1 /*
2 * Copyright (c) 2016-2021 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
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.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
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.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28
29 /*
30 * The migration of flow queue between the different states is summarised in
31 * the below state diagram. (RFC 8290)
32 *
33 * +-----------------+ +------------------+
34 * | | Empty | |
35 * | Empty |<---------------+ Old +----+
36 * | | | | |
37 * +-------+---------+ +------------------+ |
38 * | ^ ^ |Credits
39 * |Arrival | | |Exhausted
40 * v | | |
41 * +-----------------+ | | |
42 * | | Empty or | | |
43 * | New +-------------------+ +-------+
44 * | | Credits Exhausted
45 * +-----------------+
46 *
47 * In this implementation of FQ-CODEL, flow queue is a dynamically allocated
48 * object. An active flow queue goes through the above cycle of state
49 * transitions very often. To avoid the cost of frequent flow queue object
50 * allocation/free, this implementation retains the flow queue object in
51 * [Empty] state on an Empty flow queue list with an active reference in flow
52 * queue hash table. The flow queue objects on the Empty flow queue list have
53 * an associated age and are purged accordingly.
54 */
55
56 #include <sys/cdefs.h>
57 #include <sys/param.h>
58 #include <sys/mbuf.h>
59 #include <sys/socket.h>
60 #include <sys/sockio.h>
61 #include <sys/systm.h>
62 #include <sys/syslog.h>
63 #include <sys/proc.h>
64 #include <sys/errno.h>
65 #include <sys/kernel.h>
66 #include <sys/kauth.h>
67 #include <sys/sdt.h>
68 #include <kern/zalloc.h>
69 #include <netinet/in.h>
70
71 #include <net/classq/classq.h>
72 #include <net/classq/if_classq.h>
73 #include <net/pktsched/pktsched.h>
74 #include <net/pktsched/pktsched_fq_codel.h>
75 #include <net/classq/classq_fq_codel.h>
76
77 #include <netinet/tcp_var.h>
78
79 #define FQ_ZONE_MAX (32 * 1024) /* across all interfaces */
80
81 #define DTYPE_NODROP 0 /* no drop */
82 #define DTYPE_FORCED 1 /* a "forced" drop */
83 #define DTYPE_EARLY 2 /* an "unforced" (early) drop */
84
85 static uint32_t pkt_compressor = 1;
86 static uint64_t l4s_ce_threshold = 0; /* in usec */
87 static uint32_t l4s_local_ce_report = 0;
88 static uint64_t pkt_pacing_leeway = 0; /* in usec */
89 static uint64_t max_pkt_pacing_interval = 3 * NSEC_PER_SEC;
90 static uint64_t l4s_min_delay_threshold = 20 * NSEC_PER_MSEC; /* 20 ms */
91 #if (DEBUG || DEVELOPMENT)
92 SYSCTL_NODE(_net_classq, OID_AUTO, flow_q, CTLFLAG_RW | CTLFLAG_LOCKED,
93 0, "FQ-CODEL parameters");
94
95 SYSCTL_UINT(_net_classq_flow_q, OID_AUTO, pkt_compressor,
96 CTLFLAG_RW | CTLFLAG_LOCKED, &pkt_compressor, 0, "enable pkt compression");
97
98 SYSCTL_QUAD(_net_classq, OID_AUTO, l4s_ce_threshold,
99 CTLFLAG_RW | CTLFLAG_LOCKED, &l4s_ce_threshold,
100 "L4S CE threshold");
101
102 SYSCTL_UINT(_net_classq_flow_q, OID_AUTO, l4s_local_ce_report,
103 CTLFLAG_RW | CTLFLAG_LOCKED, &l4s_local_ce_report, 0,
104 "enable L4S local CE report");
105
106 SYSCTL_QUAD(_net_classq_flow_q, OID_AUTO, pkt_pacing_leeway,
107 CTLFLAG_RW | CTLFLAG_LOCKED, &pkt_pacing_leeway, "packet pacing leeway");
108
109 SYSCTL_QUAD(_net_classq_flow_q, OID_AUTO, max_pkt_pacing_interval,
110 CTLFLAG_RW | CTLFLAG_LOCKED, &max_pkt_pacing_interval, "max packet pacing interval");
