Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | ||
1da177e4 | 12 | #include <linux/module.h> |
1da177e4 LT |
13 | #include <linux/types.h> |
14 | #include <linux/kernel.h> | |
15 | #include <linux/jiffies.h> | |
16 | #include <linux/string.h> | |
1da177e4 LT |
17 | #include <linux/in.h> |
18 | #include <linux/errno.h> | |
1da177e4 | 19 | #include <linux/init.h> |
1da177e4 | 20 | #include <linux/ipv6.h> |
1da177e4 | 21 | #include <linux/skbuff.h> |
32740ddc | 22 | #include <linux/jhash.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
0ba48053 PM |
24 | #include <net/ip.h> |
25 | #include <net/netlink.h> | |
1da177e4 LT |
26 | #include <net/pkt_sched.h> |
27 | ||
28 | ||
29 | /* Stochastic Fairness Queuing algorithm. | |
30 | ======================================= | |
31 | ||
32 | Source: | |
33 | Paul E. McKenney "Stochastic Fairness Queuing", | |
34 | IEEE INFOCOMM'90 Proceedings, San Francisco, 1990. | |
35 | ||
36 | Paul E. McKenney "Stochastic Fairness Queuing", | |
37 | "Interworking: Research and Experience", v.2, 1991, p.113-131. | |
38 | ||
39 | ||
40 | See also: | |
41 | M. Shreedhar and George Varghese "Efficient Fair | |
42 | Queuing using Deficit Round Robin", Proc. SIGCOMM 95. | |
43 | ||
44 | ||
10297b99 | 45 | This is not the thing that is usually called (W)FQ nowadays. |
1da177e4 LT |
46 | It does not use any timestamp mechanism, but instead |
47 | processes queues in round-robin order. | |
48 | ||
49 | ADVANTAGE: | |
50 | ||
51 | - It is very cheap. Both CPU and memory requirements are minimal. | |
52 | ||
53 | DRAWBACKS: | |
54 | ||
10297b99 | 55 | - "Stochastic" -> It is not 100% fair. |
1da177e4 LT |
56 | When hash collisions occur, several flows are considered as one. |
57 | ||
58 | - "Round-robin" -> It introduces larger delays than virtual clock | |
59 | based schemes, and should not be used for isolating interactive | |
60 | traffic from non-interactive. It means, that this scheduler | |
61 | should be used as leaf of CBQ or P3, which put interactive traffic | |
62 | to higher priority band. | |
63 | ||
64 | We still need true WFQ for top level CSZ, but using WFQ | |
65 | for the best effort traffic is absolutely pointless: | |
66 | SFQ is superior for this purpose. | |
67 | ||
68 | IMPLEMENTATION: | |
69 | This implementation limits maximal queue length to 128; | |
70 | maximal mtu to 2^15-1; number of hash buckets to 1024. | |
71 | The only goal of this restrictions was that all data | |
72 | fit into one 4K page :-). Struct sfq_sched_data is | |
73 | organized in anti-cache manner: all the data for a bucket | |
74 | are scattered over different locations. This is not good, | |
75 | but it allowed me to put it into 4K. | |
76 | ||
77 | It is easy to increase these values, but not in flight. */ | |
78 | ||
79 | #define SFQ_DEPTH 128 | |
80 | #define SFQ_HASH_DIVISOR 1024 | |
81 | ||
82 | /* This type should contain at least SFQ_DEPTH*2 values */ | |
83 | typedef unsigned char sfq_index; | |
84 | ||
85 | struct sfq_head | |
86 | { | |
