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: src/sys/netinet/ip_dummynet.c,v 1.110.2.4 2008/10/31 12:58:12 oleg Exp $");
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
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/malloc.h>
65 #include <sys/kernel.h>
67 #include <sys/module.h>
70 #include <sys/rwlock.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
74 #include <sys/sysctl.h>
75 #include <sys/taskqueue.h>
76 #include <net/if.h> /* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */
77 #include <net/netisr.h>
78 #include <netinet/in.h>
79 #include <netinet/ip.h> /* ip_len, ip_off */
80 #include <netinet/ip_fw.h>
81 #include <netinet/ip_dummynet.h>
82 #include <netinet/ip_var.h> /* ip_output(), IP_FORWARDING */
84 #include <netinet/if_ether.h> /* various ether_* routines */
86 #include <netinet/ip6.h> /* for ip6_input, ip6_output prototypes */
87 #include <netinet6/ip6_var.h>
90 * We keep a private variable for the simulation time, but we could
91 * probably use an existing one ("softticks" in sys/kern/kern_timeout.c)
93 static dn_key curr_time = 0 ; /* current simulation time */
95 static int dn_hash_size = 64 ; /* default hash size */
97 /* statistics on number of queue searches and search steps */
98 static long searches, search_steps ;
99 static int pipe_expire = 1 ; /* expire queue if empty */
100 static int dn_max_ratio = 16 ; /* max queues/buckets ratio */
102 static long pipe_slot_limit = 100; /* Foot shooting limit for pipe queues. */
103 static long pipe_byte_limit = 1024 * 1024;
105 static int red_lookup_depth = 256; /* RED - default lookup table depth */
106 static int red_avg_pkt_size = 512; /* RED - default medium packet size */
107 static int red_max_pkt_size = 1500; /* RED - default max packet size */
109 static struct timeval prev_t, t;
110 static long tick_last; /* Last tick duration (usec). */
111 static long tick_delta; /* Last vs standard tick diff (usec). */
112 static long tick_delta_sum; /* Accumulated tick difference (usec).*/
113 static long tick_adjustment; /* Tick adjustments done. */
114 static long tick_lost; /* Lost(coalesced) ticks number. */
115 /* Adjusted vs non-adjusted curr_time difference (ticks). */
116 static long tick_diff;
119 static unsigned long io_pkt;
120 static unsigned long io_pkt_fast;
121 static unsigned long io_pkt_drop;
124 * Three heaps contain queues and pipes that the scheduler handles:
126 * ready_heap contains all dn_flow_queue related to fixed-rate pipes.
128 * wfq_ready_heap contains the pipes associated with WF2Q flows
130 * extract_heap contains pipes associated with delay lines.
134 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap");
136 static struct dn_heap ready_heap, extract_heap, wfq_ready_heap ;
138 static int heap_init(struct dn_heap *h, int size);
139 static int heap_insert (struct dn_heap *h, dn_key key1, void *p);
140 static void heap_extract(struct dn_heap *h, void *obj);
141 static void transmit_event(struct dn_pipe *pipe, struct mbuf **head,
143 static void ready_event(struct dn_flow_queue *q, struct mbuf **head,
145 static void ready_event_wfq(struct dn_pipe *p, struct mbuf **head,
149 #define HASH(num) ((((num) >> 8) ^ ((num) >> 4) ^ (num)) & 0x0f)
150 static struct dn_pipe_head pipehash[HASHSIZE]; /* all pipes */
151 static struct dn_flow_set_head flowsethash[HASHSIZE]; /* all flowsets */
153 static struct callout dn_timeout;
155 extern void (*bridge_dn_p)(struct mbuf *, struct ifnet *);
158 SYSCTL_DECL(_net_inet);
159 SYSCTL_DECL(_net_inet_ip);
161 SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet, CTLFLAG_RW, 0, "Dummynet");
162 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, hash_size,
163 CTLFLAG_RW, &dn_hash_size, 0, "Default hash table size");
164 #if 0 /* curr_time is 64 bit */
165 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, curr_time,
166 CTLFLAG_RD, &curr_time, 0, "Current tick");
168 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, ready_heap,
169 CTLFLAG_RD, &ready_heap.size, 0, "Size of ready heap");
170 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, extract_heap,
171 CTLFLAG_RD, &extract_heap.size, 0, "Size of extract heap");
172 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, searches,
173 CTLFLAG_RD, &searches, 0, "Number of queue searches");
174 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, search_steps,
175 CTLFLAG_RD, &search_steps, 0, "Number of queue search steps");
176 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, expire,
177 CTLFLAG_RW, &pipe_expire, 0, "Expire queue if empty");
178 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, max_chain_len,
179 CTLFLAG_RW, &dn_max_ratio, 0,
180 "Max ratio between dynamic queues and buckets");
181 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth,
182 CTLFLAG_RD, &red_lookup_depth, 0, "Depth of RED lookup table");
183 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size,
184 CTLFLAG_RD, &red_avg_pkt_size, 0, "RED Medium packet size");
185 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size,
186 CTLFLAG_RD, &red_max_pkt_size, 0, "RED Max packet size");
187 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta,
188 CTLFLAG_RD, &tick_delta, 0, "Last vs standard tick difference (usec).");
189 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta_sum,
190 CTLFLAG_RD, &tick_delta_sum, 0, "Accumulated tick difference (usec).");
191 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_adjustment,
192 CTLFLAG_RD, &tick_adjustment, 0, "Tick adjustments done.");
193 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_diff,
194 CTLFLAG_RD, &tick_diff, 0,
195 "Adjusted vs non-adjusted curr_time difference (ticks).");
196 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_lost,
197 CTLFLAG_RD, &tick_lost, 0,
198 "Number of ticks coalesced by dummynet taskqueue.");
199 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, io_fast,
200 CTLFLAG_RW, &io_fast, 0, "Enable fast dummynet io.");
201 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt,
202 CTLFLAG_RD, &io_pkt, 0,
203 "Number of packets passed to dummynet.");
204 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_fast,
205 CTLFLAG_RD, &io_pkt_fast, 0,
206 "Number of packets bypassed dummynet scheduler.");
207 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_drop,
208 CTLFLAG_RD, &io_pkt_drop, 0,
209 "Number of packets dropped by dummynet.");
210 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, pipe_slot_limit,
211 CTLFLAG_RW, &pipe_slot_limit, 0, "Upper limit in slots for pipe queue.");
212 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, pipe_byte_limit,
213 CTLFLAG_RW, &pipe_byte_limit, 0, "Upper limit in bytes for pipe queue.");
216 #ifdef DUMMYNET_DEBUG
217 int dummynet_debug = 0;
219 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, debug, CTLFLAG_RW, &dummynet_debug,
220 0, "control debugging printfs");
222 #define DPRINTF(X) if (dummynet_debug) printf X
227 static struct task dn_task;
228 static struct taskqueue *dn_tq = NULL;
229 static void dummynet_task(void *, int);
231 #if defined( __linux__ ) || defined( _WIN32 )
232 static DEFINE_SPINLOCK(dummynet_mtx);
234 static struct mtx dummynet_mtx;
236 #define DUMMYNET_LOCK_INIT() \
237 mtx_init(&dummynet_mtx, "dummynet", NULL, MTX_DEF)
238 #define DUMMYNET_LOCK_DESTROY() mtx_destroy(&dummynet_mtx)
239 #define DUMMYNET_LOCK() mtx_lock(&dummynet_mtx)
240 #define DUMMYNET_UNLOCK() mtx_unlock(&dummynet_mtx)
241 #define DUMMYNET_LOCK_ASSERT() mtx_assert(&dummynet_mtx, MA_OWNED)
243 static int config_pipe(struct dn_pipe *p);
244 static int ip_dn_ctl(struct sockopt *sopt);
246 static void dummynet(void *);
247 static void dummynet_flush(void);
248 static void dummynet_send(struct mbuf *);
249 void dummynet_drain(void);
250 static int dummynet_io(struct mbuf **, int , struct ip_fw_args *);
253 * Flow queue is idle if:
254 * 1) it's empty for at least 1 tick
255 * 2) it has invalid timestamp (WF2Q case)
256 * 3) parent pipe has no 'exhausted' burst.
258 #define QUEUE_IS_IDLE(q) ((q)->head == NULL && (q)->S == (q)->F + 1 && \
259 curr_time > (q)->idle_time + 1 && \
260 ((q)->numbytes + (curr_time - (q)->idle_time - 1) * \
261 (q)->fs->pipe->bandwidth >= (q)->fs->pipe->burst))
264 * Heap management functions.
266 * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2.
267 * Some macros help finding parent/children so we can optimize them.
269 * heap_init() is called to expand the heap when needed.
270 * Increment size in blocks of 16 entries.
271 * XXX failure to allocate a new element is a pretty bad failure
272 * as we basically stall a whole queue forever!!
