]>
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; | |
eeaeb068 | 70 | max mtu to 2^18-1; max 128 flows, number of hash buckets to 1024. |
1da177e4 | 71 | The only goal of this restrictions was that all data |
eda83e3b | 72 | fit into one 4K page on 32bit arches. |
1da177e4 LT |
73 | |
74 | It is easy to increase these values, but not in flight. */ | |
75 | ||
eda83e3b ED |
76 | #define SFQ_DEPTH 128 /* max number of packets per flow */ |
77 | #define SFQ_SLOTS 128 /* max number of flows */ | |
78 | #define SFQ_EMPTY_SLOT 255 | |
1da177e4 | 79 | #define SFQ_HASH_DIVISOR 1024 |
eeaeb068 ED |
80 | /* We use 16 bits to store allot, and want to handle packets up to 64K |
81 | * Scale allot by 8 (1<<3) so that no overflow occurs. | |
82 | */ | |
83 | #define SFQ_ALLOT_SHIFT 3 | |
84 | #define SFQ_ALLOT_SIZE(X) DIV_ROUND_UP(X, 1 << SFQ_ALLOT_SHIFT) | |
1da177e4 | 85 | |
eda83e3b | 86 | /* This type should contain at least SFQ_DEPTH + SFQ_SLOTS values */ |
1da177e4 LT |
87 | typedef unsigned char sfq_index; |
88 | ||
eda83e3b ED |
89 | /* |
90 | * We dont use pointers to save space. | |
91 | * Small indexes [0 ... SFQ_SLOTS - 1] are 'pointers' to slots[] array | |
92 | * while following values [SFQ_SLOTS ... SFQ_SLOTS + SFQ_DEPTH - 1] | |
93 | * are 'pointers' to dep[] array | |
94 | */ | |
cc7ec456 | 95 | struct sfq_head { |
1da177e4 LT |
96 | sfq_index next; |
97 | sfq_index prev; | |
98 | }; | |
99 | ||
eda83e3b ED |
100 | struct sfq_slot { |
101 | struct sk_buff *skblist_next; | |
102 | struct sk_buff *skblist_prev; | |
103 | sfq_index qlen; /* number of skbs in skblist */ | |
104 | sfq_index next; /* next slot in sfq chain */ | |
105 | struct sfq_head dep; /* anchor in dep[] chains */ | |
106 | unsigned short hash; /* hash value (index in ht[]) */ | |
107 | short allot; /* credit for this slot */ | |
108 | }; | |
109 | ||
cc7ec456 | 110 | struct sfq_sched_data { |
1da177e4 LT |
111 | /* Parameters */ |
112 | int perturb_period; | |
cc7ec456 | 113 | unsigned int quantum; /* Allotment per round: MUST BE >= MTU */ |
1da177e4 LT |
114 | int limit; |
115 | ||
116 | /* Variables */ | |
7d2681a6 | 117 | struct tcf_proto *filter_list; |
1da177e4 | 118 | struct timer_list perturb_timer; |
32740ddc | 119 | u32 perturbation; |
eda83e3b | 120 | sfq_index cur_depth; /* depth of longest slot */ |
eeaeb068 | 121 | unsigned short scaled_quantum; /* SFQ_ALLOT_SIZE(quantum) */ |
eda83e3b | 122 | struct sfq_slot *tail; /* current slot in round */ |
1da177e4 | 123 | sfq_index ht[SFQ_HASH_DIVISOR]; /* Hash table */ |
eda83e3b ED |
124 | struct sfq_slot slots[SFQ_SLOTS]; |
125 | struct sfq_head dep[SFQ_DEPTH]; /* Linked list of slots, indexed by depth */ | |
1da177e4 LT |
126 | }; |
127 | ||
eda83e3b ED |
128 | /* |
129 | * sfq_head are either in a sfq_slot or in dep[] array | |
130 | */ | |
