1584 s->profile = malloc(pf->oid.len,
1585 M_DUMMYNET, M_NOWAIT | M_ZERO);
1586 if (s->profile == NULL) {
1587 D("no memory for profile %d", i);
1591 /* preserve larger length XXX double check */
1592 olen = s->profile->oid.len;
1593 if (olen < pf->oid.len)
1595 bcopy(pf, s->profile, pf->oid.len);
1596 s->profile->oid.len = olen;
1603 * Delete all objects:
1606 dummynet_flush(void)
1609 /* delete all schedulers and related links/queues/flowsets */
1610 dn_ht_scan(dn_cfg.schedhash, schk_delete_cb,
1611 (void *)(uintptr_t)DN_DELETE_FS);
1612 /* delete all remaining (unlinked) flowsets */
1613 DX(4, "still %d unlinked fs", dn_cfg.fsk_count);
1614 dn_ht_free(dn_cfg.fshash, DNHT_REMOVE);
1615 fsk_detach_list(&dn_cfg.fsu, DN_DELETE_FS);
1616 /* Reinitialize system heap... */
1617 heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
1621 * Main handler for configuration. We are guaranteed to be called
1622 * with an oid which is at least a dn_id.
1623 * - the first object is the command (config, delete, flush, ...)
1624 * - config_link must be issued after the corresponding config_sched
1625 * - parameters (DN_TXT) for an object must preceed the object
1626 * processed on a config_sched.
1629 do_config(void *p, int l)
1631 struct dn_id *next, *o;
1632 int err = 0, err2 = 0;
1633 struct dn_id *arg = NULL;
1637 if (o->id != DN_API_VERSION) {
1638 D("invalid api version got %d need %d",
1639 o->id, DN_API_VERSION);
1642 for (; l >= sizeof(*o); o = next) {
1643 struct dn_id *prev = arg;
1644 if (o->len < sizeof(*o) || l < o->len) {
1645 D("bad len o->len %d len %d", o->len, l);
1650 next = (struct dn_id *)((char *)o + o->len);
1654 D("cmd %d not implemented", o->type);
1657 #ifdef EMULATE_SYSCTL
1658 /* sysctl emulation.
1659 * if we recognize the command, jump to the correct
1660 * handler and return
1663 err = kesysctl_emu_set(p, l);
1667 case DN_CMD_CONFIG: /* simply a header */
1671 /* the argument is in the first uintptr_t after o */
1672 a = (uintptr_t *)(o+1);
1673 if (o->len < sizeof(*o) + sizeof(*a)) {
1677 switch (o->subtype) {
1679 /* delete base and derived schedulers */
1681 err = delete_schk(*a);
1682 err2 = delete_schk(*a + DN_MAX_ID);
1689 D("invalid delete type %d",
1695 err = (*a <1 || *a >= DN_MAX_ID) ?
1696 EINVAL : delete_fs(*a, 0) ;
1706 case DN_TEXT: /* store argument the next block */
1711 err = config_link((struct dn_link *)o, arg);
1714 err = config_profile((struct dn_profile *)o, arg);
1717 err = config_sched((struct dn_sch *)o, arg);
1720 err = (NULL==config_fs((struct dn_fs *)o, arg, 0));
1732 compute_space(struct dn_id *cmd, struct copy_args *a)
1734 int x = 0, need = 0;
1735 int profile_size = sizeof(struct dn_profile) -
1736 ED_MAX_SAMPLES_NO*sizeof(int);
1738 /* NOTE about compute space:
1739 * NP = dn_cfg.schk_count
1740 * NSI = dn_cfg.si_count
1741 * NF = dn_cfg.fsk_count
1742 * NQ = dn_cfg.queue_count
1744 * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
1745 * link, scheduler template, flowset
1746 * integrated in scheduler and header
1748 * (NSI)*(dn_flow) all scheduler instance (includes
1749 * the queue instance)
1751 * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler
