Loading include/linux/pkt_sched.h +24 −26 Original line number Diff line number Diff line Loading @@ -93,6 +93,7 @@ struct tc_fifo_qopt /* PRIO section */ #define TCQ_PRIO_BANDS 16 #define TCQ_MIN_PRIO_BANDS 2 struct tc_prio_qopt { Loading Loading @@ -169,6 +170,7 @@ struct tc_red_qopt unsigned char Scell_log; /* cell size for idle damping */ unsigned char flags; #define TC_RED_ECN 1 #define TC_RED_HARDDROP 2 }; struct tc_red_xstats Loading @@ -194,15 +196,11 @@ enum #define TCA_GRED_MAX (__TCA_GRED_MAX - 1) #define TCA_SET_OFF TCA_GRED_PARMS struct tc_gred_qopt { __u32 limit; /* HARD maximal queue length (bytes) */ __u32 qth_min; /* Min average length threshold (bytes) */ __u32 qth_max; /* Max average length threshold (bytes) */ __u32 limit; /* HARD maximal queue length (bytes) */ __u32 qth_min; /* Min average length threshold (bytes) */ __u32 qth_max; /* Max average length threshold (bytes) */ __u32 DP; /* upto 2^32 DPs */ __u32 backlog; __u32 qave; Loading @@ -210,22 +208,22 @@ struct tc_gred_qopt __u32 early; __u32 other; __u32 pdrop; unsigned char Wlog; /* log(W) */ unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ unsigned char Scell_log; /* cell size for idle damping */ __u8 Wlog; /* log(W) */ __u8 Plog; /* log(P_max/(qth_max-qth_min)) */ __u8 Scell_log; /* cell size for idle damping */ __u8 prio; /* prio of this VQ */ __u32 packets; __u32 bytesin; }; /* gred setup */ struct tc_gred_sopt { __u32 DPs; __u32 def_DP; __u8 grio; __u8 pad1; __u16 pad2; __u8 flags; __u16 pad1; }; /* HTB section */ Loading include/net/inet_ecn.h +24 −4 Original line number Diff line number Diff line Loading @@ -2,6 +2,7 @@ #define _INET_ECN_H_ #include <linux/ip.h> #include <linux/skbuff.h> #include <net/dsfield.h> enum { Loading Loading @@ -48,7 +49,7 @@ static inline __u8 INET_ECN_encapsulate(__u8 outer, __u8 inner) (label) |= __constant_htons(INET_ECN_ECT_0 << 4); \ } while (0) static inline void IP_ECN_set_ce(struct iphdr *iph) static inline int IP_ECN_set_ce(struct iphdr *iph) { u32 check = iph->check; u32 ecn = (iph->tos + 1) & INET_ECN_MASK; Loading @@ -61,7 +62,7 @@ static inline void IP_ECN_set_ce(struct iphdr *iph) * INET_ECN_CE => 00 */ if (!(ecn & 2)) return; return !ecn; /* * The following gives us: Loading @@ -72,6 +73,7 @@ static inline void IP_ECN_set_ce(struct iphdr *iph) iph->check = check + (check>=0xFFFF); iph->tos |= INET_ECN_CE; return 1; } static inline void IP_ECN_clear(struct iphdr *iph) Loading @@ -87,11 +89,12 @@ static inline void ipv4_copy_dscp(struct iphdr *outer, struct iphdr *inner) struct ipv6hdr; static inline void IP6_ECN_set_ce(struct ipv6hdr *iph) static inline int IP6_ECN_set_ce(struct ipv6hdr *iph) { if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph))) return; return 0; *(u32*)iph |= htonl(INET_ECN_CE << 20); return 1; } static inline void IP6_ECN_clear(struct ipv6hdr *iph) Loading @@ -105,4 +108,21 @@ static inline void ipv6_copy_dscp(struct ipv6hdr *outer, struct ipv6hdr *inner) ipv6_change_dsfield(inner, INET_ECN_MASK, dscp); } static inline int INET_ECN_set_ce(struct sk_buff *skb) { switch (skb->protocol) { case __constant_htons(ETH_P_IP): if (skb->nh.raw + sizeof(struct iphdr) <= skb->tail) return IP_ECN_set_ce(skb->nh.iph); break; case __constant_htons(ETH_P_IPV6): if (skb->nh.raw + sizeof(struct ipv6hdr) <= skb->tail) return IP6_ECN_set_ce(skb->nh.ipv6h); break; } return 0; } #endif include/net/red.h 0 → 100644 +325 −0 Original line number Diff line number Diff line #ifndef __NET_SCHED_RED_H #define __NET_SCHED_RED_H #include <linux/config.h> #include <linux/types.h> #include <net/pkt_sched.h> #include <net/inet_ecn.h> #include <net/dsfield.h> /* Random Early Detection (RED) algorithm. ======================================= Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking. This file codes a "divisionless" version of RED algorithm as written down in Fig.17 of the paper. Short description. ------------------ When a new packet arrives we calculate the average queue length: avg = (1-W)*avg + W*current_queue_len, W is the filter time constant (chosen as 2^(-Wlog)), it controls the inertia of the algorithm. To allow larger bursts, W should be decreased. if (avg > th_max) -> packet marked (dropped). if (avg < th_min) -> packet passes. if (th_min < avg < th_max) we calculate probability: Pb = max_P * (avg - th_min)/(th_max-th_min) and mark (drop) packet with this probability. Pb changes from 0 (at avg==th_min) to max_P (avg==th_max). max_P should be small (not 1), usually 0.01..0.02 is good value. max_P is chosen as a number, so that max_P/(th_max-th_min) is a negative power of two in order arithmetics to contain only shifts. Parameters, settable by user: ----------------------------- qth_min - bytes (should be < qth_max/2) qth_max - bytes (should be at least 2*qth_min and less limit) Wlog - bits (<32) log(1/W). Plog - bits (<32) Plog is related to max_P by formula: max_P = (qth_max-qth_min)/2^Plog; F.e. if qth_max=128K and qth_min=32K, then Plog=22 corresponds to max_P=0.02 Scell_log Stab Lookup table for log((1-W)^(t/t_ave). NOTES: Upper bound on W. ----------------- If you want to allow bursts of L packets of size S, you should choose W: L + 1 - th_min/S < (1-(1-W)^L)/W th_min/S = 32 th_min/S = 4 log(W) L -1 33 -2 35 -3 39 -4 46 -5 57 -6 75 -7 101 -8 135 -9 190 etc. */ #define RED_STAB_SIZE 256 #define RED_STAB_MASK (RED_STAB_SIZE - 1) struct red_stats { u32 prob_drop; /* Early probability drops */ u32 prob_mark; /* Early probability marks */ u32 forced_drop; /* Forced drops, qavg > max_thresh */ u32 forced_mark; /* Forced marks, qavg > max_thresh */ u32 pdrop; /* Drops due to queue limits */ u32 other; /* Drops due to drop() calls */ u32 backlog; }; struct red_parms { /* Parameters */ u32 qth_min; /* Min avg length threshold: A scaled */ u32 qth_max; /* Max avg length threshold: A scaled */ u32 Scell_max; u32 