131 | static inline struct sfq_head *sfq_dep_head(struct sfq_sched_data *q, sfq_index val) | |
132 | { | |
133 | if (val < SFQ_SLOTS) | |
134 | return &q->slots[val].dep; | |
135 | return &q->dep[val - SFQ_SLOTS]; | |
136 | } | |
137 | ||
cc7ec456 | 138 | static unsigned int sfq_fold_hash(struct sfq_sched_data *q, u32 h, u32 h1) |
1da177e4 | 139 | { |
32740ddc | 140 | return jhash_2words(h, h1, q->perturbation) & (SFQ_HASH_DIVISOR - 1); |
1da177e4 LT |
141 | } |
142 | ||
cc7ec456 | 143 | static unsigned int sfq_hash(struct sfq_sched_data *q, struct sk_buff *skb) |
1da177e4 LT |
144 | { |
145 | u32 h, h2; | |
146 | ||
147 | switch (skb->protocol) { | |
60678040 | 148 | case htons(ETH_P_IP): |
1da177e4 | 149 | { |
f2f00981 | 150 | const struct iphdr *iph; |
b9959c2e | 151 | int poff; |
f2f00981 CG |
152 | |
153 | if (!pskb_network_may_pull(skb, sizeof(*iph))) | |
154 | goto err; | |
155 | iph = ip_hdr(skb); | |
0eae88f3 ED |
156 | h = (__force u32)iph->daddr; |
157 | h2 = (__force u32)iph->saddr ^ iph->protocol; | |
cc7ec456 | 158 | if (iph->frag_off & htons(IP_MF | IP_OFFSET)) |
b9959c2e CG |
159 | break; |
160 | poff = proto_ports_offset(iph->protocol); | |
161 | if (poff >= 0 && | |
162 | pskb_network_may_pull(skb, iph->ihl * 4 + 4 + poff)) { | |
163 | iph = ip_hdr(skb); | |
cc7ec456 | 164 | h2 ^= *(u32 *)((void *)iph + iph->ihl * 4 + poff); |
b9959c2e | 165 | } |
1da177e4 LT |
166 | break; |
167 | } | |
60678040 | 168 | case htons(ETH_P_IPV6): |
1da177e4 | 169 | { |
f2f00981 | 170 | struct ipv6hdr *iph; |
b9959c2e | 171 | int poff; |
f2f00981 CG |
172 | |
173 | if (!pskb_network_may_pull(skb, sizeof(*iph))) | |
174 | goto err; | |
175 | iph = ipv6_hdr(skb); | |
0eae88f3 ED |
176 | h = (__force u32)iph->daddr.s6_addr32[3]; |
177 | h2 = (__force u32)iph->saddr.s6_addr32[3] ^ iph->nexthdr; | |
b9959c2e CG |
178 | poff = proto_ports_offset(iph->nexthdr); |
179 | if (poff >= 0 && | |
180 | pskb_network_may_pull(skb, sizeof(*iph) + 4 + poff)) { | |
181 | iph = ipv6_hdr(skb); | |
cc7ec456 | 182 | h2 ^= *(u32 *)((void *)iph + sizeof(*iph) + poff); |
b9959c2e | 183 | } |
1da177e4 LT |
184 | break; |
185 | } | |
186 | default: | |
f2f00981 | 187 | err: |
0eae88f3 | 188 | h = (unsigned long)skb_dst(skb) ^ (__force u32)skb->protocol; |
6f9e98f7 | 189 | h2 = (unsigned long)skb->sk; |
1da177e4 | 190 | } |
6f9e98f7 | 191 | |
1da177e4 LT |
192 | return sfq_fold_hash(q, h, h2); |
193 | } | |
194 | ||
7d2681a6 PM |
195 | static unsigned int sfq_classify(struct sk_buff *skb, struct Qdisc *sch, |
196 | int *qerr) | |
197 | { | |
198 | struct sfq_sched_data *q = qdisc_priv(sch); | |
199 | struct tcf_result res; | |
200 | int result; | |
201 | ||
202 | if (TC_H_MAJ(skb->priority) == sch->handle && | |
203 | TC_H_MIN(skb->priority) > 0 && | |
204 | TC_H_MIN(skb->priority) <= SFQ_HASH_DIVISOR) | |
205 | return TC_H_MIN(skb->priority); | |
206 | ||
207 | if (!q->filter_list) | |