1752 * link, scheduler template, flowset
1753 * integrated in scheduler and header
1755 * (NSI * dn_flow) all scheduler instances
1756 * (NF * sizeof(uint_32)) space for flowset list linked to scheduler
1757 * (NQ * dn_queue) all queue [XXXfor now not listed]
1759 * (NF * dn_fs) all flowset
1760 * (NQ * dn_queue) all queues
1762 switch (cmd->subtype) {
1765 /* XXX where do LINK and SCH differ ? */
1766 /* 'ipfw sched show' could list all queues associated to
1767 * a scheduler. This feature for now is disabled
1769 case DN_LINK: /* pipe show */
1770 x = DN_C_LINK | DN_C_SCH | DN_C_FLOW;
1771 need += dn_cfg.schk_count *
1772 (sizeof(struct dn_fs) + profile_size) / 2;
1773 need += dn_cfg.fsk_count * sizeof(uint32_t);
1775 case DN_SCH: /* sched show */
1776 need += dn_cfg.schk_count *
1777 (sizeof(struct dn_fs) + profile_size) / 2;
1778 need += dn_cfg.fsk_count * sizeof(uint32_t);
1779 x = DN_C_SCH | DN_C_LINK | DN_C_FLOW;
1781 case DN_FS: /* queue show */
1782 x = DN_C_FS | DN_C_QUEUE;
1784 case DN_GET_COMPAT: /* compatibility mode */
1785 need = dn_compat_calc_size();
1790 need += dn_cfg.schk_count * sizeof(struct dn_sch) / 2;
1791 /* NOT also, each fs might be attached to a sched */
1792 need += dn_cfg.schk_count * sizeof(struct dn_id) / 2;
1795 need += dn_cfg.fsk_count * sizeof(struct dn_fs);
1796 if (x & DN_C_LINK) {
1797 need += dn_cfg.schk_count * sizeof(struct dn_link) / 2;
1800 * When exporting a queue to userland, only pass up the
1801 * struct dn_flow, which is the only visible part.
1805 need += dn_cfg.queue_count * sizeof(struct dn_flow);
1807 need += dn_cfg.si_count * (sizeof(struct dn_flow));
1812 * If compat != NULL dummynet_get is called in compatibility mode.
1813 * *compat will be the pointer to the buffer to pass to ipfw
1816 dummynet_get(struct sockopt *sopt, void **compat)
1818 int have, i, need, error;
1819 char *start = NULL, *buf;
1820 size_t sopt_valsize;
1823 struct copy_range r;
1824 int l = sizeof(struct dn_id);
1826 bzero(&a, sizeof(a));
1827 bzero(&r, sizeof(r));
1829 /* save and restore original sopt_valsize around copyin */
1830 sopt_valsize = sopt->sopt_valsize;
1835 /* copy at least an oid, and possibly a full object */
1836 error = sooptcopyin(sopt, cmd, sizeof(r), sizeof(*cmd));
1837 sopt->sopt_valsize = sopt_valsize;
1841 #ifdef EMULATE_SYSCTL
1842 /* sysctl emulation. */
1843 if (cmd->type == DN_SYSCTL_GET)
1844 return kesysctl_emu_get(sopt);
1846 if (l > sizeof(r)) {
1847 /* request larger than default, allocate buffer */
1848 cmd = malloc(l, M_DUMMYNET, M_WAITOK);
1849 error = sooptcopyin(sopt, cmd, l, l);
1850 sopt->sopt_valsize = sopt_valsize;
1854 } else { /* compatibility */
1856 cmd->type = DN_CMD_GET;
1857 cmd->len = sizeof(struct dn_id);
1858 cmd->subtype = DN_GET_COMPAT;
1859 // cmd->id = sopt_valsize;
1860 D("compatibility mode");
1862 a.extra = (struct copy_range *)cmd;
1863 if (cmd->len == sizeof(*cmd)) { /* no range, create a default */
1864 uint32_t *rp = (uint32_t *)(cmd + 1);
1865 cmd->len += 2* sizeof(uint32_t);
1867 rp[1] = DN_MAX_ID - 1;
1868 if (cmd->subtype == DN_LINK) {
1873 /* Count space (under lock) and allocate (outside lock).
1874 * Exit with lock held if we manage to get enough buffer.
