blob: f0463d757a98489f5b12138c0026d1b91742fd9d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * net/sched/sch_sfq.c Stochastic Fairness Queueing discipline.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 */
11
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/types.h>
14#include <linux/kernel.h>
15#include <linux/jiffies.h>
16#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/in.h>
18#include <linux/errno.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/init.h>
Patrick McHardy0ba48052007-07-02 22:49:07 -070020#include <linux/ipv6.h>
21#include <linux/skbuff.h>
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -070022#include <linux/jhash.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <net/ip.h>
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -070024#include <net/netlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <net/pkt_sched.h>
26
27
28/* Stochastic Fairness Queuing algorithm.
29 =======================================
30
31 Source:
32 Paul E. McKenney "Stochastic Fairness Queuing",
33 IEEE INFOCOMM'90 Proceedings, San Francisco, 1990.
34
35 Paul E. McKenney "Stochastic Fairness Queuing",
36 "Interworking: Research and Experience", v.2, 1991, p.113-131.
37
38
39 See also:
40 M. Shreedhar and George Varghese "Efficient Fair
41 Queuing using Deficit Round Robin", Proc. SIGCOMM 95.
42
43
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090044 This is not the thing that is usually called (W)FQ nowadays.
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 It does not use any timestamp mechanism, but instead
46 processes queues in round-robin order.
47
48 ADVANTAGE:
49
50 - It is very cheap. Both CPU and memory requirements are minimal.
51
52 DRAWBACKS:
53
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090054 - "Stochastic" -> It is not 100% fair.
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 When hash collisions occur, several flows are considered as one.
56
57 - "Round-robin" -> It introduces larger delays than virtual clock
58 based schemes, and should not be used for isolating interactive
59 traffic from non-interactive. It means, that this scheduler
60 should be used as leaf of CBQ or P3, which put interactive traffic
61 to higher priority band.
62
63 We still need true WFQ for top level CSZ, but using WFQ
64 for the best effort traffic is absolutely pointless:
65 SFQ is superior for this purpose.
66
67 IMPLEMENTATION:
68 This implementation limits maximal queue length to 128;
69 maximal mtu to 2^15-1; number of hash buckets to 1024.
70 The only goal of this restrictions was that all data
71 fit into one 4K page :-). Struct sfq_sched_data is
72 organized in anti-cache manner: all the data for a bucket
73 are scattered over different locations. This is not good,
74 but it allowed me to put it into 4K.
75
76 It is easy to increase these values, but not in flight. */
77
78#define SFQ_DEPTH 128
79#define SFQ_HASH_DIVISOR 1024
80
81/* This type should contain at least SFQ_DEPTH*2 values */
82typedef unsigned char sfq_index;
83
84struct sfq_head
85{
86 sfq_index next;
87 sfq_index prev;
88};
89
90struct sfq_sched_data
91{
92/* Parameters */
93 int perturb_period;
94 unsigned quantum; /* Allotment per round: MUST BE >= MTU */
95 int limit;
96
97/* Variables */
Patrick McHardy7d2681a2008-01-31 18:36:52 -080098 struct tcf_proto *filter_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 struct timer_list perturb_timer;
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700100 u32 perturbation;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 sfq_index tail; /* Index of current slot in round */
102 sfq_index max_depth; /* Maximal depth */
103
104 sfq_index ht[SFQ_HASH_DIVISOR]; /* Hash table */
105 sfq_index next[SFQ_DEPTH]; /* Active slots link */
106 short allot[SFQ_DEPTH]; /* Current allotment per slot */
107 unsigned short hash[SFQ_DEPTH]; /* Hash value indexed by slots */
108 struct sk_buff_head qs[SFQ_DEPTH]; /* Slot queue */
109 struct sfq_head dep[SFQ_DEPTH*2]; /* Linked list of slots, indexed by depth */
110};
111
112static __inline__ unsigned sfq_fold_hash(struct sfq_sched_data *q, u32 h, u32 h1)
