]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * net/sched/sch_cbq.c Class-Based 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 | */ | |
12 | ||
1da177e4 | 13 | #include <linux/module.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
1da177e4 LT |
15 | #include <linux/types.h> |
16 | #include <linux/kernel.h> | |
1da177e4 | 17 | #include <linux/string.h> |
1da177e4 | 18 | #include <linux/errno.h> |
1da177e4 | 19 | #include <linux/skbuff.h> |
0ba48053 | 20 | #include <net/netlink.h> |
1da177e4 | 21 | #include <net/pkt_sched.h> |
cf1facda | 22 | #include <net/pkt_cls.h> |
1da177e4 LT |
23 | |
24 | ||
25 | /* Class-Based Queueing (CBQ) algorithm. | |
26 | ======================================= | |
27 | ||
28 | Sources: [1] Sally Floyd and Van Jacobson, "Link-sharing and Resource | |
10297b99 | 29 | Management Models for Packet Networks", |
1da177e4 LT |
30 | IEEE/ACM Transactions on Networking, Vol.3, No.4, 1995 |
31 | ||
10297b99 | 32 | [2] Sally Floyd, "Notes on CBQ and Guaranteed Service", 1995 |
1da177e4 | 33 | |
10297b99 | 34 | [3] Sally Floyd, "Notes on Class-Based Queueing: Setting |
1da177e4 LT |
35 | Parameters", 1996 |
36 | ||
37 | [4] Sally Floyd and Michael Speer, "Experimental Results | |
38 | for Class-Based Queueing", 1998, not published. | |
39 | ||
40 | ----------------------------------------------------------------------- | |
41 | ||
42 | Algorithm skeleton was taken from NS simulator cbq.cc. | |
43 | If someone wants to check this code against the LBL version, | |
44 | he should take into account that ONLY the skeleton was borrowed, | |
45 | the implementation is different. Particularly: | |
46 | ||
47 | --- The WRR algorithm is different. Our version looks more | |
10297b99 YH |
48 | reasonable (I hope) and works when quanta are allowed to be |
49 | less than MTU, which is always the case when real time classes | |
50 | have small rates. Note, that the statement of [3] is | |
51 | incomplete, delay may actually be estimated even if class | |
52 | per-round allotment is less than MTU. Namely, if per-round | |
53 | allotment is W*r_i, and r_1+...+r_k = r < 1 | |
1da177e4 LT |
54 | |
55 | delay_i <= ([MTU/(W*r_i)]*W*r + W*r + k*MTU)/B | |
56 | ||
57 | In the worst case we have IntServ estimate with D = W*r+k*MTU | |
58 | and C = MTU*r. The proof (if correct at all) is trivial. | |
59 | ||
60 | ||
61 | --- It seems that cbq-2.0 is not very accurate. At least, I cannot | |
62 | interpret some places, which look like wrong translations | |
63 | from NS. Anyone is advised to find these differences | |
64 | and explain to me, why I am wrong 8). | |
65 | ||
66 | --- Linux has no EOI event, so that we cannot estimate true class | |
67 | idle time. Workaround is to consider the next dequeue event | |
68 | as sign that previous packet is finished. This is wrong because of | |
69 | internal device queueing, but on a permanently loaded link it is true. | |
70 | Moreover, combined with clock integrator, this scheme looks | |
71 | very close to an ideal solution. */ | |
72 | ||
73 | struct cbq_sched_data; | |
74 | ||
75 | ||
cc7ec456 | 76 | struct cbq_class { |
d77fea2e | 77 | struct Qdisc_class_common common; |
1da177e4 LT |
78 | struct cbq_class *next_alive; /* next class with backlog in this priority band */ |
79 | ||
80 | /* Parameters */ | |
1da177e4 LT |
81 | unsigned char priority; /* class priority */ |
82 | unsigned char priority2; /* priority to be used after overlimit */ | |
83 | unsigned char ewma_log; /* time constant for idle time calculation */ | |
1da177e4 LT |
84 | |
85 | u32 defmap; | |
86 | ||
87 | /* Link-sharing scheduler parameters */ | |
88 | long maxidle; /* Class parameters: see below. */ | |
89 | long offtime; | |
90 | long minidle; | |
91 | u32 avpkt; | |
92 | struct qdisc_rate_table *R_tab; | |
93 | ||
1da177e4 LT |
94 | /* General scheduler (WRR) parameters */ |
95 | long allot; | |
96 | long quantum; /* Allotment per WRR round */ | |
97 | long weight; /* Relative allotment: see below */ | |
98 | ||
99 | struct Qdisc *qdisc; /* Ptr to CBQ discipline */ | |
100 | struct cbq_class *split; /* Ptr to split node */ | |
101 | struct cbq_class *share; /* Ptr to LS parent in the class tree */ | |
102 | struct cbq_class *tparent; /* Ptr to tree parent in the class tree */ | |
103 | struct cbq_class *borrow; /* NULL if class is bandwidth limited; | |
104 | parent otherwise */ | |
105 | struct cbq_class *sibling; /* Sibling chain */ | |
106 | struct cbq_class *children; /* Pointer to children chain */ | |
107 | ||
108 | struct Qdisc *q; /* Elementary queueing discipline */ | |
109 | ||
110 | ||
111 | /* Variables */ | |
112 | unsigned char cpriority; /* Effective priority */ | |
113 | unsigned char delayed; | |
