| 1 | --- a/Documentation/Configure.help |
| 2 | +++ b/Documentation/Configure.help |
| 3 | @@ -11096,6 +11096,24 @@ CONFIG_NET_SCH_HFSC |
| 4 | whenever you want). If you want to compile it as a module, say M |
| 5 | here and read <file:Documentation/modules.txt>. |
| 6 | |
| 7 | +ESFQ queue |
| 8 | +CONFIG_NET_SCH_ESFQ |
| 9 | + Say Y here if you want to use the Stochastic Fairness Queueing (SFQ) |
| 10 | + packet scheduling algorithm for some of your network devices or as a |
| 11 | + leaf discipline for the CBQ scheduling algorithm (see the top of |
| 12 | + <file:net/sched/sch_esfq.c> for details and references about the SFQ |
| 13 | + algorithm). |
| 14 | + |
| 15 | + This is an enchanced SFQ version which allows you to control the |
| 16 | + hardcoded values in the SFQ scheduler: queue depth, hash table size, |
| 17 | + queues limit. Also adds control to the hash function used to identify |
| 18 | + packet flows. Hash by src or dst ip and original sfq hash. |
| 19 | + |
| 20 | + This code is also available as a module called sch_esfq.o ( = code |
| 21 | + which can be inserted in and removed from the running kernel |
| 22 | + whenever you want). If you want to compile it as a module, say M |
| 23 | + here and read <file:Documentation/modules.txt>. |
| 24 | + |
| 25 | CSZ packet scheduler |
| 26 | CONFIG_NET_SCH_CSZ |
| 27 | Say Y here if you want to use the Clark-Shenker-Zhang (CSZ) packet |
| 28 | --- a/include/linux/pkt_sched.h |
| 29 | +++ b/include/linux/pkt_sched.h |
| 30 | @@ -173,8 +173,36 @@ struct tc_sfq_qopt |
| 31 | * |
| 32 | * The only reason for this is efficiency, it is possible |
| 33 | * to change these parameters in compile time. |
| 34 | + * |
| 35 | + * If you need to play with these values use esfq instead. |
| 36 | */ |
| 37 | |
| 38 | +/* ESFQ section */ |
| 39 | + |
| 40 | +enum |
| 41 | +{ |
| 42 | + /* traditional */ |
| 43 | + TCA_SFQ_HASH_CLASSIC, |
| 44 | + TCA_SFQ_HASH_DST, |
| 45 | + TCA_SFQ_HASH_SRC, |
| 46 | + /* conntrack */ |
| 47 | + TCA_SFQ_HASH_CTORIGDST, |
| 48 | + TCA_SFQ_HASH_CTORIGSRC, |
| 49 | + TCA_SFQ_HASH_CTREPLDST, |
| 50 | + TCA_SFQ_HASH_CTREPLSRC, |
| 51 | + TCA_SFQ_HASH_CTNATCHG, |
| 52 | +}; |
| 53 | + |
| 54 | +struct tc_esfq_qopt |
| 55 | +{ |
| 56 | + unsigned quantum; /* Bytes per round allocated to flow */ |
| 57 | + int perturb_period; /* Period of hash perturbation */ |
| 58 | + __u32 limit; /* Maximal packets in queue */ |
| 59 | + unsigned divisor; /* Hash divisor */ |
| 60 | + unsigned flows; /* Maximal number of flows */ |
| 61 | + unsigned hash_kind; /* Hash function to use for flow identification */ |
| 62 | +}; |
| 63 | + |
| 64 | /* RED section */ |
| 65 | |
| 66 | enum |
| 67 | --- a/net/sched/Config.in |
| 68 | +++ b/net/sched/Config.in |
| 69 | @@ -12,6 +12,7 @@ fi |
| 70 | tristate ' The simplest PRIO pseudoscheduler' CONFIG_NET_SCH_PRIO |
| 71 | tristate ' RED queue' CONFIG_NET_SCH_RED |
| 72 | tristate ' SFQ queue' CONFIG_NET_SCH_SFQ |
| 73 | +tristate ' ESFQ queue' CONFIG_NET_SCH_ESFQ |
| 74 | tristate ' TEQL queue' CONFIG_NET_SCH_TEQL |
| 75 | tristate ' TBF queue' CONFIG_NET_SCH_TBF |
| 76 | tristate ' GRED queue' CONFIG_NET_SCH_GRED |
| 77 | --- a/net/sched/Makefile |
| 78 | +++ b/net/sched/Makefile |
| 79 | @@ -19,6 +19,7 @@ obj-$(CONFIG_NET_SCH_HPFQ) += sch_hpfq.o |
