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>
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 ip_dn_io_t dummynet_io;
251 static void dn_rule_delete(void *);
254 * Heap management functions.
256 * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2.
257 * Some macros help finding parent/children so we can optimize them.
259 * heap_init() is called to expand the heap when needed.
260 * Increment size in blocks of 16 entries.
261 * XXX failure to allocate a new element is a pretty bad failure
262 * as we basically stall a whole queue forever!!
263 * Returns 1 on error, 0 on success
265 #define HEAP_FATHER(x) ( ( (x) - 1 ) / 2 )
266 #define HEAP_LEFT(x) ( 2*(x) + 1 )
267 #define HEAP_IS_LEFT(x) ( (x) & 1 )
268 #define HEAP_RIGHT(x) ( 2*(x) + 2 )
269 #define HEAP_SWAP(a, b, buffer) { buffer = a ; a = b ; b = buffer ; }
270 #define HEAP_INCREMENT 15
273 heap_init(struct dn_heap *h, int new_size)
275 struct dn_heap_entry *p;
277 if (h->size >= new_size ) {
278 printf("dummynet: %s, Bogus call, have %d want %d\n", __func__,
282 new_size = (new_size + HEAP_INCREMENT ) & ~HEAP_INCREMENT ;
283 p = malloc(new_size * sizeof(*p), M_DUMMYNET, M_NOWAIT);
285 printf("dummynet: %s, resize %d failed\n", __func__, new_size );
286 return 1 ; /* error */
289 bcopy(h->p, p, h->size * sizeof(*p) );
290 free(h->p, M_DUMMYNET);
298 * Insert element in heap. Normally, p != NULL, we insert p in
299 * a new position and bubble up. If p == NULL, then the element is
300 * already in place, and key is the position where to start the
302 * Returns 1 on failure (cannot allocate new heap entry)
304 * If offset > 0 the position (index, int) of the element in the heap is
305 * also stored in the element itself at the given offset in bytes.
307 #define SET_OFFSET(heap, node) \
308 if (heap->offset > 0) \
309 *((int *)((char *)(heap->p[node].object) + heap->offset)) = node ;
311 * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value.
313 #define RESET_OFFSET(heap, node) \
314 if (heap->offset > 0) \
315 *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1 ;
317 heap_insert(struct dn_heap *h, dn_key key1, void *p)
319 int son = h->elements ;
321 if (p == NULL) /* data already there, set starting point */
323 else { /* insert new element at the end, possibly resize */
325 if (son == h->size) /* need resize... */
326 if (heap_init(h, h->elements+1) )
327 return 1 ; /* failure... */
328 h->p[son].object = p ;
329 h->p[son].key = key1 ;
332 while (son > 0) { /* bubble up */
333 int father = HEAP_FATHER(son) ;
334 struct dn_heap_entry tmp ;
336 if (DN_KEY_LT( h->p[father].key, h->p[son].key ) )
337 break ; /* found right position */
338 /* son smaller than father, swap and repeat */
339 HEAP_SWAP(h->p[son], h->p[father], tmp) ;
348 * remove top element from heap, or obj if obj != NULL
351 heap_extract(struct dn_heap *h, void *obj)
353 int child, father, max = h->elements - 1 ;
356 printf("dummynet: warning, extract from empty heap 0x%p\n", h);
359 father = 0 ; /* default: move up smallest child */
360 if (obj != NULL) { /* extract specific element, index is at offset */
362 panic("dummynet: heap_extract from middle not supported on this heap!!!\n");
363 father = *((int *)((char *)obj + h->offset)) ;
364 if (father < 0 || father >= h->elements) {
365 printf("dummynet: heap_extract, father %d out of bound 0..%d\n",
366 father, h->elements);
367 panic("dummynet: heap_extract");
370 RESET_OFFSET(h, father);
371 child = HEAP_LEFT(father) ; /* left child */
372 while (child <= max) { /* valid entry */
373 if (child != max && DN_KEY_LT(h->p[child+1].key, h->p[child].key) )
374 child = child+1 ; /* take right child, otherwise left */
375 h->p[father] = h->p[child] ;
376 SET_OFFSET(h, father);
378 child = HEAP_LEFT(child) ; /* left child for next loop */
383 * Fill hole with last entry and bubble up, reusing the insert code
385 h->p[father] = h->p[max] ;
386 heap_insert(h, father, NULL); /* this one cannot fail */
392 * change object position and update references
393 * XXX this one is never used!
396 heap_move(struct dn_heap *h, dn_key new_key, void *object)
400 int max = h->elements-1 ;
401 struct dn_heap_entry buf ;
404 panic("cannot move items on this heap");
406 i = *((int *)((char *)object + h->offset));
407 if (DN_KEY_LT(new_key, h->p[i].key) ) { /* must move up */
408 h->p[i].key = new_key ;
409 for (; i>0 && DN_KEY_LT(new_key, h->p[(temp = HEAP_FATHER(i))].key) ;
410 i = temp ) { /* bubble up */
411 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
414 } else { /* must move down */
415 h->p[i].key = new_key ;
416 while ( (temp = HEAP_LEFT(i)) <= max ) { /* found left child */
417 if ((temp != max) && DN_KEY_GT(h->p[temp].key, h->p[temp+1].key))
418 temp++ ; /* select child with min key */
419 if (DN_KEY_GT(new_key, h->p[temp].key)) { /* go down */
420 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
429 #endif /* heap_move, unused */
432 * heapify() will reorganize data inside an array to maintain the
433 * heap property. It is needed when we delete a bunch of entries.
436 heapify(struct dn_heap *h)
440 for (i = 0 ; i < h->elements ; i++ )
441 heap_insert(h, i , NULL) ;
445 * cleanup the heap and free data structure
448 heap_free(struct dn_heap *h)
451 free(h->p, M_DUMMYNET);
452 bzero(h, sizeof(*h) );
456 * --- end of heap management functions ---
460 * Return the mbuf tag holding the dummynet state. As an optimization
461 * this is assumed to be the first tag on the list. If this turns out
462 * wrong we'll need to search the list.
464 static struct dn_pkt_tag *
465 dn_tag_get(struct mbuf *m)
467 struct m_tag *mtag = m_tag_first(m);
468 KASSERT(mtag != NULL &&
469 mtag->m_tag_cookie == MTAG_ABI_COMPAT &&
470 mtag->m_tag_id == PACKET_TAG_DUMMYNET,
471 ("packet on dummynet queue w/o dummynet tag!"));
472 return (struct dn_pkt_tag *)(mtag+1);
476 * Scheduler functions:
478 * transmit_event() is called when the delay-line needs to enter
479 * the scheduler, either because of existing pkts getting ready,
480 * or new packets entering the queue. The event handled is the delivery
481 * time of the packet.
483 * ready_event() does something similar with fixed-rate queues, and the
484 * event handled is the finish time of the head pkt.
486 * wfq_ready_event() does something similar with WF2Q queues, and the
487 * event handled is the start time of the head pkt.
489 * In all cases, we make sure that the data structures are consistent
490 * before passing pkts out, because this might trigger recursive
491 * invocations of the procedures.
494 transmit_event(struct dn_pipe *pipe, struct mbuf **head, struct mbuf **tail)
497 struct dn_pkt_tag *pkt;
499 DUMMYNET_LOCK_ASSERT();
501 while ((m = pipe->head) != NULL) {
503 if (!DN_KEY_LEQ(pkt->output_time, curr_time))
506 pipe->head = m->m_nextpkt;
508 (*tail)->m_nextpkt = m;
514 (*tail)->m_nextpkt = NULL;
516 /* If there are leftover packets, put into the heap for next event. */
517 if ((m = pipe->head) != NULL) {
520 * XXX Should check errors on heap_insert, by draining the
521 * whole pipe p and hoping in the future we are more successful.
