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
59 #include <sys/limits.h>
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/malloc.h>
64 #include <sys/kernel.h>
66 #include <sys/module.h>
67 #include <sys/mutex.h>
70 #include <sys/socket.h>
71 #include <sys/socketvar.h>
73 #include <sys/sysctl.h>
74 #include <sys/taskqueue.h>
75 #include <net/if.h> /* IFNAMSIZ, struct ifaddr, ifq head */
76 #include <net/netisr.h>
77 #include <netinet/in.h>
78 #include <netinet/ip.h> /* ip_len, ip_off */
79 #include <netinet/ip_fw.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>
91 * We keep a private variable for the simulation time, but we could
92 * probably use an existing one ("softticks" in sys/kern/kern_timeout.c)
94 static dn_key curr_time = 0 ; /* current simulation time */
96 static int dn_hash_size = 64 ; /* default hash size */
98 /* statistics on number of queue searches and search steps */
99 static long searches, search_steps ;
100 static int pipe_expire = 1 ; /* expire queue if empty */
101 static int dn_max_ratio = 16 ; /* max queues/buckets ratio */
103 static long pipe_slot_limit = 100; /* Foot shooting limit for pipe queues. */
104 static long pipe_byte_limit = 1024 * 1024;
106 static int red_lookup_depth = 256; /* RED - default lookup table depth */
107 static int red_avg_pkt_size = 512; /* RED - default medium packet size */
108 static int red_max_pkt_size = 1500; /* RED - default max packet size */
110 static struct timeval prev_t, t;
111 static long tick_last; /* Last tick duration (usec). */
112 static long tick_delta; /* Last vs standard tick diff (usec). */
113 static long tick_delta_sum; /* Accumulated tick difference (usec).*/
114 static long tick_adjustment; /* Tick adjustments done. */
115 static long tick_lost; /* Lost(coalesced) ticks number. */
116 /* Adjusted vs non-adjusted curr_time difference (ticks). */
117 static long tick_diff;
120 static unsigned long io_pkt;
121 static unsigned long io_pkt_fast;
122 static unsigned long io_pkt_drop;
125 * Three heaps contain queues and pipes that the scheduler handles:
127 * ready_heap contains all dn_flow_queue related to fixed-rate pipes.
129 * wfq_ready_heap contains the pipes associated with WF2Q flows
131 * extract_heap contains pipes associated with delay lines.
135 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap");
137 static struct dn_heap ready_heap, extract_heap, wfq_ready_heap ;
139 static int heap_init(struct dn_heap *h, int size);
140 static int heap_insert (struct dn_heap *h, dn_key key1, void *p);
141 static void heap_extract(struct dn_heap *h, void *obj);
142 static void transmit_event(struct dn_pipe *pipe, struct mbuf **head,
144 static void ready_event(struct dn_flow_queue *q, struct mbuf **head,
146 static void ready_event_wfq(struct dn_pipe *p, struct mbuf **head,
150 #define HASH(num) ((((num) >> 8) ^ ((num) >> 4) ^ (num)) & 0x0f)
151 static struct dn_pipe_head pipehash[HASHSIZE]; /* all pipes */
152 static struct dn_flow_set_head flowsethash[HASHSIZE]; /* all flowsets */
154 static struct callout dn_timeout;
156 extern void (*bridge_dn_p)(struct mbuf *, struct ifnet *);
159 SYSCTL_DECL(_net_inet);
160 SYSCTL_DECL(_net_inet_ip);
162 SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet, CTLFLAG_RW, 0, "Dummynet");
163 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, hash_size,
164 CTLFLAG_RW, &dn_hash_size, 0, "Default hash table size");
165 #if 0 /* curr_time is 64 bit */
166 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, curr_time,
167 CTLFLAG_RD, &curr_time, 0, "Current tick");
169 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, ready_heap,
170 CTLFLAG_RD, &ready_heap.size, 0, "Size of ready heap");
171 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, extract_heap,
172 CTLFLAG_RD, &extract_heap.size, 0, "Size of extract heap");
173 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, searches,
174 CTLFLAG_RD, &searches, 0, "Number of queue searches");
175 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, search_steps,
176 CTLFLAG_RD, &search_steps, 0, "Number of queue search steps");
177 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, expire,
178 CTLFLAG_RW, &pipe_expire, 0, "Expire queue if empty");
179 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, max_chain_len,
180 CTLFLAG_RW, &dn_max_ratio, 0,
181 "Max ratio between dynamic queues and buckets");
182 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth,
183 CTLFLAG_RD, &red_lookup_depth, 0, "Depth of RED lookup table");
184 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size,
185 CTLFLAG_RD, &red_avg_pkt_size, 0, "RED Medium packet size");
186 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size,
187 CTLFLAG_RD, &red_max_pkt_size, 0, "RED Max packet size");
188 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta,
189 CTLFLAG_RD, &tick_delta, 0, "Last vs standard tick difference (usec).");
190 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta_sum,
191 CTLFLAG_RD, &tick_delta_sum, 0, "Accumulated tick difference (usec).");
192 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_adjustment,
193 CTLFLAG_RD, &tick_adjustment, 0, "Tick adjustments done.");
194 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_diff,
195 CTLFLAG_RD, &tick_diff, 0,
196 "Adjusted vs non-adjusted curr_time difference (ticks).");
197 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_lost,
198 CTLFLAG_RD, &tick_lost, 0,
199 "Number of ticks coalesced by dummynet taskqueue.");
200 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, io_fast,
201 CTLFLAG_RW, &io_fast, 0, "Enable fast dummynet io.");
202 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt,
203 CTLFLAG_RD, &io_pkt, 0,
204 "Number of packets passed to dummynet.");
205 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_fast,
206 CTLFLAG_RD, &io_pkt_fast, 0,
207 "Number of packets bypassed dummynet scheduler.");
208 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_drop,
209 CTLFLAG_RD, &io_pkt_drop, 0,
210 "Number of packets dropped by dummynet.");
211 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, pipe_slot_limit,
212 CTLFLAG_RW, &pipe_slot_limit, 0, "Upper limit in slots for pipe queue.");
213 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, pipe_byte_limit,
214 CTLFLAG_RW, &pipe_byte_limit, 0, "Upper limit in bytes for pipe queue.");
217 #ifdef DUMMYNET_DEBUG
218 int dummynet_debug = 0;
220 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, debug, CTLFLAG_RW, &dummynet_debug,
221 0, "control debugging printfs");
223 #define DPRINTF(X) if (dummynet_debug) printf X
228 static struct task dn_task;
229 static struct taskqueue *dn_tq = NULL;
230 static void dummynet_task(void *, int);
232 #if defined( __linux__ ) || defined( _WIN32 )
233 static DEFINE_SPINLOCK(dummynet_mtx);
235 static struct mtx dummynet_mtx;
237 #define DUMMYNET_LOCK_INIT() \
238 mtx_init(&dummynet_mtx, "dummynet", NULL, MTX_DEF)
239 #define DUMMYNET_LOCK_DESTROY() mtx_destroy(&dummynet_mtx)
240 #define DUMMYNET_LOCK() mtx_lock(&dummynet_mtx)
241 #define DUMMYNET_UNLOCK() mtx_unlock(&dummynet_mtx)
242 #define DUMMYNET_LOCK_ASSERT() mtx_assert(&dummynet_mtx, MA_OWNED)
244 static int config_pipe(struct dn_pipe *p);
245 static int ip_dn_ctl(struct sockopt *sopt);
247 static void dummynet(void *);
248 static void dummynet_flush(void);
249 static void dummynet_send(struct mbuf *);
250 void dummynet_drain(void);
251 static ip_dn_io_t dummynet_io;
252 static void dn_rule_delete(void *);
255 * Heap management functions.
257 * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2.
258 * Some macros help finding parent/children so we can optimize them.
260 * heap_init() is called to expand the heap when needed.
261 * Increment size in blocks of 16 entries.
262 * XXX failure to allocate a new element is a pretty bad failure
263 * as we basically stall a whole queue forever!!
264 * Returns 1 on error, 0 on success
266 #define HEAP_FATHER(x) ( ( (x) - 1 ) / 2 )
267 #define HEAP_LEFT(x) ( 2*(x) + 1 )
268 #define HEAP_IS_LEFT(x) ( (x) & 1 )
269 #define HEAP_RIGHT(x) ( 2*(x) + 2 )
270 #define HEAP_SWAP(a, b, buffer) { buffer = a ; a = b ; b = buffer ; }
271 #define HEAP_INCREMENT 15
274 heap_init(struct dn_heap *h, int new_size)
276 struct dn_heap_entry *p;
278 if (h->size >= new_size ) {
279 printf("dummynet: %s, Bogus call, have %d want %d\n", __func__,
283 new_size = (new_size + HEAP_INCREMENT ) & ~HEAP_INCREMENT ;
284 p = malloc(new_size * sizeof(*p), M_DUMMYNET, M_NOWAIT);
286 printf("dummynet: %s, resize %d failed\n", __func__, new_size );
287 return 1 ; /* error */
290 bcopy(h->p, p, h->size * sizeof(*p) );
291 free(h->p, M_DUMMYNET);
299 * Insert element in heap. Normally, p != NULL, we insert p in
300 * a new position and bubble up. If p == NULL, then the element is
301 * already in place, and key is the position where to start the
303 * Returns 1 on failure (cannot allocate new heap entry)
305 * If offset > 0 the position (index, int) of the element in the heap is
306 * also stored in the element itself at the given offset in bytes.
