[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 struct zone *flowq_zone = NULL;

#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_zone != NULL)
		return;

	flowq_zone = zinit(sizeof (struct flowq),
	    FQ_ZONE_MAX * sizeof (struct flowq), 0, "flowq_zone");
	if (flowq_zone == NULL) {
		panic("%s: failed to allocate flowq_zone", __func__);
		/* NOTREACHED */
	}
	zone_change(flowq_zone, Z_EXPAND, TRUE);
	zone_change(flowq_zone, Z_CALLERACCT, TRUE);
}

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

	bzero(fq, sizeof (*fq));
	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);
	zfree(flowq_zone, 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++;

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