2 * Copyright (c) 2010 Luigi Rizzo, Riccardo Panicucci, Universita` di Pisa
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * Dummynet portions related to packet handling.
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
33 #include "opt_inet6.h"
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/malloc.h>
39 #include <sys/kernel.h>
41 #include <sys/module.h>
44 #include <sys/rwlock.h>
45 #include <sys/socket.h>
47 #include <sys/sysctl.h>
49 #include <net/if.h> /* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */
50 #include <net/netisr.h>
53 #include <netinet/in.h>
54 #include <netinet/ip.h> /* ip_len, ip_off */
55 #include <netinet/ip_var.h> /* ip_output(), IP_FORWARDING */
56 #include <netinet/ip_fw.h>
57 #include <netinet/ip_dummynet.h>
58 #include <netinet/if_ether.h> /* various ether_* routines */
59 #include <netinet/ip6.h> /* for ip6_input, ip6_output prototypes */
60 #include <netinet6/ip6_var.h>
62 #include <netpfil/ipfw/ip_fw_private.h>
63 #include <netpfil/ipfw/dn_heap.h>
64 #include <netpfil/ipfw/ip_dn_private.h>
65 #include <netpfil/ipfw/dn_sched.h>
68 * We keep a private variable for the simulation time, but we could
69 * probably use an existing one ("softticks" in sys/kern/kern_timeout.c)
70 * instead of dn_cfg.curr_time
73 struct dn_parms dn_cfg;
74 //VNET_DEFINE(struct dn_parms, _base_dn_cfg);
76 static long tick_last; /* Last tick duration (usec). */
77 static long tick_delta; /* Last vs standard tick diff (usec). */
78 static long tick_delta_sum; /* Accumulated tick difference (usec).*/
79 static long tick_adjustment; /* Tick adjustments done. */
80 static long tick_lost; /* Lost(coalesced) ticks number. */
81 /* Adjusted vs non-adjusted curr_time difference (ticks). */
82 static long tick_diff;
84 static unsigned long io_pkt;
85 static unsigned long io_pkt_fast;
86 static unsigned long io_pkt_drop;
89 * We use a heap to store entities for which we have pending timer events.
90 * The heap is checked at every tick and all entities with expired events
94 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap");
96 extern void (*bridge_dn_p)(struct mbuf *, struct ifnet *);
102 SYSCTL_DECL(_net_inet);
103 SYSCTL_DECL(_net_inet_ip);
104 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet, CTLFLAG_RW, 0, "Dummynet");
106 /* wrapper to pass dn_cfg fields to SYSCTL_* */
107 //#define DC(x) (&(VNET_NAME(_base_dn_cfg).x))
108 #define DC(x) (&(dn_cfg.x))
112 sysctl_hash_size(SYSCTL_HANDLER_ARGS)
116 value = dn_cfg.hash_size;
117 error = sysctl_handle_int(oidp, &value, 0, req);
118 if (error != 0 || req->newptr == NULL)
120 if (value < 16 || value > 65536)
122 dn_cfg.hash_size = value;
126 SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, hash_size,
127 CTLTYPE_INT | CTLFLAG_RW, 0, 0, sysctl_hash_size,
128 "I", "Default hash table size");
131 sysctl_limits(SYSCTL_HANDLER_ARGS)
137 value = dn_cfg.slot_limit;
139 value = dn_cfg.byte_limit;
140 error = sysctl_handle_long(oidp, &value, 0, req);
142 if (error != 0 || req->newptr == NULL)
147 dn_cfg.slot_limit = value;
151 dn_cfg.byte_limit = value;
156 SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, pipe_slot_limit,
157 CTLTYPE_LONG | CTLFLAG_RW, 0, 1, sysctl_limits,
158 "L", "Upper limit in slots for pipe queue.");
159 SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, pipe_byte_limit,
160 CTLTYPE_LONG | CTLFLAG_RW, 0, 0, sysctl_limits,
161 "L", "Upper limit in bytes for pipe queue.");
162 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, io_fast,
163 CTLFLAG_RW, DC(io_fast), 0, "Enable fast dummynet io.");
164 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, debug,