| 80 | obj-$(CONFIG_NET_SCH_HFSC) += sch_hfsc.o |
| 81 | obj-$(CONFIG_NET_SCH_HTB) += sch_htb.o |
| 82 | obj-$(CONFIG_NET_SCH_SFQ) += sch_sfq.o |
| 83 | +obj-$(CONFIG_NET_SCH_ESFQ) += sch_esfq.o |
| 84 | obj-$(CONFIG_NET_SCH_RED) += sch_red.o |
| 85 | obj-$(CONFIG_NET_SCH_TBF) += sch_tbf.o |
| 86 | obj-$(CONFIG_NET_SCH_PRIO) += sch_prio.o |
| 87 | --- /dev/null |
| 88 | +++ b/net/sched/sch_esfq.c |
| 89 | @@ -0,0 +1,649 @@ |
| 90 | +/* |
| 91 | + * net/sched/sch_esfq.c Extended Stochastic Fairness Queueing discipline. |
| 92 | + * |
| 93 | + * This program is free software; you can redistribute it and/or |
| 94 | + * modify it under the terms of the GNU General Public License |
| 95 | + * as published by the Free Software Foundation; either version |
| 96 | + * 2 of the License, or (at your option) any later version. |
| 97 | + * |
| 98 | + * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> |
| 99 | + * |
| 100 | + * Changes: Alexander Atanasov, <alex@ssi.bg> |
| 101 | + * Added dynamic depth,limit,divisor,hash_kind options. |
| 102 | + * Added dst and src hashes. |
| 103 | + * |
| 104 | + * Alexander Clouter, <alex@digriz.org.uk> |
| 105 | + * Ported ESFQ to Linux 2.6. |
| 106 | + * |
| 107 | + * Corey Hickey, <bugfood-c@fatooh.org> |
| 108 | + * Maintenance of the Linux 2.6 port. |
| 109 | + * Added fwmark hash (thanks to Robert Kurjata). |
| 110 | + * Added usage of jhash. |
| 111 | + * |
| 112 | + */ |
| 113 | + |
| 114 | +#include <linux/config.h> |
| 115 | +#include <linux/module.h> |
| 116 | +#include <asm/uaccess.h> |
| 117 | +#include <asm/system.h> |
| 118 | +#include <linux/bitops.h> |
| 119 | +#include <linux/types.h> |
| 120 | +#include <linux/kernel.h> |
| 121 | +#include <linux/sched.h> |
| 122 | +#include <linux/string.h> |
| 123 | +#include <linux/mm.h> |
| 124 | +#include <linux/socket.h> |
| 125 | +#include <linux/sockios.h> |
| 126 | +#include <linux/in.h> |
| 127 | +#include <linux/errno.h> |
| 128 | +#include <linux/interrupt.h> |
| 129 | +#include <linux/if_ether.h> |
| 130 | +#include <linux/inet.h> |
| 131 | +#include <linux/netdevice.h> |
| 132 | +#include <linux/etherdevice.h> |
| 133 | +#include <linux/notifier.h> |
| 134 | +#include <linux/init.h> |
| 135 | +#include <net/ip.h> |
| 136 | +#include <net/route.h> |
| 137 | +#include <linux/skbuff.h> |
| 138 | +#include <net/sock.h> |
| 139 | +#include <net/pkt_sched.h> |
| 140 | +#include <linux/jhash.h> |
| 141 | + |
| 142 | +#define IPPROTO_DCCP 33 |
| 143 | +#define qdisc_priv(q) ((void *)(q->data)) |
| 144 | + |
| 145 | +#ifdef CONFIG_IP_NF_CONNTRACK |
| 146 | +/* #include <net/netfilter/nf_conntrack.h> */ |
| 147 | +#include <linux/netfilter_ipv4/ip_conntrack.h> |
| 148 | +#endif |
| 149 | + |
| 150 | +/* Stochastic Fairness Queuing algorithm. |
| 151 | + For more comments look at sch_sfq.c. |
| 152 | + The difference is that you can change limit, depth, |
| 153 | + hash table size and choose alternate hash types. |
| 154 | + |
| 155 | + classic: same as in sch_sfq.c |
| 156 | + dst: destination IP address |
| 157 | + src: source IP address |
| 158 | + ctorigdst: original destination IP address |
| 159 | + ctorigsrc: original source IP address |
| 160 | + ctrepldst: reply destination IP address |