Rmask; /* Cached random mask, see red_rmask */ u8 Scell_log; u8 Wlog; /* log(W) */ u8 Plog; /* random number bits */ u8 Stab[RED_STAB_SIZE]; /* Variables */ int qcount; /* Number of packets since last random number generation */ u32 qR; /* Cached random number */ unsigned long qavg; /* Average queue length: A scaled */ psched_time_t qidlestart; /* Start of current idle period */ }; static inline u32 red_rmask(u8 Plog) { return Plog < 32 ? ((1 << Plog) - 1) : ~0UL; } static inline void red_set_parms(struct red_parms *p, u32 qth_min, u32 qth_max, u8 Wlog, u8 Plog, u8 Scell_log, u8 *stab) { /* Reset average queue length, the value is strictly bound * to the parameters below, reseting hurts a bit but leaving * it might result in an unreasonable qavg for a while. --TGR */ p->qavg = 0; p->qcount = -1; p->qth_min = qth_min << Wlog; p->qth_max = qth_max << Wlog; p->Wlog = Wlog; p->Plog = Plog; p->Rmask = red_rmask(Plog); p->Scell_log = Scell_log; p->Scell_max = (255 << Scell_log); memcpy(p->Stab, stab, sizeof(p->Stab)); } static inline int red_is_idling(struct red_parms *p) { return !PSCHED_IS_PASTPERFECT(p->qidlestart); } static inline void red_start_of_idle_period(struct red_parms *p) { PSCHED_GET_TIME(p->qidlestart); } static inline void red_end_of_idle_period(struct red_parms *p) { PSCHED_SET_PASTPERFECT(p->qidlestart); } static inline void red_restart(struct red_parms *p) { red_end_of_idle_period(p); p->qavg = 0; p->qcount = -1; } static inline unsigned long red_calc_qavg_from_idle_time(struct red_parms *p) { psched_time_t now; long us_idle; int shift; PSCHED_GET_TIME(now); us_idle = PSCHED_TDIFF_SAFE(now, p->qidlestart, p->Scell_max); /* * The problem: ideally, average length queue recalcultion should * be done over constant clock intervals. This is too expensive, so * that the calculation is driven by outgoing packets. * When the queue is idle we have to model this clock by hand. * * SF+VJ proposed to "generate": * * m = idletime / (average_pkt_size / bandwidth) * * dummy packets as a burst after idle time, i.e. * * p->qavg *= (1-W)^m * * This is an apparently overcomplicated solution (f.e. we have to * precompute a table to make this calculation in reasonable time) * I believe that a simpler model may be used here, * but it is field for experiments. */ shift = p->Stab[(us_idle >> p->Scell_log) & RED_STAB_MASK]; if (shift) return p->qavg >> shift; else { /* Approximate initial part of exponent with linear function: * * (1-W)^m ~= 1-mW + ... * * Seems, it is the best solution to * problem of too coarse exponent tabulation. */ us_idle = (p->qavg * us_idle) >> p->Scell_log; if (us_idle < (p->qavg >> 1)) return p->qavg - us_idle; else return p->qavg >> 1; } } static inline unsigned long red_calc_qavg_no_idle_time(struct red_parms *p, unsigned int backlog) { /* * NOTE: p->qavg is fixed point number with point at Wlog. * The formula below is equvalent to floating point * version: * * qavg = qavg*(1-W) + backlog*W; * * --ANK (980924) */ return p->qavg + (backlog - (p->qavg >> p->Wlog)); } static inline unsigned long red_calc_qavg(struct red_parms *p, unsigned int backlog) { if (!red_is_idling(p)) return red_calc_qavg_no_idle_time(p, backlog); else return red_calc_qavg_from_idle_time(p); } static inline u32 red_random(struct red_parms *p) { return net_random() & p->Rmask; } static inline int red_mark_probability(struct red_parms *p, unsigned long qavg) { /* The formula used below causes questions. OK. qR is random number in the interval 0..Rmask i.e. 0..(2^Plog). If we used floating point arithmetics, it would be: (2^Plog)*rnd_num, where rnd_num is less 1. Taking into account, that qavg have fixed point at Wlog, and Plog is related to max_P by max_P = (qth_max-qth_min)/2^Plog; two lines below have the following floating point equivalent: max_P*(qavg - qth_min)/(qth_max-qth_min) < rnd/qcount Any questions? --ANK (980924) */ return !(((qavg - p->qth_min) >> p->Wlog) * p->qcount < p->qR); } enum { RED_BELOW_MIN_THRESH, RED_BETWEEN_TRESH, RED_ABOVE_MAX_TRESH, }; static inline int red_cmp_thresh(struct red_parms *p, unsigned long qavg) { if (qavg < p->qth_min) return RED_BELOW_MIN_THRESH; else if (qavg >= p->qth_max) return RED_ABOVE_MAX_TRESH; else return RED_BETWEEN_TRESH; } enum { RED_DONT_MARK, RED_PROB_MARK, RED_HARD_MARK, }; static inline int red_action(struct red_parms *p, unsigned long qavg) { switch (red_cmp_thresh(p, qavg)) { case RED_BELOW_MIN_THRESH: p->qcount = -1; return RED_DONT_MARK; case RED_BETWEEN_TRESH: if (++p->qcount) { if (red_mark_probability(p, qavg)) { p->qcount = 0; p->qR = red_random(p); return RED_PROB_MARK; } } else p->qR = red_random(p); return RED_DONT_MARK; case RED_ABOVE_MAX_TRESH: p->qcount = -1; return RED_HARD_MARK; } BUG(); return RED_DONT_MARK; } #endif net/sched/sch_gred.c +411 −430 File changed.Preview size limit exceeded, changes collapsed. Show changes net/sched/sch_red.c +107 −311 Original line number Diff line number Diff line Loading @@ -9,76 +9,23 @@ * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> * * Changes: * J Hadi Salim <hadi@nortel.com> 980914: computation fixes * J Hadi Salim 980914: computation fixes * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly. * J Hadi Salim <hadi@nortelnetworks.com> 980816: ECN support * J Hadi Salim 980816: ECN support */ #include <linux/config.h> #include <linux/module.h> #include <asm/uaccess.h> #include <asm/system.h> #include <linux/bitops.h> #include <linux/types.h> #include <linux/kernel.h> #include <linux/sched.h> #include <linux/string.h> #include <linux/mm.h> #include <linux/socket.h> #include <linux/sockios.h> #include <linux/in.h> #include <linux/errno.h> #include <linux/interrupt.h> #include <linux/if_ether.h> #include <linux/inet.h> #include <linux/netdevice.h> #include <linux/etherdevice.h> #include <linux/notifier.h> #include <net/ip.h> #include <net/route.h> #include <linux/skbuff.h> #include <net/sock.h> #include <net/pkt_sched.h> #include <net/inet_ecn.h> #include <net/dsfield.h> #include <net/red.h> /* Random Early Detection (RED) algorithm. ======================================= Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking. This file codes a "divisionless" version of RED algorithm as written down in Fig.17 of the paper. Short description. ------------------ When a new packet arrives we calculate the average queue length: avg = (1-W)*avg + W*current_queue_len, W is the filter time constant (chosen as 2^(-Wlog)), it controls the inertia of the algorithm. To allow larger bursts, W should be decreased. if (avg > th_max) -> packet marked (dropped). if (avg < th_min) -> packet passes. if (th_min < avg < th_max) we calculate probability: Pb = max_P * (avg - th_min)/(th_max-th_min) and mark (drop) packet with this probability. Pb changes from 0 (at avg==th_min) to max_P (avg==th_max). max_P should be small (not 1), usually 0.01..0.02 is good value. max_P is chosen as a number, so that max_P/(th_max-th_min) is a negative power of two in order arithmetics to contain only shifts. Parameters, settable by user: /* Parameters, settable by user: ----------------------------- limit - bytes (must be > qth_max + burst) Loading @@ -89,260 +36,111 @@ Short description. arbitrarily high (well, less than ram size) Really, this limit will never be reached if RED works correctly. qth_min - bytes (should be < qth_max/2) qth_max - bytes (should be at least 2*qth_min and less limit) Wlog - bits (<32) log(1/W). Plog - bits (<32) Plog is related to max_P by formula: max_P = (qth_max-qth_min)/2^Plog; F.e. if qth_max=128K and qth_min=32K, then Plog=22 corresponds to max_P=0.02 Scell_log Stab Lookup table for log((1-W)^(t/t_ave). NOTES: Upper bound on W. ----------------- If you want to allow bursts of L packets of size S, you should choose W: L + 1 - th_min/S < (1-(1-W)^L)/W th_min/S = 32 th_min/S = 4 log(W) L -1 33 -2 35 -3 39 -4 46 -5 57 -6 75 -7 101 -8 135 -9 190 etc. */ struct red_sched_data { /* Parameters */ u32 limit; /* HARD maximal queue length */ u32 qth_min; /* Min average length threshold: A scaled */ u32 qth_max; /* Max average length threshold: A scaled */ u32 Rmask; u32 Scell_max; unsigned char flags; char Wlog; /* log(W) */ char Plog; /* random number bits */ char Scell_log; u8 Stab[256]; /* Variables */ unsigned long qave; /* Average queue length: A scaled */ int qcount; /* Packets since last random number generation */ u32 qR; /* Cached random number */ psched_time_t qidlestart; /* Start of idle period */ struct tc_red_xstats st; struct red_parms parms; struct red_stats stats; }; static int red_ecn_mark(struct sk_buff *skb) static inline int red_use_ecn(struct red_sched_data *q) { if (skb->nh.raw + 20 > skb->tail) return 0; switch (skb->protocol) { case __constant_htons(ETH_P_IP): if (INET_ECN_is_not_ect(skb->nh.iph->tos)) return 0; IP_ECN_set_ce(skb->nh.iph); return 1; case __constant_htons(ETH_P_IPV6): if (INET_ECN_is_not_ect(ipv6_get_dsfield(skb->nh.ipv6h))) return 0; IP6_ECN_set_ce(skb->nh.ipv6h); return 1; default: return 0; return q->flags & TC_RED_ECN; } static inline int red_use_harddrop(struct red_sched_data *q) { return q->flags & TC_RED_HARDDROP; } static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch) static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch) { struct red_sched_data *q = qdisc_priv(sch); psched_time_t now; if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) { long us_idle; int shift; PSCHED_GET_TIME(now); us_idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max); PSCHED_SET_PASTPERFECT(q->qidlestart); /* The problem: ideally, average length queue recalcultion should be done over constant clock intervals. This is too expensive, so that the calculation is driven by outgoing packets. When the queue is idle we have to model this clock by hand. SF+VJ proposed to "generate" m = idletime/(average_pkt_size/bandwidth) dummy packets as a burst after idle time, i.e. q->qave *= (1-W)^m q->parms.qavg = red_calc_qavg(&q->parms, sch->qstats.backlog); This is an apparently overcomplicated solution (f.e. we have to precompute a table to make this calculation in reasonable time) I believe that a simpler model may be used here, but it is field for experiments. */ shift = q->Stab[us_idle>>q->Scell_log]; if (red_is_idling(&q->parms)) red_end_of_idle_period(&q->parms); if (shift) { q->qave >>= shift; } else { /* Approximate initial part of exponent with linear function: (1-W)^m ~= 1-mW + ... Seems, it is the best solution to problem of too coarce exponent tabulation. */ us_idle = (q->qave * us_idle)>>q->Scell_log; if (us_idle < q->qave/2) q->qave -= us_idle; else q->qave >>= 1; } } else { q->qave += sch->qstats.backlog - (q->qave >> q->Wlog); /* NOTE: q->qave is fixed point number with point at Wlog. The formulae above is equvalent to floating point version: qave = qave*(1-W) + sch->qstats.backlog*W; --ANK (980924) */ } switch (red_action(&q->parms, q->parms.qavg)) { case RED_DONT_MARK: break; if (q->qave < q->qth_min) { q->qcount = -1; enqueue: if (sch->qstats.backlog + skb->len <= q->limit) { __skb_queue_tail(&sch->q, skb); sch->qstats.backlog += skb->len; sch->bstats.bytes += skb->len; sch->bstats.packets++; return NET_XMIT_SUCCESS; } else { q->st.pdrop++; } kfree_skb(skb); sch->qstats.drops++; return NET_XMIT_DROP; } if (q->qave >= q->qth_max) { q->qcount = -1; case RED_PROB_MARK: sch->qstats.overlimits++; mark: if (!(q->flags&TC_RED_ECN) || !red_ecn_mark(skb)) { q->st.early++; goto drop; } q->st.marked++; goto enqueue; if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) { q->stats.prob_drop++; goto congestion_drop; } if (++q->qcount) { /* The formula used below causes questions. q->stats.prob_mark++; break; OK. qR is random number in the interval 0..Rmask i.e. 0..