308 #define SET_OFFSET(heap, node) \
309 if (heap->offset > 0) \
310 *((int *)((char *)(heap->p[node].object) + heap->offset)) = node ;
312 * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value.
314 #define RESET_OFFSET(heap, node) \
315 if (heap->offset > 0) \
316 *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1 ;
318 heap_insert(struct dn_heap *h, dn_key key1, void *p)
320 int son = h->elements ;
322 if (p == NULL) /* data already there, set starting point */
324 else { /* insert new element at the end, possibly resize */
326 if (son == h->size) /* need resize... */
327 if (heap_init(h, h->elements+1) )
328 return 1 ; /* failure... */
329 h->p[son].object = p ;
330 h->p[son].key = key1 ;
333 while (son > 0) { /* bubble up */
334 int father = HEAP_FATHER(son) ;
335 struct dn_heap_entry tmp ;
337 if (DN_KEY_LT( h->p[father].key, h->p[son].key ) )
338 break ; /* found right position */
339 /* son smaller than father, swap and repeat */
340 HEAP_SWAP(h->p[son], h->p[father], tmp) ;
349 * remove top element from heap, or obj if obj != NULL
352 heap_extract(struct dn_heap *h, void *obj)
354 int child, father, max = h->elements - 1 ;
357 printf("dummynet: warning, extract from empty heap 0x%p\n", h);
360 father = 0 ; /* default: move up smallest child */
361 if (obj != NULL) { /* extract specific element, index is at offset */
363 panic("dummynet: heap_extract from middle not supported on this heap!!!\n");
364 father = *((int *)((char *)obj + h->offset)) ;
365 if (father < 0 || father >= h->elements) {
366 printf("dummynet: heap_extract, father %d out of bound 0..%d\n",
367 father, h->elements);
368 panic("dummynet: heap_extract");
371 RESET_OFFSET(h, father);
372 child = HEAP_LEFT(father) ; /* left child */
373 while (child <= max) { /* valid entry */
374 if (child != max && DN_KEY_LT(h->p[child+1].key, h->p[child].key) )
375 child = child+1 ; /* take right child, otherwise left */
376 h->p[father] = h->p[child] ;
377 SET_OFFSET(h, father);
379 child = HEAP_LEFT(child) ; /* left child for next loop */
384 * Fill hole with last entry and bubble up, reusing the insert code
386 h->p[father] = h->p[max] ;
387 heap_insert(h, father, NULL); /* this one cannot fail */
393 * change object position and update references
394 * XXX this one is never used!
397 heap_move(struct dn_heap *h, dn_key new_key, void *object)
401 int max = h->elements-1 ;
402 struct dn_heap_entry buf ;
405 panic("cannot move items on this heap");
407 i = *((int *)((char *)object + h->offset));
408 if (DN_KEY_LT(new_key, h->p[i].key) ) { /* must move up */
409 h->p[i].key = new_key ;
410 for (; i>0 && DN_KEY_LT(new_key, h->p[(temp = HEAP_FATHER(i))].key) ;
411 i = temp ) { /* bubble up */
412 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
415 } else { /* must move down */
416 h->p[i].key = new_key ;
417 while ( (temp = HEAP_LEFT(i)) <= max ) { /* found left child */
418 if ((temp != max) && DN_KEY_GT(h->p[temp].key, h->p[temp+1].key))
419 temp++ ; /* select child with min key */
420 if (DN_KEY_GT(new_key, h->p[temp].key)) { /* go down */
421 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
430 #endif /* heap_move, unused */
433 * heapify() will reorganize data inside an array to maintain the
434 * heap property. It is needed when we delete a bunch of entries.
437 heapify(struct dn_heap *h)
441 for (i = 0 ; i < h->elements ; i++ )
442 heap_insert(h, i , NULL) ;
446 * cleanup the heap and free data structure
449 heap_free(struct dn_heap *h)
452 free(h->p, M_DUMMYNET);
453 bzero(h, sizeof(*h) );
457 * --- end of heap management functions ---
461 * Return the mbuf tag holding the dummynet state. As an optimization
462 * this is assumed to be the first tag on the list. If this turns out
463 * wrong we'll need to search the list.
465 static struct dn_pkt_tag *
466 dn_tag_get(struct mbuf *m)
468 struct m_tag *mtag = m_tag_first(m);
469 KASSERT(mtag != NULL &&
470 mtag->m_tag_cookie == MTAG_ABI_COMPAT &&
471 mtag->m_tag_id == PACKET_TAG_DUMMYNET,
472 ("packet on dummynet queue w/o dummynet tag!"));
473 return (struct dn_pkt_tag *)(mtag+1);
477 * Scheduler functions:
479 * transmit_event() is called when the delay-line needs to enter
480 * the scheduler, either because of existing pkts getting ready,
481 * or new packets entering the queue. The event handled is the delivery
482 * time of the packet.
484 * ready_event() does something similar with fixed-rate queues, and the
485 * event handled is the finish time of the head pkt.
487 * wfq_ready_event() does something similar with WF2Q queues, and the
488 * event handled is the start time of the head pkt.
490 * In all cases, we make sure that the data structures are consistent
491 * before passing pkts out, because this might trigger recursive
492 * invocations of the procedures.
495 transmit_event(struct dn_pipe *pipe, struct mbuf **head, struct mbuf **tail)
498 struct dn_pkt_tag *pkt;
500 DUMMYNET_LOCK_ASSERT();
502 while ((m = pipe->head) != NULL) {
504 if (!DN_KEY_LEQ(pkt->output_time, curr_time))
507 pipe->head = m->m_nextpkt;
509 (*tail)->m_nextpkt = m;
515 (*tail)->m_nextpkt = NULL;
517 /* If there are leftover packets, put into the heap for next event. */
518 if ((m = pipe->head) != NULL) {
521 * XXX Should check errors on heap_insert, by draining the
522 * whole pipe p and hoping in the future we are more successful.
524 heap_insert(&extract_heap, pkt->output_time, pipe);
529 #define div64(a, b) ((int64_t)(a) / (int64_t)(b))
531 #define DN_TO_DROP 0xffff
533 * Compute how many ticks we have to wait before being able to send
534 * a packet. This is computed as the "wire time" for the packet
535 * (length + extra bits), minus the credit available, scaled to ticks.
536 * Check that the result is not be negative (it could be if we have
537 * too much leftover credit in q->numbytes).
540 set_ticks(struct mbuf *m, struct dn_flow_queue *q, struct dn_pipe *p)
544 ret = div64( (m->m_pkthdr.len * 8 + q->extra_bits) * hz
545 - q->numbytes + p->bandwidth - 1 , p->bandwidth);
547 printf("%s %d extra_bits %d numb %d ret %d\n",
548 __FUNCTION__, __LINE__,
549 (int)(q->extra_bits & 0xffffffff),
550 (int)(q->numbytes & 0xffffffff),
551 (int)(ret & 0xffffffff));
559 * Convert the additional MAC overheads/delays into an equivalent
560 * number of bits for the given data rate. The samples are in milliseconds
561 * so we need to divide by 1000.
564 compute_extra_bits(struct mbuf *pkt, struct dn_pipe *p)
569 if (!p->samples || p->samples_no == 0)
571 index = random() % p->samples_no;
572 extra_bits = div64((dn_key)p->samples[index] * p->bandwidth, 1000);
573 if (index >= p->loss_level) {
574 struct dn_pkt_tag *dt = dn_tag_get(pkt);
576 dt->dn_dir = DN_TO_DROP;
582 free_pipe(struct dn_pipe *p)
585 free(p->samples, M_DUMMYNET);
590 * extract pkt from queue, compute output time (could be now)
591 * and put into delay line (p_queue)
594 move_pkt(struct mbuf *pkt, struct dn_flow_queue *q, struct dn_pipe *p,
597 struct dn_pkt_tag *dt = dn_tag_get(pkt);
599 q->head = pkt->m_nextpkt ;
601 q->len_bytes -= len ;
603 dt->output_time = curr_time + p->delay ;
608 p->tail->m_nextpkt = pkt;
610 p->tail->m_nextpkt = NULL;
614 * ready_event() is invoked every time the queue must enter the
615 * scheduler, either because the first packet arrives, or because
616 * a previously scheduled event fired.
617 * On invokation, drain as many pkts as possible (could be 0) and then
618 * if there are leftover packets reinsert the pkt in the scheduler.