114 | unsigned char level; /* level of the class in hierarchy: | |
115 | 0 for leaf classes, and maximal | |
116 | level of children + 1 for nodes. | |
117 | */ | |
118 | ||
119 | psched_time_t last; /* Last end of service */ | |
120 | psched_time_t undertime; | |
121 | long avgidle; | |
122 | long deficit; /* Saved deficit for WRR */ | |
1a13cb63 | 123 | psched_time_t penalized; |
c1a8f1f1 | 124 | struct gnet_stats_basic_packed bstats; |
1da177e4 | 125 | struct gnet_stats_queue qstats; |
1c0d32fd | 126 | struct net_rate_estimator __rcu *rate_est; |
1da177e4 LT |
127 | struct tc_cbq_xstats xstats; |
128 | ||
25d8c0d5 | 129 | struct tcf_proto __rcu *filter_list; |
6529eaba | 130 | struct tcf_block *block; |
1da177e4 LT |
131 | |
132 | int refcnt; | |
133 | int filters; | |
134 | ||
cc7ec456 | 135 | struct cbq_class *defaults[TC_PRIO_MAX + 1]; |
1da177e4 LT |
136 | }; |
137 | ||
cc7ec456 | 138 | struct cbq_sched_data { |
d77fea2e | 139 | struct Qdisc_class_hash clhash; /* Hash table of all classes */ |
cc7ec456 ED |
140 | int nclasses[TC_CBQ_MAXPRIO + 1]; |
141 | unsigned int quanta[TC_CBQ_MAXPRIO + 1]; | |
1da177e4 LT |
142 | |
143 | struct cbq_class link; | |
144 | ||
cc7ec456 ED |
145 | unsigned int activemask; |
146 | struct cbq_class *active[TC_CBQ_MAXPRIO + 1]; /* List of all classes | |
1da177e4 LT |
147 | with backlog */ |
148 | ||
c3bc7cff | 149 | #ifdef CONFIG_NET_CLS_ACT |
1da177e4 LT |
150 | struct cbq_class *rx_class; |
151 | #endif | |
152 | struct cbq_class *tx_class; | |
153 | struct cbq_class *tx_borrowed; | |
154 | int tx_len; | |
155 | psched_time_t now; /* Cached timestamp */ | |
cc7ec456 | 156 | unsigned int pmask; |
1da177e4 | 157 | |
2fbd3da3 | 158 | struct hrtimer delay_timer; |
88a99354 | 159 | struct qdisc_watchdog watchdog; /* Watchdog timer, |
1da177e4 LT |
160 | started when CBQ has |
161 | backlog, but cannot | |
162 | transmit just now */ | |
88a99354 | 163 | psched_tdiff_t wd_expires; |
1da177e4 LT |
164 | int toplevel; |
165 | u32 hgenerator; | |
166 | }; | |
167 | ||
168 | ||
cc7ec456 | 169 | #define L2T(cl, len) qdisc_l2t((cl)->R_tab, len) |
1da177e4 | 170 | |
cc7ec456 | 171 | static inline struct cbq_class * |
1da177e4 LT |
172 | cbq_class_lookup(struct cbq_sched_data *q, u32 classid) |
173 | { | |
d77fea2e | 174 | struct Qdisc_class_common *clc; |
1da177e4 | 175 | |
d77fea2e PM |
176 | clc = qdisc_class_find(&q->clhash, classid); |
177 | if (clc == NULL) | |
178 | return NULL; | |
179 | return container_of(clc, struct cbq_class, common); | |
1da177e4 LT |
180 | } |
181 | ||
c3bc7cff | 182 | #ifdef CONFIG_NET_CLS_ACT |
1da177e4 LT |
183 | |
184 | static struct cbq_class * | |
185 | cbq_reclassify(struct sk_buff *skb, struct cbq_class *this) | |
186 | { | |
cc7ec456 | 187 | struct cbq_class *cl; |
1da177e4 | 188 | |
cc7ec456 ED |
189 | for (cl = this->tparent; cl; cl = cl->tparent) { |
190 | struct cbq_class *new = cl->defaults[TC_PRIO_BESTEFFORT]; | |
1da177e4 | 191 | |
cc7ec456 ED |
192 | if (new != NULL && new != this) |
193 | return new; | |
194 | } | |
1da177e4 LT |
195 | return NULL; |
196 | } | |
197 | ||
198 | #endif | |
199 | ||
200 | /* Classify packet. The procedure is pretty complicated, but | |
cc7ec456 ED |
201 | * it allows us to combine link sharing and priority scheduling |
202 | * transparently. | |
203 | * | |
204 | * Namely, you can put link sharing rules (f.e. route based) at root of CBQ, | |
205 | * so that it resolves to split nodes. Then packets are classified | |
206 | * by logical priority, or a more specific classifier may be attached | |
207 | * to the split node. | |
1da177e4 LT |
208 | */ |
209 | ||
210 | static struct cbq_class * | |
211 | cbq_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr) | |
212 | { | |
213 | struct cbq_sched_data *q = qdisc_priv(sch); | |
214 | struct cbq_class *head = &q->link; | |
215 | struct cbq_class **defmap; | |
216 | struct cbq_class *cl = NULL; | |
217 | u32 prio = skb->priority; | |
25d8c0d5 | 218 | struct tcf_proto *fl; |
1da177e4 LT |
219 | struct tcf_result res; |
220 | ||
221 | /* | |
222 | * Step 1. If skb->priority points to one of our classes, use it. | |
223 | */ | |
cc7ec456 | 224 | if (TC_H_MAJ(prio ^ sch->handle) == 0 && |
1da177e4 LT |
225 | (cl = cbq_class_lookup(q, prio)) != NULL) |
226 | return cl; | |
227 | ||
c27f339a | 228 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 LT |
229 | for (;;) { |
230 | int result = 0; | |
231 | defmap = head->defaults; | |
232 | ||
25d8c0d5 | 233 | fl = rcu_dereference_bh(head->filter_list); |
1da177e4 LT |
234 | /* |
235 | * Step 2+n. Apply classifier. | |
236 | */ | |
87d83093 | 237 | result = tcf_classify(skb, fl, &res, true); |
25d8c0d5 | 238 | if (!fl || result < 0) |
