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 int tcq_stat_sc(struct tcq_if
*, cqrq_stat_sc_t
*);
74 static struct mbuf
*tcq_dequeue_cl(struct tcq_if
*, struct tcq_class
*,
75 mbuf_svc_class_t
, cqdq_op_t
);
76 static inline struct tcq_class
*tcq_clh_to_clp(struct tcq_if
*, u_int32_t
);
77 static const char *tcq_style(struct tcq_if
*);
79 #define TCQ_ZONE_MAX 32 /* maximum elements in zone */
80 #define TCQ_ZONE_NAME "pktsched_tcq" /* zone name */
82 static unsigned int tcq_size
; /* size of zone element */
83 static struct zone
*tcq_zone
; /* zone for tcq */
85 #define TCQ_CL_ZONE_MAX 32 /* maximum elements in zone */
86 #define TCQ_CL_ZONE_NAME "pktsched_tcq_cl" /* zone name */
88 static unsigned int tcq_cl_size
; /* size of zone element */
89 static struct zone
*tcq_cl_zone
; /* zone for tcq_class */
94 tcq_size
= sizeof (struct tcq_if
);
95 tcq_zone
= zinit(tcq_size
, TCQ_ZONE_MAX
* tcq_size
,
97 if (tcq_zone
== NULL
) {
98 panic("%s: failed allocating %s", __func__
, TCQ_ZONE_NAME
);
101 zone_change(tcq_zone
, Z_EXPAND
, TRUE
);
102 zone_change(tcq_zone
, Z_CALLERACCT
, TRUE
);
104 tcq_cl_size
= sizeof (struct tcq_class
);
105 tcq_cl_zone
= zinit(tcq_cl_size
, TCQ_CL_ZONE_MAX
* tcq_cl_size
,
106 0, TCQ_CL_ZONE_NAME
);
107 if (tcq_cl_zone
== NULL
) {
108 panic("%s: failed allocating %s", __func__
, TCQ_CL_ZONE_NAME
);
111 zone_change(tcq_cl_zone
, Z_EXPAND
, TRUE
);
112 zone_change(tcq_cl_zone
, Z_CALLERACCT
, TRUE
);
116 tcq_alloc(struct ifnet
*ifp
, int how
, boolean_t altq
)
120 tif
= (how
== M_WAITOK
) ? zalloc(tcq_zone
) : zalloc_noblock(tcq_zone
);
124 bzero(tif
, tcq_size
);
125 tif
->tif_maxpri
= -1;
126 tif
->tif_ifq
= &ifp
->if_snd
;
128 tif
->tif_flags
|= TCQIFF_ALTQ
;
130 if (pktsched_verbose
) {
131 log(LOG_DEBUG
, "%s: %s scheduler allocated\n",
132 if_name(ifp
), tcq_style(tif
));
139 tcq_destroy(struct tcq_if
*tif
)
141 struct ifclassq
*ifq
= tif
->tif_ifq
;
145 err
= tcq_destroy_locked(tif
);
152 tcq_destroy_locked(struct tcq_if
*tif
)
154 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
156 (void) tcq_clear_interface(tif
);
158 if (pktsched_verbose
) {
159 log(LOG_DEBUG
, "%s: %s scheduler destroyed\n",
160 if_name(TCQIF_IFP(tif
)), tcq_style(tif
));
163 zfree(tcq_zone
, tif
);
169 * bring the interface back to the initial state by discarding
170 * all the filters and classes.
173 tcq_clear_interface(struct tcq_if
*tif
)
175 struct tcq_class
*cl
;
178 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
180 /* clear out the classes */
181 for (pri
= 0; pri
<= tif
->tif_maxpri
; pri
++)
182 if ((cl
= tif
->tif_classes
[pri
]) != NULL
)
183 tcq_class_destroy(tif
, cl
);
188 /* discard all the queued packets on the interface */
190 tcq_purge(struct tcq_if
*tif
)
192 struct tcq_class
*cl
;
195 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
197 for (pri
= 0; pri
<= tif
->tif_maxpri
; pri
++) {
198 if ((cl
= tif
->tif_classes
[pri
]) != NULL
&& !qempty(&cl
->cl_q
))
199 tcq_purgeq(tif
, cl
, 0, NULL
, NULL
);
203 * This assertion is safe to be made only when PF_ALTQ is not
204 * configured; otherwise, IFCQ_LEN represents the sum of the
205 * packets managed by ifcq_disc and altq_disc instances, which
206 * is possible when transitioning between the two.
