2 * Copyright (c) 2011-2012 Apple Inc. All rights reserved.
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
33 #include <sys/cdefs.h>
34 #include <sys/param.h>
35 #include <sys/malloc.h>
37 #include <sys/systm.h>
38 #include <sys/errno.h>
39 #include <sys/kernel.h>
40 #include <sys/syslog.h>
42 #include <kern/zalloc.h>
45 #include <net/net_osdep.h>
47 #include <net/pktsched/pktsched_tcq.h>
48 #include <netinet/in.h>
53 static int tcq_enqueue_ifclassq(struct ifclassq
*, struct mbuf
*);
54 static struct mbuf
*tcq_dequeue_tc_ifclassq(struct ifclassq
*,
55 mbuf_svc_class_t
, cqdq_op_t
);
56 static int tcq_request_ifclassq(struct ifclassq
*, cqrq_t
, void *);
57 static int tcq_clear_interface(struct tcq_if
*);
58 static struct tcq_class
*tcq_class_create(struct tcq_if
*, int, u_int32_t
,
60 static int tcq_class_destroy(struct tcq_if
*, struct tcq_class
*);
61 static int tcq_destroy_locked(struct tcq_if
*);
62 static inline int tcq_addq(struct tcq_class
*, struct mbuf
*,
64 static inline struct mbuf
*tcq_getq(struct tcq_class
*);
65 static inline struct mbuf
*tcq_pollq(struct tcq_class
*);
66 static void tcq_purgeq(struct tcq_if
*, struct tcq_class
*, u_int32_t
,
67 u_int32_t
*, u_int32_t
*);
68 static void tcq_purge_sc(struct tcq_if
*, cqrq_purge_sc_t
*);
69 static void tcq_updateq(struct tcq_if
*, struct tcq_class
*, cqev_t
);
70 static int tcq_throttle(struct tcq_if
*, cqrq_throttle_t
*);
71 static int tcq_resumeq(struct tcq_if
*, struct tcq_class
*);
72 static int tcq_suspendq(struct tcq_if
*, struct tcq_class
*);
73 static struct mbuf
*tcq_dequeue_cl(struct tcq_if
*, struct tcq_class
*,
74 mbuf_svc_class_t
, cqdq_op_t
);
75 static inline struct tcq_class
*tcq_clh_to_clp(struct tcq_if
*, u_int32_t
);
76 static const char *tcq_style(struct tcq_if
*);
78 #define TCQ_ZONE_MAX 32 /* maximum elements in zone */
79 #define TCQ_ZONE_NAME "pktsched_tcq" /* zone name */
81 static unsigned int tcq_size
; /* size of zone element */
82 static struct zone
*tcq_zone
; /* zone for tcq */
84 #define TCQ_CL_ZONE_MAX 32 /* maximum elements in zone */
85 #define TCQ_CL_ZONE_NAME "pktsched_tcq_cl" /* zone name */
87 static unsigned int tcq_cl_size
; /* size of zone element */
88 static struct zone
*tcq_cl_zone
; /* zone for tcq_class */
93 tcq_size
= sizeof (struct tcq_if
);
94 tcq_zone
= zinit(tcq_size
, TCQ_ZONE_MAX
* tcq_size
,
96 if (tcq_zone
== NULL
) {
97 panic("%s: failed allocating %s", __func__
, TCQ_ZONE_NAME
);
100 zone_change(tcq_zone
, Z_EXPAND
, TRUE
);
101 zone_change(tcq_zone
, Z_CALLERACCT
, TRUE
);
103 tcq_cl_size
= sizeof (struct tcq_class
);
104 tcq_cl_zone
= zinit(tcq_cl_size
, TCQ_CL_ZONE_MAX
* tcq_cl_size
,
105 0, TCQ_CL_ZONE_NAME
);
106 if (tcq_cl_zone
== NULL
) {
107 panic("%s: failed allocating %s", __func__
, TCQ_CL_ZONE_NAME
);
110 zone_change(tcq_cl_zone
, Z_EXPAND
, TRUE
);
111 zone_change(tcq_cl_zone
, Z_CALLERACCT
, TRUE
);
115 tcq_alloc(struct ifnet
*ifp
, int how
, boolean_t altq
)
119 tif
= (how
== M_WAITOK
) ? zalloc(tcq_zone
) : zalloc_noblock(tcq_zone
);
123 bzero(tif
, tcq_size
);
124 tif
->tif_maxpri
= -1;
125 tif
->tif_ifq
= &ifp
->if_snd
;
127 tif
->tif_flags
|= TCQIFF_ALTQ
;
129 if (pktsched_verbose
) {
130 log(LOG_DEBUG
, "%s: %s scheduler allocated\n",
131 if_name(ifp
), tcq_style(tif
));
138 tcq_destroy(struct tcq_if
*tif
)
140 struct ifclassq
*ifq
= tif
->tif_ifq
;
144 err
= tcq_destroy_locked(tif
);
151 tcq_destroy_locked(struct tcq_if
*tif
)
153 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
155 (void) tcq_clear_interface(tif
);
157 if (pktsched_verbose
) {
158 log(LOG_DEBUG
, "%s: %s scheduler destroyed\n",
159 if_name(TCQIF_IFP(tif
)), tcq_style(tif
));
162 zfree(tcq_zone
, tif
);
168 * bring the interface back to the initial state by discarding
169 * all the filters and classes.