| 161 | + ctreplsrc: reply source IP |
| 162 | + ctnatchg: use the address which changed via nat |
| 163 | + |
| 164 | +*/ |
| 165 | + |
| 166 | + |
| 167 | +/* This type should contain at least SFQ_DEPTH*2 values */ |
| 168 | +typedef unsigned int esfq_index; |
| 169 | + |
| 170 | +struct esfq_head |
| 171 | +{ |
| 172 | + esfq_index next; |
| 173 | + esfq_index prev; |
| 174 | +}; |
| 175 | + |
| 176 | +struct esfq_sched_data |
| 177 | +{ |
| 178 | +/* Parameters */ |
| 179 | + int perturb_period; |
| 180 | + unsigned quantum; /* Allotment per round: MUST BE >= MTU */ |
| 181 | + int limit; |
| 182 | + unsigned depth; |
| 183 | + unsigned hash_divisor; |
| 184 | + unsigned hash_kind; |
| 185 | +/* Variables */ |
| 186 | + struct timer_list perturb_timer; |
| 187 | + int perturbation; |
| 188 | + esfq_index tail; /* Index of current slot in round */ |
| 189 | + esfq_index max_depth; /* Maximal depth */ |
| 190 | + |
| 191 | + esfq_index *ht; /* Hash table */ |
| 192 | + esfq_index *next; /* Active slots link */ |
| 193 | + short *allot; /* Current allotment per slot */ |
| 194 | + unsigned short *hash; /* Hash value indexed by slots */ |
| 195 | + struct sk_buff_head *qs; /* Slot queue */ |
| 196 | + struct esfq_head *dep; /* Linked list of slots, indexed by depth */ |
| 197 | + unsigned dyn_min; /* For dynamic divisor adjustment; minimum value seen */ |
| 198 | + unsigned dyn_max; /* maximum value seen */ |
| 199 | + unsigned dyn_range; /* saved range */ |
| 200 | +}; |
| 201 | + |
| 202 | +/* This contains the info we will hash. */ |
| 203 | +struct esfq_packet_info |
| 204 | +{ |
| 205 | + u32 proto; /* protocol or port */ |
| 206 | + u32 src; /* source from packet header */ |
| 207 | + u32 dst; /* destination from packet header */ |
| 208 | + u32 ctorigsrc; /* original source from conntrack */ |
| 209 | + u32 ctorigdst; /* original destination from conntrack */ |
| 210 | + u32 ctreplsrc; /* reply source from conntrack */ |
| 211 | + u32 ctrepldst; /* reply destination from conntrack */ |
| 212 | +}; |
| 213 | + |
| 214 | +static __inline__ unsigned esfq_jhash_1word(struct esfq_sched_data *q,u32 a) |
| 215 | +{ |
| 216 | + return jhash_1word(a, q->perturbation) & (q->hash_divisor-1); |
| 217 | +} |
| 218 | + |
| 219 | +static __inline__ unsigned esfq_jhash_2words(struct esfq_sched_data *q, u32 a, u32 b) |
| 220 | +{ |
| 221 | + return jhash_2words(a, b, q->perturbation) & (q->hash_divisor-1); |
| 222 | +} |
| 223 | + |
| 224 | +static __inline__ unsigned esfq_jhash_3words(struct esfq_sched_data *q, u32 a, u32 b, u32 c) |
| 225 | +{ |
| 226 | + return jhash_3words(a, b, c, q->perturbation) & (q->hash_divisor-1); |
| 227 | +} |
| 228 | + |
| 229 | + |
| 230 | +static unsigned esfq_hash(struct esfq_sched_data *q, struct sk_buff *skb) |
| 231 | +{ |
| 232 | + struct esfq_packet_info info; |
| 233 | +#ifdef CONFIG_IP_NF_CONNTRACK |
| 234 | + enum ip_conntrack_info ctinfo; |
| 235 | + struct ip_conntrack *ct = ip_conntrack_get(skb, &ctinfo); |
| 236 | +#endif |
| 237 | + |
| 238 | + switch (skb->protocol) { |
| 239 | + case __constant_htons(ETH_P_IP): |
| 240 | + { |
| 241 | + struct iphdr *iph = skb->nh.iph; |
| 242 | + info.dst = iph->daddr; |
| 243 | + info.src = iph->saddr; |
| 244 | + if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) && |
