--- a/Documentation/Configure.help +++ b/Documentation/Configure.help @@ -11096,6 +11096,24 @@ CONFIG_NET_SCH_HFSC whenever you want). If you want to compile it as a module, say M here and read <file:Documentation/modules.txt>. +ESFQ queue +CONFIG_NET_SCH_ESFQ + Say Y here if you want to use the Stochastic Fairness Queueing (SFQ) + packet scheduling algorithm for some of your network devices or as a + leaf discipline for the CBQ scheduling algorithm (see the top of + <file:net/sched/sch_esfq.c> for details and references about the SFQ + algorithm). + + This is an enchanced SFQ version which allows you to control the + hardcoded values in the SFQ scheduler: queue depth, hash table size, + queues limit. Also adds control to the hash function used to identify + packet flows. Hash by src or dst ip and original sfq hash. + + This code is also available as a module called sch_esfq.o ( = code + which can be inserted in and removed from the running kernel + whenever you want). If you want to compile it as a module, say M + here and read <file:Documentation/modules.txt>. + CSZ packet scheduler CONFIG_NET_SCH_CSZ Say Y here if you want to use the Clark-Shenker-Zhang (CSZ) packet --- a/include/linux/pkt_sched.h +++ b/include/linux/pkt_sched.h @@ -173,8 +173,36 @@ struct tc_sfq_qopt * * The only reason for this is efficiency, it is possible * to change these parameters in compile time. + * + * If you need to play with these values use esfq instead. */ +/* ESFQ section */ + +enum +{ + /* traditional */ + TCA_SFQ_HASH_CLASSIC, + TCA_SFQ_HASH_DST, + TCA_SFQ_HASH_SRC, + /* conntrack */ + TCA_SFQ_HASH_CTORIGDST, + TCA_SFQ_HASH_CTORIGSRC, + TCA_SFQ_HASH_CTREPLDST, + TCA_SFQ_HASH_CTREPLSRC, + TCA_SFQ_HASH_CTNATCHG, +}; + +struct tc_esfq_qopt +{ + unsigned quantum; /* Bytes per round allocated to flow */ + int perturb_period; /* Period of hash perturbation */ + __u32 limit; /* Maximal packets in queue */ + unsigned divisor; /* Hash divisor */ + unsigned flows; /* Maximal number of flows */ + unsigned hash_kind; /* Hash function to use for flow identification */ +}; + /* RED section */ enum --- a/net/sched/Config.in +++ b/net/sched/Config.in @@ -12,6 +12,7 @@ fi tristate ' The simplest PRIO pseudoscheduler' CONFIG_NET_SCH_PRIO tristate ' RED queue' CONFIG_NET_SCH_RED tristate ' SFQ queue' CONFIG_NET_SCH_SFQ +tristate ' ESFQ queue' CONFIG_NET_SCH_ESFQ tristate ' TEQL queue' CONFIG_NET_SCH_TEQL tristate ' TBF queue' CONFIG_NET_SCH_TBF tristate ' GRED queue' CONFIG_NET_SCH_GRED --- a/net/sched/Makefile +++ b/net/sched/Makefile @@ -19,6 +19,7 @@ obj-$(CONFIG_NET_SCH_HPFQ) += sch_hpfq.o obj-$(CONFIG_NET_SCH_HFSC) += sch_hfsc.o obj-$(CONFIG_NET_SCH_HTB) += sch_htb.o obj-$(CONFIG_NET_SCH_SFQ) += sch_sfq.o +obj-$(CONFIG_NET_SCH_ESFQ) += sch_esfq.o obj-$(CONFIG_NET_SCH_RED) += sch_red.o obj-$(CONFIG_NET_SCH_TBF) += sch_tbf.o obj-$(CONFIG_NET_SCH_PRIO) += sch_prio.o --- /dev/null +++ b/net/sched/sch_esfq.c @@ -0,0 +1,649 @@ +/* + * net/sched/sch_esfq.c Extended Stochastic Fairness Queueing discipline. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version + * 2 of the License, or (at your option) any later version. + * + * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> + * + * Changes: Alexander Atanasov, <alex@ssi.bg> + * Added dynamic depth,limit,divisor,hash_kind options. + * Added dst and src hashes. + * + * Alexander Clouter, <alex@digriz.org.uk> + * Ported ESFQ to Linux 2.6. + * + * Corey Hickey, <bugfood-c@fatooh.org> + * Maintenance of the Linux 2.6 port. + * Added fwmark hash (thanks to Robert Kurjata). + * Added usage of jhash. + * + */ + +#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 <linux/init.h> +#include <net/ip.h> +#include <net/route.h> +#include <linux/skbuff.h> +#include <net/sock.h> +#include <net/pkt_sched.h> +#include <linux/jhash.h> + +#define IPPROTO_DCCP 33 +#define qdisc_priv(q) ((void *)(q->data)) + +#ifdef CONFIG_IP_NF_CONNTRACK +/* #include <net/netfilter/nf_conntrack.h> */ +#include <linux/netfilter_ipv4/ip_conntrack.h> +#endif + +/* Stochastic Fairness Queuing algorithm. + For more comments look at sch_sfq.c. + The difference is that you can change limit, depth, + hash table size and choose alternate hash types. + + classic: same as in sch_sfq.c + dst: destination IP address + src: source IP address + ctorigdst: original destination IP address + ctorigsrc: original source IP address + ctrepldst: reply destination IP address + ctreplsrc: reply source IP + ctnatchg: use the address which changed via nat + +*/ + + +/* This type should contain at least SFQ_DEPTH*2 values */ +typedef unsigned int esfq_index; + +struct esfq_head +{ + esfq_index next; + esfq_index prev; +}; + +struct esfq_sched_data +{ +/* Parameters */ + int perturb_period; + unsigned quantum; /* Allotment per round: MUST BE >= MTU */ + int limit; + unsigned depth; + unsigned hash_divisor; + unsigned hash_kind; +/* Variables */ + struct timer_list perturb_timer; + int perturbation; + esfq_index tail; /* Index of current slot in round */ + esfq_index max_depth; /* Maximal depth */ + + esfq_index *ht; /* Hash table */ + esfq_index *next; /* Active slots link */ + short *allot; /* Current allotment per slot */ + unsigned short *hash; /* Hash value indexed by slots */ + struct sk_buff_head *qs; /* Slot queue */ + struct esfq_head *dep; /* Linked list of slots, indexed by depth */ + unsigned dyn_min; /* For dynamic divisor adjustment; minimum value seen */ + unsigned dyn_max; /* maximum value seen */ + unsigned dyn_range; /* saved range */ +}; + +/* This contains the info we will hash. */ +struct esfq_packet_info +{ + u32 proto; /* protocol or port */ + u32 src; /* source from packet header */ + u32 dst; /* destination from packet header */ + u32 ctorigsrc; /* original source from conntrack */ + u32 ctorigdst; /* original destination from conntrack */ + u32 ctreplsrc; /* reply source from conntrack */ + u32 ctrepldst; /* reply destination from conntrack */ +}; + +static __inline__ unsigned esfq_jhash_1word(struct esfq_sched_data *q,u32 a) +{ + return jhash_1word(a, q->perturbation) & (q->hash_divisor-1); +} + +static __inline__ unsigned esfq_jhash_2words(struct esfq_sched_data *q, u32 a, u32 b) +{ + return jhash_2words(a, b, q->perturbation) & (q->hash_divisor-1); +} + +static __inline__ unsigned esfq_jhash_3words(struct esfq_sched_data *q, u32 a, u32 b, u32 c) +{ + return jhash_3words(a, b, c, q->perturbation) & (q->hash_divisor-1); +} + + +static unsigned esfq_hash(struct esfq_sched_data *q, struct sk_buff *skb) +{ + struct esfq_packet_info info; +#ifdef CONFIG_IP_NF_CONNTRACK + enum ip_conntrack_info ctinfo; + struct ip_conntrack *ct = ip_conntrack_get(skb, &ctinfo); +#endif + + switch (skb->protocol) { + case __constant_htons(ETH_P_IP): + { + struct iphdr *iph = skb->nh.iph; + info.dst = iph->daddr; + info.src = iph->saddr; + if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) && + (iph->protocol == IPPROTO_TCP || + iph->protocol == IPPROTO_UDP || + iph->protocol == IPPROTO_SCTP || + iph->protocol == IPPROTO_DCCP || + iph->protocol == IPPROTO_ESP)) + info.proto = *(((u32*)iph) + iph->ihl); + else + info.proto = iph->protocol; + break; + } + default: + info.dst = (u32)(unsigned long)skb->dst; + info.src = (u32)(unsigned long)skb->sk; + info.proto = skb->protocol; + } + +#ifdef CONFIG_IP_NF_CONNTRACK + /* defaults if there is no conntrack info */ + info.ctorigsrc = info.src; + info.ctorigdst = info.dst; + info.ctreplsrc = info.dst; + info.ctrepldst = info.src; + /* collect conntrack info */ + IP_NF_ASSERT(ct); + if (ct) { + if (skb->protocol == __constant_htons(ETH_P_IP)) { + info.ctorigsrc = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.ip; + info.ctorigdst = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.ip; + info.ctreplsrc = ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.ip; + info.ctrepldst = ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.ip; + } + } +#endif + + switch(q->hash_kind) + { + case TCA_SFQ_HASH_CLASSIC: + return esfq_jhash_3words(q, info.dst, info.src, info.proto); + case TCA_SFQ_HASH_DST: + return esfq_jhash_1word(q, info.dst); + case TCA_SFQ_HASH_SRC: + return esfq_jhash_1word(q, info.src); +#ifdef CONFIG_IP_NF_CONNTRACK + case TCA_SFQ_HASH_CTORIGDST: + return esfq_jhash_1word(q, info.ctorigdst); + case TCA_SFQ_HASH_CTORIGSRC: + return esfq_jhash_1word(q, info.ctorigsrc); + case TCA_SFQ_HASH_CTREPLDST: + return esfq_jhash_1word(q, info.ctrepldst); + case TCA_SFQ_HASH_CTREPLSRC: + return esfq_jhash_1word(q, info.ctreplsrc); + case TCA_SFQ_HASH_CTNATCHG: + { + if (info.ctorigdst == info.ctreplsrc) + return esfq_jhash_1word(q, info.ctorigsrc); + else + return esfq_jhash_1word(q, info.ctreplsrc); + } +#endif + default: + if (net_ratelimit()) + printk(KERN_WARNING "ESFQ: Unknown hash method. Falling back to classic.