1da177e4 LT |
239 | goto fallback; |
240 | ||
cc7ec456 ED |
241 | cl = (void *)res.class; |
242 | if (!cl) { | |
1da177e4 LT |
243 | if (TC_H_MAJ(res.classid)) |
244 | cl = cbq_class_lookup(q, res.classid); | |
cc7ec456 | 245 | else if ((cl = defmap[res.classid & TC_PRIO_MAX]) == NULL) |
1da177e4 LT |
246 | cl = defmap[TC_PRIO_BESTEFFORT]; |
247 | ||
bdfc87f7 | 248 | if (cl == NULL) |
1da177e4 LT |
249 | goto fallback; |
250 | } | |
bdfc87f7 ED |
251 | if (cl->level >= head->level) |
252 | goto fallback; | |
1da177e4 LT |
253 | #ifdef CONFIG_NET_CLS_ACT |
254 | switch (result) { | |
255 | case TC_ACT_QUEUED: | |
10297b99 | 256 | case TC_ACT_STOLEN: |
e25ea21f | 257 | case TC_ACT_TRAP: |
378a2f09 | 258 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
1da177e4 LT |
259 | case TC_ACT_SHOT: |
260 | return NULL; | |
73ca4918 PM |
261 | case TC_ACT_RECLASSIFY: |
262 | return cbq_reclassify(skb, cl); | |
1da177e4 | 263 | } |
1da177e4 LT |
264 | #endif |
265 | if (cl->level == 0) | |
266 | return cl; | |
267 | ||
268 | /* | |
269 | * Step 3+n. If classifier selected a link sharing class, | |
270 | * apply agency specific classifier. | |
271 | * Repeat this procdure until we hit a leaf node. | |
272 | */ | |
273 | head = cl; | |
274 | } | |
275 | ||
276 | fallback: | |
277 | cl = head; | |
278 | ||
279 | /* | |
280 | * Step 4. No success... | |
281 | */ | |
282 | if (TC_H_MAJ(prio) == 0 && | |
cc7ec456 | 283 | !(cl = head->defaults[prio & TC_PRIO_MAX]) && |
1da177e4 LT |
284 | !(cl = head->defaults[TC_PRIO_BESTEFFORT])) |
285 | return head; | |
286 | ||
287 | return cl; | |
288 | } | |
289 | ||
290 | /* | |
cc7ec456 ED |
291 | * A packet has just been enqueued on the empty class. |
292 | * cbq_activate_class adds it to the tail of active class list | |
293 | * of its priority band. | |
1da177e4 LT |
294 | */ |
295 | ||
cc7ec456 | 296 | static inline void cbq_activate_class(struct cbq_class *cl) |
1da177e4 LT |
297 | { |
298 | struct cbq_sched_data *q = qdisc_priv(cl->qdisc); | |
299 | int prio = cl->cpriority; | |
300 | struct cbq_class *cl_tail; | |
301 | ||
302 | cl_tail = q->active[prio]; | |
303 | q->active[prio] = cl; | |
304 | ||
305 | if (cl_tail != NULL) { | |
306 | cl->next_alive = cl_tail->next_alive; | |
307 | cl_tail->next_alive = cl; | |
308 | } else { | |
309 | cl->next_alive = cl; | |
310 | q->activemask |= (1<<prio); | |
311 | } | |
312 | } | |
313 | ||
314 | /* | |
cc7ec456 ED |
315 | * Unlink class from active chain. |
316 | * Note that this same procedure is done directly in cbq_dequeue* | |
317 | * during round-robin procedure. | |
1da177e4 LT |
318 | */ |
319 | ||
320 | static void cbq_deactivate_class(struct cbq_class *this) | |
321 | { | |
322 | struct cbq_sched_data *q = qdisc_priv(this->qdisc); | |
323 | int prio = this->cpriority; | |
324 | struct cbq_class *cl; | |
325 | struct cbq_class *cl_prev = q->active[prio]; | |
326 | ||
327 | do { | |
328 | cl = cl_prev->next_alive; | |
329 | if (cl == this) { | |
330 | cl_prev->next_alive = cl->next_alive; | |
331 | cl->next_alive = NULL; | |
332 | ||
333 | if (cl == q->active[prio]) { | |
334 | q->active[prio] = cl_prev; | |
335 | if (cl == q->active[prio]) { | |
336 | q->active[prio] = NULL; | |
337 | q->activemask &= ~(1<<prio); | |
338 | return; | |
339 | } | |
340 | } | |
1da177e4 LT |
341 | return; |
342 | } | |
343 | } while ((cl_prev = cl) != q->active[prio]); | |
344 | } | |
345 | ||
346 | static void | |
347 | cbq_mark_toplevel(struct cbq_sched_data *q, struct cbq_class *cl) | |
348 | { | |
349 | int toplevel = q->toplevel; | |
350 | ||
cca605dd | 351 | if (toplevel > cl->level) { |
7201c1dd | 352 | psched_time_t now = psched_get_time(); |
1da177e4 LT |
353 | |
354 | do { | |
104e0878 | 355 | if (cl->undertime < now) { |
1da177e4 LT |
356 | q->toplevel = cl->level; |
357 | return; | |
358 | } | |
cc7ec456 | 359 | } while ((cl = cl->borrow) != NULL && toplevel > cl->level); |
1da177e4 LT |
360 | } |
361 | } | |
362 | ||
363 | static int | |
520ac30f ED |
364 | cbq_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
365 | struct sk_buff **to_free) | |
1da177e4 LT |
366 | { |
367 | struct cbq_sched_data *q = qdisc_priv(sch); | |
ddeee3ce | 368 | int uninitialized_var(ret); |
1da177e4 LT |
369 | struct cbq_class *cl = cbq_classify(skb, sch, &ret); |
370 | ||
c3bc7cff | 371 | #ifdef CONFIG_NET_CLS_ACT |
1da177e4 LT |
372 | q->rx_class = cl; |
373 | #endif | |
374 | if (cl == NULL) { | |
c27f339a | 375 | if (ret & __NET_XMIT_BYPASS) |
25331d6c | 376 | qdisc_qstats_drop(sch); |
520ac30f | 377 | __qdisc_drop(skb, to_free); |
1da177e4 LT |
378 | return ret; |
379 | } | |
380 | ||
520ac30f | 381 | ret = qdisc_enqueue(skb, cl->q, to_free); |
5f86173b | 382 | if (ret == NET_XMIT_SUCCESS) { |