| 245 | + (iph->protocol == IPPROTO_TCP || |
| 246 | + iph->protocol == IPPROTO_UDP || |
| 247 | + iph->protocol == IPPROTO_SCTP || |
| 248 | + iph->protocol == IPPROTO_DCCP || |
| 249 | + iph->protocol == IPPROTO_ESP)) |
| 250 | + info.proto = *(((u32*)iph) + iph->ihl); |
| 251 | + else |
| 252 | + info.proto = iph->protocol; |
| 253 | + break; |
| 254 | + } |
| 255 | + default: |
| 256 | + info.dst = (u32)(unsigned long)skb->dst; |
| 257 | + info.src = (u32)(unsigned long)skb->sk; |
| 258 | + info.proto = skb->protocol; |
| 259 | + } |
| 260 | + |
| 261 | +#ifdef CONFIG_IP_NF_CONNTRACK |
| 262 | + /* defaults if there is no conntrack info */ |
| 263 | + info.ctorigsrc = info.src; |
| 264 | + info.ctorigdst = info.dst; |
| 265 | + info.ctreplsrc = info.dst; |
| 266 | + info.ctrepldst = info.src; |
| 267 | + /* collect conntrack info */ |
| 268 | + IP_NF_ASSERT(ct); |
| 269 | + if (ct) { |
| 270 | + if (skb->protocol == __constant_htons(ETH_P_IP)) { |
| 271 | + info.ctorigsrc = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.ip; |
| 272 | + info.ctorigdst = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.ip; |
| 273 | + info.ctreplsrc = ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.ip; |
| 274 | + info.ctrepldst = ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.ip; |
| 275 | + } |
| 276 | + } |
| 277 | +#endif |
| 278 | + |
| 279 | + switch(q->hash_kind) |
| 280 | + { |
| 281 | + case TCA_SFQ_HASH_CLASSIC: |
| 282 | + return esfq_jhash_3words(q, info.dst, info.src, info.proto); |
| 283 | + case TCA_SFQ_HASH_DST: |
| 284 | + return esfq_jhash_1word(q, info.dst); |
| 285 | + case TCA_SFQ_HASH_SRC: |
| 286 | + return esfq_jhash_1word(q, info.src); |
| 287 | +#ifdef CONFIG_IP_NF_CONNTRACK |
| 288 | + case TCA_SFQ_HASH_CTORIGDST: |
| 289 | + return esfq_jhash_1word(q, info.ctorigdst); |
| 290 | + case TCA_SFQ_HASH_CTORIGSRC: |
| 291 | + return esfq_jhash_1word(q, info.ctorigsrc); |
| 292 | + case TCA_SFQ_HASH_CTREPLDST: |
| 293 | + return esfq_jhash_1word(q, info.ctrepldst); |
| 294 | + case TCA_SFQ_HASH_CTREPLSRC: |
| 295 | + return esfq_jhash_1word(q, info.ctreplsrc); |
| 296 | + case TCA_SFQ_HASH_CTNATCHG: |
| 297 | + { |
| 298 | + if (info.ctorigdst == info.ctreplsrc) |
| 299 | + return esfq_jhash_1word(q, info.ctorigsrc); |
| 300 | + else |
| 301 | + return esfq_jhash_1word(q, info.ctreplsrc); |
| 302 | + } |
| 303 | +#endif |
| 304 | + default: |
| 305 | + if (net_ratelimit()) |
| 306 | + printk(KERN_WARNING "ESFQ: Unknown hash method. Falling back to classic.\n"); |
| 307 | + } |
| 308 | + return esfq_jhash_3words(q, info.dst, info.src, info.proto); |
| 309 | +} |
| 310 | + |
| 311 | +static inline void esfq_link(struct esfq_sched_data *q, esfq_index x) |
| 312 | +{ |
| 313 | + esfq_index p, n; |
| 314 | + int d = q->qs[x].qlen + q->depth; |
| 315 | + |
| 316 | + p = d; |
| 317 | + n = q->dep[d].next; |
| 318 | + q->dep[x].next = n; |
| 319 | + q->dep[x].prev = p; |
| 320 | + q->dep[p].next = q->dep[n].prev = x; |
| 321 | +} |
| 322 | + |
| 323 | +static inline void esfq_dec(struct esfq_sched_data *q, esfq_index x) |
| 324 | +{ |
| 325 | + esfq_index p, n; |
| 326 | + |
| 327 | + n = q->dep[x].next; |
| 328 | + p = q->dep[x].prev; |
| 329 | + q->dep[p].next = n; |