273 * Returns 1 on error, 0 on success
275 #define HEAP_FATHER(x) ( ( (x) - 1 ) / 2 )
276 #define HEAP_LEFT(x) ( 2*(x) + 1 )
277 #define HEAP_IS_LEFT(x) ( (x) & 1 )
278 #define HEAP_RIGHT(x) ( 2*(x) + 2 )
279 #define HEAP_SWAP(a, b, buffer) { buffer = a ; a = b ; b = buffer ; }
280 #define HEAP_INCREMENT 15
283 heap_init(struct dn_heap *h, int new_size)
285 struct dn_heap_entry *p;
287 if (h->size >= new_size ) {
288 printf("dummynet: %s, Bogus call, have %d want %d\n", __func__,
292 new_size = (new_size + HEAP_INCREMENT ) & ~HEAP_INCREMENT ;
293 p = malloc(new_size * sizeof(*p), M_DUMMYNET, M_NOWAIT);
295 printf("dummynet: %s, resize %d failed\n", __func__, new_size );
296 return 1 ; /* error */
299 bcopy(h->p, p, h->size * sizeof(*p) );
300 free(h->p, M_DUMMYNET);
308 * Insert element in heap. Normally, p != NULL, we insert p in
309 * a new position and bubble up. If p == NULL, then the element is
310 * already in place, and key is the position where to start the
312 * Returns 1 on failure (cannot allocate new heap entry)
314 * If offset > 0 the position (index, int) of the element in the heap is
315 * also stored in the element itself at the given offset in bytes.
317 #define SET_OFFSET(heap, node) \
318 if (heap->offset > 0) \
319 *((int *)((char *)(heap->p[node].object) + heap->offset)) = node ;
321 * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value.
323 #define RESET_OFFSET(heap, node) \
324 if (heap->offset > 0) \
325 *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1 ;
327 heap_insert(struct dn_heap *h, dn_key key1, void *p)
329 int son = h->elements ;
331 if (p == NULL) /* data already there, set starting point */
333 else { /* insert new element at the end, possibly resize */
335 if (son == h->size) /* need resize... */
336 if (heap_init(h, h->elements+1) )
337 return 1 ; /* failure... */
338 h->p[son].object = p ;
339 h->p[son].key = key1 ;
342 while (son > 0) { /* bubble up */
343 int father = HEAP_FATHER(son) ;
344 struct dn_heap_entry tmp ;
346 if (DN_KEY_LT( h->p[father].key, h->p[son].key ) )
347 break ; /* found right position */
348 /* son smaller than father, swap and repeat */
349 HEAP_SWAP(h->p[son], h->p[father], tmp) ;
358 * remove top element from heap, or obj if obj != NULL
361 heap_extract(struct dn_heap *h, void *obj)
363 int child, father, max = h->elements - 1 ;
366 printf("dummynet: warning, extract from empty heap 0x%p\n", h);
369 father = 0 ; /* default: move up smallest child */
370 if (obj != NULL) { /* extract specific element, index is at offset */
372 panic("dummynet: heap_extract from middle not supported on this heap!!!\n");
373 father = *((int *)((char *)obj + h->offset)) ;
374 if (father < 0 || father >= h->elements) {
375 printf("dummynet: heap_extract, father %d out of bound 0..%d\n",
376 father, h->elements);
377 panic("dummynet: heap_extract");
380 RESET_OFFSET(h, father);
381 child = HEAP_LEFT(father) ; /* left child */
382 while (child <= max) { /* valid entry */
383 if (child != max && DN_KEY_LT(h->p[child+1].key, h->p[child].key) )
384 child = child+1 ; /* take right child, otherwise left */
385 h->p[father] = h->p[child] ;
386 SET_OFFSET(h, father);
388 child = HEAP_LEFT(child) ; /* left child for next loop */
393 * Fill hole with last entry and bubble up, reusing the insert code
395 h->p[father] = h->p[max] ;
396 heap_insert(h, father, NULL); /* this one cannot fail */
402 * change object position and update references
403 * XXX this one is never used!
406 heap_move(struct dn_heap *h, dn_key new_key, void *object)
410 int max = h->elements-1 ;
411 struct dn_heap_entry buf ;
414 panic("cannot move items on this heap");
416 i = *((int *)((char *)object + h->offset));
417 if (DN_KEY_LT(new_key, h->p[i].key) ) { /* must move up */
418 h->p[i].key = new_key ;
419 for (; i>0 && DN_KEY_LT(new_key, h->p[(temp = HEAP_FATHER(i))].key) ;
420 i = temp ) { /* bubble up */
421 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
424 } else { /* must move down */
425 h->p[i].key = new_key ;
426 while ( (temp = HEAP_LEFT(i)) <= max ) { /* found left child */
427 if ((temp != max) && DN_KEY_GT(h->p[temp].key, h->p[temp+1].key))
428 temp++ ; /* select child with min key */
429 if (DN_KEY_GT(new_key, h->p[temp].key)) { /* go down */
430 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
439 #endif /* heap_move, unused */
442 * heapify() will reorganize data inside an array to maintain the
443 * heap property. It is needed when we delete a bunch of entries.
446 heapify(struct dn_heap *h)
450 for (i = 0 ; i < h->elements ; i++ )
451 heap_insert(h, i , NULL) ;
455 * cleanup the heap and free data structure
458 heap_free(struct dn_heap *h)
461 free(h->p, M_DUMMYNET);
462 bzero(h, sizeof(*h) );
466 * --- end of heap management functions ---
470 * Dispose a packet in dummynet. Use an inline functions so if we
471 * need to free extra state associated to a packet, this is a
472 * central point to do it.
474 static __inline void *dn_free_pkt(struct mbuf *m)
477 netisr_dispatch(-1, m); /* -1 drop the packet */
484 static __inline void dn_free_pkts(struct mbuf *mnext)
488 while ((m = mnext) != NULL) {
489 mnext = m->m_nextpkt;
495 * Return the mbuf tag holding the dummynet state. As an optimization
496 * this is assumed to be the first tag on the list. If this turns out
497 * wrong we'll need to search the list.
499 static struct dn_pkt_tag *
500 dn_tag_get(struct mbuf *m)
502 struct m_tag *mtag = m_tag_first(m);
503 KASSERT(mtag != NULL &&
504 mtag->m_tag_cookie == MTAG_ABI_COMPAT &&
505 mtag->m_tag_id == PACKET_TAG_DUMMYNET,
506 ("packet on dummynet queue w/o dummynet tag!"));
507 return (struct dn_pkt_tag *)(mtag+1);
511 * Scheduler functions:
513 * transmit_event() is called when the delay-line needs to enter
514 * the scheduler, either because of existing pkts getting ready,
515 * or new packets entering the queue. The event handled is the delivery
516 * time of the packet.
518 * ready_event() does something similar with fixed-rate queues, and the
519 * event handled is the finish time of the head pkt.
521 * wfq_ready_event() does something similar with WF2Q queues, and the
522 * event handled is the start time of the head pkt.
524 * In all cases, we make sure that the data structures are consistent
525 * before passing pkts out, because this might trigger recursive
526 * invocations of the procedures.
529 transmit_event(struct dn_pipe *pipe, struct mbuf **head, struct mbuf **tail)
532 struct dn_pkt_tag *pkt;
534 DUMMYNET_LOCK_ASSERT();
536 while ((m = pipe->head) != NULL) {
538 if (!DN_KEY_LEQ(pkt->output_time, curr_time))
541 pipe->head = m->m_nextpkt;
543 (*tail)->m_nextpkt = m;
549 (*tail)->m_nextpkt = NULL;
551 /* If there are leftover packets, put into the heap for next event. */
552 if ((m = pipe->head) != NULL) {
555 * XXX Should check errors on heap_insert, by draining the
556 * whole pipe p and hoping in the future we are more successful.
558 heap_insert(&extract_heap, pkt->output_time, pipe);
563 #define div64(a, b) ((int64_t)(a) / (int64_t)(b))
565 #define DN_TO_DROP 0xffff
567 * Compute how many ticks we have to wait before being able to send
568 * a packet. This is computed as the "wire time" for the packet
569 * (length + extra bits), minus the credit available, scaled to ticks.
570 * Check that the result is not be negative (it could be if we have
571 * too much leftover credit in q->numbytes).
574 set_ticks(struct mbuf *m, struct dn_flow_queue *q, struct dn_pipe *p)
578 ret = div64( (m->m_pkthdr.len * 8 + q->extra_bits) * hz
579 - q->numbytes + p->bandwidth - 1 , p->bandwidth);
581 printf("%s %d extra_bits %d numb %d ret %d\n",
582 __FUNCTION__, __LINE__,
583 (int)(q->extra_bits & 0xffffffff),
584 (int)(q->numbytes & 0xffffffff),
585 (int)(ret & 0xffffffff));
593 * Convert the additional MAC overheads/delays into an equivalent
594 * number of bits for the given data rate. The samples are in milliseconds
595 * so we need to divide by 1000.