208 | return sfq_hash(q, skb) + 1; | |
209 | ||
c27f339a | 210 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
7d2681a6 PM |
211 | result = tc_classify(skb, q->filter_list, &res); |
212 | if (result >= 0) { | |
213 | #ifdef CONFIG_NET_CLS_ACT | |
214 | switch (result) { | |
215 | case TC_ACT_STOLEN: | |
216 | case TC_ACT_QUEUED: | |
378a2f09 | 217 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
7d2681a6 PM |
218 | case TC_ACT_SHOT: |
219 | return 0; | |
220 | } | |
221 | #endif | |
222 | if (TC_H_MIN(res.classid) <= SFQ_HASH_DIVISOR) | |
223 | return TC_H_MIN(res.classid); | |
224 | } | |
225 | return 0; | |
226 | } | |
227 | ||
eda83e3b ED |
228 | /* |
229 | * x : slot number [0 .. SFQ_SLOTS - 1] | |
230 | */ | |
1da177e4 LT |
231 | static inline void sfq_link(struct sfq_sched_data *q, sfq_index x) |
232 | { | |
233 | sfq_index p, n; | |
eda83e3b ED |
234 | int qlen = q->slots[x].qlen; |
235 | ||
236 | p = qlen + SFQ_SLOTS; | |
237 | n = q->dep[qlen].next; | |
1da177e4 | 238 | |
eda83e3b ED |
239 | q->slots[x].dep.next = n; |
240 | q->slots[x].dep.prev = p; | |
241 | ||
242 | q->dep[qlen].next = x; /* sfq_dep_head(q, p)->next = x */ | |
243 | sfq_dep_head(q, n)->prev = x; | |
1da177e4 LT |
244 | } |
245 | ||
eda83e3b ED |
246 | #define sfq_unlink(q, x, n, p) \ |
247 | n = q->slots[x].dep.next; \ | |
248 | p = q->slots[x].dep.prev; \ | |
249 | sfq_dep_head(q, p)->next = n; \ | |
250 | sfq_dep_head(q, n)->prev = p | |
251 | ||
252 | ||
1da177e4 LT |
253 | static inline void sfq_dec(struct sfq_sched_data *q, sfq_index x) |
254 | { | |
255 | sfq_index p, n; | |
eda83e3b | 256 | int d; |
1da177e4 | 257 | |
eda83e3b | 258 | sfq_unlink(q, x, n, p); |
1da177e4 | 259 | |
eda83e3b ED |
260 | d = q->slots[x].qlen--; |
261 | if (n == p && q->cur_depth == d) | |
262 | q->cur_depth--; | |
1da177e4 LT |
263 | sfq_link(q, x); |
264 | } | |
265 | ||
266 | static inline void sfq_inc(struct sfq_sched_data *q, sfq_index x) | |
267 | { | |
268 | sfq_index p, n; | |
269 | int d; | |
270 | ||
eda83e3b | 271 | sfq_unlink(q, x, n, p); |
1da177e4 | 272 | |
eda83e3b ED |
273 | d = ++q->slots[x].qlen; |
274 | if (q->cur_depth < d) | |
275 | q->cur_depth = d; | |
1da177e4 LT |
276 | sfq_link(q, x); |
277 | } | |
278 | ||
eda83e3b ED |
279 | /* helper functions : might be changed when/if skb use a standard list_head */ |
280 | ||
281 | /* remove one skb from tail of slot queue */ | |
282 | static inline struct sk_buff *slot_dequeue_tail(struct sfq_slot *slot) | |
283 | { | |
284 | struct sk_buff *skb = slot->skblist_prev; | |
285 | ||
286 | slot->skblist_prev = skb->prev; | |
ee09b3c1 | 287 | skb->prev->next = (struct sk_buff *)slot; |
eda83e3b ED |
288 | skb->next = skb->prev = NULL; |
289 | return skb; | |
290 | } | |
291 | ||
292 | /* remove one skb from head of slot queue */ | |
