2 * Copyright (c) 1998-2002 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: head/sys/netinet/ipfw/ip_dummynet.c 200601 2009-12-16 10:48:40Z luigi $");
31 #define DUMMYNET_DEBUG
33 #include "opt_inet6.h"
36 * This module implements IP dummynet, a bandwidth limiter/delay emulator
37 * used in conjunction with the ipfw package.
38 * Description of the data structures used is in ip_dummynet.h
39 * Here you mainly find the following blocks of code:
40 * + variable declarations;
41 * + heap management functions;
42 * + scheduler and dummynet functions;
43 * + configuration and initialization.
45 * NOTA BENE: critical sections are protected by the "dummynet lock".
47 * Most important Changes:
50 * 010124: Fixed WF2Q behaviour
51 * 010122: Fixed spl protection.
52 * 000601: WF2Q support
53 * 000106: large rewrite, use heaps to handle very many pipes.
54 * 980513: initial release
56 * include files marked with XXX are probably not needed
59 #include <sys/param.h>
60 #include <sys/systm.h>
61 #include <sys/malloc.h>
63 #include <sys/kernel.h>
65 #include <sys/module.h>
68 #include <sys/rwlock.h>
69 #include <sys/socket.h>
70 #include <sys/socketvar.h>
72 #include <sys/sysctl.h>
73 #include <sys/taskqueue.h>
74 #include <net/if.h> /* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */
75 #include <net/netisr.h>
76 #include <netinet/in.h>
77 #include <netinet/ip.h> /* ip_len, ip_off */
78 #include <netinet/ip_fw.h>
79 #include <netinet/ipfw/ip_fw_private.h>
80 #include <netinet/ip_dummynet.h>
81 #include <netinet/ip_var.h> /* ip_output(), IP_FORWARDING */
83 #include <netinet/if_ether.h> /* various ether_* routines */
85 #include <netinet/ip6.h> /* for ip6_input, ip6_output prototypes */
86 #include <netinet6/ip6_var.h>
89 * We keep a private variable for the simulation time, but we could
90 * probably use an existing one ("softticks" in sys/kern/kern_timeout.c)
92 static dn_key curr_time = 0 ; /* current simulation time */
94 static int dn_hash_size = 64 ; /* default hash size */
96 /* statistics on number of queue searches and search steps */
97 static long searches, search_steps ;
98 static int pipe_expire = 1 ; /* expire queue if empty */
99 static int dn_max_ratio = 16 ; /* max queues/buckets ratio */
101 static long pipe_slot_limit = 100; /* Foot shooting limit for pipe queues. */
102 static long pipe_byte_limit = 1024 * 1024;
104 static int red_lookup_depth = 256; /* RED - default lookup table depth */
105 static int red_avg_pkt_size = 512; /* RED - default medium packet size */
106 static int red_max_pkt_size = 1500; /* RED - default max packet size */
108 static struct timeval prev_t, t;
109 static long tick_last; /* Last tick duration (usec). */
110 static long tick_delta; /* Last vs standard tick diff (usec). */
111 static long tick_delta_sum; /* Accumulated tick difference (usec).*/
112 static long tick_adjustment; /* Tick adjustments done. */
113 static long tick_lost; /* Lost(coalesced) ticks number. */
114 /* Adjusted vs non-adjusted curr_time difference (ticks). */
115 static long tick_diff;
118 static unsigned long io_pkt;
119 static unsigned long io_pkt_fast;
120 static unsigned long io_pkt_drop;
123 * Three heaps contain queues and pipes that the scheduler handles:
125 * ready_heap contains all dn_flow_queue related to fixed-rate pipes.
127 * wfq_ready_heap contains the pipes associated with WF2Q flows
129 * extract_heap contains pipes associated with delay lines.
133 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap");
135 static struct dn_heap ready_heap, extract_heap, wfq_ready_heap ;
137 static int heap_init(struct dn_heap *h, int size);
138 static int heap_insert (struct dn_heap *h, dn_key key1, void *p);
139 static void heap_extract(struct dn_heap *h, void *obj);
140 static void transmit_event(struct dn_pipe *pipe, struct mbuf **head,
142 static void ready_event(struct dn_flow_queue *q, struct mbuf **head,
144 static void ready_event_wfq(struct dn_pipe *p, struct mbuf **head,
148 #define HASH(num) ((((num) >> 8) ^ ((num) >> 4) ^ (num)) & 0x0f)
149 static struct dn_pipe_head pipehash[HASHSIZE]; /* all pipes */
150 static struct dn_flow_set_head flowsethash[HASHSIZE]; /* all flowsets */
152 static struct callout dn_timeout;
154 extern void (*bridge_dn_p)(struct mbuf *, struct ifnet *);
157 SYSCTL_DECL(_net_inet);
158 SYSCTL_DECL(_net_inet_ip);
160 SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet, CTLFLAG_RW, 0, "Dummynet");
161 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, hash_size,
162 CTLFLAG_RW, &dn_hash_size, 0, "Default hash table size");
163 #if 0 /* curr_time is 64 bit */
164 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, curr_time,
165 CTLFLAG_RD, &curr_time, 0, "Current tick");
167 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, ready_heap,
168 CTLFLAG_RD, &ready_heap.size, 0, "Size of ready heap");
169 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, extract_heap,
170 CTLFLAG_RD, &extract_heap.size, 0, "Size of extract heap");
171 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, searches,
172 CTLFLAG_RD, &searches, 0, "Number of queue searches");
173 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, search_steps,
174 CTLFLAG_RD, &search_steps, 0, "Number of queue search steps");
175 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, expire,
176 CTLFLAG_RW, &pipe_expire, 0, "Expire queue if empty");
177 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, max_chain_len,
178 CTLFLAG_RW, &dn_max_ratio, 0,
179 "Max ratio between dynamic queues and buckets");
180 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth,
181 CTLFLAG_RD, &red_lookup_depth, 0, "Depth of RED lookup table");
182 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size,
183 CTLFLAG_RD, &red_avg_pkt_size, 0, "RED Medium packet size");
184 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size,
185 CTLFLAG_RD, &red_max_pkt_size, 0, "RED Max packet size");
186 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta,
187 CTLFLAG_RD, &tick_delta, 0, "Last vs standard tick difference (usec).");
188 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta_sum,
189 CTLFLAG_RD, &tick_delta_sum, 0, "Accumulated tick difference (usec).");
190 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_adjustment,
191 CTLFLAG_RD, &tick_adjustment, 0, "Tick adjustments done.");
192 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_diff,
193 CTLFLAG_RD, &tick_diff, 0,
194 "Adjusted vs non-adjusted curr_time difference (ticks).");
195 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_lost,
196 CTLFLAG_RD, &tick_lost, 0,
197 "Number of ticks coalesced by dummynet taskqueue.");
198 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, io_fast,
199 CTLFLAG_RW, &io_fast, 0, "Enable fast dummynet io.");
200 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt,
201 CTLFLAG_RD, &io_pkt, 0,
202 "Number of packets passed to dummynet.");
203 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_fast,
204 CTLFLAG_RD, &io_pkt_fast, 0,
205 "Number of packets bypassed dummynet scheduler.");
206 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_drop,
207 CTLFLAG_RD, &io_pkt_drop, 0,
208 "Number of packets dropped by dummynet.");
209 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, pipe_slot_limit,
210 CTLFLAG_RW, &pipe_slot_limit, 0, "Upper limit in slots for pipe queue.");
211 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, pipe_byte_limit,
212 CTLFLAG_RW, &pipe_byte_limit, 0, "Upper limit in bytes for pipe queue.");
215 #ifdef DUMMYNET_DEBUG
216 int dummynet_debug = 0;
218 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, debug, CTLFLAG_RW, &dummynet_debug,
219 0, "control debugging printfs");
221 #define DPRINTF(X) if (dummynet_debug) printf X
226 static struct task dn_task;
227 static struct taskqueue *dn_tq = NULL;
228 static void dummynet_task(void *, int);
230 #if defined( __linux__ ) || defined( _WIN32 )
231 static DEFINE_SPINLOCK(dummynet_mtx);
233 static struct mtx dummynet_mtx;
235 #define DUMMYNET_LOCK_INIT() \
236 mtx_init(&dummynet_mtx, "dummynet", NULL, MTX_DEF)
237 #define DUMMYNET_LOCK_DESTROY() mtx_destroy(&dummynet_mtx)
238 #define DUMMYNET_LOCK() mtx_lock(&dummynet_mtx)
239 #define DUMMYNET_UNLOCK() mtx_unlock(&dummynet_mtx)
240 #define DUMMYNET_LOCK_ASSERT() mtx_assert(&dummynet_mtx, MA_OWNED)
242 static int config_pipe(struct dn_pipe *p);
243 static int ip_dn_ctl(struct sockopt *sopt);
245 static void dummynet(void *);
246 static void dummynet_flush(void);
247 static void dummynet_send(struct mbuf *);
248 void dummynet_drain(void);
249 static int dummynet_io(struct mbuf **, int , struct ip_fw_args *);
252 * Flow queue is idle if:
253 * 1) it's empty for at least 1 tick
254 * 2) it has invalid timestamp (WF2Q case)
255 * 3) parent pipe has no 'exhausted' burst.