165 CTLFLAG_RW, DC(debug), 0, "Dummynet debug level");
168 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth,
169 CTLFLAG_RD, DC(red_lookup_depth), 0, "Depth of RED lookup table");
170 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size,
171 CTLFLAG_RD, DC(red_avg_pkt_size), 0, "RED Medium packet size");
172 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size,
173 CTLFLAG_RD, DC(red_max_pkt_size), 0, "RED Max packet size");
175 /* time adjustment */
176 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta,
177 CTLFLAG_RD, &tick_delta, 0, "Last vs standard tick difference (usec).");
178 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta_sum,
179 CTLFLAG_RD, &tick_delta_sum, 0, "Accumulated tick difference (usec).");
180 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_adjustment,
181 CTLFLAG_RD, &tick_adjustment, 0, "Tick adjustments done.");
182 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_diff,
183 CTLFLAG_RD, &tick_diff, 0,
184 "Adjusted vs non-adjusted curr_time difference (ticks).");
185 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_lost,
186 CTLFLAG_RD, &tick_lost, 0,
187 "Number of ticks coalesced by dummynet taskqueue.");
189 /* Drain parameters */
190 SYSCTL_UINT(_net_inet_ip_dummynet, OID_AUTO, expire,
191 CTLFLAG_RW, DC(expire), 0, "Expire empty queues/pipes");
192 SYSCTL_UINT(_net_inet_ip_dummynet, OID_AUTO, expire_cycle,
193 CTLFLAG_RD, DC(expire_cycle), 0, "Expire cycle for queues/pipes");
196 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, schk_count,
197 CTLFLAG_RD, DC(schk_count), 0, "Number of schedulers");
198 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, si_count,
199 CTLFLAG_RD, DC(si_count), 0, "Number of scheduler instances");
200 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, fsk_count,
201 CTLFLAG_RD, DC(fsk_count), 0, "Number of flowsets");
202 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, queue_count,
203 CTLFLAG_RD, DC(queue_count), 0, "Number of queues");
204 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt,
205 CTLFLAG_RD, &io_pkt, 0,
206 "Number of packets passed to dummynet.");
207 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_fast,
208 CTLFLAG_RD, &io_pkt_fast, 0,
209 "Number of packets bypassed dummynet scheduler.");
210 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_drop,
211 CTLFLAG_RD, &io_pkt_drop, 0,
212 "Number of packets dropped by dummynet.");
218 static void dummynet_send(struct mbuf *);
221 * Packets processed by dummynet have an mbuf tag associated with
222 * them that carries their dummynet state.
223 * Outside dummynet, only the 'rule' field is relevant, and it must
224 * be at the beginning of the structure.
227 struct ipfw_rule_ref rule; /* matching rule */
229 /* second part, dummynet specific */
230 int dn_dir; /* action when packet comes out.*/
231 /* see ip_fw_private.h */
232 uint64_t output_time; /* when the pkt is due for delivery*/
233 struct ifnet *ifp; /* interface, for ip_output */
234 struct _ip6dn_args ip6opt; /* XXX ipv6 options */
238 * Return the mbuf tag holding the dummynet state (it should
239 * be the first one on the list).
241 static struct dn_pkt_tag *
242 dn_tag_get(struct mbuf *m)
244 struct m_tag *mtag = m_tag_first(m);
245 KASSERT(mtag != NULL &&
246 mtag->m_tag_cookie == MTAG_ABI_COMPAT &&
247 mtag->m_tag_id == PACKET_TAG_DUMMYNET,
248 ("packet on dummynet queue w/o dummynet tag!"));
249 return (struct dn_pkt_tag *)(mtag+1);
253 mq_append(struct mq *q, struct mbuf *m)
258 q->tail->m_nextpkt = m;
264 * Dispose a list of packet. Use a functions so if we need to do
265 * more work, this is a central point to do it.