172 tcq_clear_interface(struct tcq_if
*tif
)
174 struct tcq_class
*cl
;
177 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
179 /* clear out the classes */
180 for (pri
= 0; pri
<= tif
->tif_maxpri
; pri
++)
181 if ((cl
= tif
->tif_classes
[pri
]) != NULL
)
182 tcq_class_destroy(tif
, cl
);
187 /* discard all the queued packets on the interface */
189 tcq_purge(struct tcq_if
*tif
)
191 struct tcq_class
*cl
;
194 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
196 for (pri
= 0; pri
<= tif
->tif_maxpri
; pri
++) {
197 if ((cl
= tif
->tif_classes
[pri
]) != NULL
&& !qempty(&cl
->cl_q
))
198 tcq_purgeq(tif
, cl
, 0, NULL
, NULL
);
202 * This assertion is safe to be made only when PF_ALTQ is not
203 * configured; otherwise, IFCQ_LEN represents the sum of the
204 * packets managed by ifcq_disc and altq_disc instances, which
205 * is possible when transitioning between the two.
207 VERIFY(IFCQ_LEN(tif
->tif_ifq
) == 0);
208 #endif /* !PF_ALTQ */
212 tcq_purge_sc(struct tcq_if
*tif
, cqrq_purge_sc_t
*pr
)
214 struct ifclassq
*ifq
= tif
->tif_ifq
;
217 IFCQ_LOCK_ASSERT_HELD(ifq
);
219 VERIFY(pr
->sc
== MBUF_SC_UNSPEC
|| MBUF_VALID_SC(pr
->sc
));
220 VERIFY(pr
->flow
!= 0);
222 if (pr
->sc
!= MBUF_SC_UNSPEC
) {
223 i
= MBUF_SCIDX(pr
->sc
);
224 VERIFY(i
< IFCQ_SC_MAX
);
226 tcq_purgeq(tif
, ifq
->ifcq_disc_slots
[i
].cl
,
227 pr
->flow
, &pr
->packets
, &pr
->bytes
);
234 for (i
= 0; i
< IFCQ_SC_MAX
; i
++) {
235 tcq_purgeq(tif
, ifq
->ifcq_disc_slots
[i
].cl
,
236 pr
->flow
, &cnt
, &len
);
244 tcq_event(struct tcq_if
*tif
, cqev_t ev
)
246 struct tcq_class
*cl
;
249 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
251 for (pri
= 0; pri
<= tif
->tif_maxpri
; pri
++)
252 if ((cl
= tif
->tif_classes
[pri
]) != NULL
)
253 tcq_updateq(tif
, cl
, ev
);
257 tcq_add_queue(struct tcq_if
*tif
, int priority
, u_int32_t qlimit
,
258 int flags
, u_int32_t qid
, struct tcq_class
**clp
)
260 struct tcq_class
*cl
;
262 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
264 /* check parameters */
265 if (priority
>= TCQ_MAXPRI
)
267 if (tif
->tif_classes
[priority
] != NULL
)
269 if (tcq_clh_to_clp(tif
, qid
) != NULL
)
272 cl
= tcq_class_create(tif
, priority
, qlimit
, flags
, qid
);
282 static struct tcq_class
*
283 tcq_class_create(struct tcq_if
*tif
, int pri
, u_int32_t qlimit
,
284 int flags
, u_int32_t qid
)
287 struct ifclassq
*ifq
;
288 struct tcq_class
*cl
;
290 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
292 /* Sanitize flags unless internally configured */
293 if (tif
->tif_flags
& TCQIFF_ALTQ
)
294 flags
&= TQCF_USERFLAGS
;
297 if (flags
& TQCF_RED
) {
298 log(LOG_ERR
, "%s: %s RED not available!\n",
299 if_name(TCQIF_IFP(tif
)), tcq_style(tif
));
302 #endif /* !CLASSQ_RED */
305 if (flags
& TQCF_RIO
) {
306 log(LOG_ERR
, "%s: %s RIO not available!\n",
307 if_name(TCQIF_IFP(tif
)), tcq_style(tif
));
310 #endif /* CLASSQ_RIO */
313 if (flags
& TQCF_BLUE
) {
314 log(LOG_ERR
, "%s: %s BLUE not available!\n",
315 if_name(TCQIF_IFP(tif
)), tcq_style(tif
));
318 #endif /* CLASSQ_BLUE */
320 /* These are mutually exclusive */
321 if ((flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) &&
322 (flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) != TQCF_RED
&&
323 (flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) != TQCF_RIO
&&
324 (flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) != TQCF_BLUE
