1 /*-
2 * Copyright (c) 1998-2002 Luigi Rizzo, Universita` di Pisa
3 * Portions Copyright (c) 2000 Akamba Corp.
4 * All rights reserved
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
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.
14 *
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
25 * SUCH DAMAGE.
26 *
27 * $FreeBSD: src/sys/netinet/ip_dummynet.c,v 1.83.2.4 2006/02/27 17:24:13 glebius Exp $
28 */
29
30 #define DUMMYNET_DEBUG
31
32 /*
33 * This module implements IP dummynet, a bandwidth limiter/delay emulator
34 * used in conjunction with the ipfw package.
35 * Description of the data structures used is in ip_dummynet.h
36 * Here you mainly find the following blocks of code:
37 * + variable declarations;
38 * + heap management functions;
39 * + scheduler and dummynet functions;
40 * + configuration and initialization.
41 *
42 * NOTA BENE: critical sections are protected by the "dummynet lock".
43 *
44 * Most important Changes:
45 *
46 * 011004: KLDable
47 * 010124: Fixed WF2Q behaviour
48 * 010122: Fixed spl protection.
49 * 000601: WF2Q support
50 * 000106: large rewrite, use heaps to handle very many pipes.
51 * 980513: initial release
52 *
53 * include files marked with XXX are probably not needed
54 */
55
56 #include <sys/param.h>
57 #include <sys/systm.h>
58 #include <sys/malloc.h>
59 #include <sys/mbuf.h>
60 #include <sys/kernel.h>
61 #include <sys/module.h>
62 #include <sys/proc.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/time.h>
66 #include <sys/sysctl.h>
67 #include <net/if.h>
68 #include <net/route.h>
69 #include <netinet/in.h>
70 #include <netinet/in_systm.h>
71 #include <netinet/in_var.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_fw.h>
74 #include <netinet/ip_dummynet.h>
75 #include <netinet/ip_var.h>
76
77 #include <netinet/if_ether.h> /* for struct arpcom */
78 #include <net/bridge.h>
79
80 /*
81 * We keep a private variable for the simulation time, but we could
82 * probably use an existing one ("softticks" in sys/kern/kern_timeout.c)
83 */
84 static dn_key curr_time = 0 ; /* current simulation time */
85
86 static int dn_hash_size = 64 ; /* default hash size */
87
88 /* statistics on number of queue searches and search steps */
89 static int searches, search_steps ;
90 static int pipe_expire = 1 ; /* expire queue if empty */
91 static int dn_max_ratio = 16 ; /* max queues/buckets ratio */
92
93 static int red_lookup_depth = 256; /* RED - default lookup table depth */
94 static int red_avg_pkt_size = 512; /* RED - default medium packet size */
95 static int red_max_pkt_size = 1500; /* RED - default max packet size */
96
97 /*
98 * Three heaps contain queues and pipes that the scheduler handles:
99 *
100 * ready_heap contains all dn_flow_queue related to fixed-rate pipes.
101 *
102 * wfq_ready_heap contains the pipes associated with WF2Q flows
103 *
104 * extract_heap contains pipes associated with delay lines.
105 *
106 */
107
108 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap");
109
110 static struct dn_heap ready_heap, extract_heap, wfq_ready_heap ;
111
112 static int heap_init(struct dn_heap *h, int size) ;
113 static int heap_insert (struct dn_heap *h, dn_key key1, void *p);
114 static void heap_extract(struct dn_heap *h, void *obj);
115
116 static void transmit_event(struct dn_pipe *pipe);
117 static void ready_event(struct dn_flow_queue *q);
118
119 static struct dn_pipe *all_pipes = NULL ; /* list of all pipes */
120 static struct dn_flow_set *all_flow_sets = NULL ;/* list of all flow_sets */
121
122 static struct callout dn_timeout;
123
124 extern void (*bridge_dn_p)(struct mbuf *, struct ifnet *);
125
126 #ifdef SYSCTL_NODE
127 SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet,
128 CTLFLAG_RW, 0, "Dummynet");
129 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, hash_size,
130 CTLFLAG_RW, &dn_hash_size, 0, "Default hash table size");
131 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, curr_time,
132 CTLFLAG_RD, &curr_time, 0, "Current tick");
133 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, ready_heap,
134 CTLFLAG_RD, &ready_heap.size, 0, "Size of ready heap");
135 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, extract_heap,
136 CTLFLAG_RD, &extract_heap.size, 0, "Size of extract heap");
137 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, searches,
138 CTLFLAG_RD, &searches, 0, "Number of queue searches");
139 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, search_steps,
140 CTLFLAG_RD, &search_steps, 0, "Number of queue search steps");
141 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, expire,
142 CTLFLAG_RW, &pipe_expire, 0, "Expire queue if empty");
143 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, max_chain_len,
144 CTLFLAG_RW, &dn_max_ratio, 0,
145 "Max ratio between dynamic queues and buckets");
146 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth,
147 CTLFLAG_RD, &red_lookup_depth, 0, "Depth of RED lookup table");
148 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size,
149 CTLFLAG_RD, &red_avg_pkt_size, 0, "RED Medium packet size");
150 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size,
151 CTLFLAG_RD, &red_max_pkt_size, 0, "RED Max packet size");
152 #endif
153
154 #ifdef DUMMYNET_DEBUG
155 int dummynet_debug = 0;
156 #ifdef SYSCTL_NODE
157 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, debug, CTLFLAG_RW, &dummynet_debug,
158 0, "control debugging printfs");
159 #endif
160 #define DPRINTF(X) if (dummynet_debug) printf X
161 #else
162 #define DPRINTF(X)
163 #endif
164
165 static struct mtx dummynet_mtx;
166 /*
167 * NB: Recursion is needed to deal with re-entry via ICMP. That is,
168 * a packet may be dispatched via ip_input from dummynet_io and
169 * re-enter through ip_output. Yech.
170 */
171 #define DUMMYNET_LOCK_INIT() \
172 mtx_init(&dummynet_mtx, "dummynet", NULL, MTX_DEF | MTX_RECURSE)
173 #define DUMMYNET_LOCK_DESTROY() mtx_destroy(&dummynet_mtx)
174 #define DUMMYNET_LOCK() mtx_lock(&dummynet_mtx)
175 #define DUMMYNET_UNLOCK() mtx_unlock(&dummynet_mtx)
176 #define DUMMYNET_LOCK_ASSERT() do { \
177 mtx_assert(&dummynet_mtx, MA_OWNED); \
178 NET_ASSERT_GIANT(); \
179 } while (0)
180
181 static int config_pipe(struct dn_pipe *p);
182 static int ip_dn_ctl(struct sockopt *sopt);
183
184 static void dummynet(void *);
185 static void dummynet_flush(void);
186 void dummynet_drain(void);
187 static ip_dn_io_t dummynet_io;
188 static void dn_rule_delete(void *);
189
190 int if_tx_rdy(struct ifnet *ifp);
191
192 /*
193 * Heap management functions.
194 *
195 * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2.
196 * Some macros help finding parent/children so we can optimize them.
197 *
198 * heap_init() is called to expand the heap when needed.
199 * Increment size in blocks of 16 entries.
200 * XXX failure to allocate a new element is a pretty bad failure
201 * as we basically stall a whole queue forever!!
202 * Returns 1 on error, 0 on success
203 */
204 #define HEAP_FATHER(x) ( ( (x) - 1 ) / 2 )
205 #define HEAP_LEFT(x) ( 2*(x) + 1 )
206 #define HEAP_IS_LEFT(x) ( (x) & 1 )
207 #define HEAP_RIGHT(x) ( 2*(x) + 2 )
208 #define HEAP_SWAP(a, b, buffer) { buffer = a ; a = b ; b = buffer ; }
209 #define HEAP_INCREMENT 15
210
211 static int
212 heap_init(struct dn_heap *h, int new_size)
213 {
214 struct dn_heap_entry *p;
215
216 if (h->size >= new_size ) {
217 printf("dummynet: %s, Bogus call, have %d want %d\n", __func__,
218 h->size, new_size);
219 return 0 ;
220 }
221 new_size = (new_size + HEAP_INCREMENT ) & ~HEAP_INCREMENT ;
222 p = malloc(new_size * sizeof(*p), M_DUMMYNET, M_NOWAIT);
223 if (p == NULL) {
224 printf("dummynet: %s, resize %d failed\n", __func__, new_size );
225 return 1 ; /* error */
226 }
227 if (h->size > 0) {
228 bcopy(h->p, p, h->size * sizeof(*p) );
229 free(h->p, M_DUMMYNET);
230 }
231 h->p = p ;
232 h->size = new_size ;
233 return 0 ;
234 }
235
236 /*
237 * Insert element in heap. Normally, p != NULL, we insert p in
238 * a new position and bubble up. If p == NULL, then the element is
239 * already in place, and key is the position where to start the
240 * bubble-up.
241 * Returns 1 on failure (cannot allocate new heap entry)
242 *
243 * If offset > 0 the position (index, int) of the element in the heap is
244 * also stored in the element itself at the given offset in bytes.
245 */
246 #define SET_OFFSET(heap, node) \
247 if (heap->offset > 0) \
248 *((int *)((char *)(heap->p[node].object) + heap->offset)) = node ;
249 /*
250 * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value.