523 heap_insert(&extract_heap, pkt->output_time, pipe);
528 #define div64(a, b) ((int64_t)(a) / (int64_t)(b))
530 #define DN_TO_DROP 0xffff
532 * Compute how many ticks we have to wait before being able to send
533 * a packet. This is computed as the "wire time" for the packet
534 * (length + extra bits), minus the credit available, scaled to ticks.
535 * Check that the result is not be negative (it could be if we have
536 * too much leftover credit in q->numbytes).
539 set_ticks(struct mbuf *m, struct dn_flow_queue *q, struct dn_pipe *p)
543 ret = div64( (m->m_pkthdr.len * 8 + q->extra_bits) * hz
544 - q->numbytes + p->bandwidth - 1 , p->bandwidth);
546 printf("%s %d extra_bits %d numb %d ret %d\n",
547 __FUNCTION__, __LINE__,
548 (int)(q->extra_bits & 0xffffffff),
549 (int)(q->numbytes & 0xffffffff),
550 (int)(ret & 0xffffffff));
558 * Convert the additional MAC overheads/delays into an equivalent
559 * number of bits for the given data rate. The samples are in milliseconds
560 * so we need to divide by 1000.
563 compute_extra_bits(struct mbuf *pkt, struct dn_pipe *p)
568 if (!p->samples || p->samples_no == 0)
570 index = random() % p->samples_no;
571 extra_bits = div64((dn_key)p->samples[index] * p->bandwidth, 1000);
572 if (index >= p->loss_level) {
573 struct dn_pkt_tag *dt = dn_tag_get(pkt);
575 dt->dn_dir = DN_TO_DROP;
581 free_pipe(struct dn_pipe *p)
584 free(p->samples, M_DUMMYNET);
589 * extract pkt from queue, compute output time (could be now)
590 * and put into delay line (p_queue)
593 move_pkt(struct mbuf *pkt, struct dn_flow_queue *q, struct dn_pipe *p,
596 struct dn_pkt_tag *dt = dn_tag_get(pkt);
598 q->head = pkt->m_nextpkt ;
600 q->len_bytes -= len ;
602 dt->output_time = curr_time + p->delay ;
607 p->tail->m_nextpkt = pkt;
609 p->tail->m_nextpkt = NULL;
613 * ready_event() is invoked every time the queue must enter the
614 * scheduler, either because the first packet arrives, or because
615 * a previously scheduled event fired.
616 * On invokation, drain as many pkts as possible (could be 0) and then
617 * if there are leftover packets reinsert the pkt in the scheduler.
620 ready_event(struct dn_flow_queue *q, struct mbuf **head, struct mbuf **tail)
623 struct dn_pipe *p = q->fs->pipe;
626 DUMMYNET_LOCK_ASSERT();
629 printf("dummynet: ready_event- pipe is gone\n");
632 p_was_empty = (p->head == NULL);
635 * Schedule fixed-rate queues linked to this pipe:
636 * account for the bw accumulated since last scheduling, then
637 * drain as many pkts as allowed by q->numbytes and move to
638 * the delay line (in p) computing output time.
639 * bandwidth==0 (no limit) means we can drain the whole queue,
640 * setting len_scaled = 0 does the job.
642 q->numbytes += (curr_time - q->sched_time) * p->bandwidth;
643 while ((pkt = q->head) != NULL) {
644 int len = pkt->m_pkthdr.len;
645 dn_key len_scaled = p->bandwidth ? len*8*hz
649 if (DN_KEY_GT(len_scaled, q->numbytes))
651 q->numbytes -= len_scaled;
652 move_pkt(pkt, q, p, len);
654 q->extra_bits = compute_extra_bits(q->head, p);
657 * If we have more packets queued, schedule next ready event
658 * (can only occur when bandwidth != 0, otherwise we would have
659 * flushed the whole queue in the previous loop).
660 * To this purpose we record the current time and compute how many
661 * ticks to go for the finish time of the packet.
663 if ((pkt = q->head) != NULL) { /* this implies bandwidth != 0 */
664 dn_key t = set_ticks(pkt, q, p); /* ticks i have to wait */
666 q->sched_time = curr_time;
667 heap_insert(&ready_heap, curr_time + t, (void *)q);
669 * XXX Should check errors on heap_insert, and drain the whole
670 * queue on error hoping next time we are luckier.
672 } else /* RED needs to know when the queue becomes empty. */
673 q->q_time = curr_time;
676 * If the delay line was empty call transmit_event() now.
677 * Otherwise, the scheduler will take care of it.
680 transmit_event(p, head, tail);
684 * Called when we can transmit packets on WF2Q queues. Take pkts out of
685 * the queues at their start time, and enqueue into the delay line.
686 * Packets are drained until p->numbytes < 0. As long as
687 * len_scaled >= p->numbytes, the packet goes into the delay line
688 * with a deadline p->delay. For the last packet, if p->numbytes < 0,
689 * there is an additional delay.
692 ready_event_wfq(struct dn_pipe *p, struct mbuf **head, struct mbuf **tail)
694 int p_was_empty = (p->head == NULL);
695 struct dn_heap *sch = &(p->scheduler_heap);
696 struct dn_heap *neh = &(p->not_eligible_heap);
697 int64_t p_numbytes = p->numbytes;
699 DUMMYNET_LOCK_ASSERT();
701 if (p->if_name[0] == 0) /* tx clock is simulated */
703 * Since result may not fit into p->numbytes (32bit) we
704 * are using 64bit var here.
706 p_numbytes += (curr_time - p->sched_time) * p->bandwidth;
708 * tx clock is for real,
709 * the ifq must be empty or this is a NOP.
710 * XXX not supported in Linux
712 if (1) // p->ifp && p->ifp->if_snd.ifq_head != NULL)
715 DPRINTF(("dummynet: pipe %d ready from %s --\n",
716 p->pipe_nr, p->if_name));
721 * While we have backlogged traffic AND credit, we need to do
722 * something on the queue.
724 while (p_numbytes >= 0 && (sch->elements > 0 || neh->elements > 0)) {
725 if (sch->elements > 0) {
726 /* Have some eligible pkts to send out. */
727 struct dn_flow_queue *q = sch->p[0].object;
728 struct mbuf *pkt = q->head;
729 struct dn_flow_set *fs = q->fs;
730 uint64_t len = pkt->m_pkthdr.len;
731 int len_scaled = p->bandwidth ? len * 8 * hz : 0;
733 heap_extract(sch, NULL); /* Remove queue from heap. */
734 p_numbytes -= len_scaled;
735 move_pkt(pkt, q, p, len);
737 p->V += div64((len << MY_M), p->sum); /* Update V. */
738 q->S = q->F; /* Update start time. */
740 /* Flow not backlogged any more. */
742 heap_insert(&(p->idle_heap), q->F, q);
744 /* Still backlogged. */
747 * Update F and position in backlogged queue,
748 * then put flow in not_eligible_heap
749 * (we will fix this later).
751 len = (q->head)->m_pkthdr.len;
752 q->F += div64((len << MY_M), fs->weight);
753 if (DN_KEY_LEQ(q->S, p->V))
754 heap_insert(neh, q->S, q);
756 heap_insert(sch, q->F, q);
760 * Now compute V = max(V, min(S_i)). Remember that all elements
761 * in sch have by definition S_i <= V so if sch is not empty,
762 * V is surely the max and we must not update it. Conversely,
763 * if sch is empty we only need to look at neh.
765 if (sch->elements == 0 && neh->elements > 0)
766 p->V = MAX64(p->V, neh->p[0].key);
767 /* Move from neh to sch any packets that have become eligible */
768 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V)) {
769 struct dn_flow_queue *q = neh->p[0].object;
770 heap_extract(neh, NULL);
771 heap_insert(sch, q->F, q);
774 if (p->if_name[0] != '\0') { /* Tx clock is from a real thing */
775 p_numbytes = -1; /* Mark not ready for I/O. */
779 if (sch->elements == 0 && neh->elements == 0 && p_numbytes >= 0 &&
780 p->idle_heap.elements > 0) {
782 * No traffic and no events scheduled.
783 * We can get rid of idle-heap.