208 VERIFY(IFCQ_LEN(tif
->tif_ifq
) == 0);
209 #endif /* !PF_ALTQ */
213 tcq_purge_sc(struct tcq_if
*tif
, cqrq_purge_sc_t
*pr
)
215 struct ifclassq
*ifq
= tif
->tif_ifq
;
218 IFCQ_LOCK_ASSERT_HELD(ifq
);
220 VERIFY(pr
->sc
== MBUF_SC_UNSPEC
|| MBUF_VALID_SC(pr
->sc
));
221 VERIFY(pr
->flow
!= 0);
223 if (pr
->sc
!= MBUF_SC_UNSPEC
) {
224 i
= MBUF_SCIDX(pr
->sc
);
225 VERIFY(i
< IFCQ_SC_MAX
);
227 tcq_purgeq(tif
, ifq
->ifcq_disc_slots
[i
].cl
,
228 pr
->flow
, &pr
->packets
, &pr
->bytes
);
235 for (i
= 0; i
< IFCQ_SC_MAX
; i
++) {
236 tcq_purgeq(tif
, ifq
->ifcq_disc_slots
[i
].cl
,
237 pr
->flow
, &cnt
, &len
);
245 tcq_event(struct tcq_if
*tif
, cqev_t ev
)
247 struct tcq_class
*cl
;
250 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
252 for (pri
= 0; pri
<= tif
->tif_maxpri
; pri
++)
253 if ((cl
= tif
->tif_classes
[pri
]) != NULL
)
254 tcq_updateq(tif
, cl
, ev
);
258 tcq_add_queue(struct tcq_if
*tif
, int priority
, u_int32_t qlimit
,
259 int flags
, u_int32_t qid
, struct tcq_class
**clp
)
261 struct tcq_class
*cl
;
263 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
265 /* check parameters */
266 if (priority
>= TCQ_MAXPRI
)
268 if (tif
->tif_classes
[priority
] != NULL
)
270 if (tcq_clh_to_clp(tif
, qid
) != NULL
)
273 cl
= tcq_class_create(tif
, priority
, qlimit
, flags
, qid
);
283 static struct tcq_class
*
284 tcq_class_create(struct tcq_if
*tif
, int pri
, u_int32_t qlimit
,
285 int flags
, u_int32_t qid
)
288 struct ifclassq
*ifq
;
289 struct tcq_class
*cl
;
291 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
293 /* Sanitize flags unless internally configured */
294 if (tif
->tif_flags
& TCQIFF_ALTQ
)
295 flags
&= TQCF_USERFLAGS
;
298 if (flags
& TQCF_RED
) {
299 log(LOG_ERR
, "%s: %s RED not available!\n",
300 if_name(TCQIF_IFP(tif
)), tcq_style(tif
));
303 #endif /* !CLASSQ_RED */
306 if (flags
& TQCF_RIO
) {
307 log(LOG_ERR
, "%s: %s RIO not available!\n",
308 if_name(TCQIF_IFP(tif
)), tcq_style(tif
));
311 #endif /* CLASSQ_RIO */
314 if (flags
& TQCF_BLUE
) {
315 log(LOG_ERR
, "%s: %s BLUE not available!\n",
316 if_name(TCQIF_IFP(tif
)), tcq_style(tif
));
319 #endif /* CLASSQ_BLUE */
321 /* These are mutually exclusive */
322 if ((flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) &&
323 (flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) != TQCF_RED
&&
324 (flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) != TQCF_RIO
&&
325 (flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) != TQCF_BLUE
&&
326 (flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) != TQCF_SFB
) {
327 log(LOG_ERR
, "%s: %s more than one RED|RIO|BLUE|SFB\n",
328 if_name(TCQIF_IFP(tif
)), tcq_style(tif
));
333 ifp
= TCQIF_IFP(tif
);
335 if ((cl
= tif
->tif_classes
[pri
]) != NULL
) {
336 /* modify the class instead of creating a new one */
337 if (!qempty(&cl
->cl_q
))
338 tcq_purgeq(tif
, cl
, 0, NULL
, NULL
);
340 if (q_is_rio(&cl
->cl_q
))
341 rio_destroy(cl
->cl_rio
);
342 #endif /* CLASSQ_RIO */
344 if (q_is_red(&cl
->cl_q
))
345 red_destroy(cl
->cl_red
);
346 #endif /* CLASSQ_RED */
348 if (q_is_blue(&cl
->cl_q
))
349 blue_destroy(cl
->cl_blue
);
350 #endif /* CLASSQ_BLUE */
351 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
352 sfb_destroy(cl
->cl_sfb
);
353 cl
->cl_qalg
.ptr
= NULL
;
354 qtype(&cl
->cl_q
) = Q_DROPTAIL
;
355 qstate(&cl
->cl_q
) = QS_RUNNING
;
357 cl
= zalloc(tcq_cl_zone
);
361 bzero(cl
, tcq_cl_size
);
364 tif
->tif_classes
[pri
] = cl
;
365 if (flags
& TQCF_DEFAULTCLASS
)
366 tif
->tif_default
= cl
;
367 if (qlimit
== 0 || qlimit
> IFCQ_MAXLEN(ifq
)) {
368 qlimit
= IFCQ_MAXLEN(ifq
);
370 qlimit
= DEFAULT_QLIMIT
; /* use default */
372 _qinit(&cl
->cl_q
, Q_DROPTAIL
, qlimit
);
373 cl
->cl_flags
= flags
;
375 if (pri
> tif
->tif_maxpri
)
376 tif
->tif_maxpri
= pri
;
380 if (flags
& (TQCF_RED
|TQCF_RIO
|TQCF_BLUE
|TQCF_SFB
)) {
381 #if CLASSQ_RED || CLASSQ_RIO
382 u_int64_t ifbandwidth
= ifnet_output_linkrate(ifp
);
384 #endif /* CLASSQ_RED || CLASSQ_RIO */
387 if (flags
& TQCF_ECN
) {
388 if (flags
& TQCF_BLUE
)
389 cl
->cl_qflags
|= BLUEF_ECN
;
390 else if (flags
& TQCF_SFB
)
391 cl
->cl_qflags
|= SFBF_ECN
;
392 else if (flags
& TQCF_RED
)
393 cl
->cl_qflags
|= REDF_ECN
;
394 else if (flags
& TQCF_RIO
)
395 cl
->cl_qflags
|= RIOF_ECN
;
397 if (flags
& TQCF_FLOWCTL
) {
398 if (flags
& TQCF_SFB
)
399 cl
->cl_qflags
|= SFBF_FLOWCTL
;
401 if (flags
& TQCF_DELAYBASED
) {
402 if (flags
& TQCF_SFB
)
403 cl
->cl_qflags
|= SFBF_DELAYBASED
;
405 if (flags
& TQCF_CLEARDSCP
) {
406 if (flags
& TQCF_RIO
)
407 cl
->cl_qflags
|= RIOF_CLEARDSCP
;
409 #if CLASSQ_RED || CLASSQ_RIO
411 * XXX: RED & RIO should be watching link speed and MTU
412 * events and recompute pkttime accordingly.