293 | static inline struct sk_buff *slot_dequeue_head(struct sfq_slot *slot) | |
294 | { | |
295 | struct sk_buff *skb = slot->skblist_next; | |
296 | ||
297 | slot->skblist_next = skb->next; | |
18c8d82a | 298 | skb->next->prev = (struct sk_buff *)slot; |
eda83e3b ED |
299 | skb->next = skb->prev = NULL; |
300 | return skb; | |
301 | } | |
302 | ||
303 | static inline void slot_queue_init(struct sfq_slot *slot) | |
304 | { | |
305 | slot->skblist_prev = slot->skblist_next = (struct sk_buff *)slot; | |
306 | } | |
307 | ||
308 | /* add skb to slot queue (tail add) */ | |
309 | static inline void slot_queue_add(struct sfq_slot *slot, struct sk_buff *skb) | |
310 | { | |
311 | skb->prev = slot->skblist_prev; | |
312 | skb->next = (struct sk_buff *)slot; | |
313 | slot->skblist_prev->next = skb; | |
314 | slot->skblist_prev = skb; | |
315 | } | |
316 | ||
317 | #define slot_queue_walk(slot, skb) \ | |
318 | for (skb = slot->skblist_next; \ | |
319 | skb != (struct sk_buff *)slot; \ | |
320 | skb = skb->next) | |
321 | ||
1da177e4 LT |
322 | static unsigned int sfq_drop(struct Qdisc *sch) |
323 | { | |
324 | struct sfq_sched_data *q = qdisc_priv(sch); | |
eda83e3b | 325 | sfq_index x, d = q->cur_depth; |
1da177e4 LT |
326 | struct sk_buff *skb; |
327 | unsigned int len; | |
eda83e3b | 328 | struct sfq_slot *slot; |
1da177e4 | 329 | |
eda83e3b | 330 | /* Queue is full! Find the longest slot and drop tail packet from it */ |
1da177e4 | 331 | if (d > 1) { |
eda83e3b ED |
332 | x = q->dep[d].next; |
333 | slot = &q->slots[x]; | |
334 | drop: | |
335 | skb = slot_dequeue_tail(slot); | |
0abf77e5 | 336 | len = qdisc_pkt_len(skb); |
1da177e4 | 337 | sfq_dec(q, x); |
eda83e3b | 338 | kfree_skb(skb); |
1da177e4 LT |
339 | sch->q.qlen--; |
340 | sch->qstats.drops++; | |
f5539eb8 | 341 | sch->qstats.backlog -= len; |
1da177e4 LT |
342 | return len; |
343 | } | |
344 | ||
345 | if (d == 1) { | |
346 | /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */ | |
eda83e3b ED |
347 | x = q->tail->next; |
348 | slot = &q->slots[x]; | |
349 | q->tail->next = slot->next; | |
350 | q->ht[slot->hash] = SFQ_EMPTY_SLOT; | |
351 | goto drop; | |
1da177e4 LT |
352 | } |
353 | ||
354 | return 0; | |
355 | } | |
356 | ||
357 | static int | |
6f9e98f7 | 358 | sfq_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
1da177e4 LT |
359 | { |
360 | struct sfq_sched_data *q = qdisc_priv(sch); | |
7d2681a6 | 361 | unsigned int hash; |
1da177e4 | 362 | sfq_index x; |
eda83e3b | 363 | struct sfq_slot *slot; |
7f3ff4f6 | 364 | int uninitialized_var(ret); |
7d2681a6 PM |
365 | |
366 | hash = sfq_classify(skb, sch, &ret); | |
367 | if (hash == 0) { | |
c27f339a | 368 | if (ret & __NET_XMIT_BYPASS) |
7d2681a6 PM |
369 | sch->qstats.drops++; |
370 | kfree_skb(skb); | |
371 | return ret; | |
372 | } | |
373 | hash--; | |
1da177e4 LT |
374 | |
375 | x = q->ht[hash]; | |
eda83e3b ED |