251 */
252 #define RESET_OFFSET(heap, node) \
253 if (heap->offset > 0) \
254 *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1 ;
255 static int
256 heap_insert(struct dn_heap *h, dn_key key1, void *p)
257 {
258 int son = h->elements ;
259
260 if (p == NULL) /* data already there, set starting point */
261 son = key1 ;
262 else { /* insert new element at the end, possibly resize */
263 son = h->elements ;
264 if (son == h->size) /* need resize... */
265 if (heap_init(h, h->elements+1) )
266 return 1 ; /* failure... */
267 h->p[son].object = p ;
268 h->p[son].key = key1 ;
269 h->elements++ ;
270 }
271 while (son > 0) { /* bubble up */
272 int father = HEAP_FATHER(son) ;
273 struct dn_heap_entry tmp ;
274
275 if (DN_KEY_LT( h->p[father].key, h->p[son].key ) )
276 break ; /* found right position */
277 /* son smaller than father, swap and repeat */
278 HEAP_SWAP(h->p[son], h->p[father], tmp) ;
279 SET_OFFSET(h, son);
280 son = father ;
281 }
282 SET_OFFSET(h, son);
283 return 0 ;
284 }
285
286 /*
287 * remove top element from heap, or obj if obj != NULL
288 */
289 static void
290 heap_extract(struct dn_heap *h, void *obj)
291 {
292 int child, father, max = h->elements - 1 ;
293
294 if (max < 0) {
295 printf("dummynet: warning, extract from empty heap 0x%p\n", h);
296 return ;
297 }
298 father = 0 ; /* default: move up smallest child */
299 if (obj != NULL) { /* extract specific element, index is at offset */
300 if (h->offset <= 0)
301 panic("dummynet: heap_extract from middle not supported on this heap!!!\n");
302 father = *((int *)((char *)obj + h->offset)) ;
303 if (father < 0 || father >= h->elements) {
304 printf("dummynet: heap_extract, father %d out of bound 0..%d\n",
305 father, h->elements);
306 panic("dummynet: heap_extract");
307 }
308 }
309 RESET_OFFSET(h, father);
310 child = HEAP_LEFT(father) ; /* left child */
311 while (child <= max) { /* valid entry */
312 if (child != max && DN_KEY_LT(h->p[child+1].key, h->p[child].key) )
313 child = child+1 ; /* take right child, otherwise left */
314 h->p[father] = h->p[child] ;
315 SET_OFFSET(h, father);
316 father = child ;
317 child = HEAP_LEFT(child) ; /* left child for next loop */
318 }
319 h->elements-- ;
320 if (father != max) {
321 /*
322 * Fill hole with last entry and bubble up, reusing the insert code
323 */
324 h->p[father] = h->p[max] ;
325 heap_insert(h, father, NULL); /* this one cannot fail */
326 }
327 }
328
329 #if 0
330 /*
331 * change object position and update references
332 * XXX this one is never used!
333 */
334 static void
335 heap_move(struct dn_heap *h, dn_key new_key, void *object)
336 {
337 int temp;
338 int i ;
339 int max = h->elements-1 ;
340 struct dn_heap_entry buf ;
341
342 if (h->offset <= 0)
343 panic("cannot move items on this heap");
344
345 i = *((int *)((char *)object + h->offset));
346 if (DN_KEY_LT(new_key, h->p[i].key) ) { /* must move up */
347 h->p[i].key = new_key ;
348 for (; i>0 && DN_KEY_LT(new_key, h->p[(temp = HEAP_FATHER(i))].key) ;
349 i = temp ) { /* bubble up */
350 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
351 SET_OFFSET(h, i);
352 }
353 } else { /* must move down */
354 h->p[i].key = new_key ;
355 while ( (temp = HEAP_LEFT(i)) <= max ) { /* found left child */
356 if ((temp != max) && DN_KEY_GT(h->p[temp].key, h->p[temp+1].key))
357 temp++ ; /* select child with min key */
358 if (DN_KEY_GT(new_key, h->p[temp].key)) { /* go down */
359 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
360 SET_OFFSET(h, i);
361 } else
362 break ;
363 i = temp ;
364 }
365 }
366 SET_OFFSET(h, i);
367 }
368 #endif /* heap_move, unused */
369
370 /*
371 * heapify() will reorganize data inside an array to maintain the
372 * heap property. It is needed when we delete a bunch of entries.
373 */
374 static void
375 heapify(struct dn_heap *h)
376 {
377 int i ;
378
379 for (i = 0 ; i < h->elements ; i++ )
380 heap_insert(h, i , NULL) ;
381 }
382
383 /*
384 * cleanup the heap and free data structure
385 */
386 static void
387 heap_free(struct dn_heap *h)
388 {
389 if (h->size >0 )
390 free(h->p, M_DUMMYNET);
391 bzero(h, sizeof(*h) );
392 }
393
394 /*
395 * --- end of heap management functions ---
396 */
397
398 /*
399 * Return the mbuf tag holding the dummynet state. As an optimization
400 * this is assumed to be the first tag on the list. If this turns out
401 * wrong we'll need to search the list.
402 */
403 static struct dn_pkt_tag *
404 dn_tag_get(struct mbuf *m)
405 {
406 struct m_tag *mtag = m_tag_first(m);
407 KASSERT(mtag != NULL &&
408 mtag->m_tag_cookie == MTAG_ABI_COMPAT &&
409 mtag->m_tag_id == PACKET_TAG_DUMMYNET,
410 ("packet on dummynet queue w/o dummynet tag!"));
411 return (struct dn_pkt_tag *)(mtag+1);
412 }
413
414 /*
415 * Scheduler functions:
416 *
417 * transmit_event() is called when the delay-line needs to enter
418 * the scheduler, either because of existing pkts getting ready,
419 * or new packets entering the queue. The event handled is the delivery
420 * time of the packet.
421 *
422 * ready_event() does something similar with fixed-rate queues, and the
423 * event handled is the finish time of the head pkt.
424 *
425 * wfq_ready_event() does something similar with WF2Q queues, and the
426 * event handled is the start time of the head pkt.
427 *
428 * In all cases, we make sure that the data structures are consistent
429 * before passing pkts out, because this might trigger recursive
430 * invocations of the procedures.
431 */
432 static void
433 transmit_event(struct dn_pipe *pipe)
434 {
435 struct mbuf *m ;
436 struct dn_pkt_tag *pkt ;
437 struct ip *ip;
438
439 DUMMYNET_LOCK_ASSERT();
440
441 while ( (m = pipe->head) ) {
442 pkt = dn_tag_get(m);
443 if ( !DN_KEY_LEQ(pkt->output_time, curr_time) )
444 break;
445 /*
446 * first unlink, then call procedures, since ip_input() can invoke
447 * ip_output() and viceversa, thus causing nested calls
448 */
449 pipe->head = m->m_nextpkt ;
450 m->m_nextpkt = NULL;
451
452 /* XXX: drop the lock for now to avoid LOR's */
453 DUMMYNET_UNLOCK();
454 switch (pkt->dn_dir) {
455 case DN_TO_IP_OUT:
456 (void)ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
457 break ;
458
459 case DN_TO_IP_IN :
460 ip = mtod(m, struct ip *);
461 ip->ip_len = htons(ip->ip_len);
462 ip->ip_off = htons(ip->ip_off);
463 ip_input(m) ;
464 break ;
465
466 case DN_TO_BDG_FWD :
467 /*
468 * The bridge requires/assumes the Ethernet header is
469 * contiguous in the first mbuf header. Insure this is true.
470 */
471 if (BDG_LOADED) {
472 if (m->m_len < ETHER_HDR_LEN &&
473 (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) {
474 printf("dummynet/bridge: pullup fail, dropping pkt\n");
475 break;
476 }
477 m = bdg_forward_ptr(m, pkt->ifp);
478 } else {
479 /* somebody unloaded the bridge module. Drop pkt */
480 /* XXX rate limit */
481 printf("dummynet: dropping bridged packet trapped in pipe\n");
482 }
483 if (m)
484 m_freem(m);
485 break;
486
487 case DN_TO_ETH_DEMUX:
488 /*
489 * The Ethernet code assumes the Ethernet header is
490 * contiguous in the first mbuf header. Insure this is true.
491 */
492 if (m->m_len < ETHER_HDR_LEN &&
493 (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) {
494 printf("dummynet/ether: pullup fail, dropping pkt\n");
495 break;
496 }
497 ether_demux(m->m_pkthdr.rcvif, m); /* which consumes the mbuf */
498 break ;
499
500 case DN_TO_ETH_OUT:
501 ether_output_frame(pkt->ifp, m);
502 break;
503
504 case DN_TO_IFB_FWD:
505 if (bridge_dn_p != NULL)
506 ((*bridge_dn_p)(m, pkt->ifp));
507 else
508 printf("dummynet: if_bridge not loaded\n");
509
510 break;
511
512 default:
513 printf("dummynet: bad switch %d!\n", pkt->dn_dir);
514 m_freem(m);
515 break ;
516 }
517 DUMMYNET_LOCK();
518 }
519 /* if there are leftover packets, put into the heap for next event */
520 if ( (m = pipe->head) ) {
521 pkt = dn_tag_get(m) ;
522 /* XXX should check errors on heap_insert, by draining the
523 * whole pipe p and hoping in the future we are more successful
524 */
525 heap_insert(&extract_heap, pkt->output_time, pipe ) ;
526 }
527 }
528
529 /*
530 * the following macro computes how many ticks we have to wait
531 * before being able to transmit a packet. The credit is taken from
532 * either a pipe (WF2Q) or a flow_queue (per-flow queueing)
533 */
534 #define SET_TICKS(_m, q, p) \
535 ((_m)->m_pkthdr.len*8*hz - (q)->numbytes + p->bandwidth - 1 ) / \
536 p->bandwidth ;
537
538 /*
539 * extract pkt from queue, compute output time (could be now)
540 * and put into delay line (p_queue)
541 */
542 static void
543 move_pkt(struct mbuf *pkt, struct dn_flow_queue *q,
544 struct dn_pipe *p, int len)
545 {
546 struct dn_pkt_tag *dt = dn_tag_get(pkt);
547
548 q->head = pkt->m_nextpkt ;
549 q->len-- ;
550 q->len_bytes -= len ;
551
552 dt->output_time = curr_time + p->delay ;
553
554 if (p->head == NULL)
555 p->head = pkt;
556 else
557 p->tail->m_nextpkt = pkt;
558 p->tail = pkt;
559 p->tail->m_nextpkt = NULL;
560 }
561
562 /*
563 * ready_event() is invoked every time the queue must enter the
564 * scheduler, either because the first packet arrives, or because
565 * a previously scheduled event fired.