415 pkttime
= 1000 * 1000 * 1000; /* 1 sec */
417 pkttime
= (int64_t)ifp
->if_mtu
* 1000 * 1000 * 1000 /
420 /* Test for exclusivity {RED,RIO,BLUE,SFB} was done above */
422 if (flags
& TQCF_RED
) {
423 cl
->cl_red
= red_alloc(ifp
, 0, 0,
424 qlimit(&cl
->cl_q
) * 10/100,
425 qlimit(&cl
->cl_q
) * 30/100,
426 cl
->cl_qflags
, pkttime
);
427 if (cl
->cl_red
!= NULL
)
428 qtype(&cl
->cl_q
) = Q_RED
;
430 #endif /* CLASSQ_RED */
432 if (flags
& TQCF_RIO
) {
434 rio_alloc(ifp
, 0, NULL
, cl
->cl_qflags
, pkttime
);
435 if (cl
->cl_rio
!= NULL
)
436 qtype(&cl
->cl_q
) = Q_RIO
;
438 #endif /* CLASSQ_RIO */
439 #endif /* CLASSQ_RED || CLASSQ_RIO */
441 if (flags
& TQCF_BLUE
) {
442 cl
->cl_blue
= blue_alloc(ifp
, 0, 0, cl
->cl_qflags
);
443 if (cl
->cl_blue
!= NULL
)
444 qtype(&cl
->cl_q
) = Q_BLUE
;
446 #endif /* CLASSQ_BLUE */
447 if (flags
& TQCF_SFB
) {
448 if (!(cl
->cl_flags
& TQCF_LAZY
))
449 cl
->cl_sfb
= sfb_alloc(ifp
, cl
->cl_handle
,
450 qlimit(&cl
->cl_q
), cl
->cl_qflags
);
451 if (cl
->cl_sfb
!= NULL
|| (cl
->cl_flags
& TQCF_LAZY
))
452 qtype(&cl
->cl_q
) = Q_SFB
;
456 if (pktsched_verbose
) {
457 log(LOG_DEBUG
, "%s: %s created qid=%d pri=%d qlimit=%d "
458 "flags=%b\n", if_name(ifp
), tcq_style(tif
),
459 cl
->cl_handle
, cl
->cl_pri
, qlimit
, flags
, TQCF_BITS
);
466 tcq_remove_queue(struct tcq_if
*tif
, u_int32_t qid
)
468 struct tcq_class
*cl
;
470 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
472 if ((cl
= tcq_clh_to_clp(tif
, qid
)) == NULL
)
475 return (tcq_class_destroy(tif
, cl
));
479 tcq_class_destroy(struct tcq_if
*tif
, struct tcq_class
*cl
)
481 struct ifclassq
*ifq
= tif
->tif_ifq
;
484 IFCQ_LOCK_ASSERT_HELD(ifq
);
486 if (!qempty(&cl
->cl_q
))
487 tcq_purgeq(tif
, cl
, 0, NULL
, NULL
);
489 tif
->tif_classes
[cl
->cl_pri
] = NULL
;
490 if (tif
->tif_maxpri
== cl
->cl_pri
) {
491 for (pri
= cl
->cl_pri
; pri
>= 0; pri
--)
492 if (tif
->tif_classes
[pri
] != NULL
) {
493 tif
->tif_maxpri
= pri
;
497 tif
->tif_maxpri
= -1;
500 if (tif
->tif_default
== cl
)
501 tif
->tif_default
= NULL
;
503 if (cl
->cl_qalg
.ptr
!= NULL
) {
505 if (q_is_rio(&cl
->cl_q
))
506 rio_destroy(cl
->cl_rio
);
507 #endif /* CLASSQ_RIO */
509 if (q_is_red(&cl
->cl_q
))
510 red_destroy(cl
->cl_red
);
511 #endif /* CLASSQ_RED */
513 if (q_is_blue(&cl
->cl_q
))
514 blue_destroy(cl
->cl_blue
);
515 #endif /* CLASSQ_BLUE */
516 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
517 sfb_destroy(cl
->cl_sfb
);
518 cl
->cl_qalg
.ptr
= NULL
;
519 qtype(&cl
->cl_q
) = Q_DROPTAIL
;
520 qstate(&cl
->cl_q
) = QS_RUNNING
;
523 if (pktsched_verbose
) {
524 log(LOG_DEBUG
, "%s: %s destroyed qid=%d pri=%d\n",
525 if_name(TCQIF_IFP(tif
)), tcq_style(tif
),
526 cl
->cl_handle
, cl
->cl_pri
);
529 zfree(tcq_cl_zone
, cl
);
534 tcq_enqueue(struct tcq_if
*tif
, struct tcq_class
*cl
, struct mbuf
*m
,
537 struct ifclassq
*ifq
= tif
->tif_ifq
;
540 IFCQ_LOCK_ASSERT_HELD(ifq
);
541 VERIFY(cl
== NULL
|| cl
->cl_tif
== tif
);
545 cl