257 #define QUEUE_IS_IDLE(q) ((q)->head == NULL && (q)->S == (q)->F + 1 && \
258 curr_time > (q)->idle_time + 1 && \
259 ((q)->numbytes + (curr_time - (q)->idle_time - 1) * \
260 (q)->fs->pipe->bandwidth >= (q)->fs->pipe->burst))
263 * Heap management functions.
265 * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2.
266 * Some macros help finding parent/children so we can optimize them.
268 * heap_init() is called to expand the heap when needed.
269 * Increment size in blocks of 16 entries.
270 * XXX failure to allocate a new element is a pretty bad failure
271 * as we basically stall a whole queue forever!!
272 * Returns 1 on error, 0 on success
274 #define HEAP_FATHER(x) ( ( (x) - 1 ) / 2 )
275 #define HEAP_LEFT(x) ( 2*(x) + 1 )
276 #define HEAP_IS_LEFT(x) ( (x) & 1 )
277 #define HEAP_RIGHT(x) ( 2*(x) + 2 )
278 #define HEAP_SWAP(a, b, buffer) { buffer = a ; a = b ; b = buffer ; }
279 #define HEAP_INCREMENT 15
282 heap_init(struct dn_heap *h, int new_size)
284 struct dn_heap_entry *p;
286 if (h->size >= new_size ) {
287 printf("dummynet: %s, Bogus call, have %d want %d\n", __func__,
291 new_size = (new_size + HEAP_INCREMENT ) & ~HEAP_INCREMENT ;
292 p = malloc(new_size * sizeof(*p), M_DUMMYNET, M_NOWAIT);
294 printf("dummynet: %s, resize %d failed\n", __func__, new_size );
295 return 1 ; /* error */
298 bcopy(h->p, p, h->size * sizeof(*p) );
299 free(h->p, M_DUMMYNET);
307 * Insert element in heap. Normally, p != NULL, we insert p in
308 * a new position and bubble up. If p == NULL, then the element is
309 * already in place, and key is the position where to start the
311 * Returns 1 on failure (cannot allocate new heap entry)
313 * If offset > 0 the position (index, int) of the element in the heap is
314 * also stored in the element itself at the given offset in bytes.
316 #define SET_OFFSET(heap, node) \
317 if (heap->offset > 0) \
318 *((int *)((char *)(heap->p[node].object) + heap->offset)) = node ;
320 * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value.
322 #define RESET_OFFSET(heap, node) \
323 if (heap->offset > 0) \
324 *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1 ;
326 heap_insert(struct dn_heap *h, dn_key key1, void *p)
328 int son = h->elements ;
330 if (p == NULL) /* data already there, set starting point */
332 else { /* insert new element at the end, possibly resize */
334 if (son == h->size) /* need resize... */
335 if (heap_init(h, h->elements+1) )
336 return 1 ; /* failure... */
337 h->p[son].object = p ;
338 h->p[son].key = key1 ;
341 while (son > 0) { /* bubble up */
342 int father = HEAP_FATHER(son) ;
343 struct dn_heap_entry tmp ;
345 if (DN_KEY_LT( h->p[father].key, h->p[son].key ) )
346 break ; /* found right position */
347 /* son smaller than father, swap and repeat */
348 HEAP_SWAP(h->p[son], h->p[father], tmp) ;
357 * remove top element from heap, or obj if obj != NULL
360 heap_extract(struct dn_heap *h, void *obj)
362 int child, father, max = h->elements - 1 ;
365 printf("dummynet: warning, extract from empty heap 0x%p\n", h);
368 father = 0 ; /* default: move up smallest child */
369 if (obj != NULL) { /* extract specific element, index is at offset */
371 panic("dummynet: heap_extract from middle not supported on this heap!!!\n");
372 father = *((int *)((char *)obj + h->offset)) ;
373 if (father < 0 || father >= h->elements) {
374 printf("dummynet: heap_extract, father %d out of bound 0..%d\n",
375 father, h->elements);
376 panic("dummynet: heap_extract");
379 RESET_OFFSET(h, father);
380 child = HEAP_LEFT(father) ; /* left child */
381 while (child <= max) { /* valid entry */
382 if (child != max && DN_KEY_LT(h->p[child+1].key, h->p[child].key) )
383 child = child+1 ; /* take right child, otherwise left */
384 h->p[father] = h->p[child] ;
385 SET_OFFSET(h, father);
387 child = HEAP_LEFT(child) ; /* left child for next loop */
392 * Fill hole with last entry and bubble up, reusing the insert code
394 h->p[father] = h->p[max] ;
395 heap_insert(h, father, NULL); /* this one cannot fail */
401 * change object position and update references
402 * XXX this one is never used!
405 heap_move(struct dn_heap *h, dn_key new_key, void *object)
409 int max = h->elements-1 ;
410 struct dn_heap_entry buf ;
413 panic("cannot move items on this heap");
415 i = *((int *)((char *)object + h->offset));
416 if (DN_KEY_LT(new_key, h->p[i].key) ) { /* must move up */
417 h->p[i].key = new_key ;
418 for (; i>0 && DN_KEY_LT(new_key, h->p[(temp = HEAP_FATHER(i))].key) ;
419 i = temp ) { /* bubble up */
420 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
423 } else { /* must move down */
424 h->p[i].key = new_key ;
425 while ( (temp = HEAP_LEFT(i)) <= max ) { /* found left child */
426 if ((temp != max) && DN_KEY_GT(h->p[temp].key, h->p[temp+1].key))
427 temp++ ; /* select child with min key */
428 if (DN_KEY_GT(new_key, h->p[temp].key)) { /* go down */
429 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
438 #endif /* heap_move, unused */
441 * heapify() will reorganize data inside an array to maintain the
442 * heap property. It is needed when we delete a bunch of entries.
445 heapify(struct dn_heap *h)
449 for (i = 0 ; i < h->elements ; i++ )
450 heap_insert(h, i , NULL) ;
454 * cleanup the heap and free data structure
457 heap_free(struct dn_heap *h)
460 free(h->p, M_DUMMYNET);
461 bzero(h, sizeof(*h) );
465 * --- end of heap management functions ---
469 * Dispose a list of packet. Use an inline functions so if we
470 * need to free extra state associated to a packet, this is a
471 * central point to do it.
474 static __inline void dn_free_pkts(struct mbuf *mnext)
478 while ((m = mnext) != NULL) {
479 mnext = m->m_nextpkt;
485 * Return the mbuf tag holding the dummynet state. As an optimization
486 * this is assumed to be the first tag on the list. If this turns out
487 * wrong we'll need to search the list.
489 static struct dn_pkt_tag *
490 dn_tag_get(struct mbuf *m)
492 struct m_tag *mtag = m_tag_first(m);
493 KASSERT(mtag != NULL &&
494 mtag->m_tag_cookie == MTAG_ABI_COMPAT &&
495 mtag->m_tag_id == PACKET_TAG_DUMMYNET,
496 ("packet on dummynet queue w/o dummynet tag!"));
497 return (struct dn_pkt_tag *)(mtag+1);
501 * Scheduler functions:
503 * transmit_event() is called when the delay-line needs to enter
504 * the scheduler, either because of existing pkts getting ready,
505 * or new packets entering the queue. The event handled is the delivery
506 * time of the packet.
508 * ready_event() does something similar with fixed-rate queues, and the
509 * event handled is the finish time of the head pkt.
511 * wfq_ready_event() does something similar with WF2Q queues, and the
512 * event handled is the start time of the head pkt.
514 * In all cases, we make sure that the data structures are consistent
515 * before passing pkts out, because this might trigger recursive
516 * invocations of the procedures.
519 transmit_event(struct dn_pipe *pipe, struct mbuf **head, struct mbuf **tail)
522 struct dn_pkt_tag *pkt;
524 DUMMYNET_LOCK_ASSERT();
526 while ((m = pipe->head) != NULL) {
528 if (!DN_KEY_LEQ(pkt->output_time, curr_time))
531 pipe->head = m->m_nextpkt;
533 (*tail)->m_nextpkt = m;
539 (*tail)->m_nextpkt = NULL;
541 /* If there are leftover packets, put into the heap for next event. */
542 if ((m = pipe->head) != NULL) {
545 * XXX Should check errors on heap_insert, by draining the
546 * whole pipe p and hoping in the future we are more successful.
548 heap_insert(&extract_heap, pkt->output_time, pipe);
553 #define div64(a, b) ((int64_t)(a) / (int64_t)(b))
556 * Compute how many ticks we have to wait before being able to send
557 * a packet. This is computed as the "wire time" for the packet
558 * (length + extra bits), minus the credit available, scaled to ticks.
559 * Check that the result is not be negative (it could be if we have
560 * too much leftover credit in q->numbytes).