1da177e4 | 383 | sch->q.qlen++; |
1da177e4 LT |
384 | cbq_mark_toplevel(q, cl); |
385 | if (!cl->next_alive) | |
386 | cbq_activate_class(cl); | |
387 | return ret; | |
388 | } | |
389 | ||
378a2f09 | 390 | if (net_xmit_drop_count(ret)) { |
25331d6c | 391 | qdisc_qstats_drop(sch); |
378a2f09 JP |
392 | cbq_mark_toplevel(q, cl); |
393 | cl->qstats.drops++; | |
394 | } | |
1da177e4 LT |
395 | return ret; |
396 | } | |
397 | ||
c3498d34 FW |
398 | /* Overlimit action: penalize leaf class by adding offtime */ |
399 | static void cbq_overlimit(struct cbq_class *cl) | |
1da177e4 LT |
400 | { |
401 | struct cbq_sched_data *q = qdisc_priv(cl->qdisc); | |
8edc0c31 | 402 | psched_tdiff_t delay = cl->undertime - q->now; |
1da177e4 LT |
403 | |
404 | if (!cl->delayed) { | |
405 | delay += cl->offtime; | |
406 | ||
10297b99 | 407 | /* |
cc7ec456 ED |
408 | * Class goes to sleep, so that it will have no |
409 | * chance to work avgidle. Let's forgive it 8) | |
410 | * | |
411 | * BTW cbq-2.0 has a crap in this | |
412 | * place, apparently they forgot to shift it by cl->ewma_log. | |
1da177e4 LT |
413 | */ |
414 | if (cl->avgidle < 0) | |
415 | delay -= (-cl->avgidle) - ((-cl->avgidle) >> cl->ewma_log); | |
416 | if (cl->avgidle < cl->minidle) | |
417 | cl->avgidle = cl->minidle; | |
418 | if (delay <= 0) | |
419 | delay = 1; | |
7c59e25f | 420 | cl->undertime = q->now + delay; |
1da177e4 LT |
421 | |
422 | cl->xstats.overactions++; | |
423 | cl->delayed = 1; | |
424 | } | |
425 | if (q->wd_expires == 0 || q->wd_expires > delay) | |
426 | q->wd_expires = delay; | |
427 | ||
428 | /* Dirty work! We must schedule wakeups based on | |
cc7ec456 ED |
429 | * real available rate, rather than leaf rate, |
430 | * which may be tiny (even zero). | |
1da177e4 LT |
431 | */ |
432 | if (q->toplevel == TC_CBQ_MAXLEVEL) { | |
433 | struct cbq_class *b; | |
434 | psched_tdiff_t base_delay = q->wd_expires; | |
435 | ||
436 | for (b = cl->borrow; b; b = b->borrow) { | |
8edc0c31 | 437 | delay = b->undertime - q->now; |
1da177e4 LT |
438 | if (delay < base_delay) { |
439 | if (delay <= 0) | |
440 | delay = 1; | |
441 | base_delay = delay; | |
442 | } | |
443 | } | |
444 | ||
445 | q->wd_expires = base_delay; | |
446 | } | |
447 | } | |
448 | ||
1a13cb63 PM |
449 | static psched_tdiff_t cbq_undelay_prio(struct cbq_sched_data *q, int prio, |
450 | psched_time_t now) | |
1da177e4 LT |
451 | { |
452 | struct cbq_class *cl; | |
453 | struct cbq_class *cl_prev = q->active[prio]; | |
1a13cb63 | 454 | psched_time_t sched = now; |
1da177e4 LT |
455 | |
456 | if (cl_prev == NULL) | |
e9054a33 | 457 | return 0; |
1da177e4 LT |
458 | |
459 | do { | |
460 | cl = cl_prev->next_alive; | |
1a13cb63 | 461 | if (now - cl->penalized > 0) { |
1da177e4 LT |
462 | cl_prev->next_alive = cl->next_alive; |
463 | cl->next_alive = NULL; | |
464 | cl->cpriority = cl->priority; | |
465 | cl->delayed = 0; | |
466 | cbq_activate_class(cl); | |
467 | ||
468 | if (cl == q->active[prio]) { | |
469 | q->active[prio] = cl_prev; | |
470 | if (cl == q->active[prio]) { | |
471 | q->active[prio] = NULL; | |
472 | return 0; | |
473 | } | |
474 | } | |
475 | ||
476 | cl = cl_prev->next_alive; | |
1a13cb63 | 477 | } else if (sched - cl->penalized > 0) |
1da177e4 LT |
478 | sched = cl->penalized; |
479 | } while ((cl_prev = cl) != q->active[prio]); | |
480 | ||
1a13cb63 | 481 | return sched - now; |
1da177e4 LT |
482 | } |
483 | ||
1a13cb63 | 484 | static enum hrtimer_restart cbq_undelay(struct hrtimer *timer) |
1da177e4 | 485 | { |
1a13cb63 | 486 | struct cbq_sched_data *q = container_of(timer, struct cbq_sched_data, |
2fbd3da3 | 487 | delay_timer); |
1a13cb63 PM |
488 | struct Qdisc *sch = q->watchdog.qdisc; |
489 | psched_time_t now; | |
490 | psched_tdiff_t delay = 0; | |
cc7ec456 | 491 | unsigned int pmask; |
1da177e4 | 492 | |
3bebcda2 | 493 | now = psched_get_time(); |
1a13cb63 | 494 | |
1da177e4 LT |
495 | pmask = q->pmask; |
496 | q->pmask = 0; | |
497 | ||
498 | while (pmask) { | |
499 | int prio = ffz(~pmask); | |
1a13cb63 | 500 | psched_tdiff_t tmp; |
1da177e4 LT |
501 | |
502 | pmask &= ~(1<<prio); | |
503 | ||
1a13cb63 | 504 | tmp = cbq_undelay_prio(q, prio, now); |
1da177e4 LT |
505 | if (tmp > 0) { |
506 | q->pmask |= 1<<prio; | |
507 | if (tmp < delay || delay == 0) | |
508 | delay = tmp; | |
509 | } | |
510 | } | |
511 | ||
512 | if (delay) { | |
1a13cb63 PM |
513 | ktime_t time; |
514 | ||
8b0e1953 | 515 | time = 0; |
ca44d6e6 | 516 | time = ktime_add_ns(time, PSCHED_TICKS2NS(now + delay)); |
4a8e320c | 517 | hrtimer_start(&q->delay_timer, time, HRTIMER_MODE_ABS_PINNED); |
1da177e4 LT |
518 | } |
519 | ||
8608db03 | 520 | __netif_schedule(qdisc_root(sch)); |