= tcq_clh_to_clp(tif
, t
->pftag_qid
);
547 cl
= tcq_clh_to_clp(tif
, 0);
548 #endif /* !PF_ALTQ */
550 cl
= tif
->tif_default
;
552 IFCQ_CONVERT_LOCK(ifq
);
561 ret
= tcq_addq(cl
, m
, t
);
563 if (ret
== CLASSQEQ_SUCCESS_FC
) {
564 /* packet enqueued, return advisory feedback */
567 VERIFY(ret
== CLASSQEQ_DROPPED
||
568 ret
== CLASSQEQ_DROPPED_FC
||
569 ret
== CLASSQEQ_DROPPED_SP
);
570 /* packet has been freed in tcq_addq */
571 PKTCNTR_ADD(&cl
->cl_dropcnt
, 1, len
);
572 IFCQ_DROP_ADD(ifq
, 1, len
);
574 case CLASSQEQ_DROPPED
:
576 case CLASSQEQ_DROPPED_FC
:
578 case CLASSQEQ_DROPPED_SP
:
579 return (EQSUSPENDED
);
586 /* successfully queued. */
591 * note: CLASSQDQ_POLL returns the next packet without removing the packet
592 * from the queue. CLASSQDQ_REMOVE is a normal dequeue operation.
593 * CLASSQDQ_REMOVE must return the same packet if called immediately
594 * after CLASSQDQ_POLL.
597 tcq_dequeue_tc(struct tcq_if
*tif
, mbuf_svc_class_t sc
, cqdq_op_t op
)
599 return (tcq_dequeue_cl(tif
, NULL
, sc
, op
));
603 tcq_dequeue_cl(struct tcq_if
*tif
, struct tcq_class
*cl
,
604 mbuf_svc_class_t sc
, cqdq_op_t op
)
606 struct ifclassq
*ifq
= tif
->tif_ifq
;
609 IFCQ_LOCK_ASSERT_HELD(ifq
);
612 cl
= tcq_clh_to_clp(tif
, MBUF_SCIDX(sc
));
617 if (qempty(&cl
->cl_q
))
620 VERIFY(!IFCQ_IS_EMPTY(ifq
));
622 if (op
== CLASSQDQ_POLL
)
623 return (tcq_pollq(cl
));
628 if (qempty(&cl
->cl_q
))
630 PKTCNTR_ADD(&cl
->cl_xmitcnt
, 1, m_pktlen(m
));
631 IFCQ_XMIT_ADD(ifq
, 1, m_pktlen(m
));
637 tcq_addq(struct tcq_class
*cl
, struct mbuf
*m
, struct pf_mtag
*t
)
639 struct tcq_if
*tif
= cl
->cl_tif
;
640 struct ifclassq
*ifq
= tif
->tif_ifq
;
642 IFCQ_LOCK_ASSERT_HELD(ifq
);
645 if (q_is_rio(&cl
->cl_q
))
646 return (rio_addq(cl
->cl_rio
, &cl
->cl_q
, m
, t
));
648 #endif /* CLASSQ_RIO */
650 if (q_is_red(&cl
->cl_q
))
651 return (red_addq(cl
->cl_red
, &cl
->cl_q
, m
, t
));
653 #endif /* CLASSQ_RED */
655 if (q_is_blue(&cl
->cl_q
))
656 return (blue_addq(cl
->cl_blue
, &cl
->cl_q
, m
, t
));
658 #endif /* CLASSQ_BLUE */
659 if (q_is_sfb(&cl
->cl_q
)) {
660 if (cl
->cl_sfb
== NULL
) {
661 struct ifnet
*ifp
= TCQIF_IFP(tif
);
663 VERIFY(cl
->cl_flags
& TQCF_LAZY
);
664 cl
->cl_flags
&= ~TQCF_LAZY
;
665 IFCQ_CONVERT_LOCK(ifq
);
667 cl
->cl_sfb
= sfb_alloc(ifp
, cl
->cl_handle
,
668 qlimit(&cl
->cl_q
), cl
->cl_qflags
);
669 if (cl
->cl_sfb
== NULL
) {
670 /* fall back to droptail */
671 qtype(&cl
->cl_q
) = Q_DROPTAIL
;
672 cl
->cl_flags
&= ~TQCF_SFB
;
673 cl
->cl_qflags
&= ~(SFBF_ECN
| SFBF_FLOWCTL
);
675 log(LOG_ERR
, "%s: %s SFB lazy allocation "
676 "failed for qid=%d pri=%d, falling back "
677 "to DROPTAIL\n", if_name(ifp
),
678 tcq_style(tif
), cl
->cl_handle
,
680 } else if (tif
->tif_throttle
!= IFNET_THROTTLE_OFF
) {
681 /* if there's pending throttling, set it */
682 cqrq_throttle_t tr
= { 1, tif
->tif_throttle
};
683 int err
= tcq_throttle(tif
, &tr
);
688 tr
.level
= IFNET_THROTTLE_OFF
;
689 (void) tcq_throttle(tif