| 330 | + q->dep[n].prev = p; |
| 331 | + |
| 332 | + if (n == p && q->max_depth == q->qs[x].qlen + 1) |
| 333 | + q->max_depth--; |
| 334 | + |
| 335 | + esfq_link(q, x); |
| 336 | +} |
| 337 | + |
| 338 | +static inline void esfq_inc(struct esfq_sched_data *q, esfq_index x) |
| 339 | +{ |
| 340 | + esfq_index p, n; |
| 341 | + int d; |
| 342 | + |
| 343 | + n = q->dep[x].next; |
| 344 | + p = q->dep[x].prev; |
| 345 | + q->dep[p].next = n; |
| 346 | + q->dep[n].prev = p; |
| 347 | + d = q->qs[x].qlen; |
| 348 | + if (q->max_depth < d) |
| 349 | + q->max_depth = d; |
| 350 | + |
| 351 | + esfq_link(q, x); |
| 352 | +} |
| 353 | + |
| 354 | +static unsigned int esfq_drop(struct Qdisc *sch) |
| 355 | +{ |
| 356 | + struct esfq_sched_data *q = qdisc_priv(sch); |
| 357 | + esfq_index d = q->max_depth; |
| 358 | + struct sk_buff *skb; |
| 359 | + unsigned int len; |
| 360 | + |
| 361 | + /* Queue is full! Find the longest slot and |
| 362 | + drop a packet from it */ |
| 363 | + |
| 364 | + if (d > 1) { |
| 365 | + esfq_index x = q->dep[d+q->depth].next; |
| 366 | + skb = q->qs[x].prev; |
| 367 | + len = skb->len; |
| 368 | + __skb_unlink(skb, &q->qs[x]); |
| 369 | + kfree_skb(skb); |
| 370 | + esfq_dec(q, x); |
| 371 | + sch->q.qlen--; |
| 372 | + sch->stats.drops++; |
| 373 | + sch->stats.backlog -= len; |
| 374 | + return len; |
| 375 | + } |
| 376 | + |
| 377 | + if (d == 1) { |
| 378 | + /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */ |
| 379 | + d = q->next[q->tail]; |
| 380 | + q->next[q->tail] = q->next[d]; |
| 381 | + q->allot[q->next[d]] += q->quantum; |
| 382 | + skb = q->qs[d].prev; |
| 383 | + len = skb->len; |
| 384 | + __skb_unlink(skb, &q->qs[d]); |
| 385 | + kfree_skb(skb); |
| 386 | + esfq_dec(q, d); |
| 387 | + sch->q.qlen--; |
| 388 | + q->ht[q->hash[d]] = q->depth; |
| 389 | + sch->stats.drops++; |
| 390 | + sch->stats.backlog -= len; |
| 391 | + return len; |
| 392 | + } |
| 393 | + |
| 394 | + return 0; |
| 395 | +} |
| 396 | + |
| 397 | +static int |
| 398 | +esfq_enqueue(struct sk_buff *skb, struct Qdisc* sch) |
| 399 | +{ |
| 400 | + struct esfq_sched_data *q = qdisc_priv(sch); |
| 401 | + unsigned hash = esfq_hash(q, skb); |
| 402 | + unsigned depth = q->depth; |
| 403 | + esfq_index x; |
| 404 | + |
| 405 | + x = q->ht[hash]; |
| 406 | + if (x == depth) { |
| 407 | + q->ht[hash] = x = q->dep[depth].next; |
| 408 | + q->hash[x] = hash; |
| 409 | + } |
| 410 | + sch->stats.backlog += skb->len; |
| 411 | + __skb_queue_tail(&q->qs[x], skb); |
| 412 | + esfq_inc(q, x); |
| 413 | + if (q->qs[x].qlen == 1) { /* The flow is new */ |
| 414 | + if (q->tail == depth) { /* It is the first flow */ |
| 415 | + q->tail = x; |
| 416 | + q->next[x] = x; |
| 417 | + q->allot[x] = q->quantum; |
| 418 | + } else { |
| 419 | + q->next[x] = q->next[q->tail]; |
| 420 | + q->next[q->tail] = x; |
| 421 | + q->tail = x; |
| 422 | + } |
| 423 | + } |
| 424 | + if (++sch->q.qlen < q->limit-1) { |
| 425 | + sch->stats.bytes += skb->len; |
| 426 | + sch->stats.packets++; |
| 427 | + return 0; |
| 428 | + } |
| 429 | + |
| 430 | + esfq_drop(sch); |
| 431 | + return NET_XMIT_CN; |
| 432 | +} |
| 433 | + |
| 434 | +static int |
| 435 | +esfq_requeue(struct sk_buff *skb, struct Qdisc* sch) |
| 436 | +{ |
