From a783474591f2eed0348e08b15934fa9a25e23b3e Mon Sep 17 00:00:00 2001 From: Thomas Graf Date: Sat, 5 Nov 2005 21:14:03 +0100 Subject: [PKT_SCHED]: Generic RED layer Extracts the RED algorithm from sch_red.c and puts it into include/net/red.h for use by other RED based modules. The statistics are extended to be more fine grained in order to differ between probability/forced marks/drops. We now reset the average queue length when setting new parameters, leaving it might result in an unreasonable qavg for a while depending on the value of W. Signed-off-by: Thomas Graf Signed-off-by: Arnaldo Carvalho de Melo --- include/net/red.h | 325 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 325 insertions(+) create mode 100644 include/net/red.h (limited to 'include') diff --git a/include/net/red.h b/include/net/red.h new file mode 100644 index 000000000000..2ed4358e3295 --- /dev/null +++ b/include/net/red.h @@ -0,0 +1,325 @@ +#ifndef __NET_SCHED_RED_H +#define __NET_SCHED_RED_H + +#include +#include +#include +#include +#include + +/* 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 -- cgit v1.2.3 From 2566a509cacc8b8eaea2e5b54068816c9cfb41c2 Mon Sep 17 00:00:00 2001 From: Thomas Graf Date: Sat, 5 Nov 2005 21:14:04 +0100 Subject: [NET]: Introduce INET_ECN_set_ce() function Changes IP_ECN_set_ce() and IP6_ECN_set_ce() to return 0 if the CE bits could not bet set because none of the ECT bits are set or 1 if the CE bits are already set or have been successfully set. Introduces INET_ECN_set_ce(skb) to enable CE bits for all supported protocols. Signed-off-by: Thomas Graf Signed-off-by: Arnaldo Carvalho de Melo --- include/net/inet_ecn.h | 28 ++++++++++++++++++++++++---- 1 file changed, 24 insertions(+), 4 deletions(-) (limited to 'include') diff --git a/include/net/inet_ecn.h b/include/net/inet_ecn.h index f87845e2e965..b0c47e2eccf1 100644 --- a/include/net/inet_ecn.h +++ b/include/net/inet_ecn.h @@ -2,6 +2,7 @@ #define _INET_ECN_H_ #include +#include #include enum { @@ -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; @@ -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: @@ -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) @@ -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) @@ -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 -- cgit v1.2.3 From 1e4dfaf9b99a8b652e8421936fd5fe2459da8265 Mon Sep 17 00:00:00 2001 From: Thomas Graf Date: Sat, 5 Nov 2005 21:14:25 +0100 Subject: [PKT_SCHED]: GRED: Cleanup and remove unnecessary code Removes unnecessary includes, initializers, and simplifies the code a bit. Signed-off-by: Thomas Graf Signed-off-by: Arnaldo Carvalho de Melo --- include/linux/pkt_sched.h | 48 ++++++++++------------ net/sched/sch_gred.c | 100 ++++++++++++++-------------------------------- 2 files changed, 53 insertions(+), 95 deletions(-) (limited to 'include') diff --git a/include/linux/pkt_sched.h b/include/linux/pkt_sched.h index 60ffcb9c5791..d053add3dca7 100644 --- a/include/linux/pkt_sched.h +++ b/include/linux/pkt_sched.h @@ -194,38 +194,34 @@ 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 DP; /* upto 2^32 DPs */ - __u32 backlog; - __u32 qave; - __u32 forced; - __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 prio; /* prio of this VQ */ - __u32 packets; - __u32 bytesin; + __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; + __u32 