, &tr
);
693 if (cl
->cl_sfb
!= NULL
)
694 return (sfb_addq(cl
->cl_sfb
, &cl
->cl_q
, m
, t
));
695 } else if (qlen(&cl
->cl_q
) >= qlimit(&cl
->cl_q
)) {
696 IFCQ_CONVERT_LOCK(ifq
);
698 return (CLASSQEQ_DROPPED
);
702 if (cl
->cl_flags
& TQCF_CLEARDSCP
)
703 write_dsfield(m
, t
, 0);
711 static inline struct mbuf
*
712 tcq_getq(struct tcq_class
*cl
)
714 IFCQ_LOCK_ASSERT_HELD(cl
->cl_tif
->tif_ifq
);
717 if (q_is_rio(&cl
->cl_q
))
718 return (rio_getq(cl
->cl_rio
, &cl
->cl_q
));
720 #endif /* CLASSQ_RIO */
722 if (q_is_red(&cl
->cl_q
))
723 return (red_getq(cl
->cl_red
, &cl
->cl_q
));
725 #endif /* CLASSQ_RED */
727 if (q_is_blue(&cl
->cl_q
))
728 return (blue_getq(cl
->cl_blue
, &cl
->cl_q
));
730 #endif /* CLASSQ_BLUE */
731 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
732 return (sfb_getq(cl
->cl_sfb
, &cl
->cl_q
));
734 return (_getq(&cl
->cl_q
));
737 static inline struct mbuf
*
738 tcq_pollq(struct tcq_class
*cl
)
740 IFCQ_LOCK_ASSERT_HELD(cl
->cl_tif
->tif_ifq
);
742 return (qhead(&cl
->cl_q
));
746 tcq_purgeq(struct tcq_if
*tif
, struct tcq_class
*cl
, u_int32_t flow
,
747 u_int32_t
*packets
, u_int32_t
*bytes
)
749 struct ifclassq
*ifq
= tif
->tif_ifq
;
750 u_int32_t cnt
= 0, len
= 0, qlen
;
752 IFCQ_LOCK_ASSERT_HELD(ifq
);
754 if ((qlen
= qlen(&cl
->cl_q
)) == 0)
757 /* become regular mutex before freeing mbufs */
758 IFCQ_CONVERT_LOCK(ifq
);
761 if (q_is_rio(&cl
->cl_q
))
762 rio_purgeq(cl
->cl_rio
, &cl
->cl_q
, flow
, &cnt
, &len
);
764 #endif /* CLASSQ_RIO */
766 if (q_is_red(&cl
->cl_q
))
767 red_purgeq(cl
->cl_red
, &cl
->cl_q
, flow
, &cnt
, &len
);
769 #endif /* CLASSQ_RED */
771 if (q_is_blue(&cl
->cl_q
))
772 blue_purgeq(cl
->cl_blue
, &cl
->cl_q
, flow
, &cnt
, &len
);
774 #endif /* CLASSQ_BLUE */
775 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
776 sfb_purgeq(cl
->cl_sfb
, &cl
->cl_q
, flow
, &cnt
, &len
);
778 _flushq_flow(&cl
->cl_q
, flow
, &cnt
, &len
);
781 VERIFY(qlen(&cl
->cl_q
) == (qlen
- cnt
));
783 PKTCNTR_ADD(&cl
->cl_dropcnt
, cnt
, len
);
784 IFCQ_DROP_ADD(ifq
, cnt
, len
);
786 VERIFY(((signed)IFCQ_LEN(ifq
) - cnt
) >= 0);
787 IFCQ_LEN(ifq
) -= cnt
;
789 if (pktsched_verbose
) {
790 log(LOG_DEBUG
, "%s: %s purge qid=%d pri=%d "
791 "qlen=[%d,%d] cnt=%d len=%d flow=0x%x\n",
792 if_name(TCQIF_IFP(tif
)), tcq_style(tif
),
793 cl
->cl_handle
, cl
->cl_pri
, qlen
, qlen(&cl
->cl_q
),
805 tcq_updateq(struct tcq_if
*tif
, struct tcq_class
*cl
, cqev_t ev
)
807 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
809 if (pktsched_verbose
) {
810 log(LOG_DEBUG
, "%s: %s update qid=%d pri=%d event=%s\n",
811 if_name(TCQIF_IFP(tif
)), tcq_style(tif
),
812 cl
->cl_handle
, cl
->cl_pri
, ifclassq_ev2str(ev
));
816 if (q_is_rio(&cl
->cl_q
))
817 return (rio_updateq(cl
->cl_rio
, ev
));
818 #endif /* CLASSQ_RIO */
820 if (q_is_red(&cl
->cl_q
))
821 return (red_updateq(cl
->cl_red
, ev
));
822 #endif /* CLASSQ_RED */
824 if (q_is_blue(&cl
->cl_q
))
825 return (blue_updateq(cl
