[apple/xnu.git] / bsd / net / classq / classq_fq_codel.c

/*
 * Copyright (c) 2016-2017 Apple Inc. All rights reserved.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
 *
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. The rights granted to you under the License
 * may not be used to create, or enable the creation or redistribution of,
 * unlawful or unlicensed copies of an Apple operating system, or to
 * circumvent, violate, or enable the circumvention or violation of, any
 * terms of an Apple operating system software license agreement.
 *
 * Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this file.
 *
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
 */

#include <sys/cdefs.h>
#include <sys/param.h>
#include <sys/mbuf.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/systm.h>
#include <sys/sysctl.h>
#include <sys/syslog.h>
#include <sys/proc.h>
#include <sys/errno.h>
#include <sys/kernel.h>
#include <sys/kauth.h>
#include <kern/zalloc.h>
#include <netinet/in.h>

#include <net/classq/classq.h>
#include <net/classq/if_classq.h>
#include <net/pktsched/pktsched.h>
#include <net/pktsched/pktsched_fq_codel.h>
#include <net/classq/classq_fq_codel.h>

static uint32_t flowq_size;			/* size of flowq */
static struct mcache *flowq_cache = NULL;	/* mcache for flowq */

#define	FQ_ZONE_MAX	(32 * 1024)	/* across all interfaces */

#define	DTYPE_NODROP	0	/* no drop */
#define	DTYPE_FORCED	1	/* a "forced" drop */
#define	DTYPE_EARLY	2	/* an "unforced" (early) drop */

void
fq_codel_init(void)
{
	if (flowq_cache != NULL)
		return;

	flowq_size = sizeof (fq_t);
	flowq_cache = mcache_create("fq.flowq", flowq_size, sizeof (uint64_t),
	    0, MCR_SLEEP);
	if (flowq_cache == NULL) {
		panic("%s: failed to allocate flowq_cache", __func__);
		/* NOTREACHED */
	}
}

void
fq_codel_reap_caches(boolean_t purge)
{
	mcache_reap_now(flowq_cache, purge);
}

fq_t *
fq_alloc(classq_pkt_type_t ptype)
{
	fq_t *fq = NULL;
	fq = mcache_alloc(flowq_cache, MCR_SLEEP);
	if (fq == NULL) {
		log(LOG_ERR, "%s: unable to allocate from flowq_cache\n");
		return (NULL);
	}

	bzero(fq, flowq_size);
	fq->fq_ptype = ptype;
	if (ptype == QP_MBUF) {
		MBUFQ_INIT(&fq->fq_mbufq);
	}
	return (fq);
}

void
fq_destroy(fq_t *fq)
{
	VERIFY(fq_empty(fq));
	VERIFY(!(fq->fq_flags & (FQF_NEW_FLOW | FQF_OLD_FLOW)));
	VERIFY(fq->fq_bytes == 0);
	mcache_free(flowq_cache, fq);
}

static void
fq_detect_dequeue_stall(fq_if_t *fqs, fq_t *flowq, fq_if_classq_t *fq_cl,
    u_int64_t *now)
{
	u_int64_t maxgetqtime;
	if (FQ_IS_DELAYHIGH(flowq) || flowq->fq_getqtime == 0 ||
	    fq_empty(flowq) ||
	    flowq->fq_bytes < FQ_MIN_FC_THRESHOLD_BYTES)
		return;
	maxgetqtime = flowq->fq_getqtime + fqs->fqs_update_interval;
	if ((*now) > maxgetqtime) {
		/*
		 * there was no dequeue in an update interval worth of
		 * time. It means that the queue is stalled.
		 */
		FQ_SET_DELAY_HIGH(flowq);
		fq_cl->fcl_stat.fcl_dequeue_stall++;
	}
}

void
fq_head_drop(fq_if_t *fqs, fq_t *fq)
{
	pktsched_pkt_t pkt;
	uint32_t *pkt_flags;
	uint64_t *pkt_timestamp;
	struct ifclassq *ifq = fqs->fqs_ifq;