566 * On invokation, drain as many pkts as possible (could be 0) and then
567 * if there are leftover packets reinsert the pkt in the scheduler.
568 */
569 static void
570 ready_event(struct dn_flow_queue *q)
571 {
572 struct mbuf *pkt;
573 struct dn_pipe *p = q->fs->pipe ;
574 int p_was_empty ;
575
576 DUMMYNET_LOCK_ASSERT();
577
578 if (p == NULL) {
579 printf("dummynet: ready_event- pipe is gone\n");
580 return ;
581 }
582 p_was_empty = (p->head == NULL) ;
583
584 /*
585 * schedule fixed-rate queues linked to this pipe:
586 * Account for the bw accumulated since last scheduling, then
587 * drain as many pkts as allowed by q->numbytes and move to
588 * the delay line (in p) computing output time.
589 * bandwidth==0 (no limit) means we can drain the whole queue,
590 * setting len_scaled = 0 does the job.
591 */
592 q->numbytes += ( curr_time - q->sched_time ) * p->bandwidth;
593 while ( (pkt = q->head) != NULL ) {
594 int len = pkt->m_pkthdr.len;
595 int len_scaled = p->bandwidth ? len*8*hz : 0 ;
596 if (len_scaled > q->numbytes )
597 break ;
598 q->numbytes -= len_scaled ;
599 move_pkt(pkt, q, p, len);
600 }
601 /*
602 * If we have more packets queued, schedule next ready event
603 * (can only occur when bandwidth != 0, otherwise we would have
604 * flushed the whole queue in the previous loop).
605 * To this purpose we record the current time and compute how many
606 * ticks to go for the finish time of the packet.
607 */
608 if ( (pkt = q->head) != NULL ) { /* this implies bandwidth != 0 */
609 dn_key t = SET_TICKS(pkt, q, p); /* ticks i have to wait */
610 q->sched_time = curr_time ;
611 heap_insert(&ready_heap, curr_time + t, (void *)q );
612 /* XXX should check errors on heap_insert, and drain the whole
613 * queue on error hoping next time we are luckier.
614 */
615 } else { /* RED needs to know when the queue becomes empty */
616 q->q_time = curr_time;
617 q->numbytes = 0;
618 }
619 /*
620 * If the delay line was empty call transmit_event(p) now.
621 * Otherwise, the scheduler will take care of it.
622 */
623 if (p_was_empty)
624 transmit_event(p);
625 }
626
627 /*
628 * Called when we can transmit packets on WF2Q queues. Take pkts out of
629 * the queues at their start time, and enqueue into the delay line.
630 * Packets are drained until p->numbytes < 0. As long as
631 * len_scaled >= p->numbytes, the packet goes into the delay line
632 * with a deadline p->delay. For the last packet, if p->numbytes<0,
633 * there is an additional delay.
634 */
635 static void
636 ready_event_wfq(struct dn_pipe *p)
637 {
638 int p_was_empty = (p->head == NULL) ;
639 struct dn_heap *sch = &(p->scheduler_heap);
640 struct dn_heap *neh = &(p->not_eligible_heap) ;
641
642 DUMMYNET_LOCK_ASSERT();
643
644 if (p->if_name[0] == 0) /* tx clock is simulated */
645 p->numbytes += ( curr_time - p->sched_time ) * p->bandwidth;
646 else { /* tx clock is for real, the ifq must be empty or this is a NOP */
647 if (p->ifp && p->ifp->if_snd.ifq_head != NULL)
648 return ;
649 else {
650 DPRINTF(("dummynet: pipe %d ready from %s --\n",
651 p->pipe_nr, p->if_name));
652 }
653 }
654
655 /*
656 * While we have backlogged traffic AND credit, we need to do
657 * something on the queue.
658 */
659 while ( p->numbytes >=0 && (sch->elements>0 || neh->elements >0) ) {
660 if (sch->elements > 0) { /* have some eligible pkts to send out */
661 struct dn_flow_queue *q = sch->p[0].object ;
662 struct mbuf *pkt = q->head;
663 struct dn_flow_set *fs = q->fs;
664 u_int64_t len = pkt->m_pkthdr.len;
665 int len_scaled = p->bandwidth ? len*8*hz : 0 ;
666
667 heap_extract(sch, NULL); /* remove queue from heap */
668 p->numbytes -= len_scaled ;
669 move_pkt(pkt, q, p, len);
670
671 p->V += (len<<MY_M) / p->sum ; /* update V */
672 q->S = q->F ; /* update start time */
673 if (q->len == 0) { /* Flow not backlogged any more */
674 fs->backlogged-- ;
675 heap_insert(&(p->idle_heap), q->F, q);
676 } else { /* still backlogged */
677 /*
678 * update F and position in backlogged queue, then
679 * put flow in not_eligible_heap (we will fix this later).
680 */
681 len = (q->head)->m_pkthdr.len;
682 q->F += (len<<MY_M)/(u_int64_t) fs->weight ;
683 if (DN_KEY_LEQ(q->S, p->V))
684 heap_insert(neh, q->S, q);
685 else
686 heap_insert(sch, q->F, q);
687 }
688 }
689 /*
690 * now compute V = max(V, min(S_i)). Remember that all elements in sch
691 * have by definition S_i <= V so if sch is not empty, V is surely
692 * the max and we must not update it. Conversely, if sch is empty
693 * we only need to look at neh.
694 */
695 if (sch->elements == 0 && neh->elements > 0)
696 p->V = MAX64 ( p->V, neh->p[0].key );
697 /* move from neh to sch any packets that have become eligible */
698 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V) ) {
699 struct dn_flow_queue *q = neh->p[0].object ;
700 heap_extract(neh, NULL);
701 heap_insert(sch, q->F, q);
702 }
703
704 if (p->if_name[0] != '\0') {/* tx clock is from a real thing */
705 p->numbytes = -1 ; /* mark not ready for I/O */
706 break ;
707 }
708 }
709 if (sch->elements == 0 && neh->elements == 0 && p->numbytes >= 0
710 && p->idle_heap.elements > 0) {
711 /*
712 * no traffic and no events scheduled. We can get rid of idle-heap.
713 */
714 int i ;
715
716 for (i = 0 ; i < p->idle_heap.elements ; i++) {
717 struct dn_flow_queue *q = p->idle_heap.p[i].object ;
718
719 q->F = 0 ;
720 q->S = q->F + 1 ;
721 }
722 p->sum = 0 ;
723 p->V = 0 ;
724 p->idle_heap.elements = 0 ;
725 }
726 /*
727 * If we are getting clocks from dummynet (not a real interface) and
728 * If we are under credit, schedule the next ready event.
729 * Also fix the delivery time of the last packet.
730 */
731 if (p->if_name[0]==0 && p->numbytes < 0) { /* this implies bandwidth >0 */
732 dn_key t=0 ; /* number of ticks i have to wait */
733
734 if (p->bandwidth > 0)
735 t = ( p->bandwidth -1 - p->numbytes) / p->bandwidth ;
736 dn_tag_get(p->tail)->output_time += t ;
737 p->sched_time = curr_time ;
738 heap_insert(&wfq_ready_heap, curr_time + t, (void *)p);
739 /* XXX should check errors on heap_insert, and drain the whole
740 * queue on error hoping next time we are luckier.
741 */
742 }
743 /*
744 * If the delay line was empty call transmit_event(p) now.
745 * Otherwise, the scheduler will take care of it.
746 */
747 if (p_was_empty)
748 transmit_event(p);
749 }
750
751 /*
752 * This is called once per tick, or HZ times per second. It is used to
753 * increment the current tick counter and schedule expired events.
754 */
755 static void
756 dummynet(void * __unused unused)
757 {
758 void *p ; /* generic parameter to handler */
759 struct dn_heap *h ;
760 struct dn_heap *heaps[3];
761 int i;
762 struct dn_pipe *pe ;
763
764 heaps[0] = &ready_heap ; /* fixed-rate queues */
765 heaps[1] = &wfq_ready_heap ; /* wfq queues */
766 heaps[2] = &extract_heap ; /* delay line */
767
768 DUMMYNET_LOCK();
769 curr_time++ ;
770 for (i=0; i < 3 ; i++) {
771 h = heaps[i];
772 while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time) ) {
773 if (h->p[0].key > curr_time)
774 printf("dummynet: warning, heap %d is %d ticks late\n",
775 i, (int)(curr_time - h->p[0].key));
776 p = h->p[0].object ; /* store a copy before heap_extract */
777 heap_extract(h, NULL); /* need to extract before processing */
778 if (i == 0)
779 ready_event(p) ;
780 else if (i == 1) {
781 struct dn_pipe *pipe = p;
782 if (pipe->if_name[0] != '\0')
783 printf("dummynet: bad ready_event_wfq for pipe %s\n",
784 pipe->if_name);
785 else
786 ready_event_wfq(p) ;
787 } else
788 transmit_event(p);
789 }
790 }
791 /* sweep pipes trying to expire idle flow_queues */
792 for (pe = all_pipes; pe ; pe = pe->next )
793 if (pe->idle_heap.elements > 0 &&
794 DN_KEY_LT(pe->idle_heap.p[0].key, pe->V) ) {
795 struct dn_flow_queue *q = pe->idle_heap.p[0].object ;
796
797 heap_extract(&(pe->idle_heap), NULL);
798 q->S = q->F + 1 ; /* mark timestamp as invalid */
799 pe->sum -= q->fs->weight ;
800 }
801 DUMMYNET_UNLOCK();
802
803 callout_reset(&dn_timeout, 1, dummynet, NULL);
804 }
805
806 /*
807 * called by an interface when tx_rdy occurs.