87 | sfq_index next; | |
88 | sfq_index prev; | |
89 | }; | |
90 | ||
91 | struct sfq_sched_data | |
92 | { | |
93 | /* Parameters */ | |
94 | int perturb_period; | |
95 | unsigned quantum; /* Allotment per round: MUST BE >= MTU */ | |
96 | int limit; | |
97 | ||
98 | /* Variables */ | |
7d2681a6 | 99 | struct tcf_proto *filter_list; |
1da177e4 | 100 | struct timer_list perturb_timer; |
32740ddc | 101 | u32 perturbation; |
1da177e4 LT |
102 | sfq_index tail; /* Index of current slot in round */ |
103 | sfq_index max_depth; /* Maximal depth */ | |
104 | ||
105 | sfq_index ht[SFQ_HASH_DIVISOR]; /* Hash table */ | |
106 | sfq_index next[SFQ_DEPTH]; /* Active slots link */ | |
107 | short allot[SFQ_DEPTH]; /* Current allotment per slot */ | |
108 | unsigned short hash[SFQ_DEPTH]; /* Hash value indexed by slots */ | |
109 | struct sk_buff_head qs[SFQ_DEPTH]; /* Slot queue */ | |
110 | struct sfq_head dep[SFQ_DEPTH*2]; /* Linked list of slots, indexed by depth */ | |
111 | }; | |
112 | ||
113 | static __inline__ unsigned sfq_fold_hash(struct sfq_sched_data *q, u32 h, u32 h1) | |
114 | { | |
32740ddc | 115 | return jhash_2words(h, h1, q->perturbation) & (SFQ_HASH_DIVISOR - 1); |
1da177e4 LT |
116 | } |
117 | ||
118 | static unsigned sfq_hash(struct sfq_sched_data *q, struct sk_buff *skb) | |
119 | { | |
120 | u32 h, h2; | |
121 | ||
122 | switch (skb->protocol) { | |
60678040 | 123 | case htons(ETH_P_IP): |
1da177e4 | 124 | { |
f2f00981 | 125 | const struct iphdr *iph; |
b9959c2e | 126 | int poff; |
f2f00981 CG |
127 | |
128 | if (!pskb_network_may_pull(skb, sizeof(*iph))) | |
129 | goto err; | |
130 | iph = ip_hdr(skb); | |
0eae88f3 ED |
131 | h = (__force u32)iph->daddr; |
132 | h2 = (__force u32)iph->saddr ^ iph->protocol; | |
b9959c2e CG |
133 | if (iph->frag_off & htons(IP_MF|IP_OFFSET)) |
134 | break; | |
135 | poff = proto_ports_offset(iph->protocol); | |
136 | if (poff >= 0 && | |
137 | pskb_network_may_pull(skb, iph->ihl * 4 + 4 + poff)) { | |
138 | iph = ip_hdr(skb); | |
139 | h2 ^= *(u32*)((void *)iph + iph->ihl * 4 + poff); | |
140 | } | |
1da177e4 LT |
141 | break; |
142 | } | |
60678040 | 143 | case htons(ETH_P_IPV6): |
1da177e4 | 144 | { |
f2f00981 | 145 | struct ipv6hdr *iph; |
b9959c2e | 146 | int poff; |
f2f00981 CG |
147 | |
148 | if (!pskb_network_may_pull(skb, sizeof(*iph))) | |
149 | goto err; | |
150 | iph = ipv6_hdr(skb); | |
0eae88f3 ED |
151 | h = (__force u32)iph->daddr.s6_addr32[3]; |
152 | h2 = (__force u32)iph->saddr.s6_addr32[3] ^ iph->nexthdr; | |
b9959c2e CG |
153 | poff = proto_ports_offset(iph->nexthdr); |
154 | if (poff >= 0 && | |
155 | pskb_network_may_pull(skb, sizeof(*iph) + 4 + poff)) { | |
156 | iph = ipv6_hdr(skb); | |
157 | h2 ^= *(u32*)((void *)iph + sizeof(*iph) + poff); | |
158 | } | |
1da177e4 LT |
159 | break; |
160 | } | |
161 | default: | |
f2f00981 | 162 | err: |
0eae88f3 | 163 | h = (unsigned long)skb_dst(skb) ^ (__force u32)skb->protocol; |
6f9e98f7 | 164 | h2 = (unsigned long)skb->sk; |
1da177e4 | 165 | } |
6f9e98f7 | 166 | |
1da177e4 LT |