&&
325 (flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) != TQCF_SFB
) {
326 log(LOG_ERR
, "%s: %s more than one RED|RIO|BLUE|SFB\n",
327 if_name(TCQIF_IFP(tif
)), tcq_style(tif
));
332 ifp
= TCQIF_IFP(tif
);
334 if ((cl
= tif
->tif_classes
[pri
]) != NULL
) {
335 /* modify the class instead of creating a new one */
336 if (!qempty(&cl
->cl_q
))
337 tcq_purgeq(tif
, cl
, 0, NULL
, NULL
);
339 if (q_is_rio(&cl
->cl_q
))
340 rio_destroy(cl
->cl_rio
);
341 #endif /* CLASSQ_RIO */
343 if (q_is_red(&cl
->cl_q
))
344 red_destroy(cl
->cl_red
);
345 #endif /* CLASSQ_RED */
347 if (q_is_blue(&cl
->cl_q
))
348 blue_destroy(cl
->cl_blue
);
349 #endif /* CLASSQ_BLUE */
350 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
351 sfb_destroy(cl
->cl_sfb
);
352 cl
->cl_qalg
.ptr
= NULL
;
353 qtype(&cl
->cl_q
) = Q_DROPTAIL
;
354 qstate(&cl
->cl_q
) = QS_RUNNING
;
356 cl
= zalloc(tcq_cl_zone
);
360 bzero(cl
, tcq_cl_size
);
363 tif
->tif_classes
[pri
] = cl
;
364 if (flags
& TQCF_DEFAULTCLASS
)
365 tif
->tif_default
= cl
;
366 if (qlimit
== 0 || qlimit
> IFCQ_MAXLEN(ifq
)) {
367 qlimit
= IFCQ_MAXLEN(ifq
);
369 qlimit
= DEFAULT_QLIMIT
; /* use default */
371 _qinit(&cl
->cl_q
, Q_DROPTAIL
, qlimit
);
372 cl
->cl_flags
= flags
;
374 if (pri
> tif
->tif_maxpri
)
375 tif
->tif_maxpri
= pri
;
379 if (flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) {
380 #if CLASSQ_RED || CLASSQ_RIO
381 u_int64_t ifbandwidth
= ifnet_output_linkrate(ifp
);
383 #endif /* CLASSQ_RED || CLASSQ_RIO */
386 if (flags
& TQCF_ECN
) {
387 if (flags
& TQCF_BLUE
)
388 cl
->cl_qflags
|= BLUEF_ECN
;
389 else if (flags
& TQCF_SFB
)
390 cl
->cl_qflags
|= SFBF_ECN
;
391 else if (flags
& TQCF_RED
)
392 cl
->cl_qflags
|= REDF_ECN
;
393 else if (flags
& TQCF_RIO
)
394 cl
->cl_qflags
|= RIOF_ECN
;
396 if (flags
& TQCF_FLOWCTL
) {
397 if (flags
& TQCF_SFB
)
398 cl
->cl_qflags
|= SFBF_FLOWCTL
;
400 if (flags
& TQCF_CLEARDSCP
) {
401 if (flags
& TQCF_RIO
)
402 cl
->cl_qflags
|= RIOF_CLEARDSCP
;
404 #if CLASSQ_RED || CLASSQ_RIO
406 * XXX: RED & RIO should be watching link speed and MTU
407 * events and recompute pkttime accordingly.
410 pkttime
= 1000 * 1000 * 1000; /* 1 sec */
412 pkttime
= (int64_t)ifp
->if_mtu
* 1000 * 1000 * 1000 /
415 /* Test for exclusivity {RED,RIO,BLUE,SFB} was done above */
417 if (flags
& TQCF_RED
) {
418 cl
->cl_red
= red_alloc(ifp
, 0, 0,
419 qlimit(&cl
->cl_q
) * 10/100,
420 qlimit(&cl
->cl_q
) * 30/100,
421 cl
->cl_qflags
, pkttime
);
422 if (cl
->cl_red
!= NULL
)
423 qtype(&cl
->cl_q
) = Q_RED
;
425 #endif /* CLASSQ_RED */
427 if (flags
& TQCF_RIO
) {
429 rio_alloc(ifp
, 0, NULL
, cl
->cl_qflags
, pkttime
);
430 if (cl
->cl_rio
!= NULL
)
431 qtype(&cl
->cl_q
) = Q_RIO
;
433 #endif /* CLASSQ_RIO */
434 #endif /* CLASSQ_RED || CLASSQ_RIO */
436 if (flags
& TQCF_BLUE
) {
437 cl
->cl_blue
= blue_alloc(ifp
, 0, 0, cl
->cl_qflags
);
438 if (cl
->cl_blue
!= NULL
)
439 qtype(&cl
->cl_q
) = Q_BLUE
;
441 #endif /* CLASSQ_BLUE */
442 if (flags
& TQCF_SFB
) {
443 if (!(cl
->cl_flags
& TQCF_LAZY
))
444 cl
->cl_sfb
= sfb_alloc(ifp
, cl
->cl_handle
,
445 qlimit(&cl
->cl_q
), cl
->cl_qflags
);
446 if (cl
->cl_sfb
!= NULL
|| (cl
->cl_flags
& TQCF_LAZY
))
447 qtype(&cl
->cl_q
) = Q_SFB
;
451 if (pktsched_verbose
) {
452 log(LOG_DEBUG
, "%s: %s created qid=%d pri=%d qlimit=%d "
453 "flags=%b\n", if_name(ifp
), tcq_style(tif
),
454 cl
->cl_handle
, cl