| 437 | + struct esfq_sched_data *q = qdisc_priv(sch); |
| 438 | + unsigned hash = esfq_hash(q, skb); |
| 439 | + unsigned depth = q->depth; |
| 440 | + esfq_index x; |
| 441 | + |
| 442 | + x = q->ht[hash]; |
| 443 | + if (x == depth) { |
| 444 | + q->ht[hash] = x = q->dep[depth].next; |
| 445 | + q->hash[x] = hash; |
| 446 | + } |
| 447 | + sch->stats.backlog += skb->len; |
| 448 | + __skb_queue_head(&q->qs[x], skb); |
| 449 | + esfq_inc(q, x); |
| 450 | + if (q->qs[x].qlen == 1) { /* The flow is new */ |
| 451 | + if (q->tail == depth) { /* It is the first flow */ |
| 452 | + q->tail = x; |
| 453 | + q->next[x] = x; |
| 454 | + q->allot[x] = q->quantum; |
| 455 | + } else { |
| 456 | + q->next[x] = q->next[q->tail]; |
| 457 | + q->next[q->tail] = x; |
| 458 | + q->tail = x; |
| 459 | + } |
| 460 | + } |
| 461 | + if (++sch->q.qlen < q->limit - 1) { |
| 462 | + return 0; |
| 463 | + } |
| 464 | + |
| 465 | + sch->stats.drops++; |
| 466 | + esfq_drop(sch); |
| 467 | + return NET_XMIT_CN; |
| 468 | +} |
| 469 | + |
| 470 | + |
| 471 | + |
| 472 | + |
| 473 | +static struct sk_buff * |
| 474 | +esfq_dequeue(struct Qdisc* sch) |
| 475 | +{ |
| 476 | + struct esfq_sched_data *q = qdisc_priv(sch); |
| 477 | + struct sk_buff *skb; |
| 478 | + unsigned depth = q->depth; |
| 479 | + esfq_index a, old_a; |
| 480 | + |
| 481 | + /* No active slots */ |
| 482 | + if (q->tail == depth) |
| 483 | + return NULL; |
| 484 | + |
| 485 | + a = old_a = q->next[q->tail]; |
| 486 | + |
| 487 | + /* Grab packet */ |
| 488 | + skb = __skb_dequeue(&q->qs[a]); |
| 489 | + esfq_dec(q, a); |
| 490 | + sch->q.qlen--; |
| 491 | + sch->stats.backlog -= skb->len; |
| 492 | + |
| 493 | + /* Is the slot empty? */ |
| 494 | + if (q->qs[a].qlen == 0) { |
| 495 | + q->ht[q->hash[a]] = depth; |
| 496 | + a = q->next[a]; |
| 497 | + if (a == old_a) { |
| 498 | + q->tail = depth; |
| 499 | + return skb; |
| 500 | + } |
| 501 | + q->next[q->tail] = a; |
| 502 | + q->allot[a] += q->quantum; |
| 503 | + } else if ((q->allot[a] -= skb->len) <= 0) { |
| 504 | + q->tail = a; |
| 505 | + a = q->next[a]; |
| 506 | + q->allot[a] += q->quantum; |
| 507 | + } |
| 508 | + |
| 509 | + return skb; |
| 510 | +} |
| 511 | + |
| 512 | +static void |
| 513 | +esfq_reset(struct Qdisc* sch) |
| 514 | +{ |
| 515 | + struct sk_buff *skb; |
| 516 | + |
| 517 | + while ((skb = esfq_dequeue(sch)) != NULL) |
| 518 | + kfree_skb(skb); |
| 519 | +} |
| 520 | + |
| 521 | +static void esfq_perturbation(unsigned long arg) |
| 522 | +{ |
| 523 | + struct Qdisc *sch = (struct Qdisc*)arg; |
| 524 | + struct esfq_sched_data *q = qdisc_priv(sch); |
| 525 | + |
| 526 | + q->perturbation = net_random()&0x1F; |
| 527 | + |
| 528 | + if (q->perturb_period) { |
| 529 | + q->perturb_timer.expires = jiffies + q->perturb_period; |
| 530 | + add_timer(&q->perturb_timer); |
| 531 | + } |
| 532 | +} |
| 533 | + |
| 534 | +static int esfq_change(struct Qdisc *sch, struct rtattr *opt) |
| 535 | +{ |
| 536 | + struct esfq_sched_data *q = qdisc_priv(sch); |
| 537 | + struct tc_esfq_qopt *ctl = RTA_DATA(opt); |
| 538 | + int old_perturb = q->perturb_period; |
| 539 | + |
| 540 | + if (opt->rta_len < RTA_LENGTH(sizeof(*ctl))) |
| 541 | + return -EINVAL; |
| 542 | + |
| 543 | + sch_tree_lock(sch); |