forced; + __u32 early; + __u32 other; + __u32 pdrop; + __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; + __u32 DPs; + __u32 def_DP; + __u8 grio; + __u8 pad1; + __u16 pad2; }; /* HTB section */ diff --git a/net/sched/sch_gred.c b/net/sched/sch_gred.c index 897e6df81b1f..1fb34be32f7c 100644 --- a/net/sched/sch_gred.c +++ b/net/sched/sch_gred.c @@ -15,50 +15,18 @@ * from Ren Liu * - More error checks * - * - * - * For all the glorious comments look at Alexey's sch_red.c + * For all the glorious comments look at include/net/red.h */ #include #include -#include -#include -#include #include #include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include #include -#include -#include -#include -#include #include -#include #include #include -#if 1 /* control */ -#define DPRINTK(format,args...) printk(KERN_DEBUG format,##args) -#else -#define DPRINTK(format,args...) -#endif - -#if 0 /* data */ -#define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args) -#else -#define D2PRINTK(format,args...) -#endif - #define GRED_DEF_PRIO (MAX_DPs / 2) #define GRED_VQ_MASK (MAX_DPs - 1) @@ -72,7 +40,7 @@ struct gred_sched_data u32 bytesin; /* bytes seen on virtualQ so far*/ u32 packetsin; /* packets seen on virtualQ so far*/ u32 backlog; /* bytes on the virtualQ */ - u8 prio; /* the prio of this vq */ + u8 prio; /* the prio of this vq */ struct red_parms parms; struct red_stats stats; @@ -87,8 +55,8 @@ struct gred_sched { struct gred_sched_data *tab[MAX_DPs]; unsigned long flags; - u32 DPs; - u32 def; + u32 DPs; + u32 def; struct red_parms wred_set; }; @@ -172,13 +140,11 @@ static inline void gred_store_wred_set(struct gred_sched *table, table->wred_set.qavg = q->parms.qavg; } -static int -gred_enqueue(struct sk_buff *skb, struct Qdisc* sch) +static int gred_enqueue(struct sk_buff *skb, struct Qdisc* sch) { struct gred_sched_data *q=NULL; struct gred_sched *t= qdisc_priv(sch); unsigned long qavg = 0; - int i=0; u16 dp = tc_index_to_dp(skb); if (dp >= t->DPs || (q = t->tab[dp]) == NULL) { @@ -200,26 +166,23 @@ gred_enqueue(struct sk_buff *skb, struct Qdisc* sch) skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp; } - /* sum up all the qaves of prios <= to ours to get the new qave*/ + /* sum up all the qaves of prios <= to ours to get the new qave */ if (!gred_wred_mode(t) && gred_rio_mode(t)) { - for (i=0;iDPs;i++) { - if ((!t->tab[i]) || (i==q->DP)) - continue; - - if (t->tab[i]->prio < q->prio && + int i; + + for (i = 0; i < t->DPs; i++) { + if (t->tab[i] && t->tab[i]->prio < q->prio && !red_is_idling(&t->tab[i]->parms)) qavg +=t->tab[i]->parms.qavg; } - + } q->packetsin++; - q->bytesin+=skb->len; + q->bytesin += skb->len; - if (gred_wred_mode(t)) { - qavg = 0; + if (gred_wred_mode(t)) gred_load_wred_set(t, q); - } q->parms.qavg = red_calc_qavg(&q->parms, gred_backlog(t, q, sch)); @@ -258,8 +221,7 @@ congestion_drop: return NET_XMIT_CN; } -static int -gred_requeue(struct sk_buff *skb, struct Qdisc* sch) +static int gred_requeue(struct sk_buff *skb, struct Qdisc* sch) { struct gred_sched *t = qdisc_priv(sch); struct gred_sched_data *q; @@ -279,16 +241,15 @@ gred_requeue(struct sk_buff *skb, struct Qdisc* sch) return qdisc_requeue(skb, sch); } -static struct sk_buff * -gred_dequeue(struct Qdisc* sch) +static struct sk_buff *gred_dequeue(struct Qdisc* sch) { struct sk_buff *skb; - struct gred_sched_data *q; - struct gred_sched *t= qdisc_priv(sch); + struct gred_sched *t = qdisc_priv(sch); skb = qdisc_dequeue_head(sch); if (skb) { + struct gred_sched_data *q; u16 dp = tc_index_to_dp(skb); if (dp >= t->DPs || (q = t->tab[dp]) == NULL) { @@ -315,13 +276,12 @@ gred_dequeue(struct Qdisc* sch) static unsigned int gred_drop(struct Qdisc* sch) { struct sk_buff *skb; - - struct gred_sched_data *q; - struct gred_sched *t= qdisc_priv(sch); + struct gred_sched *t = qdisc_priv(sch); skb = qdisc_dequeue_tail(sch); if (skb) { unsigned int len = skb->len; + struct gred_sched_data *q; u16 dp = tc_index_to_dp(skb); if (dp >= t->DPs || (q = t->tab[dp]) == NULL) { @@ -351,15 +311,16 @@ static unsigned int gred_drop(struct Qdisc* sch) static void gred_reset(struct Qdisc* sch) { int i; - struct gred_sched_data *q; - struct gred_sched *t= qdisc_priv(sch); + struct gred_sched *t = qdisc_priv(sch); qdisc_reset_queue(sch); - for (i=0;iDPs;i++) { - q= t->tab[i]; - if (!q) - continue; + for (i = 0; i < t->DPs; i++) { + struct gred_sched_data *q = t->tab[i]; + + if (!q) + continue; + red_restart(&q->parms); q->backlog = 0; } @@ -590,15 +551,13 @@ static void gred_destroy(struct Qdisc *sch) struct gred_sched *table = qdisc_priv(sch); int i; - for (i = 0;i < table->DPs; i++) { + for (i = 0; i < table->DPs; i++) { if (table->tab[i]) gred_destroy_vq(table->tab[i]); } } static struct Qdisc_ops gred_qdisc_ops = { - .next = NULL, - .cl_ops = NULL, .id = "gred", .priv_size = sizeof(struct gred_sched), .enqueue = gred_enqueue, @@ -617,10 +576,13 @@ static int __init gred_module_init(void) { return register_qdisc(&gred_qdisc_ops); } -static void __exit gred_module_exit(void) + +static void __exit gred_module_exit(void) { unregister_qdisc(&gred_qdisc_ops); } + module_init(gred_module_init) module_exit(gred_module_exit) + MODULE_LICENSE("GPL"); -- cgit v1.2.3 From b38c7eef7e536d12051cc3d5864032f2f907cdfe Mon Sep 17 00:00:00 2001 From: Thomas Graf Date: Sat, 5 Nov 2005 21:14:27 +0100 Subject: [PKT_SCHED]: GRED: Support ECN marking Adds a new u8 flags in a unused padding area of the netlink message. Adds ECN marking support to be used instead of dropping packets immediately. Signed-off-by: Thomas Graf Signed-off-by: Arnaldo Carvalho de Melo --- include/linux/pkt_sched.h | 4 ++-- net/sched/sch_gred.c | 25 +++++++++++++++++++++---- 2 files changed, 23 insertions(+), 6 deletions(-) (limited to 'include') diff --git a/include/linux/pkt_sched.h b/include/linux/pkt_sched.h index d053add3dca7..0ebe320223e2 100644 --- a/include/linux/pkt_sched.h +++ b/include/linux/pkt_sched.h @@ -220,8 +220,8 @@ struct tc_gred_sopt __u32 DPs; __u32 def_DP; __u8 grio; - __u8 pad1; - __u16 pad2; + __u8 flags; + __u16 pad1; }; /* HTB section */ diff --git a/net/sched/sch_gred.c b/net/sched/sch_gred.c index 69f0fd45d4c3..079b0a4ea1c2 100644 --- a/net/sched/sch_gred.c +++ b/net/sched/sch_gred.c @@ -55,6 +55,7 @@ struct gred_sched { struct gred_sched_data *tab[MAX_DPs]; unsigned long flags; + u32 red_flags; u32 DPs; u32 def; struct red_parms wred_set; @@ -140,6 +141,11 @@ static inline void gred_store_wred_set(struct gred_sched *table, table->wred_set.qavg = q->parms.qavg; } +static inline int gred_use_ecn(struct gred_sched *t) +{ + return t->red_flags & TC_RED_ECN; +} + static int gred_enqueue(struct sk_buff *skb, struct Qdisc* sch) { struct gred_sched_data *q=NULL; @@ -198,13 +204,22 @@ static int gred_enqueue(struct sk_buff *skb, struct Qdisc* sch) case RED_PROB_MARK: sch->qstats.overlimits++; - q->stats.prob_drop++; - goto congestion_drop; + if (!gred_use_ecn(t) || !INET_ECN_set_ce(skb)) { + q->stats.prob_drop++; + goto congestion_drop; + } + + q->stats.prob_mark++; + break; case RED_HARD_MARK: sch->qstats.overlimits++; - q->stats.forced_drop++; - goto congestion_drop; + if (!gred_use_ecn(t) || !INET_ECN_set_ce(skb)) { + q->stats.forced_drop++; + goto congestion_drop; + } + q->stats.forced_mark++; + break; } if (q->backlog + skb->len <= q->limit) { @@ -348,6 +363,7 @@ static inline int gred_change_table_def(struct Qdisc *sch, struct rtattr *dps) sch_tree_lock(sch); table->DPs = sopt->DPs; table->def = sopt->def_DP; + table->red_flags = sopt->flags; /* * Every entry point to GRED is synchronized with the above code @@ -489,6 +505,7 @@ static int gred_dump(struct Qdisc *sch, struct sk_buff *skb) .DPs = table->DPs, .def_DP = table->def, .grio = gred_rio_mode(table), + .flags = table->red_flags, }; opts = RTA_NEST(skb, TCA_OPTIONS); -- cgit v1.2.3 From bdc450a0bb1d48144ced1f899cc8366ec8e85024 Mon Sep 17 00:00:00 2001 From: Thomas Graf Date: Sat, 5 Nov 2005 21:14:28 +0100 Subject: [PKT_SCHED]: (G)RED: Introduce hard dropping Introduces a new flag TC_RED_HARDDROP which specifies that if ECN marking is enabled packets should still be dropped once the average queue length exceeds the maximum threshold. This _may_ help to avoid global synchronisation during small bursts of peers advertising but not caring about ECN. Use this option very carefully, it does more harm than good if (qth_max - qth_min) does not cover at least two average burst cycles. The difference to the current behaviour, in which we'd run into the hard queue limit, is that due to the low pass filter of RED short bursts are less likely to cause a global synchronisation. Signed-off-by: Thomas Graf Signed-off-by: Arnaldo Carvalho de Melo --- include/linux/pkt_sched.h | 2 ++ net/sched/sch_gred.c | 8 +++++++- net/sched/sch_red.c | 8 +++++++- 3 files changed, 16 insertions(+), 2 deletions(-) (limited to 'include') diff --git a/include/linux/pkt_sched.h b/include/linux/pkt_sched.h index 0ebe320223e2..e87b233615b3 100644 --- a/include/linux/pkt_sched.h +++ b/include/linux/pkt_sched.h @@ -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 { @@ -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 diff --git a/net/sched/sch_gred.c b/net/sched/sch_gred.c index 079b0a4ea1c2..29a2dd9f3029 100644 --- a/net/sched/sch_gred.c +++ b/net/sched/sch_gred.c @@ -146,6 +146,11 @@ static inline int gred_use_ecn(struct gred_sched *t) return t->red_flags & TC_RED_ECN; } +static inline int gred_use_harddrop(struct gred_sched *t) +{ + return t->red_flags & TC_RED_HARDDROP; +} + static int gred_enqueue(struct sk_buff *skb, struct Qdisc* sch) { struct gred_sched_data *q=NULL; @@ -214,7 +219,8 @@ static int gred_enqueue(struct sk_buff *skb, struct Qdisc* sch) case RED_HARD_MARK: sch->qstats.overlimits++; - if (!gred_use_ecn(t) || !INET_ECN_set_ce(skb)) { + if (gred_use_harddrop(t) || !gred_use_ecn(t) || + !INET_ECN_set_ce(skb)) { q->stats.forced_drop++; goto congestion_drop; } diff --git