->cl_pri
, qlimit
, flags
, TQCF_BITS
);
461 tcq_remove_queue(struct tcq_if
*tif
, u_int32_t qid
)
463 struct tcq_class
*cl
;
465 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
467 if ((cl
= tcq_clh_to_clp(tif
, qid
)) == NULL
)
470 return (tcq_class_destroy(tif
, cl
));
474 tcq_class_destroy(struct tcq_if
*tif
, struct tcq_class
*cl
)
476 struct ifclassq
*ifq
= tif
->tif_ifq
;
479 IFCQ_LOCK_ASSERT_HELD(ifq
);
481 if (!qempty(&cl
->cl_q
))
482 tcq_purgeq(tif
, cl
, 0, NULL
, NULL
);
484 tif
->tif_classes
[cl
->cl_pri
] = NULL
;
485 if (tif
->tif_maxpri
== cl
->cl_pri
) {
486 for (pri
= cl
->cl_pri
; pri
>= 0; pri
--)
487 if (tif
->tif_classes
[pri
] != NULL
) {
488 tif
->tif_maxpri
= pri
;
492 tif
->tif_maxpri
= -1;
495 if (tif
->tif_default
== cl
)
496 tif
->tif_default
= NULL
;
498 if (cl
->cl_qalg
.ptr
!= NULL
) {
500 if (q_is_rio(&cl
->cl_q
))
501 rio_destroy(cl
->cl_rio
);
502 #endif /* CLASSQ_RIO */
504 if (q_is_red(&cl
->cl_q
))
505 red_destroy(cl
->cl_red
);
506 #endif /* CLASSQ_RED */
508 if (q_is_blue(&cl
->cl_q
))
509 blue_destroy(cl
->cl_blue
);
510 #endif /* CLASSQ_BLUE */
511 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
512 sfb_destroy(cl
->cl_sfb
);
513 cl
->cl_qalg
.ptr
= NULL
;
514 qtype(&cl
->cl_q
) = Q_DROPTAIL
;
515 qstate(&cl
->cl_q
) = QS_RUNNING
;
518 if (pktsched_verbose
) {
519 log(LOG_DEBUG
, "%s: %s destroyed qid=%d pri=%d\n",
520 if_name(TCQIF_IFP(tif
)), tcq_style(tif
),
521 cl
->cl_handle
, cl
->cl_pri
);
524 zfree(tcq_cl_zone
, cl
);
529 tcq_enqueue(struct tcq_if
*tif
, struct tcq_class
*cl
, struct mbuf
*m
,
532 struct ifclassq
*ifq
= tif
->tif_ifq
;
535 IFCQ_LOCK_ASSERT_HELD(ifq
);
536 VERIFY(cl
== NULL
|| cl
->cl_tif
== tif
);
539 cl
= tcq_clh_to_clp(tif
, t
->pftag_qid
);
541 cl
= tif
->tif_default
;
543 IFCQ_CONVERT_LOCK(ifq
);
552 ret
= tcq_addq(cl
, m
, t
);
554 if (ret
== CLASSQEQ_SUCCESS_FC
) {
555 /* packet enqueued, return advisory feedback */
558 VERIFY(ret
== CLASSQEQ_DROPPED
||
559 ret
== CLASSQEQ_DROPPED_FC
||
560 ret
== CLASSQEQ_DROPPED_SP
);
561 /* packet has been freed in tcq_addq */
562 PKTCNTR_ADD(&cl
->cl_dropcnt
, 1, len
);
563 IFCQ_DROP_ADD(ifq
, 1, len
);
565 case CLASSQEQ_DROPPED
:
567 case CLASSQEQ_DROPPED_FC
:
569 case CLASSQEQ_DROPPED_SP
:
570 return (EQSUSPENDED
);
577 /* successfully queued. */
582 * note: CLASSQDQ_POLL returns the next packet without removing the packet
583 * from the queue. CLASSQDQ_REMOVE is a normal dequeue operation.
584 * CLASSQDQ_REMOVE must return the same packet if called immediately
585 * after CLASSQDQ_POLL.
588 tcq_dequeue_tc(struct tcq_if
*tif
, mbuf_svc_class_t sc
, cqdq_op_t op
)
590 return (tcq_dequeue_cl(tif
, NULL
, sc
, op
));
594 tcq_dequeue_cl(struct tcq_if
*tif
, struct tcq_class
*cl
,
595 mbuf_svc_class_t sc
, cqdq_op_t op
)
597 struct ifclassq
*ifq
= tif
->tif_ifq
;
600 IFCQ_LOCK_ASSERT_HELD(ifq
);
603 cl
= tcq_clh_to_clp(tif
, MBUF_SCIDX(sc
));
608 if (qempty(&cl
->cl_q
))
611 VERIFY(!IFCQ_IS_EMPTY(ifq
));
613 if (op
== CLASSQDQ_POLL
)
614 return (tcq_pollq(cl
));
619 if (qempty(&cl
->cl_q
))
621 PKTCNTR_ADD(&cl
->cl_xmitcnt
, 1, m_pktlen(m
));
622 IFCQ_XMIT_ADD(ifq
, 1, m_pktlen(m
));
628 tcq_addq(struct tcq_class
*cl
, struct mbuf
*m
, struct pf_mtag
*t
)
630 struct tcq_if
*tif
= cl
->cl_tif
;
631 struct ifclassq
*ifq
= tif
->tif_ifq
;
633 IFCQ_LOCK_ASSERT_HELD(ifq
);
636 if (q_is_rio(&cl
->cl_q
))