267 void dn_free_pkts(struct mbuf *mnext)
271 while ((m = mnext) != NULL) {
272 mnext = m->m_nextpkt;
278 red_drops (struct dn_queue *q, int len)
283 * RED calculates the average queue size (avg) using a low-pass filter
284 * with an exponential weighted (w_q) moving average:
285 * avg <- (1-w_q) * avg + w_q * q_size
286 * where q_size is the queue length (measured in bytes or * packets).
288 * If q_size == 0, we compute the idle time for the link, and set
289 * avg = (1 - w_q)^(idle/s)
290 * where s is the time needed for transmitting a medium-sized packet.
292 * Now, if avg < min_th the packet is enqueued.
293 * If avg > max_th the packet is dropped. Otherwise, the packet is
294 * dropped with probability P function of avg.
297 struct dn_fsk *fs = q->fs;
300 /* Queue in bytes or packets? */
301 uint32_t q_size = (fs->fs.flags & DN_QSIZE_BYTES) ?
302 q->ni.len_bytes : q->ni.length;
304 /* Average queue size estimation. */
306 /* Queue is not empty, avg <- avg + (q_size - avg) * w_q */
307 int diff = SCALE(q_size) - q->avg;
308 int64_t v = SCALE_MUL((int64_t)diff, (int64_t)fs->w_q);
313 * Queue is empty, find for how long the queue has been
314 * empty and use a lookup table for computing
315 * (1 - * w_q)^(idle_time/s) where s is the time to send a
320 u_int t = div64((dn_cfg.curr_time - q->q_time), fs->lookup_step);
322 q->avg = (t < fs->lookup_depth) ?
323 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0;
328 if (q->avg < fs->min_th) {
330 return (0); /* accept packet */
332 if (q->avg >= fs->max_th) { /* average queue >= max threshold */
333 if (fs->fs.flags & DN_IS_GENTLE_RED) {
335 * According to Gentle-RED, if avg is greater than
336 * max_th the packet is dropped with a probability
337 * p_b = c_3 * avg - c_4
338 * where c_3 = (1 - max_p) / max_th
339 * c_4 = 1 - 2 * max_p
341 p_b = SCALE_MUL((int64_t)fs->c_3, (int64_t)q->avg) -
347 } else if (q->avg > fs->min_th) {
349 * We compute p_b using the linear dropping function
350 * p_b = c_1 * avg - c_2
351 * where c_1 = max_p / (max_th - min_th)
352 * c_2 = max_p * min_th / (max_th - min_th)
354 p_b = SCALE_MUL((int64_t)fs->c_1, (int64_t)q->avg) - fs->c_2;
357 if (fs->fs.flags & DN_QSIZE_BYTES)
358 p_b = div64((p_b * len) , fs->max_pkt_size);
360 q->random = random() & 0xffff;
363 * q->count counts packets arrived since last drop, so a greater
364 * value of q->count means a greater packet drop probability.
366 if (SCALE_MUL(p_b, SCALE((int64_t)q->count)) > q->random) {
368 /* After a drop we calculate a new random value. */
369 q->random = random() & 0xffff;
370 return (1); /* drop */
373 /* End of RED algorithm. */
375 return (0); /* accept */
380 * Enqueue a packet in q, subject to space and queue management policy
381 * (whose parameters are in q->fs).
382 * Update stats for the queue and the scheduler.
383 * Return 0 on success, 1 on drop. The packet is consumed anyways.