	_PKTSCHED_PKT_INIT(&pkt);
	if (fq_getq_flow_internal(fqs, fq, &pkt) == NULL)
		return;

	pktsched_get_pkt_vars(&pkt, &pkt_flags, &pkt_timestamp, NULL, NULL,
	    NULL, NULL);

	*pkt_timestamp = 0;
	if (pkt.pktsched_ptype == QP_MBUF)
		*pkt_flags &= ~PKTF_PRIV_GUARDED;

	IFCQ_DROP_ADD(ifq, 1, pktsched_get_pkt_len(&pkt));
	IFCQ_CONVERT_LOCK(ifq);
	pktsched_free_pkt(&pkt);
}

int
fq_addq(fq_if_t *fqs, pktsched_pkt_t *pkt, fq_if_classq_t *fq_cl)
{
	int droptype = DTYPE_NODROP, fc_adv = 0, ret = CLASSQEQ_SUCCESS;
	u_int64_t now;
	fq_t *fq = NULL;
	uint64_t *pkt_timestamp;
	uint32_t *pkt_flags;
	uint32_t pkt_flowid, pkt_tx_start_seq;
	uint8_t pkt_proto, pkt_flowsrc;

	pktsched_get_pkt_vars(pkt, &pkt_flags, &pkt_timestamp, &pkt_flowid,
	    &pkt_flowsrc, &pkt_proto, &pkt_tx_start_seq);

	if (pkt->pktsched_ptype == QP_MBUF) {
		/* See comments in <rdar://problem/14040693> */
		VERIFY(!(*pkt_flags & PKTF_PRIV_GUARDED));
		*pkt_flags |= PKTF_PRIV_GUARDED;
	}

	if (*pkt_timestamp > 0) {
		now = *pkt_timestamp;
	} else {
		struct timespec now_ts;
		nanouptime(&now_ts);
		now = (now_ts.tv_sec * NSEC_PER_SEC) + now_ts.tv_nsec;
		*pkt_timestamp = now;
	}

	/* find the flowq for this packet */
	fq = fq_if_hash_pkt(fqs, pkt_flowid, pktsched_get_pkt_svc(pkt),
	    now, TRUE, pkt->pktsched_ptype);
	if (fq == NULL) {
		/* drop the packet if we could not allocate a flow queue */
		fq_cl->fcl_stat.fcl_drop_memfailure++;
		IFCQ_CONVERT_LOCK(fqs->fqs_ifq);
		return (CLASSQEQ_DROP);
	}
	VERIFY(fq->fq_ptype == pkt->pktsched_ptype);

	fq_detect_dequeue_stall(fqs, fq, fq_cl, &now);

	if (FQ_IS_DELAYHIGH(fq)) {
		if ((fq->fq_flags & FQF_FLOWCTL_CAPABLE) &&
		    (*pkt_flags & PKTF_FLOW_ADV)) {
			fc_adv = 1;
			/*
			 * If the flow is suspended or it is not
			 * TCP, drop the packet
			 */
			if (pkt_proto != IPPROTO_TCP) {
				droptype = DTYPE_EARLY;
				fq_cl->fcl_stat.fcl_drop_early++;
			}
		} else {
			/*
			 * Need to drop a packet, instead of dropping this
			 * one, try to drop from the head of the queue
			 */
			if (!fq_empty(fq)) {
				fq_head_drop(fqs, fq);
				droptype = DTYPE_NODROP;
			} else {
				droptype = DTYPE_EARLY;
			}
			fq_cl->fcl_stat.fcl_drop_early++;
		}

	}

	/* Set the return code correctly */
	if (fc_adv == 1 && droptype != DTYPE_FORCED) {
		if (fq_if_add_fcentry(fqs, pkt, pkt_flowid, pkt_flowsrc,
		    fq_cl)) {
			fq->fq_flags |= FQF_FLOWCTL_ON;
			/* deliver flow control advisory error */
			if (droptype == DTYPE_NODROP) {
				ret = CLASSQEQ_SUCCESS_FC;
			} else {
				/* dropped due to flow control */
				ret = CLASSQEQ_DROP_FC;
			}
		} else {
			/*
			 * if we could not flow control the flow, it is
			 * better to drop
			 */
			droptype = DTYPE_FORCED;
			ret = CLASSQEQ_DROP_FC;
			fq_cl->fcl_stat.fcl_flow_control_fail++;
		}
	}