563 set_ticks(struct mbuf *m, struct dn_flow_queue *q, struct dn_pipe *p)
567 ret = div64( (m->m_pkthdr.len * 8 + q->extra_bits) * hz
568 - q->numbytes + p->bandwidth - 1 , p->bandwidth);
575 * Convert the additional MAC overheads/delays into an equivalent
576 * number of bits for the given data rate. The samples are in milliseconds
577 * so we need to divide by 1000.
580 compute_extra_bits(struct mbuf *pkt, struct dn_pipe *p)
585 if (!p->samples || p->samples_no == 0)
587 index = random() % p->samples_no;
588 extra_bits = div64((dn_key)p->samples[index] * p->bandwidth, 1000);
589 if (index >= p->loss_level) {
590 struct dn_pkt_tag *dt = dn_tag_get(pkt);
592 dt->dn_dir = DIR_DROP;
598 free_pipe(struct dn_pipe *p)
601 free(p->samples, M_DUMMYNET);
606 * extract pkt from queue, compute output time (could be now)
607 * and put into delay line (p_queue)
610 move_pkt(struct mbuf *pkt, struct dn_flow_queue *q, struct dn_pipe *p,
613 struct dn_pkt_tag *dt = dn_tag_get(pkt);
615 q->head = pkt->m_nextpkt ;
617 q->len_bytes -= len ;
619 dt->output_time = curr_time + p->delay ;
624 p->tail->m_nextpkt = pkt;
626 p->tail->m_nextpkt = NULL;
630 * ready_event() is invoked every time the queue must enter the
631 * scheduler, either because the first packet arrives, or because
632 * a previously scheduled event fired.
633 * On invokation, drain as many pkts as possible (could be 0) and then
634 * if there are leftover packets reinsert the pkt in the scheduler.
637 ready_event(struct dn_flow_queue *q, struct mbuf **head, struct mbuf **tail)
640 struct dn_pipe *p = q->fs->pipe;
643 DUMMYNET_LOCK_ASSERT();
646 printf("dummynet: ready_event- pipe is gone\n");
649 p_was_empty = (p->head == NULL);
652 * Schedule fixed-rate queues linked to this pipe:
653 * account for the bw accumulated since last scheduling, then
654 * drain as many pkts as allowed by q->numbytes and move to
655 * the delay line (in p) computing output time.
656 * bandwidth==0 (no limit) means we can drain the whole queue,
657 * setting len_scaled = 0 does the job.
659 q->numbytes += (curr_time - q->sched_time) * p->bandwidth;
660 while ((pkt = q->head) != NULL) {
661 int len = pkt->m_pkthdr.len;
662 dn_key len_scaled = p->bandwidth ? len*8*hz
666 if (DN_KEY_GT(len_scaled, q->numbytes))
668 q->numbytes -= len_scaled;
669 move_pkt(pkt, q, p, len);
671 q->extra_bits = compute_extra_bits(q->head, p);
674 * If we have more packets queued, schedule next ready event
675 * (can only occur when bandwidth != 0, otherwise we would have
676 * flushed the whole queue in the previous loop).
677 * To this purpose we record the current time and compute how many
678 * ticks to go for the finish time of the packet.
680 if ((pkt = q->head) != NULL) { /* this implies bandwidth != 0 */
681 dn_key t = set_ticks(pkt, q, p); /* ticks i have to wait */
683 q->sched_time = curr_time;
684 heap_insert(&ready_heap, curr_time + t, (void *)q);
686 * XXX Should check errors on heap_insert, and drain the whole
687 * queue on error hoping next time we are luckier.
689 } else /* RED needs to know when the queue becomes empty. */
690 q->idle_time = curr_time;
693 * If the delay line was empty call transmit_event() now.
694 * Otherwise, the scheduler will take care of it.
697 transmit_event(p, head, tail);
701 * Called when we can transmit packets on WF2Q queues. Take pkts out of
702 * the queues at their start time, and enqueue into the delay line.
703 * Packets are drained until p->numbytes < 0. As long as
704 * len_scaled >= p->numbytes, the packet goes into the delay line
705 * with a deadline p->delay. For the last packet, if p->numbytes < 0,
706 * there is an additional delay.
709 ready_event_wfq(struct dn_pipe *p, struct mbuf **head, struct mbuf **tail)
711 int p_was_empty = (p->head == NULL);
712 struct dn_heap *sch = &(p->scheduler_heap);
713 struct dn_heap *neh = &(p->not_eligible_heap);
714 int64_t p_numbytes = p->numbytes;
717 * p->numbytes is only 32bits in FBSD7, but we might need 64 bits.
718 * Use a local variable for the computations, and write back the
719 * results when done, saturating if needed.
720 * The local variable has no impact on performance and helps
721 * reducing diffs between the various branches.
724 DUMMYNET_LOCK_ASSERT();
726 if (p->if_name[0] == 0) /* tx clock is simulated */
727 p_numbytes += (curr_time - p->sched_time) * p->bandwidth;
729 * tx clock is for real,
730 * the ifq must be empty or this is a NOP.
735 if (p->ifp && p->ifp->if_snd.ifq_head != NULL)
738 DPRINTF(("dummynet: pipe %d ready from %s --\n",
739 p->pipe_nr, p->if_name));
745 * While we have backlogged traffic AND credit, we need to do
746 * something on the queue.
748 while (p_numbytes >= 0 && (sch->elements > 0 || neh->elements > 0)) {
749 if (sch->elements > 0) {
750 /* Have some eligible pkts to send out. */
751 struct dn_flow_queue *q = sch->p[0].object;
752 struct mbuf *pkt = q->head;
753 struct dn_flow_set *fs = q->fs;
754 uint64_t len = pkt->m_pkthdr.len;
755 int len_scaled = p->bandwidth ? len * 8 * hz : 0;
757 heap_extract(sch, NULL); /* Remove queue from heap. */
758 p_numbytes -= len_scaled;
759 move_pkt(pkt, q, p, len);
761 p->V += div64((len << MY_M), p->sum); /* Update V. */
762 q->S = q->F; /* Update start time. */
764 /* Flow not backlogged any more. */
766 heap_insert(&(p->idle_heap), q->F, q);
768 /* Still backlogged. */
771 * Update F and position in backlogged queue,
772 * then put flow in not_eligible_heap
773 * (we will fix this later).
775 len = (q->head)->m_pkthdr.len;
776 q->F += div64((len << MY_M), fs->weight);
777 if (DN_KEY_LEQ(q->S, p->V))
778 heap_insert(neh, q->S, q);
780 heap_insert(sch, q->F, q);
784 * Now compute V = max(V, min(S_i)). Remember that all elements
785 * in sch have by definition S_i <= V so if sch is not empty,
786 * V is surely the max and we must not update it. Conversely,
787 * if sch is empty we only need to look at neh.
789 if (sch->elements == 0 && neh->elements > 0)
790 p->V = MAX64(p->V, neh->p[0].key);
791 /* Move from neh to sch any packets that have become eligible */
792 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V)) {
793 struct dn_flow_queue *q = neh->p[0].object;
794 heap_extract(neh, NULL);
795 heap_insert(sch, q->F, q);
798 if (p->if_name[0] != '\0') { /* Tx clock is from a real thing */
799 p_numbytes = -1; /* Mark not ready for I/O. */
803 if (sch->elements == 0 && neh->elements == 0 && p_numbytes >= 0) {
804 p->idle_time = curr_time;
806 * No traffic and no events scheduled.
807 * We can get rid of idle-heap.
809 if (p->idle_heap.elements > 0) {
812 for (i = 0; i < p->idle_heap.elements; i++) {
813 struct dn_flow_queue *q;
815 q = p->idle_heap.p[i].object;
821 p->idle_heap.elements = 0;
825 * If we are getting clocks from dummynet (not a real interface) and
826 * If we are under credit, schedule the next ready event.
827 * Also fix the delivery time of the last packet.
829 if (p->if_name[0]==0 && p_numbytes < 0) { /* This implies bw > 0. */
830 dn_key t = 0; /* Number of ticks i have to wait. */
832 if (p->bandwidth > 0)
833 t = div64(p->bandwidth - 1 - p_numbytes, p->bandwidth);
834 dn_tag_get(p->tail)->output_time += t;
835 p->sched_time = curr_time;
836 heap_insert(&wfq_ready_heap, curr_time + t, (void *)p);
838 * XXX Should check errors on heap_insert, and drain the whole
839 * queue on error hoping next time we are luckier.
843 /* Write back p_numbytes (adjust 64->32bit if necessary). */
844 p->numbytes = p_numbytes;
847 * If the delay line was empty call transmit_event() now.
848 * Otherwise, the scheduler will take care of it.
851 transmit_event(p, head, tail);
855 * This is called one tick, after previous run. It is used to
859 dummynet(void * __unused unused)
862 taskqueue_enqueue(dn_tq, &dn_task);
866 * The main dummynet processing function.