1a13cb63 | 521 | return HRTIMER_NORESTART; |
1da177e4 LT |
522 | } |
523 | ||
10297b99 | 524 | /* |
cc7ec456 ED |
525 | * It is mission critical procedure. |
526 | * | |
527 | * We "regenerate" toplevel cutoff, if transmitting class | |
528 | * has backlog and it is not regulated. It is not part of | |
529 | * original CBQ description, but looks more reasonable. | |
530 | * Probably, it is wrong. This question needs further investigation. | |
531 | */ | |
1da177e4 | 532 | |
cc7ec456 | 533 | static inline void |
1da177e4 LT |
534 | cbq_update_toplevel(struct cbq_sched_data *q, struct cbq_class *cl, |
535 | struct cbq_class *borrowed) | |
536 | { | |
537 | if (cl && q->toplevel >= borrowed->level) { | |
538 | if (cl->q->q.qlen > 1) { | |
539 | do { | |
a084980d | 540 | if (borrowed->undertime == PSCHED_PASTPERFECT) { |
1da177e4 LT |
541 | q->toplevel = borrowed->level; |
542 | return; | |
543 | } | |
cc7ec456 | 544 | } while ((borrowed = borrowed->borrow) != NULL); |
1da177e4 | 545 | } |
10297b99 | 546 | #if 0 |
1da177e4 LT |
547 | /* It is not necessary now. Uncommenting it |
548 | will save CPU cycles, but decrease fairness. | |
549 | */ | |
550 | q->toplevel = TC_CBQ_MAXLEVEL; | |
551 | #endif | |
552 | } | |
553 | } | |
554 | ||
555 | static void | |
556 | cbq_update(struct cbq_sched_data *q) | |
557 | { | |
558 | struct cbq_class *this = q->tx_class; | |
559 | struct cbq_class *cl = this; | |
560 | int len = q->tx_len; | |
73d0f37a | 561 | psched_time_t now; |
1da177e4 LT |
562 | |
563 | q->tx_class = NULL; | |
73d0f37a VA |
564 | /* Time integrator. We calculate EOS time |
565 | * by adding expected packet transmission time. | |
566 | */ | |
567 | now = q->now + L2T(&q->link, len); | |
1da177e4 LT |
568 | |
569 | for ( ; cl; cl = cl->share) { | |
570 | long avgidle = cl->avgidle; | |
571 | long idle; | |
572 | ||
573 | cl->bstats.packets++; | |
574 | cl->bstats.bytes += len; | |
575 | ||
576 | /* | |
cc7ec456 ED |
577 | * (now - last) is total time between packet right edges. |
578 | * (last_pktlen/rate) is "virtual" busy time, so that | |
579 | * | |
580 | * idle = (now - last) - last_pktlen/rate | |
1da177e4 LT |
581 | */ |
582 | ||
73d0f37a | 583 | idle = now - cl->last; |
1da177e4 LT |
584 | if ((unsigned long)idle > 128*1024*1024) { |
585 | avgidle = cl->maxidle; | |
586 | } else { | |
587 | idle -= L2T(cl, len); | |
588 | ||
589 | /* true_avgidle := (1-W)*true_avgidle + W*idle, | |
cc7ec456 ED |
590 | * where W=2^{-ewma_log}. But cl->avgidle is scaled: |
591 | * cl->avgidle == true_avgidle/W, | |
592 | * hence: | |
1da177e4 LT |
593 | */ |
594 | avgidle += idle - (avgidle>>cl->ewma_log); | |
595 | } | |
596 | ||
597 | if (avgidle <= 0) { | |
598 | /* Overlimit or at-limit */ | |
599 | ||
600 | if (avgidle < cl->minidle) | |
601 | avgidle = cl->minidle; | |
602 | ||
603 | cl->avgidle = avgidle; | |
604 | ||
605 | /* Calculate expected time, when this class | |
cc7ec456 ED |
606 | * will be allowed to send. |
607 | * It will occur, when: | |
608 | * (1-W)*true_avgidle + W*delay = 0, i.e. | |
609 | * idle = (1/W - 1)*(-true_avgidle) | |
610 | * or | |
611 | * idle = (1 - W)*(-cl->avgidle); | |
1da177e4 LT |
612 | */ |
613 | idle = (-avgidle) - ((-avgidle) >> cl->ewma_log); | |
614 | ||
615 | /* | |
cc7ec456 ED |
616 | * That is not all. |
617 | * To maintain the rate allocated to the class, | |
618 | * we add to undertime virtual clock, | |
619 | * necessary to complete transmitted packet. | |
620 | * (len/phys_bandwidth has been already passed | |
621 | * to the moment of cbq_update) | |
1da177e4 LT |
622 | */ |
623 | ||
624 | idle -= L2T(&q->link, len); | |
625 | idle += L2T(cl, len); | |
626 | ||
73d0f37a | 627 | cl->undertime = now + idle; |
1da177e4 LT |
628 | } else { |
629 | /* Underlimit */ | |
630 | ||
a084980d | 631 | cl->undertime = PSCHED_PASTPERFECT; |
1da177e4 LT |
632 | if (avgidle > cl->maxidle) |
633 | cl->avgidle = cl->maxidle; | |
634 | else | |
635 | cl->avgidle = avgidle; | |
636 | } | |
73d0f37a VA |
637 | if ((s64)(now - cl->last) > 0) |
638 | cl->last = now; | |
1da177e4 LT |
639 | } |
640 | ||
641 | cbq_update_toplevel(q, this, q->tx_borrowed); | |
642 | } | |
643 | ||
cc7ec456 | 644 | static inline struct cbq_class * |
1da177e4 LT |
645 | cbq_under_limit(struct cbq_class *cl) |
646 | { | |
647 | struct cbq_sched_data *q = qdisc_priv(cl->qdisc); | |
648 | struct cbq_class *this_cl = cl; | |
649 | ||
650 | if (cl->tparent == NULL) | |
651 | return cl; | |
652 | ||
a084980d | 653 | if (cl->undertime == PSCHED_PASTPERFECT || q->now >= cl->undertime) { |
1da177e4 LT |
654 | cl->delayed = 0; |
655 | return cl; | |
656 | } | |
657 | ||
658 | do { | |
659 | /* It is very suspicious place. Now overlimit | |
cc7ec456 ED |