621 ready_event(struct dn_flow_queue *q, struct mbuf **head, struct mbuf **tail)
624 struct dn_pipe *p = q->fs->pipe;
627 DUMMYNET_LOCK_ASSERT();
630 printf("dummynet: ready_event- pipe is gone\n");
633 p_was_empty = (p->head == NULL);
636 * Schedule fixed-rate queues linked to this pipe:
637 * account for the bw accumulated since last scheduling, then
638 * drain as many pkts as allowed by q->numbytes and move to
639 * the delay line (in p) computing output time.
640 * bandwidth==0 (no limit) means we can drain the whole queue,
641 * setting len_scaled = 0 does the job.
643 q->numbytes += (curr_time - q->sched_time) * p->bandwidth;
644 while ((pkt = q->head) != NULL) {
645 int len = pkt->m_pkthdr.len;
646 dn_key len_scaled = p->bandwidth ? len*8*hz
650 if (DN_KEY_GT(len_scaled, q->numbytes))
652 q->numbytes -= len_scaled;
653 move_pkt(pkt, q, p, len);
655 q->extra_bits = compute_extra_bits(q->head, p);
658 * If we have more packets queued, schedule next ready event
659 * (can only occur when bandwidth != 0, otherwise we would have
660 * flushed the whole queue in the previous loop).
661 * To this purpose we record the current time and compute how many
662 * ticks to go for the finish time of the packet.
664 if ((pkt = q->head) != NULL) { /* this implies bandwidth != 0 */
665 dn_key t = set_ticks(pkt, q, p); /* ticks i have to wait */
667 q->sched_time = curr_time;
668 heap_insert(&ready_heap, curr_time + t, (void *)q);
670 * XXX Should check errors on heap_insert, and drain the whole
671 * queue on error hoping next time we are luckier.
673 } else /* RED needs to know when the queue becomes empty. */
674 q->q_time = curr_time;
677 * If the delay line was empty call transmit_event() now.
678 * Otherwise, the scheduler will take care of it.
681 transmit_event(p, head, tail);
685 * Called when we can transmit packets on WF2Q queues. Take pkts out of
686 * the queues at their start time, and enqueue into the delay line.
687 * Packets are drained until p->numbytes < 0. As long as
688 * len_scaled >= p->numbytes, the packet goes into the delay line
689 * with a deadline p->delay. For the last packet, if p->numbytes < 0,
690 * there is an additional delay.
693 ready_event_wfq(struct dn_pipe *p, struct mbuf **head, struct mbuf **tail)
695 int p_was_empty = (p->head == NULL);
696 struct dn_heap *sch = &(p->scheduler_heap);
697 struct dn_heap *neh = &(p->not_eligible_heap);
698 int64_t p_numbytes = p->numbytes;
700 DUMMYNET_LOCK_ASSERT();
702 if (p->if_name[0] == 0) /* tx clock is simulated */
704 * Since result may not fit into p->numbytes (32bit) we
705 * are using 64bit var here.
707 p_numbytes += (curr_time - p->sched_time) * p->bandwidth;
709 * tx clock is for real,
710 * the ifq must be empty or this is a NOP.
711 * XXX not supported in Linux
713 if (1) // p->ifp && p->ifp->if_snd.ifq_head != NULL)
716 DPRINTF(("dummynet: pipe %d ready from %s --\n",
717 p->pipe_nr, p->if_name));
722 * While we have backlogged traffic AND credit, we need to do
723 * something on the queue.
725 while (p_numbytes >= 0 && (sch->elements > 0 || neh->elements > 0)) {
726 if (sch->elements > 0) {
727 /* Have some eligible pkts to send out. */
728 struct dn_flow_queue *q = sch->p[0].object;
729 struct mbuf *pkt = q->head;
730 struct dn_flow_set *fs = q->fs;
731 uint64_t len = pkt->m_pkthdr.len;
732 int len_scaled = p->bandwidth ? len * 8 * hz : 0;
734 heap_extract(sch, NULL); /* Remove queue from heap. */
735 p_numbytes -= len_scaled;
736 move_pkt(pkt, q, p, len);
738 p->V += div64((len << MY_M), p->sum); /* Update V. */
739 q->S = q->F; /* Update start time. */
741 /* Flow not backlogged any more. */
743 heap_insert(&(p->idle_heap), q->F, q);
745 /* Still backlogged. */
748 * Update F and position in backlogged queue,
749 * then put flow in not_eligible_heap
750 * (we will fix this later).
752 len = (q->head)->m_pkthdr.len;
753 q->F += div64((len << MY_M), fs->weight);
754 if (DN_KEY_LEQ(q->S, p->V))
755 heap_insert(neh, q->S, q);
757 heap_insert(sch, q->F, q);
761 * Now compute V = max(V, min(S_i)). Remember that all elements
762 * in sch have by definition S_i <= V so if sch is not empty,
763 * V is surely the max and we must not update it. Conversely,
764 * if sch is empty we only need to look at neh.
766 if (sch->elements == 0 && neh->elements > 0)
767 p->V = MAX64(p->V, neh->p[0].key);
768 /* Move from neh to sch any packets that have become eligible */
769 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V)) {
770 struct dn_flow_queue *q = neh->p[0].object;
771 heap_extract(neh, NULL);
772 heap_insert(sch, q->F, q);
775 if (p->if_name[0] != '\0') { /* Tx clock is from a real thing */
776 p_numbytes = -1; /* Mark not ready for I/O. */
780 if (sch->elements == 0 && neh->elements == 0 && p_numbytes >= 0 &&
781 p->idle_heap.elements > 0) {
783 * No traffic and no events scheduled.
784 * We can get rid of idle-heap.
788 for (i = 0; i < p->idle_heap.elements; i++) {
789 struct dn_flow_queue *q = p->idle_heap.p[i].object;
796 p->idle_heap.elements = 0;
799 * If we are getting clocks from dummynet (not a real interface) and
800 * If we are under credit, schedule the next ready event.
801 * Also fix the delivery time of the last packet.
803 if (p->if_name[0]==0 && p_numbytes < 0) { /* This implies bw > 0. */
804 dn_key t = 0; /* Number of ticks i have to wait. */
806 if (p->bandwidth > 0)
807 t = div64(p->bandwidth - 1 - p_numbytes, p->bandwidth);
808 dn_tag_get(p->tail)->output_time += t;
809 p->sched_time = curr_time;
810 heap_insert(&wfq_ready_heap, curr_time + t, (void *)p);
812 * XXX Should check errors on heap_insert, and drain the whole
813 * queue on error hoping next time we are luckier.
817 /* Fit (adjust if necessary) 64bit result into 32bit variable. */
818 if (p_numbytes > INT_MAX)
819 p->numbytes = INT_MAX;
820 else if (p_numbytes < INT_MIN)
821 p->numbytes = INT_MIN;
823 p->numbytes = p_numbytes;
826 * If the delay line was empty call transmit_event() now.
827 * Otherwise, the scheduler will take care of it.
830 transmit_event(p, head, tail);
834 * This is called one tick, after previous run. It is used to
838 dummynet(void * __unused unused)
841 taskqueue_enqueue(dn_tq, &dn_task);
845 * The main dummynet processing function.
848 dummynet_task(void *context, int pending)
850 struct mbuf *head = NULL, *tail = NULL;
851 struct dn_pipe *pipe;
852 struct dn_heap *heaps[3];
854 void *p; /* generic parameter to handler */
859 heaps[0] = &ready_heap; /* fixed-rate queues */
860 heaps[1] = &wfq_ready_heap; /* wfq queues */
861 heaps[2] = &extract_heap; /* delay line */
863 /* Update number of lost(coalesced) ticks. */
864 tick_lost += pending - 1;
867 /* Last tick duration (usec). */
868 tick_last = (t.tv_sec - prev_t.tv_sec) * 1000000 +
869 (t.tv_usec - prev_t.tv_usec);
870 /* Last tick vs standard tick difference (usec). */
871 tick_delta = (tick_last * hz - 1000000) / hz;
872 /* Accumulated tick difference (usec). */
873 tick_delta_sum += tick_delta;
878 * Adjust curr_time if accumulated tick difference greater than
879 * 'standard' tick. Since curr_time should be monotonically increasing,
880 * we do positive adjustment as required and throttle curr_time in
881 * case of negative adjustment.