111
112 SYSCTL_QUAD(_net_classq_flow_q, OID_AUTO, l4s_min_delay_threshold,
113 CTLFLAG_RW | CTLFLAG_LOCKED, &l4s_min_delay_threshold, "l4s min delay threshold");
114 #endif /* (DEBUG || DEVELOPMENT) */
115
116 void
fq_codel_init(void)117 fq_codel_init(void)
118 {
119 _CASSERT(AQM_KTRACE_AON_FLOW_HIGH_DELAY == 0x8300004);
120 _CASSERT(AQM_KTRACE_AON_THROTTLE == 0x8300008);
121 _CASSERT(AQM_KTRACE_AON_FLOW_OVERWHELMING == 0x830000c);
122 _CASSERT(AQM_KTRACE_AON_FLOW_DQ_STALL == 0x8300010);
123
124 _CASSERT(AQM_KTRACE_STATS_FLOW_ENQUEUE == 0x8310004);
125 _CASSERT(AQM_KTRACE_STATS_FLOW_DEQUEUE == 0x8310008);
126 _CASSERT(AQM_KTRACE_STATS_FLOW_CTL == 0x831000c);
127 _CASSERT(AQM_KTRACE_STATS_FLOW_ALLOC == 0x8310010);
128 _CASSERT(AQM_KTRACE_STATS_FLOW_DESTROY == 0x8310014);
129 _CASSERT(AQM_KTRACE_STATS_FLOW_REPORT_CE == 0x8310018);
130 _CASSERT(AQM_KTRACE_STATS_GET_QLEN == 0x831001c);
131 _CASSERT(AQM_KTRACE_TX_NOT_READY == 0x8310020);
132 _CASSERT(AQM_KTRACE_TX_PACEMAKER == 0x8310024);
133 }
134
135 fq_t *
fq_alloc(classq_pkt_type_t ptype)136 fq_alloc(classq_pkt_type_t ptype)
137 {
138 fq_t *fq = NULL;
139
140 fq = kalloc_type(fq_t, Z_WAITOK_ZERO);
141 if (ptype == QP_MBUF) {
142 MBUFQ_INIT(&fq->fq_mbufq);
143 }
144 #if SKYWALK
145 else {
146 VERIFY(ptype == QP_PACKET);
147 KPKTQ_INIT(&fq->fq_kpktq);
148 }
149 #endif /* SKYWALK */
150 CLASSQ_PKT_INIT(&fq->fq_dq_head);
151 CLASSQ_PKT_INIT(&fq->fq_dq_tail);
152 fq->fq_in_dqlist = false;
153
154 return fq;
155 }
156
157 void
fq_destroy(fq_t * fq,classq_pkt_type_t ptype)158 fq_destroy(fq_t *fq, classq_pkt_type_t ptype)
159 {
160 VERIFY(!fq->fq_in_dqlist);
161 VERIFY(fq_empty(fq, ptype));
162 VERIFY(!(fq->fq_flags & (FQF_NEW_FLOW | FQF_OLD_FLOW |
163 FQF_EMPTY_FLOW)));
164 VERIFY(fq->fq_bytes == 0);
165 kfree_type(fq_t, fq);
166 }
167
168 static inline void
fq_detect_dequeue_stall(fq_if_t * fqs,fq_t * flowq,fq_if_classq_t * fq_cl,u_int64_t * now)169 fq_detect_dequeue_stall(fq_if_t *fqs, fq_t *flowq, fq_if_classq_t *fq_cl,
170 u_int64_t *now)
171 {
172 u_int64_t maxgetqtime, update_interval;
173 if (FQ_IS_DELAY_HIGH(flowq) || flowq->fq_getqtime == 0 ||
174 fq_empty(flowq, fqs->fqs_ptype) ||
175 flowq->fq_bytes < FQ_MIN_FC_THRESHOLD_BYTES) {
176 return;
177 }
178
179 update_interval = FQ_UPDATE_INTERVAL(flowq);
180 maxgetqtime = flowq->fq_getqtime + update_interval;
181 if ((*now) > maxgetqtime) {
182 /*
183 * there was no dequeue in an update interval worth of
184 * time. It means that the queue is stalled.
185 */
186 FQ_SET_DELAY_HIGH(flowq);
187 fq_cl->fcl_stat.fcl_dequeue_stall++;
188 os_log_error(OS_LOG_DEFAULT, "%s:num: %d, "
189 "scidx: %d, flow: 0x%x, iface: %s grp: %hhu", __func__,
190 fq_cl->fcl_stat.fcl_dequeue_stall, flowq->fq_sc_index,
191 flowq->fq_flowhash, if_name(fqs->fqs_ifq->ifcq_ifp),
192 FQ_GROUP(flowq)->fqg_index);
193 KDBG(AQM_KTRACE_AON_FLOW_DQ_STALL, flowq->fq_flowhash,
194 AQM_KTRACE_FQ_GRP_SC_IDX(flowq), flowq->fq_bytes,
195 (*now) - flowq->fq_getqtime);
196 }
197 }
198
199 void
fq_head_drop(fq_if_t * fqs,fq_t * fq)200 fq_head_drop(fq_if_t *fqs, fq_t *fq)
201 {
202 pktsched_pkt_t pkt;
203 volatile uint32_t *__single pkt_flags;
204 uint64_t *__single pkt_timestamp;
205 struct ifclassq *ifq = fqs->fqs_ifq;
206
207 _PKTSCHED_PKT_INIT(&pkt);
208 fq_getq_flow_internal(fqs, fq, &pkt);
209 if (pkt.pktsched_pkt_mbuf == NULL) {