386 dn_enqueue(struct dn_queue *q, struct mbuf* m, int drop)
389 struct dn_flow *ni; /* stats for scheduler instance */
392 if (q->fs == NULL || q->_si == NULL) {
393 printf("%s fs %p si %p, dropping\n",
394 __FUNCTION__, q->fs, q->_si);
400 len = m->m_pkthdr.len;
401 /* Update statistics, then check reasons to drop pkt. */
402 q->ni.tot_bytes += len;
404 ni->tot_bytes += len;
408 if (f->plr && random() < f->plr)
410 if (f->flags & DN_IS_RED && red_drops(q, m->m_pkthdr.len))
412 if (f->flags & DN_QSIZE_BYTES) {
413 if (q->ni.len_bytes > f->qsize)
415 } else if (q->ni.length >= f->qsize) {
418 mq_append(&q->mq, m);
420 q->ni.len_bytes += len;
422 ni->len_bytes += len;
434 * Fetch packets from the delay line which are due now. If there are
435 * leftover packets, reinsert the delay line in the heap.
436 * Runs under scheduler lock.
439 transmit_event(struct mq *q, struct delay_line *dline, uint64_t now)
442 struct dn_pkt_tag *pkt = NULL;
444 dline->oid.subtype = 0; /* not in heap */
445 while ((m = dline->mq.head) != NULL) {
447 if (!DN_KEY_LEQ(pkt->output_time, now))
449 dline->mq.head = m->m_nextpkt;
453 dline->oid.subtype = 1; /* in heap */
454 heap_insert(&dn_cfg.evheap, pkt->output_time, dline);
459 * Convert the additional MAC overheads/delays into an equivalent
460 * number of bits for the given data rate. The samples are
461 * in milliseconds so we need to divide by 1000.
464 extra_bits(struct mbuf *m, struct dn_schk *s)
468 struct dn_profile *pf = s->profile;
470 if (!pf || pf->samples_no == 0)
472 index = random() % pf->samples_no;
473 bits = div64((uint64_t)pf->samples[index] * s->link.bandwidth, 1000);
474 if (index >= pf->loss_level) {
475 struct dn_pkt_tag *dt = dn_tag_get(m);
477 dt->dn_dir = DIR_DROP;
483 * Send traffic from a scheduler instance due by 'now'.
484 * Return a pointer to the head of the queue.
487 serve_sched(struct mq *q, struct dn_sch_inst *si, uint64_t now)
490 struct dn_schk *s = si->sched;
491 struct mbuf *m = NULL;
492 int delay_line_idle = (si->dline.mq.head == NULL);
500 bw = s->link.bandwidth;
501 si->kflags &= ~DN_ACTIVE;
504 si->credit += (now - si->sched_time) * bw;
507 si->sched_time = now;
509 while (si->credit >= 0 && (m = s->fp->dequeue(si)) != NULL) {
513 len_scaled = (bw == 0) ? 0 : hz *
514 (m->m_pkthdr.len * 8 + extra_bits(m, s));
515 si->credit -= len_scaled;
516 /* Move packet in the delay line */
517 dn_tag_get(m)->output_time = dn_cfg.curr_time + s->link.delay ;
518 mq_append(&si->dline.mq, m);
522 * If credit >= 0 the instance is idle, mark time.
523 * Otherwise put back in the heap, and adjust the output
524 * time of the last inserted packet, m, which was too early.
526 if (si->credit >= 0) {
530 KASSERT (bw > 0, ("bw=0 and credit<0 ?"));
531 t = div64(bw - 1 - si->credit, bw);
533 dn_tag_get(m)->output_time += t;
534 si->kflags |= DN_ACTIVE;
535 heap_insert(&dn_cfg.evheap, now + t, si);
537 if (delay_line_idle && done)
538 transmit_event(q, &si->dline, now);
543 * The timer handler for dummynet. Time is computed in ticks, but
544 * but the code is tolerant to the actual rate at which this is called.
545 * Once complete, the function reschedules itself for the next tick.