->cl_blue
, ev
));
826 #endif /* CLASSQ_BLUE */
827 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
828 return (sfb_updateq(cl
->cl_sfb
, ev
));
832 tcq_get_class_stats(struct tcq_if
*tif
, u_int32_t qid
,
833 struct tcq_classstats
*sp
)
835 struct tcq_class
*cl
;
837 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
839 if ((cl
= tcq_clh_to_clp(tif
, qid
)) == NULL
)
842 sp
->class_handle
= cl
->cl_handle
;
843 sp
->priority
= cl
->cl_pri
;
844 sp
->qlength
= qlen(&cl
->cl_q
);
845 sp
->qlimit
= qlimit(&cl
->cl_q
);
846 sp
->period
= cl
->cl_period
;
847 sp
->xmitcnt
= cl
->cl_xmitcnt
;
848 sp
->dropcnt
= cl
->cl_dropcnt
;
850 sp
->qtype
= qtype(&cl
->cl_q
);
851 sp
->qstate
= qstate(&cl
->cl_q
);
853 if (q_is_red(&cl
->cl_q
))
854 red_getstats(cl
->cl_red
, &sp
->red
[0]);
855 #endif /* CLASSQ_RED */
857 if (q_is_rio(&cl
->cl_q
))
858 rio_getstats(cl
->cl_rio
, &sp
->red
[0]);
859 #endif /* CLASSQ_RIO */
861 if (q_is_blue(&cl
->cl_q
))
862 blue_getstats(cl
->cl_blue
, &sp
->blue
);
863 #endif /* CLASSQ_BLUE */
864 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
865 sfb_getstats(cl
->cl_sfb
, &sp
->sfb
);
871 tcq_stat_sc(struct tcq_if
*tif
, cqrq_stat_sc_t
*sr
)
873 struct ifclassq
*ifq
= tif
->tif_ifq
;
874 struct tcq_class
*cl
;
877 IFCQ_LOCK_ASSERT_HELD(ifq
);
879 VERIFY(sr
->sc
== MBUF_SC_UNSPEC
|| MBUF_VALID_SC(sr
->sc
));
881 i
= MBUF_SCIDX(sr
->sc
);
882 VERIFY(i
< IFCQ_SC_MAX
);
884 cl
= ifq
->ifcq_disc_slots
[i
].cl
;
885 sr
->packets
= qlen(&cl
->cl_q
);
886 sr
->bytes
= qsize(&cl
->cl_q
);
891 /* convert a class handle to the corresponding class pointer */
892 static inline struct tcq_class
*
893 tcq_clh_to_clp(struct tcq_if
*tif
, u_int32_t chandle
)
895 struct tcq_class
*cl
;
898 IFCQ_LOCK_ASSERT_HELD(tif
->tif_ifq
);
900 for (idx
= tif
->tif_maxpri
; idx
>= 0; idx
--)
901 if ((cl
= tif
->tif_classes
[idx
]) != NULL
&&
902 cl
->cl_handle
== chandle
)
909 tcq_style(struct tcq_if
*tif
)
911 return ((tif
->tif_flags
& TCQIFF_ALTQ
) ? "ALTQ_TCQ" : "TCQ");
915 * tcq_enqueue_ifclassq is an enqueue function to be registered to
916 * (*ifcq_enqueue) in struct ifclassq.
919 tcq_enqueue_ifclassq(struct ifclassq
*ifq
, struct mbuf
*m
)
923 IFCQ_LOCK_ASSERT_HELD(ifq
);
925 if (!(m
->m_flags
& M_PKTHDR
)) {
926 /* should not happen */
927 log(LOG_ERR
, "%s: packet does not have pkthdr\n",
928 if_name(ifq
->ifcq_ifp
));
929 IFCQ_CONVERT_LOCK(ifq
);
934 i
= MBUF_SCIDX(mbuf_get_service_class(m
));
935 VERIFY((u_int32_t
)i
< IFCQ_SC_MAX
);
937 return (tcq_enqueue(ifq
->ifcq_disc
,
938 ifq
->ifcq_disc_slots
[i
].cl
, m
, m_pftag(m
)));
942 * tcq_dequeue_tc_ifclassq is a dequeue function to be registered to
943 * (*ifcq_dequeue) in struct ifclass.
945 * note: CLASSQDQ_POLL returns the next packet without removing the packet
946 * from the queue. CLASSQDQ_REMOVE is a normal dequeue operation.
947 * CLASSQDQ_REMOVE must return the same packet if called immediately
948 * after CLASSQDQ_POLL.