869 dummynet_task(void *context, int pending)
871 struct mbuf *head = NULL, *tail = NULL;
872 struct dn_pipe *pipe;
873 struct dn_heap *heaps[3];
875 void *p; /* generic parameter to handler */
880 heaps[0] = &ready_heap; /* fixed-rate queues */
881 heaps[1] = &wfq_ready_heap; /* wfq queues */
882 heaps[2] = &extract_heap; /* delay line */
884 /* Update number of lost(coalesced) ticks. */
885 tick_lost += pending - 1;
888 /* Last tick duration (usec). */
889 tick_last = (t.tv_sec - prev_t.tv_sec) * 1000000 +
890 (t.tv_usec - prev_t.tv_usec);
891 /* Last tick vs standard tick difference (usec). */
892 tick_delta = (tick_last * hz - 1000000) / hz;
893 /* Accumulated tick difference (usec). */
894 tick_delta_sum += tick_delta;
899 * Adjust curr_time if accumulated tick difference greater than
900 * 'standard' tick. Since curr_time should be monotonically increasing,
901 * we do positive adjustment as required and throttle curr_time in
902 * case of negative adjustment.
905 if (tick_delta_sum - tick >= 0) {
906 int diff = tick_delta_sum / tick;
910 tick_delta_sum %= tick;
912 } else if (tick_delta_sum + tick <= 0) {
915 tick_delta_sum += tick;
919 for (i = 0; i < 3; i++) {
921 while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time)) {
922 if (h->p[0].key > curr_time)
923 printf("dummynet: warning, "
924 "heap %d is %d ticks late\n",
925 i, (int)(curr_time - h->p[0].key));
926 /* store a copy before heap_extract */
928 /* need to extract before processing */
929 heap_extract(h, NULL);
931 ready_event(p, &head, &tail);
933 struct dn_pipe *pipe = p;
934 if (pipe->if_name[0] != '\0')
935 printf("dummynet: bad ready_event_wfq "
936 "for pipe %s\n", pipe->if_name);
938 ready_event_wfq(p, &head, &tail);
940 transmit_event(p, &head, &tail);
944 /* Sweep pipes trying to expire idle flow_queues. */
945 for (i = 0; i < HASHSIZE; i++) {
946 SLIST_FOREACH(pipe, &pipehash[i], next) {
947 if (pipe->idle_heap.elements > 0 &&
948 DN_KEY_LT(pipe->idle_heap.p[0].key, pipe->V)) {
949 struct dn_flow_queue *q =
950 pipe->idle_heap.p[0].object;
952 heap_extract(&(pipe->idle_heap), NULL);
953 /* Mark timestamp as invalid. */
955 pipe->sum -= q->fs->weight;
965 callout_reset(&dn_timeout, 1, dummynet, NULL);
969 dummynet_send(struct mbuf *m)
973 for (; m != NULL; m = n) {
974 struct ifnet *ifp = NULL;
980 tag = m_tag_first(m);
984 struct dn_pkt_tag *pkt = dn_tag_get(m);
985 /* extract the dummynet info, rename the tag */
988 /* rename the tag so it carries reinject info */
989 tag->m_tag_cookie = MTAG_IPFW_RULE;
995 SET_HOST_IPLEN(mtod(m, struct ip *));
996 ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
999 /* put header in network format for ip_input() */
1000 //SET_NET_IPLEN(mtod(m, struct ip *));
1001 netisr_dispatch(NETISR_IP, m);
1004 case DIR_IN | PROTO_IPV6:
1005 netisr_dispatch(NETISR_IPV6, m);
1008 case DIR_OUT | PROTO_IPV6:
1009 SET_HOST_IPLEN(mtod(m, struct ip *));
1010 ip6_output(m, NULL, NULL, IPV6_FORWARDING, NULL, NULL, NULL);
1013 case DIR_FWD | PROTO_IFB: /* DN_TO_IFB_FWD: */
1014 if (bridge_dn_p != NULL)
1015 ((*bridge_dn_p)(m, ifp));
1017 printf("dummynet: if_bridge not loaded\n");
1020 case DIR_IN | PROTO_LAYER2: /* DN_TO_ETH_DEMUX: */
1022 * The Ethernet code assumes the Ethernet header is
1023 * contiguous in the first mbuf header.
1024 * Insure this is true.
1026 if (m->m_len < ETHER_HDR_LEN &&
1027 (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) {
1028 printf("dummynet/ether: pullup failed, "
1029 "dropping packet\n");
1032 ether_demux(m->m_pkthdr.rcvif, m);
1034 case DIR_OUT | PROTO_LAYER2: /* N_TO_ETH_OUT: */
1035 ether_output_frame(ifp, m);
1039 /* drop the packet after some time */
1044 printf("dummynet: bad switch %d!\n", dst);
1052 * Unconditionally expire empty queues in case of shortage.
1053 * Returns the number of queues freed.
1056 expire_queues(struct dn_flow_set *fs)
1058 struct dn_flow_queue *q, *prev ;
1059 int i, initial_elements = fs->rq_elements ;
1061 if (fs->last_expired == time_uptime)
1063 fs->last_expired = time_uptime ;
1064 for (i = 0 ; i <= fs->rq_size ; i++) { /* last one is overflow */
1065 for (prev=NULL, q = fs->rq[i] ; q != NULL ; ) {
1066 if (!QUEUE_IS_IDLE(q)) {
1069 } else { /* entry is idle, expire it */
1070 struct dn_flow_queue *old_q = q ;
1073 prev->next = q = q->next ;
1075 fs->rq[i] = q = q->next ;
1077 free(old_q, M_DUMMYNET);
1081 return initial_elements - fs->rq_elements ;
1085 * If room, create a new queue and put at head of slot i;
1086 * otherwise, create or use the default queue.
1088 static struct dn_flow_queue *
1089 create_queue(struct dn_flow_set *fs, int i)
1091 struct dn_flow_queue *q;
1093 if (fs->rq_elements > fs->rq_size * dn_max_ratio &&
1094 expire_queues(fs) == 0) {
1095 /* No way to get room, use or create overflow queue. */
1097 if (fs->rq[i] != NULL)
1100 q = malloc(sizeof(*q), M_DUMMYNET, M_NOWAIT | M_ZERO);
1102 printf("dummynet: sorry, cannot allocate queue for new flow\n");
1107 q->next = fs->rq[i];
1108 q->S = q->F + 1; /* hack - mark timestamp as invalid. */
1109 q->numbytes = fs->pipe->burst + (io_fast ? fs->pipe->bandwidth : 0);
1116 * Given a flow_set and a pkt in last_pkt, find a matching queue
1117 * after appropriate masking. The queue is moved to front
1118 * so that further searches take less time.