210 return;
211 }
212
213 pktsched_get_pkt_vars(&pkt, &pkt_flags, &pkt_timestamp, NULL, NULL,
214 NULL, NULL, NULL);
215
216 *pkt_timestamp = 0;
217 switch (pkt.pktsched_ptype) {
218 case QP_MBUF:
219 *pkt_flags &= ~PKTF_PRIV_GUARDED;
220 break;
221 #if SKYWALK
222 case QP_PACKET:
223 /* sanity check */
224 ASSERT((*pkt_flags & ~PKT_F_COMMON_MASK) == 0);
225 break;
226 #endif /* SKYWALK */
227 default:
228 VERIFY(0);
229 /* NOTREACHED */
230 __builtin_unreachable();
231 }
232
233 IFCQ_DROP_ADD(ifq, 1, pktsched_get_pkt_len(&pkt));
234 IFCQ_CONVERT_LOCK(ifq);
235 pktsched_free_pkt(&pkt);
236 }
237
238
239 static int
fq_compressor(fq_if_t * fqs,fq_t * fq,fq_if_classq_t * fq_cl,pktsched_pkt_t * pkt)240 fq_compressor(fq_if_t *fqs, fq_t *fq, fq_if_classq_t *fq_cl,
241 pktsched_pkt_t *pkt)
242 {
243 classq_pkt_type_t ptype = fqs->fqs_ptype;
244 uint32_t comp_gencnt = 0;
245 uint64_t *__single pkt_timestamp;
246 uint64_t old_timestamp = 0;
247 uint32_t old_pktlen = 0;
248 struct ifclassq *ifq = fqs->fqs_ifq;
249
250 if (__improbable(pkt_compressor == 0)) {
251 return 0;
252 }
253
254 pktsched_get_pkt_vars(pkt, NULL, &pkt_timestamp, NULL, NULL, NULL,
255 &comp_gencnt, NULL);
256
257 if (comp_gencnt == 0) {
258 return 0;
259 }
260
261 fq_cl->fcl_stat.fcl_pkts_compressible++;
262
263 if (fq_empty(fq, fqs->fqs_ptype)) {
264 return 0;
265 }
266
267 if (ptype == QP_MBUF) {
268 struct mbuf *m = MBUFQ_LAST(&fq->fq_mbufq);
269
270 if (comp_gencnt != m->m_pkthdr.comp_gencnt) {
271 return 0;
272 }
273
274 /* If we got until here, we should merge/replace the segment */
275 MBUFQ_REMOVE(&fq->fq_mbufq, m);
276 old_pktlen = m_pktlen(m);
277 old_timestamp = m->m_pkthdr.pkt_timestamp;
278
279 IFCQ_CONVERT_LOCK(fqs->fqs_ifq);
280 m_freem(m);
281 }
282 #if SKYWALK
283 else {
284 struct __kern_packet *kpkt = KPKTQ_LAST(&fq->fq_kpktq);
285
286 if (comp_gencnt != kpkt->pkt_comp_gencnt) {
287 return 0;
288 }
289
290 /* If we got until here, we should merge/replace the segment */
291 KPKTQ_REMOVE(&fq->fq_kpktq, kpkt);
292 old_pktlen = kpkt->pkt_length;
293 old_timestamp = kpkt->pkt_timestamp;
294
295 IFCQ_CONVERT_LOCK(fqs->fqs_ifq);
296 struct kern_pbufpool *pp =
297 __DECONST(struct kern_pbufpool *, ((struct __kern_quantum *)kpkt)->qum_pp);
298 pp_free_packet(pp, (uint64_t)kpkt);
299 }
300 #endif /* SKYWALK */
301
302 fq->fq_bytes -= old_pktlen;
303 fq_cl->fcl_stat.fcl_byte_cnt -= old_pktlen;
304 fq_cl->fcl_stat.fcl_pkt_cnt--;
305 IFCQ_DEC_LEN(ifq);
306 IFCQ_DEC_BYTES(ifq, old_pktlen);
307
308 FQ_GRP_DEC_LEN(fq);
309 FQ_GRP_DEC_BYTES(fq, old_pktlen);
310
311 *pkt_timestamp = old_timestamp;
312
313 return CLASSQEQ_COMPRESSED;
314 }
315
316 int
fq_addq(fq_if_t * fqs,fq_if_group_t * fq_grp,pktsched_pkt_t * pkt,fq_if_classq_t * fq_cl)317 fq_addq(fq_if_t *fqs, fq_if_group_t *fq_grp, pktsched_pkt_t *pkt,
318 fq_if_classq_t *fq_cl)
319 {
320 int droptype = DTYPE_NODROP, fc_adv = 0, ret = CLASSQEQ_SUCCESS;
321 u_int64_t now;
322 fq_t *fq = NULL;
323 uint64_t *__single pkt_timestamp;
324 volatile uint32_t *__single pkt_flags;
325 uint32_t pkt_flowid, cnt;
326 uint8_t pkt_proto, pkt_flowsrc;
327 fq_tfc_type_t tfc_type = FQ_TFC_C;
328
329 cnt = pkt->pktsched_pcnt;
330 pktsched_get_pkt_vars(pkt, &pkt_flags, &pkt_timestamp, &pkt_flowid,
331 &pkt_flowsrc, &pkt_proto, NULL, NULL);
332
333 /*
334 * XXX Not walking the chain to set this flag on every packet.