598 compute_extra_bits(struct mbuf *pkt, struct dn_pipe *p)
603 if (!p->samples || p->samples_no == 0)
605 index = random() % p->samples_no;
606 extra_bits = div64((dn_key)p->samples[index] * p->bandwidth, 1000);
607 if (index >= p->loss_level) {
608 struct dn_pkt_tag *dt = dn_tag_get(pkt);
610 dt->dn_dir = DN_TO_DROP;
616 free_pipe(struct dn_pipe *p)
619 free(p->samples, M_DUMMYNET);
624 * extract pkt from queue, compute output time (could be now)
625 * and put into delay line (p_queue)
628 move_pkt(struct mbuf *pkt, struct dn_flow_queue *q, struct dn_pipe *p,
631 struct dn_pkt_tag *dt = dn_tag_get(pkt);
633 q->head = pkt->m_nextpkt ;
635 q->len_bytes -= len ;
637 dt->output_time = curr_time + p->delay ;
642 p->tail->m_nextpkt = pkt;
644 p->tail->m_nextpkt = NULL;
648 * ready_event() is invoked every time the queue must enter the
649 * scheduler, either because the first packet arrives, or because
650 * a previously scheduled event fired.
651 * On invokation, drain as many pkts as possible (could be 0) and then
652 * if there are leftover packets reinsert the pkt in the scheduler.
655 ready_event(struct dn_flow_queue *q, struct mbuf **head, struct mbuf **tail)
658 struct dn_pipe *p = q->fs->pipe;
661 DUMMYNET_LOCK_ASSERT();
664 printf("dummynet: ready_event- pipe is gone\n");
667 p_was_empty = (p->head == NULL);
670 * Schedule fixed-rate queues linked to this pipe:
671 * account for the bw accumulated since last scheduling, then
672 * drain as many pkts as allowed by q->numbytes and move to
673 * the delay line (in p) computing output time.
674 * bandwidth==0 (no limit) means we can drain the whole queue,
675 * setting len_scaled = 0 does the job.
677 q->numbytes += (curr_time - q->sched_time) * p->bandwidth;
678 while ((pkt = q->head) != NULL) {
679 int len = pkt->m_pkthdr.len;
680 dn_key len_scaled = p->bandwidth ? len*8*hz
684 if (DN_KEY_GT(len_scaled, q->numbytes))
686 q->numbytes -= len_scaled;
687 move_pkt(pkt, q, p, len);
689 q->extra_bits = compute_extra_bits(q->head, p);
692 * If we have more packets queued, schedule next ready event
693 * (can only occur when bandwidth != 0, otherwise we would have
694 * flushed the whole queue in the previous loop).
695 * To this purpose we record the current time and compute how many
696 * ticks to go for the finish time of the packet.
698 if ((pkt = q->head) != NULL) { /* this implies bandwidth != 0 */
699 dn_key t = set_ticks(pkt, q, p); /* ticks i have to wait */
701 q->sched_time = curr_time;
702 heap_insert(&ready_heap, curr_time + t, (void *)q);
704 * XXX Should check errors on heap_insert, and drain the whole
705 * queue on error hoping next time we are luckier.
707 } else /* RED needs to know when the queue becomes empty. */
708 q->idle_time = curr_time;
711 * If the delay line was empty call transmit_event() now.
712 * Otherwise, the scheduler will take care of it.
715 transmit_event(p, head, tail);
719 * Called when we can transmit packets on WF2Q queues. Take pkts out of
720 * the queues at their start time, and enqueue into the delay line.
721 * Packets are drained until p->numbytes < 0. As long as
722 * len_scaled >= p->numbytes, the packet goes into the delay line
723 * with a deadline p->delay. For the last packet, if p->numbytes < 0,
724 * there is an additional delay.
727 ready_event_wfq(struct dn_pipe *p, struct mbuf **head, struct mbuf **tail)
729 int p_was_empty = (p->head == NULL);
730 struct dn_heap *sch = &(p->scheduler_heap);
731 struct dn_heap *neh = &(p->not_eligible_heap);
732 int64_t p_numbytes = p->numbytes;
735 * p->numbytes is only 32bits in FBSD7, but we might need 64 bits.
736 * Use a local variable for the computations, and write back the
737 * results when done, saturating if needed.
738 * The local variable has no impact on performance and helps
739 * reducing diffs between the various branches.
742 DUMMYNET_LOCK_ASSERT();
744 if (p->if_name[0] == 0) /* tx clock is simulated */
745 p_numbytes += (curr_time - p->sched_time) * p->bandwidth;
747 * tx clock is for real,
748 * the ifq must be empty or this is a NOP.
749 * XXX not supported in Linux
751 if (1) // p->ifp && p->ifp->if_snd.ifq_head != NULL)
754 DPRINTF(("dummynet: pipe %d ready from %s --\n",
755 p->pipe_nr, p->if_name));
760 * While we have backlogged traffic AND credit, we need to do
761 * something on the queue.
763 while (p_numbytes >= 0 && (sch->elements > 0 || neh->elements > 0)) {
764 if (sch->elements > 0) {
765 /* Have some eligible pkts to send out. */
766 struct dn_flow_queue *q = sch->p[0].object;
767 struct mbuf *pkt = q->head;
768 struct dn_flow_set *fs = q->fs;
769 uint64_t len = pkt->m_pkthdr.len;
770 int len_scaled = p->bandwidth ? len * 8 * hz : 0;
772 heap_extract(sch, NULL); /* Remove queue from heap. */
773 p_numbytes -= len_scaled;
774 move_pkt(pkt, q, p, len);
776 p->V += div64((len << MY_M), p->sum); /* Update V. */
777 q->S = q->F; /* Update start time. */
779 /* Flow not backlogged any more. */
781 heap_insert(&(p->idle_heap), q->F, q);
783 /* Still backlogged. */
786 * Update F and position in backlogged queue,
787 * then put flow in not_eligible_heap
788 * (we will fix this later).
790 len = (q->head)->m_pkthdr.len;
791 q->F += div64((len << MY_M), fs->weight);
792 if (DN_KEY_LEQ(q->S, p->V))
793 heap_insert(neh, q->S, q);
795 heap_insert(sch, q->F, q);
799 * Now compute V = max(V, min(S_i)). Remember that all elements
800 * in sch have by definition S_i <= V so if sch is not empty,
801 * V is surely the max and we must not update it. Conversely,
802 * if sch is empty we only need to look at neh.
804 if (sch->elements == 0 && neh->elements > 0)
805 p->V = MAX64(p->V, neh->p[0].key);
806 /* Move from neh to sch any packets that have become eligible */
807 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V)) {
808 struct dn_flow_queue *q = neh->p[0].object;
809 heap_extract(neh, NULL);
810 heap_insert(sch, q->F, q);
813 if (p->if_name[0] != '\0') { /* Tx clock is from a real thing */
814 p_numbytes = -1; /* Mark not ready for I/O. */
818 if (sch->elements == 0 && neh->elements == 0 && p_numbytes >= 0) {
819 p->idle_time = curr_time;
821 * No traffic and no events scheduled.
822 * We can get rid of idle-heap.
824 if (p->idle_heap.elements > 0) {
827 for (i = 0; i < p->idle_heap.elements; i++) {
828 struct dn_flow_queue *q;
830 q = p->idle_heap.p[i].object;
836 p->idle_heap.elements = 0;
840 * If we are getting clocks from dummynet (not a real interface) and
841 * If we are under credit, schedule the next ready event.
842 * Also fix the delivery time of the last packet.
844 if (p->if_name[0]==0 && p_numbytes < 0) { /* This implies bw > 0. */
845 dn_key t = 0; /* Number of ticks i have to wait. */
847 if (p->bandwidth > 0)
848 t = div64(p->bandwidth - 1 - p_numbytes, p->bandwidth);
849 dn_tag_get(p->tail)->output_time += t;
850 p->sched_time = curr_time;
851 heap_insert(&wfq_ready_heap, curr_time + t, (void *)p);
853 * XXX Should check errors on heap_insert, and drain the whole
854 * queue on error hoping next time we are luckier.
858 /* Write back p_numbytes (adjust 64->32bit if necessary). */
859 p->numbytes = p_numbytes;
862 * If the delay line was empty call transmit_event() now.
863 * Otherwise, the scheduler will take care of it.
866 transmit_event(p, head, tail);
870 * This is called one tick, after previous run. It is used to
874 dummynet(void * __unused unused)
877 taskqueue_enqueue(dn_tq, &dn_task);
881 * The main dummynet processing function.