113{
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700114 return jhash_2words(h, h1, q->perturbation) & (SFQ_HASH_DIVISOR - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115}
116
117static unsigned sfq_hash(struct sfq_sched_data *q, struct sk_buff *skb)
118{
119 u32 h, h2;
120
121 switch (skb->protocol) {
122 case __constant_htons(ETH_P_IP):
123 {
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700124 const struct iphdr *iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 h = iph->daddr;
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800126 h2 = iph->saddr ^ iph->protocol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
128 (iph->protocol == IPPROTO_TCP ||
129 iph->protocol == IPPROTO_UDP ||
Patrick McHardya8d0f952007-02-07 15:07:43 -0800130 iph->protocol == IPPROTO_UDPLITE ||
Patrick McHardyae82af52006-01-17 13:01:06 -0800131 iph->protocol == IPPROTO_SCTP ||
132 iph->protocol == IPPROTO_DCCP ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 iph->protocol == IPPROTO_ESP))
134 h2 ^= *(((u32*)iph) + iph->ihl);
135 break;
136 }
137 case __constant_htons(ETH_P_IPV6):
138 {
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700139 struct ipv6hdr *iph = ipv6_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 h = iph->daddr.s6_addr32[3];
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800141 h2 = iph->saddr.s6_addr32[3] ^ iph->nexthdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 if (iph->nexthdr == IPPROTO_TCP ||
143 iph->nexthdr == IPPROTO_UDP ||
Patrick McHardya8d0f952007-02-07 15:07:43 -0800144 iph->nexthdr == IPPROTO_UDPLITE ||
Patrick McHardyae82af52006-01-17 13:01:06 -0800145 iph->nexthdr == IPPROTO_SCTP ||
146 iph->nexthdr == IPPROTO_DCCP ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 iph->nexthdr == IPPROTO_ESP)
148 h2 ^= *(u32*)&iph[1];
149 break;
150 }
151 default:
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800152 h = (unsigned long)skb->dst ^ skb->protocol;
153 h2 = (unsigned long)skb->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800155
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 return sfq_fold_hash(q, h, h2);
157}
158
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800159static unsigned int sfq_classify(struct sk_buff *skb, struct Qdisc *sch,
160 int *qerr)
161{
162 struct sfq_sched_data *q = qdisc_priv(sch);
163 struct tcf_result res;
164 int result;
165
166 if (TC_H_MAJ(skb->priority) == sch->handle &&
167 TC_H_MIN(skb->priority) > 0 &&
168 TC_H_MIN(skb->priority) <= SFQ_HASH_DIVISOR)
169 return TC_H_MIN(skb->priority);
170
171 if (!q->filter_list)
172 return sfq_hash(q, skb) + 1;
173
174 *qerr = NET_XMIT_BYPASS;
175 result = tc_classify(skb, q->filter_list, &res);
176 if (result >= 0) {
177#ifdef CONFIG_NET_CLS_ACT
178 switch (result) {
179 case TC_ACT_STOLEN:
180 case TC_ACT_QUEUED:
181 *qerr = NET_XMIT_SUCCESS;
182 case TC_ACT_SHOT:
183 return 0;
184 }
185#endif
186 if (TC_H_MIN(res.classid) <= SFQ_HASH_DIVISOR)
187 return TC_H_MIN(res.classid);
188 }
189 return 0;
190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192static inline void sfq_link(struct sfq_sched_data *q, sfq_index x)
193{
194 sfq_index p, n;
195 int d = q->qs[x].qlen + SFQ_DEPTH;
196
197 p = d;
198 n = q->dep[d].next;
199 q->dep[x].next = n;
200 q->dep[x].prev = p;
201 q->dep[p].next = q->dep[n].prev = x;
202}
203
204static inline void sfq_dec(struct sfq_sched_data *q, sfq_index x)
205{
206 sfq_index p, n;
207
208 n = q->dep[x].next;
209 p = q->dep[x].prev;
210 q->dep[p].next = n;
211 q->dep[n].prev = p;
212
213 if (n == p && q->max_depth == q->qs[x].qlen + 1)
214 q->max_depth--;
215
216 sfq_link(q, x);
217}
218
219static inline void sfq_inc(struct sfq_sched_data *q, sfq_index x)
220{
221 sfq_index p, n;
222 int d;
223
224 n = q->dep[x].next;
225 p = q->dep[x].prev;
226 q->dep[p].next = n;
227 q->dep[n].prev = p;
228 d = q->qs[x].qlen;
229 if (q->max_depth < d)
230 q->max_depth = d;
231
232 sfq_link(q, x);
233}
234
235static unsigned int sfq_drop(struct Qdisc *sch)
236{
237 struct sfq_sched_data *q = qdisc_priv(sch);
238 sfq_index d = q->max_depth;