167 | return sfq_fold_hash(q, h, h2); |
168 | } | |
169 | ||
7d2681a6 PM |
170 | static unsigned int sfq_classify(struct sk_buff *skb, struct Qdisc *sch, |
171 | int *qerr) | |
172 | { | |
173 | struct sfq_sched_data *q = qdisc_priv(sch); | |
174 | struct tcf_result res; | |
175 | int result; | |
176 | ||
177 | if (TC_H_MAJ(skb->priority) == sch->handle && | |
178 | TC_H_MIN(skb->priority) > 0 && | |
179 | TC_H_MIN(skb->priority) <= SFQ_HASH_DIVISOR) | |
180 | return TC_H_MIN(skb->priority); | |
181 | ||
182 | if (!q->filter_list) | |
183 | return sfq_hash(q, skb) + 1; | |
184 | ||
c27f339a | 185 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
7d2681a6 PM |
186 | result = tc_classify(skb, q->filter_list, &res); |
187 | if (result >= 0) { | |
188 | #ifdef CONFIG_NET_CLS_ACT | |
189 | switch (result) { | |
190 | case TC_ACT_STOLEN: | |
191 | case TC_ACT_QUEUED: | |
378a2f09 | 192 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
7d2681a6 PM |
193 | case TC_ACT_SHOT: |
194 | return 0; | |
195 | } | |
196 | #endif | |
197 | if (TC_H_MIN(res.classid) <= SFQ_HASH_DIVISOR) | |
198 | return TC_H_MIN(res.classid); | |
199 | } | |
200 | return 0; | |
201 | } | |
202 | ||
1da177e4 LT |
203 | static inline void sfq_link(struct sfq_sched_data *q, sfq_index x) |
204 | { | |
205 | sfq_index p, n; | |
206 | int d = q->qs[x].qlen + SFQ_DEPTH; | |
207 | ||
208 | p = d; | |
209 | n = q->dep[d].next; | |
210 | q->dep[x].next = n; | |
211 | q->dep[x].prev = p; | |
212 | q->dep[p].next = q->dep[n].prev = x; | |
213 | } | |
214 | ||
215 | static inline void sfq_dec(struct sfq_sched_data *q, sfq_index x) | |
216 | { | |
217 | sfq_index p, n; | |
218 | ||
219 | n = q->dep[x].next; | |
220 | p = q->dep[x].prev; | |
221 | q->dep[p].next = n; | |
222 | q->dep[n].prev = p; | |
223 | ||
224 | if (n == p && q->max_depth == q->qs[x].qlen + 1) | |
225 | q->max_depth--; | |
226 | ||
227 | sfq_link(q, x); | |
228 | } | |
229 | ||
230 | static inline void sfq_inc(struct sfq_sched_data *q, sfq_index x) | |
231 | { | |
232 | sfq_index p, n; | |
233 | int d; | |
234 | ||
235 | n = q->dep[x].next; | |
236 | p = q->dep[x].prev; | |
237 | q->dep[p].next = n; | |
238 | q->dep[n].prev = p; | |
239 | d = q->qs[x].qlen; | |
240 | if (q->max_depth < d) | |
241 | q->max_depth = d; | |
242 | ||
243 | sfq_link(q, x); | |
244 | } | |
245 | ||
246 | static unsigned int sfq_drop(struct Qdisc *sch) | |
247 | { | |
248 | struct sfq_sched_data *q = qdisc_priv(sch); | |
249 | sfq_index d = q->max_depth; | |
250 | struct sk_buff *skb; | |
251 | unsigned int len; | |
252 | ||
253 | /* Queue is full! Find the longest slot and | |
254 | drop a packet from it */ | |
255 | ||
256 | if (d > 1) { | |
6f9e98f7 | 257 | sfq_index x = q->dep[d + SFQ_DEPTH].next; |
1da177e4 | 258 | skb = q->qs[x].prev; |
0abf77e5 | 259 | len = qdisc_pkt_len(skb); |
1da177e4 LT |
260 | __skb_unlink(skb, &q->qs[x]); |
261 | kfree_skb(skb); | |
262 | sfq_dec(q, x); | |
263 | sch->q.qlen--; | |
264 | sch->qstats.drops++; | |
f5539eb8 | 265 | sch->qstats.backlog -= len; |