951 tcq_dequeue_tc_ifclassq(struct ifclassq
*ifq
, mbuf_svc_class_t sc
,
954 u_int32_t i
= MBUF_SCIDX(sc
);
956 VERIFY((u_int32_t
)i
< IFCQ_SC_MAX
);
958 return (tcq_dequeue_cl(ifq
->ifcq_disc
,
959 ifq
->ifcq_disc_slots
[i
].cl
, sc
, op
));
963 tcq_request_ifclassq(struct ifclassq
*ifq
, cqrq_t req
, void *arg
)
965 struct tcq_if
*tif
= (struct tcq_if
*)ifq
->ifcq_disc
;
968 IFCQ_LOCK_ASSERT_HELD(ifq
);
975 case CLASSQRQ_PURGE_SC
:
976 tcq_purge_sc(tif
, (cqrq_purge_sc_t
*)arg
);
980 tcq_event(tif
, (cqev_t
)arg
);
983 case CLASSQRQ_THROTTLE
:
984 err
= tcq_throttle(tif
, (cqrq_throttle_t
*)arg
);
987 case CLASSQRQ_STAT_SC
:
988 err
= tcq_stat_sc(tif
, (cqrq_stat_sc_t
*)arg
);
995 tcq_setup_ifclassq(struct ifclassq
*ifq
, u_int32_t flags
)
997 struct ifnet
*ifp
= ifq
->ifcq_ifp
;
998 struct tcq_class
*cl0
, *cl1
, *cl2
, *cl3
;
1000 u_int32_t maxlen
= 0, qflags
= 0;
1003 IFCQ_LOCK_ASSERT_HELD(ifq
);
1004 VERIFY(ifq
->ifcq_disc
== NULL
);
1005 VERIFY(ifq
->ifcq_type
== PKTSCHEDT_NONE
);
1007 if (flags
& PKTSCHEDF_QALG_RED
)
1009 if (flags
& PKTSCHEDF_QALG_RIO
)
1011 if (flags
& PKTSCHEDF_QALG_BLUE
)
1012 qflags
|= TQCF_BLUE
;
1013 if (flags
& PKTSCHEDF_QALG_SFB
)
1015 if (flags
& PKTSCHEDF_QALG_ECN
)
1017 if (flags
& PKTSCHEDF_QALG_FLOWCTL
)
1018 qflags
|= TQCF_FLOWCTL
;
1019 if (flags
& PKTSCHEDF_QALG_DELAYBASED
)
1020 qflags
|= TQCF_DELAYBASED
;
1022 tif
= tcq_alloc(ifp
, M_WAITOK
, FALSE
);
1026 if ((maxlen
= IFCQ_MAXLEN(ifq
)) == 0)
1027 maxlen
= if_sndq_maxlen
;
1029 if ((err
= tcq_add_queue(tif
, 0, maxlen
,
1030 qflags
| PRCF_LAZY
, SCIDX_BK
, &cl0
)) != 0)
1033 if ((err
= tcq_add_queue(tif
, 1, maxlen
,
1034 qflags
| TQCF_DEFAULTCLASS
, SCIDX_BE
, &cl1
)) != 0)
1037 if ((err
= tcq_add_queue(tif
, 2, maxlen
,
1038 qflags
| PRCF_LAZY
, SCIDX_VI
, &cl2
)) != 0)
1041 if ((err
= tcq_add_queue(tif
, 3, maxlen
,
1042 qflags
, SCIDX_VO
, &cl3
)) != 0)
1045 err
= ifclassq_attach(ifq
, PKTSCHEDT_TCQ
, tif
,
1046 tcq_enqueue_ifclassq
, NULL
, tcq_dequeue_tc_ifclassq
,
1047 tcq_request_ifclassq
);
1049 /* cache these for faster lookup */
1051 /* Map {BK_SYS,BK} to TC_BK */
1052 ifq
->ifcq_disc_slots
[SCIDX_BK_SYS
].qid
= SCIDX_BK
;
1053 ifq
->ifcq_disc_slots
[SCIDX_BK_SYS
].cl
= cl0
;
1055 ifq
->ifcq_disc_slots
[SCIDX_BK
].qid
= SCIDX_BK
;
1056 ifq
->ifcq_disc_slots
[SCIDX_BK
].cl
= cl0
;
1058 /* Map {BE,RD,OAM} to TC_BE */
1059 ifq
->ifcq_disc_slots
[SCIDX_BE
].qid
= SCIDX_BE
;
1060 ifq
->ifcq_disc_slots
[SCIDX_BE
].cl
= cl1
;
1062 ifq
->ifcq_disc_slots
[SCIDX_RD
].qid
= SCIDX_BE
;
1063 ifq
->ifcq_disc_slots
[SCIDX_RD
].cl
= cl1
;
1065 ifq
->ifcq_disc_slots
[SCIDX_OAM
].qid
= SCIDX_BE
;
1066 ifq
->ifcq_disc_slots
[SCIDX_OAM
].cl
= cl1
;
1068 /* Map {AV,RV,VI} to TC_VI */
1069 ifq
->ifcq_disc_slots
[SCIDX_AV
].qid
= SCIDX_VI
;
1070 ifq
->ifcq_disc_slots
[SCIDX_AV
].cl
= cl2
;
1072 ifq
->ifcq_disc_slots
[SCIDX_RV
].qid
= SCIDX_VI
;
1073 ifq
->ifcq_disc_slots
[SCIDX_RV
].cl
= cl2
;
1075 ifq
->ifcq_disc_slots
[SCIDX_VI
].qid
= SCIDX_VI
;
1076 ifq
->ifcq_disc_slots
[SCIDX_VI
].cl
= cl2
;
1078 /* Map {VO,CTL} to TC_VO */
1079 ifq
->ifcq_disc_slots
[SCIDX_VO
].qid
= SCIDX_VO
;
1080 ifq
->ifcq_disc_slots
[SCIDX_VO
].cl
= cl3
;
1082 ifq
->ifcq_disc_slots
[SCIDX_CTL
].qid
= SCIDX_VO
;
1083 ifq
->ifcq_disc_slots
[SCIDX_CTL
].cl
= cl3
;
1088 (void) tcq_destroy_locked(tif
);