376 | slot = &q->slots[x]; |
377 | if (x == SFQ_EMPTY_SLOT) { | |
378 | x = q->dep[0].next; /* get a free slot */ | |
379 | q->ht[hash] = x; | |
380 | slot = &q->slots[x]; | |
381 | slot->hash = hash; | |
1da177e4 | 382 | } |
6f9e98f7 | 383 | |
eda83e3b | 384 | /* If selected queue has length q->limit, do simple tail drop, |
32740ddc AK |
385 | * i.e. drop _this_ packet. |
386 | */ | |
eda83e3b | 387 | if (slot->qlen >= q->limit) |
32740ddc AK |
388 | return qdisc_drop(skb, sch); |
389 | ||
0abf77e5 | 390 | sch->qstats.backlog += qdisc_pkt_len(skb); |
eda83e3b | 391 | slot_queue_add(slot, skb); |
1da177e4 | 392 | sfq_inc(q, x); |
eda83e3b ED |
393 | if (slot->qlen == 1) { /* The flow is new */ |
394 | if (q->tail == NULL) { /* It is the first flow */ | |
395 | slot->next = x; | |
1da177e4 | 396 | } else { |
eda83e3b ED |
397 | slot->next = q->tail->next; |
398 | q->tail->next = x; | |
1da177e4 | 399 | } |
eda83e3b | 400 | q->tail = slot; |
eeaeb068 | 401 | slot->allot = q->scaled_quantum; |
1da177e4 | 402 | } |
5588b40d | 403 | if (++sch->q.qlen <= q->limit) { |
bfe0d029 | 404 | qdisc_bstats_update(sch, skb); |
9871e50e | 405 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
406 | } |
407 | ||
408 | sfq_drop(sch); | |
409 | return NET_XMIT_CN; | |
410 | } | |
411 | ||
48a8f519 PM |
412 | static struct sk_buff * |
413 | sfq_peek(struct Qdisc *sch) | |
414 | { | |
415 | struct sfq_sched_data *q = qdisc_priv(sch); | |
1da177e4 | 416 | |
48a8f519 | 417 | /* No active slots */ |
eda83e3b | 418 | if (q->tail == NULL) |
48a8f519 | 419 | return NULL; |
1da177e4 | 420 | |
eda83e3b | 421 | return q->slots[q->tail->next].skblist_next; |
48a8f519 | 422 | } |
1da177e4 LT |
423 | |
424 | static struct sk_buff * | |
6f9e98f7 | 425 | sfq_dequeue(struct Qdisc *sch) |
1da177e4 LT |
426 | { |
427 | struct sfq_sched_data *q = qdisc_priv(sch); | |
428 | struct sk_buff *skb; | |
aa3e2199 | 429 | sfq_index a, next_a; |
eda83e3b | 430 | struct sfq_slot *slot; |
1da177e4 LT |
431 | |
432 | /* No active slots */ | |
eda83e3b | 433 | if (q->tail == NULL) |
1da177e4 LT |
434 | return NULL; |
435 | ||
eeaeb068 | 436 | next_slot: |
eda83e3b ED |
437 | a = q->tail->next; |
438 | slot = &q->slots[a]; | |
eeaeb068 ED |
439 | if (slot->allot <= 0) { |
440 | q->tail = slot; | |
441 | slot->allot += q->scaled_quantum; | |
442 | goto next_slot; | |
443 | } | |
eda83e3b | 444 | skb = slot_dequeue_head(slot); |
1da177e4 LT |
445 | sfq_dec(q, a); |
446 | sch->q.qlen--; | |
0abf77e5 | 447 | sch->qstats.backlog -= qdisc_pkt_len(skb); |
1da177e4 LT |
448 | |
449 | /* Is the slot empty? */ | |
eda83e3b ED |
450 | if (slot->qlen == 0) { |
451 | q->ht[slot->hash] = SFQ_EMPTY_SLOT; | |
452 | next_a = slot->next; | |
aa3e2199 | 453 | if (a == next_a) { |
eda83e3b | 454 | q->tail = NULL; /* no more active slots */ |
1da177e4 LT |
455 | return skb; |