787 for (i = 0; i < p->idle_heap.elements; i++) {
788 struct dn_flow_queue *q = p->idle_heap.p[i].object;
795 p->idle_heap.elements = 0;
798 * If we are getting clocks from dummynet (not a real interface) and
799 * If we are under credit, schedule the next ready event.
800 * Also fix the delivery time of the last packet.
802 if (p->if_name[0]==0 && p_numbytes < 0) { /* This implies bw > 0. */
803 dn_key t = 0; /* Number of ticks i have to wait. */
805 if (p->bandwidth > 0)
806 t = div64(p->bandwidth - 1 - p_numbytes, p->bandwidth);
807 dn_tag_get(p->tail)->output_time += t;
808 p->sched_time = curr_time;
809 heap_insert(&wfq_ready_heap, curr_time + t, (void *)p);
811 * XXX Should check errors on heap_insert, and drain the whole
812 * queue on error hoping next time we are luckier.
816 /* Fit (adjust if necessary) 64bit result into 32bit variable. */
817 if (p_numbytes > INT_MAX)
818 p->numbytes = INT_MAX;
819 else if (p_numbytes < INT_MIN)
820 p->numbytes = INT_MIN;
822 p->numbytes = p_numbytes;
825 * If the delay line was empty call transmit_event() now.
826 * Otherwise, the scheduler will take care of it.
829 transmit_event(p, head, tail);
833 * This is called one tick, after previous run. It is used to
837 dummynet(void * __unused unused)
840 taskqueue_enqueue(dn_tq, &dn_task);
844 * The main dummynet processing function.
847 dummynet_task(void *context, int pending)
849 struct mbuf *head = NULL, *tail = NULL;
850 struct dn_pipe *pipe;
851 struct dn_heap *heaps[3];
853 void *p; /* generic parameter to handler */
858 heaps[0] = &ready_heap; /* fixed-rate queues */
859 heaps[1] = &wfq_ready_heap; /* wfq queues */
860 heaps[2] = &extract_heap; /* delay line */
862 /* Update number of lost(coalesced) ticks. */
863 tick_lost += pending - 1;
866 /* Last tick duration (usec). */
867 tick_last = (t.tv_sec - prev_t.tv_sec) * 1000000 +
868 (t.tv_usec - prev_t.tv_usec);
869 /* Last tick vs standard tick difference (usec). */
870 tick_delta = (tick_last * hz - 1000000) / hz;
871 /* Accumulated tick difference (usec). */
872 tick_delta_sum += tick_delta;
877 * Adjust curr_time if accumulated tick difference greater than
878 * 'standard' tick. Since curr_time should be monotonically increasing,
879 * we do positive adjustment as required and throttle curr_time in
880 * case of negative adjustment.
883 if (tick_delta_sum - tick >= 0) {
884 int diff = tick_delta_sum / tick;
888 tick_delta_sum %= tick;
890 } else if (tick_delta_sum + tick <= 0) {
893 tick_delta_sum += tick;
897 for (i = 0; i < 3; i++) {
899 while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time)) {
900 if (h->p[0].key > curr_time)
901 printf("dummynet: warning, "
902 "heap %d is %d ticks late\n",
903 i, (int)(curr_time - h->p[0].key));
904 /* store a copy before heap_extract */
906 /* need to extract before processing */
907 heap_extract(h, NULL);
909 ready_event(p, &head, &tail);
911 struct dn_pipe *pipe = p;
912 if (pipe->if_name[0] != '\0')
913 printf("dummynet: bad ready_event_wfq "
914 "for pipe %s\n", pipe->if_name);
916 ready_event_wfq(p, &head, &tail);
918 transmit_event(p, &head, &tail);
922 /* Sweep pipes trying to expire idle flow_queues. */
923 for (i = 0; i < HASHSIZE; i++)
924 SLIST_FOREACH(pipe, &pipehash[i], next)
925 if (pipe->idle_heap.elements > 0 &&
926 DN_KEY_LT(pipe->idle_heap.p[0].key, pipe->V)) {
927 struct dn_flow_queue *q =
928 pipe->idle_heap.p[0].object;
930 heap_extract(&(pipe->idle_heap), NULL);
931 /* Mark timestamp as invalid. */
933 pipe->sum -= q->fs->weight;
941 callout_reset(&dn_timeout, 1, dummynet, NULL);
945 dummynet_send(struct mbuf *m)
947 struct dn_pkt_tag *pkt;
952 for (; m != NULL; m = n) {
955 if (m_tag_first(m) == NULL) {
956 pkt = NULL; /* probably unnecessary */
964 ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
967 ip = mtod(m, struct ip *);
968 #ifndef __linux__ /* restore net format for FreeBSD */
969 ip->ip_len = htons(ip->ip_len);
970 ip->ip_off = htons(ip->ip_off);
972 netisr_dispatch(NETISR_IP, m);
976 netisr_dispatch(NETISR_IPV6, m);
980 ip6_output(m, NULL, NULL, IPV6_FORWARDING, NULL, NULL, NULL);
984 if (bridge_dn_p != NULL)
985 ((*bridge_dn_p)(m, pkt->ifp));
987 printf("dummynet: if_bridge not loaded\n");
990 case DN_TO_ETH_DEMUX:
992 * The Ethernet code assumes the Ethernet header is
993 * contiguous in the first mbuf header.
994 * Insure this is true.
996 if (m->m_len < ETHER_HDR_LEN &&
997 (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) {
998 printf("dummynet/ether: pullup failed, "
999 "dropping packet\n");
1002 ether_demux(m->m_pkthdr.rcvif, m);
1005 ether_output_frame(pkt->ifp, m);
1009 /* drop the packet after some time */
1011 netisr_dispatch(-1, m); /* -1 drop the packet */
1018 printf("dummynet: bad switch %d!\n", pkt->dn_dir);
1026 * Unconditionally expire empty queues in case of shortage.
1027 * Returns the number of queues freed.
1030 expire_queues(struct dn_flow_set *fs)
1032 struct dn_flow_queue *q, *prev ;
1033 int i, initial_elements = fs->rq_elements ;
1035 if (fs->last_expired == time_uptime)
1037 fs->last_expired = time_uptime ;
1038 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is overflow */
1039 for (prev=NULL, q = fs->rq[i] ; q != NULL ; )
1040 if (q->head != NULL || q->S != q->F+1) {
1043 } else { /* entry is idle, expire it */
1044 struct dn_flow_queue *old_q = q ;
1047 prev->next = q = q->next ;
1049 fs->rq[i] = q = q->next ;
1051 free(old_q, M_DUMMYNET);
1053 return initial_elements - fs->rq_elements ;
1057 * If room, create a new queue and put at head of slot i;
1058 * otherwise, create or use the default queue.
1060 static struct dn_flow_queue *
1061 create_queue(struct dn_flow_set *fs, int i)
1063 struct dn_flow_queue *q;
1065 if (fs->rq_elements > fs->rq_size * dn_max_ratio &&
1066 expire_queues(fs) == 0) {
1067 /* No way to get room, use or create overflow queue. */
1069 if (fs->rq[i] != NULL)
1072 q = malloc(sizeof(*q), M_DUMMYNET, M_NOWAIT | M_ZERO);
1074 printf("dummynet: sorry, cannot allocate queue for new flow\n");
1079 q->next = fs->rq[i];
1080 q->S = q->F + 1; /* hack - mark timestamp as invalid. */
1081 q->numbytes = io_fast ? fs->pipe->bandwidth : 0;
1088 * Given a flow_set and a pkt in last_pkt, find a matching queue
1089 * after appropriate masking. The queue is moved to front
1090 * so that further searches take less time.