1120 static struct dn_flow_queue *
1121 find_queue(struct dn_flow_set *fs, struct ipfw_flow_id *id)
1123 int i = 0 ; /* we need i and q for new allocations */
1124 struct dn_flow_queue *q, *prev;
1125 int is_v6 = IS_IP6_FLOW_ID(id);
1127 if ( !(fs->flags_fs & DN_HAVE_FLOW_MASK) )
1130 /* first, do the masking, then hash */
1131 id->dst_port &= fs->flow_mask.dst_port ;
1132 id->src_port &= fs->flow_mask.src_port ;
1133 id->proto &= fs->flow_mask.proto ;
1134 id->flags = 0 ; /* we don't care about this one */
1136 APPLY_MASK(&id->dst_ip6, &fs->flow_mask.dst_ip6);
1137 APPLY_MASK(&id->src_ip6, &fs->flow_mask.src_ip6);
1138 id->flow_id6 &= fs->flow_mask.flow_id6;
1140 i = ((id->dst_ip6.__u6_addr.__u6_addr32[0]) & 0xffff)^
1141 ((id->dst_ip6.__u6_addr.__u6_addr32[1]) & 0xffff)^
1142 ((id->dst_ip6.__u6_addr.__u6_addr32[2]) & 0xffff)^
1143 ((id->dst_ip6.__u6_addr.__u6_addr32[3]) & 0xffff)^
1145 ((id->dst_ip6.__u6_addr.__u6_addr32[0] >> 15) & 0xffff)^
1146 ((id->dst_ip6.__u6_addr.__u6_addr32[1] >> 15) & 0xffff)^
1147 ((id->dst_ip6.__u6_addr.__u6_addr32[2] >> 15) & 0xffff)^
1148 ((id->dst_ip6.__u6_addr.__u6_addr32[3] >> 15) & 0xffff)^
1150 ((id->src_ip6.__u6_addr.__u6_addr32[0] << 1) & 0xfffff)^
1151 ((id->src_ip6.__u6_addr.__u6_addr32[1] << 1) & 0xfffff)^
1152 ((id->src_ip6.__u6_addr.__u6_addr32[2] << 1) & 0xfffff)^
1153 ((id->src_ip6.__u6_addr.__u6_addr32[3] << 1) & 0xfffff)^
1155 ((id->src_ip6.__u6_addr.__u6_addr32[0] << 16) & 0xffff)^
1156 ((id->src_ip6.__u6_addr.__u6_addr32[1] << 16) & 0xffff)^
1157 ((id->src_ip6.__u6_addr.__u6_addr32[2] << 16) & 0xffff)^
1158 ((id->src_ip6.__u6_addr.__u6_addr32[3] << 16) & 0xffff)^
1160 (id->dst_port << 1) ^ (id->src_port) ^
1164 id->dst_ip &= fs->flow_mask.dst_ip ;
1165 id->src_ip &= fs->flow_mask.src_ip ;
1167 i = ( (id->dst_ip) & 0xffff ) ^
1168 ( (id->dst_ip >> 15) & 0xffff ) ^
1169 ( (id->src_ip << 1) & 0xffff ) ^
1170 ( (id->src_ip >> 16 ) & 0xffff ) ^
1171 (id->dst_port << 1) ^ (id->src_port) ^
1174 i = i % fs->rq_size ;
1175 /* finally, scan the current list for a match */
1177 for (prev=NULL, q = fs->rq[i] ; q ; ) {
1180 IN6_ARE_ADDR_EQUAL(&id->dst_ip6,&q->id.dst_ip6) &&
1181 IN6_ARE_ADDR_EQUAL(&id->src_ip6,&q->id.src_ip6) &&
1182 id->dst_port == q->id.dst_port &&
1183 id->src_port == q->id.src_port &&
1184 id->proto == q->id.proto &&
1185 id->flags == q->id.flags &&
1186 id->flow_id6 == q->id.flow_id6)
1189 if (!is_v6 && id->dst_ip == q->id.dst_ip &&
1190 id->src_ip == q->id.src_ip &&
1191 id->dst_port == q->id.dst_port &&
1192 id->src_port == q->id.src_port &&
1193 id->proto == q->id.proto &&
1194 id->flags == q->id.flags)
1197 /* No match. Check if we can expire the entry */
1198 if (pipe_expire && QUEUE_IS_IDLE(q)) {
1199 /* entry is idle and not in any heap, expire it */
1200 struct dn_flow_queue *old_q = q ;
1203 prev->next = q = q->next ;
1205 fs->rq[i] = q = q->next ;
1207 free(old_q, M_DUMMYNET);
1213 if (q && prev != NULL) { /* found and not in front */
1214 prev->next = q->next ;
1215 q->next = fs->rq[i] ;
1219 if (q == NULL) { /* no match, need to allocate a new entry */
1220 q = create_queue(fs, i);
1228 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len)
1233 * RED calculates the average queue size (avg) using a low-pass filter
1234 * with an exponential weighted (w_q) moving average:
1235 * avg <- (1-w_q) * avg + w_q * q_size
1236 * where q_size is the queue length (measured in bytes or * packets).
1238 * If q_size == 0, we compute the idle time for the link, and set
1239 * avg = (1 - w_q)^(idle/s)
1240 * where s is the time needed for transmitting a medium-sized packet.
1242 * Now, if avg < min_th the packet is enqueued.
1243 * If avg > max_th the packet is dropped. Otherwise, the packet is
1244 * dropped with probability P function of avg.
1249 /* Queue in bytes or packets? */
1250 u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ?
1251 q->len_bytes : q->len;
1253 DPRINTF(("\ndummynet: %d q: %2u ", (int)curr_time, q_size));
1255 /* Average queue size estimation. */
1257 /* Queue is not empty, avg <- avg + (q_size - avg) * w_q */
1258 int diff = SCALE(q_size) - q->avg;
1259 int64_t v = SCALE_MUL((int64_t)diff, (int64_t)fs->w_q);
1264 * Queue is empty, find for how long the queue has been
1265 * empty and use a lookup table for computing
1266 * (1 - * w_q)^(idle_time/s) where s is the time to send a
1268 * XXX check wraps...
1271 u_int t = div64(curr_time - q->idle_time,
1274 q->avg = (t < fs->lookup_depth) ?
1275 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0;
1278 DPRINTF(("dummynet: avg: %u ", SCALE_VAL(q->avg)));
1280 /* Should i drop? */
1281 if (q->avg < fs->min_th) {
1283 return (0); /* accept packet */
1285 if (q->avg >= fs->max_th) { /* average queue >= max threshold */
1286 if (fs->flags_fs & DN_IS_GENTLE_RED) {
1288 * According to Gentle-RED, if avg is greater than
1289 * max_th the packet is dropped with a probability
1290 * p_b = c_3 * avg - c_4
1291 * where c_3 = (1 - max_p) / max_th
1292 * c_4 = 1 - 2 * max_p
1294 p_b = SCALE_MUL((int64_t)fs->c_3, (int64_t)q->avg) -
1298 DPRINTF(("dummynet: - drop"));
1301 } else if (q->avg > fs->min_th) {
1303 * We compute p_b using the linear dropping function
1304 * p_b = c_1 * avg - c_2
1305 * where c_1 = max_p / (max_th - min_th)
1306 * c_2 = max_p * min_th / (max_th - min_th)
1308 p_b = SCALE_MUL((int64_t)fs->c_1, (int64_t)q->avg) - fs->c_2;
1311 if (fs->flags_fs & DN_QSIZE_IS_BYTES)
1312 p_b = div64(p_b * len, fs->max_pkt_size);
1313 if (++q->count == 0)
1314 q->random = random() & 0xffff;
1317 * q->count counts packets arrived since last drop, so a greater
1318 * value of q->count means a greater packet drop probability.
1320 if (SCALE_MUL(p_b, SCALE((int64_t)q->count)) > q->random) {
1322 DPRINTF(("dummynet: - red drop"));
1323 /* After a drop we calculate a new random value. */
1324 q->random = random() & 0xffff;
1325 return (1); /* drop */
1328 /* End of RED algorithm. */
1330 return (0); /* accept */
1333 static __inline struct dn_flow_set *
1334 locate_flowset(int fs_nr)
1336 struct dn_flow_set *fs;
1338 SLIST_FOREACH(fs, &flowsethash[HASH(fs_nr)], next)
1339 if (fs->fs_nr == fs_nr)
1345 static __inline struct dn_pipe *
1346 locate_pipe(int pipe_nr)
1348 struct dn_pipe *pipe;
1350 SLIST_FOREACH(pipe, &pipehash[HASH(pipe_nr)], next)
1351 if (pipe->pipe_nr == pipe_nr)
1358 * dummynet hook for packets. Below 'pipe' is a pipe or a queue
1359 * depending on whether WF2Q or fixed bw is used.
1361 * pipe_nr pipe or queue the packet is destined for.
1362 * dir where shall we send the packet after dummynet.
1363 * m the mbuf with the packet
1364 * ifp the 'ifp' parameter from the caller.
1365 * NULL in ip_input, destination interface in ip_output,
1366 * rule matching rule, in case of multiple passes
1369 dummynet_io(struct mbuf **m0, int dir, struct ip_fw_args *fwa)
1371 struct mbuf *m = *m0, *head = NULL, *tail = NULL;
1372 struct dn_pkt_tag *pkt;
1374 struct dn_flow_set *fs = NULL;
1375 struct dn_pipe *pipe;
1376 uint64_t len = m->m_pkthdr.len;
1377 struct dn_flow_queue *q = NULL;
1378 int is_pipe = fwa->rule.info & IPFW_IS_PIPE;
1380 KASSERT(m->m_nextpkt == NULL,
1381 ("dummynet_io: mbuf queue passed to dummynet"));
1386 * This is a dummynet rule, so we expect an O_PIPE or O_QUEUE rule.
1389 pipe = locate_pipe(fwa->rule.info & IPFW_INFO_MASK);
1393 fs = locate_flowset(fwa->rule.info & IPFW_INFO_MASK);
1396 goto dropit; /* This queue/pipe does not exist! */
1398 if (pipe == NULL) { /* Must be a queue, try find a matching pipe. */
1399 pipe = locate_pipe(fs->parent_nr);
1403 printf("dummynet: no pipe %d for queue %d, drop pkt\n",
1404 fs->parent_nr, fs->fs_nr);
1408 q = find_queue(fs, &(fwa->f_id));
1410 goto dropit; /* Cannot allocate queue. */
1412 /* Update statistics, then check reasons to drop pkt. */
1413 q->tot_bytes += len;
1415 if (fs->plr && random() < fs->plr)
1416 goto dropit; /* Random pkt drop. */
1417 if (fs->flags_fs & DN_QSIZE_IS_BYTES) {
1418 if (q->len_bytes > fs->qsize)
1419 goto dropit; /* Queue size overflow. */
1421 if (q->len >= fs->qsize)
1422 goto dropit; /* Queue count overflow. */
1424 if (fs->flags_fs & DN_IS_RED && red_drops(fs, q, len))
1427 /* XXX expensive to zero, see if we can remove it. */
1428 mtag = m_tag_get(PACKET_TAG_DUMMYNET,
1429 sizeof(struct dn_pkt_tag), M_NOWAIT | M_ZERO);
1431 goto dropit; /* Cannot allocate packet header. */
1432 m_tag_prepend(m, mtag); /* Attach to mbuf chain. */
1434 pkt = (struct dn_pkt_tag *)(mtag + 1);
1436 * Ok, i can handle the pkt now...