456 | } | |
eda83e3b | 457 | q->tail->next = next_a; |
eeaeb068 ED |
458 | } else { |
459 | slot->allot -= SFQ_ALLOT_SIZE(qdisc_pkt_len(skb)); | |
1da177e4 LT |
460 | } |
461 | return skb; | |
462 | } | |
463 | ||
464 | static void | |
6f9e98f7 | 465 | sfq_reset(struct Qdisc *sch) |
1da177e4 LT |
466 | { |
467 | struct sk_buff *skb; | |
468 | ||
469 | while ((skb = sfq_dequeue(sch)) != NULL) | |
470 | kfree_skb(skb); | |
471 | } | |
472 | ||
473 | static void sfq_perturbation(unsigned long arg) | |
474 | { | |
6f9e98f7 | 475 | struct Qdisc *sch = (struct Qdisc *)arg; |
1da177e4 LT |
476 | struct sfq_sched_data *q = qdisc_priv(sch); |
477 | ||
d46f8dd8 | 478 | q->perturbation = net_random(); |
1da177e4 | 479 | |
32740ddc AK |
480 | if (q->perturb_period) |
481 | mod_timer(&q->perturb_timer, jiffies + q->perturb_period); | |
1da177e4 LT |
482 | } |
483 | ||
1e90474c | 484 | static int sfq_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
485 | { |
486 | struct sfq_sched_data *q = qdisc_priv(sch); | |
1e90474c | 487 | struct tc_sfq_qopt *ctl = nla_data(opt); |
5e50da01 | 488 | unsigned int qlen; |
1da177e4 | 489 | |
1e90474c | 490 | if (opt->nla_len < nla_attr_size(sizeof(*ctl))) |
1da177e4 LT |
491 | return -EINVAL; |
492 | ||
493 | sch_tree_lock(sch); | |
5ce2d488 | 494 | q->quantum = ctl->quantum ? : psched_mtu(qdisc_dev(sch)); |
eeaeb068 | 495 | q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum); |
6f9e98f7 | 496 | q->perturb_period = ctl->perturb_period * HZ; |
1da177e4 | 497 | if (ctl->limit) |
32740ddc | 498 | q->limit = min_t(u32, ctl->limit, SFQ_DEPTH - 1); |
1da177e4 | 499 | |
5e50da01 | 500 | qlen = sch->q.qlen; |
5588b40d | 501 | while (sch->q.qlen > q->limit) |
1da177e4 | 502 | sfq_drop(sch); |
5e50da01 | 503 | qdisc_tree_decrease_qlen(sch, qlen - sch->q.qlen); |
1da177e4 LT |
504 | |
505 | del_timer(&q->perturb_timer); | |
506 | if (q->perturb_period) { | |
32740ddc | 507 | mod_timer(&q->perturb_timer, jiffies + q->perturb_period); |
d46f8dd8 | 508 | q->perturbation = net_random(); |
1da177e4 LT |
509 | } |
510 | sch_tree_unlock(sch); | |
511 | return 0; | |
512 | } | |
513 | ||
1e90474c | 514 | static int sfq_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
515 | { |
516 | struct sfq_sched_data *q = qdisc_priv(sch); | |
517 | int i; | |
518 | ||
d3e99483 | 519 | q->perturb_timer.function = sfq_perturbation; |
c19a28e1 | 520 | q->perturb_timer.data = (unsigned long)sch; |
d3e99483 | 521 | init_timer_deferrable(&q->perturb_timer); |
1da177e4 | 522 | |
6f9e98f7 | 523 | for (i = 0; i < SFQ_HASH_DIVISOR; i++) |
eda83e3b | 524 | q->ht[i] = SFQ_EMPTY_SLOT; |
6f9e98f7 SH |
525 | |
526 | for (i = 0; i < SFQ_DEPTH; i++) { | |
eda83e3b ED |
527 | q->dep[i].next = i + SFQ_SLOTS; |
528 | q->dep[i].prev = i + SFQ_SLOTS; | |
1da177e4 | 529 | } |
6f9e98f7 | 530 | |
32740ddc | 531 | q->limit = SFQ_DEPTH - 1; |