808 */
809 int
810 if_tx_rdy(struct ifnet *ifp)
811 {
812 struct dn_pipe *p;
813
814 DUMMYNET_LOCK();
815 for (p = all_pipes; p ; p = p->next )
816 if (p->ifp == ifp)
817 break ;
818 if (p == NULL) {
819 for (p = all_pipes; p ; p = p->next )
820 if (!strcmp(p->if_name, ifp->if_xname) ) {
821 p->ifp = ifp ;
822 DPRINTF(("dummynet: ++ tx rdy from %s (now found)\n",
823 ifp->if_xname));
824 break ;
825 }
826 }
827 if (p != NULL) {
828 DPRINTF(("dummynet: ++ tx rdy from %s - qlen %d\n", ifp->if_xname,
829 ifp->if_snd.ifq_len));
830 p->numbytes = 0 ; /* mark ready for I/O */
831 ready_event_wfq(p);
832 }
833 DUMMYNET_UNLOCK();
834
835 return 0;
836 }
837
838 /*
839 * Unconditionally expire empty queues in case of shortage.
840 * Returns the number of queues freed.
841 */
842 static int
843 expire_queues(struct dn_flow_set *fs)
844 {
845 struct dn_flow_queue *q, *prev ;
846 int i, initial_elements = fs->rq_elements ;
847
848 if (fs->last_expired == time_second)
849 return 0 ;
850 fs->last_expired = time_second ;
851 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is overflow */
852 for (prev=NULL, q = fs->rq[i] ; q != NULL ; )
853 if (q->head != NULL || q->S != q->F+1) {
854 prev = q ;
855 q = q->next ;
856 } else { /* entry is idle, expire it */
857 struct dn_flow_queue *old_q = q ;
858
859 if (prev != NULL)
860 prev->next = q = q->next ;
861 else
862 fs->rq[i] = q = q->next ;
863 fs->rq_elements-- ;
864 free(old_q, M_DUMMYNET);
865 }
866 return initial_elements - fs->rq_elements ;
867 }
868
869 /*
870 * If room, create a new queue and put at head of slot i;
871 * otherwise, create or use the default queue.
872 */
873 static struct dn_flow_queue *
874 create_queue(struct dn_flow_set *fs, int i)
875 {
876 struct dn_flow_queue *q ;
877
878 if (fs->rq_elements > fs->rq_size * dn_max_ratio &&
879 expire_queues(fs) == 0) {
880 /*
881 * No way to get room, use or create overflow queue.
882 */
883 i = fs->rq_size ;
884 if ( fs->rq[i] != NULL )
885 return fs->rq[i] ;
886 }
887 q = malloc(sizeof(*q), M_DUMMYNET, M_NOWAIT | M_ZERO);
888 if (q == NULL) {
889 printf("dummynet: sorry, cannot allocate queue for new flow\n");
890 return NULL ;
891 }
892 q->fs = fs ;
893 q->hash_slot = i ;
894 q->next = fs->rq[i] ;
895 q->S = q->F + 1; /* hack - mark timestamp as invalid */
896 fs->rq[i] = q ;
897 fs->rq_elements++ ;
898 return q ;
899 }
900
901 /*
902 * Given a flow_set and a pkt in last_pkt, find a matching queue
903 * after appropriate masking. The queue is moved to front
904 * so that further searches take less time.
905 */
906 static struct dn_flow_queue *
907 find_queue(struct dn_flow_set *fs, struct ipfw_flow_id *id)
908 {
909 int i = 0 ; /* we need i and q for new allocations */
910 struct dn_flow_queue *q, *prev;
911
912 if ( !(fs->flags_fs & DN_HAVE_FLOW_MASK) )
913 q = fs->rq[0] ;
914 else {
915 /* first, do the masking */
916 id->dst_ip &= fs->flow_mask.dst_ip ;
917 id->src_ip &= fs->flow_mask.src_ip ;
918 id->dst_port &= fs->flow_mask.dst_port ;
919 id->src_port &= fs->flow_mask.src_port ;
920 id->proto &= fs->flow_mask.proto ;
921 id->flags = 0 ; /* we don't care about this one */
922 /* then, hash function */
923 i = ( (id->dst_ip) & 0xffff ) ^
924 ( (id->dst_ip >> 15) & 0xffff ) ^
925 ( (id->src_ip << 1) & 0xffff ) ^
926 ( (id->src_ip >> 16 ) & 0xffff ) ^
927 (id->dst_port << 1) ^ (id->src_port) ^
928 (id->proto );
929 i = i % fs->rq_size ;
930 /* finally, scan the current list for a match */
931 searches++ ;
932 for (prev=NULL, q = fs->rq[i] ; q ; ) {
933 search_steps++;
934 if (id->dst_ip == q->id.dst_ip &&
935 id->src_ip == q->id.src_ip &&
936 id->dst_port == q->id.dst_port &&
937 id->src_port == q->id.src_port &&
938 id->proto == q->id.proto &&
939 id->flags == q->id.flags)
940 break ; /* found */
941 else if (pipe_expire && q->head == NULL && q->S == q->F+1 ) {
942 /* entry is idle and not in any heap, expire it */
943 struct dn_flow_queue *old_q = q ;
944
945 if (prev != NULL)
946 prev->next = q = q->next ;
947 else
948 fs->rq[i] = q = q->next ;
949 fs->rq_elements-- ;
950 free(old_q, M_DUMMYNET);
951 continue ;
952 }
953 prev = q ;
954 q = q->next ;
955 }
956 if (q && prev != NULL) { /* found and not in front */
957 prev->next = q->next ;
958 q->next = fs->rq[i] ;
959 fs->rq[i] = q ;
960 }
961 }
962 if (q == NULL) { /* no match, need to allocate a new entry */
963 q = create_queue(fs, i);
964 if (q != NULL)
965 q->id = *id ;
966 }
967 return q ;
968 }
969
970 static int
971 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len)
972 {
973 /*
974 * RED algorithm
975 *
976 * RED calculates the average queue size (avg) using a low-pass filter
977 * with an exponential weighted (w_q) moving average:
978 * avg <- (1-w_q) * avg + w_q * q_size
979 * where q_size is the queue length (measured in bytes or * packets).
980 *
981 * If q_size == 0, we compute the idle time for the link, and set
982 * avg = (1 - w_q)^(idle/s)
983 * where s is the time needed for transmitting a medium-sized packet.
984 *
985 * Now, if avg < min_th the packet is enqueued.
986 * If avg > max_th the packet is dropped. Otherwise, the packet is
987 * dropped with probability P function of avg.
988 *
989 */
990
991 int64_t p_b = 0;
992 /* queue in bytes or packets ? */
993 u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ? q->len_bytes : q->len;
994
995 DPRINTF(("\ndummynet: %d q: %2u ", (int) curr_time, q_size));
996
997 /* average queue size estimation */
998 if (q_size != 0) {
999 /*
1000 * queue is not empty, avg <- avg + (q_size - avg) * w_q
1001 */
1002 int diff = SCALE(q_size) - q->avg;
1003 int64_t v = SCALE_MUL((int64_t) diff, (int64_t) fs->w_q);
1004
1005 q->avg += (int) v;
1006 } else {
1007 /*
1008 * queue is empty, find for how long the queue has been
1009 * empty and use a lookup table for computing
1010 * (1 - * w_q)^(idle_time/s) where s is the time to send a
1011 * (small) packet.
1012 * XXX check wraps...
1013 */
1014 if (q->avg) {
1015 u_int t = (curr_time - q->q_time) / fs->lookup_step;
1016
1017 q->avg = (t < fs->lookup_depth) ?
1018 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0;
1019 }
1020 }
1021 DPRINTF(("dummynet: avg: %u ", SCALE_VAL(q->avg)));
1022
1023 /* should i drop ? */
1024
1025 if (q->avg < fs->min_th) {
1026 q->count = -1;
1027 return 0; /* accept packet ; */
1028 }
1029 if (q->avg >= fs->max_th) { /* average queue >= max threshold */
1030 if (fs->flags_fs & DN_IS_GENTLE_RED) {
1031 /*
1032 * According to Gentle-RED, if avg is greater than max_th the
1033 * packet is dropped with a probability
1034 * p_b = c_3 * avg - c_4
1035 * where c_3 = (1 - max_p) / max_th, and c_4 = 1 - 2 * max_p
1036 */
1037 p_b = SCALE_MUL((int64_t) fs->c_3, (int64_t) q->avg) - fs->c_4;
1038 } else {
1039 q->count = -1;
1040 DPRINTF(("dummynet: - drop"));
1041 return 1 ;
1042 }
1043 } else if (q->avg > fs->min_th) {
1044 /*
1045 * we compute p_b using the linear dropping function p_b = c_1 *
1046 * avg - c_2, where c_1 = max_p / (max_th - min_th), and c_2 =
1047 * max_p * min_th / (max_th - min_th)
1048 */
1049 p_b = SCALE_MUL((int64_t) fs->c_1, (int64_t) q->avg) - fs->c_2;
1050 }
1051 if (fs->flags_fs & DN_QSIZE_IS_BYTES)
1052 p_b = (p_b * len) / fs->max_pkt_size;
1053 if (++q->count == 0)
1054 q->random = random() & 0xffff;
1055 else {
1056 /*
1057 * q->count counts packets arrived since last drop, so a greater
1058 * value of q->count means a greater packet drop probability.