1437 * Build and enqueue packet + parameters.
1439 pkt->rule = fwa->rule;
1440 pkt->rule.info &= IPFW_ONEPASS; /* only keep this info */
1442 pkt->ifp = fwa->oif;
1444 if (q->head == NULL)
1447 q->tail->m_nextpkt = m;
1450 q->len_bytes += len;
1452 if (q->head != m) /* Flow was not idle, we are done. */
1455 if (is_pipe) { /* Fixed rate queues. */
1456 if (q->idle_time < curr_time) {
1457 /* Calculate available burst size. */
1459 (curr_time - q->idle_time - 1) * pipe->bandwidth;
1460 if (q->numbytes > pipe->burst)
1461 q->numbytes = pipe->burst;
1463 q->numbytes += pipe->bandwidth;
1465 } else { /* WF2Q. */
1466 if (pipe->idle_time < curr_time &&
1467 pipe->scheduler_heap.elements == 0 &&
1468 pipe->not_eligible_heap.elements == 0) {
1469 /* Calculate available burst size. */
1471 (curr_time - pipe->idle_time - 1) * pipe->bandwidth;
1472 if (pipe->numbytes > 0 && pipe->numbytes > pipe->burst)
1473 pipe->numbytes = pipe->burst;
1475 pipe->numbytes += pipe->bandwidth;
1477 pipe->idle_time = curr_time;
1479 /* Necessary for both: fixed rate & WF2Q queues. */
1480 q->idle_time = curr_time;
1483 * If we reach this point the flow was previously idle, so we need
1484 * to schedule it. This involves different actions for fixed-rate or
1488 /* Fixed-rate queue: just insert into the ready_heap. */
1491 if (pipe->bandwidth) {
1492 q->extra_bits = compute_extra_bits(m, pipe);
1493 t = set_ticks(m, q, pipe);
1495 q->sched_time = curr_time;
1496 if (t == 0) /* Must process it now. */
1497 ready_event(q, &head, &tail);
1499 heap_insert(&ready_heap, curr_time + t , q);
1502 * WF2Q. First, compute start time S: if the flow was
1503 * idle (S = F + 1) set S to the virtual time V for the
1504 * controlling pipe, and update the sum of weights for the pipe;
1505 * otherwise, remove flow from idle_heap and set S to max(F,V).
1506 * Second, compute finish time F = S + len / weight.
1507 * Third, if pipe was idle, update V = max(S, V).
1508 * Fourth, count one more backlogged flow.
1510 if (DN_KEY_GT(q->S, q->F)) { /* Means timestamps are invalid. */
1512 pipe->sum += fs->weight; /* Add weight of new queue. */
1514 heap_extract(&(pipe->idle_heap), q);
1515 q->S = MAX64(q->F, pipe->V);
1517 q->F = q->S + div64(len << MY_M, fs->weight);
1519 if (pipe->not_eligible_heap.elements == 0 &&
1520 pipe->scheduler_heap.elements == 0)
1521 pipe->V = MAX64(q->S, pipe->V);
1524 * Look at eligibility. A flow is not eligibile if S>V (when
1525 * this happens, it means that there is some other flow already
1526 * scheduled for the same pipe, so the scheduler_heap cannot be
1527 * empty). If the flow is not eligible we just store it in the
1528 * not_eligible_heap. Otherwise, we store in the scheduler_heap
1529 * and possibly invoke ready_event_wfq() right now if there is
1531 * Note that for all flows in scheduler_heap (SCH), S_i <= V,
1532 * and for all flows in not_eligible_heap (NEH), S_i > V.
1533 * So when we need to compute max(V, min(S_i)) forall i in
1534 * SCH+NEH, we only need to look into NEH.
1536 if (DN_KEY_GT(q->S, pipe->V)) { /* Not eligible. */
1537 if (pipe->scheduler_heap.elements == 0)
1538 printf("dummynet: ++ ouch! not eligible but empty scheduler!\n");
1539 heap_insert(&(pipe->not_eligible_heap), q->S, q);
1541 heap_insert(&(pipe->scheduler_heap), q->F, q);
1542 if (pipe->numbytes >= 0) { /* Pipe is idle. */
1543 if (pipe->scheduler_heap.elements != 1)
1544 printf("dummynet: OUCH! pipe should have been idle!\n");
1545 DPRINTF(("dummynet: waking up pipe %d at %d\n",
1546 pipe->pipe_nr, (int)(q->F >> MY_M)));
1547 pipe->sched_time = curr_time;
1548 ready_event_wfq(pipe, &head, &tail);
1553 if (head == m && (dir & PROTO_LAYER2) == 0 ) {
1556 if (m->m_nextpkt != NULL)
1557 printf("dummynet: fast io: pkt chain detected!\n");
1558 head = m->m_nextpkt = NULL;
1560 *m0 = NULL; /* Normal io. */
1564 dummynet_send(head);
1574 return ((fs && (fs->flags_fs & DN_NOERROR)) ? 0 : ENOBUFS);
1578 * Dispose all packets and flow_queues on a flow_set.
1579 * If all=1, also remove red lookup table and other storage,
1580 * including the descriptor itself.
1581 * For the one in dn_pipe MUST also cleanup ready_heap...
1584 purge_flow_set(struct dn_flow_set *fs, int all)
1586 struct dn_flow_queue *q, *qn;
1589 DUMMYNET_LOCK_ASSERT();
1591 for (i = 0; i <= fs->rq_size; i++) {
1592 for (q = fs->rq[i]; q != NULL; q = qn) {
1593 dn_free_pkts(q->head);
1595 free(q, M_DUMMYNET);
1600 fs->rq_elements = 0;
1602 /* RED - free lookup table. */
1603 if (fs->w_q_lookup != NULL)
1604 free(fs->w_q_lookup, M_DUMMYNET);
1606 free(fs->rq, M_DUMMYNET);
1607 /* If this fs is not part of a pipe, free it. */
1608 if (fs->pipe == NULL || fs != &(fs->pipe->fs))
1609 free(fs, M_DUMMYNET);
1614 * Dispose all packets queued on a pipe (not a flow_set).
1615 * Also free all resources associated to a pipe, which is about
1619 purge_pipe(struct dn_pipe *pipe)
1622 purge_flow_set( &(pipe->fs), 1 );
1624 dn_free_pkts(pipe->head);
1626 heap_free( &(pipe->scheduler_heap) );
1627 heap_free( &(pipe->not_eligible_heap) );
1628 heap_free( &(pipe->idle_heap) );
1632 * Delete all pipes and heaps returning memory. Must also
1633 * remove references from all ipfw rules to all pipes.
1636 dummynet_flush(void)
1638 struct dn_pipe *pipe, *pipe1;
1639 struct dn_flow_set *fs, *fs1;
1643 /* Free heaps so we don't have unwanted events. */
1644 heap_free(&ready_heap);
1645 heap_free(&wfq_ready_heap);
1646 heap_free(&extract_heap);
1649 * Now purge all queued pkts and delete all pipes.
1651 * XXXGL: can we merge the for(;;) cycles into one or not?