239 struct sk_buff *skb;
240 unsigned int len;
241
242 /* Queue is full! Find the longest slot and
243 drop a packet from it */
244
245 if (d > 1) {
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800246 sfq_index x = q->dep[d + SFQ_DEPTH].next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 skb = q->qs[x].prev;
248 len = skb->len;
249 __skb_unlink(skb, &q->qs[x]);
250 kfree_skb(skb);
251 sfq_dec(q, x);
252 sch->q.qlen--;
253 sch->qstats.drops++;
Patrick McHardyf5539eb2006-03-20 19:01:38 -0800254 sch->qstats.backlog -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 return len;
256 }
257
258 if (d == 1) {
259 /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
260 d = q->next[q->tail];
261 q->next[q->tail] = q->next[d];
262 q->allot[q->next[d]] += q->quantum;
263 skb = q->qs[d].prev;
264 len = skb->len;
265 __skb_unlink(skb, &q->qs[d]);
266 kfree_skb(skb);
267 sfq_dec(q, d);
268 sch->q.qlen--;
269 q->ht[q->hash[d]] = SFQ_DEPTH;
270 sch->qstats.drops++;
Patrick McHardyf5539eb2006-03-20 19:01:38 -0800271 sch->qstats.backlog -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 return len;
273 }
274
275 return 0;
276}
277
278static int
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800279sfq_enqueue(struct sk_buff *skb, struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
281 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800282 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 sfq_index x;
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800284 int ret;
285
286 hash = sfq_classify(skb, sch, &ret);
287 if (hash == 0) {
288 if (ret == NET_XMIT_BYPASS)
289 sch->qstats.drops++;
290 kfree_skb(skb);
291 return ret;
292 }
293 hash--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
295 x = q->ht[hash];
296 if (x == SFQ_DEPTH) {
297 q->ht[hash] = x = q->dep[SFQ_DEPTH].next;
298 q->hash[x] = hash;
299 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800300
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700301 /* If selected queue has length q->limit, this means that
302 * all another queues are empty and that we do simple tail drop,
303 * i.e. drop _this_ packet.
304 */
305 if (q->qs[x].qlen >= q->limit)
306 return qdisc_drop(skb, sch);
307
Patrick McHardyf5539eb2006-03-20 19:01:38 -0800308 sch->qstats.backlog += skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 __skb_queue_tail(&q->qs[x], skb);
310 sfq_inc(q, x);
311 if (q->qs[x].qlen == 1) { /* The flow is new */
312 if (q->tail == SFQ_DEPTH) { /* It is the first flow */
313 q->tail = x;
314 q->next[x] = x;
315 q->allot[x] = q->quantum;
316 } else {
317 q->next[x] = q->next[q->tail];
318 q->next[q->tail] = x;
319 q->tail = x;
320 }
321 }
Alexey Kuznetsov5588b402007-09-19 10:42:03 -0700322 if (++sch->q.qlen <= q->limit) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 sch->bstats.bytes += skb->len;
324 sch->bstats.packets++;
325 return 0;
326 }
327
328 sfq_drop(sch);
329 return NET_XMIT_CN;
330}
331
332static int
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800333sfq_requeue(struct sk_buff *skb, struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800336 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 sfq_index x;
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800338 int ret;
339
340 hash = sfq_classify(skb, sch, &ret);
341 if (hash == 0) {
342 if (ret == NET_XMIT_BYPASS)
343 sch->qstats.drops++;
344 kfree_skb(skb);
345 return ret;
346 }
347 hash--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 x = q->ht[hash];
350 if (x == SFQ_DEPTH) {
351 q->ht[hash] = x = q->dep[SFQ_DEPTH].next;
352 q->hash[x] = hash;
353 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800354
Patrick McHardyf5539eb2006-03-20 19:01:38 -0800355 sch->qstats.backlog += skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 __skb_queue_head(&q->qs[x], skb);
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700357 /* If selected queue has length q->limit+1, this means that
358 * all another queues are empty and we do simple tail drop.
359 * This packet is still requeued at head of queue, tail packet
360 * is dropped.