1da177e4 LT |
266 | return len; |
267 | } | |
268 | ||
269 | if (d == 1) { | |
270 | /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */ | |
271 | d = q->next[q->tail]; | |
272 | q->next[q->tail] = q->next[d]; | |
1da177e4 | 273 | skb = q->qs[d].prev; |
0abf77e5 | 274 | len = qdisc_pkt_len(skb); |
1da177e4 LT |
275 | __skb_unlink(skb, &q->qs[d]); |
276 | kfree_skb(skb); | |
277 | sfq_dec(q, d); | |
278 | sch->q.qlen--; | |
279 | q->ht[q->hash[d]] = SFQ_DEPTH; | |
280 | sch->qstats.drops++; | |
f5539eb8 | 281 | sch->qstats.backlog -= len; |
1da177e4 LT |
282 | return len; |
283 | } | |
284 | ||
285 | return 0; | |
286 | } | |
287 | ||
288 | static int | |
6f9e98f7 | 289 | sfq_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
1da177e4 LT |
290 | { |
291 | struct sfq_sched_data *q = qdisc_priv(sch); | |
7d2681a6 | 292 | unsigned int hash; |
1da177e4 | 293 | sfq_index x; |
7f3ff4f6 | 294 | int uninitialized_var(ret); |
7d2681a6 PM |
295 | |
296 | hash = sfq_classify(skb, sch, &ret); | |
297 | if (hash == 0) { | |
c27f339a | 298 | if (ret & __NET_XMIT_BYPASS) |
7d2681a6 PM |
299 | sch->qstats.drops++; |
300 | kfree_skb(skb); | |
301 | return ret; | |
302 | } | |
303 | hash--; | |
1da177e4 LT |
304 | |
305 | x = q->ht[hash]; | |
306 | if (x == SFQ_DEPTH) { | |
307 | q->ht[hash] = x = q->dep[SFQ_DEPTH].next; | |
308 | q->hash[x] = hash; | |
309 | } | |
6f9e98f7 | 310 | |
32740ddc AK |
311 | /* If selected queue has length q->limit, this means that |
312 | * all another queues are empty and that we do simple tail drop, | |
313 | * i.e. drop _this_ packet. | |
314 | */ | |
315 | if (q->qs[x].qlen >= q->limit) | |
316 | return qdisc_drop(skb, sch); | |
317 | ||
0abf77e5 | 318 | sch->qstats.backlog += qdisc_pkt_len(skb); |
1da177e4 LT |
319 | __skb_queue_tail(&q->qs[x], skb); |
320 | sfq_inc(q, x); | |
321 | if (q->qs[x].qlen == 1) { /* The flow is new */ | |
322 | if (q->tail == SFQ_DEPTH) { /* It is the first flow */ | |
1da177e4 | 323 | q->next[x] = x; |
1da177e4 LT |
324 | } else { |
325 | q->next[x] = q->next[q->tail]; | |
326 | q->next[q->tail] = x; | |
1da177e4 | 327 | } |
aa3e2199 ED |
328 | q->tail = x; |
329 | q->allot[x] = q->quantum; | |
1da177e4 | 330 | } |
5588b40d | 331 | if (++sch->q.qlen <= q->limit) { |
0abf77e5 | 332 | sch->bstats.bytes += qdisc_pkt_len(skb); |
1da177e4 | 333 | sch->bstats.packets++; |
9871e50e | 334 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
335 | } |
336 | ||
337 | sfq_drop(sch); | |
338 | return NET_XMIT_CN; | |
339 | } | |
340 | ||
48a8f519 PM |
341 | static struct sk_buff * |
342 | sfq_peek(struct Qdisc *sch) | |
343 | { | |
344 | struct sfq_sched_data *q = qdisc_priv(sch); | |
345 | sfq_index a; | |
1da177e4 | 346 | |
48a8f519 PM |
347 | /* No active slots */ |
348 | if (q->tail == SFQ_DEPTH) | |
349 | return NULL; | |
1da177e4 | 350 | |
48a8f519 PM |
351 | a = q->next[q->tail]; |
352 | return skb_peek(&q->qs[a]); | |
353 | } | |
1da177e4 LT |
354 | |
355 | static struct sk_buff * | |
6f9e98f7 | 356 | sfq_dequeue(struct Qdisc *sch) |