660 | * action is generated for not bounded classes |
661 | * only if link is completely congested. | |
662 | * Though it is in agree with ancestor-only paradigm, | |
663 | * it looks very stupid. Particularly, | |
664 | * it means that this chunk of code will either | |
665 | * never be called or result in strong amplification | |
666 | * of burstiness. Dangerous, silly, and, however, | |
667 | * no another solution exists. | |
1da177e4 | 668 | */ |
cc7ec456 ED |
669 | cl = cl->borrow; |
670 | if (!cl) { | |
1da177e4 | 671 | this_cl->qstats.overlimits++; |
c3498d34 | 672 | cbq_overlimit(this_cl); |
1da177e4 LT |
673 | return NULL; |
674 | } | |
675 | if (cl->level > q->toplevel) | |
676 | return NULL; | |
a084980d | 677 | } while (cl->undertime != PSCHED_PASTPERFECT && q->now < cl->undertime); |
1da177e4 LT |
678 | |
679 | cl->delayed = 0; | |
680 | return cl; | |
681 | } | |
682 | ||
cc7ec456 | 683 | static inline struct sk_buff * |
1da177e4 LT |
684 | cbq_dequeue_prio(struct Qdisc *sch, int prio) |
685 | { | |
686 | struct cbq_sched_data *q = qdisc_priv(sch); | |
687 | struct cbq_class *cl_tail, *cl_prev, *cl; | |
688 | struct sk_buff *skb; | |
689 | int deficit; | |
690 | ||
691 | cl_tail = cl_prev = q->active[prio]; | |
692 | cl = cl_prev->next_alive; | |
693 | ||
694 | do { | |
695 | deficit = 0; | |
696 | ||
697 | /* Start round */ | |
698 | do { | |
699 | struct cbq_class *borrow = cl; | |
700 | ||
701 | if (cl->q->q.qlen && | |
702 | (borrow = cbq_under_limit(cl)) == NULL) | |
703 | goto skip_class; | |
704 | ||
705 | if (cl->deficit <= 0) { | |
706 | /* Class exhausted its allotment per | |
cc7ec456 | 707 | * this round. Switch to the next one. |
1da177e4 LT |
708 | */ |
709 | deficit = 1; | |
710 | cl->deficit += cl->quantum; | |
711 | goto next_class; | |
712 | } | |
713 | ||
714 | skb = cl->q->dequeue(cl->q); | |
715 | ||
716 | /* Class did not give us any skb :-( | |
cc7ec456 ED |
717 | * It could occur even if cl->q->q.qlen != 0 |
718 | * f.e. if cl->q == "tbf" | |
1da177e4 LT |
719 | */ |
720 | if (skb == NULL) | |
721 | goto skip_class; | |
722 | ||
0abf77e5 | 723 | cl->deficit -= qdisc_pkt_len(skb); |
1da177e4 LT |
724 | q->tx_class = cl; |
725 | q->tx_borrowed = borrow; | |
726 | if (borrow != cl) { | |
727 | #ifndef CBQ_XSTATS_BORROWS_BYTES | |
728 | borrow->xstats.borrows++; | |
729 | cl->xstats.borrows++; | |
730 | #else | |
0abf77e5 JK |
731 | borrow->xstats.borrows += qdisc_pkt_len(skb); |
732 | cl->xstats.borrows += qdisc_pkt_len(skb); | |
1da177e4 LT |
733 | #endif |
734 | } | |
0abf77e5 | 735 | q->tx_len = qdisc_pkt_len(skb); |
1da177e4 LT |
736 | |
737 | if (cl->deficit <= 0) { | |
738 | q->active[prio] = cl; | |
739 | cl = cl->next_alive; | |
740 | cl->deficit += cl->quantum; | |
741 | } | |
742 | return skb; | |
743 | ||
744 | skip_class: | |
745 | if (cl->q->q.qlen == 0 || prio != cl->cpriority) { | |
746 | /* Class is empty or penalized. | |
cc7ec456 | 747 | * Unlink it from active chain. |
1da177e4 LT |
748 | */ |
749 | cl_prev->next_alive = cl->next_alive; | |
750 | cl->next_alive = NULL; | |
751 | ||
752 | /* Did cl_tail point to it? */ | |
753 | if (cl == cl_tail) { | |
754 | /* Repair it! */ | |
755 | cl_tail = cl_prev; | |
756 | ||
757 | /* Was it the last class in this band? */ | |
758 | if (cl == cl_tail) { | |
759 | /* Kill the band! */ | |
760 | q->active[prio] = NULL; | |
761 | q->activemask &= ~(1<<prio); | |
762 | if (cl->q->q.qlen) | |
763 | cbq_activate_class(cl); | |
764 | return NULL; | |
765 | } | |
766 | ||
767 | q->active[prio] = cl_tail; | |
768 | } | |
769 | if (cl->q->q.qlen) | |
770 | cbq_activate_class(cl); | |
771 | ||
772 | cl = cl_prev; | |
773 | } | |
774 | ||
775 | next_class: | |
776 | cl_prev = cl; | |
777 | cl = cl->next_alive; | |
778 | } while (cl_prev != cl_tail); | |
779 | } while (deficit); | |
780 | ||
781 | q->active[prio] = cl_prev; | |
782 | ||
783 | return NULL; | |
784 | } | |
785 | ||
cc7ec456 | 786 | static inline struct sk_buff * |
1da177e4 LT |
787 | cbq_dequeue_1(struct Qdisc *sch) |
788 | { | |
789 | struct cbq_sched_data *q = qdisc_priv(sch); | |
790 | struct sk_buff *skb; | |
cc7ec456 | 791 | unsigned int activemask; |
1da177e4 | 792 | |
cc7ec456 | 793 | activemask = q->activemask & 0xFF; |
1da177e4 LT |
794 | while (activemask) { |
795 | int prio = ffz(~activemask); | |
796 | activemask &= ~(1<<prio); | |
797 | skb = cbq_dequeue_prio(sch, prio); | |
798 | if (skb) | |
799 | return skb; | |
800 | } | |
801 | return NULL; | |
802 | } | |
803 | ||
804 | static struct sk_buff * | |
805 | cbq_dequeue(struct Qdisc *sch) | |
806 | { | |
807 | struct sk_buff *skb; | |
808 | struct cbq_sched_data *q = qdisc_priv(sch); | |
809 | psched_time_t now; | |
1da177e4 | 810 | |