a/net/sched/sch_red.c b/net/sched/sch_red.c index 0d89dee751a9..dccfa44c2d71 100644 --- a/net/sched/sch_red.c +++ b/net/sched/sch_red.c @@ -51,6 +51,11 @@ static inline int red_use_ecn(struct red_sched_data *q) 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) { struct red_sched_data *q = qdisc_priv(sch); @@ -76,7 +81,8 @@ static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch) case RED_HARD_MARK: sch->qstats.overlimits++; - if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) { + if (red_use_harddrop(q) || !red_use_ecn(q) || + !INET_ECN_set_ce(skb)) { q->stats.forced_drop++; goto congestion_drop; } -- cgit v1.2.3 From 300ce174ebc2fcf2b5111a50fa42f79d891927dd Mon Sep 17 00:00:00 2001 From: Stephen Hemminger Date: Sun, 30 Oct 2005 13:47:34 -0800 Subject: [NETEM]: Support time based reordering Change netem to support packets getting reordered because of variations in delay. Introduce a special case version of FIFO that queues packets in order based on the netem delay. Since netem is classful, those users that don't want jitter based reordering can just insert a pfifo instead of the default. This required changes to generic skbuff code to allow finer grain manipulation of sk_buff_head. Insertion into the middle and reverse walk. Signed-off-by: Stephen Hemminger Signed-off-by: Arnaldo Carvalho de Melo --- include/linux/skbuff.h | 38 +++++++++++++++++----- net/sched/sch_netem.c | 85 +++++++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 114 insertions(+), 9 deletions(-) (limited to 'include') diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h index 4286d832166f..fdfb8fe8c38c 100644 --- a/include/linux/skbuff.h +++ b/include/linux/skbuff.h @@ -603,29 +603,46 @@ static inline void skb_queue_head_init(struct sk_buff_head *list) */ /** - * __skb_queue_head - queue a buffer at the list head + * __skb_queue_after - queue a buffer at the list head * @list: list to use + * @prev: place after this buffer * @newsk: buffer to queue * - * Queue a buffer at the start of a list. This function takes no locks + * Queue a buffer int the middle of a list. This function takes no locks * and you must therefore hold required locks before calling it. * * A buffer cannot be placed on two lists at the same time. */ -extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); -static inline void __skb_queue_head(struct sk_buff_head *list, - struct sk_buff *newsk) +static inline void __skb_queue_after(struct sk_buff_head *list, + struct sk_buff *prev, + struct sk_buff *newsk) { - struct sk_buff *prev, *next; - + struct sk_buff *next; list->qlen++; - prev = (struct sk_buff *)list; + next = prev->next; newsk->next = next; newsk->prev = prev; next->prev = prev->next = newsk; } +/** + * __skb_queue_head - queue a buffer at the list head + * @list: list to use + * @newsk: buffer to queue + * + * Queue a buffer at the start of a list. This function takes no locks + * and you must therefore hold required locks before calling it. + * + * A buffer cannot be placed on two lists at the same time. + */ +extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); +static inline void __skb_queue_head(struct sk_buff_head *list, + struct sk_buff *newsk) +{ + __skb_queue_after(list, (struct sk_buff *)list, newsk); +} + /** * __skb_queue_tail - queue a buffer at the list tail * @list: list to use @@ -1203,6 +1220,11 @@ static inline void kunmap_skb_frag(void *vaddr) prefetch(skb->next), (skb != (struct sk_buff *)(queue)); \ skb = skb->next) +#define skb_queue_reverse_walk(queue, skb) \ + for (skb = (queue)->prev; \ + prefetch(skb->prev), (skb != (struct sk_buff *)(queue)); \ + skb = skb->prev) + extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, int *err); diff --git a/net/sched/sch_netem.c b/net/sched/sch_netem.c index d871fe7f81a9..7c10ef3457d7 100644 --- a/net/sched/sch_netem.c +++ b/net/sched/sch_netem.c @@ -300,11 +300,16 @@ static void netem_reset(struct Qdisc *sch) del_timer_sync(&q->timer); } +/* Pass size change message down to embedded FIFO */ static int set_fifo_limit(struct Qdisc *q, int limit) { struct rtattr *rta; int ret = -ENOMEM; + /* Hack to avoid sending change message to non-FIFO */ + if (strncmp(q->ops->id + 1, "fifo", 4) != 0) + return 0; + rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)), GFP_KERNEL); if (rta) { rta->rta_type = RTM_NEWQDISC; @@ -436,6 +441,84 @@ static int netem_change(struct Qdisc *sch, struct rtattr *opt) return 0; } +/* + * Special case version of FIFO queue for use by netem. + * It queues in order based on timestamps in skb's + */ +struct fifo_sched_data { + u32 limit; +}; + +static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) +{ + struct fifo_sched_data *q = qdisc_priv(sch); + struct sk_buff_head *list = &sch->q; + const struct netem_skb_cb *ncb + = (const struct netem_skb_cb *)nskb->cb; + struct sk_buff *skb; + + if (likely(skb_queue_len(list) < q->limit)) { + skb_queue_reverse_walk(list, skb) { + const struct netem_skb_cb *cb + = (const struct netem_skb_cb *)skb->cb; + + if (PSCHED_TLESS(cb->time_to_send, ncb->time_to_send)) + break; + } + + __skb_queue_after(list, skb, nskb); + + sch->qstats.backlog += nskb->len; + sch->bstats.bytes += nskb->len; + sch->bstats.packets++; + + return NET_XMIT_SUCCESS; + } + + return qdisc_drop(nskb, sch); +} + +static int tfifo_init(struct Qdisc *sch, struct rtattr *opt) +{ + struct fifo_sched_data *q = qdisc_priv(sch); + + if (opt) { + struct tc_fifo_qopt *ctl = RTA_DATA(opt); + if (RTA_PAYLOAD(opt) < sizeof(*ctl)) + return -EINVAL; + + q->limit = ctl->limit; + } else + q->limit = max_t(u32, sch->dev->tx_queue_len, 1); + + return 0; +} + +static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb) +{ + struct fifo_sched_data *q = qdisc_priv(sch); + struct tc_fifo_qopt opt = { .limit = q->limit }; + + RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); + return skb->len; + +rtattr_failure: + return -1; +} + +static struct Qdisc_ops tfifo_qdisc_ops = { + .id = "tfifo", + .priv_size = sizeof(struct fifo_sched_data), + .enqueue = tfifo_enqueue, + .dequeue = qdisc_dequeue_head, + .requeue = qdisc_requeue, + .drop = qdisc_queue_drop, + .init = tfifo_init, + .reset = qdisc_reset_queue, + .change = tfifo_init, + .dump = tfifo_dump, +}; + static int netem_init(struct Qdisc *sch, struct rtattr *opt) { struct netem_sched_data *q = qdisc_priv(sch); @@ -448,7 +531,7 @@ static int netem_init(struct Qdisc *sch, struct rtattr *opt) q->timer.function = netem_watchdog; q->timer.data = (unsigned long) sch; - q->qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops); + q->qdisc = qdisc_create_dflt(sch->dev, &tfifo_qdisc_ops); if (!q->qdisc) { pr_debug("netem: qdisc create failed\n"); return -ENOMEM; -- cgit v1.2.3 From 6df716340da3a6fdd33d73d7ed4c6f7590ca1c42 Mon Sep 17 00:00:00 2001 From: Stephen Hemminger Date: Thu, 3 Nov 2005 16:33:23 -0800 Subject: [TCP/DCCP]: Randomize port selection This patch randomizes the port selected on bind() for connections to help with possible security attacks. It should also be faster in most cases because there is no need for a global lock. Signed-off-by: Stephen Hemminger Signed-off-by: Arnaldo Carvalho de Melo --- include/net/inet_hashtables.h | 2 -- net/dccp/ipv4.c | 32 +++----------------------------- net/ipv4/inet_connection_sock.c | 14 +++----------- net/ipv4/tcp.c | 1 - net/ipv4/tcp_ipv4.c | 2 -- net/ipv6/tcp_ipv6.c | 15 ++++----------- 6 files changed, 10 insertions(+), 56 deletions(-) (limited to 'include') diff --git a/include/net/inet_hashtables.h b/include/net/inet_hashtables.h index f50f95968340..07840baa9341 100644 --- a/include/net/inet_hashtables.h +++ b/include/net/inet_hashtables.h @@ -125,9 +125,7 @@ struct inet_hashinfo { rwlock_t lhash_lock ____cacheline_aligned; atomic_t lhash_users; wait_queue_head_t lhash_wait; - spinlock_t portalloc_lock; kmem_cache_t *bind_bucket_cachep; - int port_rover; }; static inline unsigned int inet_ehashfn(const __u32 laddr, const __u16 lport, diff --git a/net/dccp/ipv4.c b/net/dccp/ipv4.c index 6298cf58ff9e..4b9bc81ae1a3 100644 --- a/net/dccp/ipv4.c +++ b/net/dccp/ipv4.c @@ -31,8 +31,6 @@ struct inet_hashinfo __cacheline_aligned dccp_hashinfo = { .lhash_lock = RW_LOCK_UNLOCKED, .lhash_users = ATOMIC_INIT(0), .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait), - .portalloc_lock = SPIN_LOCK_UNLOCKED, - .port_rover = 1024 - 1, }; EXPORT_SYMBOL_GPL(dccp_hashinfo); @@ -125,36 +123,15 @@ static int dccp_v4_hash_connect(struct sock *sk) int ret; if (snum == 0) { - int rover; int low = sysctl_local_port_range[0]; int high = sysctl_local_port_range[1]; int remaining = (high - low) + 1; + int rover = net_random() % (high - low) + low; struct hlist_node *node; struct inet_timewait_sock *tw = NULL; local_bh_disable(); - - /* TODO. Actually it is not so bad idea to remove - * dccp_hashinfo.portalloc_lock before next submission to - * Linus. - * As soon as we touch this place at all it is time to think. - * - * Now it protects single _advisory_ variable - * dccp_hashinfo.port_rover, hence it is mostly useless. - * Code will work nicely if we just delete it, but - * I am afraid in contented case it will work not better or - * even worse: another cpu just will hit the same bucket - * and spin there. - * So some cpu salt could remove both contention and - * memory pingpong. Any ideas how to do this in a nice way? - */ - spin_lock(&dccp_hashinfo.portalloc_lock); - rover = dccp_hashinfo.port_rover; - do { - rover++; - if ((rover < low) || (rover > high)) - rover = low; head = &dccp_hashinfo.bhash[inet_bhashfn(rover, dccp_hashinfo.bhash_size)]; spin_lock(&head->lock); @@ -187,9 +164,9 @@ static int dccp_v4_hash_connect(struct sock *sk) next_port: spin_unlock(&head->lock); + if (++rover > high) + rover = low; } while (--remaining > 0); - dccp_hashinfo.port_rover = rover; - spin_unlock(&dccp_hashinfo.portalloc_lock); local_bh_enable(); @@ -197,9 +174,6 @@ static int dccp_v4_hash_connect(struct sock *sk) ok: /* All locks still held