1092 static struct dn_flow_queue *
1093 find_queue(struct dn_flow_set *fs, struct ipfw_flow_id *id)
1095 int i = 0 ; /* we need i and q for new allocations */
1096 struct dn_flow_queue *q, *prev;
1097 int is_v6 = IS_IP6_FLOW_ID(id);
1099 if ( !(fs->flags_fs & DN_HAVE_FLOW_MASK) )
1102 /* first, do the masking, then hash */
1103 id->dst_port &= fs->flow_mask.dst_port ;
1104 id->src_port &= fs->flow_mask.src_port ;
1105 id->proto &= fs->flow_mask.proto ;
1106 id->flags = 0 ; /* we don't care about this one */
1108 APPLY_MASK(&id->dst_ip6, &fs->flow_mask.dst_ip6);
1109 APPLY_MASK(&id->src_ip6, &fs->flow_mask.src_ip6);
1110 id->flow_id6 &= fs->flow_mask.flow_id6;
1112 i = ((id->dst_ip6.__u6_addr.__u6_addr32[0]) & 0xffff)^
1113 ((id->dst_ip6.__u6_addr.__u6_addr32[1]) & 0xffff)^
1114 ((id->dst_ip6.__u6_addr.__u6_addr32[2]) & 0xffff)^
1115 ((id->dst_ip6.__u6_addr.__u6_addr32[3]) & 0xffff)^
1117 ((id->dst_ip6.__u6_addr.__u6_addr32[0] >> 15) & 0xffff)^
1118 ((id->dst_ip6.__u6_addr.__u6_addr32[1] >> 15) & 0xffff)^
1119 ((id->dst_ip6.__u6_addr.__u6_addr32[2] >> 15) & 0xffff)^
1120 ((id->dst_ip6.__u6_addr.__u6_addr32[3] >> 15) & 0xffff)^
1122 ((id->src_ip6.__u6_addr.__u6_addr32[0] << 1) & 0xfffff)^
1123 ((id->src_ip6.__u6_addr.__u6_addr32[1] << 1) & 0xfffff)^
1124 ((id->src_ip6.__u6_addr.__u6_addr32[2] << 1) & 0xfffff)^
1125 ((id->src_ip6.__u6_addr.__u6_addr32[3] << 1) & 0xfffff)^
1127 ((id->src_ip6.__u6_addr.__u6_addr32[0] << 16) & 0xffff)^
1128 ((id->src_ip6.__u6_addr.__u6_addr32[1] << 16) & 0xffff)^
1129 ((id->src_ip6.__u6_addr.__u6_addr32[2] << 16) & 0xffff)^
1130 ((id->src_ip6.__u6_addr.__u6_addr32[3] << 16) & 0xffff)^
1132 (id->dst_port << 1) ^ (id->src_port) ^
1136 id->dst_ip &= fs->flow_mask.dst_ip ;
1137 id->src_ip &= fs->flow_mask.src_ip ;
1139 i = ( (id->dst_ip) & 0xffff ) ^
1140 ( (id->dst_ip >> 15) & 0xffff ) ^
1141 ( (id->src_ip << 1) & 0xffff ) ^
1142 ( (id->src_ip >> 16 ) & 0xffff ) ^
1143 (id->dst_port << 1) ^ (id->src_port) ^
1146 i = i % fs->rq_size ;
1147 /* finally, scan the current list for a match */
1149 for (prev=NULL, q = fs->rq[i] ; q ; ) {
1152 IN6_ARE_ADDR_EQUAL(&id->dst_ip6,&q->id.dst_ip6) &&
1153 IN6_ARE_ADDR_EQUAL(&id->src_ip6,&q->id.src_ip6) &&
1154 id->dst_port == q->id.dst_port &&
1155 id->src_port == q->id.src_port &&
1156 id->proto == q->id.proto &&
1157 id->flags == q->id.flags &&
1158 id->flow_id6 == q->id.flow_id6)
1161 if (!is_v6 && id->dst_ip == q->id.dst_ip &&
1162 id->src_ip == q->id.src_ip &&
1163 id->dst_port == q->id.dst_port &&
1164 id->src_port == q->id.src_port &&
1165 id->proto == q->id.proto &&
1166 id->flags == q->id.flags)
1169 /* No match. Check if we can expire the entry */
1170 if (pipe_expire && q->head == NULL && q->S == q->F+1 ) {
1171 /* entry is idle and not in any heap, expire it */
1172 struct dn_flow_queue *old_q = q ;
1175 prev->next = q = q->next ;
1177 fs->rq[i] = q = q->next ;
1179 free(old_q, M_DUMMYNET);
1185 if (q && prev != NULL) { /* found and not in front */
1186 prev->next = q->next ;
1187 q->next = fs->rq[i] ;
1191 if (q == NULL) { /* no match, need to allocate a new entry */
1192 q = create_queue(fs, i);
1200 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len)
1205 * RED calculates the average queue size (avg) using a low-pass filter
1206 * with an exponential weighted (w_q) moving average:
1207 * avg <- (1-w_q) * avg + w_q * q_size
1208 * where q_size is the queue length (measured in bytes or * packets).
1210 * If q_size == 0, we compute the idle time for the link, and set
1211 * avg = (1 - w_q)^(idle/s)
1212 * where s is the time needed for transmitting a medium-sized packet.
1214 * Now, if avg < min_th the packet is enqueued.
1215 * If avg > max_th the packet is dropped. Otherwise, the packet is
1216 * dropped with probability P function of avg.
1221 /* Queue in bytes or packets? */
1222 u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ?
1223 q->len_bytes : q->len;
1225 DPRINTF(("\ndummynet: %d q: %2u ", (int)curr_time, q_size));
1227 /* Average queue size estimation. */
1229 /* Queue is not empty, avg <- avg + (q_size - avg) * w_q */
1230 int diff = SCALE(q_size) - q->avg;
1231 int64_t v = SCALE_MUL((int64_t)diff, (int64_t)fs->w_q);
1236 * Queue is empty, find for how long the queue has been
1237 * empty and use a lookup table for computing
1238 * (1 - * w_q)^(idle_time/s) where s is the time to send a
1240 * XXX check wraps...
1243 u_int t = div64(curr_time - q->q_time,
1246 q->avg = (t >= 0 && t < fs->lookup_depth) ?
1247 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0;
1250 DPRINTF(("dummynet: avg: %u ", SCALE_VAL(q->avg)));
1252 /* Should i drop? */
1253 if (q->avg < fs->min_th) {
1255 return (0); /* accept packet */
1257 if (q->avg >= fs->max_th) { /* average queue >= max threshold */
1258 if (fs->flags_fs & DN_IS_GENTLE_RED) {
1260 * According to Gentle-RED, if avg is greater than
1261 * max_th the packet is dropped with a probability
1262 * p_b = c_3 * avg - c_4
1263 * where c_3 = (1 - max_p) / max_th
1264 * c_4 = 1 - 2 * max_p
1266 p_b = SCALE_MUL((int64_t)fs->c_3, (int64_t)q->avg) -
1270 DPRINTF(("dummynet: - drop"));
1273 } else if (q->avg > fs->min_th) {
1275 * We compute p_b using the linear dropping function
1276 * p_b = c_1 * avg - c_2
1277 * where c_1 = max_p / (max_th - min_th)
1278 * c_2 = max_p * min_th / (max_th - min_th)
1280 p_b = SCALE_MUL((int64_t)fs->c_1, (int64_t)q->avg) - fs->c_2;
1283 if (fs->flags_fs & DN_QSIZE_IS_BYTES)
1284 p_b = div64(p_b * len, fs->max_pkt_size);
1285 if (++q->count == 0)
1286 q->random = random() & 0xffff;
1289 * q->count counts packets arrived since last drop, so a greater
1290 * value of q->count means a greater packet drop probability.
1292 if (SCALE_MUL(p_b, SCALE((int64_t)q->count)) > q->random) {
1294 DPRINTF(("dummynet: - red drop"));
1295 /* After a drop we calculate a new random value. */
1296 q->random = random() & 0xffff;
1297 return (1); /* drop */
1300 /* End of RED algorithm. */
1302 return (0); /* accept */
1305 static __inline struct dn_flow_set *
1306 locate_flowset(int fs_nr)
1308 struct dn_flow_set *fs;
1310 SLIST_FOREACH(fs, &flowsethash[HASH(fs_nr)], next)
1311 if (fs->fs_nr == fs_nr)
1317 static __inline struct dn_pipe *
1318 locate_pipe(int pipe_nr)
1320 struct dn_pipe *pipe;
1322 SLIST_FOREACH(pipe, &pipehash[HASH(pipe_nr)], next)
1323 if (pipe->pipe_nr == pipe_nr)
1330 * dummynet hook for packets. Below 'pipe' is a pipe or a queue
1331 * depending on whether WF2Q or fixed bw is used.