1da177e4 LT |
357 | { |
358 | struct sfq_sched_data *q = qdisc_priv(sch); | |
359 | struct sk_buff *skb; | |
aa3e2199 | 360 | sfq_index a, next_a; |
1da177e4 LT |
361 | |
362 | /* No active slots */ | |
363 | if (q->tail == SFQ_DEPTH) | |
364 | return NULL; | |
365 | ||
aa3e2199 | 366 | a = q->next[q->tail]; |
1da177e4 LT |
367 | |
368 | /* Grab packet */ | |
369 | skb = __skb_dequeue(&q->qs[a]); | |
370 | sfq_dec(q, a); | |
371 | sch->q.qlen--; | |
0abf77e5 | 372 | sch->qstats.backlog -= qdisc_pkt_len(skb); |
1da177e4 LT |
373 | |
374 | /* Is the slot empty? */ | |
375 | if (q->qs[a].qlen == 0) { | |
376 | q->ht[q->hash[a]] = SFQ_DEPTH; | |
aa3e2199 ED |
377 | next_a = q->next[a]; |
378 | if (a == next_a) { | |
1da177e4 LT |
379 | q->tail = SFQ_DEPTH; |
380 | return skb; | |
381 | } | |
aa3e2199 | 382 | q->next[q->tail] = next_a; |
0abf77e5 | 383 | } else if ((q->allot[a] -= qdisc_pkt_len(skb)) <= 0) { |
1da177e4 | 384 | q->allot[a] += q->quantum; |
aa3e2199 | 385 | q->tail = a; |
1da177e4 LT |
386 | } |
387 | return skb; | |
388 | } | |
389 | ||
390 | static void | |
6f9e98f7 | 391 | sfq_reset(struct Qdisc *sch) |
1da177e4 LT |
392 | { |
393 | struct sk_buff *skb; | |
394 | ||
395 | while ((skb = sfq_dequeue(sch)) != NULL) | |
396 | kfree_skb(skb); | |
397 | } | |
398 | ||
399 | static void sfq_perturbation(unsigned long arg) | |
400 | { | |
6f9e98f7 | 401 | struct Qdisc *sch = (struct Qdisc *)arg; |
1da177e4 LT |
402 | struct sfq_sched_data *q = qdisc_priv(sch); |
403 | ||
d46f8dd8 | 404 | q->perturbation = net_random(); |
1da177e4 | 405 | |
32740ddc AK |
406 | if (q->perturb_period) |
407 | mod_timer(&q->perturb_timer, jiffies + q->perturb_period); | |
1da177e4 LT |
408 | } |
409 | ||
1e90474c | 410 | static int sfq_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
411 | { |
412 | struct sfq_sched_data *q = qdisc_priv(sch); | |
1e90474c | 413 | struct tc_sfq_qopt *ctl = nla_data(opt); |
5e50da01 | 414 | unsigned int qlen; |
1da177e4 | 415 | |
1e90474c | 416 | if (opt->nla_len < nla_attr_size(sizeof(*ctl))) |
1da177e4 LT |
417 | return -EINVAL; |
418 | ||
419 | sch_tree_lock(sch); | |
5ce2d488 | 420 | q->quantum = ctl->quantum ? : psched_mtu(qdisc_dev(sch)); |
6f9e98f7 | 421 | q->perturb_period = ctl->perturb_period * HZ; |
1da177e4 | 422 | if (ctl->limit) |
32740ddc | 423 | q->limit = min_t(u32, ctl->limit, SFQ_DEPTH - 1); |
1da177e4 | 424 | |
5e50da01 | 425 | qlen = sch->q.qlen; |
5588b40d | 426 | while (sch->q.qlen > q->limit) |
1da177e4 | 427 | sfq_drop(sch); |
5e50da01 | 428 | qdisc_tree_decrease_qlen(sch, qlen - sch->q.qlen); |
1da177e4 LT |
429 | |
430 | del_timer(&q->perturb_timer); | |
431 | if (q->perturb_period) { | |
32740ddc | 432 | mod_timer(&q->perturb_timer, jiffies + q->perturb_period); |
d46f8dd8 | 433 | q->perturbation = net_random(); |
1da177e4 LT |
434 | } |
435 | sch_tree_unlock(sch); | |
436 | return 0; | |
437 | } | |
438 | ||
1e90474c | 439 | static int sfq_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
440 | { |