335 * This flag is only used for debugging. Nothing is affected if it's
336 * not set.
337 */
338 switch (pkt->pktsched_ptype) {
339 case QP_MBUF:
340 /* See comments in <rdar://problem/14040693> */
341 VERIFY(!(*pkt_flags & PKTF_PRIV_GUARDED));
342 *pkt_flags |= PKTF_PRIV_GUARDED;
343 break;
344 #if SKYWALK
345 case QP_PACKET:
346 /* sanity check */
347 ASSERT((*pkt_flags & ~PKT_F_COMMON_MASK) == 0);
348 break;
349 #endif /* SKYWALK */
350 default:
351 VERIFY(0);
352 /* NOTREACHED */
353 __builtin_unreachable();
354 }
355
356 if (ifclassq_enable_l4s) {
357 tfc_type = pktsched_is_pkt_l4s(pkt) ? FQ_TFC_L4S : FQ_TFC_C;
358 }
359
360 /*
361 * Timestamps for every packet must be set prior to entering this path.
362 */
363 now = *pkt_timestamp;
364 ASSERT(now > 0);
365
366 /* find the flowq for this packet */
367 fq = fq_if_hash_pkt(fqs, fq_grp, pkt_flowid, pktsched_get_pkt_svc(pkt),
368 now, true, tfc_type);
369 if (__improbable(fq == NULL)) {
370 DTRACE_IP1(memfail__drop, fq_if_t *, fqs);
371 /* drop the packet if we could not allocate a flow queue */
372 fq_cl->fcl_stat.fcl_drop_memfailure += cnt;
373 return CLASSQEQ_DROP;
374 }
375 VERIFY(fq->fq_group == fq_grp);
376 VERIFY(fqs->fqs_ptype == pkt->pktsched_ptype);
377
378 KDBG(AQM_KTRACE_STATS_FLOW_ENQUEUE, fq->fq_flowhash,
379 AQM_KTRACE_FQ_GRP_SC_IDX(fq),
380 fq->fq_bytes, pktsched_get_pkt_len(pkt));
381
382 fq_detect_dequeue_stall(fqs, fq, fq_cl, &now);
383
384 /*
385 * Skip the dropping part if it's L4S. Flow control or ECN marking decision
386 * will be made at dequeue time.
387 */
388 if (ifclassq_enable_l4s && tfc_type == FQ_TFC_L4S) {
389 fq_cl->fcl_stat.fcl_l4s_pkts += cnt;
390 droptype = DTYPE_NODROP;
391 }
392
393 if (__improbable(FQ_IS_DELAY_HIGH(fq) || FQ_IS_OVERWHELMING(fq))) {
394 if ((fq->fq_flags & FQF_FLOWCTL_CAPABLE) &&
395 (*pkt_flags & PKTF_FLOW_ADV)) {
396 fc_adv = 1;
397 /*
398 * If the flow is suspended or it is not
399 * TCP/QUIC, drop the chain.
400 */
401 if ((pkt_proto != IPPROTO_TCP) &&
402 (pkt_proto != IPPROTO_QUIC)) {
403 droptype = DTYPE_EARLY;
404 fq_cl->fcl_stat.fcl_drop_early += cnt;
405 IFCQ_DROP_ADD(fqs->fqs_ifq, cnt, pktsched_get_pkt_len(pkt));
406 }
407 DTRACE_IP6(flow__adv, fq_if_t *, fqs,
408 fq_if_classq_t *, fq_cl, fq_t *, fq,
409 int, droptype, pktsched_pkt_t *, pkt,
410 uint32_t, cnt);
411 } else {
412 /*
413 * Need to drop packets to make room for the new
414 * ones. Try to drop from the head of the queue
415 * instead of the latest packets.
416 */
417 if (!fq_empty(fq, fqs->fqs_ptype)) {
418 uint32_t i;
419
420 for (i = 0; i < cnt; i++) {
421 fq_head_drop(fqs, fq);
422 }
423 droptype = DTYPE_NODROP;
424 } else {
425 droptype = DTYPE_EARLY;
426 }
427 fq_cl->fcl_stat.fcl_drop_early += cnt;
428
429 DTRACE_IP6(no__flow__adv, fq_if_t *, fqs,
430 fq_if_classq_t *, fq_cl, fq_t *, fq,
431 int, droptype, pktsched_pkt_t *, pkt,
432 uint32_t, cnt);
433 }
434 }
435
436 /* Set the return code correctly */
437 if (__improbable(fc_adv == 1 && droptype != DTYPE_FORCED)) {
438 if (fq_if_add_fcentry(fqs, pkt, pkt_flowsrc, fq, fq_cl)) {
439 fq->fq_flags |= FQF_FLOWCTL_ON;
440 /* deliver flow control advisory error */
441 if (droptype == DTYPE_NODROP) {
442 ret = CLASSQEQ_SUCCESS_FC;
443 } else {
444 /* dropped due to flow control */
445 ret = CLASSQEQ_DROP_FC;
446 }
447 } else {
448 /*
449 * if we could not flow control the flow, it is
450 * better to drop
451 */
452 droptype = DTYPE_FORCED;
453 ret = CLASSQEQ_DROP_FC;
454 fq_cl->fcl_stat.fcl_flow_control_fail++;
455 }
456 DTRACE_IP3(fc__ret, fq_if_t *, fqs, int, droptype, int, ret);
457 }
458
459 /*
460 * If the queue length hits the queue limit, drop a chain with the
461 * same number of packets from the front of the queue for a flow with
462 * maximum number of bytes. This will penalize heavy and unresponsive
463 * flows. It will also avoid a tail drop.