1333 * pipe_nr pipe or queue the packet is destined for.
1334 * dir where shall we send the packet after dummynet.
1335 * m the mbuf with the packet
1336 * ifp the 'ifp' parameter from the caller.
1337 * NULL in ip_input, destination interface in ip_output,
1338 * rule matching rule, in case of multiple passes
1341 dummynet_io(struct mbuf **m0, int dir, struct ip_fw_args *fwa)
1343 struct mbuf *m = *m0, *head = NULL, *tail = NULL;
1344 struct dn_pkt_tag *pkt;
1346 struct dn_flow_set *fs = NULL;
1347 struct dn_pipe *pipe;
1348 uint64_t len = m->m_pkthdr.len;
1349 struct dn_flow_queue *q = NULL;
1351 ipfw_insn *cmd = ACTION_PTR(fwa->rule);
1353 KASSERT(m->m_nextpkt == NULL,
1354 ("dummynet_io: mbuf queue passed to dummynet"));
1356 if (cmd->opcode == O_LOG)
1358 if (cmd->opcode == O_ALTQ)
1360 if (cmd->opcode == O_TAG)
1362 is_pipe = (cmd->opcode == O_PIPE);
1367 * This is a dummynet rule, so we expect an O_PIPE or O_QUEUE rule.
1369 * XXXGL: probably the pipe->fs and fs->pipe logic here
1370 * below can be simplified.
1373 pipe = locate_pipe(fwa->cookie);
1377 fs = locate_flowset(fwa->cookie);
1380 goto dropit; /* This queue/pipe does not exist! */
1382 if (pipe == NULL) { /* Must be a queue, try find a matching pipe. */
1383 pipe = locate_pipe(fs->parent_nr);
1387 printf("dummynet: no pipe %d for queue %d, drop pkt\n",
1388 fs->parent_nr, fs->fs_nr);
1392 q = find_queue(fs, &(fwa->f_id));
1394 goto dropit; /* Cannot allocate queue. */
1396 /* Update statistics, then check reasons to drop pkt. */
1397 q->tot_bytes += len;
1399 if (fs->plr && random() < fs->plr)
1400 goto dropit; /* Random pkt drop. */
1401 if (fs->flags_fs & DN_QSIZE_IS_BYTES) {
1402 if (q->len_bytes > fs->qsize)
1403 goto dropit; /* Queue size overflow. */
1405 if (q->len >= fs->qsize)
1406 goto dropit; /* Queue count overflow. */
1408 if (fs->flags_fs & DN_IS_RED && red_drops(fs, q, len))
1411 /* XXX expensive to zero, see if we can remove it. */
1412 mtag = m_tag_get(PACKET_TAG_DUMMYNET,
1413 sizeof(struct dn_pkt_tag), M_NOWAIT | M_ZERO);
1415 goto dropit; /* Cannot allocate packet header. */
1416 m_tag_prepend(m, mtag); /* Attach to mbuf chain. */
1418 pkt = (struct dn_pkt_tag *)(mtag + 1);
1420 * Ok, i can handle the pkt now...
1421 * Build and enqueue packet + parameters.
1423 pkt->rule = fwa->rule;
1426 pkt->ifp = fwa->oif;
1428 if (q->head == NULL)
1431 q->tail->m_nextpkt = m;
1434 q->len_bytes += len;
1436 if (q->head != m) /* Flow was not idle, we are done. */
1439 if (q->q_time < (uint32_t)curr_time)
1440 q->numbytes = io_fast ? fs->pipe->bandwidth : 0;
1441 q->q_time = curr_time;
1444 * If we reach this point the flow was previously idle, so we need
1445 * to schedule it. This involves different actions for fixed-rate or
1449 /* Fixed-rate queue: just insert into the ready_heap. */
1452 if (pipe->bandwidth) {
1453 q->extra_bits = compute_extra_bits(m, pipe);
1454 t = set_ticks(m, q, pipe);
1456 q->sched_time = curr_time;
1457 if (t == 0) /* Must process it now. */
1458 ready_event(q, &head, &tail);
1460 heap_insert(&ready_heap, curr_time + t , q);
1463 * WF2Q. First, compute start time S: if the flow was
1464 * idle (S = F + 1) set S to the virtual time V for the
1465 * controlling pipe, and update the sum of weights for the pipe;
1466 * otherwise, remove flow from idle_heap and set S to max(F,V).
1467 * Second, compute finish time F = S + len / weight.
1468 * Third, if pipe was idle, update V = max(S, V).
1469 * Fourth, count one more backlogged flow.
1471 if (DN_KEY_GT(q->S, q->F)) { /* Means timestamps are invalid. */
1473 pipe->sum += fs->weight; /* Add weight of new queue. */
1475 heap_extract(&(pipe->idle_heap), q);
1476 q->S = MAX64(q->F, pipe->V);
1478 q->F = div64(q->S + (len << MY_M), fs->weight);
1480 if (pipe->not_eligible_heap.elements == 0 &&
1481 pipe->scheduler_heap.elements == 0)
1482 pipe->V = MAX64(q->S, pipe->V);
1485 * Look at eligibility. A flow is not eligibile if S>V (when
1486 * this happens, it means that there is some other flow already
1487 * scheduled for the same pipe, so the scheduler_heap cannot be
1488 * empty). If the flow is not eligible we just store it in the
1489 * not_eligible_heap. Otherwise, we store in the scheduler_heap
1490 * and possibly invoke ready_event_wfq() right now if there is
1492 * Note that for all flows in scheduler_heap (SCH), S_i <= V,
1493 * and for all flows in not_eligible_heap (NEH), S_i > V.
1494 * So when we need to compute max(V, min(S_i)) forall i in
1495 * SCH+NEH, we only need to look into NEH.
1497 if (DN_KEY_GT(q->S, pipe->V)) { /* Not eligible. */
1498 if (pipe->scheduler_heap.elements == 0)
1499 printf("dummynet: ++ ouch! not eligible but empty scheduler!\n");
1500 heap_insert(&(pipe->not_eligible_heap), q->S, q);
1502 heap_insert(&(pipe->scheduler_heap), q->F, q);
1503 if (pipe->numbytes >= 0) { /* Pipe is idle. */
1504 if (pipe->scheduler_heap.elements != 1)
1505 printf("dummynet: OUCH! pipe should have been idle!\n");
1506 DPRINTF(("dummynet: waking up pipe %d at %d\n",
1507 pipe->pipe_nr, (int)(q->F >> MY_M)));
1508 pipe->sched_time = curr_time;
1509 ready_event_wfq(pipe, &head, &tail);
1514 if (head == m && dir != DN_TO_IFB_FWD && dir != DN_TO_ETH_DEMUX &&
1515 dir != DN_TO_ETH_OUT) { /* Fast io. */
1517 if (m->m_nextpkt != NULL)
1518 printf("dummynet: fast io: pkt chain detected!\n");
1519 head = m->m_nextpkt = NULL;
1521 *m0 = NULL; /* Normal io. */
1525 dummynet_send(head);
1534 * set the tag, if present. dn_tag_get cannot fail
1535 * so we need to check first
1537 if (m_tag_first(m)) {
1538 pkt = dn_tag_get(m);
1539 pkt->dn_dir = DN_TO_DROP;
1541 dummynet_send(m); /* drop the packet */
1543 return ((fs && (fs->flags_fs & DN_NOERROR)) ? 0 : ENOBUFS);
1547 * Below, the rt_unref is only needed when (pkt->dn_dir == DN_TO_IP_OUT)
1548 * Doing this would probably save us the initial bzero of dn_pkt
1550 #define DN_FREE_PKT(_m) do { \
1555 * Dispose all packets and flow_queues on a flow_set.