884 if (tick_delta_sum - tick >= 0) {
885 int diff = tick_delta_sum / tick;
889 tick_delta_sum %= tick;
891 } else if (tick_delta_sum + tick <= 0) {
894 tick_delta_sum += tick;
898 for (i = 0; i < 3; i++) {
900 while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time)) {
901 if (h->p[0].key > curr_time)
902 printf("dummynet: warning, "
903 "heap %d is %d ticks late\n",
904 i, (int)(curr_time - h->p[0].key));
905 /* store a copy before heap_extract */
907 /* need to extract before processing */
908 heap_extract(h, NULL);
910 ready_event(p, &head, &tail);
912 struct dn_pipe *pipe = p;
913 if (pipe->if_name[0] != '\0')
914 printf("dummynet: bad ready_event_wfq "
915 "for pipe %s\n", pipe->if_name);
917 ready_event_wfq(p, &head, &tail);
919 transmit_event(p, &head, &tail);
923 /* Sweep pipes trying to expire idle flow_queues. */
924 for (i = 0; i < HASHSIZE; i++)
925 SLIST_FOREACH(pipe, &pipehash[i], next)
926 if (pipe->idle_heap.elements > 0 &&
927 DN_KEY_LT(pipe->idle_heap.p[0].key, pipe->V)) {
928 struct dn_flow_queue *q =
929 pipe->idle_heap.p[0].object;
931 heap_extract(&(pipe->idle_heap), NULL);
932 /* Mark timestamp as invalid. */
934 pipe->sum -= q->fs->weight;
942 callout_reset(&dn_timeout, 1, dummynet, NULL);
946 dummynet_send(struct mbuf *m)
948 struct dn_pkt_tag *pkt;
953 for (; m != NULL; m = n) {
956 if (m_tag_first(m) == NULL) {
957 pkt = NULL; /* probably unnecessary */
966 ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
969 ip = mtod(m, struct ip *);
970 #ifndef __linux__ /* restore net format for FreeBSD */
971 ip->ip_len = htons(ip->ip_len);
972 ip->ip_off = htons(ip->ip_off);
974 netisr_dispatch(NETISR_IP, m);
978 netisr_dispatch(NETISR_IPV6, m);
982 ip6_output(m, NULL, NULL, IPV6_FORWARDING, NULL, NULL, NULL);
986 if (bridge_dn_p != NULL)
987 ((*bridge_dn_p)(m, pkt->ifp));
989 printf("dummynet: if_bridge not loaded\n");
992 case DN_TO_ETH_DEMUX:
994 * The Ethernet code assumes the Ethernet header is
995 * contiguous in the first mbuf header.
996 * Insure this is true.
998 if (m->m_len < ETHER_HDR_LEN &&
999 (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) {
1000 printf("dummynet/ether: pullup failed, "
1001 "dropping packet\n");
1004 ether_demux(m->m_pkthdr.rcvif, m);
1007 ether_output_frame(pkt->ifp, m);
1011 /* drop the packet after some time */
1013 netisr_dispatch(-1, m); /* -1 drop the packet */
1020 printf("dummynet: bad switch %d!\n", pkt->dn_dir);
1028 * Unconditionally expire empty queues in case of shortage.
1029 * Returns the number of queues freed.
1032 expire_queues(struct dn_flow_set *fs)
1034 struct dn_flow_queue *q, *prev ;
1035 int i, initial_elements = fs->rq_elements ;
1037 if (fs->last_expired == time_uptime)
1039 fs->last_expired = time_uptime ;
1040 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is overflow */
1041 for (prev=NULL, q = fs->rq[i] ; q != NULL ; )
1042 if (q->head != NULL || q->S != q->F+1) {
1045 } else { /* entry is idle, expire it */
1046 struct dn_flow_queue *old_q = q ;
1049 prev->next = q = q->next ;
1051 fs->rq[i] = q = q->next ;
1053 free(old_q, M_DUMMYNET);
1055 return initial_elements - fs->rq_elements ;
1059 * If room, create a new queue and put at head of slot i;
1060 * otherwise, create or use the default queue.
1062 static struct dn_flow_queue *
1063 create_queue(struct dn_flow_set *fs, int i)
1065 struct dn_flow_queue *q;
1067 if (fs->rq_elements > fs->rq_size * dn_max_ratio &&
1068 expire_queues(fs) == 0) {
1069 /* No way to get room, use or create overflow queue. */
1071 if (fs->rq[i] != NULL)
1074 q = malloc(sizeof(*q), M_DUMMYNET, M_NOWAIT | M_ZERO);
1076 printf("dummynet: sorry, cannot allocate queue for new flow\n");
1081 q->next = fs->rq[i];
1082 q->S = q->F + 1; /* hack - mark timestamp as invalid. */
1083 q->numbytes = io_fast ? fs->pipe->bandwidth : 0;
1090 * Given a flow_set and a pkt in last_pkt, find a matching queue
1091 * after appropriate masking. The queue is moved to front
1092 * so that further searches take less time.
1094 static struct dn_flow_queue *
1095 find_queue(struct dn_flow_set *fs, struct ipfw_flow_id *id)
1097 int i = 0 ; /* we need i and q for new allocations */
1098 struct dn_flow_queue *q, *prev;
1099 int is_v6 = IS_IP6_FLOW_ID(id);
1101 if ( !(fs->flags_fs & DN_HAVE_FLOW_MASK) )
1104 /* first, do the masking, then hash */
1105 id->dst_port &= fs->flow_mask.dst_port ;
1106 id->src_port &= fs->flow_mask.src_port ;
1107 id->proto &= fs->flow_mask.proto ;
1108 id->flags = 0 ; /* we don't care about this one */
1110 APPLY_MASK(&id->dst_ip6, &fs->flow_mask.dst_ip6);
1111 APPLY_MASK(&id->src_ip6, &fs->flow_mask.src_ip6);
1112 id->flow_id6 &= fs->flow_mask.flow_id6;
1114 i = ((id->dst_ip6.__u6_addr.__u6_addr32[0]) & 0xffff)^
1115 ((id->dst_ip6.__u6_addr.__u6_addr32[1]) & 0xffff)^
1116 ((id->dst_ip6.__u6_addr.__u6_addr32[2]) & 0xffff)^
1117 ((id->dst_ip6.__u6_addr.__u6_addr32[3]) & 0xffff)^
1119 ((id->dst_ip6.__u6_addr.__u6_addr32[0] >> 15) & 0xffff)^
1120 ((id->dst_ip6.__u6_addr.__u6_addr32[1] >> 15) & 0xffff)^
1121 ((id->dst_ip6.__u6_addr.__u6_addr32[2] >> 15) & 0xffff)^
1122 ((id->dst_ip6.__u6_addr.__u6_addr32[3] >> 15) & 0xffff)^
1124 ((id->src_ip6.__u6_addr.__u6_addr32[0] << 1) & 0xfffff)^
1125 ((id->src_ip6.__u6_addr.__u6_addr32[1] << 1) & 0xfffff)^
1126 ((id->src_ip6.__u6_addr.__u6_addr32[2] << 1) & 0xfffff)^
1127 ((id->src_ip6.__u6_addr.__u6_addr32[3] << 1) & 0xfffff)^
1129 ((id->src_ip6.__u6_addr.__u6_addr32[0] << 16) & 0xffff)^
1130 ((id->src_ip6.__u6_addr.__u6_addr32[1] << 16) & 0xffff)^
1131 ((id->src_ip6.__u6_addr.__u6_addr32[2] << 16) & 0xffff)^
1132 ((id->src_ip6.__u6_addr.__u6_addr32[3] << 16) & 0xffff)^
1134 (id->dst_port << 1) ^ (id->src_port) ^
1138 id->dst_ip &= fs->flow_mask.dst_ip ;
1139 id->src_ip &= fs->flow_mask.src_ip ;
1141 i = ( (id->dst_ip) & 0xffff ) ^
1142 ( (id->dst_ip >> 15) & 0xffff ) ^
1143 ( (id->src_ip << 1) & 0xffff ) ^
1144 ( (id->src_ip >> 16 ) & 0xffff ) ^
1145 (id->dst_port << 1) ^ (id->src_port) ^
1148 i = i % fs->rq_size ;
1149 /* finally, scan the current list for a match */
1151 for (prev=NULL, q = fs->rq[i] ; q ; ) {
1154 IN6_ARE_ADDR_EQUAL(&id->dst_ip6,&q->id.dst_ip6) &&
1155 IN6_ARE_ADDR_EQUAL(&id->src_ip6,&q->id.src_ip6) &&
1156 id->dst_port == q->id.dst_port &&
1157 id->src_port == q->id.src_port &&
1158 id->proto == q->id.proto &&
1159 id->flags == q->id.flags &&
1160 id->flow_id6 == q->id.flow_id6)
1163 if (!is_v6 && id->dst_ip == q->id.dst_ip &&
1164 id->src_ip == q->id.src_ip &&
1165 id->dst_port == q->id.dst_port &&
1166 id->src_port == q->id.src_port &&
1167 id->proto == q->id.proto &&
1168 id->flags == q->id.flags)
1171 /* No match. Check if we can expire the entry */
1172 if (pipe_expire && q->head == NULL && q->S == q->F+1 ) {
1173 /* entry is idle and not in any heap, expire it */
1174 struct dn_flow_queue *old_q = q ;
1177 prev->next = q = q->next ;
1179 fs->rq[i] = q = q->next ;
1181 free(old_q, M_DUMMYNET);
1187 if (q && prev != NULL) { /* found and not in front */
1188 prev->next = q->next ;
1189 q->next = fs->rq[i] ;
1193 if (q == NULL) { /* no match, need to allocate a new entry */
1194 q = create_queue(fs, i);
1202 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len)
1207 * RED calculates the average queue size (avg) using a low-pass filter
1208 * with an exponential weighted (w_q) moving average:
1209 * avg <- (1-w_q) * avg + w_q * q_size
1210 * where q_size is the queue length (measured in bytes or * packets).