1653 for (i = 0; i < HASHSIZE; i++)
1654 SLIST_FOREACH_SAFE(fs, &flowsethash[i], next, fs1) {
1655 SLIST_REMOVE(&flowsethash[i], fs, dn_flow_set, next);
1656 purge_flow_set(fs, 1);
1658 for (i = 0; i < HASHSIZE; i++)
1659 SLIST_FOREACH_SAFE(pipe, &pipehash[i], next, pipe1) {
1660 SLIST_REMOVE(&pipehash[i], pipe, dn_pipe, next);
1668 * setup RED parameters
1671 config_red(struct dn_flow_set *p, struct dn_flow_set *x)
1676 x->min_th = SCALE(p->min_th);
1677 x->max_th = SCALE(p->max_th);
1678 x->max_p = p->max_p;
1680 x->c_1 = p->max_p / (p->max_th - p->min_th);
1681 x->c_2 = SCALE_MUL(x->c_1, SCALE(p->min_th));
1683 if (x->flags_fs & DN_IS_GENTLE_RED) {
1684 x->c_3 = (SCALE(1) - p->max_p) / p->max_th;
1685 x->c_4 = SCALE(1) - 2 * p->max_p;
1688 /* If the lookup table already exist, free and create it again. */
1689 if (x->w_q_lookup) {
1690 free(x->w_q_lookup, M_DUMMYNET);
1691 x->w_q_lookup = NULL;
1693 if (red_lookup_depth == 0) {
1694 printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1696 free(x, M_DUMMYNET);
1699 x->lookup_depth = red_lookup_depth;
1700 x->w_q_lookup = (u_int *)malloc(x->lookup_depth * sizeof(int),
1701 M_DUMMYNET, M_NOWAIT);
1702 if (x->w_q_lookup == NULL) {
1703 printf("dummynet: sorry, cannot allocate red lookup table\n");
1704 free(x, M_DUMMYNET);
1708 /* Fill the lookup table with (1 - w_q)^x */
1709 x->lookup_step = p->lookup_step;
1710 x->lookup_weight = p->lookup_weight;
1711 x->w_q_lookup[0] = SCALE(1) - x->w_q;
1713 for (i = 1; i < x->lookup_depth; i++)
1715 SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight);
1717 if (red_avg_pkt_size < 1)
1718 red_avg_pkt_size = 512;
1719 x->avg_pkt_size = red_avg_pkt_size;
1720 if (red_max_pkt_size < 1)
1721 red_max_pkt_size = 1500;
1722 x->max_pkt_size = red_max_pkt_size;
1727 alloc_hash(struct dn_flow_set *x, struct dn_flow_set *pfs)
1729 if (x->flags_fs & DN_HAVE_FLOW_MASK) { /* allocate some slots */
1730 int l = pfs->rq_size;
1736 else if (l > DN_MAX_HASH_SIZE)
1737 l = DN_MAX_HASH_SIZE;
1739 } else /* one is enough for null mask */
1741 x->rq = malloc((1 + x->rq_size) * sizeof(struct dn_flow_queue *),
1742 M_DUMMYNET, M_NOWAIT | M_ZERO);
1743 if (x->rq == NULL) {
1744 printf("dummynet: sorry, cannot allocate queue\n");
1752 set_fs_parms(struct dn_flow_set *x, struct dn_flow_set *src)
1754 x->flags_fs = src->flags_fs;
1755 x->qsize = src->qsize;
1757 x->flow_mask = src->flow_mask;
1758 if (x->flags_fs & DN_QSIZE_IS_BYTES) {
1759 if (x->qsize > pipe_byte_limit)
1760 x->qsize = 1024 * 1024;
1764 if (x->qsize > pipe_slot_limit)
1767 /* Configuring RED. */
1768 if (x->flags_fs & DN_IS_RED)
1769 config_red(src, x); /* XXX should check errors */
1773 * Setup pipe or queue parameters.
1776 config_pipe(struct dn_pipe *p)
1778 struct dn_flow_set *pfs = &(p->fs);
1779 struct dn_flow_queue *q;
1783 * The config program passes parameters as follows:
1784 * bw = bits/second (0 means no limits),
1785 * delay = ms, must be translated into ticks.
1786 * qsize = slots/bytes
1788 p->delay = (p->delay * hz) / 1000;
1789 /* Scale burst size: bytes -> bits * hz */
1791 /* We need either a pipe number or a flow_set number. */
1792 if (p->pipe_nr == 0 && pfs->fs_nr == 0)
1794 if (p->pipe_nr != 0 && pfs->fs_nr != 0)
1796 if (p->pipe_nr != 0) { /* this is a pipe */
1797 struct dn_pipe *pipe;
1800 pipe = locate_pipe(p->pipe_nr); /* locate pipe */
1802 if (pipe == NULL) { /* new pipe */
1803 pipe = malloc(sizeof(struct dn_pipe), M_DUMMYNET,
1807 printf("dummynet: no memory for new pipe\n");
1810 pipe->pipe_nr = p->pipe_nr;
1811 pipe->fs.pipe = pipe;
1813 * idle_heap is the only one from which
1814 * we extract from the middle.
1816 pipe->idle_heap.size = pipe->idle_heap.elements = 0;
1817 pipe->idle_heap.offset =
1818 offsetof(struct dn_flow_queue, heap_pos);
1820 /* Flush accumulated credit for all queues. */
1821 for (i = 0; i <= pipe->fs.rq_size; i++) {
1822 for (q = pipe->fs.rq[i]; q; q = q->next) {
1823 q->numbytes = p->burst +
1824 (io_fast ? p->bandwidth : 0);
1829 pipe->bandwidth = p->bandwidth;
1830 pipe->burst = p->burst;
1831 pipe->numbytes = pipe->burst + (io_fast ? pipe->bandwidth : 0);
1832 bcopy(p->if_name, pipe->if_name, sizeof(p->if_name));
1833 pipe->ifp = NULL; /* reset interface ptr */
1834 pipe->delay = p->delay;
1835 set_fs_parms(&(pipe->fs), pfs);
1837 /* Handle changes in the delay profile. */
1838 if (p->samples_no > 0) {
1839 if (pipe->samples_no != p->samples_no) {
1840 if (pipe->samples != NULL)
1841 free(pipe->samples, M_DUMMYNET);
1843 malloc(p->samples_no*sizeof(dn_key),
1844 M_DUMMYNET, M_NOWAIT | M_ZERO);
1845 if (pipe->samples == NULL) {
1847 printf("dummynet: no memory "
1848 "for new samples\n");
1851 pipe->samples_no = p->samples_no;
1854 strncpy(pipe->name,p->name,sizeof(pipe->name));
1855 pipe->loss_level = p->loss_level;
1856 for (i = 0; i<pipe->samples_no; ++i)
1857 pipe->samples[i] = p->samples[i];
1858 } else if (pipe->samples != NULL) {
1859 free(pipe->samples, M_DUMMYNET);
1860 pipe->samples = NULL;
1861 pipe->samples_no = 0;
1864 if (pipe->fs.rq == NULL) { /* a new pipe */
1865 error = alloc_hash(&(pipe->fs), pfs);
1871 SLIST_INSERT_HEAD(&pipehash[HASH(pipe->pipe_nr)],
1875 } else { /* config queue */
1876 struct dn_flow_set *fs;
1879 fs = locate_flowset(pfs->fs_nr); /* locate flow_set */
1881 if (fs == NULL) { /* new */
1882 if (pfs->parent_nr == 0) { /* need link to a pipe */
1886 fs = malloc(sizeof(struct dn_flow_set), M_DUMMYNET,
1891 "dummynet: no memory for new flow_set\n");
1894 fs->fs_nr = pfs->fs_nr;
1895 fs->parent_nr = pfs->parent_nr;
1896 fs->weight = pfs->weight;
1897 if (fs->weight == 0)
1899 else if (fs->weight > 100)
1903 * Change parent pipe not allowed;
1904 * must delete and recreate.
1906 if (pfs->parent_nr != 0 &&
1907 fs->parent_nr != pfs->parent_nr) {
1913 set_fs_parms(fs, pfs);
1915 if (fs->rq == NULL) { /* a new flow_set */
1916 error = alloc_hash(fs, pfs);
1919 free(fs, M_DUMMYNET);
1922 SLIST_INSERT_HEAD(&flowsethash[HASH(fs->fs_nr)],
1931 * Helper function to remove from a heap queues which are linked to
1932 * a flow_set about to be deleted.
1935 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs)
1939 for (i = found = 0 ; i < h->elements ;) {
1940 if ( ((struct dn_flow_queue *)h->p[i].object)->fs == fs) {
1942 h->p[i] = h->p[h->elements] ;
1952 * helper function to remove a pipe from a heap (can be there at most once)
1955 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p)
1959 for (i=0; i < h->elements ; i++ ) {
1960 if (h->p[i].object == p) { /* found it */
1962 h->p[i] = h->p[h->elements] ;
1970 * drain all queues. Called in case of severe mbuf shortage.
1973 dummynet_drain(void)
1975 struct dn_flow_set *fs;
1976 struct dn_pipe *pipe;
1979 DUMMYNET_LOCK_ASSERT();
1981 heap_free(&ready_heap);
1982 heap_free(&wfq_ready_heap);
1983 heap_free(&extract_heap);
1984 /* remove all references to this pipe from flow_sets */
1985 for (i = 0; i < HASHSIZE; i++)
1986 SLIST_FOREACH(fs, &flowsethash[i], next)
1987 purge_flow_set(fs, 0);
1989 for (i = 0; i < HASHSIZE; i++) {
1990 SLIST_FOREACH(pipe, &pipehash[i], next) {
1991 purge_flow_set(&(pipe->fs), 0);
1992 dn_free_pkts(pipe->head);
1993 pipe->head = pipe->tail = NULL;
1999 * Fully delete a pipe or a queue, cleaning up associated info.