	/*
	 * If the queue length hits the queue limit, drop a packet from the
	 * front of the queue for a flow with maximum number of bytes. This
	 * will penalize heavy and unresponsive flows. It will also avoid a
	 * tail drop.
	 */
	if (droptype == DTYPE_NODROP && fq_if_at_drop_limit(fqs)) {
		if (fqs->fqs_large_flow == fq) {
			/*
			 * Drop from the head of the current fq. Since a
			 * new packet will be added to the tail, it is ok
			 * to leave fq in place.
			 */
			fq_head_drop(fqs, fq);
		} else {
			if (fqs->fqs_large_flow == NULL) {
				droptype = DTYPE_FORCED;
				fq_cl->fcl_stat.fcl_drop_overflow++;
				ret = CLASSQEQ_DROP;

				/*
				 * if this fq was freshly created and there
				 * is nothing to enqueue, free it
				 */
				if (fq_empty(fq) && !(fq->fq_flags &
				    (FQF_NEW_FLOW | FQF_OLD_FLOW))) {
					fq_if_destroy_flow(fqs, fq_cl, fq);
					fq = NULL;
				}
			} else {
				fq_if_drop_packet(fqs);
			}
		}
	}

	if (droptype == DTYPE_NODROP) {
		uint32_t pkt_len = pktsched_get_pkt_len(pkt);
		fq_enqueue(fq, pkt->pktsched_pkt);
		fq->fq_bytes += pkt_len;
		fq_cl->fcl_stat.fcl_byte_cnt += pkt_len;
		fq_cl->fcl_stat.fcl_pkt_cnt++;

		/*
		 * check if this queue will qualify to be the next
		 * victim queue
		 */
		fq_if_is_flow_heavy(fqs, fq);
	} else {
		IFCQ_CONVERT_LOCK(fqs->fqs_ifq);
		return ((ret != CLASSQEQ_SUCCESS) ? ret : CLASSQEQ_DROP);
	}

	/*
	 * If the queue is not currently active, add it to the end of new
	 * flows list for that service class.
	 */
	if ((fq->fq_flags & (FQF_NEW_FLOW|FQF_OLD_FLOW)) == 0) {
		VERIFY(STAILQ_NEXT(fq, fq_actlink) == NULL);
		STAILQ_INSERT_TAIL(&fq_cl->fcl_new_flows, fq, fq_actlink);
		fq->fq_flags |= FQF_NEW_FLOW;

		fq_cl->fcl_stat.fcl_newflows_cnt++;

		fq->fq_deficit = fq_cl->fcl_quantum;
	}
	return (ret);
}

void *
fq_getq_flow_internal(fq_if_t *fqs, fq_t *fq, pktsched_pkt_t *pkt)
{
	void *p;
	uint32_t plen;
	fq_if_classq_t *fq_cl;
	struct ifclassq *ifq = fqs->fqs_ifq;

	fq_dequeue(fq, p);
	if (p == NULL)
		return (NULL);

	pktsched_pkt_encap(pkt, fq->fq_ptype, p);
	plen = pktsched_get_pkt_len(pkt);

	VERIFY(fq->fq_bytes >= plen);
	fq->fq_bytes -= plen;

	fq_cl = &fqs->fqs_classq[fq->fq_sc_index];
	fq_cl->fcl_stat.fcl_byte_cnt -= plen;
	fq_cl->fcl_stat.fcl_pkt_cnt--;
	IFCQ_DEC_LEN(ifq);
	IFCQ_DEC_BYTES(ifq, plen);