884 dummynet_task(void *context, int pending)
886 struct mbuf *head = NULL, *tail = NULL;
887 struct dn_pipe *pipe;
888 struct dn_heap *heaps[3];
890 void *p; /* generic parameter to handler */
895 heaps[0] = &ready_heap; /* fixed-rate queues */
896 heaps[1] = &wfq_ready_heap; /* wfq queues */
897 heaps[2] = &extract_heap; /* delay line */
899 /* Update number of lost(coalesced) ticks. */
900 tick_lost += pending - 1;
903 /* Last tick duration (usec). */
904 tick_last = (t.tv_sec - prev_t.tv_sec) * 1000000 +
905 (t.tv_usec - prev_t.tv_usec);
906 /* Last tick vs standard tick difference (usec). */
907 tick_delta = (tick_last * hz - 1000000) / hz;
908 /* Accumulated tick difference (usec). */
909 tick_delta_sum += tick_delta;
914 * Adjust curr_time if accumulated tick difference greater than
915 * 'standard' tick. Since curr_time should be monotonically increasing,
916 * we do positive adjustment as required and throttle curr_time in
917 * case of negative adjustment.
920 if (tick_delta_sum - tick >= 0) {
921 int diff = tick_delta_sum / tick;
925 tick_delta_sum %= tick;
927 } else if (tick_delta_sum + tick <= 0) {
930 tick_delta_sum += tick;
934 for (i = 0; i < 3; i++) {
936 while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time)) {
937 if (h->p[0].key > curr_time)
938 printf("dummynet: warning, "
939 "heap %d is %d ticks late\n",
940 i, (int)(curr_time - h->p[0].key));
941 /* store a copy before heap_extract */
943 /* need to extract before processing */
944 heap_extract(h, NULL);
946 ready_event(p, &head, &tail);
948 struct dn_pipe *pipe = p;
949 if (pipe->if_name[0] != '\0')
950 printf("dummynet: bad ready_event_wfq "
951 "for pipe %s\n", pipe->if_name);
953 ready_event_wfq(p, &head, &tail);
955 transmit_event(p, &head, &tail);
959 /* Sweep pipes trying to expire idle flow_queues. */
960 for (i = 0; i < HASHSIZE; i++)
961 SLIST_FOREACH(pipe, &pipehash[i], next)
962 if (pipe->idle_heap.elements > 0 &&
963 DN_KEY_LT(pipe->idle_heap.p[0].key, pipe->V)) {
964 struct dn_flow_queue *q =
965 pipe->idle_heap.p[0].object;
967 heap_extract(&(pipe->idle_heap), NULL);
968 /* Mark timestamp as invalid. */
970 pipe->sum -= q->fs->weight;
978 callout_reset(&dn_timeout, 1, dummynet, NULL);
982 dummynet_send(struct mbuf *m)
984 struct dn_pkt_tag *pkt;
989 for (; m != NULL; m = n) {
992 if (m_tag_first(m) == NULL) {
993 pkt = NULL; /* probably unnecessary */
1002 ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
1005 ip = mtod(m, struct ip *);
1006 #ifndef __linux__ /* restore net format for FreeBSD */
1007 ip->ip_len = htons(ip->ip_len);
1008 ip->ip_off = htons(ip->ip_off);
1010 netisr_dispatch(NETISR_IP, m);
1014 netisr_dispatch(NETISR_IPV6, m);
1018 ip6_output(m, NULL, NULL, IPV6_FORWARDING, NULL, NULL, NULL);
1022 if (bridge_dn_p != NULL)
1023 ((*bridge_dn_p)(m, pkt->ifp));
1025 printf("dummynet: if_bridge not loaded\n");
1028 case DN_TO_ETH_DEMUX:
1030 * The Ethernet code assumes the Ethernet header is
1031 * contiguous in the first mbuf header.
1032 * Insure this is true.
1034 if (m->m_len < ETHER_HDR_LEN &&
1035 (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) {
1036 printf("dummynet/ether: pullup failed, "
1037 "dropping packet\n");
1040 ether_demux(m->m_pkthdr.rcvif, m);
1043 ether_output_frame(pkt->ifp, m);
1047 /* drop the packet after some time */
1052 printf("dummynet: bad switch %d!\n", pkt->dn_dir);
1060 * Unconditionally expire empty queues in case of shortage.
1061 * Returns the number of queues freed.
1064 expire_queues(struct dn_flow_set *fs)
1066 struct dn_flow_queue *q, *prev ;
1067 int i, initial_elements = fs->rq_elements ;
1069 if (fs->last_expired == time_uptime)
1071 fs->last_expired = time_uptime ;
1072 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is overflow */
1073 for (prev=NULL, q = fs->rq[i] ; q != NULL ; )
1074 if (!QUEUE_IS_IDLE(q)) {
1077 } else { /* entry is idle, expire it */
1078 struct dn_flow_queue *old_q = q ;
1081 prev->next = q = q->next ;
1083 fs->rq[i] = q = q->next ;
1085 free(old_q, M_DUMMYNET);
1087 return initial_elements - fs->rq_elements ;
1091 * If room, create a new queue and put at head of slot i;
1092 * otherwise, create or use the default queue.
1094 static struct dn_flow_queue *
1095 create_queue(struct dn_flow_set *fs, int i)
1097 struct dn_flow_queue *q;
1099 if (fs->rq_elements > fs->rq_size * dn_max_ratio &&
1100 expire_queues(fs) == 0) {
1101 /* No way to get room, use or create overflow queue. */
1103 if (fs->rq[i] != NULL)
1106 q = malloc(sizeof(*q), M_DUMMYNET, M_NOWAIT | M_ZERO);
1108 printf("dummynet: sorry, cannot allocate queue for new flow\n");
1113 q->next = fs->rq[i];
1114 q->S = q->F + 1; /* hack - mark timestamp as invalid. */
1115 q->numbytes = fs->pipe->burst + (io_fast ? fs->pipe->bandwidth : 0);
1122 * Given a flow_set and a pkt in last_pkt, find a matching queue
1123 * after appropriate masking. The queue is moved to front
1124 * so that further searches take less time.
1126 static struct dn_flow_queue *
1127 find_queue(struct dn_flow_set *fs, struct ipfw_flow_id *id)
1129 int i = 0 ; /* we need i and q for new allocations */
1130 struct dn_flow_queue *q, *prev;
1131 int is_v6 = IS_IP6_FLOW_ID(id);
1133 if ( !(fs->flags_fs & DN_HAVE_FLOW_MASK) )
1136 /* first, do the masking, then hash */
1137 id->dst_port &= fs->flow_mask.dst_port ;
1138 id->src_port &= fs->flow_mask.src_port ;
1139 id->proto &= fs->flow_mask.proto ;
1140 id->flags = 0 ; /* we don't care about this one */
1142 APPLY_MASK(&id->dst_ip6, &fs->flow_mask.dst_ip6);
1143 APPLY_MASK(&id->src_ip6, &fs->flow_mask.src_ip6);
1144 id->flow_id6 &= fs->flow_mask.flow_id6;
1146 i = ((id->dst_ip6.__u6_addr.__u6_addr32[0]) & 0xffff)^
1147 ((id->dst_ip6.__u6_addr.__u6_addr32[1]) & 0xffff)^
1148 ((id->dst_ip6.__u6_addr.__u6_addr32[2]) & 0xffff)^
1149 ((id->dst_ip6.__u6_addr.__u6_addr32[3]) & 0xffff)^
1151 ((id->dst_ip6.__u6_addr.__u6_addr32[0] >> 15) & 0xffff)^
1152 ((id->dst_ip6.__u6_addr.__u6_addr32[1] >> 15) & 0xffff)^
1153 ((id->dst_ip6.__u6_addr.__u6_addr32[2] >> 15) & 0xffff)^
1154 ((id->dst_ip6.__u6_addr.__u6_addr32[3] >> 15) & 0xffff)^
1156 ((id->src_ip6.__u6_addr.__u6_addr32[0] << 1) & 0xfffff)^
1157 ((id->src_ip6.__u6_addr.__u6_addr32[1] << 1) & 0xfffff)^
1158 ((id->src_ip6.__u6_addr.__u6_addr32[2] << 1) & 0xfffff)^
1159 ((id->src_ip6.__u6_addr.__u6_addr32[3] << 1) & 0xfffff)^
1161 ((id->src_ip6.__u6_addr.__u6_addr32[0] << 16) & 0xffff)^
1162 ((id->src_ip6.__u6_addr.__u6_addr32[1] << 16) & 0xffff)^
1163 ((id->src_ip6.__u6_addr.__u6_addr32[2] << 16) & 0xffff)^
1164 ((id->src_ip6.__u6_addr.__u6_addr32[3] << 16) & 0xffff)^
1166 (id->dst_port << 1) ^ (id->src_port) ^
1170 id->dst_ip &= fs->flow_mask.dst_ip ;
1171 id->src_ip &= fs->flow_mask.src_ip ;
1173 i = ( (id->dst_ip) & 0xffff ) ^
1174 ( (id->dst_ip >> 15) & 0xffff ) ^
1175 ( (id->src_ip << 1) & 0xffff ) ^
1176 ( (id->src_ip >> 16 ) & 0xffff ) ^
1177 (id->dst_port << 1) ^ (id->src_port) ^
1180 i = i % fs->rq_size ;
1181 /* finally, scan the current list for a match */
1183 for (prev=NULL, q = fs->rq[i] ; q ; ) {
1186 IN6_ARE_ADDR_EQUAL(&id->dst_ip6,&q->id.dst_ip6) &&
1187 IN6_ARE_ADDR_EQUAL(&id->src_ip6,&q->id.src_ip6) &&
1188 id->dst_port == q->id.dst_port &&
1189 id->src_port == q->id.src_port &&
1190 id->proto == q->id.proto &&
1191 id->flags == q->id.flags &&
1192 id->flow_id6 == q->id.flow_id6)
1195 if (!is_v6 && id->dst_ip == q->id.dst_ip &&
1196 id->src_ip == q->id.src_ip &&
1197 id->dst_port == q->id.dst_port &&
1198 id->src_port == q->id.src_port &&
1199 id->proto == q->id.proto &&
1200 id->flags == q->id.flags)
1203 /* No match. Check if we can expire the entry */
1204 if (pipe_expire && QUEUE_IS_IDLE(q)) {
1205 /* entry is idle and not in any heap, expire it */
1206 struct dn_flow_queue *old_q = q ;
1209 prev->next = q = q->next ;
1211 fs->rq[i] = q = q->next ;
1213 free(old_q, M_DUMMYNET);
1219 if (q && prev != NULL) { /* found and not in front */
1220 prev->next = q->next ;
1221 q->next = fs->rq[i] ;
1225 if (q == NULL) { /* no match, need to allocate a new entry */
1226 q = create_queue(fs, i);
1234 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len)
1239 * RED calculates the average queue size (avg) using a low-pass filter
1240 * with an exponential weighted (w_q) moving average:
1241 * avg <- (1-w_q) * avg + w_q * q_size
1242 * where q_size is the queue length (measured in bytes or * packets).