464 */
465 if (__improbable(droptype == DTYPE_NODROP &&
466 fq_if_at_drop_limit(fqs))) {
467 uint32_t i;
468
469 if (fqs->fqs_large_flow == fq) {
470 /*
471 * Drop from the head of the current fq. Since a
472 * new packet will be added to the tail, it is ok
473 * to leave fq in place.
474 */
475 DTRACE_IP5(large__flow, fq_if_t *, fqs,
476 fq_if_classq_t *, fq_cl, fq_t *, fq,
477 pktsched_pkt_t *, pkt, uint32_t, cnt);
478
479 for (i = 0; i < cnt; i++) {
480 fq_head_drop(fqs, fq);
481 }
482 fq_cl->fcl_stat.fcl_drop_overflow += cnt;
483
484 /*
485 * TCP and QUIC will react to the loss of those head dropped pkts
486 * and adjust send rate.
487 */
488 if ((fq->fq_flags & FQF_FLOWCTL_CAPABLE) &&
489 (*pkt_flags & PKTF_FLOW_ADV) &&
490 (pkt_proto != IPPROTO_TCP) &&
491 (pkt_proto != IPPROTO_QUIC)) {
492 if (fq_if_add_fcentry(fqs, pkt, pkt_flowsrc, fq, fq_cl)) {
493 fq->fq_flags |= FQF_FLOWCTL_ON;
494 FQ_SET_OVERWHELMING(fq);
495 fq_cl->fcl_stat.fcl_overwhelming++;
496 /* deliver flow control advisory error */
497 ret = CLASSQEQ_SUCCESS_FC;
498 }
499 }
500 } else {
501 if (fqs->fqs_large_flow == NULL) {
502 droptype = DTYPE_FORCED;
503 fq_cl->fcl_stat.fcl_drop_overflow += cnt;
504 ret = CLASSQEQ_DROP;
505
506 DTRACE_IP5(no__large__flow, fq_if_t *, fqs,
507 fq_if_classq_t *, fq_cl, fq_t *, fq,
508 pktsched_pkt_t *, pkt, uint32_t, cnt);
509
510 /*
511 * if this fq was freshly created and there
512 * is nothing to enqueue, move it to empty list
513 */
514 if (fq_empty(fq, fqs->fqs_ptype) &&
515 !(fq->fq_flags & (FQF_NEW_FLOW |
516 FQF_OLD_FLOW))) {
517 fq_if_move_to_empty_flow(fqs, fq_cl,
518 fq, now);
519 fq = NULL;
520 }
521 } else {
522 DTRACE_IP5(different__large__flow,
523 fq_if_t *, fqs, fq_if_classq_t *, fq_cl,
524 fq_t *, fq, pktsched_pkt_t *, pkt,
525 uint32_t, cnt);
526
527 for (i = 0; i < cnt; i++) {
528 fq_if_drop_packet(fqs, now);
529 }
530 }
531 }
532 }
533
534 fq_cl->fcl_flags &= ~FCL_PACED;
535
536 if (__probable(droptype == DTYPE_NODROP)) {
537 uint32_t chain_len = pktsched_get_pkt_len(pkt);
538 int ret_compress = 0;
539
540 /*
541 * We do not compress if we are enqueuing a chain.
542 * Traversing the chain to look for acks would defeat the
543 * purpose of batch enqueueing.
544 */
545 if (cnt == 1) {
546 ret_compress = fq_compressor(fqs, fq, fq_cl, pkt);
547 if (ret_compress == CLASSQEQ_COMPRESSED) {
548 fq_cl->fcl_stat.fcl_pkts_compressed++;
549 }
550 }
551 DTRACE_IP5(fq_enqueue, fq_if_t *, fqs, fq_if_classq_t *, fq_cl,
552 fq_t *, fq, pktsched_pkt_t *, pkt, uint32_t, cnt);
553 fq_enqueue(fq, pkt->pktsched_pkt, pkt->pktsched_tail, cnt,
554 pkt->pktsched_ptype);
555
556 fq->fq_bytes += chain_len;
557 fq_cl->fcl_stat.fcl_byte_cnt += chain_len;
558 fq_cl->fcl_stat.fcl_pkt_cnt += cnt;
559
560 /*
561 * check if this queue will qualify to be the next
562 * victim queue
563 */
564 fq_if_is_flow_heavy(fqs, fq);
565 } else {
566 DTRACE_IP3(fq_drop, fq_if_t *, fqs, int, droptype, int, ret);
567 return (ret != CLASSQEQ_SUCCESS) ? ret : CLASSQEQ_DROP;
568 }
569
570 /*
571 * If the queue is not currently active, add it to the end of new
572 * flows list for that service class.