1059 */
1060 if (SCALE_MUL(p_b, SCALE((int64_t) q->count)) > q->random) {
1061 q->count = 0;
1062 DPRINTF(("dummynet: - red drop"));
1063 /* after a drop we calculate a new random value */
1064 q->random = random() & 0xffff;
1065 return 1; /* drop */
1066 }
1067 }
1068 /* end of RED algorithm */
1069 return 0 ; /* accept */
1070 }
1071
1072 static __inline
1073 struct dn_flow_set *
1074 locate_flowset(int pipe_nr, struct ip_fw *rule)
1075 {
1076 #if IPFW2
1077 struct dn_flow_set *fs;
1078 ipfw_insn *cmd = rule->cmd + rule->act_ofs;
1079
1080 if (cmd->opcode == O_LOG)
1081 cmd += F_LEN(cmd);
1082 #ifdef __i386__
1083 fs = ((ipfw_insn_pipe *)cmd)->pipe_ptr;
1084 #else
1085 bcopy(& ((ipfw_insn_pipe *)cmd)->pipe_ptr, &fs, sizeof(fs));
1086 #endif
1087
1088 if (fs != NULL)
1089 return fs;
1090
1091 if (cmd->opcode == O_QUEUE)
1092 #else /* !IPFW2 */
1093 struct dn_flow_set *fs = NULL ;
1094
1095 if ( (rule->fw_flg & IP_FW_F_COMMAND) == IP_FW_F_QUEUE )
1096 #endif /* !IPFW2 */
1097 for (fs=all_flow_sets; fs && fs->fs_nr != pipe_nr; fs=fs->next)
1098 ;
1099 else {
1100 struct dn_pipe *p1;
1101 for (p1 = all_pipes; p1 && p1->pipe_nr != pipe_nr; p1 = p1->next)
1102 ;
1103 if (p1 != NULL)
1104 fs = &(p1->fs) ;
1105 }
1106 /* record for the future */
1107 #if IPFW2
1108 #ifdef __i386__
1109 ((ipfw_insn_pipe *)cmd)->pipe_ptr = fs;
1110 #else
1111 bcopy(&fs, & ((ipfw_insn_pipe *)cmd)->pipe_ptr, sizeof(fs));
1112 #endif
1113 #else
1114 if (fs != NULL)
1115 rule->pipe_ptr = fs;
1116 #endif
1117 return fs ;
1118 }
1119
1120 /*
1121 * dummynet hook for packets. Below 'pipe' is a pipe or a queue
1122 * depending on whether WF2Q or fixed bw is used.
1123 *
1124 * pipe_nr pipe or queue the packet is destined for.
1125 * dir where shall we send the packet after dummynet.
1126 * m the mbuf with the packet
1127 * ifp the 'ifp' parameter from the caller.
1128 * NULL in ip_input, destination interface in ip_output,
1129 * real_dst in bdg_forward
1130 * rule matching rule, in case of multiple passes
1131 *
1132 */
1133 static int
1134 dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
1135 {
1136 struct dn_pkt_tag *pkt;
1137 struct m_tag *mtag;
1138 struct dn_flow_set *fs;
1139 struct dn_pipe *pipe ;
1140 u_int64_t len = m->m_pkthdr.len ;
1141 struct dn_flow_queue *q = NULL ;
1142 int is_pipe;
1143 #if IPFW2
1144 ipfw_insn *cmd = fwa->rule->cmd + fwa->rule->act_ofs;
1145 #endif
1146
1147 KASSERT(m->m_nextpkt == NULL,
1148 ("dummynet_io: mbuf queue passed to dummynet"));
1149
1150 #if IPFW2
1151 if (cmd->opcode == O_LOG)
1152 cmd += F_LEN(cmd);
1153 is_pipe = (cmd->opcode == O_PIPE);
1154 #else
1155 is_pipe = (fwa->rule->fw_flg & IP_FW_F_COMMAND) == IP_FW_F_PIPE;
1156 #endif
1157
1158 pipe_nr &= 0xffff ;
1159
1160 DUMMYNET_LOCK();
1161 /*
1162 * This is a dummynet rule, so we expect an O_PIPE or O_QUEUE rule.
1163 */
1164 fs = locate_flowset(pipe_nr, fwa->rule);
1165 if (fs == NULL)
1166 goto dropit ; /* this queue/pipe does not exist! */
1167 pipe = fs->pipe ;
1168 if (pipe == NULL) { /* must be a queue, try find a matching pipe */
1169 for (pipe = all_pipes; pipe && pipe->pipe_nr != fs->parent_nr;
1170 pipe = pipe->next)
1171 ;
1172 if (pipe != NULL)
1173 fs->pipe = pipe ;
1174 else {
1175 printf("dummynet: no pipe %d for queue %d, drop pkt\n",
1176 fs->parent_nr, fs->fs_nr);
1177 goto dropit ;
1178 }
1179 }
1180 q = find_queue(fs, &(fwa->f_id));
1181 if ( q == NULL )
1182 goto dropit ; /* cannot allocate queue */
1183 /*
1184 * update statistics, then check reasons to drop pkt
1185 */
1186 q->tot_bytes += len ;
1187 q->tot_pkts++ ;
1188 if ( fs->plr && random() < fs->plr )
1189 goto dropit ; /* random pkt drop */
1190 if ( fs->flags_fs & DN_QSIZE_IS_BYTES) {
1191 if (q->len_bytes > fs->qsize)
1192 goto dropit ; /* queue size overflow */
1193 } else {
1194 if (q->len >= fs->qsize)
1195 goto dropit ; /* queue count overflow */
1196 }
1197 if ( fs->flags_fs & DN_IS_RED && red_drops(fs, q, len) )
1198 goto dropit ;
1199
1200 /* XXX expensive to zero, see if we can remove it*/
1201 mtag = m_tag_get(PACKET_TAG_DUMMYNET,
1202 sizeof(struct dn_pkt_tag), M_NOWAIT|M_ZERO);
1203 if ( mtag == NULL )
1204 goto dropit ; /* cannot allocate packet header */
1205 m_tag_prepend(m, mtag); /* attach to mbuf chain */
1206
1207 pkt = (struct dn_pkt_tag *)(mtag+1);
1208 /* ok, i can handle the pkt now... */
1209 /* build and enqueue packet + parameters */
1210 pkt->rule = fwa->rule ;
1211 pkt->dn_dir = dir ;
1212
1213 pkt->ifp = fwa->oif;
1214 if (q->head == NULL)
1215 q->head = m;
1216 else
1217 q->tail->m_nextpkt = m;
1218 q->tail = m;
1219 q->len++;
1220 q->len_bytes += len ;
1221
1222 if ( q->head != m ) /* flow was not idle, we are done */
1223 goto done;
1224 /*
1225 * If we reach this point the flow was previously idle, so we need
1226 * to schedule it. This involves different actions for fixed-rate or
1227 * WF2Q queues.
1228 */
1229 if (is_pipe) {
1230 /*
1231 * Fixed-rate queue: just insert into the ready_heap.
1232 */
1233 dn_key t = 0 ;
1234 if (pipe->bandwidth)
1235 t = SET_TICKS(m, q, pipe);
1236 q->sched_time = curr_time ;
1237 if (t == 0) /* must process it now */
1238 ready_event( q );
1239 else
1240 heap_insert(&ready_heap, curr_time + t , q );
1241 } else {
1242 /*
1243 * WF2Q. First, compute start time S: if the flow was idle (S=F+1)
1244 * set S to the virtual time V for the controlling pipe, and update
1245 * the sum of weights for the pipe; otherwise, remove flow from
1246 * idle_heap and set S to max(F,V).
1247 * Second, compute finish time F = S + len/weight.
1248 * Third, if pipe was idle, update V=max(S, V).
1249 * Fourth, count one more backlogged flow.
1250 */
1251 if (DN_KEY_GT(q->S, q->F)) { /* means timestamps are invalid */
1252 q->S = pipe->V ;
1253 pipe->sum += fs->weight ; /* add weight of new queue */
1254 } else {
1255 heap_extract(&(pipe->idle_heap), q);
1256 q->S = MAX64(q->F, pipe->V ) ;
1257 }
1258 q->F = q->S + ( len<<MY_M )/(u_int64_t) fs->weight;
1259
1260 if (pipe->not_eligible_heap.elements == 0 &&
1261 pipe->scheduler_heap.elements == 0)
1262 pipe->V = MAX64 ( q->S, pipe->V );
1263 fs->backlogged++ ;
1264 /*
1265 * Look at eligibility. A flow is not eligibile if S>V (when
1266 * this happens, it means that there is some other flow already
1267 * scheduled for the same pipe, so the scheduler_heap cannot be
1268 * empty). If the flow is not eligible we just store it in the
1269 * not_eligible_heap. Otherwise, we store in the scheduler_heap
1270 * and possibly invoke ready_event_wfq() right now if there is
1271 * leftover credit.
1272 * Note that for all flows in scheduler_heap (SCH), S_i <= V,
1273 * and for all flows in not_eligible_heap (NEH), S_i > V .
1274 * So when we need to compute max( V, min(S_i) ) forall i in SCH+NEH,
1275 * we only need to look into NEH.
1276 */
1277 if (DN_KEY_GT(q->S, pipe->V) ) { /* not eligible */
1278 if (pipe->scheduler_heap.elements == 0)
1279 printf("dummynet: ++ ouch! not eligible but empty scheduler!\n");
1280 heap_insert(&(pipe->not_eligible_heap), q->S, q);
1281 } else {
1282 heap_insert(&(pipe->scheduler_heap), q->F, q);
1283 if (pipe->numbytes >= 0) { /* pipe is idle */
1284 if (pipe->scheduler_heap.elements != 1)
1285 printf("dummynet: OUCH! pipe should have been idle!\n");
1286 DPRINTF(("dummynet: waking up pipe %d at %d\n",
1287 pipe->pipe_nr, (int)(q->F >> MY_M)));
1288 pipe->sched_time = curr_time ;
1289 ready_event_wfq(pipe);
1290 }
1291 }
1292 }
1293 done:
1294 DUMMYNET_UNLOCK();
1295 return 0;
1296
1297 dropit:
1298 if (q)
1299 q->drops++ ;
1300 DUMMYNET_UNLOCK();
1301 m_freem(m);
1302 return ( (fs && (fs->flags_fs & DN_NOERROR)) ? 0 : ENOBUFS);
1303 }
1304
1305 /*
1306 * Below, the rt_unref is only needed when (pkt->dn_dir == DN_TO_IP_OUT)
1307 * Doing this would probably save us the initial bzero of dn_pkt
1308 */
1309 #define DN_FREE_PKT(_m) do { \
1310 m_freem(_m); \
1311 } while (0)
1312
1313 /*
1314 * Dispose all packets and flow_queues on a flow_set.