	/* Reset getqtime so that we don't count idle times */
	if (fq_empty(fq))
		fq->fq_getqtime = 0;

	return (p);
}

void *
fq_getq_flow(fq_if_t *fqs, fq_t *fq, pktsched_pkt_t *pkt)
{
	void *p;
	fq_if_classq_t *fq_cl;
	u_int64_t now;
	int64_t qdelay = 0;
	struct timespec now_ts;
	uint32_t *pkt_flags, pkt_tx_start_seq;
	uint64_t *pkt_timestamp;

	p = fq_getq_flow_internal(fqs, fq, pkt);
	if (p == NULL)
		return (NULL);

	pktsched_get_pkt_vars(pkt, &pkt_flags, &pkt_timestamp, NULL, NULL,
	    NULL, &pkt_tx_start_seq);

	nanouptime(&now_ts);
	now = (now_ts.tv_sec * NSEC_PER_SEC) + now_ts.tv_nsec;

	/* this will compute qdelay in nanoseconds */
	if (now > *pkt_timestamp)
		qdelay = now - *pkt_timestamp;
	fq_cl = &fqs->fqs_classq[fq->fq_sc_index];

	if (fq->fq_min_qdelay == 0 ||
	    (qdelay > 0 && (u_int64_t)qdelay < fq->fq_min_qdelay))
		fq->fq_min_qdelay = qdelay;
	if (now >= fq->fq_updatetime) {
		if (fq->fq_min_qdelay > fqs->fqs_target_qdelay) {
			if (!FQ_IS_DELAYHIGH(fq))
				FQ_SET_DELAY_HIGH(fq);
		} else {
			FQ_CLEAR_DELAY_HIGH(fq);
		}


		/* Reset measured queue delay and update time */
		fq->fq_updatetime = now + fqs->fqs_update_interval;
		fq->fq_min_qdelay = 0;
	}
	if (!FQ_IS_DELAYHIGH(fq) || fq_empty(fq)) {
		FQ_CLEAR_DELAY_HIGH(fq);
		if (fq->fq_flags & FQF_FLOWCTL_ON) {
			fq_if_flow_feedback(fqs, fq, fq_cl);
		}
	}

	if (fq_empty(fq)) {
		/* Reset getqtime so that we don't count idle times */
		fq->fq_getqtime = 0;
	} else {
		fq->fq_getqtime = now;
	}
	fq_if_is_flow_heavy(fqs, fq);

	*pkt_timestamp = 0;
	if (pkt->pktsched_ptype == QP_MBUF)
		*pkt_flags &= ~PKTF_PRIV_GUARDED;