3bebcda2 | 811 | now = psched_get_time(); |
73d0f37a VA |
812 | |
813 | if (q->tx_class) | |
1da177e4 | 814 | cbq_update(q); |
73d0f37a VA |
815 | |
816 | q->now = now; | |
1da177e4 LT |
817 | |
818 | for (;;) { | |
819 | q->wd_expires = 0; | |
820 | ||
821 | skb = cbq_dequeue_1(sch); | |
822 | if (skb) { | |
9190b3b3 | 823 | qdisc_bstats_update(sch, skb); |
1da177e4 | 824 | sch->q.qlen--; |
1da177e4 LT |
825 | return skb; |
826 | } | |
827 | ||
828 | /* All the classes are overlimit. | |
cc7ec456 ED |
829 | * |
830 | * It is possible, if: | |
831 | * | |
832 | * 1. Scheduler is empty. | |
833 | * 2. Toplevel cutoff inhibited borrowing. | |
834 | * 3. Root class is overlimit. | |
835 | * | |
836 | * Reset 2d and 3d conditions and retry. | |
837 | * | |
838 | * Note, that NS and cbq-2.0 are buggy, peeking | |
839 | * an arbitrary class is appropriate for ancestor-only | |
840 | * sharing, but not for toplevel algorithm. | |
841 | * | |
842 | * Our version is better, but slower, because it requires | |
843 | * two passes, but it is unavoidable with top-level sharing. | |
844 | */ | |
1da177e4 LT |
845 | |
846 | if (q->toplevel == TC_CBQ_MAXLEVEL && | |
a084980d | 847 | q->link.undertime == PSCHED_PASTPERFECT) |
1da177e4 LT |
848 | break; |
849 | ||
850 | q->toplevel = TC_CBQ_MAXLEVEL; | |
a084980d | 851 | q->link.undertime = PSCHED_PASTPERFECT; |
1da177e4 LT |
852 | } |
853 | ||
854 | /* No packets in scheduler or nobody wants to give them to us :-( | |
cc7ec456 ED |
855 | * Sigh... start watchdog timer in the last case. |
856 | */ | |
1da177e4 LT |
857 | |
858 | if (sch->q.qlen) { | |
25331d6c | 859 | qdisc_qstats_overlimit(sch); |
88a99354 PM |
860 | if (q->wd_expires) |
861 | qdisc_watchdog_schedule(&q->watchdog, | |
bb239acf | 862 | now + q->wd_expires); |
1da177e4 LT |
863 | } |
864 | return NULL; | |
865 | } | |
866 | ||
867 | /* CBQ class maintanance routines */ | |
868 | ||
869 | static void cbq_adjust_levels(struct cbq_class *this) | |
870 | { | |
871 | if (this == NULL) | |
872 | return; | |
873 | ||
874 | do { | |
875 | int level = 0; | |
876 | struct cbq_class *cl; | |
877 | ||
cc7ec456 ED |
878 | cl = this->children; |
879 | if (cl) { | |
1da177e4 LT |
880 | do { |
881 | if (cl->level > level) | |
882 | level = cl->level; | |
883 | } while ((cl = cl->sibling) != this->children); | |
884 | } | |
cc7ec456 | 885 | this->level = level + 1; |
1da177e4 LT |
886 | } while ((this = this->tparent) != NULL); |
887 | } | |
888 | ||
889 | static void cbq_normalize_quanta(struct cbq_sched_data *q, int prio) | |
890 | { | |
891 | struct cbq_class *cl; | |
d77fea2e | 892 | unsigned int h; |
1da177e4 LT |
893 | |
894 | if (q->quanta[prio] == 0) | |
895 | return; | |
896 | ||
d77fea2e | 897 | for (h = 0; h < q->clhash.hashsize; h++) { |
b67bfe0d | 898 | hlist_for_each_entry(cl, &q->clhash.hash[h], common.hnode) { |
1da177e4 | 899 | /* BUGGGG... Beware! This expression suffer of |
cc7ec456 | 900 | * arithmetic overflows! |
1da177e4 LT |
901 | */ |
902 | if (cl->priority == prio) { | |
903 | cl->quantum = (cl->weight*cl->allot*q->nclasses[prio])/ | |
904 | q->quanta[prio]; | |
905 | } | |
833fa743 YY |
906 | if (cl->quantum <= 0 || |
907 | cl->quantum > 32*qdisc_dev(cl->qdisc)->mtu) { | |
c17988a9 YY |
908 | pr_warn("CBQ: class %08x has bad quantum==%ld, repaired.\n", |
909 | cl->common.classid, cl->quantum); | |
5ce2d488 | 910 | cl->quantum = qdisc_dev(cl->qdisc)->mtu/2 + 1; |
1da177e4 LT |
911 | } |
912 | } | |
913 | } | |
914 | } | |
915 | ||
916 | static void cbq_sync_defmap(struct cbq_class *cl) | |
917 | { | |
918 | struct cbq_sched_data *q = qdisc_priv(cl->qdisc); | |
919 | struct cbq_class *split = cl->split; | |
cc7ec456 | 920 | unsigned int h; |
1da177e4 LT |
921 | int i; |
922 | ||
923 | if (split == NULL) | |
924 | return; | |
925 | ||
cc7ec456 ED |
926 | for (i = 0; i <= TC_PRIO_MAX; i++) { |
927 | if (split->defaults[i] == cl && !(cl->defmap & (1<<i))) | |
1da177e4 LT |
928 | split->defaults[i] = NULL; |
929 | } | |
930 | ||
cc7ec456 | 931 | for (i = 0; i <= TC_PRIO_MAX; i++) { |
1da177e4 LT |
932 | int level = split->level; |
933 | ||
934 | if (split->defaults[i]) | |
935 | continue; | |
936 | ||
d77fea2e | 937 | for (h = 0; h < q->clhash.hashsize; h++) { |
1da177e4 LT |
938 | struct cbq_class *c; |
939 | ||
b67bfe0d | 940 | hlist_for_each_entry(c, &q->clhash.hash[h], |
d77fea2e | 941 | common.hnode) { |
1da177e4 | 942 | if (c->split == split && c->level < level && |
cc7ec456 | 943 | c->defmap & (1<<i)) { |
1da177e4 LT |
944 | split->defaults[i] = c; |
945 | level = c->level; | |
946 | } | |
947 | } | |
948 | } | |
949 | } | |
950 | } | |
951 | ||
952 | static void cbq_change_defmap(struct cbq_class *cl, u32 splitid, u32 def, u32 mask) | |