1315 * If all=1, also remove red lookup table and other storage,
1316 * including the descriptor itself.
1317 * For the one in dn_pipe MUST also cleanup ready_heap...
1318 */
1319 static void
1320 purge_flow_set(struct dn_flow_set *fs, int all)
1321 {
1322 struct dn_flow_queue *q, *qn ;
1323 int i ;
1324
1325 DUMMYNET_LOCK_ASSERT();
1326
1327 for (i = 0 ; i <= fs->rq_size ; i++ ) {
1328 for (q = fs->rq[i] ; q ; q = qn ) {
1329 struct mbuf *m, *mnext;
1330
1331 mnext = q->head;
1332 while ((m = mnext) != NULL) {
1333 mnext = m->m_nextpkt;
1334 DN_FREE_PKT(m);
1335 }
1336 qn = q->next ;
1337 free(q, M_DUMMYNET);
1338 }
1339 fs->rq[i] = NULL ;
1340 }
1341 fs->rq_elements = 0 ;
1342 if (all) {
1343 /* RED - free lookup table */
1344 if (fs->w_q_lookup)
1345 free(fs->w_q_lookup, M_DUMMYNET);
1346 if (fs->rq)
1347 free(fs->rq, M_DUMMYNET);
1348 /* if this fs is not part of a pipe, free it */
1349 if (fs->pipe && fs != &(fs->pipe->fs) )
1350 free(fs, M_DUMMYNET);
1351 }
1352 }
1353
1354 /*
1355 * Dispose all packets queued on a pipe (not a flow_set).
1356 * Also free all resources associated to a pipe, which is about
1357 * to be deleted.
1358 */
1359 static void
1360 purge_pipe(struct dn_pipe *pipe)
1361 {
1362 struct mbuf *m, *mnext;
1363
1364 purge_flow_set( &(pipe->fs), 1 );
1365
1366 mnext = pipe->head;
1367 while ((m = mnext) != NULL) {
1368 mnext = m->m_nextpkt;
1369 DN_FREE_PKT(m);
1370 }
1371
1372 heap_free( &(pipe->scheduler_heap) );
1373 heap_free( &(pipe->not_eligible_heap) );
1374 heap_free( &(pipe->idle_heap) );
1375 }
1376
1377 /*
1378 * Delete all pipes and heaps returning memory. Must also
1379 * remove references from all ipfw rules to all pipes.
1380 */
1381 static void
1382 dummynet_flush()
1383 {
1384 struct dn_pipe *curr_p, *p ;
1385 struct dn_flow_set *fs, *curr_fs;
1386
1387 DUMMYNET_LOCK();
1388 /* remove all references to pipes ...*/
1389 flush_pipe_ptrs(NULL);
1390 /* prevent future matches... */
1391 p = all_pipes ;
1392 all_pipes = NULL ;
1393 fs = all_flow_sets ;
1394 all_flow_sets = NULL ;
1395 /* and free heaps so we don't have unwanted events */
1396 heap_free(&ready_heap);
1397 heap_free(&wfq_ready_heap);
1398 heap_free(&extract_heap);
1399
1400 /*
1401 * Now purge all queued pkts and delete all pipes
1402 */
1403 /* scan and purge all flow_sets. */
1404 for ( ; fs ; ) {
1405 curr_fs = fs ;
1406 fs = fs->next ;
1407 purge_flow_set(curr_fs, 1);
1408 }
1409 for ( ; p ; ) {
1410 purge_pipe(p);
1411 curr_p = p ;
1412 p = p->next ;
1413 free(curr_p, M_DUMMYNET);
1414 }
1415 DUMMYNET_UNLOCK();
1416 }
1417
1418
1419 extern struct ip_fw *ip_fw_default_rule ;
1420 static void
1421 dn_rule_delete_fs(struct dn_flow_set *fs, void *r)
1422 {
1423 int i ;
1424 struct dn_flow_queue *q ;
1425 struct mbuf *m ;
1426
1427 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is ovflow */
1428 for (q = fs->rq[i] ; q ; q = q->next )
1429 for (m = q->head ; m ; m = m->m_nextpkt ) {
1430 struct dn_pkt_tag *pkt = dn_tag_get(m) ;
1431 if (pkt->rule == r)
1432 pkt->rule = ip_fw_default_rule ;
1433 }
1434 }
1435 /*
1436 * when a firewall rule is deleted, scan all queues and remove the flow-id
1437 * from packets matching this rule.
1438 */
1439 void
1440 dn_rule_delete(void *r)
1441 {
1442 struct dn_pipe *p ;
1443 struct dn_flow_set *fs ;
1444 struct dn_pkt_tag *pkt ;
1445 struct mbuf *m ;
1446
1447 DUMMYNET_LOCK();
1448 /*
1449 * If the rule references a queue (dn_flow_set), then scan
1450 * the flow set, otherwise scan pipes. Should do either, but doing
1451 * both does not harm.
1452 */
1453 for ( fs = all_flow_sets ; fs ; fs = fs->next )
1454 dn_rule_delete_fs(fs, r);
1455 for ( p = all_pipes ; p ; p = p->next ) {
1456 fs = &(p->fs) ;
1457 dn_rule_delete_fs(fs, r);
1458 for (m = p->head ; m ; m = m->m_nextpkt ) {
1459 pkt = dn_tag_get(m) ;
1460 if (pkt->rule == r)
1461 pkt->rule = ip_fw_default_rule ;
1462 }
1463 }
1464 DUMMYNET_UNLOCK();
1465 }
1466
1467 /*
1468 * setup RED parameters
1469 */
1470 static int
1471 config_red(struct dn_flow_set *p, struct dn_flow_set * x)
1472 {
1473 int i;
1474
1475 x->w_q = p->w_q;
1476 x->min_th = SCALE(p->min_th);
1477 x->max_th = SCALE(p->max_th);
1478 x->max_p = p->max_p;
1479
1480 x->c_1 = p->max_p / (p->max_th - p->min_th);
1481 x->c_2 = SCALE_MUL(x->c_1, SCALE(p->min_th));
1482 if (x->flags_fs & DN_IS_GENTLE_RED) {
1483 x->c_3 = (SCALE(1) - p->max_p) / p->max_th;
1484 x->c_4 = (SCALE(1) - 2 * p->max_p);
1485 }
1486
1487 /* if the lookup table already exist, free and create it again */
1488 if (x->w_q_lookup) {
1489 free(x->w_q_lookup, M_DUMMYNET);
1490 x->w_q_lookup = NULL ;
1491 }
1492 if (red_lookup_depth == 0) {
1493 printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth must be > 0\n");
1494 free(x, M_DUMMYNET);
1495 return EINVAL;
1496 }
1497 x->lookup_depth = red_lookup_depth;
1498 x->w_q_lookup = (u_int *) malloc(x->lookup_depth * sizeof(int),
1499 M_DUMMYNET, M_NOWAIT);
1500 if (x->w_q_lookup == NULL) {
1501 printf("dummynet: sorry, cannot allocate red lookup table\n");
1502 free(x, M_DUMMYNET);
1503 return ENOSPC;
1504 }
1505
1506 /* fill the lookup table with (1 - w_q)^x */
1507 x->lookup_step = p->lookup_step ;
1508 x->lookup_weight = p->lookup_weight ;
1509 x->w_q_lookup[0] = SCALE(1) - x->w_q;
1510 for (i = 1; i < x->lookup_depth; i++)
1511 x->w_q_lookup[i] = SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight);
1512 if (red_avg_pkt_size < 1)
1513 red_avg_pkt_size = 512 ;
1514 x->avg_pkt_size = red_avg_pkt_size ;
1515 if (red_max_pkt_size < 1)
1516 red_max_pkt_size = 1500 ;
1517 x->max_pkt_size = red_max_pkt_size ;
1518 return 0 ;
1519 }
1520
1521 static int
1522 alloc_hash(struct dn_flow_set *x, struct dn_flow_set *pfs)
1523 {
1524 if (x->flags_fs & DN_HAVE_FLOW_MASK) { /* allocate some slots */
1525 int l = pfs->rq_size;
1526
1527 if (l == 0)
1528 l = dn_hash_size;
1529 if (l < 4)
1530 l = 4;
1531 else if (l > DN_MAX_HASH_SIZE)
1532 l = DN_MAX_HASH_SIZE;
1533 x->rq_size = l;
1534 } else /* one is enough for null mask */
1535 x->rq_size = 1;
1536 x->rq = malloc((1 + x->rq_size) * sizeof(struct dn_flow_queue *),
1537 M_DUMMYNET, M_NOWAIT | M_ZERO);
1538 if (x->rq == NULL) {
1539 printf("dummynet: sorry, cannot allocate queue\n");
1540 return ENOSPC;
1541 }
1542 x->rq_elements = 0;
1543 return 0 ;
1544 }
1545
1546 static void
1547 set_fs_parms(struct dn_flow_set *x, struct dn_flow_set *src)
1548 {
1549 x->flags_fs = src->flags_fs;
1550 x->qsize = src->qsize;
1551 x->plr = src->plr;
1552 x->flow_mask = src->flow_mask;
1553 if (x->flags_fs & DN_QSIZE_IS_BYTES) {
1554 if (x->qsize > 1024*1024)
1555 x->qsize = 1024*1024 ;
1556 } else {
1557 if (x->qsize == 0)
1558 x->qsize = 50 ;
1559 if (x->qsize > 100)
1560 x->qsize = 50 ;
1561 }
1562 /* configuring RED */
1563 if ( x->flags_fs & DN_IS_RED )
1564 config_red(src, x) ; /* XXX should check errors */
1565 }
1566
1567 /*
1568 * setup pipe or queue parameters.