	return (p);
}
Commit	Line	Data
39037602	1	/*
5ba3f43e	2	* Copyright (c) 2016-2017 Apple Inc. All rights reserved.
39037602 A	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	#include <sys/cdefs.h>
	30	#include <sys/param.h>
	31	#include <sys/mbuf.h>
	32	#include <sys/socket.h>
	33	#include <sys/sockio.h>
	34	#include <sys/systm.h>
	35	#include <sys/sysctl.h>
	36	#include <sys/syslog.h>
	37	#include <sys/proc.h>
	38	#include <sys/errno.h>
	39	#include <sys/kernel.h>
	40	#include <sys/kauth.h>
	41	#include <kern/zalloc.h>
	42	#include <netinet/in.h>
	43
5ba3f43e A	44	#include <net/classq/classq.h>
	45	#include <net/classq/if_classq.h>
	46	#include <net/pktsched/pktsched.h>
39037602 A	47	#include <net/pktsched/pktsched_fq_codel.h>
	48	#include <net/classq/classq_fq_codel.h>
	49
a39ff7e2 A	50	static uint32_t flowq_size; /* size of flowq */
a39ff7e2 A	51	static struct mcache flowq_cache = NULL; / mcache for flowq */
39037602 A	52
39037602 A	53	#define FQ_ZONE_MAX (32 * 1024) /* across all interfaces */
5ba3f43e A	54
	55	#define DTYPE_NODROP 0 /* no drop */
	56	#define DTYPE_FORCED 1 /* a "forced" drop */
	57	#define DTYPE_EARLY 2 /* an "unforced" (early) drop */
39037602 A	58
	59	void
	60	fq_codel_init(void)
	61	{
a39ff7e2	62	if (flowq_cache != NULL)
39037602 A	63	return;
39037602 A	64
a39ff7e2 A	65	flowq_size = sizeof (fq_t);
	66	flowq_cache = mcache_create("fq.flowq", flowq_size, sizeof (uint64_t),
	67	0, MCR_SLEEP);
	68	if (flowq_cache == NULL) {
	69	panic("%s: failed to allocate flowq_cache", __func__);
39037602 A	70	/* NOTREACHED */
39037602 A	71	}
a39ff7e2 A	72	}
	73
	74	void
	75	fq_codel_reap_caches(boolean_t purge)
	76	{
	77	mcache_reap_now(flowq_cache, purge);
39037602 A	78	}
	79
	80	fq_t *
5ba3f43e	81	fq_alloc(classq_pkt_type_t ptype)
39037602 A	82	{
39037602 A	83	fq_t *fq = NULL;
a39ff7e2	84	fq = mcache_alloc(flowq_cache, MCR_SLEEP);
39037602	85	if (fq == NULL) {
a39ff7e2	86	log(LOG_ERR, "%s: unable to allocate from flowq_cache\n");
39037602 A	87	return (NULL);
	88	}
	89
a39ff7e2	90	bzero(fq, flowq_size);
5ba3f43e A	91	fq->fq_ptype = ptype;
	92	if (ptype == QP_MBUF) {
	93	MBUFQ_INIT(&fq->fq_mbufq);
	94	}
39037602 A	95	return (fq);
	96	}
	97
	98	void
	99	fq_destroy(fq_t *fq)
	100	{
5ba3f43e	101	VERIFY(fq_empty(fq));
39037602	102	VERIFY(!(fq->fq_flags & (FQF_NEW_FLOW \| FQF_OLD_FLOW)));
5ba3f43e	103	VERIFY(fq->fq_bytes == 0);
a39ff7e2	104	mcache_free(flowq_cache, fq);
39037602 A	105	}
	106
	107	static void
	108	fq_detect_dequeue_stall(fq_if_t fqs, fq_t flowq, fq_if_classq_t *fq_cl,
	109	u_int64_t *now)
	110	{
	111	u_int64_t maxgetqtime;
	112	if (FQ_IS_DELAYHIGH(flowq) \|\| flowq->fq_getqtime == 0 \|\|
5ba3f43e	113	fq_empty(flowq) \|\|
39037602 A	114	flowq->fq_bytes < FQ_MIN_FC_THRESHOLD_BYTES)
	115	return;
	116	maxgetqtime = flowq->fq_getqtime + fqs->fqs_update_interval;
	117	if ((*now) > maxgetqtime) {
	118	/*
	119	* there was no dequeue in an update interval worth of