\n"); + } + return esfq_jhash_3words(q, info.dst, info.src, info.proto); +} + +static inline void esfq_link(struct esfq_sched_data *q, esfq_index x) +{ + esfq_index p, n; + int d = q->qs[x].qlen + q->depth; + + p = d; + n = q->dep[d].next; + q->dep[x].next = n; + q->dep[x].prev = p; + q->dep[p].next = q->dep[n].prev = x; +} + +static inline void esfq_dec(struct esfq_sched_data *q, esfq_index x) +{ + esfq_index p, n; + + n = q->dep[x].next; + p = q->dep[x].prev; + q->dep[p].next = n; + q->dep[n].prev = p; + + if (n == p && q->max_depth == q->qs[x].qlen + 1) + q->max_depth--; + + esfq_link(q, x); +} + +static inline void esfq_inc(struct esfq_sched_data *q, esfq_index x) +{ + esfq_index p, n; + int d; + + n = q->dep[x].next; + p = q->dep[x].prev; + q->dep[p].next = n; + q->dep[n].prev = p; + d = q->qs[x].qlen; + if (q->max_depth < d) + q->max_depth = d; + + esfq_link(q, x); +} + +static unsigned int esfq_drop(struct Qdisc *sch) +{ + struct esfq_sched_data *q = qdisc_priv(sch); + esfq_index d = q->max_depth; + struct sk_buff *skb; + unsigned int len; + + /* Queue is full! Find the longest slot and + drop a packet from it */ + + if (d > 1) { + esfq_index x = q->dep[d+q->depth].next; + skb = q->qs[x].prev; + len = skb->len; + __skb_unlink(skb, &q->qs[x]); + kfree_skb(skb); + esfq_dec(q, x); + sch->q.qlen--; + sch->stats.drops++; + sch->stats.backlog -= len; + return len; + } + + if (d == 1) { + /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */ + d = q->next[q->tail]; + q->next[q->tail] = q->next[d]; + q->allot[q->next[d]] += q->quantum; + skb = q->qs[d].prev; + len = skb->len; + __skb_unlink(skb, &q->qs[d]); + kfree_skb(skb); + esfq_dec(q, d); + sch->q.qlen--; + q->ht[q->hash[d]] = q->depth; + sch->stats.drops++; + sch->stats.backlog -= len; + return len; + } + + return 0; +} + +static int +esfq_enqueue(struct sk_buff *skb, struct Qdisc* sch) +{ + struct esfq_sched_data *q = qdisc_priv(sch); + unsigned hash = esfq_hash(q, skb); + unsigned depth = q->depth; + esfq_index x; + + x = q->ht[hash]; + if (x == depth) { + q->ht[hash] = x = q->dep[depth].next; + q->hash[x] = hash; + } + sch->stats.backlog += skb->len; + __skb_queue_tail(&q->qs[x], skb); + esfq_inc(q, x); + if (q->qs[x].qlen == 1) { /* The flow is new */ + if (q->tail == depth) { /* It is the first flow */ + q->tail = x; + q->next[x] = x; + q->allot[x] = q->quantum; + } else { + q->next[x] = q->next[q->tail]; + q->next[q->tail] = x; + q->tail = x; + } + } + if (++sch->q.qlen < q->limit-1) { + sch->stats.bytes += skb->len; + sch->stats.packets++; + return 0; + } + + esfq_drop(sch); + return NET_XMIT_CN; +} + +static int +esfq_requeue(struct sk_buff *skb, struct Qdisc* sch) +{ + struct esfq_sched_data *q = qdisc_priv(sch); + unsigned hash = esfq_hash(q, skb); + unsigned depth = q->depth; + esfq_index x; + + x = q->ht[hash]; + if (x == depth) { + q->ht[hash] = x = q->dep[depth].next; + q->hash[x] = hash; + } + sch->stats.backlog += skb->len; + __skb_queue_head(&q->qs[x], skb); + esfq_inc(q, x); + if (q->qs[x].qlen == 1) { /* The flow is new */ + if (q->tail == depth) { /* It is the first flow */ + q->tail = x; + q->next[x] = x; + q->allot[x] = q->quantum; + } else { + q->next[x] = q->next[q->tail]; + q->next[q->tail] = x; + q->tail = x; + } + } + if (++sch->q.qlen < q->limit - 1) { + return 0; + } + + sch->stats.drops++; + esfq_drop(sch); + return NET_XMIT_CN; +} + + + + +static