blob: cb21380c0605fe813b4a4ff0f489d9fdc08c2428 [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) {
Arnaldo Carvalho de Melo60678042008-09-20 22:20:49 -0700122 case htons(ETH_P_IP):
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 {
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 }
Arnaldo Carvalho de Melo60678042008-09-20 22:20:49 -0700137 case htons(ETH_P_IPV6):
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 {
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:
Eric Dumazetadf30902009-06-02 05:19:30 +0000152 h = (unsigned long)skb_dst(skb) ^ skb->protocol;
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800153 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
Jarek Poplawskic27f3392008-08-04 22:39:11 -0700174 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800175 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:
Jarek Poplawski378a2f02008-08-04 22:31:03 -0700181 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800182 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;
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700248 len = qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 __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;
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700264 len = qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 __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;
Jarek Poplawski7f3ff4f2008-12-21 20:14:48 -0800284 int uninitialized_var(ret);
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800285
286 hash = sfq_classify(skb, sch, &ret);
287 if (hash == 0) {
Jarek Poplawskic27f3392008-08-04 22:39:11 -0700288 if (ret & __NET_XMIT_BYPASS)
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800289 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
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700308 sch->qstats.backlog += qdisc_pkt_len(skb);
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) {
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700323 sch->bstats.bytes += qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 sch->bstats.packets++;
325 return 0;
326 }
327
328 sfq_drop(sch);
329 return NET_XMIT_CN;
330}
331
Patrick McHardy48a8f512008-10-31 00:44:18 -0700332static struct sk_buff *
333sfq_peek(struct Qdisc *sch)
334{
335 struct sfq_sched_data *q = qdisc_priv(sch);
336 sfq_index a;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Patrick McHardy48a8f512008-10-31 00:44:18 -0700338 /* No active slots */
339 if (q->tail == SFQ_DEPTH)
340 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
Patrick McHardy48a8f512008-10-31 00:44:18 -0700342 a = q->next[q->tail];
343 return skb_peek(&q->qs[a]);
344}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
346static struct sk_buff *
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800347sfq_dequeue(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
349 struct sfq_sched_data *q = qdisc_priv(sch);
350 struct sk_buff *skb;
351 sfq_index a, old_a;
352
353 /* No active slots */
354 if (q->tail == SFQ_DEPTH)
355 return NULL;
356
357 a = old_a = q->next[q->tail];
358
359 /* Grab packet */
360 skb = __skb_dequeue(&q->qs[a]);
361 sfq_dec(q, a);
362 sch->q.qlen--;
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700363 sch->qstats.backlog -= qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
365 /* Is the slot empty? */
366 if (q->qs[a].qlen == 0) {
367 q->ht[q->hash[a]] = SFQ_DEPTH;
368 a = q->next[a];
369 if (a == old_a) {
370 q->tail = SFQ_DEPTH;
371 return skb;
372 }
373 q->next[q->tail] = a;
374 q->allot[a] += q->quantum;
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700375 } else if ((q->allot[a] -= qdisc_pkt_len(skb)) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 q->tail = a;
377 a = q->next[a];
378 q->allot[a] += q->quantum;
379 }
380 return skb;
381}
382
383static void
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800384sfq_reset(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385{
386 struct sk_buff *skb;
387
388 while ((skb = sfq_dequeue(sch)) != NULL)
389 kfree_skb(skb);
390}
391
392static void sfq_perturbation(unsigned long arg)
393{
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800394 struct Qdisc *sch = (struct Qdisc *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 struct sfq_sched_data *q = qdisc_priv(sch);
396
Stephen Hemmingerd46f8dd2008-01-20 17:19:43 -0800397 q->perturbation = net_random();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700399 if (q->perturb_period)
400 mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401}
402
Patrick McHardy1e904742008-01-22 22:11:17 -0800403static int sfq_change(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
405 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800406 struct tc_sfq_qopt *ctl = nla_data(opt);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800407 unsigned int qlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Patrick McHardy1e904742008-01-22 22:11:17 -0800409 if (opt->nla_len < nla_attr_size(sizeof(*ctl)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 return -EINVAL;
411
412 sch_tree_lock(sch);
David S. Miller5ce2d482008-07-08 17:06:30 -0700413 q->quantum = ctl->quantum ? : psched_mtu(qdisc_dev(sch));
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800414 q->perturb_period = ctl->perturb_period * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 if (ctl->limit)
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700416 q->limit = min_t(u32, ctl->limit, SFQ_DEPTH - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
Patrick McHardy5e50da02006-11-29 17:36:20 -0800418 qlen = sch->q.qlen;
Alexey Kuznetsov5588b402007-09-19 10:42:03 -0700419 while (sch->q.qlen > q->limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 sfq_drop(sch);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800421 qdisc_tree_decrease_qlen(sch, qlen - sch->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
423 del_timer(&q->perturb_timer);
424 if (q->perturb_period) {
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700425 mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
Stephen Hemmingerd46f8dd2008-01-20 17:19:43 -0800426 q->perturbation = net_random();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 }