2002 delete_pipe(struct dn_pipe *p)
2005 if (p->pipe_nr == 0 && p->fs.fs_nr == 0)
2007 if (p->pipe_nr != 0 && p->fs.fs_nr != 0)
2009 if (p->pipe_nr != 0) { /* this is an old-style pipe */
2010 struct dn_pipe *pipe;
2011 struct dn_flow_set *fs;
2015 pipe = locate_pipe(p->pipe_nr); /* locate pipe */
2019 return (ENOENT); /* not found */
2022 /* Unlink from list of pipes. */
2023 SLIST_REMOVE(&pipehash[HASH(pipe->pipe_nr)], pipe, dn_pipe, next);
2025 /* Remove all references to this pipe from flow_sets. */
2026 for (i = 0; i < HASHSIZE; i++) {
2027 SLIST_FOREACH(fs, &flowsethash[i], next) {
2028 if (fs->pipe == pipe) {
2029 printf("dummynet: ++ ref to pipe %d from fs %d\n",
2030 p->pipe_nr, fs->fs_nr);
2032 purge_flow_set(fs, 0);
2036 fs_remove_from_heap(&ready_heap, &(pipe->fs));
2037 purge_pipe(pipe); /* remove all data associated to this pipe */
2038 /* remove reference to here from extract_heap and wfq_ready_heap */
2039 pipe_remove_from_heap(&extract_heap, pipe);
2040 pipe_remove_from_heap(&wfq_ready_heap, pipe);
2044 } else { /* this is a WF2Q queue (dn_flow_set) */
2045 struct dn_flow_set *fs;
2048 fs = locate_flowset(p->fs.fs_nr); /* locate set */
2052 return (ENOENT); /* not found */
2055 /* Unlink from list of flowsets. */
2056 SLIST_REMOVE( &flowsethash[HASH(fs->fs_nr)], fs, dn_flow_set, next);
2058 if (fs->pipe != NULL) {
2059 /* Update total weight on parent pipe and cleanup parent heaps. */
2060 fs->pipe->sum -= fs->weight * fs->backlogged ;
2061 fs_remove_from_heap(&(fs->pipe->not_eligible_heap), fs);
2062 fs_remove_from_heap(&(fs->pipe->scheduler_heap), fs);
2063 #if 1 /* XXX should i remove from idle_heap as well ? */
2064 fs_remove_from_heap(&(fs->pipe->idle_heap), fs);
2067 purge_flow_set(fs, 1);
2074 * helper function used to copy data from kernel in DUMMYNET_GET
2077 dn_copy_set(struct dn_flow_set *set, char *bp)
2080 struct dn_flow_queue *q, *qp = (struct dn_flow_queue *)bp;
2082 DUMMYNET_LOCK_ASSERT();
2084 for (i = 0 ; i <= set->rq_size ; i++) {
2085 for (q = set->rq[i] ; q ; q = q->next, qp++ ) {
2086 if (q->hash_slot != i)
2087 printf("dummynet: ++ at %d: wrong slot (have %d, "
2088 "should be %d)\n", copied, q->hash_slot, i);
2090 printf("dummynet: ++ at %d: wrong fs ptr (have %p, should be %p)\n",
2093 bcopy(q, qp, sizeof( *q ) );
2094 /* cleanup pointers */
2096 qp->head = qp->tail = NULL ;
2100 if (copied != set->rq_elements)
2101 printf("dummynet: ++ wrong count, have %d should be %d\n",
2102 copied, set->rq_elements);
2109 struct dn_flow_set *fs;
2110 struct dn_pipe *pipe;
2114 DUMMYNET_LOCK_ASSERT();
2116 * Compute size of data structures: list of pipes and flow_sets.
2118 for (i = 0; i < HASHSIZE; i++) {
2119 SLIST_FOREACH(pipe, &pipehash[i], next)
2120 size += sizeof(*pipe) +
2121 pipe->fs.rq_elements * sizeof(struct dn_flow_queue);
2122 SLIST_FOREACH(fs, &flowsethash[i], next)
2123 size += sizeof (*fs) +
2124 fs->rq_elements * sizeof(struct dn_flow_queue);
2130 dummynet_get(struct sockopt *sopt)
2132 char *buf, *bp ; /* bp is the "copy-pointer" */
2134 struct dn_flow_set *fs;
2135 struct dn_pipe *pipe;
2138 /* XXX lock held too long */
2141 * XXX: Ugly, but we need to allocate memory with M_WAITOK flag and we
2142 * cannot use this flag while holding a mutex.
2144 for (i = 0; i < 10; i++) {
2145 size = dn_calc_size();
2147 buf = malloc(size, M_TEMP, M_WAITOK);
2149 if (size >= dn_calc_size())
2159 for (i = 0; i < HASHSIZE; i++) {
2160 SLIST_FOREACH(pipe, &pipehash[i], next) {
2161 struct dn_pipe *pipe_bp = (struct dn_pipe *)bp;
2164 * Copy pipe descriptor into *bp, convert delay back to ms,
2165 * then copy the flow_set descriptor(s) one at a time.
2166 * After each flow_set, copy the queue descriptor it owns.
2168 bcopy(pipe, bp, sizeof(*pipe));
2169 pipe_bp->delay = (pipe_bp->delay * 1000) / hz;
2170 pipe_bp->burst = div64(pipe_bp->burst, 8 * hz);
2172 * XXX the following is a hack based on ->next being the
2173 * first field in dn_pipe and dn_flow_set. The correct
2174 * solution would be to move the dn_flow_set to the beginning
2175 * of struct dn_pipe.
2177 pipe_bp->next.sle_next = (struct dn_pipe *)DN_IS_PIPE;
2178 /* Clean pointers. */
2179 pipe_bp->head = pipe_bp->tail = NULL;
2180 pipe_bp->fs.next.sle_next = NULL;
2181 pipe_bp->fs.pipe = NULL;
2182 pipe_bp->fs.rq = NULL;
2183 pipe_bp->samples = NULL;
2185 bp += sizeof(*pipe) ;
2186 bp = dn_copy_set(&(pipe->fs), bp);
2190 for (i = 0; i < HASHSIZE; i++) {
2191 SLIST_FOREACH(fs, &flowsethash[i], next) {
2192 struct dn_flow_set *fs_bp = (struct dn_flow_set *)bp;
2194 bcopy(fs, bp, sizeof(*fs));
2195 /* XXX same hack as above */
2196 fs_bp->next.sle_next = (struct dn_flow_set *)DN_IS_QUEUE;
2200 bp = dn_copy_set(fs, bp);
2206 error = sooptcopyout(sopt, buf, size);
2212 * Handler for the various dummynet socket options (get, flush, config, del)
2215 ip_dn_ctl(struct sockopt *sopt)
2218 struct dn_pipe *p = NULL;
2220 error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2224 /* Disallow sets in really-really secure mode. */
2225 if (sopt->sopt_dir == SOPT_SET) {
2226 #if __FreeBSD_version >= 500034
2227 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2231 if (securelevel >= 3)
2236 switch (sopt->sopt_name) {
2238 printf("dummynet: -- unknown option %d", sopt->sopt_name);
2242 case IP_DUMMYNET_GET :
2243 error = dummynet_get(sopt);
2246 case IP_DUMMYNET_FLUSH :
2250 case IP_DUMMYNET_CONFIGURE :
2251 p = malloc(sizeof(struct dn_pipe_max), M_TEMP, M_WAITOK);
2252 error = sooptcopyin(sopt, p, sizeof(struct dn_pipe_max), sizeof *p);
2255 if (p->samples_no > 0)
2256 p->samples = &(((struct dn_pipe_max *)p)->samples[0]);
2258 error = config_pipe(p);
2261 case IP_DUMMYNET_DEL : /* remove a pipe or queue */
2262 p = malloc(sizeof(struct dn_pipe), M_TEMP, M_WAITOK);
2263 error = sooptcopyin(sopt, p, sizeof(struct dn_pipe), sizeof *p);
2267 error = delete_pipe(p);
2283 printf("DUMMYNET with IPv6 initialized (040826)\n");
2285 DUMMYNET_LOCK_INIT();
2287 for (i = 0; i < HASHSIZE; i++) {
2288 SLIST_INIT(&pipehash[i]);
2289 SLIST_INIT(&flowsethash[i]);
2291 ready_heap.size = ready_heap.elements = 0;
2292 ready_heap.offset = 0;
2294 wfq_ready_heap.size = wfq_ready_heap.elements = 0;
2295 wfq_ready_heap.offset = 0;
2297 extract_heap.size = extract_heap.elements = 0;
2298 extract_heap.offset = 0;
2300 ip_dn_ctl_ptr = ip_dn_ctl;
2301 ip_dn_io_ptr = dummynet_io;
2303 TASK_INIT(&dn_task, 0, dummynet_task, NULL);
2304 dn_tq = taskqueue_create_fast("dummynet", M_NOWAIT,
2305 taskqueue_thread_enqueue, &dn_tq);
2306 taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2308 callout_init(&dn_timeout, CALLOUT_MPSAFE);
2309 callout_reset(&dn_timeout, 1, dummynet, NULL);
2311 /* Initialize curr_time adjustment mechanics. */
2312 getmicrouptime(&prev_t);
2319 ip_dn_ctl_ptr = NULL;
2320 ip_dn_io_ptr = NULL;
2323 callout_stop(&dn_timeout);
2325 taskqueue_drain(dn_tq, &dn_task);
2326 taskqueue_free(dn_tq);
2330 DUMMYNET_LOCK_DESTROY();
2332 #endif /* KLD_MODULE */
2335 dummynet_modevent(module_t mod, int type, void *data)
2341 printf("DUMMYNET already loaded\n");
2348 #if !defined(KLD_MODULE)
2349 printf("dummynet statically compiled, cannot unload\n");
2362 static moduledata_t dummynet_mod = {
2367 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
2368 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
2369 MODULE_VERSION(dummynet, 1);