struct sk_buff * +esfq_dequeue(struct Qdisc* sch) +{ + struct esfq_sched_data *q = qdisc_priv(sch); + struct sk_buff *skb; + unsigned depth = q->depth; + esfq_index a, old_a; + + /* No active slots */ + if (q->tail == depth) + return NULL; + + a = old_a = q->next[q->tail]; + + /* Grab packet */ + skb = __skb_dequeue(&q->qs[a]); + esfq_dec(q, a); + sch->q.qlen--; + sch->stats.backlog -= skb->len; + + /* Is the slot empty? */ + if (q->qs[a].qlen == 0) { + q->ht[q->hash[a]] = depth; + a = q->next[a]; + if (a == old_a) { + q->tail = depth; + return skb; + } + q->next[q->tail] = a; + q->allot[a] += q->quantum; + } else if ((q->allot[a] -= skb->len) <= 0) { + q->tail = a; + a = q->next[a]; + q->allot[a] += q->quantum; + } + + return skb; +} + +static void +esfq_reset(struct Qdisc* sch) +{ + struct sk_buff *skb; + + while ((skb = esfq_dequeue(sch)) != NULL) + kfree_skb(skb); +} + +static void esfq_perturbation(unsigned long arg) +{ + struct Qdisc *sch = (struct Qdisc*)arg; + struct esfq_sched_data *q = qdisc_priv(sch); + + q->perturbation = net_random()&0x1F; + + if (q->perturb_period) { + q->perturb_timer.expires = jiffies + q->perturb_period; + add_timer(&q->perturb_timer); + } +} + +static int esfq_change(struct Qdisc *sch, struct rtattr *opt) +{ + struct esfq_sched_data *q = qdisc_priv(sch); + struct tc_esfq_qopt *ctl = RTA_DATA(opt); + int old_perturb = q->perturb_period; + + if (opt->rta_len < RTA_LENGTH(sizeof(*ctl))) + return -EINVAL; + + sch_tree_lock(sch); + q->quantum = ctl->quantum ? : psched_mtu(sch->dev); + q->perturb_period = ctl->perturb_period*HZ; +// q->hash_divisor = ctl->divisor; +// q->tail = q->limit = q->depth = ctl->flows; + + if (ctl->limit) + q->limit = min_t(u32, ctl->limit, q->depth); + + if (ctl->hash_kind) { + q->hash_kind = ctl->hash_kind; + if (q->hash_kind != TCA_SFQ_HASH_CLASSIC) + q->perturb_period = 0; + } + + // is sch_tree_lock enough to do this ? + while (sch->q.qlen >= q->limit-1) + esfq_drop(sch); + + if (old_perturb) + del_timer(&q->perturb_timer); + if (q->perturb_period) { + q->perturb_timer.expires = jiffies + q->perturb_period; + add_timer(&q->perturb_timer); + } else { + q->perturbation = 0; + } + sch_tree_unlock(sch); + return 0; +} + +static int esfq_init(struct Qdisc *sch, struct rtattr *opt) +{ + struct esfq_sched_data *q = qdisc_priv(sch); + struct tc_esfq_qopt *ctl; + esfq_index p = ~0U/2; + int i; + + if (opt && opt->rta_len < RTA_LENGTH(sizeof(*ctl))) + return -EINVAL; + + init_timer(&q->perturb_timer); + q->perturb_timer.data = (unsigned long)sch; + q->perturb_timer.function = esfq_perturbation; + q->perturbation = 0; + q->hash_kind = TCA_SFQ_HASH_CLASSIC; + q->max_depth = 0; + q->dyn_min = ~0U; /* maximum value for this type */ + q->dyn_max = 0; /* dyn_min/dyn_max will be set properly upon first packet */ + if (opt == NULL) { + q->quantum = psched_mtu(sch->dev); + q->perturb_period = 0; + q->hash_divisor = 1024; + q->tail = q->limit = q->depth = 128; + + } else { + ctl = RTA_DATA(opt); + q->quantum = ctl->quantum ? : psched_mtu(sch->dev); + q->perturb_period = ctl->perturb_period*HZ; + q->hash_divisor = ctl->divisor ? : 1024; + q->tail = q->limit = q->depth = ctl->flows ? : 128; + + if ( q->depth > p - 1 ) + return -EINVAL; + + if (ctl->limit) + q->limit = min_t(u32, ctl->limit, q->depth); + + if (ctl->hash_kind) { + q->hash_kind = ctl->hash_kind; + } + + if (q->perturb_period) { + q->perturb_timer.expires = jiffies + q->perturb_period; + add_timer(&q->perturb_timer); + } + } + + q->ht = kmalloc(q->hash_divisor*sizeof(esfq_index), GFP_KERNEL); + if (!q->ht) + goto err_case; + + q->dep = kmalloc((1+q->depth*2)*sizeof(struct esfq_head), GFP_KERNEL); + if (!q->dep) + goto err_case; + q->next = kmalloc(q->depth*sizeof(esfq_index), GFP_KERNEL); + if (!q->next) + goto err_case; + + q->allot = kmalloc(q->depth*sizeof(short), GFP_KERNEL); + if (!q->allot) + goto err_case; + q->hash = kmalloc(q->depth*sizeof(unsigned short), GFP_KERNEL); + if (!q->hash) + goto err_case; + q->qs = kmalloc(q->depth*sizeof(struct sk_buff_head), GFP_KERNEL); + if (!q->qs) + goto err_case; + + for (i=0; i< q->hash_divisor; i++) + q->ht[i] = q->depth; + for (i=0; i<q->depth; i++) { + skb_queue_head_init(&q->qs[i]); + q->dep[i+q->depth].next = i+q->depth; + q->dep[i+q->depth].prev = i+q->depth; + } + + for (i=0; i<q->depth; i++) + esfq_link(q, i); + return 0; +err_case: + del_timer(&q->perturb_timer); + if (q->ht) + kfree(q->ht); + if (q->dep) + kfree(q->dep); + if (q->next) + kfree(q->next); + if (q->allot) + kfree(q->allot); + if (q->hash) + kfree(q->hash); + if (q->qs) + kfree(q->qs); + return -ENOBUFS; +} + +static void esfq_destroy(struct Qdisc *sch) +{ + struct esfq_sched_data *q = qdisc_priv(sch); + del_timer(&q->perturb_timer); + if(q->ht) + kfree(q->ht); + if(q->dep) + kfree(q->dep); + if(q->next) + kfree(q->next); + if(q->allot) + kfree(q->allot); + if(q->hash) + kfree(q->hash); + if(q->qs) + kfree(q->qs); +} + +static int esfq_dump(struct Qdisc *sch, struct sk_buff *skb) +{ + struct esfq_sched_data *q = qdisc_priv(sch); + unsigned char *b = skb->tail; + struct tc_esfq_qopt opt; + + opt.quantum = q->quantum; + opt.perturb_period = q->perturb_period/HZ; + + opt.limit = q->limit; + opt.divisor = q->hash_divisor; + opt.flows = q->depth; + opt.hash_kind = q->hash_kind; + + RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); + + return skb->len; + +rtattr_failure: + skb_trim(skb, b - skb->data); + return -1; +} + +static struct Qdisc_ops esfq_qdisc_ops = +{ + .next = NULL, + .cl_ops = NULL, + .id = "esfq", + .priv_size = sizeof(struct esfq_sched_data), + .enqueue = esfq_enqueue, + .dequeue = esfq_dequeue, + .requeue = esfq_requeue, + .drop = esfq_drop, + .init = esfq_init, + .reset = esfq_reset, + .destroy = esfq_destroy, + .change = NULL, /* esfq_change - needs more work */ + .dump = esfq_dump, +}; + +static int __init esfq_module_init(void) +{ + return register_qdisc(&esfq_qdisc_ops); +} +static void __exit esfq_module_exit(void) +{ + unregister_qdisc(&esfq_qdisc_ops); +} +module_init(esfq_module_init) +module_exit(esfq_module_exit) +MODULE_LICENSE("GPL");