361 */
362 if (q->qs[x].qlen > q->limit) {
363 skb = q->qs[x].prev;
364 __skb_unlink(skb, &q->qs[x]);
365 sch->qstats.drops++;
366 sch->qstats.backlog -= skb->len;
367 kfree_skb(skb);
368 return NET_XMIT_CN;
369 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 sfq_inc(q, x);
372 if (q->qs[x].qlen == 1) { /* The flow is new */
373 if (q->tail == SFQ_DEPTH) { /* It is the first flow */
374 q->tail = x;
375 q->next[x] = x;
376 q->allot[x] = q->quantum;
377 } else {
378 q->next[x] = q->next[q->tail];
379 q->next[q->tail] = x;
380 q->tail = x;
381 }
382 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800383
Alexey Kuznetsov5588b402007-09-19 10:42:03 -0700384 if (++sch->q.qlen <= q->limit) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 sch->qstats.requeues++;
386 return 0;
387 }
388
389 sch->qstats.drops++;
390 sfq_drop(sch);
391 return NET_XMIT_CN;
392}
393
394
395
396
397static struct sk_buff *
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800398sfq_dequeue(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
400 struct sfq_sched_data *q = qdisc_priv(sch);
401 struct sk_buff *skb;
402 sfq_index a, old_a;
403
404 /* No active slots */
405 if (q->tail == SFQ_DEPTH)
406 return NULL;
407
408 a = old_a = q->next[q->tail];
409
410 /* Grab packet */
411 skb = __skb_dequeue(&q->qs[a]);
412 sfq_dec(q, a);
413 sch->q.qlen--;
Patrick McHardyf5539eb2006-03-20 19:01:38 -0800414 sch->qstats.backlog -= skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
416 /* Is the slot empty? */
417 if (q->qs[a].qlen == 0) {
418 q->ht[q->hash[a]] = SFQ_DEPTH;
419 a = q->next[a];
420 if (a == old_a) {
421 q->tail = SFQ_DEPTH;
422 return skb;
423 }
424 q->next[q->tail] = a;
425 q->allot[a] += q->quantum;
426 } else if ((q->allot[a] -= skb->len) <= 0) {
427 q->tail = a;
428 a = q->next[a];
429 q->allot[a] += q->quantum;
430 }
431 return skb;
432}
433
434static void
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800435sfq_reset(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
437 struct sk_buff *skb;
438
439 while ((skb = sfq_dequeue(sch)) != NULL)
440 kfree_skb(skb);
441}
442
443static void sfq_perturbation(unsigned long arg)
444{
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800445 struct Qdisc *sch = (struct Qdisc *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 struct sfq_sched_data *q = qdisc_priv(sch);
447
Stephen Hemmingerd46f8dd2008-01-20 17:19:43 -0800448 q->perturbation = net_random();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700450 if (q->perturb_period)
451 mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452}
453
Patrick McHardy1e904742008-01-22 22:11:17 -0800454static int sfq_change(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
456 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800457 struct tc_sfq_qopt *ctl = nla_data(opt);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800458 unsigned int qlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Patrick McHardy1e904742008-01-22 22:11:17 -0800460 if (opt->nla_len < nla_attr_size(sizeof(*ctl)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 return -EINVAL;
462
463 sch_tree_lock(sch);
464 q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800465 q->perturb_period = ctl->perturb_period * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 if (ctl->limit)
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700467 q->limit = min_t(u32, ctl->limit, SFQ_DEPTH - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Patrick McHardy5e50da02006-11-29 17:36:20 -0800469 qlen = sch->q.qlen;
Alexey Kuznetsov5588b402007-09-19 10:42:03 -0700470 while (sch->q.qlen > q->limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 sfq_drop(sch);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800472 qdisc_tree_decrease_qlen(sch, qlen - sch->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474 del_timer(&q->perturb_timer);
475 if (q->perturb_period) {
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700476 mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
Stephen Hemmingerd46f8dd2008-01-20 17:19:43 -0800477 q->perturbation = net_random();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
479 sch_tree_unlock(sch);
480 return 0;
481}
482
Patrick McHardy1e904742008-01-22 22:11:17 -0800483static int sfq_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
485 struct sfq_sched_data *q = qdisc_priv(sch);
486 int i;
487
Stephen Hemmingerd3e99482008-01-20 17:18:45 -0800488 q->perturb_timer.function = sfq_perturbation;
489 q->perturb_timer.data = (unsigned long)sch;;
490 init_timer_deferrable(&q->perturb_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800492 for (i = 0; i < SFQ_HASH_DIVISOR; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 q->ht[i] = SFQ_DEPTH;
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800494
495 for (i = 0; i < SFQ_DEPTH; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 skb_queue_head_init(&q->qs[i]);
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800497 q->dep[i + SFQ_DEPTH].next = i + SFQ_DEPTH;