eda83e3b ED |
532 | q->cur_depth = 0; |
533 | q->tail = NULL; | |
1da177e4 | 534 | if (opt == NULL) { |
5ce2d488 | 535 | q->quantum = psched_mtu(qdisc_dev(sch)); |
eeaeb068 | 536 | q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum); |
1da177e4 | 537 | q->perturb_period = 0; |
d46f8dd8 | 538 | q->perturbation = net_random(); |
1da177e4 LT |
539 | } else { |
540 | int err = sfq_change(sch, opt); | |
541 | if (err) | |
542 | return err; | |
543 | } | |
6f9e98f7 | 544 | |
18c8d82a ED |
545 | for (i = 0; i < SFQ_SLOTS; i++) { |
546 | slot_queue_init(&q->slots[i]); | |
1da177e4 | 547 | sfq_link(q, i); |
18c8d82a | 548 | } |
1da177e4 LT |
549 | return 0; |
550 | } | |
551 | ||
552 | static void sfq_destroy(struct Qdisc *sch) | |
553 | { | |
554 | struct sfq_sched_data *q = qdisc_priv(sch); | |
7d2681a6 | 555 | |
ff31ab56 | 556 | tcf_destroy_chain(&q->filter_list); |
980c478d JP |
557 | q->perturb_period = 0; |
558 | del_timer_sync(&q->perturb_timer); | |
1da177e4 LT |
559 | } |
560 | ||
561 | static int sfq_dump(struct Qdisc *sch, struct sk_buff *skb) | |
562 | { | |
563 | struct sfq_sched_data *q = qdisc_priv(sch); | |
27a884dc | 564 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
565 | struct tc_sfq_qopt opt; |
566 | ||
567 | opt.quantum = q->quantum; | |
6f9e98f7 | 568 | opt.perturb_period = q->perturb_period / HZ; |
1da177e4 LT |
569 | |
570 | opt.limit = q->limit; | |
571 | opt.divisor = SFQ_HASH_DIVISOR; | |
cdec7e50 | 572 | opt.flows = q->limit; |
1da177e4 | 573 | |
1e90474c | 574 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); |
1da177e4 LT |
575 | |
576 | return skb->len; | |
577 | ||
1e90474c | 578 | nla_put_failure: |
dc5fc579 | 579 | nlmsg_trim(skb, b); |
1da177e4 LT |
580 | return -1; |
581 | } | |
582 | ||
41065fba JP |
583 | static struct Qdisc *sfq_leaf(struct Qdisc *sch, unsigned long arg) |
584 | { | |
585 | return NULL; | |
586 | } | |
587 | ||
7d2681a6 PM |
588 | static unsigned long sfq_get(struct Qdisc *sch, u32 classid) |
589 | { | |
590 | return 0; | |
591 | } | |
592 | ||
eb4a5527 JP |
593 | static unsigned long sfq_bind(struct Qdisc *sch, unsigned long parent, |
594 | u32 classid) | |
595 | { | |
596 | return 0; | |
597 | } | |
598 | ||
da7115d9 JP |
599 | static void sfq_put(struct Qdisc *q, unsigned long cl) |
600 | { | |
601 | } | |
602 | ||
7d2681a6 PM |
603 | static struct tcf_proto **sfq_find_tcf(struct Qdisc *sch, unsigned long cl) |
604 | { | |
605 | struct sfq_sched_data *q = qdisc_priv(sch); | |
606 | ||
607 | if (cl) | |
608 | return NULL; | |
609 | return &q->filter_list; | |
610 | } | |
611 | ||
94de78d1 PM |
612 | static int sfq_dump_class(struct Qdisc *sch, unsigned long cl, |
613 | struct sk_buff *skb, struct tcmsg *tcm) | |
614 | { | |
615 | tcm->tcm_handle |= TC_H_MIN(cl); | |
616 | return 0; | |
617 | } | |
618 | ||