1244 * If q_size == 0, we compute the idle time for the link, and set
1245 * avg = (1 - w_q)^(idle/s)
1246 * where s is the time needed for transmitting a medium-sized packet.
1248 * Now, if avg < min_th the packet is enqueued.
1249 * If avg > max_th the packet is dropped. Otherwise, the packet is
1250 * dropped with probability P function of avg.
1255 /* Queue in bytes or packets? */
1256 u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ?
1257 q->len_bytes : q->len;
1259 DPRINTF(("\ndummynet: %d q: %2u ", (int)curr_time, q_size));
1261 /* Average queue size estimation. */
1263 /* Queue is not empty, avg <- avg + (q_size - avg) * w_q */
1264 int diff = SCALE(q_size) - q->avg;
1265 int64_t v = SCALE_MUL((int64_t)diff, (int64_t)fs->w_q);
1270 * Queue is empty, find for how long the queue has been
1271 * empty and use a lookup table for computing
1272 * (1 - * w_q)^(idle_time/s) where s is the time to send a
1274 * XXX check wraps...
1277 u_int t = div64(curr_time - q->idle_time,
1280 q->avg = (t < fs->lookup_depth) ?
1281 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0;
1284 DPRINTF(("dummynet: avg: %u ", SCALE_VAL(q->avg)));
1286 /* Should i drop? */
1287 if (q->avg < fs->min_th) {
1289 return (0); /* accept packet */
1291 if (q->avg >= fs->max_th) { /* average queue >= max threshold */
1292 if (fs->flags_fs & DN_IS_GENTLE_RED) {
1294 * According to Gentle-RED, if avg is greater than
1295 * max_th the packet is dropped with a probability
1296 * p_b = c_3 * avg - c_4
1297 * where c_3 = (1 - max_p) / max_th
1298 * c_4 = 1 - 2 * max_p
1300 p_b = SCALE_MUL((int64_t)fs->c_3, (int64_t)q->avg) -
1304 DPRINTF(("dummynet: - drop"));
1307 } else if (q->avg > fs->min_th) {
1309 * We compute p_b using the linear dropping function
1310 * p_b = c_1 * avg - c_2
1311 * where c_1 = max_p / (max_th - min_th)
1312 * c_2 = max_p * min_th / (max_th - min_th)
1314 p_b = SCALE_MUL((int64_t)fs->c_1, (int64_t)q->avg) - fs->c_2;
1317 if (fs->flags_fs & DN_QSIZE_IS_BYTES)
1318 p_b = div64(p_b * len, fs->max_pkt_size);
1319 if (++q->count == 0)
1320 q->random = random() & 0xffff;
1323 * q->count counts packets arrived since last drop, so a greater
1324 * value of q->count means a greater packet drop probability.
1326 if (SCALE_MUL(p_b, SCALE((int64_t)q->count)) > q->random) {
1328 DPRINTF(("dummynet: - red drop"));
1329 /* After a drop we calculate a new random value. */
1330 q->random = random() & 0xffff;
1331 return (1); /* drop */
1334 /* End of RED algorithm. */
1336 return (0); /* accept */
1339 static __inline struct dn_flow_set *
1340 locate_flowset(int fs_nr)
1342 struct dn_flow_set *fs;
1344 SLIST_FOREACH(fs, &flowsethash[HASH(fs_nr)], next)
1345 if (fs->fs_nr == fs_nr)
1351 static __inline struct dn_pipe *
1352 locate_pipe(int pipe_nr)
1354 struct dn_pipe *pipe;
1356 SLIST_FOREACH(pipe, &pipehash[HASH(pipe_nr)], next)
1357 if (pipe->pipe_nr == pipe_nr)
1364 * dummynet hook for packets. Below 'pipe' is a pipe or a queue
1365 * depending on whether WF2Q or fixed bw is used.
1367 * pipe_nr pipe or queue the packet is destined for.
1368 * dir where shall we send the packet after dummynet.
1369 * m the mbuf with the packet
1370 * ifp the 'ifp' parameter from the caller.
1371 * NULL in ip_input, destination interface in ip_output,
1372 * rule matching rule, in case of multiple passes
1375 dummynet_io(struct mbuf **m0, int dir, struct ip_fw_args *fwa)
1377 struct mbuf *m = *m0, *head = NULL, *tail = NULL;
1378 struct dn_pkt_tag *pkt;
1380 struct dn_flow_set *fs = NULL;
1381 struct dn_pipe *pipe;
1382 uint64_t len = m->m_pkthdr.len;
1383 struct dn_flow_queue *q = NULL;
1385 ipfw_insn *cmd = ACTION_PTR(fwa->rule);
1387 KASSERT(m->m_nextpkt == NULL,
1388 ("dummynet_io: mbuf queue passed to dummynet"));
1390 if (cmd->opcode == O_LOG)
1392 if (cmd->opcode == O_ALTQ)
1394 if (cmd->opcode == O_TAG)
1396 is_pipe = (cmd->opcode == O_PIPE);
1401 * This is a dummynet rule, so we expect an O_PIPE or O_QUEUE rule.
1403 * XXXGL: probably the pipe->fs and fs->pipe logic here
1404 * below can be simplified.
1407 pipe = locate_pipe(fwa->cookie);
1411 fs = locate_flowset(fwa->cookie);
1414 goto dropit; /* This queue/pipe does not exist! */
1416 if (pipe == NULL) { /* Must be a queue, try find a matching pipe. */
1417 pipe = locate_pipe(fs->parent_nr);
1421 printf("dummynet: no pipe %d for queue %d, drop pkt\n",
1422 fs->parent_nr, fs->fs_nr);
1426 q = find_queue(fs, &(fwa->f_id));
1428 goto dropit; /* Cannot allocate queue. */
1430 /* Update statistics, then check reasons to drop pkt. */
1431 q->tot_bytes += len;
1433 if (fs->plr && random() < fs->plr)
1434 goto dropit; /* Random pkt drop. */
1435 if (fs->flags_fs & DN_QSIZE_IS_BYTES) {
1436 if (q->len_bytes > fs->qsize)
1437 goto dropit; /* Queue size overflow. */
1439 if (q->len >= fs->qsize)
1440 goto dropit; /* Queue count overflow. */
1442 if (fs->flags_fs & DN_IS_RED && red_drops(fs, q, len))
1445 /* XXX expensive to zero, see if we can remove it. */
1446 mtag = m_tag_get(PACKET_TAG_DUMMYNET,
1447 sizeof(struct dn_pkt_tag), M_NOWAIT | M_ZERO);
1449 goto dropit; /* Cannot allocate packet header. */
1450 m_tag_prepend(m, mtag); /* Attach to mbuf chain. */
1452 pkt = (struct dn_pkt_tag *)(mtag + 1);
1454 * Ok, i can handle the pkt now...
1455 * Build and enqueue packet + parameters.