637 return (rio_addq(cl
->cl_rio
, &cl
->cl_q
, m
, t
));
639 #endif /* CLASSQ_RIO */
641 if (q_is_red(&cl
->cl_q
))
642 return (red_addq(cl
->cl_red
, &cl
->cl_q
, m
, t
));
644 #endif /* CLASSQ_RED */
646 if (q_is_blue(&cl
->cl_q
))
647 return (blue_addq(cl
->cl_blue
, &cl
->cl_q
, m
, t
));
649 #endif /* CLASSQ_BLUE */
650 if (q_is_sfb(&cl
->cl_q
)) {
651 if (cl
->cl_sfb
== NULL
) {
652 struct ifnet
*ifp
= TCQIF_IFP(tif
);
654 VERIFY(cl
->cl_flags
& TQCF_LAZY
);
655 cl
->cl_flags
&= ~TQCF_LAZY
;
656 IFCQ_CONVERT_LOCK(ifq
);
658 cl
->cl_sfb
= sfb_alloc(ifp
, cl
->cl_handle
,
659 qlimit(&cl
->cl_q
), cl
->cl_qflags
);
660 if (cl
->cl_sfb
== NULL
) {
661 /* fall back to droptail */
662 qtype(&cl
->cl_q
) = Q_DROPTAIL
;
663 cl
->cl_flags
&= ~TQCF_SFB
;
664 cl
->cl_qflags
&= ~(SFBF_ECN
| SFBF_FLOWCTL
);
666 log(LOG_ERR
, "%s: %s SFB lazy allocation "
667 "failed for qid=%d pri=%d, falling back "
668 "to DROPTAIL\n", if_name(ifp
),
669 tcq_style(tif
), cl
->cl_handle
,
671 } else if (tif
->tif_throttle
!= IFNET_THROTTLE_OFF
) {
672 /* if there's pending throttling, set it */
673 cqrq_throttle_t tr
= { 1, tif
->tif_throttle
};
674 int err
= tcq_throttle(tif
, &tr
);
679 tr
.level
= IFNET_THROTTLE_OFF
;
680 (void) tcq_throttle(tif
, &tr
);
684 if (cl
->cl_sfb
!= NULL
)
685 return (sfb_addq(cl
->cl_sfb
, &cl
->cl_q
, m
, t
));
686 } else if (qlen(&cl
->cl_q
) >= qlimit(&cl
->cl_q
)) {
687 IFCQ_CONVERT_LOCK(ifq
);
689 return (CLASSQEQ_DROPPED
);
692 if (cl
->cl_flags
& TQCF_CLEARDSCP
)
693 write_dsfield(m
, t
, 0);
700 static inline struct mbuf
*
701 tcq_getq(struct tcq_class
*cl
)
703 IFCQ_LOCK_ASSERT_HELD(cl
->cl_tif
->tif_ifq
);
706 if (q_is_rio(&cl
->cl_q
))
707 return (rio_getq(cl
->cl_rio
, &cl
->cl_q
));
709 #endif /* CLASSQ_RIO */
711 if (q_is_red(&cl
->cl_q
))
712 return (red_getq(cl
->cl_red
, &cl
->cl_q
));
714 #endif /* CLASSQ_RED */
716 if (q_is_blue(&cl
->cl_q
))
717 return (blue_getq(cl
->cl_blue
, &cl
->cl_q
));
719 #endif /* CLASSQ_BLUE */
720 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
721 return (sfb_getq(cl
->cl_sfb
, &cl
->cl_q
));
723 return (_getq(&cl
->cl_q
));
726 static inline struct mbuf
*
727 tcq_pollq(struct tcq_class
*cl
)
729 IFCQ_LOCK_ASSERT_HELD(cl
->cl_tif
->tif_ifq
);
731 return (qhead(&cl
->cl_q
));
735 tcq_purgeq(struct tcq_if
*tif
, struct tcq_class
*cl
, u_int32_t flow
,
736 u_int32_t
*packets
, u_int32_t
*bytes
)
738 struct ifclassq
*ifq
= tif
->tif_ifq
;
739 u_int32_t cnt
= 0, len
= 0, qlen
;
741 IFCQ_LOCK_ASSERT_HELD(ifq
);
743 if ((qlen
= qlen(&cl
->cl_q
)) == 0)
746 /* become regular mutex before freeing mbufs */
747 IFCQ_CONVERT_LOCK(ifq
);
750 if (q_is_rio(&cl
->cl_q
))
751 rio_purgeq(cl
->cl_rio
, &cl
->cl_q
, flow
, &cnt
, &len
);
753 #endif /* CLASSQ_RIO */
755 if (q_is_red(&cl
->cl_q
))
756 red_purgeq(cl
->cl_red
, &cl
->cl_q
, flow
, &cnt
, &len
);
758 #endif /* CLASSQ_RED */
760 if (q_is_blue(&cl
->cl_q
))
761 blue_purgeq(cl
->cl_blue
, &cl
->cl_q
, flow
, &cnt
, &len
);
763 #endif /* CLASSQ_BLUE */
764 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
765 sfb_purgeq(cl
->cl_sfb
, &cl
->cl_q
, flow
, &cnt
, &len
);
767 _flushq_flow(&cl
->cl_q
, flow
, &cnt
, &len
);
770 VERIFY(qlen(&cl
->cl_q
) == (qlen
- cnt
));
772 PKTCNTR_ADD(&cl
->cl_dropcnt
, cnt
, len
);
773 IFCQ_DROP_ADD(ifq
, cnt
, len
);
775 VERIFY(((signed)IFCQ_LEN(ifq
) - cnt
) >= 0);
776 IFCQ_LEN(ifq
) -= cnt
;