| 544 | + q->quantum = ctl->quantum ? : psched_mtu(sch->dev); |
| 545 | + q->perturb_period = ctl->perturb_period*HZ; |
| 546 | +// q->hash_divisor = ctl->divisor; |
| 547 | +// q->tail = q->limit = q->depth = ctl->flows; |
| 548 | + |
| 549 | + if (ctl->limit) |
| 550 | + q->limit = min_t(u32, ctl->limit, q->depth); |
| 551 | + |
| 552 | + if (ctl->hash_kind) { |
| 553 | + q->hash_kind = ctl->hash_kind; |
| 554 | + if (q->hash_kind != TCA_SFQ_HASH_CLASSIC) |
| 555 | + q->perturb_period = 0; |
| 556 | + } |
| 557 | + |
| 558 | + // is sch_tree_lock enough to do this ? |
| 559 | + while (sch->q.qlen >= q->limit-1) |
| 560 | + esfq_drop(sch); |
| 561 | + |
| 562 | + if (old_perturb) |
| 563 | + del_timer(&q->perturb_timer); |
| 564 | + if (q->perturb_period) { |
| 565 | + q->perturb_timer.expires = jiffies + q->perturb_period; |
| 566 | + add_timer(&q->perturb_timer); |
| 567 | + } else { |
| 568 | + q->perturbation = 0; |
| 569 | + } |
| 570 | + sch_tree_unlock(sch); |
| 571 | + return 0; |
| 572 | +} |
| 573 | + |
| 574 | +static int esfq_init(struct Qdisc *sch, struct rtattr *opt) |
| 575 | +{ |
| 576 | + struct esfq_sched_data *q = qdisc_priv(sch); |
| 577 | + struct tc_esfq_qopt *ctl; |
| 578 | + esfq_index p = ~0U/2; |
| 579 | + int i; |
| 580 | + |
| 581 | + if (opt && opt->rta_len < RTA_LENGTH(sizeof(*ctl))) |
| 582 | + return -EINVAL; |
| 583 | + |
| 584 | + init_timer(&q->perturb_timer); |
| 585 | + q->perturb_timer.data = (unsigned long)sch; |
| 586 | + q->perturb_timer.function = esfq_perturbation; |
| 587 | + q->perturbation = 0; |
| 588 | + q->hash_kind = TCA_SFQ_HASH_CLASSIC; |
| 589 | + q->max_depth = 0; |
| 590 | + q->dyn_min = ~0U; /* maximum value for this type */ |
| 591 | + q->dyn_max = 0; /* dyn_min/dyn_max will be set properly upon first packet */ |
| 592 | + if (opt == NULL) { |
| 593 | + q->quantum = psched_mtu(sch->dev); |
| 594 | + q->perturb_period = 0; |
| 595 | + q->hash_divisor = 1024; |
| 596 | + q->tail = q->limit = q->depth = 128; |
| 597 | + |
| 598 | + } else { |
| 599 | + ctl = RTA_DATA(opt); |
| 600 | + q->quantum = ctl->quantum ? : psched_mtu(sch->dev); |
| 601 | + q->perturb_period = ctl->perturb_period*HZ; |
| 602 | + q->hash_divisor = ctl->divisor ? : 1024; |
| 603 | + q->tail = q->limit = q->depth = ctl->flows ? : 128; |
| 604 | + |
| 605 | + if ( q->depth > p - 1 ) |
| 606 | + return -EINVAL; |
| 607 | + |
| 608 | + if (ctl->limit) |
| 609 | + q->limit = min_t(u32, ctl->limit, q->depth); |
| 610 | + |
| 611 | + if (ctl->hash_kind) { |
| 612 | + q->hash_kind = ctl->hash_kind; |
| 613 | + } |
| 614 | + |
| 615 | + if (q->perturb_period) { |
| 616 | + q->perturb_timer.expires = jiffies + q->perturb_period; |
| 617 | + add_timer(&q->perturb_timer); |
| 618 | + } |
| 619 | + } |
| 620 | + |
| 621 | + q->ht = kmalloc(q->hash_divisor*sizeof(esfq_index), GFP_KERNEL); |
| 622 | + if (!q->ht) |
| 623 | + goto err_case; |
| 624 | + |
| 625 | + q->dep = kmalloc((1+q->depth*2)*sizeof(struct esfq_head), GFP_KERNEL); |
| 626 | + if (!q->dep) |
| 627 | + goto err_case; |
| 628 | + q->next = kmalloc(q->depth*sizeof(esfq_index), GFP_KERNEL); |
| 629 | + if (!q->next) |
| 630 | + goto err_case; |
| 631 | + |
| 632 | + q->allot = kmalloc(q->depth*sizeof(short), GFP_KERNEL); |
| 633 | + if (!q->allot) |
| 634 | + goto err_case; |