1457 pkt->rule = fwa->rule;
1458 pkt->rule_id = fwa->rule_id;
1459 pkt->chain_id = fwa->chain_id;
1462 pkt->ifp = fwa->oif;
1464 if (q->head == NULL)
1467 q->tail->m_nextpkt = m;
1470 q->len_bytes += len;
1472 if (q->head != m) /* Flow was not idle, we are done. */
1475 if (is_pipe) { /* Fixed rate queues. */
1476 if (q->idle_time < curr_time) {
1477 /* Calculate available burst size. */
1479 (curr_time - q->idle_time - 1) * pipe->bandwidth;
1480 if (q->numbytes > pipe->burst)
1481 q->numbytes = pipe->burst;
1483 q->numbytes += pipe->bandwidth;
1485 } else { /* WF2Q. */
1486 if (pipe->idle_time < curr_time &&
1487 pipe->scheduler_heap.elements == 0 &&
1488 pipe->not_eligible_heap.elements == 0) {
1489 /* Calculate available burst size. */
1491 (curr_time - pipe->idle_time - 1) * pipe->bandwidth;
1492 if (pipe->numbytes > 0 && pipe->numbytes > pipe->burst)
1493 pipe->numbytes = pipe->burst;
1495 pipe->numbytes += pipe->bandwidth;
1497 pipe->idle_time = curr_time;
1499 /* Necessary for both: fixed rate & WF2Q queues. */
1500 q->idle_time = curr_time;
1503 * If we reach this point the flow was previously idle, so we need
1504 * to schedule it. This involves different actions for fixed-rate or
1508 /* Fixed-rate queue: just insert into the ready_heap. */
1511 if (pipe->bandwidth) {
1512 q->extra_bits = compute_extra_bits(m, pipe);
1513 t = set_ticks(m, q, pipe);
1515 q->sched_time = curr_time;
1516 if (t == 0) /* Must process it now. */
1517 ready_event(q, &head, &tail);
1519 heap_insert(&ready_heap, curr_time + t , q);
1522 * WF2Q. First, compute start time S: if the flow was
1523 * idle (S = F + 1) set S to the virtual time V for the
1524 * controlling pipe, and update the sum of weights for the pipe;
1525 * otherwise, remove flow from idle_heap and set S to max(F,V).
1526 * Second, compute finish time F = S + len / weight.
1527 * Third, if pipe was idle, update V = max(S, V).
1528 * Fourth, count one more backlogged flow.
1530 if (DN_KEY_GT(q->S, q->F)) { /* Means timestamps are invalid. */
1532 pipe->sum += fs->weight; /* Add weight of new queue. */
1534 heap_extract(&(pipe->idle_heap), q);
1535 q->S = MAX64(q->F, pipe->V);
1537 q->F = q->S + div64(len << MY_M, fs->weight);
1539 if (pipe->not_eligible_heap.elements == 0 &&
1540 pipe->scheduler_heap.elements == 0)
1541 pipe->V = MAX64(q->S, pipe->V);
1544 * Look at eligibility. A flow is not eligibile if S>V (when
1545 * this happens, it means that there is some other flow already
1546 * scheduled for the same pipe, so the scheduler_heap cannot be
1547 * empty). If the flow is not eligible we just store it in the
1548 * not_eligible_heap. Otherwise, we store in the scheduler_heap
1549 * and possibly invoke ready_event_wfq() right now if there is
1551 * Note that for all flows in scheduler_heap (SCH), S_i <= V,
1552 * and for all flows in not_eligible_heap (NEH), S_i > V.
1553 * So when we need to compute max(V, min(S_i)) forall i in
1554 * SCH+NEH, we only need to look into NEH.
1556 if (DN_KEY_GT(q->S, pipe->V)) { /* Not eligible. */
1557 if (pipe->scheduler_heap.elements == 0)
1558 printf("dummynet: ++ ouch! not eligible but empty scheduler!\n");
1559 heap_insert(&(pipe->not_eligible_heap), q->S, q);
1561 heap_insert(&(pipe->scheduler_heap), q->F, q);
1562 if (pipe->numbytes >= 0) { /* Pipe is idle. */
1563 if (pipe->scheduler_heap.elements != 1)
1564 printf("dummynet: OUCH! pipe should have been idle!\n");
1565 DPRINTF(("dummynet: waking up pipe %d at %d\n",
1566 pipe->pipe_nr, (int)(q->F >> MY_M)));
1567 pipe->sched_time = curr_time;
1568 ready_event_wfq(pipe, &head, &tail);
1573 if (head == m && dir != DN_TO_IFB_FWD && dir != DN_TO_ETH_DEMUX &&
1574 dir != DN_TO_ETH_OUT) { /* Fast io. */
1576 if (m->m_nextpkt != NULL)
1577 printf("dummynet: fast io: pkt chain detected!\n");
1578 head = m->m_nextpkt = NULL;
1580 *m0 = NULL; /* Normal io. */
1584 dummynet_send(head);
1592 *m0 = dn_free_pkt(m);
1593 return ((fs && (fs->flags_fs & DN_NOERROR)) ? 0 : ENOBUFS);
1597 * Dispose all packets and flow_queues on a flow_set.
1598 * If all=1, also remove red lookup table and other storage,
1599 * including the descriptor itself.
1600 * For the one in dn_pipe MUST also cleanup ready_heap...
1603 purge_flow_set(struct dn_flow_set *fs, int all)
1605 struct dn_flow_queue *q, *qn;
1608 DUMMYNET_LOCK_ASSERT();
1610 for (i = 0; i <= fs->rq_size; i++) {
1611 for (q = fs->rq[i]; q != NULL; q = qn) {
1612 dn_free_pkts(q->head);
1614 free(q, M_DUMMYNET);
1619 fs->rq_elements = 0;
1621 /* RED - free lookup table. */
1622 if (fs->w_q_lookup != NULL)
1623 free(fs->w_q_lookup, M_DUMMYNET);
1625 free(fs->rq, M_DUMMYNET);
1626 /* If this fs is not part of a pipe, free it. */
1627 if (fs->pipe == NULL || fs != &(fs->pipe->fs))
1628 free(fs, M_DUMMYNET);
1633 * Dispose all packets queued on a pipe (not a flow_set).
1634 * Also free all resources associated to a pipe, which is about
1638 purge_pipe(struct dn_pipe *pipe)
1641 purge_flow_set( &(pipe->fs), 1 );
1643 dn_free_pkts(pipe->head);
1645 heap_free( &(pipe->scheduler_heap) );
1646 heap_free( &(pipe->not_eligible_heap) );
1647 heap_free( &(pipe->idle_heap) );
1651 * Delete all pipes and heaps returning memory. Must also
1652 * remove references from all ipfw rules to all pipes.
1655 dummynet_flush(void)
1657 struct dn_pipe *pipe, *pipe1;
1658 struct dn_flow_set *fs, *fs1;
1662 /* Free heaps so we don't have unwanted events. */
1663 heap_free(&ready_heap);
1664 heap_free(&wfq_ready_heap);
1665 heap_free(&extract_heap);
1668 * Now purge all queued pkts and delete all pipes.
1670 * XXXGL: can we merge the for(;;) cycles into one or not?
1672 for (i = 0; i < HASHSIZE; i++)
1673 SLIST_FOREACH_SAFE(fs, &flowsethash[i], next, fs1) {
1674 SLIST_REMOVE(&flowsethash[i], fs, dn_flow_set, next);
1675 purge_flow_set(fs, 1);
1677 for (i = 0; i < HASHSIZE; i++)
1678 SLIST_FOREACH_SAFE(pipe, &pipehash[i], next, pipe1) {
1679 SLIST_REMOVE(&pipehash[i], pipe, dn_pipe, next);
1687 * setup RED parameters
1690 config_red(struct dn_flow_set *p, struct dn_flow_set *x)
1695 x->min_th = SCALE(p->min_th);
1696 x->max_th = SCALE(p->max_th);
1697 x->max_p = p->max_p;
1699 x->c_1 = p->max_p / (p->max_th - p->min_th);
1700 x->c_2 = SCALE_MUL(x->c_1, SCALE(p->min_th));
1702 if (x->flags_fs & DN_IS_GENTLE_RED) {
1703 x->c_3 = (SCALE(1) - p->max_p) / p->max_th;
1704 x->c_4 = SCALE(1) - 2 * p->max_p;
1707 /* If the lookup table already exist, free and create it again. */
1708 if (x->w_q_lookup) {
1709 free(x->w_q_lookup, M_DUMMYNET);
1710 x->w_q_lookup = NULL;
1712 if (red_lookup_depth == 0) {
1713 printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1715 free(x, M_DUMMYNET);
1718 x->lookup_depth = red_lookup_depth;
1719 x->w_q_lookup = (u_int *)malloc(x->lookup_depth * sizeof(int),
1720 M_DUMMYNET, M_NOWAIT);
1721 if (x->w_q_lookup == NULL) {
1722 printf("dummynet: sorry, cannot allocate red lookup table\n");
1723 free(x, M_DUMMYNET);
1727 /* Fill the lookup table with (1 - w_q)^x */
1728 x->lookup_step = p->lookup_step;
1729 x->lookup_weight = p->lookup_weight;
1730 x->w_q_lookup[0] = SCALE(1) - x->w_q;
1732 for (i = 1; i < x->lookup_depth; i++)
1734 SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight);
1736 if (red_avg_pkt_size < 1)
1737 red_avg_pkt_size = 512;
1738 x->avg_pkt_size = red_avg_pkt_size;
1739 if (red_max_pkt_size < 1)
1740 red_max_pkt_size = 1500;
1741 x->max_pkt_size = red_max_pkt_size;
1746 alloc_hash(struct dn_flow_set *x, struct dn_flow_set *pfs)
1748 if (x->flags_fs & DN_HAVE_FLOW_MASK) { /* allocate some slots */
1749 int l = pfs->rq_size;
1755 else if (l > DN_MAX_HASH_SIZE)
1756 l = DN_MAX_HASH_SIZE;
1758 } else /* one is enough for null mask */
1760 x->rq = malloc((1 + x->rq_size) * sizeof(struct dn_flow_queue *),
1761 M_DUMMYNET, M_NOWAIT | M_ZERO);
1762 if (x->rq == NULL) {
1763 printf("dummynet: sorry, cannot allocate queue\n");
1771 set_fs_parms(struct dn_flow_set *x, struct dn_flow_set *src)
1773 x->flags_fs = src->flags_fs;
1774 x->qsize = src->qsize;
1776 x->flow_mask = src->flow_mask;
1777 if (x->flags_fs & DN_QSIZE_IS_BYTES) {
1778 if (x->qsize > pipe_byte_limit)
1779 x->qsize = 1024 * 1024;
1783 if (x->qsize > pipe_slot_limit)
1786 /* Configuring RED. */
1787 if (x->flags_fs & DN_IS_RED)
1788 config_red(src, x); /* XXX should check errors */
1792 * Setup pipe or queue parameters.