1875 * Try a few times then give up.
1877 for (have = 0, i = 0; i < 10; i++) {
1879 need = compute_space(cmd, &a);
1881 /* if there is a range, ignore value from compute_space() */
1882 if (l > sizeof(*cmd))
1883 need = sopt_valsize - sizeof(*cmd);
1890 need += sizeof(*cmd);
1897 free(start, M_DUMMYNET);
1899 if (need > sopt_valsize)
1903 start = malloc(have, M_DUMMYNET, M_WAITOK | M_ZERO);
1906 if (start == NULL) {
1911 error = sooptcopyout(sopt, cmd, sizeof(*cmd));
1915 ND("have %d:%d sched %d, %d:%d links %d, %d:%d flowsets %d, "
1916 "%d:%d si %d, %d:%d queues %d",
1917 dn_cfg.schk_count, sizeof(struct dn_sch), DN_SCH,
1918 dn_cfg.schk_count, sizeof(struct dn_link), DN_LINK,
1919 dn_cfg.fsk_count, sizeof(struct dn_fs), DN_FS,
1920 dn_cfg.si_count, sizeof(struct dn_flow), DN_SCH_I,
1921 dn_cfg.queue_count, sizeof(struct dn_queue), DN_QUEUE);
1922 sopt->sopt_valsize = sopt_valsize;
1923 a.type = cmd->subtype;
1925 if (compat == NULL) {
1926 bcopy(cmd, start, sizeof(*cmd));
1927 ((struct dn_id*)(start))->len = sizeof(struct dn_id);
1928 buf = start + sizeof(*cmd);
1932 a.end = start + have;
1933 /* start copying other objects */
1935 a.type = DN_COMPAT_PIPE;
1936 dn_ht_scan(dn_cfg.schedhash, copy_data_helper_compat, &a);
1937 a.type = DN_COMPAT_QUEUE;
1938 dn_ht_scan(dn_cfg.fshash, copy_data_helper_compat, &a);
1939 } else if (a.type == DN_FS) {
1940 dn_ht_scan(dn_cfg.fshash, copy_data_helper, &a);
1942 dn_ht_scan(dn_cfg.schedhash, copy_data_helper, &a);
1948 sopt->sopt_valsize = buf - start;
1949 /* free() is done by ip_dummynet_compat() */
1950 start = NULL; //XXX hack
1952 error = sooptcopyout(sopt, start, buf - start);
1955 if (cmd && cmd != &r.o)
1956 free(cmd, M_DUMMYNET);
1958 free(start, M_DUMMYNET);
1962 /* Callback called on scheduler instance to delete it if idle */
1964 drain_scheduler_cb(void *_si, void *arg)
1966 struct dn_sch_inst *si = _si;
1968 if ((si->kflags & DN_ACTIVE) || si->dline.mq.head != NULL)
1971 if (si->sched->fp->flags & DN_MULTIQUEUE) {
1972 if (si->q_count == 0)
1973 return si_destroy(si, NULL);
1976 } else { /* !DN_MULTIQUEUE */
1977 if ((si+1)->ni.length == 0)
1978 return si_destroy(si, NULL);
1982 return 0; /* unreachable */
1985 /* Callback called on scheduler to check if it has instances */
1987 drain_scheduler_sch_cb(void *_s, void *arg)
1989 struct dn_schk *s = _s;
1991 if (s->sch.flags & DN_HAVE_MASK) {
1992 dn_ht_scan_bucket(s->siht, &s->drain_bucket,
1993 drain_scheduler_cb, NULL);
1997 if (drain_scheduler_cb(s->siht, NULL) == DNHT_SCAN_DEL)
2004 /* Called every tick, try to delete a 'bucket' of scheduler */
2006 dn_drain_scheduler(void)
2008 dn_ht_scan_bucket(dn_cfg.schedhash, &dn_cfg.drain_sch,
2009 drain_scheduler_sch_cb, NULL);