1212 * If q_size == 0, we compute the idle time for the link, and set
1213 * avg = (1 - w_q)^(idle/s)
1214 * where s is the time needed for transmitting a medium-sized packet.
1216 * Now, if avg < min_th the packet is enqueued.
1217 * If avg > max_th the packet is dropped. Otherwise, the packet is
1218 * dropped with probability P function of avg.
1223 /* Queue in bytes or packets? */
1224 u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ?
1225 q->len_bytes : q->len;
1227 DPRINTF(("\ndummynet: %d q: %2u ", (int)curr_time, q_size));
1229 /* Average queue size estimation. */
1231 /* Queue is not empty, avg <- avg + (q_size - avg) * w_q */
1232 int diff = SCALE(q_size) - q->avg;
1233 int64_t v = SCALE_MUL((int64_t)diff, (int64_t)fs->w_q);
1238 * Queue is empty, find for how long the queue has been
1239 * empty and use a lookup table for computing
1240 * (1 - * w_q)^(idle_time/s) where s is the time to send a
1242 * XXX check wraps...
1245 u_int t = div64(curr_time - q->q_time,
1248 q->avg = (t >= 0 && t < fs->lookup_depth) ?
1249 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0;
1252 DPRINTF(("dummynet: avg: %u ", SCALE_VAL(q->avg)));
1254 /* Should i drop? */
1255 if (q->avg < fs->min_th) {
1257 return (0); /* accept packet */
1259 if (q->avg >= fs->max_th) { /* average queue >= max threshold */
1260 if (fs->flags_fs & DN_IS_GENTLE_RED) {
1262 * According to Gentle-RED, if avg is greater than
1263 * max_th the packet is dropped with a probability
1264 * p_b = c_3 * avg - c_4
1265 * where c_3 = (1 - max_p) / max_th
1266 * c_4 = 1 - 2 * max_p
1268 p_b = SCALE_MUL((int64_t)fs->c_3, (int64_t)q->avg) -
1272 DPRINTF(("dummynet: - drop"));
1275 } else if (q->avg > fs->min_th) {
1277 * We compute p_b using the linear dropping function
1278 * p_b = c_1 * avg - c_2
1279 * where c_1 = max_p / (max_th - min_th)
1280 * c_2 = max_p * min_th / (max_th - min_th)
1282 p_b = SCALE_MUL((int64_t)fs->c_1, (int64_t)q->avg) - fs->c_2;
1285 if (fs->flags_fs & DN_QSIZE_IS_BYTES)
1286 p_b = div64(p_b * len, fs->max_pkt_size);
1287 if (++q->count == 0)
1288 q->random = random() & 0xffff;
1291 * q->count counts packets arrived since last drop, so a greater
1292 * value of q->count means a greater packet drop probability.
1294 if (SCALE_MUL(p_b, SCALE((int64_t)q->count)) > q->random) {
1296 DPRINTF(("dummynet: - red drop"));
1297 /* After a drop we calculate a new random value. */
1298 q->random = random() & 0xffff;
1299 return (1); /* drop */
1302 /* End of RED algorithm. */
1304 return (0); /* accept */
1307 static __inline struct dn_flow_set *
1308 locate_flowset(int fs_nr)
1310 struct dn_flow_set *fs;
1312 SLIST_FOREACH(fs, &flowsethash[HASH(fs_nr)], next)
1313 if (fs->fs_nr == fs_nr)
1319 static __inline struct dn_pipe *
1320 locate_pipe(int pipe_nr)
1322 struct dn_pipe *pipe;
1324 SLIST_FOREACH(pipe, &pipehash[HASH(pipe_nr)], next)
1325 if (pipe->pipe_nr == pipe_nr)
1332 * dummynet hook for packets. Below 'pipe' is a pipe or a queue
1333 * depending on whether WF2Q or fixed bw is used.
1335 * pipe_nr pipe or queue the packet is destined for.
1336 * dir where shall we send the packet after dummynet.
1337 * m the mbuf with the packet
1338 * ifp the 'ifp' parameter from the caller.
1339 * NULL in ip_input, destination interface in ip_output,
1340 * rule matching rule, in case of multiple passes
1343 dummynet_io(struct mbuf **m0, int dir, struct ip_fw_args *fwa)
1345 struct mbuf *m = *m0, *head = NULL, *tail = NULL;
1346 struct dn_pkt_tag *pkt;
1348 struct dn_flow_set *fs = NULL;
1349 struct dn_pipe *pipe;
1350 uint64_t len = m->m_pkthdr.len;
1351 struct dn_flow_queue *q = NULL;
1353 ipfw_insn *cmd = ACTION_PTR(fwa->rule);
1355 KASSERT(m->m_nextpkt == NULL,
1356 ("dummynet_io: mbuf queue passed to dummynet"));
1358 if (cmd->opcode == O_LOG)
1360 if (cmd->opcode == O_ALTQ)
1362 if (cmd->opcode == O_TAG)
1364 is_pipe = (cmd->opcode == O_PIPE);
1369 * This is a dummynet rule, so we expect an O_PIPE or O_QUEUE rule.
1371 * XXXGL: probably the pipe->fs and fs->pipe logic here
1372 * below can be simplified.
1375 pipe = locate_pipe(fwa->cookie);
1379 fs = locate_flowset(fwa->cookie);
1382 goto dropit; /* This queue/pipe does not exist! */
1384 if (pipe == NULL) { /* Must be a queue, try find a matching pipe. */
1385 pipe = locate_pipe(fs->parent_nr);
1389 printf("dummynet: no pipe %d for queue %d, drop pkt\n",
1390 fs->parent_nr, fs->fs_nr);
1394 q = find_queue(fs, &(fwa->f_id));
1396 goto dropit; /* Cannot allocate queue. */
1398 /* Update statistics, then check reasons to drop pkt. */
1399 q->tot_bytes += len;
1401 if (fs->plr && random() < fs->plr)
1402 goto dropit; /* Random pkt drop. */
1403 if (fs->flags_fs & DN_QSIZE_IS_BYTES) {
1404 if (q->len_bytes > fs->qsize)
1405 goto dropit; /* Queue size overflow. */
1407 if (q->len >= fs->qsize)
1408 goto dropit; /* Queue count overflow. */
1410 if (fs->flags_fs & DN_IS_RED && red_drops(fs, q, len))
1413 /* XXX expensive to zero, see if we can remove it. */
1414 mtag = m_tag_get(PACKET_TAG_DUMMYNET,
1415 sizeof(struct dn_pkt_tag), M_NOWAIT | M_ZERO);
1417 goto dropit; /* Cannot allocate packet header. */
1418 m_tag_prepend(m, mtag); /* Attach to mbuf chain. */
1420 pkt = (struct dn_pkt_tag *)(mtag + 1);
1422 * Ok, i can handle the pkt now...
1423 * Build and enqueue packet + parameters.
1425 pkt->rule = fwa->rule;
1428 pkt->ifp = fwa->oif;
1430 if (q->head == NULL)
1433 q->tail->m_nextpkt = m;
1436 q->len_bytes += len;
1438 if (q->head != m) /* Flow was not idle, we are done. */
1441 if (q->q_time < (uint32_t)curr_time)
1442 q->numbytes = io_fast ? fs->pipe->bandwidth : 0;
1443 q->q_time = curr_time;
1446 * If we reach this point the flow was previously idle, so we need
1447 * to schedule it. This involves different actions for fixed-rate or
1451 /* Fixed-rate queue: just insert into the ready_heap. */
1454 if (pipe->bandwidth) {
1455 q->extra_bits = compute_extra_bits(m, pipe);
1456 t = set_ticks(m, q, pipe);
1458 q->sched_time = curr_time;
1459 if (t == 0) /* Must process it now. */
1460 ready_event(q, &head, &tail);
1462 heap_insert(&ready_heap, curr_time + t , q);
1465 * WF2Q. First, compute start time S: if the flow was
1466 * idle (S = F + 1) set S to the virtual time V for the
1467 * controlling pipe, and update the sum of weights for the pipe;
1468 * otherwise, remove flow from idle_heap and set S to max(F,V).
1469 * Second, compute finish time F = S + len / weight.
1470 * Third, if pipe was idle, update V = max(S, V).
1471 * Fourth, count one more backlogged flow.