441 | struct sfq_sched_data *q = qdisc_priv(sch); | |
442 | int i; | |
443 | ||
d3e99483 | 444 | q->perturb_timer.function = sfq_perturbation; |
c19a28e1 | 445 | q->perturb_timer.data = (unsigned long)sch; |
d3e99483 | 446 | init_timer_deferrable(&q->perturb_timer); |
1da177e4 | 447 | |
6f9e98f7 | 448 | for (i = 0; i < SFQ_HASH_DIVISOR; i++) |
1da177e4 | 449 | q->ht[i] = SFQ_DEPTH; |
6f9e98f7 SH |
450 | |
451 | for (i = 0; i < SFQ_DEPTH; i++) { | |
1da177e4 | 452 | skb_queue_head_init(&q->qs[i]); |
6f9e98f7 SH |
453 | q->dep[i + SFQ_DEPTH].next = i + SFQ_DEPTH; |
454 | q->dep[i + SFQ_DEPTH].prev = i + SFQ_DEPTH; | |
1da177e4 | 455 | } |
6f9e98f7 | 456 | |
32740ddc | 457 | q->limit = SFQ_DEPTH - 1; |
1da177e4 LT |
458 | q->max_depth = 0; |
459 | q->tail = SFQ_DEPTH; | |
460 | if (opt == NULL) { | |
5ce2d488 | 461 | q->quantum = psched_mtu(qdisc_dev(sch)); |
1da177e4 | 462 | q->perturb_period = 0; |
d46f8dd8 | 463 | q->perturbation = net_random(); |
1da177e4 LT |
464 | } else { |
465 | int err = sfq_change(sch, opt); | |
466 | if (err) | |
467 | return err; | |
468 | } | |
6f9e98f7 SH |
469 | |
470 | for (i = 0; i < SFQ_DEPTH; i++) | |
1da177e4 LT |
471 | sfq_link(q, i); |
472 | return 0; | |
473 | } | |
474 | ||
475 | static void sfq_destroy(struct Qdisc *sch) | |
476 | { | |
477 | struct sfq_sched_data *q = qdisc_priv(sch); | |
7d2681a6 | 478 | |
ff31ab56 | 479 | tcf_destroy_chain(&q->filter_list); |
980c478d JP |
480 | q->perturb_period = 0; |
481 | del_timer_sync(&q->perturb_timer); | |
1da177e4 LT |
482 | } |
483 | ||
484 | static int sfq_dump(struct Qdisc *sch, struct sk_buff *skb) | |
485 | { | |
486 | struct sfq_sched_data *q = qdisc_priv(sch); | |
27a884dc | 487 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
488 | struct tc_sfq_qopt opt; |
489 | ||
490 | opt.quantum = q->quantum; | |
6f9e98f7 | 491 | opt.perturb_period = q->perturb_period / HZ; |
1da177e4 LT |
492 | |
493 | opt.limit = q->limit; | |
494 | opt.divisor = SFQ_HASH_DIVISOR; | |
cdec7e50 | 495 | opt.flows = q->limit; |
1da177e4 | 496 | |
1e90474c | 497 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); |
1da177e4 LT |
498 | |
499 | return skb->len; | |
500 | ||
1e90474c | 501 | nla_put_failure: |
dc5fc579 | 502 | nlmsg_trim(skb, b); |
1da177e4 LT |
503 | return -1; |
504 | } | |
505 | ||
41065fba JP |
506 | static struct Qdisc *sfq_leaf(struct Qdisc *sch, unsigned long arg) |
507 | { | |
508 | return NULL; | |
509 | } | |
510 | ||
7d2681a6 PM |
511 | static unsigned long sfq_get(struct Qdisc *sch, u32 classid) |
512 | { | |
513 | return 0; | |
514 | } | |
515 | ||
eb4a5527 JP |
516 | static unsigned long sfq_bind(struct Qdisc *sch, unsigned long parent, |
517 | u32 classid) | |
518 | { | |
519 | return 0; | |
520 | } | |
521 | ||
da7115d9 JP |
522 | static void sfq_put(struct Qdisc *q, unsigned long cl) |
523 | { | |
524 | } | |
525 | ||
7d2681a6 PM |
526 | static struct tcf_proto **sfq_find_tcf(struct Qdisc *sch, unsigned long cl) |
527 | { | |
528 | struct sfq_sched_data *q = qdisc_priv(sch); | |
529 | ||
530 | if (cl) | |