573 */
574 if ((fq->fq_flags & (FQF_NEW_FLOW | FQF_OLD_FLOW)) == 0) {
575 VERIFY(STAILQ_NEXT(fq, fq_actlink) == NULL);
576 STAILQ_INSERT_TAIL(&fq_cl->fcl_new_flows, fq, fq_actlink);
577 fq->fq_flags |= FQF_NEW_FLOW;
578
579 fq_cl->fcl_stat.fcl_newflows_cnt++;
580
581 fq->fq_deficit = fq_cl->fcl_quantum;
582 }
583 return ret;
584 }
585
586 void
fq_getq_flow_internal(fq_if_t * fqs,fq_t * fq,pktsched_pkt_t * pkt)587 fq_getq_flow_internal(fq_if_t *fqs, fq_t *fq, pktsched_pkt_t *pkt)
588 {
589 classq_pkt_t p = CLASSQ_PKT_INITIALIZER(p);
590 uint32_t plen;
591 fq_if_classq_t *fq_cl;
592 struct ifclassq *ifq = fqs->fqs_ifq;
593
594 fq_dequeue(fq, &p, fqs->fqs_ptype);
595 if (p.cp_ptype == QP_INVALID) {
596 VERIFY(p.cp_mbuf == NULL);
597 return;
598 }
599
600 fq->fq_next_tx_time = FQ_INVALID_TX_TS;
601
602 pktsched_pkt_encap(pkt, &p);
603 plen = pktsched_get_pkt_len(pkt);
604
605 VERIFY(fq->fq_bytes >= plen);
606 fq->fq_bytes -= plen;
607
608 fq_cl = &FQ_CLASSQ(fq);
609 fq_cl->fcl_stat.fcl_byte_cnt -= plen;
610 fq_cl->fcl_stat.fcl_pkt_cnt--;
611 fq_cl->fcl_flags &= ~FCL_PACED;
612
613 IFCQ_DEC_LEN(ifq);
614 IFCQ_DEC_BYTES(ifq, plen);
615
616 FQ_GRP_DEC_LEN(fq);
617 FQ_GRP_DEC_BYTES(fq, plen);
618
619 /* Reset getqtime so that we don't count idle times */
620 if (fq_empty(fq, fqs->fqs_ptype)) {
621 fq->fq_getqtime = 0;
622 }
623 }
624
625 /*
626 * fq_get_next_tx_time returns FQ_INVALID_TX_TS when there is no tx time in fq
627 */
628 static uint64_t
fq_get_next_tx_time(fq_if_t * fqs,fq_t * fq)629 fq_get_next_tx_time(fq_if_t *fqs, fq_t *fq)
630 {
631 uint64_t tx_time = FQ_INVALID_TX_TS;
632
633 /*
634 * Check the cached value in fq
635 */
636 if (fq->fq_next_tx_time != FQ_INVALID_TX_TS) {
637 return fq->fq_next_tx_time;
638 }
639
640 switch (fqs->fqs_ptype) {
641 case QP_MBUF: {
642 struct mbuf *m;
643 if ((m = MBUFQ_FIRST(&fq->fq_mbufq)) != NULL) {
644 struct m_tag *tag;
645 tag = m_tag_locate(m, KERNEL_MODULE_TAG_ID,
646 KERNEL_TAG_TYPE_AQM);
647 if (tag != NULL) {
648 tx_time = *(uint64_t *)tag->m_tag_data;
649 }
650 }
651 break;
652 }
653 case QP_PACKET: {
654 struct __kern_packet *p = KPKTQ_FIRST(&fq->fq_kpktq);
655 if (__probable(p != NULL)) {
656 tx_time = __packet_get_tx_timestamp(SK_PKT2PH(p));
657 }
658 break;
659 }
660 default:
661 VERIFY(0);
662 /* NOTREACHED */
663 __builtin_unreachable();
664 }
665
666 /*
667 * Cache the tx time in fq. The cache will be clear after dequeue or drop
668 * from the fq.
669 */
670 fq->fq_next_tx_time = tx_time;
671
672 return tx_time;
673 }
674
675 /*
676 * fq_tx_time_ready returns true if the fq is empty so that it doesn't
677 * affect caller logics that handles empty flow.