1473 if (DN_KEY_GT(q->S, q->F)) { /* Means timestamps are invalid. */
1475 pipe->sum += fs->weight; /* Add weight of new queue. */
1477 heap_extract(&(pipe->idle_heap), q);
1478 q->S = MAX64(q->F, pipe->V);
1480 q->F = div64(q->S + (len << MY_M), fs->weight);
1482 if (pipe->not_eligible_heap.elements == 0 &&
1483 pipe->scheduler_heap.elements == 0)
1484 pipe->V = MAX64(q->S, pipe->V);
1487 * Look at eligibility. A flow is not eligibile if S>V (when
1488 * this happens, it means that there is some other flow already
1489 * scheduled for the same pipe, so the scheduler_heap cannot be
1490 * empty). If the flow is not eligible we just store it in the
1491 * not_eligible_heap. Otherwise, we store in the scheduler_heap
1492 * and possibly invoke ready_event_wfq() right now if there is
1494 * Note that for all flows in scheduler_heap (SCH), S_i <= V,
1495 * and for all flows in not_eligible_heap (NEH), S_i > V.
1496 * So when we need to compute max(V, min(S_i)) forall i in
1497 * SCH+NEH, we only need to look into NEH.
1499 if (DN_KEY_GT(q->S, pipe->V)) { /* Not eligible. */
1500 if (pipe->scheduler_heap.elements == 0)
1501 printf("dummynet: ++ ouch! not eligible but empty scheduler!\n");
1502 heap_insert(&(pipe->not_eligible_heap), q->S, q);
1504 heap_insert(&(pipe->scheduler_heap), q->F, q);
1505 if (pipe->numbytes >= 0) { /* Pipe is idle. */
1506 if (pipe->scheduler_heap.elements != 1)
1507 printf("dummynet: OUCH! pipe should have been idle!\n");
1508 DPRINTF(("dummynet: waking up pipe %d at %d\n",
1509 pipe->pipe_nr, (int)(q->F >> MY_M)));
1510 pipe->sched_time = curr_time;
1511 ready_event_wfq(pipe, &head, &tail);
1516 if (head == m && dir != DN_TO_IFB_FWD && dir != DN_TO_ETH_DEMUX &&
1517 dir != DN_TO_ETH_OUT) { /* Fast io. */
1519 if (m->m_nextpkt != NULL)
1520 printf("dummynet: fast io: pkt chain detected!\n");
1521 head = m->m_nextpkt = NULL;
1523 *m0 = NULL; /* Normal io. */
1527 dummynet_send(head);
1536 * set the tag, if present. dn_tag_get cannot fail
1537 * so we need to check first
1539 if (m_tag_first(m)) {
1540 pkt = dn_tag_get(m);
1541 pkt->dn_dir = DN_TO_DROP;
1543 dummynet_send(m); /* drop the packet */
1545 return ((fs && (fs->flags_fs & DN_NOERROR)) ? 0 : ENOBUFS);
1549 * Below, the rt_unref is only needed when (pkt->dn_dir == DN_TO_IP_OUT)
1550 * Doing this would probably save us the initial bzero of dn_pkt
1552 #define DN_FREE_PKT(_m) do { \
1557 * Dispose all packets and flow_queues on a flow_set.
1558 * If all=1, also remove red lookup table and other storage,
1559 * including the descriptor itself.
1560 * For the one in dn_pipe MUST also cleanup ready_heap...
1563 purge_flow_set(struct dn_flow_set *fs, int all)
1565 struct dn_flow_queue *q, *qn;
1568 DUMMYNET_LOCK_ASSERT();
1570 for (i = 0; i <= fs->rq_size; i++) {
1571 for (q = fs->rq[i]; q != NULL; q = qn) {
1572 struct mbuf *m, *mnext;
1575 while ((m = mnext) != NULL) {
1576 mnext = m->m_nextpkt;
1580 free(q, M_DUMMYNET);
1585 fs->rq_elements = 0;
1587 /* RED - free lookup table. */
1588 if (fs->w_q_lookup != NULL)
1589 free(fs->w_q_lookup, M_DUMMYNET);
1591 free(fs->rq, M_DUMMYNET);
1592 /* If this fs is not part of a pipe, free it. */
1593 if (fs->pipe == NULL || fs != &(fs->pipe->fs))
1594 free(fs, M_DUMMYNET);
1599 * Dispose all packets queued on a pipe (not a flow_set).
1600 * Also free all resources associated to a pipe, which is about
1604 purge_pipe(struct dn_pipe *pipe)
1606 struct mbuf *m, *mnext;
1608 purge_flow_set( &(pipe->fs), 1 );
1611 while ((m = mnext) != NULL) {
1612 mnext = m->m_nextpkt;
1616 heap_free( &(pipe->scheduler_heap) );
1617 heap_free( &(pipe->not_eligible_heap) );
1618 heap_free( &(pipe->idle_heap) );
1622 * Delete all pipes and heaps returning memory. Must also
1623 * remove references from all ipfw rules to all pipes.
1626 dummynet_flush(void)
1628 struct dn_pipe *pipe, *pipe1;
1629 struct dn_flow_set *fs, *fs1;
1633 /* Free heaps so we don't have unwanted events. */
1634 heap_free(&ready_heap);
1635 heap_free(&wfq_ready_heap);
1636 heap_free(&extract_heap);
1639 * Now purge all queued pkts and delete all pipes.
1641 * XXXGL: can we merge the for(;;) cycles into one or not?
1643 for (i = 0; i < HASHSIZE; i++)
1644 SLIST_FOREACH_SAFE(fs, &flowsethash[i], next, fs1) {
1645 SLIST_REMOVE(&flowsethash[i], fs, dn_flow_set, next);
1646 purge_flow_set(fs, 1);
1648 for (i = 0; i < HASHSIZE; i++)
1649 SLIST_FOREACH_SAFE(pipe, &pipehash[i], next, pipe1) {
1650 SLIST_REMOVE(&pipehash[i], pipe, dn_pipe, next);
1657 extern struct ip_fw *ip_fw_default_rule;
1659 dn_rule_delete_fs(struct dn_flow_set *fs, void *r)
1662 struct dn_flow_queue *q ;
1665 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is ovflow */
1666 for (q = fs->rq[i] ; q ; q = q->next )
1667 for (m = q->head ; m ; m = m->m_nextpkt ) {
1668 struct dn_pkt_tag *pkt = dn_tag_get(m) ;
1670 pkt->rule = ip_fw_default_rule ;
1675 * When a firewall rule is deleted, scan all queues and remove the pointer
1676 * to the rule from matching packets, making them point to the default rule.
1677 * The pointer is used to reinject packets in case one_pass = 0.
1680 dn_rule_delete(void *r)
1682 struct dn_pipe *pipe;
1683 struct dn_flow_set *fs;
1684 struct dn_pkt_tag *pkt;
1690 * If the rule references a queue (dn_flow_set), then scan
1691 * the flow set, otherwise scan pipes. Should do either, but doing
1692 * both does not harm.
1694 for (i = 0; i < HASHSIZE; i++)
1695 SLIST_FOREACH(fs, &flowsethash[i], next)
1696 dn_rule_delete_fs(fs, r);
1698 for (i = 0; i < HASHSIZE; i++)
1699 SLIST_FOREACH(pipe, &pipehash[i], next) {
1701 dn_rule_delete_fs(fs, r);
1702 for (m = pipe->head ; m ; m = m->m_nextpkt ) {
1703 pkt = dn_tag_get(m);
1705 pkt->rule = ip_fw_default_rule;
1712 * setup RED parameters
1715 config_red(struct dn_flow_set *p, struct dn_flow_set *x)
1720 x->min_th = SCALE(p->min_th);
1721 x->max_th = SCALE(p->max_th);
1722 x->max_p = p->max_p;
1724 x->c_1 = p->max_p / (p->max_th - p->min_th);
1725 x->c_2 = SCALE_MUL(x->c_1, SCALE(p->min_th));
1727 if (x->flags_fs & DN_IS_GENTLE_RED) {
1728 x->c_3 = (SCALE(1) - p->max_p) / p->max_th;
1729 x->c_4 = SCALE(1) - 2 * p->max_p;
1732 /* If the lookup table already exist, free and create it again. */
1733 if (x->w_q_lookup) {
1734 free(x->w_q_lookup, M_DUMMYNET);
1735 x->w_q_lookup = NULL;
1737 if (red_lookup_depth == 0) {
1738 printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1740 free(x, M_DUMMYNET);
1743 x->lookup_depth = red_lookup_depth;
1744 x->w_q_lookup = (u_int *)malloc(x->lookup_depth * sizeof(int),
1745 M_DUMMYNET, M_NOWAIT);
1746 if (x->w_q_lookup == NULL) {
1747 printf("dummynet: sorry, cannot allocate red lookup table\n");
1748 free(x, M_DUMMYNET);
1752 /* Fill the lookup table with (1 - w_q)^x */
1753 x->lookup_step = p->lookup_step;
1754 x->lookup_weight = p->lookup_weight;
1755 x->w_q_lookup[0] = SCALE(1) - x->w_q;
1757 for (i = 1; i < x->lookup_depth; i++)
1759 SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight);
1761 if (red_avg_pkt_size < 1)
1762 red_avg_pkt_size = 512;
1763 x->avg_pkt_size = red_avg_pkt_size;
1764 if (red_max_pkt_size < 1)
1765 red_max_pkt_size = 1500;
1766 x->max_pkt_size = red_max_pkt_size;
1771 alloc_hash(struct dn_flow_set *x, struct dn_flow_set *pfs)
1773 if (x->flags_fs & DN_HAVE_FLOW_MASK) { /* allocate some slots */
1774 int l = pfs->rq_size;
1780 else if (l > DN_MAX_HASH_SIZE)
1781 l = DN_MAX_HASH_SIZE;
1783 } else /* one is enough for null mask */
1785 x->rq = malloc((1 + x->rq_size) * sizeof(struct dn_flow_queue *),
1786 M_DUMMYNET, M_NOWAIT | M_ZERO);
1787 if (x->rq == NULL) {
1788 printf("dummynet: sorry, cannot allocate queue\n");
1796 set_fs_parms(struct dn_flow_set *x, struct dn_flow_set *src)
1798 x->flags_fs = src->flags_fs;
1799 x->qsize = src->qsize;
1801 x->flow_mask = src->flow_mask;
1802 if (x->flags_fs & DN_QSIZE_IS_BYTES) {
1803 if (x->qsize > pipe_byte_limit)
1804 x->qsize = 1024 * 1024;
1808 if (x->qsize > pipe_slot_limit)
1811 /* Configuring RED. */
1812 if (x->flags_fs & DN_IS_RED)
1813 config_red(src, x); /* XXX should check errors */
1817 * Setup pipe or queue parameters.