and bhs disabled */ - dccp_hashinfo.port_rover = rover; - spin_unlock(&dccp_hashinfo.portalloc_lock); - inet_bind_hash(sk, tb, rover); if (sk_unhashed(sk)) { inet_sk(sk)->sport = htons(rover); diff --git a/net/ipv4/inet_connection_sock.c b/net/ipv4/inet_connection_sock.c index 94468a76c5b4..3fe021f1a566 100644 --- a/net/ipv4/inet_connection_sock.c +++ b/net/ipv4/inet_connection_sock.c @@ -78,17 +78,9 @@ int inet_csk_get_port(struct inet_hashinfo *hashinfo, int low = sysctl_local_port_range[0]; int high = sysctl_local_port_range[1]; int remaining = (high - low) + 1; - int rover; + int rover = net_random() % (high - low) + low; - spin_lock(&hashinfo->portalloc_lock); - if (hashinfo->port_rover < low) - rover = low; - else - rover = hashinfo->port_rover; do { - rover++; - if (rover > high) - rover = low; head = &hashinfo->bhash[inet_bhashfn(rover, hashinfo->bhash_size)]; spin_lock(&head->lock); inet_bind_bucket_for_each(tb, node, &head->chain) @@ -97,9 +89,9 @@ int inet_csk_get_port(struct inet_hashinfo *hashinfo, break; next: spin_unlock(&head->lock); + if (++rover > high) + rover = low; } while (--remaining > 0); - hashinfo->port_rover = rover; - spin_unlock(&hashinfo->portalloc_lock); /* Exhausted local port range during search? It is not * possible for us to be holding one of the bind hash diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index f3f0013a9580..72b7c22e1ea5 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -2112,7 +2112,6 @@ void __init tcp_init(void) sysctl_tcp_max_orphans >>= (3 - order); sysctl_max_syn_backlog = 128; } - tcp_hashinfo.port_rover = sysctl_local_port_range[0] - 1; sysctl_tcp_mem[0] = 768 << order; sysctl_tcp_mem[1] = 1024 << order; diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c index c85819d8474b..49d67cd75edd 100644 --- a/net/ipv4/tcp_ipv4.c +++ b/net/ipv4/tcp_ipv4.c @@ -93,8 +93,6 @@ struct inet_hashinfo __cacheline_aligned tcp_hashinfo = { .lhash_lock = RW_LOCK_UNLOCKED, .lhash_users = ATOMIC_INIT(0), .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait), - .portalloc_lock = SPIN_LOCK_UNLOCKED, - .port_rover = 1024 - 1, }; static int tcp_v4_get_port(struct sock *sk, unsigned short snum) diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c index d693cb988b78..d746d3b27efb 100644 --- a/net/ipv6/tcp_ipv6.c +++ b/net/ipv6/tcp_ipv6.c @@ -114,16 +114,9 @@ static int tcp_v6_get_port(struct sock *sk, unsigned short snum) int low = sysctl_local_port_range[0]; int high = sysctl_local_port_range[1]; int remaining = (high - low) + 1; - int rover; + int rover = net_random() % (high - low) + low; - spin_lock(&tcp_hashinfo.portalloc_lock); - if (tcp_hashinfo.port_rover < low) - rover = low; - else - rover = tcp_hashinfo.port_rover; - do { rover++; - if (rover > high) - rover = low; + do { head = &tcp_hashinfo.bhash[inet_bhashfn(rover, tcp_hashinfo.bhash_size)]; spin_lock(&head->lock); inet_bind_bucket_for_each(tb, node, &head->chain) @@ -132,9 +125,9 @@ static int tcp_v6_get_port(struct sock *sk, unsigned short snum) break; next: spin_unlock(&head->lock); + if (++rover > high) + rover = low; } while (--remaining > 0); - tcp_hashinfo.port_rover = rover; - spin_unlock(&tcp_hashinfo.portalloc_lock); /* Exhausted local port range during search? It is not * possible for us to be holding one of the bind hash -- cgit v1.2.3