1569 */
1570
1571 static int
1572 config_pipe(struct dn_pipe *p)
1573 {
1574 int i, r;
1575 struct dn_flow_set *pfs = &(p->fs);
1576 struct dn_flow_queue *q;
1577
1578 /*
1579 * The config program passes parameters as follows:
1580 * bw = bits/second (0 means no limits),
1581 * delay = ms, must be translated into ticks.
1582 * qsize = slots/bytes
1583 */
1584 p->delay = ( p->delay * hz ) / 1000 ;
1585 /* We need either a pipe number or a flow_set number */
1586 if (p->pipe_nr == 0 && pfs->fs_nr == 0)
1587 return EINVAL ;
1588 if (p->pipe_nr != 0 && pfs->fs_nr != 0)
1589 return EINVAL ;
1590 if (p->pipe_nr != 0) { /* this is a pipe */
1591 struct dn_pipe *x, *a, *b;
1592
1593 DUMMYNET_LOCK();
1594 /* locate pipe */
1595 for (a = NULL , b = all_pipes ; b && b->pipe_nr < p->pipe_nr ;
1596 a = b , b = b->next) ;
1597
1598 if (b == NULL || b->pipe_nr != p->pipe_nr) { /* new pipe */
1599 x = malloc(sizeof(struct dn_pipe), M_DUMMYNET, M_NOWAIT | M_ZERO);
1600 if (x == NULL) {
1601 DUMMYNET_UNLOCK();
1602 printf("dummynet: no memory for new pipe\n");
1603 return ENOSPC;
1604 }
1605 x->pipe_nr = p->pipe_nr;
1606 x->fs.pipe = x ;
1607 /* idle_heap is the only one from which we extract from the middle.
1608 */
1609 x->idle_heap.size = x->idle_heap.elements = 0 ;
1610 x->idle_heap.offset=OFFSET_OF(struct dn_flow_queue, heap_pos);
1611 } else {
1612 x = b;
1613 /* Flush accumulated credit for all queues */
1614 for (i = 0; i <= x->fs.rq_size; i++)
1615 for (q = x->fs.rq[i]; q; q = q->next)
1616 q->numbytes = 0;
1617 }
1618
1619 x->bandwidth = p->bandwidth ;
1620 x->numbytes = 0; /* just in case... */
1621 bcopy(p->if_name, x->if_name, sizeof(p->if_name) );
1622 x->ifp = NULL ; /* reset interface ptr */
1623 x->delay = p->delay ;
1624 set_fs_parms(&(x->fs), pfs);
1625
1626
1627 if ( x->fs.rq == NULL ) { /* a new pipe */
1628 r = alloc_hash(&(x->fs), pfs) ;
1629 if (r) {
1630 DUMMYNET_UNLOCK();
1631 free(x, M_DUMMYNET);
1632 return r ;
1633 }
1634 x->next = b ;
1635 if (a == NULL)
1636 all_pipes = x ;
1637 else
1638 a->next = x ;
1639 }
1640 DUMMYNET_UNLOCK();
1641 } else { /* config queue */
1642 struct dn_flow_set *x, *a, *b ;
1643
1644 DUMMYNET_LOCK();
1645 /* locate flow_set */
1646 for (a=NULL, b=all_flow_sets ; b && b->fs_nr < pfs->fs_nr ;
1647 a = b , b = b->next) ;
1648
1649 if (b == NULL || b->fs_nr != pfs->fs_nr) { /* new */
1650 if (pfs->parent_nr == 0) { /* need link to a pipe */
1651 DUMMYNET_UNLOCK();
1652 return EINVAL ;
1653 }
1654 x = malloc(sizeof(struct dn_flow_set), M_DUMMYNET, M_NOWAIT|M_ZERO);
1655 if (x == NULL) {
1656 DUMMYNET_UNLOCK();
1657 printf("dummynet: no memory for new flow_set\n");
1658 return ENOSPC;
1659 }
1660 x->fs_nr = pfs->fs_nr;
1661 x->parent_nr = pfs->parent_nr;
1662 x->weight = pfs->weight ;
1663 if (x->weight == 0)
1664 x->weight = 1 ;
1665 else if (x->weight > 100)
1666 x->weight = 100 ;
1667 } else {
1668 /* Change parent pipe not allowed; must delete and recreate */
1669 if (pfs->parent_nr != 0 && b->parent_nr != pfs->parent_nr) {
1670 DUMMYNET_UNLOCK();
1671 return EINVAL ;
1672 }
1673 x = b;
1674 }
1675 set_fs_parms(x, pfs);
1676
1677 if ( x->rq == NULL ) { /* a new flow_set */
1678 r = alloc_hash(x, pfs) ;
1679 if (r) {
1680 DUMMYNET_UNLOCK();
1681 free(x, M_DUMMYNET);
1682 return r ;
1683 }
1684 x->next = b;
1685 if (a == NULL)
1686 all_flow_sets = x;
1687 else
1688 a->next = x;
1689 }
1690 DUMMYNET_UNLOCK();
1691 }
1692 return 0 ;
1693 }
1694
1695 /*
1696 * Helper function to remove from a heap queues which are linked to
1697 * a flow_set about to be deleted.
1698 */
1699 static void
1700 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs)
1701 {
1702 int i = 0, found = 0 ;
1703 for (; i < h->elements ;)
1704 if ( ((struct dn_flow_queue *)h->p[i].object)->fs == fs) {
1705 h->elements-- ;
1706 h->p[i] = h->p[h->elements] ;
1707 found++ ;
1708 } else
1709 i++ ;
1710 if (found)
1711 heapify(h);
1712 }
1713
1714 /*
1715 * helper function to remove a pipe from a heap (can be there at most once)
1716 */
1717 static void
1718 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p)
1719 {
1720 if (h->elements > 0) {
1721 int i = 0 ;
1722 for (i=0; i < h->elements ; i++ ) {
1723 if (h->p[i].object == p) { /* found it */
1724 h->elements-- ;
1725 h->p[i] = h->p[h->elements] ;
1726 heapify(h);
1727 break ;
1728 }
1729 }
1730 }
1731 }
1732
1733 /*
1734 * drain all queues. Called in case of severe mbuf shortage.
1735 */
1736 void
1737 dummynet_drain()
1738 {
1739 struct dn_flow_set *fs;
1740 struct dn_pipe *p;
1741 struct mbuf *m, *mnext;
1742
1743 DUMMYNET_LOCK_ASSERT();
1744
1745 heap_free(&ready_heap);
1746 heap_free(&wfq_ready_heap);
1747 heap_free(&extract_heap);
1748 /* remove all references to this pipe from flow_sets */
1749 for (fs = all_flow_sets; fs; fs= fs->next )
1750 purge_flow_set(fs, 0);
1751
1752 for (p = all_pipes; p; p= p->next ) {
1753 purge_flow_set(&(p->fs), 0);
1754
1755 mnext = p->head;
1756 while ((m = mnext) != NULL) {
1757 mnext = m->m_nextpkt;
1758 DN_FREE_PKT(m);
1759 }
1760 p->head = p->tail = NULL ;
1761 }
1762 }
1763
1764 /*
1765 * Fully delete a pipe or a queue, cleaning up associated info.
1766 */
1767 static int
1768 delete_pipe(struct dn_pipe *p)
1769 {
1770 if (p->pipe_nr == 0 && p->fs.fs_nr == 0)
1771 return EINVAL ;
1772 if (p->pipe_nr != 0 && p->fs.fs_nr != 0)
1773 return EINVAL ;
1774 if (p->pipe_nr != 0) { /* this is an old-style pipe */
1775 struct dn_pipe *a, *b;
1776 struct dn_flow_set *fs;
1777
1778 DUMMYNET_LOCK();
1779 /* locate pipe */
1780 for (a = NULL , b = all_pipes ; b && b->pipe_nr < p->pipe_nr ;
1781 a = b , b = b->next) ;
1782 if (b == NULL || (b->pipe_nr != p->pipe_nr) ) {
1783 DUMMYNET_UNLOCK();
1784 return EINVAL ; /* not found */
1785 }
1786
1787 /* unlink from list of pipes */
1788 if (a == NULL)
1789 all_pipes = b->next ;
1790 else
1791 a->next = b->next ;
1792 /* remove references to this pipe from the ip_fw rules. */
1793 flush_pipe_ptrs(&(b->fs));
1794
1795 /* remove all references to this pipe from flow_sets */
1796 for (fs = all_flow_sets; fs; fs= fs->next )
1797 if (fs->pipe == b) {
1798 printf("dummynet: ++ ref to pipe %d from fs %d\n",
1799 p->pipe_nr, fs->fs_nr);
1800 fs->pipe = NULL ;
1801 purge_flow_set(fs, 0);
1802 }
1803 fs_remove_from_heap(&ready_heap, &(b->fs));
1804 purge_pipe(b); /* remove all data associated to this pipe */
1805 /* remove reference to here from extract_heap and wfq_ready_heap */
1806 pipe_remove_from_heap(&extract_heap, b);
1807 pipe_remove_from_heap(&wfq_ready_heap, b);
1808 DUMMYNET_UNLOCK();
1809
1810 free(b, M_DUMMYNET);
1811 } else { /* this is a WF2Q queue (dn_flow_set) */
1812 struct dn_flow_set *a, *b;
1813
1814 DUMMYNET_LOCK();
1815 /* locate set */
1816 for (a = NULL, b = all_flow_sets ; b && b->fs_nr < p->fs.fs_nr ;
1817 a = b , b = b->next) ;
1818 if (b == NULL || (b->fs_nr != p->fs.fs_nr) ) {
1819 DUMMYNET_UNLOCK();
1820 return EINVAL ; /* not found */
1821 }
1822
1823 if (a == NULL)
1824 all_flow_sets = b->next ;
1825 else
1826 a->next = b->next ;
1827 /* remove references to this flow_set from the ip_fw rules. */
1828 flush_pipe_ptrs(b);
1829
1830 if (b->pipe != NULL) {
1831 /* Update total weight on parent pipe and cleanup parent heaps */
1832 b->pipe->sum -= b->weight * b->backlogged ;
1833 fs_remove_from_heap(&(b->pipe->not_eligible_heap), b);
1834 fs_remove_from_heap(&(b->pipe->scheduler_heap), b);
1835 #if 1 /* XXX should i remove from idle_heap as well ? */
1836 fs_remove_from_heap(&(b->pipe->idle_heap), b);
1837 #endif
1838 }
1839 purge_flow_set(b, 1);
1840 DUMMYNET_UNLOCK();
1841 }
1842 return 0 ;
1843 }
1844
1845 /*
1846 * helper function used to copy data from kernel in DUMMYNET_GET
1847 */
1848 static char *
1849 dn_copy_set(struct dn_flow_set *set, char *bp)
1850 {
1851 int i, copied = 0 ;
1852 struct dn_flow_queue *q, *qp = (struct dn_flow_queue *)bp;
1853
1854 DUMMYNET_LOCK_ASSERT();
1855
1856 for (i = 0 ; i <= set->rq_size ; i++)
1857 for (q = set->rq[i] ; q ; q = q->next, qp++ ) {
1858 if (q->hash_slot != i)
1859 printf("dummynet: ++ at %d: wrong slot (have %d, "
1860 "should be %d)\n", copied, q->hash_slot, i);
1861 if (q->fs != set)
1862 printf("dummynet: ++ at %d: wrong fs ptr (have %p, should be %p)\n",
1863 i, q->fs, set);
1864 copied++ ;
1865 bcopy(q, qp, sizeof( *q ) );
1866 /* cleanup pointers */
1867 qp->next = NULL ;
1868 qp->head = qp->tail = NULL ;
1869 qp->fs = NULL ;
1870 }
1871 if (copied != set->rq_elements)
1872 printf("dummynet: ++ wrong count, have %d should be %d\n",
1873 copied, set->rq_elements);
1874 return (char *)qp ;
1875 }
1876
1877 static size_t
1878 dn_calc_size(void)
1879 {
1880 struct dn_flow_set *set ;
1881 struct dn_pipe *p ;
1882 size_t size ;
1883
1884 DUMMYNET_LOCK_ASSERT();
1885 /*
1886 * compute size of data structures: list of pipes and flow_sets.