678 */
679 boolean_t
fq_tx_time_ready(fq_if_t * fqs,fq_t * fq,uint64_t now,uint64_t * ready_time)680 fq_tx_time_ready(fq_if_t *fqs, fq_t *fq, uint64_t now, uint64_t *ready_time)
681 {
682 uint64_t pkt_tx_time;
683 fq_if_classq_t *fq_cl = &FQ_CLASSQ(fq);
684
685 if (!ifclassq_enable_pacing || !ifclassq_enable_l4s || fq->fq_tfc_type != FQ_TFC_L4S) {
686 return TRUE;
687 }
688
689 pkt_tx_time = fq_get_next_tx_time(fqs, fq);
690 if (ready_time != NULL) {
691 *ready_time = pkt_tx_time;
692 }
693
694 if (pkt_tx_time <= now + pkt_pacing_leeway ||
695 pkt_tx_time == FQ_INVALID_TX_TS) {
696 return TRUE;
697 }
698
699 /*
700 * Ignore the tx time if it's scheduled too far in the future
701 */
702 if (pkt_tx_time > max_pkt_pacing_interval + now) {
703 fq_cl->fcl_stat.fcl_ignore_tx_time++;
704 return TRUE;
705 }
706
707 ASSERT(pkt_tx_time != FQ_INVALID_TX_TS);
708 KDBG(AQM_KTRACE_TX_NOT_READY, fq->fq_flowhash,
709 AQM_KTRACE_FQ_GRP_SC_IDX(fq), now, pkt_tx_time);
710 return FALSE;
711 }
712
713 void
fq_getq_flow(fq_if_t * fqs,fq_t * fq,pktsched_pkt_t * pkt,uint64_t now)714 fq_getq_flow(fq_if_t *fqs, fq_t *fq, pktsched_pkt_t *pkt, uint64_t now)
715 {
716 fq_if_classq_t *fq_cl = &FQ_CLASSQ(fq);
717 int64_t qdelay = 0;
718 volatile uint32_t *__single pkt_flags;
719 uint64_t *__single pkt_timestamp, pkt_tx_time = 0, pacing_delay = 0;
720 uint64_t fq_min_delay_threshold = FQ_TARGET_DELAY(fq);
721 uint8_t pkt_flowsrc;
722 boolean_t l4s_pkt;
723
724 fq_getq_flow_internal(fqs, fq, pkt);
725 if (pkt->pktsched_ptype == QP_INVALID) {
726 VERIFY(pkt->pktsched_pkt_mbuf == NULL);
727 return;
728 }
729
730 pktsched_get_pkt_vars(pkt, &pkt_flags, &pkt_timestamp, NULL, &pkt_flowsrc,
731 NULL, NULL, &pkt_tx_time);
732 l4s_pkt = pktsched_is_pkt_l4s(pkt);
733 if (ifclassq_enable_pacing && ifclassq_enable_l4s) {
734 if (pkt_tx_time > *pkt_timestamp) {
735 pacing_delay = pkt_tx_time - *pkt_timestamp;
736 fq_cl->fcl_stat.fcl_paced_pkts++;
737 DTRACE_SKYWALK3(aqm__pacing__delta, uint64_t, now - pkt_tx_time,
738 fq_if_t *, fqs, fq_t *, fq);
739 }
740 #if (DEVELOPMENT || DEBUG)
741 else if (pkt_tx_time != 0) {
742 DTRACE_SKYWALK5(aqm__miss__pacing__delay, uint64_t, *pkt_timestamp,
743 uint64_t, pkt_tx_time, uint64_t, now, fq_if_t *,
744 fqs, fq_t *, fq);
745 }
746 #endif // (DEVELOPMENT || DEBUG)
747 }
748
749 /* this will compute qdelay in nanoseconds */
750 if (now > *pkt_timestamp) {
751 qdelay = now - *pkt_timestamp;
752 }
753
754 /* Update min/max/avg qdelay for the respective class */
755 if (fq_cl->fcl_stat.fcl_min_qdelay == 0 ||
756 (qdelay > 0 && (u_int64_t)qdelay < fq_cl->fcl_stat.fcl_min_qdelay)) {
757 fq_cl->fcl_stat.fcl_min_qdelay = qdelay;
758 }
759
760 if (fq_cl->fcl_stat.fcl_max_qdelay == 0 ||
761 (qdelay > 0 && (u_int64_t)qdelay > fq_cl->fcl_stat.fcl_max_qdelay)) {
762 fq_cl->fcl_stat.fcl_max_qdelay = qdelay;
763 }
764
765 uint64_t num_dequeues = fq_cl->fcl_stat.fcl_dequeue;
766
767 if (num_dequeues == 0) {
768 fq_cl->fcl_stat.fcl_avg_qdelay = qdelay;
769 } else if (qdelay > 0) {
770 uint64_t res = 0;
771 if (os_add_overflow(num_dequeues, 1, &res)) {
772 /* Reset the dequeue num and dequeue bytes */
773 fq_cl->fcl_stat.fcl_dequeue = num_dequeues = 0;
774 fq_cl->fcl_stat.fcl_dequeue_bytes = 0;
775 fq_cl->fcl_stat.fcl_avg_qdelay = qdelay;
776 os_log_info(OS_LOG_DEFAULT, "%s: dequeue num overflow, "
777 "flow: 0x%x, iface: %s", __func__, fq->fq_flowhash,
778 if_name(fqs->fqs_ifq->ifcq_ifp));
779 } else {
780 uint64_t product = 0;
781 if (os_mul_overflow(fq_cl->fcl_stat.fcl_avg_qdelay,
782 num_dequeues, &product) || os_add_overflow(product, qdelay, &res)) {
783 fq_cl->fcl_stat.fcl_avg_qdelay = qdelay;
784 } else {
785 fq_cl->fcl_stat.fcl_avg_qdelay = res /
786 (num_dequeues + 1);
787 }
788 }
789 }
790
791 fq->fq_pkts_since_last_report++;
792 if (ifclassq_enable_l4s && l4s_pkt) {
793 /*
794 * A safe guard to make sure that L4S is not going to build a huge
795 * queue if we encounter unexpected problems (for eg., if ACKs don't
796 * arrive in timely manner due to congestion in reverse path).