548 dummynet_task(void *context, int pending)
551 struct mq q = { NULL, NULL }; /* queue to accumulate results */
553 CURVNET_SET((struct vnet *)context);
557 /* Update number of lost(coalesced) ticks. */
558 tick_lost += pending - 1;
561 /* Last tick duration (usec). */
562 tick_last = (t.tv_sec - dn_cfg.prev_t.tv_sec) * 1000000 +
563 (t.tv_usec - dn_cfg.prev_t.tv_usec);
564 /* Last tick vs standard tick difference (usec). */
565 tick_delta = (tick_last * hz - 1000000) / hz;
566 /* Accumulated tick difference (usec). */
567 tick_delta_sum += tick_delta;
572 * Adjust curr_time if the accumulated tick difference is
573 * greater than the 'standard' tick. Since curr_time should
574 * be monotonically increasing, we do positive adjustments
575 * as required, and throttle curr_time in case of negative
579 if (tick_delta_sum - tick >= 0) {
580 int diff = tick_delta_sum / tick;
582 dn_cfg.curr_time += diff;
584 tick_delta_sum %= tick;
586 } else if (tick_delta_sum + tick <= 0) {
589 tick_delta_sum += tick;
593 /* serve pending events, accumulate in q */
595 struct dn_id *p; /* generic parameter to handler */
597 if (dn_cfg.evheap.elements == 0 ||
598 DN_KEY_LT(dn_cfg.curr_time, HEAP_TOP(&dn_cfg.evheap)->key))
600 p = HEAP_TOP(&dn_cfg.evheap)->object;
601 heap_extract(&dn_cfg.evheap, NULL);
603 if (p->type == DN_SCH_I) {
604 serve_sched(&q, (struct dn_sch_inst *)p, dn_cfg.curr_time);
605 } else { /* extracted a delay line */
606 transmit_event(&q, (struct delay_line *)p, dn_cfg.curr_time);
609 if (dn_cfg.expire && ++dn_cfg.expire_cycle >= dn_cfg.expire) {
610 dn_cfg.expire_cycle = 0;
611 dn_drain_scheduler();
618 dummynet_send(q.head);
623 * forward a chain of packets to the proper destination.
624 * This runs outside the dummynet lock.
627 dummynet_send(struct mbuf *m)
631 for (; m != NULL; m = n) {
632 struct ifnet *ifp = NULL; /* gcc 3.4.6 complains */
638 tag = m_tag_first(m);
639 if (tag == NULL) { /* should not happen */
642 struct dn_pkt_tag *pkt = dn_tag_get(m);
643 /* extract the dummynet info, rename the tag
644 * to carry reinject info.
648 tag->m_tag_cookie = MTAG_IPFW_RULE;
654 SET_HOST_IPLEN(mtod(m, struct ip *));
655 ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
659 /* put header in network format for ip_input() */
660 //SET_NET_IPLEN(mtod(m, struct ip *));
661 netisr_dispatch(NETISR_IP, m);
665 case DIR_IN | PROTO_IPV6:
666 netisr_dispatch(NETISR_IPV6, m);
669 case DIR_OUT | PROTO_IPV6:
670 ip6_output(m, NULL, NULL, IPV6_FORWARDING, NULL, NULL, NULL);
674 case DIR_FWD | PROTO_IFB: /* DN_TO_IFB_FWD: */
675 if (bridge_dn_p != NULL)
676 ((*bridge_dn_p)(m, ifp));
678 printf("dummynet: if_bridge not loaded\n");
682 case DIR_IN | PROTO_LAYER2: /* DN_TO_ETH_DEMUX: */
684 * The Ethernet code assumes the Ethernet header is
685 * contiguous in the first mbuf header.
686 * Insure this is true.
688 if (m->m_len < ETHER_HDR_LEN &&
689 (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) {
690 printf("dummynet/ether: pullup failed, "
691 "dropping packet\n");
694 ether_demux(m->m_pkthdr.rcvif, m);
697 case DIR_OUT | PROTO_LAYER2: /* N_TO_ETH_OUT: */
698 ether_output_frame(ifp, m);
702 /* drop the packet after some time */
707 printf("dummynet: bad switch %d!\n", dst);
715 tag_mbuf(struct mbuf *m, int dir, struct ip_fw_args *fwa)
717 struct dn_pkt_tag *dt;
720 mtag = m_tag_get(PACKET_TAG_DUMMYNET,
721 sizeof(*dt), M_NOWAIT | M_ZERO);
723 return 1; /* Cannot allocate packet header. */
724 m_tag_prepend(m, mtag); /* Attach to mbuf chain. */
725 dt = (struct dn_pkt_tag *)(mtag + 1);
726 dt->rule = fwa->rule;
727 dt->rule.info &= IPFW_ONEPASS; /* only keep this info */
730 /* dt->output tame is updated as we move through */
731 dt->output_time = dn_cfg.curr_time;
737 * dummynet hook for packets.