1556 * If all=1, also remove red lookup table and other storage,
1557 * including the descriptor itself.
1558 * For the one in dn_pipe MUST also cleanup ready_heap...
1561 purge_flow_set(struct dn_flow_set *fs, int all)
1563 struct dn_flow_queue *q, *qn;
1566 DUMMYNET_LOCK_ASSERT();
1568 for (i = 0; i <= fs->rq_size; i++) {
1569 for (q = fs->rq[i]; q != NULL; q = qn) {
1570 struct mbuf *m, *mnext;
1573 while ((m = mnext) != NULL) {
1574 mnext = m->m_nextpkt;
1578 free(q, M_DUMMYNET);
1583 fs->rq_elements = 0;
1585 /* RED - free lookup table. */
1586 if (fs->w_q_lookup != NULL)
1587 free(fs->w_q_lookup, M_DUMMYNET);
1589 free(fs->rq, M_DUMMYNET);
1590 /* If this fs is not part of a pipe, free it. */
1591 if (fs->pipe == NULL || fs != &(fs->pipe->fs))
1592 free(fs, M_DUMMYNET);
1597 * Dispose all packets queued on a pipe (not a flow_set).
1598 * Also free all resources associated to a pipe, which is about
1602 purge_pipe(struct dn_pipe *pipe)
1604 struct mbuf *m, *mnext;
1606 purge_flow_set( &(pipe->fs), 1 );
1609 while ((m = mnext) != NULL) {
1610 mnext = m->m_nextpkt;
1614 heap_free( &(pipe->scheduler_heap) );
1615 heap_free( &(pipe->not_eligible_heap) );
1616 heap_free( &(pipe->idle_heap) );
1620 * Delete all pipes and heaps returning memory. Must also
1621 * remove references from all ipfw rules to all pipes.
1624 dummynet_flush(void)
1626 struct dn_pipe *pipe, *pipe1;
1627 struct dn_flow_set *fs, *fs1;
1631 /* Free heaps so we don't have unwanted events. */
1632 heap_free(&ready_heap);
1633 heap_free(&wfq_ready_heap);
1634 heap_free(&extract_heap);
1637 * Now purge all queued pkts and delete all pipes.
1639 * XXXGL: can we merge the for(;;) cycles into one or not?
1641 for (i = 0; i < HASHSIZE; i++)
1642 SLIST_FOREACH_SAFE(fs, &flowsethash[i], next, fs1) {
1643 SLIST_REMOVE(&flowsethash[i], fs, dn_flow_set, next);
1644 purge_flow_set(fs, 1);
1646 for (i = 0; i < HASHSIZE; i++)
1647 SLIST_FOREACH_SAFE(pipe, &pipehash[i], next, pipe1) {
1648 SLIST_REMOVE(&pipehash[i], pipe, dn_pipe, next);
1655 extern struct ip_fw *ip_fw_default_rule;
1657 dn_rule_delete_fs(struct dn_flow_set *fs, void *r)
1660 struct dn_flow_queue *q ;
1663 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is ovflow */
1664 for (q = fs->rq[i] ; q ; q = q->next )
1665 for (m = q->head ; m ; m = m->m_nextpkt ) {
1666 struct dn_pkt_tag *pkt = dn_tag_get(m) ;
1668 pkt->rule = ip_fw_default_rule ;
1673 * When a firewall rule is deleted, scan all queues and remove the pointer
1674 * to the rule from matching packets, making them point to the default rule.
1675 * The pointer is used to reinject packets in case one_pass = 0.
1678 dn_rule_delete(void *r)
1680 struct dn_pipe *pipe;
1681 struct dn_flow_set *fs;
1682 struct dn_pkt_tag *pkt;
1688 * If the rule references a queue (dn_flow_set), then scan
1689 * the flow set, otherwise scan pipes. Should do either, but doing
1690 * both does not harm.
1692 for (i = 0; i < HASHSIZE; i++)
1693 SLIST_FOREACH(fs, &flowsethash[i], next)
1694 dn_rule_delete_fs(fs, r);
1696 for (i = 0; i < HASHSIZE; i++)
1697 SLIST_FOREACH(pipe, &pipehash[i], next) {
1699 dn_rule_delete_fs(fs, r);
1700 for (m = pipe->head ; m ; m = m->m_nextpkt ) {
1701 pkt = dn_tag_get(m);
1703 pkt->rule = ip_fw_default_rule;
1710 * setup RED parameters
1713 config_red(struct dn_flow_set *p, struct dn_flow_set *x)
1718 x->min_th = SCALE(p->min_th);
1719 x->max_th = SCALE(p->max_th);
1720 x->max_p = p->max_p;
1722 x->c_1 = p->max_p / (p->max_th - p->min_th);
1723 x->c_2 = SCALE_MUL(x->c_1, SCALE(p->min_th));
1725 if (x->flags_fs & DN_IS_GENTLE_RED) {
1726 x->c_3 = (SCALE(1) - p->max_p) / p->max_th;
1727 x->c_4 = SCALE(1) - 2 * p->max_p;
1730 /* If the lookup table already exist, free and create it again. */
1731 if (x->w_q_lookup) {
1732 free(x->w_q_lookup, M_DUMMYNET);
1733 x->w_q_lookup = NULL;
1735 if (red_lookup_depth == 0) {
1736 printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth"
1738 free(x, M_DUMMYNET);
1741 x->lookup_depth = red_lookup_depth;
1742 x->w_q_lookup = (u_int *)malloc(x->lookup_depth * sizeof(int),
1743 M_DUMMYNET, M_NOWAIT);
1744 if (x->w_q_lookup == NULL) {
1745 printf("dummynet: sorry, cannot allocate red lookup table\n");
1746 free(x, M_DUMMYNET);
1750 /* Fill the lookup table with (1 - w_q)^x */
1751 x->lookup_step = p->lookup_step;
1752 x->lookup_weight = p->lookup_weight;
1753 x->w_q_lookup[0] = SCALE(1) - x->w_q;
1755 for (i = 1; i < x->lookup_depth; i++)
1757 SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight);
1759 if (red_avg_pkt_size < 1)
1760 red_avg_pkt_size = 512;
1761 x->avg_pkt_size = red_avg_pkt_size;
1762 if (red_max_pkt_size < 1)
1763 red_max_pkt_size = 1500;
1764 x->max_pkt_size = red_max_pkt_size;
1769 alloc_hash(struct dn_flow_set *x, struct dn_flow_set *pfs)
1771 if (x->flags_fs & DN_HAVE_FLOW_MASK) { /* allocate some slots */
1772 int l = pfs->rq_size;
1778 else if (l > DN_MAX_HASH_SIZE)
1779 l = DN_MAX_HASH_SIZE;
1781 } else /* one is enough for null mask */
1783 x->rq = malloc((1 + x->rq_size) * sizeof(struct dn_flow_queue *),
1784 M_DUMMYNET, M_NOWAIT | M_ZERO);
1785 if (x->rq == NULL) {
1786 printf("dummynet: sorry, cannot allocate queue\n");
1794 set_fs_parms(struct dn_flow_set *x, struct dn_flow_set *src)
1796 x->flags_fs = src->flags_fs;
1797 x->qsize = src->qsize;
1799 x->flow_mask = src->flow_mask;
1800 if (x->flags_fs & DN_QSIZE_IS_BYTES) {
1801 if (x->qsize > pipe_byte_limit)
1802 x->qsize = 1024 * 1024;
1806 if (x->qsize > pipe_slot_limit)
1809 /* Configuring RED. */
1810 if (x->flags_fs & DN_IS_RED)
1811 config_red(src, x); /* XXX should check errors */
1815 * Setup pipe or queue parameters.