1795 config_pipe(struct dn_pipe *p)
1797 struct dn_flow_set *pfs = &(p->fs);
1798 struct dn_flow_queue *q;
1802 * The config program passes parameters as follows:
1803 * bw = bits/second (0 means no limits),
1804 * delay = ms, must be translated into ticks.
1805 * qsize = slots/bytes
1807 p->delay = (p->delay * hz) / 1000;
1808 /* Scale burst size: bytes -> bits * hz */
1810 /* We need either a pipe number or a flow_set number. */
1811 if (p->pipe_nr == 0 && pfs->fs_nr == 0)
1813 if (p->pipe_nr != 0 && pfs->fs_nr != 0)
1815 if (p->pipe_nr != 0) { /* this is a pipe */
1816 struct dn_pipe *pipe;
1819 pipe = locate_pipe(p->pipe_nr); /* locate pipe */
1821 if (pipe == NULL) { /* new pipe */
1822 pipe = malloc(sizeof(struct dn_pipe), M_DUMMYNET,
1826 printf("dummynet: no memory for new pipe\n");
1829 pipe->pipe_nr = p->pipe_nr;
1830 pipe->fs.pipe = pipe;
1832 * idle_heap is the only one from which
1833 * we extract from the middle.
1835 pipe->idle_heap.size = pipe->idle_heap.elements = 0;
1836 pipe->idle_heap.offset =
1837 offsetof(struct dn_flow_queue, heap_pos);
1839 /* Flush accumulated credit for all queues. */
1840 for (i = 0; i <= pipe->fs.rq_size; i++)
1841 for (q = pipe->fs.rq[i]; q; q = q->next) {
1842 q->numbytes = p->burst +
1843 (io_fast ? p->bandwidth : 0);
1846 pipe->bandwidth = p->bandwidth;
1847 pipe->burst = p->burst;
1848 pipe->numbytes = pipe->burst + (io_fast ? pipe->bandwidth : 0);
1849 bcopy(p->if_name, pipe->if_name, sizeof(p->if_name));
1850 pipe->ifp = NULL; /* reset interface ptr */
1851 pipe->delay = p->delay;
1852 set_fs_parms(&(pipe->fs), pfs);
1854 /* Handle changes in the delay profile. */
1855 if (p->samples_no > 0) {
1856 if (pipe->samples_no != p->samples_no) {
1857 if (pipe->samples != NULL)
1858 free(pipe->samples, M_DUMMYNET);
1860 malloc(p->samples_no*sizeof(dn_key),
1861 M_DUMMYNET, M_NOWAIT | M_ZERO);
1862 if (pipe->samples == NULL) {
1864 printf("dummynet: no memory "
1865 "for new samples\n");
1868 pipe->samples_no = p->samples_no;
1871 strncpy(pipe->name,p->name,sizeof(pipe->name));
1872 pipe->loss_level = p->loss_level;
1873 for (i = 0; i<pipe->samples_no; ++i)
1874 pipe->samples[i] = p->samples[i];
1875 } else if (pipe->samples != NULL) {
1876 free(pipe->samples, M_DUMMYNET);
1877 pipe->samples = NULL;
1878 pipe->samples_no = 0;
1881 if (pipe->fs.rq == NULL) { /* a new pipe */
1882 error = alloc_hash(&(pipe->fs), pfs);
1888 SLIST_INSERT_HEAD(&pipehash[HASH(pipe->pipe_nr)],
1892 } else { /* config queue */
1893 struct dn_flow_set *fs;
1896 fs = locate_flowset(pfs->fs_nr); /* locate flow_set */
1898 if (fs == NULL) { /* new */
1899 if (pfs->parent_nr == 0) { /* need link to a pipe */
1903 fs = malloc(sizeof(struct dn_flow_set), M_DUMMYNET,
1908 "dummynet: no memory for new flow_set\n");
1911 fs->fs_nr = pfs->fs_nr;
1912 fs->parent_nr = pfs->parent_nr;
1913 fs->weight = pfs->weight;
1914 if (fs->weight == 0)
1916 else if (fs->weight > 100)
1920 * Change parent pipe not allowed;
1921 * must delete and recreate.
1923 if (pfs->parent_nr != 0 &&
1924 fs->parent_nr != pfs->parent_nr) {
1930 set_fs_parms(fs, pfs);
1932 if (fs->rq == NULL) { /* a new flow_set */
1933 error = alloc_hash(fs, pfs);
1936 free(fs, M_DUMMYNET);
1939 SLIST_INSERT_HEAD(&flowsethash[HASH(fs->fs_nr)],
1948 * Helper function to remove from a heap queues which are linked to
1949 * a flow_set about to be deleted.
1952 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs)
1954 int i = 0, found = 0 ;
1955 for (; i < h->elements ;)
1956 if ( ((struct dn_flow_queue *)h->p[i].object)->fs == fs) {
1958 h->p[i] = h->p[h->elements] ;
1967 * helper function to remove a pipe from a heap (can be there at most once)
1970 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p)
1972 if (h->elements > 0) {
1974 for (i=0; i < h->elements ; i++ ) {
1975 if (h->p[i].object == p) { /* found it */
1977 h->p[i] = h->p[h->elements] ;
1986 * drain all queues. Called in case of severe mbuf shortage.
1989 dummynet_drain(void)
1991 struct dn_flow_set *fs;
1992 struct dn_pipe *pipe;
1995 DUMMYNET_LOCK_ASSERT();
1997 heap_free(&ready_heap);
1998 heap_free(&wfq_ready_heap);
1999 heap_free(&extract_heap);
2000 /* remove all references to this pipe from flow_sets */
2001 for (i = 0; i < HASHSIZE; i++)
2002 SLIST_FOREACH(fs, &flowsethash[i], next)
2003 purge_flow_set(fs, 0);
2005 for (i = 0; i < HASHSIZE; i++) {
2006 SLIST_FOREACH(pipe, &pipehash[i], next) {
2007 purge_flow_set(&(pipe->fs), 0);
2008 dn_free_pkts(pipe->head);
2009 pipe->head = pipe->tail = NULL;
2015 * Fully delete a pipe or a queue, cleaning up associated info.