	120	* time. It means that the queue is stalled.
	121	*/
	122	FQ_SET_DELAY_HIGH(flowq);
	123	fq_cl->fcl_stat.fcl_dequeue_stall++;
	124	}
	125	}
	126
	127	void
	128	fq_head_drop(fq_if_t fqs, fq_t fq)
	129	{
5ba3f43e A	130	pktsched_pkt_t pkt;
	131	uint32_t *pkt_flags;
	132	uint64_t *pkt_timestamp;
39037602 A	133	struct ifclassq *ifq = fqs->fqs_ifq;
39037602 A	134
5ba3f43e A	135	_PKTSCHED_PKT_INIT(&pkt);
5ba3f43e A	136	if (fq_getq_flow_internal(fqs, fq, &pkt) == NULL)
39037602 A	137	return;
39037602 A	138
5ba3f43e A	139	pktsched_get_pkt_vars(&pkt, &pkt_flags, &pkt_timestamp, NULL, NULL,
	140	NULL, NULL);
	141
	142	*pkt_timestamp = 0;
	143	if (pkt.pktsched_ptype == QP_MBUF)
	144	*pkt_flags &= ~PKTF_PRIV_GUARDED;
	145
	146	IFCQ_DROP_ADD(ifq, 1, pktsched_get_pkt_len(&pkt));
39037602	147	IFCQ_CONVERT_LOCK(ifq);
5ba3f43e	148	pktsched_free_pkt(&pkt);
39037602 A	149	}
	150
	151	int
5ba3f43e	152	fq_addq(fq_if_t fqs, pktsched_pkt_t pkt, fq_if_classq_t *fq_cl)
39037602	153	{
39037602 A	154	int droptype = DTYPE_NODROP, fc_adv = 0, ret = CLASSQEQ_SUCCESS;
	155	u_int64_t now;
	156	fq_t *fq = NULL;
5ba3f43e A	157	uint64_t *pkt_timestamp;
	158	uint32_t *pkt_flags;
	159	uint32_t pkt_flowid, pkt_tx_start_seq;
	160	uint8_t pkt_proto, pkt_flowsrc;
	161
	162	pktsched_get_pkt_vars(pkt, &pkt_flags, &pkt_timestamp, &pkt_flowid,
	163	&pkt_flowsrc, &pkt_proto, &pkt_tx_start_seq);
	164
	165	if (pkt->pktsched_ptype == QP_MBUF) {
	166	/* See comments in <rdar://problem/14040693> */
	167	VERIFY(!(*pkt_flags & PKTF_PRIV_GUARDED));
	168	*pkt_flags \|= PKTF_PRIV_GUARDED;
	169	}
39037602	170
5ba3f43e A	171	if (*pkt_timestamp > 0) {
5ba3f43e A	172	now = *pkt_timestamp;
39037602	173	} else {
5ba3f43e A	174	struct timespec now_ts;
	175	nanouptime(&now_ts);
	176	now = (now_ts.tv_sec * NSEC_PER_SEC) + now_ts.tv_nsec;
	177	*pkt_timestamp = now;
39037602 A	178	}
	179
	180	/* find the flowq for this packet */
5ba3f43e A	181	fq = fq_if_hash_pkt(fqs, pkt_flowid, pktsched_get_pkt_svc(pkt),
5ba3f43e A	182	now, TRUE, pkt->pktsched_ptype);
39037602 A	183	if (fq == NULL) {
	184	/* drop the packet if we could not allocate a flow queue */
	185	fq_cl->fcl_stat.fcl_drop_memfailure++;
	186	IFCQ_CONVERT_LOCK(fqs->fqs_ifq);
5ba3f43e	187	return (CLASSQEQ_DROP);
39037602	188	}
5ba3f43e	189	VERIFY(fq->fq_ptype == pkt->pktsched_ptype);
39037602 A	190
	191	fq_detect_dequeue_stall(fqs, fq, fq_cl, &now);
	192
	193	if (FQ_IS_DELAYHIGH(fq)) {
	194	if ((fq->fq_flags & FQF_FLOWCTL_CAPABLE) &&
5ba3f43e	195	(*pkt_flags & PKTF_FLOW_ADV)) {
39037602 A	196	fc_adv = 1;
	197	/*
	198	* If the flow is suspended or it is not
	199	* TCP, drop the packet
	200	*/
5ba3f43e	201	if (pkt_proto != IPPROTO_TCP) {
39037602 A	202	droptype = DTYPE_EARLY;
	203	fq_cl->fcl_stat.fcl_drop_early++;
	204	}
	205	} else {
	206	/*
	207	* Need to drop a packet, instead of dropping this
	208	* one, try to drop from the head of the queue
	209	*/
5ba3f43e	210	if (!fq_empty(fq)) {