2013 /* Callback called on queue to delete if it is idle */
2015 drain_queue_cb(void *_q, void *arg)
2017 struct dn_queue *q = _q;
2019 if (q->ni.length == 0) {
2020 dn_delete_queue(q, DN_DESTROY);
2021 return DNHT_SCAN_DEL; /* queue is deleted */
2024 return 0; /* queue isn't deleted */
2027 /* Callback called on flowset used to check if it has queues */
2029 drain_queue_fs_cb(void *_fs, void *arg)
2031 struct dn_fsk *fs = _fs;
2033 if (fs->fs.flags & DN_QHT_HASH) {
2034 /* Flowset has a hash table for queues */
2035 dn_ht_scan_bucket(fs->qht, &fs->drain_bucket,
2036 drain_queue_cb, NULL);
2039 /* No hash table for this flowset, null the pointer
2040 * if the queue is deleted
2043 if (drain_queue_cb(fs->qht, NULL) == DNHT_SCAN_DEL)
2050 /* Called every tick, try to delete a 'bucket' of queue */
2052 dn_drain_queue(void)
2054 /* scan a bucket of flowset */
2055 dn_ht_scan_bucket(dn_cfg.fshash, &dn_cfg.drain_fs,
2056 drain_queue_fs_cb, NULL);
2061 * Handler for the various dummynet socket options
2064 ip_dn_ctl(struct sockopt *sopt)
2069 error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2073 /* Disallow sets in really-really secure mode. */
2074 if (sopt->sopt_dir == SOPT_SET) {
2075 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2080 switch (sopt->sopt_name) {
2082 D("dummynet: unknown option %d", sopt->sopt_name);
2086 case IP_DUMMYNET_FLUSH:
2087 case IP_DUMMYNET_CONFIGURE:
2088 case IP_DUMMYNET_DEL: /* remove a pipe or queue */
2089 case IP_DUMMYNET_GET:
2090 D("dummynet: compat option %d", sopt->sopt_name);
2091 error = ip_dummynet_compat(sopt);
2095 if (sopt->sopt_dir == SOPT_GET) {
2096 error = dummynet_get(sopt, NULL);
2099 l = sopt->sopt_valsize;
2100 if (l < sizeof(struct dn_id) || l > 12000) {
2101 D("argument len %d invalid", l);
2104 p = malloc(l, M_TEMP, M_WAITOK); // XXX can it fail ?
2105 error = sooptcopyin(sopt, p, l, l);
2108 error = do_config(p, l);
2122 if (dn_cfg.init_done)
2124 printf("DUMMYNET %p with IPv6 initialized (100409)\n", curvnet);
2125 dn_cfg.init_done = 1;
2126 /* Set defaults here. MSVC does not accept initializers,
2127 * and this is also useful for vimages
2130 dn_cfg.slot_limit = 100; /* Foot shooting limit for queues. */
2131 dn_cfg.byte_limit = 1024 * 1024;
2134 /* RED parameters */
2135 dn_cfg.red_lookup_depth = 256; /* default lookup table depth */
2136 dn_cfg.red_avg_pkt_size = 512; /* default medium packet size */
2137 dn_cfg.red_max_pkt_size = 1500; /* default max packet size */
2140 dn_cfg.max_hash_size = 65536; /* max in the hash tables */
2141 dn_cfg.hash_size = 64; /* default hash size */
2143 /* create hash tables for schedulers and flowsets.
2144 * In both we search by key and by pointer.