778 if (pktsched_verbose
) {
779 log(LOG_DEBUG
, "%s: %s purge qid=%d pri=%d "
780 "qlen=[%d,%d] cnt=%d len=%d flow=0x%x\n",
781 if_name(TCQIF_IFP(tif
)), tcq_style(tif
),
782 cl
->cl_handle
, cl
->cl_pri
, qlen
, qlen(&cl
->cl_q
),
794 tcq_updateq(struct tcq_if
*tif
, struct tcq_class
*cl
, cqev_t ev
)
796 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
798 if (pktsched_verbose
) {
799 log(LOG_DEBUG
, "%s: %s update qid=%d pri=%d event=%s\n",
800 if_name(TCQIF_IFP(tif
)), tcq_style(tif
),
801 cl
->cl_handle
, cl
->cl_pri
, ifclassq_ev2str(ev
));
805 if (q_is_rio(&cl
->cl_q
))
806 return (rio_updateq(cl
->cl_rio
, ev
));
807 #endif /* CLASSQ_RIO */
809 if (q_is_red(&cl
->cl_q
))
810 return (red_updateq(cl
->cl_red
, ev
));
811 #endif /* CLASSQ_RED */
813 if (q_is_blue(&cl
->cl_q
))
814 return (blue_updateq(cl
->cl_blue
, ev
));
815 #endif /* CLASSQ_BLUE */
816 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
817 return (sfb_updateq(cl
->cl_sfb
, ev
));
821 tcq_get_class_stats(struct tcq_if
*tif
, u_int32_t qid
,
822 struct tcq_classstats
*sp
)
824 struct tcq_class
*cl
;
826 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
828 if ((cl
= tcq_clh_to_clp(tif
, qid
)) == NULL
)
831 sp
->class_handle
= cl
->cl_handle
;
832 sp
->priority
= cl
->cl_pri
;
833 sp
->qlength
= qlen(&cl
->cl_q
);
834 sp
->qlimit
= qlimit(&cl
->cl_q
);
835 sp
->period
= cl
->cl_period
;
836 sp
->xmitcnt
= cl
->cl_xmitcnt
;
837 sp
->dropcnt
= cl
->cl_dropcnt
;
839 sp
->qtype
= qtype(&cl
->cl_q
);
840 sp
->qstate
= qstate(&cl
->cl_q
);
842 if (q_is_red(&cl
->cl_q
))
843 red_getstats(cl
->cl_red
, &sp
->red
[0]);
844 #endif /* CLASSQ_RED */
846 if (q_is_rio(&cl
->cl_q
))
847 rio_getstats(cl
->cl_rio
, &sp
->red
[0]);
848 #endif /* CLASSQ_RIO */
850 if (q_is_blue(&cl
->cl_q
))
851 blue_getstats(cl
->cl_blue
, &sp
->blue
);
852 #endif /* CLASSQ_BLUE */
853 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
854 sfb_getstats(cl
->cl_sfb
, &sp
->sfb
);
859 /* convert a class handle to the corresponding class pointer */
860 static inline struct tcq_class
*
861 tcq_clh_to_clp(struct tcq_if
*tif
, u_int32_t chandle
)
863 struct tcq_class
*cl
;
866 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
868 for (idx
= tif
->tif_maxpri
; idx
>= 0; idx
--)
869 if ((cl
= tif
->tif_classes
[idx
]) != NULL
&&
870 cl
->cl_handle
== chandle
)
877 tcq_style(struct tcq_if
*tif
)
879 return ((tif
->tif_flags
& TCQIFF_ALTQ
) ? "ALTQ_TCQ" : "TCQ");
883 * tcq_enqueue_ifclassq is an enqueue function to be registered to
884 * (*ifcq_enqueue) in struct ifclassq.
887 tcq_enqueue_ifclassq(struct ifclassq
*ifq
, struct mbuf
*m
)
891 IFCQ_LOCK_ASSERT_HELD(ifq
);
893 if (!(m
->m_flags
& M_PKTHDR
)) {
894 /* should not happen */
895 log(LOG_ERR
, "%s: packet does not have pkthdr\n",
896 if_name(ifq
->ifcq_ifp
));
897 IFCQ_CONVERT_LOCK(ifq
);
902 i
= MBUF_SCIDX(mbuf_get_service_class(m
));
903 VERIFY((u_int32_t
)i
< IFCQ_SC_MAX
);
905 return (tcq_enqueue(ifq
->ifcq_disc
,
906 ifq
->ifcq_disc_slots
[i
].cl
, m
, m_pftag(m
)));
910 * tcq_dequeue_tc_ifclassq is a dequeue function to be registered to
911 * (*ifcq_dequeue) in struct ifclass.
913 * note: CLASSQDQ_POLL returns the next packet without removing the packet
914 * from the queue. CLASSQDQ_REMOVE is a normal dequeue operation.
915 * CLASSQDQ_REMOVE must return the same packet if called immediately
916 * after CLASSQDQ_POLL.