39037602 A	211	fq_head_drop(fqs, fq);
	212	droptype = DTYPE_NODROP;
	213	} else {
	214	droptype = DTYPE_EARLY;
	215	}
	216	fq_cl->fcl_stat.fcl_drop_early++;
	217	}
	218
	219	}
	220
39037602 A	221	/* Set the return code correctly */
39037602 A	222	if (fc_adv == 1 && droptype != DTYPE_FORCED) {
5ba3f43e A	223	if (fq_if_add_fcentry(fqs, pkt, pkt_flowid, pkt_flowsrc,
5ba3f43e A	224	fq_cl)) {
39037602 A	225	fq->fq_flags \|= FQF_FLOWCTL_ON;
	226	/* deliver flow control advisory error */
	227	if (droptype == DTYPE_NODROP) {
	228	ret = CLASSQEQ_SUCCESS_FC;
	229	} else {
	230	/* dropped due to flow control */
5ba3f43e	231	ret = CLASSQEQ_DROP_FC;
39037602 A	232	}
	233	} else {
	234	/*
	235	* if we could not flow control the flow, it is
	236	* better to drop
	237	*/
	238	droptype = DTYPE_FORCED;
5ba3f43e	239	ret = CLASSQEQ_DROP_FC;
39037602 A	240	fq_cl->fcl_stat.fcl_flow_control_fail++;
	241	}
	242	}
	243
	244	/*
	245	* If the queue length hits the queue limit, drop a packet from the
	246	* front of the queue for a flow with maximum number of bytes. This
	247	* will penalize heavy and unresponsive flows. It will also avoid a
	248	* tail drop.
	249	*/
	250	if (droptype == DTYPE_NODROP && fq_if_at_drop_limit(fqs)) {
5ba3f43e A	251	if (fqs->fqs_large_flow == fq) {
	252	/*
	253	* Drop from the head of the current fq. Since a
	254	* new packet will be added to the tail, it is ok
	255	* to leave fq in place.
	256	*/
	257	fq_head_drop(fqs, fq);
	258	} else {
	259	if (fqs->fqs_large_flow == NULL) {
	260	droptype = DTYPE_FORCED;
	261	fq_cl->fcl_stat.fcl_drop_overflow++;
5c9f4661	262	ret = CLASSQEQ_DROP;
5ba3f43e A	263
	264	/*
	265	* if this fq was freshly created and there
	266	* is nothing to enqueue, free it
	267	*/
	268	if (fq_empty(fq) && !(fq->fq_flags &
	269	(FQF_NEW_FLOW \| FQF_OLD_FLOW))) {
	270	fq_if_destroy_flow(fqs, fq_cl, fq);
	271	fq = NULL;
	272	}
	273	} else {
	274	fq_if_drop_packet(fqs);
	275	}
	276	}
39037602 A	277	}
	278
	279	if (droptype == DTYPE_NODROP) {
5ba3f43e A	280	uint32_t pkt_len = pktsched_get_pkt_len(pkt);
	281	fq_enqueue(fq, pkt->pktsched_pkt);
	282	fq->fq_bytes += pkt_len;
	283	fq_cl->fcl_stat.fcl_byte_cnt += pkt_len;
39037602 A	284	fq_cl->fcl_stat.fcl_pkt_cnt++;
	285
	286	/*
	287	* check if this queue will qualify to be the next
	288	* victim queue
	289	*/
	290	fq_if_is_flow_heavy(fqs, fq);
	291	} else {
	292	IFCQ_CONVERT_LOCK(fqs->fqs_ifq);
5ba3f43e	293	return ((ret != CLASSQEQ_SUCCESS) ? ret : CLASSQEQ_DROP);
39037602 A	294	}
	295
	296	/*
	297	* If the queue is not currently active, add it to the end of new
	298	* flows list for that service class.
	299	*/
	300	if ((fq->fq_flags & (FQF_NEW_FLOW\|FQF_OLD_FLOW)) == 0) {
	301	VERIFY(STAILQ_NEXT(fq, fq_actlink) == NULL);
	302	STAILQ_INSERT_TAIL(&fq_cl->fcl_new_flows, fq, fq_actlink);
	303	fq->fq_flags \|= FQF_NEW_FLOW;
	304
	305	fq_cl->fcl_stat.fcl_newflows_cnt++;
	306
	307	fq->fq_deficit = fq_cl->fcl_quantum;
	308	}
	309	return (ret);
	310	}