2146 dn_cfg.schedhash = dn_ht_init(NULL, dn_cfg.hash_size,
2147 offsetof(struct dn_schk, schk_next),
2148 schk_hash, schk_match, schk_new);
2149 dn_cfg.fshash = dn_ht_init(NULL, dn_cfg.hash_size,
2150 offsetof(struct dn_fsk, fsk_next),
2151 fsk_hash, fsk_match, fsk_new);
2153 /* bucket index to drain object */
2154 dn_cfg.drain_fs = 0;
2155 dn_cfg.drain_sch = 0;
2157 heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id));
2158 SLIST_INIT(&dn_cfg.fsu);
2159 SLIST_INIT(&dn_cfg.schedlist);
2163 TASK_INIT(&dn_task, 0, dummynet_task, curvnet);
2164 dn_tq = taskqueue_create_fast("dummynet", M_WAITOK,
2165 taskqueue_thread_enqueue, &dn_tq);
2166 taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2168 callout_init(&dn_timeout, CALLOUT_MPSAFE);
2171 /* Initialize curr_time adjustment mechanics. */
2172 getmicrouptime(&dn_cfg.prev_t);
2176 ip_dn_destroy(int last)
2178 callout_drain(&dn_timeout);
2182 ND("removing last instance\n");
2183 ip_dn_ctl_ptr = NULL;
2184 ip_dn_io_ptr = NULL;
2189 taskqueue_drain(dn_tq, &dn_task);
2190 taskqueue_free(dn_tq);
2192 dn_ht_free(dn_cfg.schedhash, 0);
2193 dn_ht_free(dn_cfg.fshash, 0);
2194 heap_free(&dn_cfg.evheap);
2200 dummynet_modevent(module_t mod, int type, void *data)
2203 if (type == MOD_LOAD) {
2205 printf("DUMMYNET already loaded\n");
2209 ip_dn_ctl_ptr = ip_dn_ctl;
2210 ip_dn_io_ptr = dummynet_io;
2212 } else if (type == MOD_UNLOAD) {
2213 ip_dn_destroy(1 /* last */);
2219 /* modevent helpers for the modules */
2221 load_dn_sched(struct dn_alg *d)
2226 return 1; /* error */
2227 ip_dn_init(); /* just in case, we need the lock */
2229 /* Check that mandatory funcs exists */
2230 if (d->enqueue == NULL || d->dequeue == NULL) {
2231 D("missing enqueue or dequeue for %s", d->name);
2235 /* Search if scheduler already exists */
2237 SLIST_FOREACH(s, &dn_cfg.schedlist, next) {
2238 if (strcmp(s->name, d->name) == 0) {
2239 D("%s already loaded", d->name);
2240 break; /* scheduler already exists */
2244 SLIST_INSERT_HEAD(&dn_cfg.schedlist, d, next);
2246 D("dn_sched %s %sloaded", d->name, s ? "not ":"");
2251 unload_dn_sched(struct dn_alg *s)
2253 struct dn_alg *tmp, *r;
2256 ND("called for %s", s->name);
2259 SLIST_FOREACH_SAFE(r, &dn_cfg.schedlist, next, tmp) {
2260 if (strcmp(s->name, r->name) != 0)
2262 ND("ref_count = %d", r->ref_count);
2263 err = (r->ref_count != 0) ? EBUSY : 0;
2265 SLIST_REMOVE(&dn_cfg.schedlist, r, dn_alg, next);
2269 D("dn_sched %s %sunloaded", s->name, err ? "not ":"");
2274 dn_sched_modevent(module_t mod, int cmd, void *arg)
2276 struct dn_alg *sch = arg;
2278 if (cmd == MOD_LOAD)
2279 return load_dn_sched(sch);
2280 else if (cmd == MOD_UNLOAD)
2281 return unload_dn_sched(sch);
2286 static moduledata_t dummynet_mod = {
2287 "dummynet", dummynet_modevent, NULL
2290 #define DN_SI_SUB SI_SUB_PROTO_IFATTACHDOMAIN
2291 #define DN_MODEV_ORD (SI_ORDER_ANY - 128) /* after ipfw */
2292 DECLARE_MODULE(dummynet, dummynet_mod, DN_SI_SUB, DN_MODEV_ORD);
2293 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
2294 MODULE_VERSION(dummynet, 3);
2297 * Starting up. Done in order after dummynet_modevent() has been called.
2298 * VNET_SYSINIT is also called for each existing vnet and each new vnet.
2300 //VNET_SYSINIT(vnet_dn_init, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_init, NULL);
2303 * Shutdown handlers up shop. These are done in REVERSE ORDER, but still
2304 * after dummynet_modevent() has been called. Not called on reboot.
2305 * VNET_SYSUNINIT is also called for each exiting vnet as it exits.
2306 * or when the module is unloaded.
2308 //VNET_SYSUNINIT(vnet_dn_uninit, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_destroy, NULL);