797 */
798 fq_min_delay_threshold = l4s_min_delay_threshold;
799
800 if ((l4s_ce_threshold != 0 && qdelay > l4s_ce_threshold + pacing_delay) ||
801 (l4s_ce_threshold == 0 && qdelay > FQ_TARGET_DELAY(fq) + pacing_delay)) {
802 DTRACE_SKYWALK4(aqm__mark__ce, uint64_t, qdelay, uint64_t, pacing_delay,
803 fq_if_t *, fqs, fq_t *, fq);
804 KDBG(AQM_KTRACE_STATS_FLOW_REPORT_CE, fq->fq_flowhash,
805 AQM_KTRACE_FQ_GRP_SC_IDX(fq), qdelay, pacing_delay);
806 /*
807 * The packet buffer that pktsched_mark_ecn writes to can be pageable.
808 * Since it is not safe to write to pageable memory while preemption
809 * is disabled, convert the spin lock into mutex.
810 */
811 IFCQ_CONVERT_LOCK(fqs->fqs_ifq);
812 if (__improbable(l4s_local_ce_report != 0) &&
813 (*pkt_flags & PKTF_FLOW_ADV) != 0 &&
814 fq_if_report_ce(fqs, pkt, 1, fq->fq_pkts_since_last_report)) {
815 fq->fq_pkts_since_last_report = 0;
816 fq_cl->fcl_stat.fcl_ce_reported++;
817 } else if (pktsched_mark_ecn(pkt) == 0) {
818 fq_cl->fcl_stat.fcl_ce_marked++;
819 } else {
820 fq_cl->fcl_stat.fcl_ce_mark_failures++;
821 }
822 }
823 }
824
825 ASSERT(pacing_delay <= INT64_MAX);
826 qdelay = MAX(0, qdelay - (int64_t)pacing_delay);
827 if (fq->fq_min_qdelay == 0 ||
828 (u_int64_t)qdelay < fq->fq_min_qdelay) {
829 fq->fq_min_qdelay = qdelay;
830 }
831
832 if (now >= fq->fq_updatetime) {
833 if (fq->fq_min_qdelay > fq_min_delay_threshold) {
834 if (!FQ_IS_DELAY_HIGH(fq)) {
835 FQ_SET_DELAY_HIGH(fq);
836 }
837 } else {
838 FQ_CLEAR_DELAY_HIGH(fq);
839 }
840 /* Reset measured queue delay and update time */
841 fq->fq_updatetime = now + FQ_UPDATE_INTERVAL(fq);
842 fq->fq_min_qdelay = 0;
843 }
844
845 if (fqs->fqs_large_flow != fq || !fq_if_almost_at_drop_limit(fqs)) {
846 FQ_CLEAR_OVERWHELMING(fq);
847 }
848 if (!FQ_IS_DELAY_HIGH(fq) || fq_empty(fq, fqs->fqs_ptype)) {
849 FQ_CLEAR_DELAY_HIGH(fq);
850 }
851
852 if ((fq->fq_flags & FQF_FLOWCTL_ON) &&
853 !FQ_IS_DELAY_HIGH(fq) && !FQ_IS_OVERWHELMING(fq)) {
854 fq_if_flow_feedback(fqs, fq, fq_cl);
855 }
856
857 if (fq_empty(fq, fqs->fqs_ptype)) {
858 /* Reset getqtime so that we don't count idle times */
859 fq->fq_getqtime = 0;
860 } else {
861 fq->fq_getqtime = now;
862 }
863 fq_if_is_flow_heavy(fqs, fq);
864
865 *pkt_timestamp = 0;
866 switch (pkt->pktsched_ptype) {
867 case QP_MBUF:
868 *pkt_flags &= ~PKTF_PRIV_GUARDED;
869 break;
870 #if SKYWALK
871 case QP_PACKET:
872 /* sanity check */
873 ASSERT((*pkt_flags & ~PKT_F_COMMON_MASK) == 0);
874 break;
875 #endif /* SKYWALK */
876 default:
877 VERIFY(0);
878 /* NOTREACHED */
879 __builtin_unreachable();
880 }
881 }
882