(2^Plog). If we used floating point arithmetics, it would be: (2^Plog)*rnd_num, where rnd_num is less 1. case RED_HARD_MARK: sch->qstats.overlimits++; if (red_use_harddrop(q) || !red_use_ecn(q) || !INET_ECN_set_ce(skb)) { q->stats.forced_drop++; goto congestion_drop; } Taking into account, that qave have fixed point at Wlog, and Plog is related to max_P by max_P = (qth_max-qth_min)/2^Plog; two lines below have the following floating point equivalent: q->stats.forced_mark++; break; } max_P*(qave - qth_min)/(qth_max-qth_min) < rnd/qcount if (sch->qstats.backlog + skb->len <= q->limit) return qdisc_enqueue_tail(skb, sch); Any questions? --ANK (980924) */ if (((q->qave - q->qth_min)>>q->Wlog)*q->qcount < q->qR) goto enqueue; q->qcount = 0; q->qR = net_random()&q->Rmask; sch->qstats.overlimits++; goto mark; } q->qR = net_random()&q->Rmask; goto enqueue; q->stats.pdrop++; return qdisc_drop(skb, sch); drop: kfree_skb(skb); sch->qstats.drops++; congestion_drop: qdisc_drop(skb, sch); return NET_XMIT_CN; } static int red_requeue(struct sk_buff *skb, struct Qdisc* sch) static int red_requeue(struct sk_buff *skb, struct Qdisc* sch) { struct red_sched_data *q = qdisc_priv(sch); PSCHED_SET_PASTPERFECT(q->qidlestart); if (red_is_idling(&q->parms)) red_end_of_idle_period(&q->parms); __skb_queue_head(&sch->q, skb); sch->qstats.backlog += skb->len; sch->qstats.requeues++; return 0; return qdisc_requeue(skb, sch); } static struct sk_buff * red_dequeue(struct Qdisc* sch) static struct sk_buff * red_dequeue(struct Qdisc* sch) { struct sk_buff *skb; struct red_sched_data *q = qdisc_priv(sch); skb = __skb_dequeue(&sch->q); if (skb) { sch->qstats.backlog -= skb->len; skb = qdisc_dequeue_head(sch); if (skb == NULL && !red_is_idling(&q->parms)) red_start_of_idle_period(&q->parms); return skb; } PSCHED_GET_TIME(q->qidlestart); return NULL; } static unsigned int red_drop(struct Qdisc* sch) { struct sk_buff *skb; struct red_sched_data *q = qdisc_priv(sch); skb = __skb_dequeue_tail(&sch->q); skb = qdisc_dequeue_tail(sch); if (skb) { unsigned int len = skb->len; sch->qstats.backlog -= len; sch->qstats.drops++; q->st.other++; kfree_skb(skb); q->stats.other++; qdisc_drop(skb, sch); return len; } PSCHED_GET_TIME(q->qidlestart); if (!red_is_idling(&q->parms)) red_start_of_idle_period(&q->parms); return 0; } Loading @@ -350,43 +148,38 @@ static void red_reset(struct Qdisc* sch) { struct red_sched_data *q = qdisc_priv(sch); __skb_queue_purge(&sch->q); sch->qstats.backlog = 0; PSCHED_SET_PASTPERFECT(q->qidlestart); q->qave = 0; q->qcount = -1; qdisc_reset_queue(sch); red_restart(&q->parms); } static int red_change(struct Qdisc *sch, struct rtattr *opt) { struct red_sched_data *q = qdisc_priv(sch); struct rtattr *tb[TCA_RED_STAB]; struct rtattr *tb[TCA_RED_MAX]; struct tc_red_qopt *ctl; if (opt == NULL || rtattr_parse_nested(tb, TCA_RED_STAB, opt) || tb[TCA_RED_PARMS-1] == 0 || tb[TCA_RED_STAB-1] == 0 || if (opt == NULL || rtattr_parse_nested(tb, TCA_RED_MAX, opt)) return -EINVAL; if (tb[TCA_RED_PARMS-1] == NULL || RTA_PAYLOAD(tb[TCA_RED_PARMS-1]) < sizeof(*ctl) || RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < 256) tb[TCA_RED_STAB-1] == NULL || RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < RED_STAB_SIZE) return -EINVAL; ctl = RTA_DATA(tb[TCA_RED_PARMS-1]); sch_tree_lock(sch); q->flags = ctl->flags; q->Wlog = ctl->Wlog; q->Plog = ctl->Plog; q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL; q->Scell_log = ctl->Scell_log; q->Scell_max = (255<<q->Scell_log); q->qth_min = ctl->qth_min<<ctl->Wlog; q->qth_max = ctl->qth_max<<ctl->Wlog; q->limit = ctl->limit; memcpy(q->Stab, RTA_DATA(tb[TCA_RED_STAB-1]), 256); q->qcount = -1; red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog, ctl->Scell_log, RTA_DATA(tb[TCA_RED_STAB-1])); if (skb_queue_empty(&sch->q)) PSCHED_SET_PASTPERFECT(q->qidlestart); red_end_of_idle_period(&q->parms); sch_tree_unlock(sch); return 0; } Loading @@ -399,39 +192,39 @@ static int red_init(struct Qdisc* sch, struct rtattr *opt) static int red_dump(struct Qdisc *sch, struct sk_buff *skb) { struct red_sched_data *q = qdisc_priv(sch); unsigned char *b = skb->tail; struct rtattr *rta; struct tc_red_qopt opt; rta = (struct rtattr*)b; RTA_PUT(skb, TCA_OPTIONS, 0, NULL); opt.limit = q->limit; opt.qth_min = q->qth_min>>q->Wlog; opt.qth_max = q->qth_max>>q->Wlog; opt.Wlog = q->Wlog; opt.Plog = q->Plog; opt.Scell_log = q->Scell_log; opt.flags = q->flags; RTA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt); rta->rta_len = skb->tail - b; struct rtattr *opts = NULL; struct tc_red_qopt opt = { .limit = q->limit, .flags = q->flags, .qth_min = q->parms.qth_min >> q->parms.Wlog, .qth_max = q->parms.qth_max >> q->parms.Wlog, .Wlog = q->parms.Wlog, .Plog = q->parms.Plog, .Scell_log = q->parms.Scell_log, }; return skb->len; opts = RTA_NEST(skb, TCA_OPTIONS); RTA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt); return RTA_NEST_END(skb, opts); rtattr_failure: skb_trim(skb, b - skb->data); return -1; return RTA_NEST_CANCEL(skb, opts); } static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) { struct red_sched_data *q = qdisc_priv(sch); struct tc_red_xstats st = { .early = q->stats.prob_drop + q->stats.forced_drop, .pdrop = q->stats.pdrop, .other = q->stats.other, .marked = q->stats.prob_mark + q->stats.forced_mark, }; return gnet_stats_copy_app(d, &q->st, sizeof(q->st)); return gnet_stats_copy_app(d, &st, sizeof(st)); } static struct Qdisc_ops red_qdisc_ops = { .next = NULL, .cl_ops = NULL, .id = "red", .priv_size = sizeof(struct red_sched_data), .enqueue = red_enqueue, Loading @@ -450,10 +243,13 @@ static int __init red_module_init(void) { return register_qdisc(&red_qdisc_ops); } static void __exit red_module_exit(void) { unregister_qdisc(&red_qdisc_ops); } module_init(red_module_init) module_exit(red_module_exit) MODULE_LICENSE("GPL"); Loading