| 635 | + q->hash = kmalloc(q->depth*sizeof(unsigned short), GFP_KERNEL); |
| 636 | + if (!q->hash) |
| 637 | + goto err_case; |
| 638 | + q->qs = kmalloc(q->depth*sizeof(struct sk_buff_head), GFP_KERNEL); |
| 639 | + if (!q->qs) |
| 640 | + goto err_case; |
| 641 | + |
| 642 | + for (i=0; i< q->hash_divisor; i++) |
| 643 | + q->ht[i] = q->depth; |
| 644 | + for (i=0; i<q->depth; i++) { |
| 645 | + skb_queue_head_init(&q->qs[i]); |
| 646 | + q->dep[i+q->depth].next = i+q->depth; |
| 647 | + q->dep[i+q->depth].prev = i+q->depth; |
| 648 | + } |
| 649 | + |
| 650 | + for (i=0; i<q->depth; i++) |
| 651 | + esfq_link(q, i); |
| 652 | + return 0; |
| 653 | +err_case: |
| 654 | + del_timer(&q->perturb_timer); |
| 655 | + if (q->ht) |
| 656 | + kfree(q->ht); |
| 657 | + if (q->dep) |
| 658 | + kfree(q->dep); |
| 659 | + if (q->next) |
| 660 | + kfree(q->next); |
| 661 | + if (q->allot) |
| 662 | + kfree(q->allot); |
| 663 | + if (q->hash) |
| 664 | + kfree(q->hash); |
| 665 | + if (q->qs) |
| 666 | + kfree(q->qs); |
| 667 | + return -ENOBUFS; |
| 668 | +} |
| 669 | + |
| 670 | +static void esfq_destroy(struct Qdisc *sch) |
| 671 | +{ |
| 672 | + struct esfq_sched_data *q = qdisc_priv(sch); |
| 673 | + del_timer(&q->perturb_timer); |
| 674 | + if(q->ht) |
| 675 | + kfree(q->ht); |
| 676 | + if(q->dep) |
| 677 | + kfree(q->dep); |
| 678 | + if(q->next) |
| 679 | + kfree(q->next); |
| 680 | + if(q->allot) |
| 681 | + kfree(q->allot); |
| 682 | + if(q->hash) |
| 683 | + kfree(q->hash); |
| 684 | + if(q->qs) |
| 685 | + kfree(q->qs); |
| 686 | +} |
| 687 | + |
| 688 | +static int esfq_dump(struct Qdisc *sch, struct sk_buff *skb) |
| 689 | +{ |
| 690 | + struct esfq_sched_data *q = qdisc_priv(sch); |
| 691 | + unsigned char *b = skb->tail; |
| 692 | + struct tc_esfq_qopt opt; |
| 693 | + |
| 694 | + opt.quantum = q->quantum; |
| 695 | + opt.perturb_period = q->perturb_period/HZ; |
| 696 | + |
| 697 | + opt.limit = q->limit; |
| 698 | + opt.divisor = q->hash_divisor; |
| 699 | + opt.flows = q->depth; |
| 700 | + opt.hash_kind = q->hash_kind; |
| 701 | + |
| 702 | + RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); |
| 703 | + |
| 704 | + return skb->len; |
| 705 | + |
| 706 | +rtattr_failure: |
| 707 | + skb_trim(skb, b - skb->data); |
| 708 | + return -1; |
| 709 | +} |
| 710 | + |
| 711 | +static struct Qdisc_ops esfq_qdisc_ops = |
| 712 | +{ |
| 713 | + .next = NULL, |
| 714 | + .cl_ops = NULL, |
| 715 | + .id = "esfq", |
| 716 | + .priv_size = sizeof(struct esfq_sched_data), |
| 717 | + .enqueue = esfq_enqueue, |
| 718 | + .dequeue = esfq_dequeue, |
| 719 | + .requeue = esfq_requeue, |
| 720 | + .drop = esfq_drop, |
| 721 | + .init = esfq_init, |
| 722 | + .reset = esfq_reset, |
| 723 | + .destroy = esfq_destroy, |
| 724 | + .change = NULL, /* esfq_change - needs more work */ |
| 725 | + .dump = esfq_dump, |
| 726 | +}; |
| 727 | + |
| 728 | +static int __init esfq_module_init(void) |
| 729 | +{ |
| 730 | + return register_qdisc(&esfq_qdisc_ops); |
| 731 | +} |
| 732 | +static void __exit esfq_module_exit(void) |
| 733 | +{ |
| 734 | + unregister_qdisc(&esfq_qdisc_ops); |
| 735 | +} |
| 736 | +module_init(esfq_module_init) |
| 737 | +module_exit(esfq_module_exit) |
| 738 | +MODULE_LICENSE("GPL"); |
| 739 | |