2018 delete_pipe(struct dn_pipe *p)
2021 if (p->pipe_nr == 0 && p->fs.fs_nr == 0)
2023 if (p->pipe_nr != 0 && p->fs.fs_nr != 0)
2025 if (p->pipe_nr != 0) { /* this is an old-style pipe */
2026 struct dn_pipe *pipe;
2027 struct dn_flow_set *fs;
2031 pipe = locate_pipe(p->pipe_nr); /* locate pipe */
2035 return (ENOENT); /* not found */
2038 /* Unlink from list of pipes. */
2039 SLIST_REMOVE(&pipehash[HASH(pipe->pipe_nr)], pipe, dn_pipe, next);
2041 /* Remove all references to this pipe from flow_sets. */
2042 for (i = 0; i < HASHSIZE; i++)
2043 SLIST_FOREACH(fs, &flowsethash[i], next)
2044 if (fs->pipe == pipe) {
2045 printf("dummynet: ++ ref to pipe %d from fs %d\n",
2046 p->pipe_nr, fs->fs_nr);
2048 purge_flow_set(fs, 0);
2050 fs_remove_from_heap(&ready_heap, &(pipe->fs));
2051 purge_pipe(pipe); /* remove all data associated to this pipe */
2052 /* remove reference to here from extract_heap and wfq_ready_heap */
2053 pipe_remove_from_heap(&extract_heap, pipe);
2054 pipe_remove_from_heap(&wfq_ready_heap, pipe);
2058 } else { /* this is a WF2Q queue (dn_flow_set) */
2059 struct dn_flow_set *fs;
2062 fs = locate_flowset(p->fs.fs_nr); /* locate set */
2066 return (ENOENT); /* not found */
2069 /* Unlink from list of flowsets. */
2070 SLIST_REMOVE( &flowsethash[HASH(fs->fs_nr)], fs, dn_flow_set, next);
2072 if (fs->pipe != NULL) {
2073 /* Update total weight on parent pipe and cleanup parent heaps. */
2074 fs->pipe->sum -= fs->weight * fs->backlogged ;
2075 fs_remove_from_heap(&(fs->pipe->not_eligible_heap), fs);
2076 fs_remove_from_heap(&(fs->pipe->scheduler_heap), fs);
2077 #if 1 /* XXX should i remove from idle_heap as well ? */
2078 fs_remove_from_heap(&(fs->pipe->idle_heap), fs);
2081 purge_flow_set(fs, 1);
2088 * helper function used to copy data from kernel in DUMMYNET_GET
2091 dn_copy_set(struct dn_flow_set *set, char *bp)
2094 struct dn_flow_queue *q, *qp = (struct dn_flow_queue *)bp;
2096 DUMMYNET_LOCK_ASSERT();
2098 for (i = 0 ; i <= set->rq_size ; i++)
2099 for (q = set->rq[i] ; q ; q = q->next, qp++ ) {
2100 if (q->hash_slot != i)
2101 printf("dummynet: ++ at %d: wrong slot (have %d, "
2102 "should be %d)\n", copied, q->hash_slot, i);
2104 printf("dummynet: ++ at %d: wrong fs ptr (have %p, should be %p)\n",
2107 bcopy(q, qp, sizeof( *q ) );
2108 /* cleanup pointers */
2110 qp->head = qp->tail = NULL ;
2113 if (copied != set->rq_elements)
2114 printf("dummynet: ++ wrong count, have %d should be %d\n",
2115 copied, set->rq_elements);
2122 struct dn_flow_set *fs;
2123 struct dn_pipe *pipe;
2127 DUMMYNET_LOCK_ASSERT();
2129 * Compute size of data structures: list of pipes and flow_sets.
2131 for (i = 0; i < HASHSIZE; i++) {
2132 SLIST_FOREACH(pipe, &pipehash[i], next)
2133 size += sizeof(*pipe) +
2134 pipe->fs.rq_elements * sizeof(struct dn_flow_queue);
2135 SLIST_FOREACH(fs, &flowsethash[i], next)
2136 size += sizeof (*fs) +
2137 fs->rq_elements * sizeof(struct dn_flow_queue);
2143 dummynet_get(struct sockopt *sopt)
2145 char *buf, *bp ; /* bp is the "copy-pointer" */
2147 struct dn_flow_set *fs;
2148 struct dn_pipe *pipe;
2151 /* XXX lock held too long */
2154 * XXX: Ugly, but we need to allocate memory with M_WAITOK flag and we
2155 * cannot use this flag while holding a mutex.
2157 for (i = 0; i < 10; i++) {
2158 size = dn_calc_size();
2160 buf = malloc(size, M_TEMP, M_WAITOK);
2162 if (size == dn_calc_size())
2172 for (i = 0; i < HASHSIZE; i++)
2173 SLIST_FOREACH(pipe, &pipehash[i], next) {
2174 struct dn_pipe *pipe_bp = (struct dn_pipe *)bp;
2177 * Copy pipe descriptor into *bp, convert delay back to ms,
2178 * then copy the flow_set descriptor(s) one at a time.
2179 * After each flow_set, copy the queue descriptor it owns.
2181 bcopy(pipe, bp, sizeof(*pipe));
2182 pipe_bp->delay = (pipe_bp->delay * 1000) / hz;
2183 pipe_bp->burst = div64(pipe_bp->burst, 8 * hz);
2185 * XXX the following is a hack based on ->next being the
2186 * first field in dn_pipe and dn_flow_set. The correct
2187 * solution would be to move the dn_flow_set to the beginning
2188 * of struct dn_pipe.
2190 pipe_bp->next.sle_next = (struct dn_pipe *)DN_IS_PIPE;
2191 /* Clean pointers. */
2192 pipe_bp->head = pipe_bp->tail = NULL;
2193 pipe_bp->fs.next.sle_next = NULL;
2194 pipe_bp->fs.pipe = NULL;
2195 pipe_bp->fs.rq = NULL;
2196 pipe_bp->samples = NULL;
2198 bp += sizeof(*pipe) ;
2199 bp = dn_copy_set(&(pipe->fs), bp);
2202 for (i = 0; i < HASHSIZE; i++)
2203 SLIST_FOREACH(fs, &flowsethash[i], next) {
2204 struct dn_flow_set *fs_bp = (struct dn_flow_set *)bp;
2206 bcopy(fs, bp, sizeof(*fs));
2207 /* XXX same hack as above */
2208 fs_bp->next.sle_next = (struct dn_flow_set *)DN_IS_QUEUE;
2212 bp = dn_copy_set(fs, bp);
2217 error = sooptcopyout(sopt, buf, size);
2223 * Handler for the various dummynet socket options (get, flush, config, del)
2226 ip_dn_ctl(struct sockopt *sopt)
2229 struct dn_pipe *p = NULL;
2231 error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2235 /* Disallow sets in really-really secure mode. */
2236 if (sopt->sopt_dir == SOPT_SET) {
2237 #if __FreeBSD_version >= 500034
2238 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2242 if (securelevel >= 3)
2247 switch (sopt->sopt_name) {
2249 printf("dummynet: -- unknown option %d", sopt->sopt_name);
2253 case IP_DUMMYNET_GET :
2254 error = dummynet_get(sopt);
2257 case IP_DUMMYNET_FLUSH :
2261 case IP_DUMMYNET_CONFIGURE :
2262 p = malloc(sizeof(struct dn_pipe_max), M_TEMP, M_WAITOK);
2263 error = sooptcopyin(sopt, p, sizeof(struct dn_pipe_max), sizeof *p);
2266 if (p->samples_no > 0)
2267 p->samples = &( ((struct dn_pipe_max*) p)->samples[0] );
2269 error = config_pipe(p);
2272 case IP_DUMMYNET_DEL : /* remove a pipe or queue */
2273 p = malloc(sizeof(struct dn_pipe), M_TEMP, M_WAITOK);
2274 error = sooptcopyin(sopt, p, sizeof (struct dn_pipe), sizeof *p);
2278 error = delete_pipe(p);
2294 printf("DUMMYNET with IPv6 initialized (040826)\n");
2296 DUMMYNET_LOCK_INIT();
2298 for (i = 0; i < HASHSIZE; i++) {
2299 SLIST_INIT(&pipehash[i]);
2300 SLIST_INIT(&flowsethash[i]);
2302 ready_heap.size = ready_heap.elements = 0;
2303 ready_heap.offset = 0;
2305 wfq_ready_heap.size = wfq_ready_heap.elements = 0;
2306 wfq_ready_heap.offset = 0;
2308 extract_heap.size = extract_heap.elements = 0;
2309 extract_heap.offset = 0;
2311 ip_dn_ctl_ptr = ip_dn_ctl;
2312 ip_dn_io_ptr = dummynet_io;
2314 TASK_INIT(&dn_task, 0, dummynet_task, NULL);
2315 dn_tq = taskqueue_create_fast("dummynet", M_NOWAIT,
2316 taskqueue_thread_enqueue, &dn_tq);
2317 taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2319 callout_init(&dn_timeout, CALLOUT_MPSAFE);
2320 callout_reset(&dn_timeout, 1, dummynet, NULL);
2322 /* Initialize curr_time adjustment mechanics. */
2323 getmicrouptime(&prev_t);
2330 ip_dn_ctl_ptr = NULL;
2331 ip_dn_io_ptr = NULL;
2334 callout_stop(&dn_timeout);
2336 taskqueue_drain(dn_tq, &dn_task);
2337 taskqueue_free(dn_tq);
2341 DUMMYNET_LOCK_DESTROY();
2343 #endif /* KLD_MODULE */
2346 dummynet_modevent(module_t mod, int type, void *data)
2352 printf("DUMMYNET already loaded\n");
2359 #if !defined(KLD_MODULE)
2360 printf("dummynet statically compiled, cannot unload\n");
2373 static moduledata_t dummynet_mod = {
2378 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
2379 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
2380 MODULE_VERSION(dummynet, 1);