919 tcq_dequeue_tc_ifclassq(struct ifclassq
*ifq
, mbuf_svc_class_t sc
,
922 u_int32_t i
= MBUF_SCIDX(sc
);
924 VERIFY((u_int32_t
)i
< IFCQ_SC_MAX
);
926 return (tcq_dequeue_cl(ifq
->ifcq_disc
,
927 ifq
->ifcq_disc_slots
[i
].cl
, sc
, op
));
931 tcq_request_ifclassq(struct ifclassq
*ifq
, cqrq_t req
, void *arg
)
933 struct tcq_if
*tif
= (struct tcq_if
*)ifq
->ifcq_disc
;
936 IFCQ_LOCK_ASSERT_HELD(ifq
);
943 case CLASSQRQ_PURGE_SC
:
944 tcq_purge_sc(tif
, (cqrq_purge_sc_t
*)arg
);
948 tcq_event(tif
, (cqev_t
)arg
);
951 case CLASSQRQ_THROTTLE
:
952 err
= tcq_throttle(tif
, (cqrq_throttle_t
*)arg
);
959 tcq_setup_ifclassq(struct ifclassq
*ifq
, u_int32_t flags
)
961 struct ifnet
*ifp
= ifq
->ifcq_ifp
;
962 struct tcq_class
*cl0
, *cl1
, *cl2
, *cl3
;
964 u_int32_t maxlen
= 0, qflags
= 0;
967 IFCQ_LOCK_ASSERT_HELD(ifq
);
968 VERIFY(ifq
->ifcq_disc
== NULL
);
969 VERIFY(ifq
->ifcq_type
== PKTSCHEDT_NONE
);
971 if (flags
& PKTSCHEDF_QALG_RED
)
973 if (flags
& PKTSCHEDF_QALG_RIO
)
975 if (flags
& PKTSCHEDF_QALG_BLUE
)
977 if (flags
& PKTSCHEDF_QALG_SFB
)
979 if (flags
& PKTSCHEDF_QALG_ECN
)
981 if (flags
& PKTSCHEDF_QALG_FLOWCTL
)
982 qflags
|= TQCF_FLOWCTL
;
984 tif
= tcq_alloc(ifp
, M_WAITOK
, FALSE
);
988 if ((maxlen
= IFCQ_MAXLEN(ifq
)) == 0)
989 maxlen
= if_sndq_maxlen
;
991 if ((err
= tcq_add_queue(tif
, 0, maxlen
,
992 qflags
| PRCF_LAZY
, SCIDX_BK
, &cl0
)) != 0)
995 if ((err
= tcq_add_queue(tif
, 1, maxlen
,
996 qflags
| TQCF_DEFAULTCLASS
, SCIDX_BE
, &cl1
)) != 0)
999 if ((err
= tcq_add_queue(tif
, 2, maxlen
,
1000 qflags
| PRCF_LAZY
, SCIDX_VI
, &cl2
)) != 0)
1003 if ((err
= tcq_add_queue(tif
, 3, maxlen
,
1004 qflags
, SCIDX_VO
, &cl3
)) != 0)
1007 err
= ifclassq_attach(ifq
, PKTSCHEDT_TCQ
, tif
,
1008 tcq_enqueue_ifclassq
, NULL
, tcq_dequeue_tc_ifclassq
,
1009 tcq_request_ifclassq
);
1011 /* cache these for faster lookup */
1013 /* Map {BK_SYS,BK} to TC_BK */
1014 ifq
->ifcq_disc_slots
[SCIDX_BK_SYS
].qid
= SCIDX_BK
;
1015 ifq
->ifcq_disc_slots
[SCIDX_BK_SYS
].cl
= cl0
;
1017 ifq
->ifcq_disc_slots
[SCIDX_BK
].qid
= SCIDX_BK
;
1018 ifq
->ifcq_disc_slots
[SCIDX_BK
].cl
= cl0
;
1020 /* Map {BE,RD,OAM} to TC_BE */
1021 ifq
->ifcq_disc_slots
[SCIDX_BE
].qid
= SCIDX_BE
;
1022 ifq
->ifcq_disc_slots
[SCIDX_BE
].cl
= cl1
;
1024 ifq
->ifcq_disc_slots
[SCIDX_RD
].qid
= SCIDX_BE
;
1025 ifq
->ifcq_disc_slots
[SCIDX_RD
].cl
= cl1
;
1027 ifq
->ifcq_disc_slots
[SCIDX_OAM
].qid
= SCIDX_BE
;
1028 ifq
->ifcq_disc_slots
[SCIDX_OAM
].cl
= cl1
;
1030 /* Map {AV,RV,VI} to TC_VI */
1031 ifq
->ifcq_disc_slots
[SCIDX_AV
].qid
= SCIDX_VI
;
1032 ifq
->ifcq_disc_slots
[SCIDX_AV
].cl
= cl2
;
1034 ifq
->ifcq_disc_slots
[SCIDX_RV
].qid
= SCIDX_VI
;
1035 ifq
->ifcq_disc_slots
[SCIDX_RV
].cl
= cl2
;
1037 ifq
->ifcq_disc_slots
[SCIDX_VI
].qid
= SCIDX_VI
;
1038 ifq
->ifcq_disc_slots
[SCIDX_VI
].cl
= cl2
;
1040 /* Map {VO,CTL} to TC_VO */
1041 ifq
->ifcq_disc_slots
[SCIDX_VO
].qid
= SCIDX_VO
;
1042 ifq
->ifcq_disc_slots
[SCIDX_VO
].cl
= cl3
;
1044 ifq
->ifcq_disc_slots
[SCIDX_CTL
].qid
= SCIDX_VO
;
1045 ifq
->ifcq_disc_slots
[SCIDX_CTL
].cl
= cl3
;
1050 (void) tcq_destroy_locked(tif
);
1056 tcq_teardown_ifclassq(struct ifclassq
*ifq
)
1058 struct tcq_if
*tif
= ifq
->ifcq_disc
;