738 * We use the argument to locate the flowset fs and the sched_set sch
739 * associated to it. The we apply flow_mask and sched_mask to
740 * determine the queue and scheduler instances.
742 * dir where shall we send the packet after dummynet.
743 * *m0 the mbuf with the packet
744 * ifp the 'ifp' parameter from the caller.
745 * NULL in ip_input, destination interface in ip_output,
748 dummynet_io(struct mbuf **m0, int dir, struct ip_fw_args *fwa)
750 struct mbuf *m = *m0;
751 struct dn_fsk *fs = NULL;
752 struct dn_sch_inst *si;
753 struct dn_queue *q = NULL; /* default */
755 int fs_id = (fwa->rule.info & IPFW_INFO_MASK) +
756 ((fwa->rule.info & IPFW_IS_PIPE) ? 2*DN_MAX_ID : 0);
759 /* we could actually tag outside the lock, but who cares... */
760 if (tag_mbuf(m, dir, fwa))
763 /* if the upper half is busy doing something expensive,
764 * lets queue the packet and move forward
766 mq_append(&dn_cfg.pending, m);
767 m = *m0 = NULL; /* consumed */
768 goto done; /* already active, nothing to do */
770 /* XXX locate_flowset could be optimised with a direct ref. */
771 fs = dn_ht_find(dn_cfg.fshash, fs_id, 0, NULL);
773 goto dropit; /* This queue/pipe does not exist! */
774 if (fs->sched == NULL) /* should not happen */
776 /* find scheduler instance, possibly applying sched_mask */
777 si = ipdn_si_find(fs->sched, &(fwa->f_id));
781 * If the scheduler supports multiple queues, find the right one
782 * (otherwise it will be ignored by enqueue).
784 if (fs->sched->fp->flags & DN_MULTIQUEUE) {
785 q = ipdn_q_find(fs, si, &(fwa->f_id));
789 if (fs->sched->fp->enqueue(si, q, m)) {
790 /* packet was dropped by enqueue() */
795 if (si->kflags & DN_ACTIVE) {
796 m = *m0 = NULL; /* consumed */
797 goto done; /* already active, nothing to do */
800 /* compute the initial allowance */
801 if (si->idle_time < dn_cfg.curr_time) {
802 /* Do this only on the first packet on an idle pipe */
803 struct dn_link *p = &fs->sched->link;
805 si->sched_time = dn_cfg.curr_time;
806 si->credit = dn_cfg.io_fast ? p->bandwidth : 0;
808 uint64_t burst = (dn_cfg.curr_time - si->idle_time) * p->bandwidth;
809 if (burst > p->burst)
814 /* pass through scheduler and delay line */
815 m = serve_sched(NULL, si, dn_cfg.curr_time);
817 /* optimization -- pass it back to ipfw for immediate send */
818 /* XXX Don't call dummynet_send() if scheduler return the packet
819 * just enqueued. This avoid a lock order reversal.
822 if (/*dn_cfg.io_fast &&*/ m == *m0 && (dir & PROTO_LAYER2) == 0 ) {
823 /* fast io, rename the tag * to carry reinject info. */
824 struct m_tag *tag = m_tag_first(m);
826 tag->m_tag_cookie = MTAG_IPFW_RULE;
829 if (m->m_nextpkt != NULL) {
830 printf("dummynet: fast io: pkt chain detected!\n");
849 return (fs && (fs->fs.flags & DN_NOERROR)) ? 0 : ENOBUFS;