498 q->dep[i + SFQ_DEPTH].prev = i + SFQ_DEPTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800500
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700501 q->limit = SFQ_DEPTH - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 q->max_depth = 0;
503 q->tail = SFQ_DEPTH;
504 if (opt == NULL) {
505 q->quantum = psched_mtu(sch->dev);
506 q->perturb_period = 0;
Stephen Hemmingerd46f8dd2008-01-20 17:19:43 -0800507 q->perturbation = net_random();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 } else {
509 int err = sfq_change(sch, opt);
510 if (err)
511 return err;
512 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800513
514 for (i = 0; i < SFQ_DEPTH; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 sfq_link(q, i);
516 return 0;
517}
518
519static void sfq_destroy(struct Qdisc *sch)
520{
521 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800522
523 tcf_destroy_chain(q->filter_list);
Jarek Poplawski980c4782008-04-29 03:29:03 -0700524 q->perturb_period = 0;
525 del_timer_sync(&q->perturb_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526}
527
528static int sfq_dump(struct Qdisc *sch, struct sk_buff *skb)
529{
530 struct sfq_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700531 unsigned char *b = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 struct tc_sfq_qopt opt;
533
534 opt.quantum = q->quantum;
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800535 opt.perturb_period = q->perturb_period / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
537 opt.limit = q->limit;
538 opt.divisor = SFQ_HASH_DIVISOR;
539 opt.flows = q->limit;
540
Patrick McHardy1e904742008-01-22 22:11:17 -0800541 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
543 return skb->len;
544
Patrick McHardy1e904742008-01-22 22:11:17 -0800545nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700546 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 return -1;
548}
549
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800550static int sfq_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
551 struct nlattr **tca, unsigned long *arg)
552{
553 return -EOPNOTSUPP;
554}
555
556static unsigned long sfq_get(struct Qdisc *sch, u32 classid)
557{
558 return 0;
559}
560
561static struct tcf_proto **sfq_find_tcf(struct Qdisc *sch, unsigned long cl)
562{
563 struct sfq_sched_data *q = qdisc_priv(sch);
564
565 if (cl)
566 return NULL;
567 return &q->filter_list;
568}
569
Patrick McHardy94de78d2008-01-31 18:37:16 -0800570static int sfq_dump_class(struct Qdisc *sch, unsigned long cl,
571 struct sk_buff *skb, struct tcmsg *tcm)
572{
573 tcm->tcm_handle |= TC_H_MIN(cl);
574 return 0;
575}
576
577static int sfq_dump_class_stats(struct Qdisc *sch, unsigned long cl,
578 struct gnet_dump *d)
579{
580 struct sfq_sched_data *q = qdisc_priv(sch);
581 sfq_index idx = q->ht[cl-1];
582 struct gnet_stats_queue qs = { .qlen = q->qs[idx].qlen };
583 struct tc_sfq_xstats xstats = { .allot = q->allot[idx] };
584
585 if (gnet_stats_copy_queue(d, &qs) < 0)
586 return -1;
587 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
588}
589
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800590static void sfq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
591{
Patrick McHardy94de78d2008-01-31 18:37:16 -0800592 struct sfq_sched_data *q = qdisc_priv(sch);
593 unsigned int i;
594
595 if (arg->stop)
596 return;
597
598 for (i = 0; i < SFQ_HASH_DIVISOR; i++) {
599 if (q->ht[i] == SFQ_DEPTH ||
600 arg->count < arg->skip) {
601 arg->count++;
602 continue;
603 }
604 if (arg->fn(sch, i + 1, arg) < 0) {
605 arg->stop = 1;
606 break;
607 }
608 arg->count++;
609 }
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800610}
611
612static const struct Qdisc_class_ops sfq_class_ops = {
613 .get = sfq_get,
614 .change = sfq_change_class,
615 .tcf_chain = sfq_find_tcf,
Patrick McHardy94de78d2008-01-31 18:37:16 -0800616 .dump = sfq_dump_class,
617 .dump_stats = sfq_dump_class_stats,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800618 .walk = sfq_walk,
619};
620
Eric Dumazet20fea082007-11-14 01:44:41 -0800621static struct Qdisc_ops sfq_qdisc_ops __read_mostly = {
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800622 .cl_ops = &sfq_class_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 .id = "sfq",
624 .priv_size = sizeof(struct sfq_sched_data),
625 .enqueue = sfq_enqueue,
626 .dequeue = sfq_dequeue,
627 .requeue = sfq_requeue,
628 .drop = sfq_drop,
629 .init = sfq_init,
630 .reset = sfq_reset,
631 .destroy = sfq_destroy,
632 .change = NULL,
633 .dump = sfq_dump,
634 .owner = THIS_MODULE,
635};
636
637static int __init sfq_module_init(void)
638{
639 return register_qdisc(&sfq_qdisc_ops);
640}
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900641static void __exit sfq_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
643 unregister_qdisc(&sfq_qdisc_ops);
644}
645module_init(sfq_module_init)
646module_exit(sfq_module_exit)
647MODULE_LICENSE("GPL");