1094 tcq_teardown_ifclassq(struct ifclassq
*ifq
)
1096 struct tcq_if
*tif
= ifq
->ifcq_disc
;
1099 IFCQ_LOCK_ASSERT_HELD(ifq
);
1100 VERIFY(tif
!= NULL
&& ifq
->ifcq_type
== PKTSCHEDT_TCQ
);
1102 (void) tcq_destroy_locked(tif
);
1104 ifq
->ifcq_disc
= NULL
;
1105 for (i
= 0; i
< IFCQ_SC_MAX
; i
++) {
1106 ifq
->ifcq_disc_slots
[i
].qid
= 0;
1107 ifq
->ifcq_disc_slots
[i
].cl
= NULL
;
1110 return (ifclassq_detach(ifq
));
1114 tcq_getqstats_ifclassq(struct ifclassq
*ifq
, u_int32_t slot
,
1115 struct if_ifclassq_stats
*ifqs
)
1117 struct tcq_if
*tif
= ifq
->ifcq_disc
;
1119 IFCQ_LOCK_ASSERT_HELD(ifq
);
1120 VERIFY(ifq
->ifcq_type
== PKTSCHEDT_TCQ
);
1122 if (slot
>= IFCQ_SC_MAX
)
1125 return (tcq_get_class_stats(tif
, ifq
->ifcq_disc_slots
[slot
].qid
,
1126 &ifqs
->ifqs_tcq_stats
));
1130 tcq_throttle(struct tcq_if
*tif
, cqrq_throttle_t
*tr
)
1132 struct ifclassq
*ifq
= tif
->tif_ifq
;
1133 struct tcq_class
*cl
;
1136 IFCQ_LOCK_ASSERT_HELD(ifq
);
1137 VERIFY(!(tif
->tif_flags
& TCQIFF_ALTQ
));
1140 tr
->level
= tif
->tif_throttle
;
1144 if (tr
->level
== tif
->tif_throttle
)
1147 /* Current throttling levels only involve BK_SYS class */
1148 cl
= ifq
->ifcq_disc_slots
[SCIDX_BK_SYS
].cl
;
1150 switch (tr
->level
) {
1151 case IFNET_THROTTLE_OFF
:
1152 err
= tcq_resumeq(tif
, cl
);
1155 case IFNET_THROTTLE_OPPORTUNISTIC
:
1156 err
= tcq_suspendq(tif
, cl
);
1164 if (err
== 0 || err
== ENXIO
) {
1165 if (pktsched_verbose
) {
1166 log(LOG_DEBUG
, "%s: %s throttling %slevel set %d->%d\n",
1167 if_name(TCQIF_IFP(tif
)), tcq_style(tif
),
1168 (err
== 0) ? "" : "lazy ", tif
->tif_throttle
,
1171 tif
->tif_throttle
= tr
->level
;
1175 tcq_purgeq(tif
, cl
, 0, NULL
, NULL
);
1177 log(LOG_ERR
, "%s: %s unable to set throttling level "
1178 "%d->%d [error=%d]\n", if_name(TCQIF_IFP(tif
)),
1179 tcq_style(tif
), tif
->tif_throttle
, tr
->level
, err
);
1186 tcq_resumeq(struct tcq_if
*tif
, struct tcq_class
*cl
)
1188 struct ifclassq
*ifq
= tif
->tif_ifq
;
1191 IFCQ_LOCK_ASSERT_HELD(ifq
);
1194 if (q_is_rio(&cl
->cl_q
))
1195 err
= rio_suspendq(cl
->cl_rio
, &cl
->cl_q
, FALSE
);
1197 #endif /* CLASSQ_RIO */
1199 if (q_is_red(&cl
->cl_q
))
1200 err
= red_suspendq(cl
->cl_red
, &cl
->cl_q
, FALSE
);
1202 #endif /* CLASSQ_RED */
1204 if (q_is_blue(&cl
->cl_q
))
1205 err
= blue_suspendq(cl
->cl_blue
, &cl
->cl_q
, FALSE
);
1207 #endif /* CLASSQ_BLUE */
1208 if (q_is_sfb(&cl
->cl_q
) && cl
->cl_sfb
!= NULL
)
1209 err
= sfb_suspendq(cl
->cl_sfb
, &cl
->cl_q
, FALSE
);
1212 qstate(&cl
->cl_q
) = QS_RUNNING
;
1218 tcq_suspendq(struct tcq_if
*tif
, struct tcq_class
*cl
)
1220 struct ifclassq
*ifq
= tif
->tif_ifq
;
1223 IFCQ_LOCK_ASSERT_HELD(ifq
);
1226 if (q_is_rio(&cl
->cl_q
))
1227 err
= rio_suspendq(cl
->cl_rio
, &cl
->cl_q
, TRUE
);
1229 #endif /* CLASSQ_RIO */
1231 if (q_is_red(&cl
->cl_q
))
1232 err
= red_suspendq(cl
->cl_red
, &cl
->cl_q
, TRUE
);
1234 #endif /* CLASSQ_RED */
1236 if (q_is_blue(&cl
->cl_q
))
1237 err
= blue_suspendq(cl
->cl_blue
, &cl
->cl_q
, TRUE
);
1239 #endif /* CLASSQ_BLUE */
1240 if (q_is_sfb(&cl
->cl_q
)) {
1241 if (cl
->cl_sfb
!= NULL
) {
1242 err
= sfb_suspendq(cl
->cl_sfb
, &cl
->cl_q
, TRUE
);
1244 VERIFY(cl
->cl_flags
& TQCF_LAZY
);
1245 err
= ENXIO
; /* delayed throttling */
1249 if (err
== 0 || err
== ENXIO
)
1250 qstate(&cl
->cl_q
) = QS_SUSPENDED
;