619 | static int sfq_dump_class_stats(struct Qdisc *sch, unsigned long cl, | |
620 | struct gnet_dump *d) | |
621 | { | |
622 | struct sfq_sched_data *q = qdisc_priv(sch); | |
ee09b3c1 ED |
623 | sfq_index idx = q->ht[cl - 1]; |
624 | struct gnet_stats_queue qs = { 0 }; | |
625 | struct tc_sfq_xstats xstats = { 0 }; | |
c4266263 ED |
626 | struct sk_buff *skb; |
627 | ||
ee09b3c1 ED |
628 | if (idx != SFQ_EMPTY_SLOT) { |
629 | const struct sfq_slot *slot = &q->slots[idx]; | |
94de78d1 | 630 | |
eeaeb068 | 631 | xstats.allot = slot->allot << SFQ_ALLOT_SHIFT; |
ee09b3c1 ED |
632 | qs.qlen = slot->qlen; |
633 | slot_queue_walk(slot, skb) | |
634 | qs.backlog += qdisc_pkt_len(skb); | |
635 | } | |
94de78d1 PM |
636 | if (gnet_stats_copy_queue(d, &qs) < 0) |
637 | return -1; | |
638 | return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); | |
639 | } | |
640 | ||
7d2681a6 PM |
641 | static void sfq_walk(struct Qdisc *sch, struct qdisc_walker *arg) |
642 | { | |
94de78d1 PM |
643 | struct sfq_sched_data *q = qdisc_priv(sch); |
644 | unsigned int i; | |
645 | ||
646 | if (arg->stop) | |
647 | return; | |
648 | ||
649 | for (i = 0; i < SFQ_HASH_DIVISOR; i++) { | |
eda83e3b | 650 | if (q->ht[i] == SFQ_EMPTY_SLOT || |
94de78d1 PM |
651 | arg->count < arg->skip) { |
652 | arg->count++; | |
653 | continue; | |
654 | } | |
655 | if (arg->fn(sch, i + 1, arg) < 0) { | |
656 | arg->stop = 1; | |
657 | break; | |
658 | } | |
659 | arg->count++; | |
660 | } | |
7d2681a6 PM |
661 | } |
662 | ||
663 | static const struct Qdisc_class_ops sfq_class_ops = { | |
41065fba | 664 | .leaf = sfq_leaf, |
7d2681a6 | 665 | .get = sfq_get, |
da7115d9 | 666 | .put = sfq_put, |
7d2681a6 | 667 | .tcf_chain = sfq_find_tcf, |
eb4a5527 | 668 | .bind_tcf = sfq_bind, |
da7115d9 | 669 | .unbind_tcf = sfq_put, |
94de78d1 PM |
670 | .dump = sfq_dump_class, |
671 | .dump_stats = sfq_dump_class_stats, | |
7d2681a6 PM |
672 | .walk = sfq_walk, |
673 | }; | |
674 | ||
20fea08b | 675 | static struct Qdisc_ops sfq_qdisc_ops __read_mostly = { |
7d2681a6 | 676 | .cl_ops = &sfq_class_ops, |
1da177e4 LT |
677 | .id = "sfq", |
678 | .priv_size = sizeof(struct sfq_sched_data), | |
679 | .enqueue = sfq_enqueue, | |
680 | .dequeue = sfq_dequeue, | |
48a8f519 | 681 | .peek = sfq_peek, |
1da177e4 LT |
682 | .drop = sfq_drop, |
683 | .init = sfq_init, | |
684 | .reset = sfq_reset, | |
685 | .destroy = sfq_destroy, | |
686 | .change = NULL, | |
687 | .dump = sfq_dump, | |
688 | .owner = THIS_MODULE, | |
689 | }; | |
690 | ||
691 | static int __init sfq_module_init(void) | |
692 | { | |
693 | return register_qdisc(&sfq_qdisc_ops); | |
694 | } | |
10297b99 | 695 | static void __exit sfq_module_exit(void) |
1da177e4 LT |
696 | { |
697 | unregister_qdisc(&sfq_qdisc_ops); | |
698 | } | |
699 | module_init(sfq_module_init) | |
700 | module_exit(sfq_module_exit) | |
701 | MODULE_LICENSE("GPL"); |