1887 */
1888 for (p = all_pipes, size = 0 ; p ; p = p->next )
1889 size += sizeof( *p ) +
1890 p->fs.rq_elements * sizeof(struct dn_flow_queue);
1891 for (set = all_flow_sets ; set ; set = set->next )
1892 size += sizeof ( *set ) +
1893 set->rq_elements * sizeof(struct dn_flow_queue);
1894 return size ;
1895 }
1896
1897 static int
1898 dummynet_get(struct sockopt *sopt)
1899 {
1900 char *buf, *bp ; /* bp is the "copy-pointer" */
1901 size_t size ;
1902 struct dn_flow_set *set ;
1903 struct dn_pipe *p ;
1904 int error=0, i ;
1905
1906 /* XXX lock held too long */
1907 DUMMYNET_LOCK();
1908 /*
1909 * XXX: Ugly, but we need to allocate memory with M_WAITOK flag and we
1910 * cannot use this flag while holding a mutex.
1911 */
1912 for (i = 0; i < 10; i++) {
1913 size = dn_calc_size();
1914 DUMMYNET_UNLOCK();
1915 buf = malloc(size, M_TEMP, M_WAITOK);
1916 DUMMYNET_LOCK();
1917 if (size == dn_calc_size())
1918 break;
1919 free(buf, M_TEMP);
1920 buf = NULL;
1921 }
1922 if (buf == NULL) {
1923 DUMMYNET_UNLOCK();
1924 return ENOBUFS ;
1925 }
1926 for (p = all_pipes, bp = buf ; p ; p = p->next ) {
1927 struct dn_pipe *pipe_bp = (struct dn_pipe *)bp ;
1928
1929 /*
1930 * copy pipe descriptor into *bp, convert delay back to ms,
1931 * then copy the flow_set descriptor(s) one at a time.
1932 * After each flow_set, copy the queue descriptor it owns.
1933 */
1934 bcopy(p, bp, sizeof( *p ) );
1935 pipe_bp->delay = (pipe_bp->delay * 1000) / hz ;
1936 /*
1937 * XXX the following is a hack based on ->next being the
1938 * first field in dn_pipe and dn_flow_set. The correct
1939 * solution would be to move the dn_flow_set to the beginning
1940 * of struct dn_pipe.
1941 */
1942 pipe_bp->next = (struct dn_pipe *)DN_IS_PIPE ;
1943 /* clean pointers */
1944 pipe_bp->head = pipe_bp->tail = NULL ;
1945 pipe_bp->fs.next = NULL ;
1946 pipe_bp->fs.pipe = NULL ;
1947 pipe_bp->fs.rq = NULL ;
1948
1949 bp += sizeof( *p ) ;
1950 bp = dn_copy_set( &(p->fs), bp );
1951 }
1952 for (set = all_flow_sets ; set ; set = set->next ) {
1953 struct dn_flow_set *fs_bp = (struct dn_flow_set *)bp ;
1954 bcopy(set, bp, sizeof( *set ) );
1955 /* XXX same hack as above */
1956 fs_bp->next = (struct dn_flow_set *)DN_IS_QUEUE ;
1957 fs_bp->pipe = NULL ;
1958 fs_bp->rq = NULL ;
1959 bp += sizeof( *set ) ;
1960 bp = dn_copy_set( set, bp );
1961 }
1962 DUMMYNET_UNLOCK();
1963
1964 error = sooptcopyout(sopt, buf, size);
1965 free(buf, M_TEMP);
1966 return error ;
1967 }
1968
1969 /*
1970 * Handler for the various dummynet socket options (get, flush, config, del)
1971 */
1972 static int
1973 ip_dn_ctl(struct sockopt *sopt)
1974 {
1975 int error = 0 ;
1976 struct dn_pipe *p, tmp_pipe;
1977
1978 /* Disallow sets in really-really secure mode. */
1979 if (sopt->sopt_dir == SOPT_SET) {
1980 #if __FreeBSD_version >= 500034
1981 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
1982 if (error)
1983 return (error);
1984 #else
1985 if (securelevel >= 3)
1986 return (EPERM);
1987 #endif
1988 }
1989
1990 switch (sopt->sopt_name) {
1991 default :
1992 printf("dummynet: -- unknown option %d", sopt->sopt_name);
1993 return EINVAL ;
1994
1995 case IP_DUMMYNET_GET :
1996 error = dummynet_get(sopt);
1997 break ;
1998
1999 case IP_DUMMYNET_FLUSH :
2000 dummynet_flush() ;
2001 break ;
2002
2003 case IP_DUMMYNET_CONFIGURE :
2004 p = &tmp_pipe ;
2005 error = sooptcopyin(sopt, p, sizeof *p, sizeof *p);
2006 if (error)
2007 break ;
2008 error = config_pipe(p);
2009 break ;
2010
2011 case IP_DUMMYNET_DEL : /* remove a pipe or queue */
2012 p = &tmp_pipe ;
2013 error = sooptcopyin(sopt, p, sizeof *p, sizeof *p);
2014 if (error)
2015 break ;
2016
2017 error = delete_pipe(p);
2018 break ;
2019 }
2020 return error ;
2021 }
2022
2023 static void
2024 ip_dn_init(void)
2025 {
2026 if (bootverbose)
2027 printf("DUMMYNET initialized (011031)\n");
2028
2029 DUMMYNET_LOCK_INIT();
2030
2031 all_pipes = NULL ;
2032 all_flow_sets = NULL ;
2033 ready_heap.size = ready_heap.elements = 0 ;
2034 ready_heap.offset = 0 ;
2035
2036 wfq_ready_heap.size = wfq_ready_heap.elements = 0 ;
2037 wfq_ready_heap.offset = 0 ;
2038
2039 extract_heap.size = extract_heap.elements = 0 ;
2040 extract_heap.offset = 0 ;
2041
2042 ip_dn_ctl_ptr = ip_dn_ctl;
2043 ip_dn_io_ptr = dummynet_io;
2044 ip_dn_ruledel_ptr = dn_rule_delete;
2045
2046 callout_init(&dn_timeout, debug_mpsafenet ? CALLOUT_MPSAFE : 0);
2047 callout_reset(&dn_timeout, 1, dummynet, NULL);
2048 }
2049
2050 #ifdef KLD_MODULE
2051 static void
2052 ip_dn_destroy(void)
2053 {
2054 ip_dn_ctl_ptr = NULL;
2055 ip_dn_io_ptr = NULL;
2056 ip_dn_ruledel_ptr = NULL;
2057
2058 callout_stop(&dn_timeout);
2059 dummynet_flush();
2060
2061 DUMMYNET_LOCK_DESTROY();
2062 }
2063 #endif /* KLD_MODULE */
2064
2065 static int
2066 dummynet_modevent(module_t mod, int type, void *data)
2067 {
2068 switch (type) {
2069 case MOD_LOAD:
2070 if (DUMMYNET_LOADED) {
2071 printf("DUMMYNET already loaded\n");
2072 return EEXIST ;
2073 }
2074 ip_dn_init();
2075 break;
2076
2077 case MOD_UNLOAD:
2078 #if !defined(KLD_MODULE)
2079 printf("dummynet statically compiled, cannot unload\n");
2080 return EINVAL ;
2081 #else
2082 ip_dn_destroy();
2083 #endif
2084 break ;
2085 default:
2086 return EOPNOTSUPP;
2087 break ;
2088 }
2089 return 0 ;
2090 }
2091
2092 static moduledata_t dummynet_mod = {
2093 "dummynet",
2094 dummynet_modevent,
2095 NULL
2096 };
2097 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY);
2098 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2);
2099 MODULE_VERSION(dummynet, 1);
Cache object: e681e27d605a5f29732eb104b3f48df2
|