428 sch_tree_unlock(sch);
429 return 0;
430}
431
Patrick McHardy1e904742008-01-22 22:11:17 -0800432static int sfq_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
434 struct sfq_sched_data *q = qdisc_priv(sch);
435 int i;
436
Stephen Hemmingerd3e99482008-01-20 17:18:45 -0800437 q->perturb_timer.function = sfq_perturbation;
Fernando Carrijoc19a28e2009-01-07 18:09:08 -0800438 q->perturb_timer.data = (unsigned long)sch;
Stephen Hemmingerd3e99482008-01-20 17:18:45 -0800439 init_timer_deferrable(&q->perturb_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800441 for (i = 0; i < SFQ_HASH_DIVISOR; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 q->ht[i] = SFQ_DEPTH;
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800443
444 for (i = 0; i < SFQ_DEPTH; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 skb_queue_head_init(&q->qs[i]);
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800446 q->dep[i + SFQ_DEPTH].next = i + SFQ_DEPTH;
447 q->dep[i + SFQ_DEPTH].prev = i + SFQ_DEPTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800449
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700450 q->limit = SFQ_DEPTH - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 q->max_depth = 0;
452 q->tail = SFQ_DEPTH;
453 if (opt == NULL) {
David S. Miller5ce2d482008-07-08 17:06:30 -0700454 q->quantum = psched_mtu(qdisc_dev(sch));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 q->perturb_period = 0;
Stephen Hemmingerd46f8dd2008-01-20 17:19:43 -0800456 q->perturbation = net_random();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 } else {
458 int err = sfq_change(sch, opt);
459 if (err)
460 return err;
461 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800462
463 for (i = 0; i < SFQ_DEPTH; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 sfq_link(q, i);
465 return 0;
466}
467
468static void sfq_destroy(struct Qdisc *sch)
469{
470 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800471
Patrick McHardyff31ab52008-07-01 19:52:38 -0700472 tcf_destroy_chain(&q->filter_list);
Jarek Poplawski980c4782008-04-29 03:29:03 -0700473 q->perturb_period = 0;
474 del_timer_sync(&q->perturb_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
477static int sfq_dump(struct Qdisc *sch, struct sk_buff *skb)
478{
479 struct sfq_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700480 unsigned char *b = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 struct tc_sfq_qopt opt;
482
483 opt.quantum = q->quantum;
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800484 opt.perturb_period = q->perturb_period / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
486 opt.limit = q->limit;
487 opt.divisor = SFQ_HASH_DIVISOR;
David S. Millercdec7e52008-07-26 02:28:09 -0700488 opt.flows = q->limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Patrick McHardy1e904742008-01-22 22:11:17 -0800490 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
492 return skb->len;
493
Patrick McHardy1e904742008-01-22 22:11:17 -0800494nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700495 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 return -1;
497}
498
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800499static unsigned long sfq_get(struct Qdisc *sch, u32 classid)
500{
501 return 0;
502}
503
504static struct tcf_proto **sfq_find_tcf(struct Qdisc *sch, unsigned long cl)
505{
506 struct sfq_sched_data *q = qdisc_priv(sch);
507
508 if (cl)
509 return NULL;
510 return &q->filter_list;
511}
512
Patrick McHardy94de78d2008-01-31 18:37:16 -0800513static int sfq_dump_class(struct Qdisc *sch, unsigned long cl,
514 struct sk_buff *skb, struct tcmsg *tcm)
515{
516 tcm->tcm_handle |= TC_H_MIN(cl);
517 return 0;
518}
519
520static int sfq_dump_class_stats(struct Qdisc *sch, unsigned long cl,
521 struct gnet_dump *d)
522{
523 struct sfq_sched_data *q = qdisc_priv(sch);
524 sfq_index idx = q->ht[cl-1];
525 struct gnet_stats_queue qs = { .qlen = q->qs[idx].qlen };
526 struct tc_sfq_xstats xstats = { .allot = q->allot[idx] };
527
528 if (gnet_stats_copy_queue(d, &qs) < 0)
529 return -1;
530 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
531}
532
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800533static void sfq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
534{
Patrick McHardy94de78d2008-01-31 18:37:16 -0800535 struct sfq_sched_data *q = qdisc_priv(sch);
536 unsigned int i;
537
538 if (arg->stop)
539 return;
540
541 for (i = 0; i < SFQ_HASH_DIVISOR; i++) {
542 if (q->ht[i] == SFQ_DEPTH ||
543 arg->count < arg->skip) {
544 arg->count++;
545 continue;
546 }
547 if (arg->fn(sch, i + 1, arg) < 0) {
548 arg->stop = 1;
549 break;
550 }
551 arg->count++;
552 }
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800553}
554
555static const struct Qdisc_class_ops sfq_class_ops = {
556 .get = sfq_get,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800557 .tcf_chain = sfq_find_tcf,
Patrick McHardy94de78d2008-01-31 18:37:16 -0800558 .dump = sfq_dump_class,
559 .dump_stats = sfq_dump_class_stats,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800560 .walk = sfq_walk,
561};
562
Eric Dumazet20fea082007-11-14 01:44:41 -0800563static struct Qdisc_ops sfq_qdisc_ops __read_mostly = {
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800564 .cl_ops = &sfq_class_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 .id = "sfq",
566 .priv_size = sizeof(struct sfq_sched_data),
567 .enqueue = sfq_enqueue,
568 .dequeue = sfq_dequeue,
Patrick McHardy48a8f512008-10-31 00:44:18 -0700569 .peek = sfq_peek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 .drop = sfq_drop,
571 .init = sfq_init,
572 .reset = sfq_reset,
573 .destroy = sfq_destroy,
574 .change = NULL,
575 .dump = sfq_dump,
576 .owner = THIS_MODULE,
577};
578
579static int __init sfq_module_init(void)
580{
581 return register_qdisc(&sfq_qdisc_ops);
582}
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900583static void __exit sfq_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584{
585 unregister_qdisc(&sfq_qdisc_ops);
586}
587module_init(sfq_module_init)
588module_exit(sfq_module_exit)
589MODULE_LICENSE("GPL");