1061 IFCQ_LOCK_ASSERT_HELD(ifq
);
1062 VERIFY(tif
!= NULL
&& ifq
->ifcq_type
== PKTSCHEDT_TCQ
);
1064 (void) tcq_destroy_locked(tif
);
1066 ifq
->ifcq_disc
= NULL
;
1067 for (i
= 0; i
< IFCQ_SC_MAX
; i
++) {
1068 ifq
->ifcq_disc_slots
[i
].qid
= 0;
1069 ifq
->ifcq_disc_slots
[i
].cl
= NULL
;
1072 return (ifclassq_detach(ifq
));
1076 tcq_getqstats_ifclassq(struct ifclassq
*ifq
, u_int32_t slot
,
1077 struct if_ifclassq_stats
*ifqs
)
1079 struct tcq_if
*tif
= ifq
->ifcq_disc
;
1081 IFCQ_LOCK_ASSERT_HELD(ifq
);
1082 VERIFY(ifq
->ifcq_type
== PKTSCHEDT_TCQ
);
1084 if (slot
>= IFCQ_SC_MAX
)
1087 return (tcq_get_class_stats(tif
, ifq
->ifcq_disc_slots
[slot
].qid
,
1088 &ifqs
->ifqs_tcq_stats
));
1092 tcq_throttle(struct tcq_if
*tif
, cqrq_throttle_t
*tr
)
1094 struct ifclassq
*ifq
= tif
->tif_ifq
;
1095 struct tcq_class
*cl
;
1098 IFCQ_LOCK_ASSERT_HELD(ifq
);
1099 VERIFY(!(tif
->tif_flags
& TCQIFF_ALTQ
));
1102 tr
->level
= tif
->tif_throttle
;
1106 if (tr
->level
== tif
->tif_throttle
)
1109 /* Current throttling levels only involve BK_SYS class */
1110 cl
= ifq
->ifcq_disc_slots
[SCIDX_BK_SYS
].cl
;
1112 switch (tr
->level
) {
1113 case IFNET_THROTTLE_OFF
:
1114 err
= tcq_resumeq(tif
, cl
);
1117 case IFNET_THROTTLE_OPPORTUNISTIC
:
1118 err
= tcq_suspendq(tif
, cl
);
1126 if (err
== 0 || err
== ENXIO
) {
1127 if (pktsched_verbose
) {
1128 log(LOG_DEBUG
, "%s: %s throttling %slevel set %d->%d\n",
1129 if_name(TCQIF_IFP(tif
)), tcq_style(tif
),
1130 (err
== 0) ? "" : "lazy ", tif
->tif_throttle
,
1133 tif
->tif_throttle
= tr
->level
;
1137 tcq_purgeq(tif
, cl
, 0, NULL
, NULL
);
1139 log(LOG_ERR
, "%s: %s unable to set throttling level "
1140 "%d->%d [error=%d]\n", if_name(TCQIF_IFP(tif
)),
1141 tcq_style(tif
), tif
->tif_throttle
, tr
->level
, err
);
1148 tcq_resumeq(struct tcq_if
*tif
, struct tcq_class
*cl
)
1150 struct ifclassq
*ifq
= tif
->tif_ifq
;
1153 IFCQ_LOCK_ASSERT_HELD(ifq
);
1156 if (q_is_rio(&cl
->cl_q
))
1157 err
= rio_suspendq(cl
->cl_rio
, &cl
->cl_q
, FALSE
);
1159 #endif /* CLASSQ_RIO */
1161 if (q_is_red(&cl
->cl_q
))
1162 err
= red_suspendq(cl
->cl_red
, &cl
->cl_q
, FALSE
);
1164 #endif /* CLASSQ_RED */
1166 if (q_is_blue(&cl
->cl_q
))
1167 err
= blue_suspendq(cl
->cl_blue
, &cl
->cl_q
, FALSE
);
1169 #endif /* CLASSQ_BLUE */
1170 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
1171 err
= sfb_suspendq(cl
->cl_sfb
, &cl
->cl_q
, FALSE
);
1174 qstate(&cl
->cl_q
) = QS_RUNNING
;
1180 tcq_suspendq(struct tcq_if
*tif
, struct tcq_class
*cl
)
1182 struct ifclassq
*ifq
= tif
->tif_ifq
;
1185 IFCQ_LOCK_ASSERT_HELD(ifq
);
1188 if (q_is_rio(&cl
->cl_q
))
1189 err
= rio_suspendq(cl
->cl_rio
, &cl
->cl_q
, TRUE
);
1191 #endif /* CLASSQ_RIO */
1193 if (q_is_red(&cl
->cl_q
))
1194 err
= red_suspendq(cl
->cl_red
, &cl
->cl_q
, TRUE
);
1196 #endif /* CLASSQ_RED */
1198 if (q_is_blue(&cl
->cl_q
))
1199 err
= blue_suspendq(cl
->cl_blue
, &cl
->cl_q
, TRUE
);
1201 #endif /* CLASSQ_BLUE */
1202 if (q_is_sfb(&cl
->cl_q
)) {
1203 if (cl
->cl_sfb
!= NULL
) {
1204 err
= sfb_suspendq(cl
->cl_sfb
, &cl
->cl_q
, TRUE
);
1206 VERIFY(cl
->cl_flags
& TQCF_LAZY
);
1207 err
= ENXIO
; /* delayed throttling */
1211 if (err
== 0 || err
== ENXIO
)
1212 qstate(&cl
->cl_q
) = QS_SUSPENDED
;