include/linux/pkt_sched.h +24 −26 Original line number Diff line number Diff line Loading @@ -93,6 +93,7 @@ struct tc_fifo_qopt /* PRIO section */ #define TCQ_PRIO_BANDS 16 #define TCQ_MIN_PRIO_BANDS 2 struct tc_prio_qopt { Loading Loading @@ -169,6 +170,7 @@ struct tc_red_qopt unsigned char Scell_log; /* cell size for idle damping */ unsigned char flags; #define TC_RED_ECN 1 #define TC_RED_HARDDROP 2 }; struct tc_red_xstats Loading @@ -194,15 +196,11 @@ enum #define TCA_GRED_MAX (__TCA_GRED_MAX - 1) #define TCA_SET_OFF TCA_GRED_PARMS struct tc_gred_qopt { __u32 limit; /* HARD maximal queue length (bytes) */ __u32 qth_min; /* Min average length threshold (bytes) */ __u32 qth_max; /* Max average length threshold (bytes) */ __u32 limit; /* HARD maximal queue length (bytes) */ __u32 qth_min; /* Min average length threshold (bytes) */ __u32 qth_max; /* Max average length threshold (bytes) */ __u32 DP; /* upto 2^32 DPs */ __u32 backlog; __u32 qave; Loading @@ -210,22 +208,22 @@ struct tc_gred_qopt __u32 early; __u32 other; __u32 pdrop; unsigned char Wlog; /* log(W) */ unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ unsigned char Scell_log; /* cell size for idle damping */ __u8 Wlog; /* log(W) */ __u8 Plog; /* log(P_max/(qth_max-qth_min)) */ __u8 Scell_log; /* cell size for idle damping */ __u8 prio; /* prio of this VQ */ __u32 packets; __u32 bytesin; }; /* gred setup */ struct tc_gred_sopt { __u32 DPs; __u32 def_DP; __u8 grio; __u8 pad1; __u16 pad2; __u8 flags; __u16 pad1; }; /* HTB section */ Loading
include/net/inet_ecn.h +24 −4 Original line number Diff line number Diff line Loading @@ -2,6 +2,7 @@ #define _INET_ECN_H_ #include <linux/ip.h> #include <linux/skbuff.h> #include <net/dsfield.h> enum { Loading Loading @@ -48,7 +49,7 @@ static inline __u8 INET_ECN_encapsulate(__u8 outer, __u8 inner) (label) |= __constant_htons(INET_ECN_ECT_0 << 4); \ } while (0) static inline void IP_ECN_set_ce(struct iphdr *iph) static inline int IP_ECN_set_ce(struct iphdr *iph) { u32 check = iph->check; u32 ecn = (iph->tos + 1) & INET_ECN_MASK; Loading @@ -61,7 +62,7 @@ static inline void IP_ECN_set_ce(struct iphdr *iph) * INET_ECN_CE => 00 */ if (!(ecn & 2)) return; return !ecn; /* * The following gives us: Loading @@ -72,6 +73,7 @@ static inline void IP_ECN_set_ce(struct iphdr *iph) iph->check = check + (check>=0xFFFF); iph->tos |= INET_ECN_CE; return 1; } static inline void IP_ECN_clear(struct iphdr *iph) Loading @@ -87,11 +89,12 @@ static inline void ipv4_copy_dscp(struct iphdr *outer, struct iphdr *inner) struct ipv6hdr; static inline void IP6_ECN_set_ce(struct ipv6hdr *iph) static inline int IP6_ECN_set_ce(struct ipv6hdr *iph) { if (INET_ECN_is_not_ect(ipv6_get_dsfield(iph))) return; return 0; *(u32*)iph |= htonl(INET_ECN_CE << 20); return 1; } static inline void IP6_ECN_clear(struct ipv6hdr *iph) Loading @@ -105,4 +108,21 @@ static inline void ipv6_copy_dscp(struct ipv6hdr *outer, struct ipv6hdr *inner) ipv6_change_dsfield(inner, INET_ECN_MASK, dscp); } static inline int INET_ECN_set_ce(struct sk_buff *skb) { switch (skb->protocol) { case __constant_htons(ETH_P_IP): if (skb->nh.raw + sizeof(struct iphdr) <= skb->tail) return IP_ECN_set_ce(skb->nh.iph); break; case __constant_htons(ETH_P_IPV6): if (skb->nh.raw + sizeof(struct ipv6hdr) <= skb->tail) return IP6_ECN_set_ce(skb->nh.ipv6h); break; } return 0; } #endif
include/net/red.h 0 → 100644 +325 −0 Original line number Diff line number Diff line #ifndef __NET_SCHED_RED_H #define __NET_SCHED_RED_H #include <linux/config.h> #include <linux/types.h> #include <net/pkt_sched.h> #include <net/inet_ecn.h> #include <net/dsfield.h> /* Random Early Detection (RED) algorithm. ======================================= Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking. This file codes a "divisionless" version of RED algorithm as written down in Fig.17 of the paper. Short description. ------------------ When a new packet arrives we calculate the average queue length: avg = (1-W)*avg + W*current_queue_len, W is the filter time constant (chosen as 2^(-Wlog)), it controls the inertia of the algorithm. To allow larger bursts, W should be decreased. if (avg > th_max) -> packet marked (dropped). if (avg < th_min) -> packet passes. if (th_min < avg < th_max) we calculate probability: Pb = max_P * (avg - th_min)/(th_max-th_min) and mark (drop) packet with this probability. Pb changes from 0 (at avg==th_min) to max_P (avg==th_max). max_P should be small (not 1), usually 0.01..0.02 is good value. max_P is chosen as a number, so that max_P/(th_max-th_min) is a negative power of two in order arithmetics to contain only shifts. Parameters, settable by user: ----------------------------- qth_min - bytes (should be < qth_max/2) qth_max - bytes (should be at least 2*qth_min and less limit) Wlog - bits (<32) log(1/W). Plog - bits (<32) Plog is related to max_P by formula: max_P = (qth_max-qth_min)/2^Plog; F.e. if qth_max=128K and qth_min=32K, then Plog=22 corresponds to max_P=0.02 Scell_log Stab Lookup table for log((1-W)^(t/t_ave). NOTES: Upper bound on W. ----------------- If you want to allow bursts of L packets of size S, you should choose W: L + 1 - th_min/S < (1-(1-W)^L)/W th_min/S = 32 th_min/S = 4 log(W) L -1 33 -2 35 -3 39 -4 46 -5 57 -6 75 -7 101 -8 135 -9 190 etc. */ #define RED_STAB_SIZE 256 #define RED_STAB_MASK (RED_STAB_SIZE - 1) struct red_stats { u32 prob_drop; /* Early probability drops */ u32 prob_mark; /* Early probability marks */ u32 forced_drop; /* Forced drops, qavg > max_thresh */ u32 forced_mark; /* Forced marks, qavg > max_thresh */ u32 pdrop; /* Drops due to queue limits */ u32 other; /* Drops due to drop() calls */ u32 backlog; }; struct red_parms { /* Parameters */ u32 qth_min; /* Min avg length threshold: A scaled */ u32 qth_max; /* Max avg length threshold: A scaled */ u32 Scell_max; u32 