1820 config_pipe(struct dn_pipe *p)
1822 struct dn_flow_set *pfs = &(p->fs);
1823 struct dn_flow_queue *q;
1827 * The config program passes parameters as follows:
1828 * bw = bits/second (0 means no limits),
1829 * delay = ms, must be translated into ticks.
1830 * qsize = slots/bytes
1832 p->delay = (p->delay * hz) / 1000;
1833 /* We need either a pipe number or a flow_set number. */
1834 if (p->pipe_nr == 0 && pfs->fs_nr == 0)
1836 if (p->pipe_nr != 0 && pfs->fs_nr != 0)
1838 if (p->pipe_nr != 0) { /* this is a pipe */
1839 struct dn_pipe *pipe;
1842 pipe = locate_pipe(p->pipe_nr); /* locate pipe */
1844 if (pipe == NULL) { /* new pipe */
1845 pipe = malloc(sizeof(struct dn_pipe), M_DUMMYNET,
1849 printf("dummynet: no memory for new pipe\n");
1852 pipe->pipe_nr = p->pipe_nr;
1853 pipe->fs.pipe = pipe;
1855 * idle_heap is the only one from which
1856 * we extract from the middle.
1858 pipe->idle_heap.size = pipe->idle_heap.elements = 0;
1859 pipe->idle_heap.offset =
1860 offsetof(struct dn_flow_queue, heap_pos);
1862 /* Flush accumulated credit for all queues. */
1863 for (i = 0; i <= pipe->fs.rq_size; i++)
1864 for (q = pipe->fs.rq[i]; q; q = q->next)
1865 q->numbytes = io_fast ? p->bandwidth : 0;
1867 pipe->bandwidth = p->bandwidth;
1868 pipe->numbytes = 0; /* just in case... */
1869 bcopy(p->if_name, pipe->if_name, sizeof(p->if_name));
1870 pipe->ifp = NULL; /* reset interface ptr */
1871 pipe->delay = p->delay;
1872 set_fs_parms(&(pipe->fs), pfs);
1874 /* Handle changes in the delay profile. */
1875 if (p->samples_no > 0) {
1876 if (pipe->samples_no != p->samples_no) {
1877 if (pipe->samples != NULL)
1878 free(pipe->samples, M_DUMMYNET);
1880 malloc(p->samples_no*sizeof(dn_key),
1881 M_DUMMYNET, M_NOWAIT | M_ZERO);
1882 if (pipe->samples == NULL) {
1884 printf("dummynet: no memory "
1885 "for new samples\n");
1888 pipe->samples_no = p->samples_no;
1891 strncpy(pipe->name,p->name,sizeof(pipe->name));
1892 pipe->loss_level = p->loss_level;
1893 for (i = 0; i<pipe->samples_no; ++i)
1894 pipe->samples[i] = p->samples[i];
1895 } else if (pipe->samples != NULL) {
1896 free(pipe->samples, M_DUMMYNET);
1897 pipe->samples = NULL;
1898 pipe->samples_no = 0;
1901 if (pipe->fs.rq == NULL) { /* a new pipe */
1902 error = alloc_hash(&(pipe->fs), pfs);
1908 SLIST_INSERT_HEAD(&pipehash[HASH(pipe->pipe_nr)],
1912 } else { /* config queue */
1913 struct dn_flow_set *fs;
1916 fs = locate_flowset(pfs->fs_nr); /* locate flow_set */
1918 if (fs == NULL) { /* new */
1919 if (pfs->parent_nr == 0) { /* need link to a pipe */
1923 fs = malloc(sizeof(struct dn_flow_set), M_DUMMYNET,
1928 "dummynet: no memory for new flow_set\n");
1931 fs->fs_nr = pfs->fs_nr;
1932 fs->parent_nr = pfs->parent_nr;
1933 fs->weight = pfs->weight;
1934 if (fs->weight == 0)
1936 else if (fs->weight > 100)
1940 * Change parent pipe not allowed;
1941 * must delete and recreate.
1943 if (pfs->parent_nr != 0 &&
1944 fs->parent_nr != pfs->parent_nr) {
1950 set_fs_parms(fs, pfs);
1952 if (fs->rq == NULL) { /* a new flow_set */
1953 error = alloc_hash(fs, pfs);
1956 free(fs, M_DUMMYNET);
1959 SLIST_INSERT_HEAD(&flowsethash[HASH(fs->fs_nr)],
1968 * Helper function to remove from a heap queues which are linked to
1969 * a flow_set about to be deleted.
1972 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs)
1974 int i = 0, found = 0 ;
1975 for (; i < h->elements ;)
1976 if ( ((struct dn_flow_queue *)h->p[i].object)->fs == fs) {
1978 h->p[i] = h->p[h->elements] ;
1987 * helper function to remove a pipe from a heap (can be there at most once)
1990 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p)
1992 if (h->elements > 0) {
1994 for (i=0; i < h->elements ; i++ ) {
1995 if (h->p[i].object == p) { /* found it */
1997 h->p[i] = h->p[h->elements] ;
2006 * drain all queues. Called in case of severe mbuf shortage.
2009 dummynet_drain(void)
2011 struct dn_flow_set *fs;
2012 struct dn_pipe *pipe;
2013 struct mbuf *m, *mnext;
2016 DUMMYNET_LOCK_ASSERT();
2018 heap_free(&ready_heap);
2019 heap_free(&wfq_ready_heap);
2020 heap_free(&extract_heap);
2021 /* remove all references to this pipe from flow_sets */
2022 for (i = 0; i < HASHSIZE; i++)
2023 SLIST_FOREACH(fs, &flowsethash[i], next)
2024 purge_flow_set(fs, 0);
2026 for (i = 0; i < HASHSIZE; i++) {
2027 SLIST_FOREACH(pipe, &pipehash[i], next) {
2028 purge_flow_set(&(pipe->fs), 0);
2031 while ((m = mnext) != NULL) {
2032 mnext = m->m_nextpkt;
2035 pipe->head = pipe->tail = NULL;
2041 * Fully delete a pipe or a queue, cleaning up associated info.