1818 config_pipe(struct dn_pipe *p)
1820 struct dn_flow_set *pfs = &(p->fs);
1821 struct dn_flow_queue *q;
1825 * The config program passes parameters as follows:
1826 * bw = bits/second (0 means no limits),
1827 * delay = ms, must be translated into ticks.
1828 * qsize = slots/bytes
1830 p->delay = (p->delay * hz) / 1000;
1831 /* We need either a pipe number or a flow_set number. */
1832 if (p->pipe_nr == 0 && pfs->fs_nr == 0)
1834 if (p->pipe_nr != 0 && pfs->fs_nr != 0)
1836 if (p->pipe_nr != 0) { /* this is a pipe */
1837 struct dn_pipe *pipe;
1840 pipe = locate_pipe(p->pipe_nr); /* locate pipe */
1842 if (pipe == NULL) { /* new pipe */
1843 pipe = malloc(sizeof(struct dn_pipe), M_DUMMYNET,
1847 printf("dummynet: no memory for new pipe\n");
1850 pipe->pipe_nr = p->pipe_nr;
1851 pipe->fs.pipe = pipe;
1853 * idle_heap is the only one from which
1854 * we extract from the middle.
1856 pipe->idle_heap.size = pipe->idle_heap.elements = 0;
1857 pipe->idle_heap.offset =
1858 offsetof(struct dn_flow_queue, heap_pos);
1860 /* Flush accumulated credit for all queues. */
1861 for (i = 0; i <= pipe->fs.rq_size; i++)
1862 for (q = pipe->fs.rq[i]; q; q = q->next)
1863 q->numbytes = io_fast ? p->bandwidth : 0;
1865 pipe->bandwidth = p->bandwidth;
1866 pipe->numbytes = 0; /* just in case... */
1867 bcopy(p->if_name, pipe->if_name, sizeof(p->if_name));
1868 pipe->ifp = NULL; /* reset interface ptr */
1869 pipe->delay = p->delay;
1870 set_fs_parms(&(pipe->fs), pfs);
1872 /* Handle changes in the delay profile. */
1873 if (p->samples_no > 0) {
1874 if (pipe->samples_no != p->samples_no) {
1875 if (pipe->samples != NULL)
1876 free(pipe->samples, M_DUMMYNET);
1878 malloc(p->samples_no*sizeof(dn_key),
1879 M_DUMMYNET, M_NOWAIT | M_ZERO);
1880 if (pipe->samples == NULL) {
1882 printf("dummynet: no memory "
1883 "for new samples\n");
1886 pipe->samples_no = p->samples_no;
1889 strncpy(pipe->name,p->name,sizeof(pipe->name));
1890 pipe->loss_level = p->loss_level;
1891 for (i = 0; i<pipe->samples_no; ++i)
1892 pipe->samples[i] = p->samples[i];
1893 } else if (pipe->samples != NULL) {
1894 free(pipe->samples, M_DUMMYNET);
1895 pipe->samples = NULL;
1896 pipe->samples_no = 0;
1899 if (pipe->fs.rq == NULL) { /* a new pipe */
1900 error = alloc_hash(&(pipe->fs), pfs);
1906 SLIST_INSERT_HEAD(&pipehash[HASH(pipe->pipe_nr)],
1910 } else { /* config queue */
1911 struct dn_flow_set *fs;
1914 fs = locate_flowset(pfs->fs_nr); /* locate flow_set */
1916 if (fs == NULL) { /* new */
1917 if (pfs->parent_nr == 0) { /* need link to a pipe */
1921 fs = malloc(sizeof(struct dn_flow_set), M_DUMMYNET,
1926 "dummynet: no memory for new flow_set\n");
1929 fs->fs_nr = pfs->fs_nr;
1930 fs->parent_nr = pfs->parent_nr;
1931 fs->weight = pfs->weight;
1932 if (fs->weight == 0)
1934 else if (fs->weight > 100)
1938 * Change parent pipe not allowed;
1939 * must delete and recreate.
1941 if (pfs->parent_nr != 0 &&
1942 fs->parent_nr != pfs->parent_nr) {
1948 set_fs_parms(fs, pfs);
1950 if (fs->rq == NULL) { /* a new flow_set */
1951 error = alloc_hash(fs, pfs);
1954 free(fs, M_DUMMYNET);
1957 SLIST_INSERT_HEAD(&flowsethash[HASH(fs->fs_nr)],
1966 * Helper function to remove from a heap queues which are linked to
1967 * a flow_set about to be deleted.
1970 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs)
1972 int i = 0, found = 0 ;
1973 for (; i < h->elements ;)
1974 if ( ((struct dn_flow_queue *)h->p[i].object)->fs == fs) {
1976 h->p[i] = h->p[h->elements] ;
1985 * helper function to remove a pipe from a heap (can be there at most once)
1988 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p)
1990 if (h->elements > 0) {
1992 for (i=0; i < h->elements ; i++ ) {
1993 if (h->p[i].object == p) { /* found it */
1995 h->p[i] = h->p[h->elements] ;
2004 * drain all queues. Called in case of severe mbuf shortage.
2007 dummynet_drain(void)
2009 struct dn_flow_set *fs;
2010 struct dn_pipe *pipe;
2011 struct mbuf *m, *mnext;
2014 DUMMYNET_LOCK_ASSERT();
2016 heap_free(&ready_heap);
2017 heap_free(&wfq_ready_heap);
2018 heap_free(&extract_heap);
2019 /* remove all references to this pipe from flow_sets */
2020 for (i = 0; i < HASHSIZE; i++)
2021 SLIST_FOREACH(fs, &flowsethash[i], next)
2022 purge_flow_set(fs, 0);
2024 for (i = 0; i < HASHSIZE; i++) {
2025 SLIST_FOREACH(pipe, &pipehash[i], next) {
2026 purge_flow_set(&(pipe->fs), 0);
2029 while ((m = mnext) != NULL) {
2030 mnext = m->m_nextpkt;
2033 pipe->head = pipe->tail = NULL;
2039 * Fully delete a pipe or a queue, cleaning up associated info.
2042 delete_pipe(struct dn_pipe *p)
2045 if (p->pipe_nr == 0 && p->fs.fs_nr == 0)
2047 if (p->pipe_nr != 0 && p->fs.fs_nr != 0)
2049 if (p->pipe_nr != 0) { /* this is an old-style pipe */
2050 struct dn_pipe *pipe;
2051 struct dn_flow_set *fs;
2055 pipe = locate_pipe(p->pipe_nr); /* locate pipe */
2059 return (ENOENT); /* not found */
2062 /* Unlink from list of pipes. */
2063 SLIST_REMOVE(&pipehash[HASH(pipe->pipe_nr)], pipe, dn_pipe, next);
2065 /* Remove all references to this pipe from flow_sets. */
2066 for (i = 0; i < HASHSIZE; i++)
2067 SLIST_FOREACH(fs, &flowsethash[i], next)
2068 if (fs->pipe == pipe) {
2069 printf("dummynet: ++ ref to pipe %d from fs %d\n",
2070 p->pipe_nr, fs->fs_nr);
2072 purge_flow_set(fs, 0);
2074 fs_remove_from_heap(&ready_heap, &(pipe->fs));
2075 purge_pipe(pipe); /* remove all data associated to this pipe */
2076 /* remove reference to here from extract_heap and wfq_ready_heap */
2077 pipe_remove_from_heap(&extract_heap, pipe);
2078 pipe_remove_from_heap(&wfq_ready_heap, pipe);
2082 } else { /* this is a WF2Q queue (dn_flow_set) */
2083 struct dn_flow_set *fs;
2086 fs = locate_flowset(p->fs.fs_nr); /* locate set */
2090 return (ENOENT); /* not found */
2093 /* Unlink from list of flowsets. */
2094 SLIST_REMOVE( &flowsethash[HASH(fs->fs_nr)], fs, dn_flow_set, next);
2096 if (fs->pipe != NULL) {
2097 /* Update total weight on parent pipe and cleanup parent heaps. */
2098 fs->pipe->sum -= fs->weight * fs->backlogged ;
2099 fs_remove_from_heap(&(fs->pipe->not_eligible_heap), fs);
2100 fs_remove_from_heap(&(fs->pipe->scheduler_heap), fs);
2101 #if 1 /* XXX should i remove from idle_heap as well ? */
2102 fs_remove_from_heap(&(fs->pipe->idle_heap), fs);
2105 purge_flow_set(fs, 1);
2112 * helper function used to copy data from kernel in DUMMYNET_GET
2115 dn_copy_set(struct dn_flow_set *set, char *bp)
2118 struct dn_flow_queue *q, *qp = (struct dn_flow_queue *)bp;
2120 DUMMYNET_LOCK_ASSERT();
2122 for (i = 0 ; i <= set->rq_size ; i++)
2123 for (q = set->rq[i] ; q ; q = q->next, qp++ ) {
2124 if (q->hash_slot != i)
2125 printf("dummynet: ++ at %d: wrong slot (have %d, "
2126 "should be %d)\n", copied, q->hash_slot, i);
2128 printf("dummynet: ++ at %d: wrong fs ptr (have %p, should be %p)\n",
2131 bcopy(q, qp, sizeof( *q ) );
2132 /* cleanup pointers */
2134 qp->head = qp->tail = NULL ;
2137 if (copied != set->rq_elements)
2138 printf("dummynet: ++ wrong count, have %d should be %d\n",
2139 copied, set->rq_elements);
2146 struct dn_flow_set *fs;
2147 struct dn_pipe *pipe;
2151 DUMMYNET_LOCK_ASSERT();
2153 * Compute size of data structures: list of pipes and flow_sets.