Rmask; /* Cached random mask, see red_rmask */ u8 Scell_log; u8 Wlog; /* log(W) */ u8 Plog; /* random number bits */ u8 Stab[RED_STAB_SIZE]; /* Variables */ int qcount; /* Number of packets since last random number generation */ u32 qR; /* Cached random number */ unsigned long qavg; /* Average queue length: A scaled */ psched_time_t qidlestart; /* Start of current idle period */ }; static inline u32 red_rmask(u8 Plog) { return Plog < 32 ? ((1 << Plog) - 1) : ~0UL; } static inline void red_set_parms(struct red_parms *p, u32 qth_min, u32 qth_max, u8 Wlog, u8 Plog, u8 Scell_log, u8 *stab) { /* Reset average queue length, the value is strictly bound * to the parameters below, reseting hurts a bit but leaving * it might result in an unreasonable qavg for a while. --TGR */ p->qavg = 0; p->qcount = -1; p->qth_min = qth_min << Wlog; p->qth_max = qth_max << Wlog; p->Wlog = Wlog; p->Plog = Plog; p->Rmask = red_rmask(Plog); p->Scell_log = Scell_log; p->Scell_max = (255 << Scell_log); memcpy(p->Stab, stab, sizeof(p->Stab)); } static inline int red_is_idling(struct red_parms *p) { return !PSCHED_IS_PASTPERFECT(p->qidlestart); } static inline void red_start_of_idle_period(struct red_parms *p) { PSCHED_GET_TIME(p->qidlestart); } static inline void red_end_of_idle_period(struct red_parms *p) { PSCHED_SET_PASTPERFECT(p->qidlestart); } static inline void red_restart(struct red_parms *p) { red_end_of_idle_period(p); p->qavg = 0; p->qcount = -1; } static inline unsigned long red_calc_qavg_from_idle_time(struct red_parms *p) { psched_time_t now; long us_idle; int shift; PSCHED_GET_TIME(now); us_idle = PSCHED_TDIFF_SAFE(now, p->qidlestart, p->Scell_max); /* * The problem: ideally, average length queue recalcultion should * be done over constant clock intervals. This is too expensive, so * that the calculation is driven by outgoing packets. * When the queue is idle we have to model this clock by hand. * * SF+VJ proposed to "generate": * * m = idletime / (average_pkt_size / bandwidth) * * dummy packets as a burst after idle time, i.e. * * p->qavg *= (1-W)^m * * This is an apparently overcomplicated solution (f.e. we have to * precompute a table to make this calculation in reasonable time) * I believe that a simpler model may be used here, * but it is field for experiments. */ shift = p->Stab[(us_idle >> p->Scell_log) & RED_STAB_MASK]; if (shift) return p->qavg >> shift; else { /* Approximate initial part of exponent with linear function: * * (1-W)^m ~= 1-mW + ... * * Seems, it is the best solution to * problem of too coarse exponent tabulation. */ us_idle = (p->qavg * us_idle) >> p->Scell_log; if (us_idle < (p->qavg >> 1)) return p->qavg - us_idle; else return p->qavg >> 1; } } static inline unsigned long red_calc_qavg_no_idle_time(struct red_parms *p, unsigned int backlog) { /* * NOTE: p->qavg is fixed point number with point at Wlog. * The formula below is equvalent to floating point * version: * * qavg = qavg*(1-W) + backlog*W; * * --ANK (980924) */ return p->qavg + (backlog - (p->qavg >> p->Wlog)); } static inline unsigned long red_calc_qavg(struct red_parms *p, unsigned int backlog) { if (!red_is_idling(p)) return red_calc_qavg_no_idle_time(p, backlog); else return red_calc_qavg_from_idle_time(p); } static inline u32 red_random(struct red_parms *p) { return net_random() & p->Rmask; } static inline int red_mark_probability(struct red_parms *p, unsigned long qavg) { /* The formula used below causes questions. OK. qR is random number in the interval 0..Rmask i.e. 0..(2^Plog). If we used floating point arithmetics, it would be: (2^Plog)*rnd_num, where rnd_num is less 1. Taking into account, that qavg have fixed point at Wlog, and Plog is related to max_P by max_P = (qth_max-qth_min)/2^Plog; two lines below have the following floating point equivalent: max_P*(qavg - qth_min)/(qth_max-qth_min) < rnd/qcount Any questions? --ANK (980924) */ return !(((qavg - p->qth_min) >> p->Wlog) * p->qcount < p->qR); } enum { RED_BELOW_MIN_THRESH, RED_BETWEEN_TRESH, RED_ABOVE_MAX_TRESH, }; static inline int red_cmp_thresh(struct red_parms *p, unsigned long qavg) { if (qavg < p->qth_min) return RED_BELOW_MIN_THRESH; else if (qavg >= p->qth_max) return RED_ABOVE_MAX_TRESH; else return RED_BETWEEN_TRESH; } enum { RED_DONT_MARK, RED_PROB_MARK, RED_HARD_MARK, }; static inline int red_action(struct red_parms *p, unsigned long qavg) { switch (red_cmp_thresh(p, qavg)) { case RED_BELOW_MIN_THRESH: p->qcount = -1; return RED_DONT_MARK; case RED_BETWEEN_TRESH: if (++p->qcount) { if (red_mark_probability(p, qavg)) { p->qcount = 0; p->qR = red_random(p); return RED_PROB_MARK; } } else p->qR = red_random(p); return RED_DONT_MARK; case RED_ABOVE_MAX_TRESH: p->qcount = -1; return RED_HARD_MARK; } BUG(); return RED_DONT_MARK; } #endif
net/sched/sch_gred.c +411 −430 File changed.Preview size limit exceeded, changes collapsed. Show changes
net/sched/sch_red.c +107 −311 Original line number Diff line number Diff line Loading @@ -9,76 +9,23 @@ * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> * * Changes: * J Hadi Salim <hadi@nortel.com> 980914: computation fixes * J Hadi Salim 980914: computation fixes * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly. * J Hadi Salim <hadi@nortelnetworks.com> 980816: ECN support * J Hadi Salim 980816: ECN support */ #include <linux/config.h> #include <linux/module.h> #include <asm/uaccess.h> #include <asm/system.h> #include <linux/bitops.h> #include <linux/types.h> #include <linux/kernel.h> #include <linux/sched.h> #include <linux/string.h> #include <linux/mm.h> #include <linux/socket.h> #include <linux/sockios.h> #include <linux/in.h> #include <linux/errno.h> #include <linux/interrupt.h> #include <linux/if_ether.h> #include <linux/inet.h> #include <linux/netdevice.h> #include <linux/etherdevice.h> #include <linux/notifier.h> #include <net/ip.h> #include <net/route.h> #include <linux/skbuff.h> #include <net/sock.h> #include <net/pkt_sched.h> #include <net/inet_ecn.h> #include <net/dsfield.h> #include <net/red.h> /* Random Early Detection (RED) algorithm. ======================================= Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking. This file codes a "divisionless" version of RED algorithm as written down in Fig.17 of the paper. Short description. ------------------ When a new packet arrives we calculate the average queue length: avg = (1-W)*avg + W*current_queue_len, W is the filter time constant (chosen as 2^(-Wlog)), it controls the inertia of the algorithm. To allow larger bursts, W should be decreased. if (avg > th_max) -> packet marked (dropped). if (avg < th_min) -> packet passes. if (th_min < avg < th_max) we calculate probability: Pb = max_P * (avg - th_min)/(th_max-th_min) and mark (drop) packet with this probability. Pb changes from 0 (at avg==th_min) to max_P (avg==th_max). max_P should be small (not 1), usually 0.01..0.02 is good value. max_P is chosen as a number, so that max_P/(th_max-th_min) is a negative power of two in order arithmetics to contain only shifts. Parameters, settable by user: /* Parameters, settable by user: ----------------------------- limit - bytes (must be > qth_max + burst) Loading @@ -89,260 +36,111 @@ Short description. arbitrarily high (well, less than ram size) Really, this limit will never be reached if RED works correctly. qth_min - bytes (should be < qth_max/2) qth_max - bytes (should be at least 2*qth_min and less limit) Wlog - bits (<32) log(1/W). Plog - bits (<32) Plog is related to max_P by formula: max_P = (qth_max-qth_min)/2^Plog; F.e. if qth_max=128K and qth_min=32K, then Plog=22 corresponds to max_P=0.02 Scell_log Stab Lookup table for log((1-W)^(t/t_ave). NOTES: Upper bound on W. ----------------- If you want to allow bursts of L packets of size S, you should choose W: L + 1 - th_min/S < (1-(1-W)^L)/W th_min/S = 32 th_min/S = 4 log(W) L -1 33 -2 35 -3 39 -4 46 -5 57 -6 75 -7 101 -8 135 -9 190 etc. */ struct red_sched_data { /* Parameters */ u32 limit; /* HARD maximal queue length */ u32 qth_min; /* Min average length threshold: A scaled */ u32 qth_max; /* Max average length threshold: A scaled */ u32 Rmask; u32 Scell_max; unsigned char flags; char Wlog; /* log(W) */ char Plog; /* random number bits */ char Scell_log; u8 Stab[256]; /* Variables */ unsigned long qave; /* Average queue length: A scaled */ int qcount; /* Packets since last random number generation */ u32 qR; /* Cached random number */ psched_time_t qidlestart; /* Start of idle period */ struct tc_red_xstats st; struct red_parms parms; struct red_stats stats; }; static int red_ecn_mark(struct sk_buff *skb) static inline int red_use_ecn(struct red_sched_data *q) { if (skb->nh.raw + 20 > skb->tail) return 0; switch (skb->protocol) { case __constant_htons(ETH_P_IP): if (INET_ECN_is_not_ect(skb->nh.iph->tos)) return 0; IP_ECN_set_ce(skb->nh.iph); return 1; case __constant_htons(ETH_P_IPV6): if (INET_ECN_is_not_ect(ipv6_get_dsfield(skb->nh.ipv6h))) return 0; IP6_ECN_set_ce(skb->nh.ipv6h); return 1; default: return 0; return q->flags & TC_RED_ECN; } static inline int red_use_harddrop(struct red_sched_data *q) { return q->flags & TC_RED_HARDDROP; } static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch) static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch) { struct red_sched_data *q = qdisc_priv(sch); psched_time_t now; if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) { long us_idle; int shift; PSCHED_GET_TIME(now); us_idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max); PSCHED_SET_PASTPERFECT(q->qidlestart); /* The problem: ideally, average length queue recalcultion should be done over constant clock intervals. This is too expensive, so that the calculation is driven by outgoing packets. When the queue is idle we have to model this clock by hand. SF+VJ proposed to "generate" m = idletime/(average_pkt_size/bandwidth) dummy packets as a burst after idle time, i.e. q->qave *= (1-W)^m q->parms.qavg = red_calc_qavg(&q->parms, sch->qstats.backlog); This is an apparently overcomplicated solution (f.e. we have to precompute a table to make this calculation in reasonable time) I believe that a simpler model may be used here, but it is field for experiments. */ shift = q->Stab[us_idle>>q->Scell_log]; if (red_is_idling(&q->parms)) red_end_of_idle_period(&q->parms); if (shift) { q->qave >>= shift; } else { /* Approximate initial part of exponent with linear function: (1-W)^m ~= 1-mW + ... Seems, it is the best solution to problem of too coarce exponent tabulation. */ us_idle = (q->qave * us_idle)>>q->Scell_log; if (us_idle < q->qave/2) q->qave -= us_idle; else q->qave >>= 1; } } else { q->qave += sch->qstats.backlog - (q->qave >> q->Wlog); /* NOTE: q->qave is fixed point number with point at Wlog. The formulae above is equvalent to floating point version: qave = qave*(1-W) + sch->qstats.backlog*W; --ANK (980924) */ } switch (red_action(&q->parms, q->parms.qavg)) { case RED_DONT_MARK: break; if (q->qave < q->qth_min) { q->qcount = -1; enqueue: if (sch->qstats.backlog + skb->len <= q->limit) { __skb_queue_tail(&sch->q, skb); sch->qstats.backlog += skb->len; sch->bstats.bytes += skb->len; sch->bstats.packets++; return NET_XMIT_SUCCESS; } else { q->st.pdrop++; } kfree_skb(skb); sch->qstats.drops++; return NET_XMIT_DROP; } if (q->qave >= q->qth_max) { q->qcount = -1; case RED_PROB_MARK: sch->qstats.overlimits++; mark: if (!(q->flags&TC_RED_ECN) || !red_ecn_mark(skb)) { q->st.early++; goto drop; } q->st.marked++; goto enqueue; if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) { q->stats.prob_drop++; goto congestion_drop; } if (++q->qcount) { /* The formula used below causes questions. q->stats.prob_mark++; break; OK. qR is random number in the interval 0..Rmask i.e. 0..