2044 delete_pipe(struct dn_pipe *p)
2047 if (p->pipe_nr == 0 && p->fs.fs_nr == 0)
2049 if (p->pipe_nr != 0 && p->fs.fs_nr != 0)
2051 if (p->pipe_nr != 0) { /* this is an old-style pipe */
2052 struct dn_pipe *pipe;
2053 struct dn_flow_set *fs;
2057 pipe = locate_pipe(p->pipe_nr); /* locate pipe */
2061 return (ENOENT); /* not found */
2064 /* Unlink from list of pipes. */
2065 SLIST_REMOVE(&pipehash[HASH(pipe->pipe_nr)], pipe, dn_pipe, next);
2067 /* Remove all references to this pipe from flow_sets. */
2068 for (i = 0; i < HASHSIZE; i++)
2069 SLIST_FOREACH(fs, &flowsethash[i], next)
2070 if (fs->pipe == pipe) {
2071 printf("dummynet: ++ ref to pipe %d from fs %d\n",
2072 p->pipe_nr, fs->fs_nr);
2074 purge_flow_set(fs, 0);
2076 fs_remove_from_heap(&ready_heap, &(pipe->fs));
2077 purge_pipe(pipe); /* remove all data associated to this pipe */
2078 /* remove reference to here from extract_heap and wfq_ready_heap */
2079 pipe_remove_from_heap(&extract_heap, pipe);
2080 pipe_remove_from_heap(&wfq_ready_heap, pipe);
2084 } else { /* this is a WF2Q queue (dn_flow_set) */
2085 struct dn_flow_set *fs;
2088 fs = locate_flowset(p->fs.fs_nr); /* locate set */
2092 return (ENOENT); /* not found */
2095 /* Unlink from list of flowsets. */
2096 SLIST_REMOVE( &flowsethash[HASH(fs->fs_nr)], fs, dn_flow_set, next);
2098 if (fs->pipe != NULL) {
2099 /* Update total weight on parent pipe and cleanup parent heaps. */
2100 fs->pipe->sum -= fs->weight * fs->backlogged ;
2101 fs_remove_from_heap(&(fs->pipe->not_eligible_heap), fs);
2102 fs_remove_from_heap(&(fs->pipe->scheduler_heap), fs);
2103 #if 1 /* XXX should i remove from idle_heap as well ? */
2104 fs_remove_from_heap(&(fs->pipe->idle_heap), fs);
2107 purge_flow_set(fs, 1);
2114 * helper function used to copy data from kernel in DUMMYNET_GET
2117 dn_copy_set(struct dn_flow_set *set, char *bp)
2120 struct dn_flow_queue *q, *qp = (struct dn_flow_queue *)bp;
2122 DUMMYNET_LOCK_ASSERT();
2124 for (i = 0 ; i <= set->rq_size ; i++)
2125 for (q = set->rq[i] ; q ; q = q->next, qp++ ) {
2126 if (q->hash_slot != i)
2127 printf("dummynet: ++ at %d: wrong slot (have %d, "
2128 "should be %d)\n", copied, q->hash_slot, i);
2130 printf("dummynet: ++ at %d: wrong fs ptr (have %p, should be %p)\n",
2133 bcopy(q, qp, sizeof( *q ) );
2134 /* cleanup pointers */
2136 qp->head = qp->tail = NULL ;
2139 if (copied != set->rq_elements)
2140 printf("dummynet: ++ wrong count, have %d should be %d\n",
2141 copied, set->rq_elements);
2148 struct dn_flow_set *fs;
2149 struct dn_pipe *pipe;
2153 DUMMYNET_LOCK_ASSERT();
2155 * Compute size of data structures: list of pipes and flow_sets.
2157 for (i = 0; i < HASHSIZE; i++) {
2158 SLIST_FOREACH(pipe, &pipehash[i], next)
2159 size += sizeof(*pipe) +
2160 pipe->fs.rq_elements * sizeof(struct dn_flow_queue);
2161 SLIST_FOREACH(fs, &flowsethash[i], next)
2162 size += sizeof (*fs) +
2163 fs->rq_elements * sizeof(struct dn_flow_queue);
2169 dummynet_get(struct sockopt *sopt)
2171 char *buf, *bp ; /* bp is the "copy-pointer" */
2173 struct dn_flow_set *fs;
2174 struct dn_pipe *pipe;
2177 /* XXX lock held too long */
2180 * XXX: Ugly, but we need to allocate memory with M_WAITOK flag and we
2181 * cannot use this flag while holding a mutex.
2183 for (i = 0; i < 10; i++) {
2184 size = dn_calc_size();
2186 buf = malloc(size, M_TEMP, M_WAITOK);
2188 if (size == dn_calc_size())
2198 for (i = 0; i < HASHSIZE; i++)
2199 SLIST_FOREACH(pipe, &pipehash[i], next) {
2200 struct dn_pipe *pipe_bp = (struct dn_pipe *)bp;
2203 * Copy pipe descriptor into *bp, convert delay back to ms,
2204 * then copy the flow_set descriptor(s) one at a time.
2205 * After each flow_set, copy the queue descriptor it owns.
2207 bcopy(pipe, bp, sizeof(*pipe));
2208 pipe_bp->delay = (pipe_bp->delay * 1000) / hz;
2210 * XXX the following is a hack based on ->next being the
2211 * first field in dn_pipe and dn_flow_set. The correct
2212 * solution would be to move the dn_flow_set to the beginning
2213 * of struct dn_pipe.
2215 pipe_bp->next.sle_next = (struct dn_pipe *)DN_IS_PIPE;
2216 /* Clean pointers. */
2217 pipe_bp->head = pipe_bp->tail = NULL;
2218 pipe_bp->fs.next.sle_next = NULL;
2219 pipe_bp->fs.pipe = NULL;
2220 pipe_bp->fs.rq = NULL;
2221 pipe_bp->samples = NULL;
2223 bp += sizeof(*pipe) ;
2224 bp = dn_copy_set(&(pipe->fs), bp);
2227 for (i = 0; i < HASHSIZE; i++)
2228 SLIST_FOREACH(fs, &flowsethash[i], next) {
2229 struct dn_flow_set *fs_bp = (struct dn_flow_set *)bp;
2231 bcopy(fs, bp, sizeof(*fs));
2232 /* XXX same hack as above */
2233 fs_bp->next.sle_next = (struct dn_flow_set *)DN_IS_QUEUE;
2237 bp = dn_copy_set(fs, bp);
2242 error = sooptcopyout(sopt, buf, size);
2248 * Handler for the various dummynet socket options (get, flush, config, del)
2251 ip_dn_ctl(struct sockopt *sopt)
2254 struct dn_pipe *p = NULL;
2256 error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2260 /* Disallow sets in really-really secure mode. */
2261 if (sopt->sopt_dir == SOPT_SET) {
2262 #if __FreeBSD_version >= 500034
2263 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2267 if (securelevel >= 3)
2272 switch (sopt->sopt_name) {
2274 printf("dummynet: -- unknown option %d", sopt->sopt_name);
2278 case IP_DUMMYNET_GET :
2279 error = dummynet_get(sopt);
2282 case IP_DUMMYNET_FLUSH :
2286 case IP_DUMMYNET_CONFIGURE :
2287 p = malloc(sizeof(struct dn_pipe_max), M_TEMP, M_WAITOK);
2288 error = sooptcopyin(sopt, p, sizeof(struct dn_pipe_max), sizeof *p);
2291 if (p->samples_no > 0)
2292 p->samples = &( ((struct dn_pipe_max*) p)->samples[0] );
2294 error = config_pipe(p);
2297 case IP_DUMMYNET_DEL : /* remove a pipe or queue */
2298 p = malloc(sizeof(struct dn_pipe_max), M_TEMP, M_WAITOK);
2299 error = sooptcopyin(sopt, p, sizeof *p, sizeof *p);
2303 error = delete_pipe(p);
2319 printf("DUMMYNET with IPv6 initialized (040826)\n");
2321 DUMMYNET_LOCK_INIT();
2323 for (i = 0; i < HASHSIZE; i++) {
2324 SLIST_INIT(&pipehash[i]);
2325 SLIST_INIT(&flowsethash[i]);
2327 ready_heap.size = ready_heap.elements = 0;
2328 ready_heap.offset = 0;
2330 wfq_ready_heap.size = wfq_ready_heap.elements = 0;
2331 wfq_ready_heap.offset = 0;
2333 extract_heap.size = extract_heap.elements = 0;
2334 extract_heap.offset = 0;
2336 ip_dn_ctl_ptr = ip_dn_ctl;
2337 ip_dn_io_ptr = dummynet_io;
2338 ip_dn_ruledel_ptr = dn_rule_delete;
2340 TASK_INIT(&dn_task, 0, dummynet_task, NULL);
2341 dn_tq = taskqueue_create_fast("dummynet", M_NOWAIT,
2342 taskqueue_thread_enqueue, &dn_tq);
2343 taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2345 callout_init(&dn_timeout, CALLOUT_MPSAFE);
2346 callout_reset(&dn_timeout, 1, dummynet, NULL);
2348 /* Initialize curr_time adjustment mechanics. */
2349 getmicrouptime(&prev_t);
2356 ip_dn_ctl_ptr = NULL;
2357 ip_dn_io_ptr = NULL;
2358 ip_dn_ruledel_ptr = NULL;
2361 callout_stop(&dn_timeout);
2363 taskqueue_drain(dn_tq, &dn_task);
2364 taskqueue_free(dn_tq);
2368 DUMMYNET_LOCK_DESTROY();
2370 #endif /* KLD_MODULE */
2373 dummynet_modevent(module_t mod, int type, void *data)
2379 printf("DUMMYNET already loaded\n");
2386 #if !defined(KLD_MODULE)
2387 printf("dummynet statically compiled, cannot unload\n");
2400 static moduledata_t dummynet_mod = {
2405 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
2406 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
2407 MODULE_VERSION(dummynet, 1);