	311
5ba3f43e A	312	void *
5ba3f43e A	313	fq_getq_flow_internal(fq_if_t fqs, fq_t fq, pktsched_pkt_t *pkt)
39037602	314	{
5ba3f43e A	315	void *p;
5ba3f43e A	316	uint32_t plen;
39037602	317	fq_if_classq_t *fq_cl;
5ba3f43e	318	struct ifclassq *ifq = fqs->fqs_ifq;
39037602	319
5ba3f43e A	320	fq_dequeue(fq, p);
5ba3f43e A	321	if (p == NULL)
39037602 A	322	return (NULL);
39037602 A	323
5ba3f43e A	324	pktsched_pkt_encap(pkt, fq->fq_ptype, p);
5ba3f43e A	325	plen = pktsched_get_pkt_len(pkt);
39037602	326
5ba3f43e A	327	VERIFY(fq->fq_bytes >= plen);
5ba3f43e A	328	fq->fq_bytes -= plen;
39037602 A	329
39037602 A	330	fq_cl = &fqs->fqs_classq[fq->fq_sc_index];
5ba3f43e	331	fq_cl->fcl_stat.fcl_byte_cnt -= plen;
39037602 A	332	fq_cl->fcl_stat.fcl_pkt_cnt--;
39037602 A	333	IFCQ_DEC_LEN(ifq);
5ba3f43e	334	IFCQ_DEC_BYTES(ifq, plen);
39037602	335
5ba3f43e A	336	/* Reset getqtime so that we don't count idle times */
	337	if (fq_empty(fq))
	338	fq->fq_getqtime = 0;
	339
	340	return (p);
	341	}
	342
	343	void *
	344	fq_getq_flow(fq_if_t fqs, fq_t fq, pktsched_pkt_t *pkt)
	345	{
	346	void *p;
	347	fq_if_classq_t *fq_cl;
	348	u_int64_t now;
	349	int64_t qdelay = 0;
	350	struct timespec now_ts;
	351	uint32_t *pkt_flags, pkt_tx_start_seq;
	352	uint64_t *pkt_timestamp;
	353
	354	p = fq_getq_flow_internal(fqs, fq, pkt);
	355	if (p == NULL)
	356	return (NULL);
	357
	358	pktsched_get_pkt_vars(pkt, &pkt_flags, &pkt_timestamp, NULL, NULL,
	359	NULL, &pkt_tx_start_seq);
	360
	361	nanouptime(&now_ts);
	362	now = (now_ts.tv_sec * NSEC_PER_SEC) + now_ts.tv_nsec;
39037602 A	363
39037602 A	364	/* this will compute qdelay in nanoseconds */
5ba3f43e A	365	if (now > *pkt_timestamp)
	366	qdelay = now - *pkt_timestamp;
	367	fq_cl = &fqs->fqs_classq[fq->fq_sc_index];
39037602 A	368
	369	if (fq->fq_min_qdelay == 0 \|\|
	370	(qdelay > 0 && (u_int64_t)qdelay < fq->fq_min_qdelay))
	371	fq->fq_min_qdelay = qdelay;
5ba3f43e A	372	if (now >= fq->fq_updatetime) {
5ba3f43e A	373	if (fq->fq_min_qdelay > fqs->fqs_target_qdelay) {
39037602 A	374	if (!FQ_IS_DELAYHIGH(fq))
39037602 A	375	FQ_SET_DELAY_HIGH(fq);
5ba3f43e	376	} else {
39037602	377	FQ_CLEAR_DELAY_HIGH(fq);
39037602 A	378	}
39037602 A	379
5ba3f43e	380
39037602 A	381	/* Reset measured queue delay and update time */
	382	fq->fq_updatetime = now + fqs->fqs_update_interval;
	383	fq->fq_min_qdelay = 0;
	384	}
5ba3f43e A	385	if (!FQ_IS_DELAYHIGH(fq) \|\| fq_empty(fq)) {
	386	FQ_CLEAR_DELAY_HIGH(fq);
	387	if (fq->fq_flags & FQF_FLOWCTL_ON) {
	388	fq_if_flow_feedback(fqs, fq, fq_cl);
	389	}
	390	}
39037602	391
5ba3f43e	392	if (fq_empty(fq)) {
39037602 A	393	/* Reset getqtime so that we don't count idle times */
	394	fq->fq_getqtime = 0;
	395	} else {
	396	fq->fq_getqtime = now;
	397	}
5ba3f43e A	398	fq_if_is_flow_heavy(fqs, fq);
	399
	400	*pkt_timestamp = 0;
	401	if (pkt->pktsched_ptype == QP_MBUF)
	402	*pkt_flags &= ~PKTF_PRIV_GUARDED;
39037602	403
5ba3f43e	404	return (p);
39037602	405	}