2155 for (i = 0; i < HASHSIZE; i++) {
2156 SLIST_FOREACH(pipe, &pipehash[i], next)
2157 size += sizeof(*pipe) +
2158 pipe->fs.rq_elements * sizeof(struct dn_flow_queue);
2159 SLIST_FOREACH(fs, &flowsethash[i], next)
2160 size += sizeof (*fs) +
2161 fs->rq_elements * sizeof(struct dn_flow_queue);
2167 dummynet_get(struct sockopt *sopt)
2169 char *buf, *bp ; /* bp is the "copy-pointer" */
2171 struct dn_flow_set *fs;
2172 struct dn_pipe *pipe;
2175 /* XXX lock held too long */
2178 * XXX: Ugly, but we need to allocate memory with M_WAITOK flag and we
2179 * cannot use this flag while holding a mutex.
2181 for (i = 0; i < 10; i++) {
2182 size = dn_calc_size();
2184 buf = malloc(size, M_TEMP, M_WAITOK);
2186 if (size == dn_calc_size())
2196 for (i = 0; i < HASHSIZE; i++)
2197 SLIST_FOREACH(pipe, &pipehash[i], next) {
2198 struct dn_pipe *pipe_bp = (struct dn_pipe *)bp;
2201 * Copy pipe descriptor into *bp, convert delay back to ms,
2202 * then copy the flow_set descriptor(s) one at a time.
2203 * After each flow_set, copy the queue descriptor it owns.
2205 bcopy(pipe, bp, sizeof(*pipe));
2206 pipe_bp->delay = (pipe_bp->delay * 1000) / hz;
2208 * XXX the following is a hack based on ->next being the
2209 * first field in dn_pipe and dn_flow_set. The correct
2210 * solution would be to move the dn_flow_set to the beginning
2211 * of struct dn_pipe.
2213 pipe_bp->next.sle_next = (struct dn_pipe *)DN_IS_PIPE;
2214 /* Clean pointers. */
2215 pipe_bp->head = pipe_bp->tail = NULL;
2216 pipe_bp->fs.next.sle_next = NULL;
2217 pipe_bp->fs.pipe = NULL;
2218 pipe_bp->fs.rq = NULL;
2219 pipe_bp->samples = NULL;
2221 bp += sizeof(*pipe) ;
2222 bp = dn_copy_set(&(pipe->fs), bp);
2225 for (i = 0; i < HASHSIZE; i++)
2226 SLIST_FOREACH(fs, &flowsethash[i], next) {
2227 struct dn_flow_set *fs_bp = (struct dn_flow_set *)bp;
2229 bcopy(fs, bp, sizeof(*fs));
2230 /* XXX same hack as above */
2231 fs_bp->next.sle_next = (struct dn_flow_set *)DN_IS_QUEUE;
2235 bp = dn_copy_set(fs, bp);
2240 error = sooptcopyout(sopt, buf, size);
2246 * Handler for the various dummynet socket options (get, flush, config, del)
2249 ip_dn_ctl(struct sockopt *sopt)
2252 struct dn_pipe *p = NULL;
2254 error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET);
2258 /* Disallow sets in really-really secure mode. */
2259 if (sopt->sopt_dir == SOPT_SET) {
2260 #if __FreeBSD_version >= 500034
2261 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2265 if (securelevel >= 3)
2270 switch (sopt->sopt_name) {
2272 printf("dummynet: -- unknown option %d", sopt->sopt_name);
2276 case IP_DUMMYNET_GET :
2277 error = dummynet_get(sopt);
2280 case IP_DUMMYNET_FLUSH :
2284 case IP_DUMMYNET_CONFIGURE :
2285 p = malloc(sizeof(struct dn_pipe_max), M_TEMP, M_WAITOK);
2286 error = sooptcopyin(sopt, p, sizeof(struct dn_pipe_max), sizeof *p);
2289 if (p->samples_no > 0)
2290 p->samples = &( ((struct dn_pipe_max*) p)->samples[0] );
2292 error = config_pipe(p);
2295 case IP_DUMMYNET_DEL : /* remove a pipe or queue */
2296 p = malloc(sizeof(struct dn_pipe_max), M_TEMP, M_WAITOK);
2297 error = sooptcopyin(sopt, p, sizeof *p, sizeof *p);
2301 error = delete_pipe(p);
2317 printf("DUMMYNET with IPv6 initialized (040826)\n");
2319 DUMMYNET_LOCK_INIT();
2321 for (i = 0; i < HASHSIZE; i++) {
2322 SLIST_INIT(&pipehash[i]);
2323 SLIST_INIT(&flowsethash[i]);
2325 ready_heap.size = ready_heap.elements = 0;
2326 ready_heap.offset = 0;
2328 wfq_ready_heap.size = wfq_ready_heap.elements = 0;
2329 wfq_ready_heap.offset = 0;
2331 extract_heap.size = extract_heap.elements = 0;
2332 extract_heap.offset = 0;
2334 ip_dn_ctl_ptr = ip_dn_ctl;
2335 ip_dn_io_ptr = dummynet_io;
2336 ip_dn_ruledel_ptr = dn_rule_delete;
2338 TASK_INIT(&dn_task, 0, dummynet_task, NULL);
2339 dn_tq = taskqueue_create_fast("dummynet", M_NOWAIT,
2340 taskqueue_thread_enqueue, &dn_tq);
2341 taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet");
2343 callout_init(&dn_timeout, CALLOUT_MPSAFE);
2344 callout_reset(&dn_timeout, 1, dummynet, NULL);
2346 /* Initialize curr_time adjustment mechanics. */
2347 getmicrouptime(&prev_t);
2354 ip_dn_ctl_ptr = NULL;
2355 ip_dn_io_ptr = NULL;
2356 ip_dn_ruledel_ptr = NULL;
2359 callout_stop(&dn_timeout);
2361 taskqueue_drain(dn_tq, &dn_task);
2362 taskqueue_free(dn_tq);
2366 DUMMYNET_LOCK_DESTROY();
2368 #endif /* KLD_MODULE */
2371 dummynet_modevent(module_t mod, int type, void *data)
2377 printf("DUMMYNET already loaded\n");
2384 #if !defined(KLD_MODULE)
2385 printf("dummynet statically compiled, cannot unload\n");
2398 static moduledata_t dummynet_mod = {
2403 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
2404 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
2405 MODULE_VERSION(dummynet, 1);