(2^Plog). If we used floating point arithmetics, it would be: (2^Plog)*rnd_num, where rnd_num is less 1. case RED_HARD_MARK: sch->qstats.overlimits++; if (red_use_harddrop(q) || !red_use_ecn(q) || !INET_ECN_set_ce(skb)) { q->stats.forced_drop++; goto congestion_drop; } Taking into account, that qave have fixed point at Wlog, and Plog is related to max_P by max_P = (qth_max-qth_min)/2^Plog; two lines below have the following floating point equivalent: q->stats.forced_mark++; break; } max_P*(qave - qth_min)/(qth_max-qth_min) < rnd/qcount if (sch->qstats.backlog + skb->len <= q->limit) return qdisc_enqueue_tail(skb, sch); Any questions? --ANK (980924) */ if (((q->qave - q->qth_min)>>q->Wlog)*q->qcount < q->qR) goto enqueue; q->qcount = 0; q->qR = net_random()&q->Rmask; sch->qstats.overlimits++; goto mark; } q->qR = net_random()&q->Rmask; goto enqueue; q->stats.pdrop++; return qdisc_drop(skb, sch); drop: kfree_skb(skb); sch->qstats.drops++; congestion_drop: qdisc_drop(skb, sch); return NET_XMIT_CN; } static int red_requeue(struct sk_buff *skb, struct Qdisc* sch) static int red_requeue(struct sk_buff *skb, struct Qdisc* sch) { struct red_sched_data *q = qdisc_priv(sch); PSCHED_SET_PASTPERFECT(q->qidlestart); if (red_is_idling(&q->parms)) red_end_of_idle_period(&q->parms); __skb_queue_head(&sch->q, skb); sch->qstats.backlog += skb->len; sch->qstats.requeues++; return 0; return qdisc_requeue(skb, sch); } static struct sk_buff * red_dequeue(struct Qdisc* sch) static struct sk_buff * red_dequeue(struct Qdisc* sch) { struct sk_buff *skb; struct red_sched_data *q = qdisc_priv(sch); skb = __skb_dequeue(&sch->q); if (skb) { sch->qstats.backlog -= skb->len; skb = qdisc_dequeue_head(sch); if (skb == NULL && !red_is_idling(&q->parms)) red_start_of_idle_period(&q->parms); return skb; } PSCHED_GET_TIME(q->qidlestart); return NULL; } static unsigned int red_drop(struct Qdisc* sch) { struct sk_buff *skb; struct red_sched_data *q = qdisc_priv(sch); skb = __skb_dequeue_tail(&sch->q); skb = qdisc_dequeue_tail(sch); if (skb) { unsigned int len = skb->len; sch->qstats.backlog -= len; sch->qstats.drops++; q->st.other++; kfree_skb(skb); q->stats.other++; qdisc_drop(skb, sch); return len; } PSCHED_GET_TIME(q->qidlestart); if (!red_is_idling(&q->parms)) red_start_of_idle_period(&q->parms); return 0; } Loading @@ -350,43 +148,38 @@ static void red_reset(struct Qdisc* sch) { struct red_sched_data *q = qdisc_priv(sch); __skb_queue_purge(&sch->q); sch->qstats.backlog = 0; PSCHED_SET_PASTPERFECT(q->qidlestart); q->qave = 0; q->qcount = -1; qdisc_reset_queue(sch); red_restart(&q->parms); } static int red_change(struct Qdisc *sch, struct rtattr *opt) { struct red_sched_data *q = qdisc_priv(sch); struct rtattr *tb[TCA_RED_STAB]; struct rtattr *tb[TCA_RED_MAX]; struct tc_red_qopt *ctl; if (opt == NULL || rtattr_parse_nested(tb, TCA_RED_STAB, opt) || tb[TCA_RED_PARMS-1] == 0 || tb[TCA_RED_STAB-1] == 0 || if (opt == NULL || rtattr_parse_nested(tb, TCA_RED_MAX, opt)) return -EINVAL; if (tb[TCA_RED_PARMS-1] == NULL || RTA_PAYLOAD(tb[TCA_RED_PARMS-1]) < sizeof(*ctl) || RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < 256) tb[TCA_RED_STAB-1] == NULL || RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < RED_STAB_SIZE) return -EINVAL; ctl = RTA_DATA(tb[TCA_RED_PARMS-1]); sch_tree_lock(sch); q->flags = ctl->flags; q->Wlog = ctl->Wlog; q->Plog = ctl->Plog; q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL; q->Scell_log = ctl->Scell_log; q->Scell_max = (255<<q->Scell_log); q->qth_min = ctl->qth_min<<ctl->Wlog; q->qth_max = ctl->qth_max<<ctl->Wlog; q->limit = ctl->limit; memcpy(q->Stab, RTA_DATA(tb[TCA_RED_STAB-1]), 256); q->qcount = -1; red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog, ctl->Scell_log, RTA_DATA(tb[TCA_RED_STAB-1])); if (skb_queue_empty(&sch->q)) PSCHED_SET_PASTPERFECT(q->qidlestart); red_end_of_idle_period(&q->parms); sch_tree_unlock(sch); return 0; } Loading @@ -399,39 +192,39 @@ static int red_init(struct Qdisc* sch, struct rtattr *opt) static int red_dump(struct Qdisc *sch, struct sk_buff *skb) { struct red_sched_data *q = qdisc_priv(sch); unsigned char *b = skb->tail; struct rtattr *rta; struct tc_red_qopt opt; rta = (struct rtattr*)b; RTA_PUT(skb, TCA_OPTIONS, 0, NULL); opt.limit = q->limit; opt.qth_min = q->qth_min>>q->Wlog; opt.qth_max = q->qth_max>>q->Wlog; opt.Wlog = q->Wlog; opt.Plog = q->Plog; opt.Scell_log = q->Scell_log; opt.flags = q->flags; RTA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt); rta->rta_len = skb->tail - b; struct rtattr *opts = NULL; struct tc_red_qopt opt = { .limit = q->limit, .flags = q->flags, .qth_min = q->parms.qth_min >> q->parms.Wlog, .qth_max = q->parms.qth_max >> q->parms.Wlog, .Wlog = q->parms.Wlog, .Plog = q->parms.Plog, .Scell_log = q->parms.Scell_log, }; return skb->len; opts = RTA_NEST(skb, TCA_OPTIONS); RTA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt); return RTA_NEST_END(skb, opts); rtattr_failure: skb_trim(skb, b - skb->data); return -1; return RTA_NEST_CANCEL(skb, opts); } static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) { struct red_sched_data *q = qdisc_priv(sch); struct tc_red_xstats st = { .early = q->stats.prob_drop + q->stats.forced_drop, .pdrop = q->stats.pdrop, .other = q->stats.other, .marked = q->stats.prob_mark + q->stats.forced_mark, }; return gnet_stats_copy_app(d, &q->st, sizeof(q->st)); return gnet_stats_copy_app(d, &st, sizeof(st)); } static struct Qdisc_ops red_qdisc_ops = { .next = NULL, .cl_ops = NULL, .id = "red", .priv_size = sizeof(struct red_sched_data), .enqueue = red_enqueue, Loading @@ -450,10 +243,13 @@ static int __init red_module_init(void) { return register_qdisc(&red_qdisc_ops); } static void __exit red_module_exit(void) { unregister_qdisc(&red_qdisc_ops); } module_init(red_module_init) module_exit(red_module_exit) MODULE_LICENSE("GPL");