531 | return NULL; | |
532 | return &q->filter_list; | |
533 | } | |
534 | ||
94de78d1 PM |
535 | static int sfq_dump_class(struct Qdisc *sch, unsigned long cl, |
536 | struct sk_buff *skb, struct tcmsg *tcm) | |
537 | { | |
538 | tcm->tcm_handle |= TC_H_MIN(cl); | |
539 | return 0; | |
540 | } | |
541 | ||
542 | static int sfq_dump_class_stats(struct Qdisc *sch, unsigned long cl, | |
543 | struct gnet_dump *d) | |
544 | { | |
545 | struct sfq_sched_data *q = qdisc_priv(sch); | |
546 | sfq_index idx = q->ht[cl-1]; | |
c4266263 ED |
547 | struct sk_buff_head *list = &q->qs[idx]; |
548 | struct gnet_stats_queue qs = { .qlen = list->qlen }; | |
94de78d1 | 549 | struct tc_sfq_xstats xstats = { .allot = q->allot[idx] }; |
c4266263 ED |
550 | struct sk_buff *skb; |
551 | ||
552 | skb_queue_walk(list, skb) | |
553 | qs.backlog += qdisc_pkt_len(skb); | |
94de78d1 PM |
554 | |
555 | if (gnet_stats_copy_queue(d, &qs) < 0) | |
556 | return -1; | |
557 | return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); | |
558 | } | |
559 | ||
7d2681a6 PM |
560 | static void sfq_walk(struct Qdisc *sch, struct qdisc_walker *arg) |
561 | { | |
94de78d1 PM |
562 | struct sfq_sched_data *q = qdisc_priv(sch); |
563 | unsigned int i; | |
564 | ||
565 | if (arg->stop) | |
566 | return; | |
567 | ||
568 | for (i = 0; i < SFQ_HASH_DIVISOR; i++) { | |
569 | if (q->ht[i] == SFQ_DEPTH || | |
570 | arg->count < arg->skip) { | |
571 | arg->count++; | |
572 | continue; | |
573 | } | |
574 | if (arg->fn(sch, i + 1, arg) < 0) { | |
575 | arg->stop = 1; | |
576 | break; | |
577 | } | |
578 | arg->count++; | |
579 | } | |
7d2681a6 PM |
580 | } |
581 | ||
582 | static const struct Qdisc_class_ops sfq_class_ops = { | |
41065fba | 583 | .leaf = sfq_leaf, |
7d2681a6 | 584 | .get = sfq_get, |
da7115d9 | 585 | .put = sfq_put, |
7d2681a6 | 586 | .tcf_chain = sfq_find_tcf, |
eb4a5527 | 587 | .bind_tcf = sfq_bind, |
da7115d9 | 588 | .unbind_tcf = sfq_put, |
94de78d1 PM |
589 | .dump = sfq_dump_class, |
590 | .dump_stats = sfq_dump_class_stats, | |
7d2681a6 PM |
591 | .walk = sfq_walk, |
592 | }; | |
593 | ||
20fea08b | 594 | static struct Qdisc_ops sfq_qdisc_ops __read_mostly = { |
7d2681a6 | 595 | .cl_ops = &sfq_class_ops, |
1da177e4 LT |
596 | .id = "sfq", |
597 | .priv_size = sizeof(struct sfq_sched_data), | |
598 | .enqueue = sfq_enqueue, | |
599 | .dequeue = sfq_dequeue, | |
48a8f519 | 600 | .peek = sfq_peek, |
1da177e4 LT |
601 | .drop = sfq_drop, |
602 | .init = sfq_init, | |
603 | .reset = sfq_reset, | |
604 | .destroy = sfq_destroy, | |
605 | .change = NULL, | |
606 | .dump = sfq_dump, | |
607 | .owner = THIS_MODULE, | |
608 | }; | |
609 | ||
610 | static int __init sfq_module_init(void) | |
611 | { | |
612 | return register_qdisc(&sfq_qdisc_ops); | |
613 | } | |
10297b99 | 614 | static void __exit sfq_module_exit(void) |
1da177e4 LT |
615 | { |
616 | unregister_qdisc(&sfq_qdisc_ops); | |
617 | } | |
618 | module_init(sfq_module_init) | |
619 | module_exit(sfq_module_exit) | |
620 | MODULE_LICENSE("GPL"); |