953 | { | |
954 | struct cbq_class *split = NULL; | |
955 | ||
956 | if (splitid == 0) { | |
cc7ec456 ED |
957 | split = cl->split; |
958 | if (!split) | |
1da177e4 | 959 | return; |
d77fea2e | 960 | splitid = split->common.classid; |
1da177e4 LT |
961 | } |
962 | ||
d77fea2e | 963 | if (split == NULL || split->common.classid != splitid) { |
1da177e4 | 964 | for (split = cl->tparent; split; split = split->tparent) |
d77fea2e | 965 | if (split->common.classid == splitid) |
1da177e4 LT |
966 | break; |
967 | } | |
968 | ||
969 | if (split == NULL) | |
970 | return; | |
971 | ||
972 | if (cl->split != split) { | |
973 | cl->defmap = 0; | |
974 | cbq_sync_defmap(cl); | |
975 | cl->split = split; | |
cc7ec456 | 976 | cl->defmap = def & mask; |
1da177e4 | 977 | } else |
cc7ec456 | 978 | cl->defmap = (cl->defmap & ~mask) | (def & mask); |
1da177e4 LT |
979 | |
980 | cbq_sync_defmap(cl); | |
981 | } | |
982 | ||
983 | static void cbq_unlink_class(struct cbq_class *this) | |
984 | { | |
985 | struct cbq_class *cl, **clp; | |
986 | struct cbq_sched_data *q = qdisc_priv(this->qdisc); | |
987 | ||
d77fea2e | 988 | qdisc_class_hash_remove(&q->clhash, &this->common); |
1da177e4 LT |
989 | |
990 | if (this->tparent) { | |
cc7ec456 | 991 | clp = &this->sibling; |
1da177e4 LT |
992 | cl = *clp; |
993 | do { | |
994 | if (cl == this) { | |
995 | *clp = cl->sibling; | |
996 | break; | |
997 | } | |
998 | clp = &cl->sibling; | |
999 | } while ((cl = *clp) != this->sibling); | |
1000 | ||
1001 | if (this->tparent->children == this) { | |
1002 | this->tparent->children = this->sibling; | |
1003 | if (this->sibling == this) | |
1004 | this->tparent->children = NULL; | |
1005 | } | |
1006 | } else { | |
547b792c | 1007 | WARN_ON(this->sibling != this); |
1da177e4 LT |
1008 | } |
1009 | } | |
1010 | ||
1011 | static void cbq_link_class(struct cbq_class *this) | |
1012 | { | |
1013 | struct cbq_sched_data *q = qdisc_priv(this->qdisc); | |
1da177e4 LT |
1014 | struct cbq_class *parent = this->tparent; |
1015 | ||
1016 | this->sibling = this; | |
d77fea2e | 1017 | qdisc_class_hash_insert(&q->clhash, &this->common); |
1da177e4 LT |
1018 | |
1019 | if (parent == NULL) | |
1020 | return; | |
1021 | ||
1022 | if (parent->children == NULL) { | |
1023 | parent->children = this; | |
1024 | } else { | |
1025 | this->sibling = parent->children->sibling; | |
1026 | parent->children->sibling = this; | |
1027 | } | |
1028 | } | |
1029 | ||
1da177e4 | 1030 | static void |
cc7ec456 | 1031 | cbq_reset(struct Qdisc *sch) |
1da177e4 LT |
1032 | { |
1033 | struct cbq_sched_data *q = qdisc_priv(sch); | |
1034 | struct cbq_class *cl; | |
1035 | int prio; | |
cc7ec456 | 1036 | unsigned int h; |
1da177e4 LT |
1037 | |
1038 | q->activemask = 0; | |
1039 | q->pmask = 0; | |
1040 | q->tx_class = NULL; | |
1041 | q->tx_borrowed = NULL; | |
88a99354 | 1042 | qdisc_watchdog_cancel(&q->watchdog); |
2fbd3da3 | 1043 | hrtimer_cancel(&q->delay_timer); |
1da177e4 | 1044 | q->toplevel = TC_CBQ_MAXLEVEL; |
3bebcda2 | 1045 | q->now = psched_get_time(); |
1da177e4 LT |
1046 | |
1047 | for (prio = 0; prio <= TC_CBQ_MAXPRIO; prio++) | |
1048 | q->active[prio] = NULL; | |
1049 | ||
d77fea2e | 1050 | for (h = 0; h < q->clhash.hashsize; h++) { |
b67bfe0d | 1051 | hlist_for_each_entry(cl, &q->clhash.hash[h], common.hnode) { |
1da177e4 LT |
1052 | qdisc_reset(cl->q); |
1053 | ||
1054 | cl->next_alive = NULL; | |
a084980d | 1055 | cl->undertime = PSCHED_PASTPERFECT; |
1da177e4 LT |
1056 | cl->avgidle = cl->maxidle; |
1057 | cl->deficit = cl->quantum; | |
1058 | cl->cpriority = cl->priority; | |
1059 | } | |
1060 | } | |
1061 | sch->q.qlen = 0; | |
1062 | } | |
1063 | ||
1064 | ||
1065 | static int cbq_set_lss(struct cbq_class *cl, struct tc_cbq_lssopt *lss) | |
1066 | { | |
cc7ec456 ED |
1067 | if (lss->change & TCF_CBQ_LSS_FLAGS) { |
1068 | cl->share = (lss->flags & TCF_CBQ_LSS_ISOLATED) ? NULL : cl->tparent; | |
1069 | cl->borrow = (lss->flags & TCF_CBQ_LSS_BOUNDED) ? NULL : cl->tparent; | |
1da177e4 | 1070 | } |
cc7ec456 | 1071 | if (lss->change & TCF_CBQ_LSS_EWMA) |
1da177e4 | 1072 | cl->ewma_log = lss->ewma_log; |
cc7ec456 | 1073 | if (lss->change & TCF_CBQ_LSS_AVPKT) |
1da177e4 | 1074 | cl->avpkt = lss->avpkt; |
cc7ec456 | 1075 | if (lss->change & TCF_CBQ_LSS_MINIDLE) |
1da177e4 | 1076 | cl->minidle = -(long)lss->minidle; |
cc7ec456 | 1077 | if (lss->change & TCF_CBQ_LSS_MAXIDLE) { |
1da177e4 LT |
1078 | cl->maxidle = lss->maxidle; |
1079 | cl->avgidle = lss->maxidle; | |
1080 | } | |
cc7ec456 | 1081 | if (lss->change & TCF_CBQ_LSS_OFFTIME) |
1da177e4 LT |
1082 | cl->offtime = lss->offtime; |
1083 | return 0; | |
1084 | } | |
1085 | ||
1086 | static void cbq_rmprio(struct cbq_sched_data *q, struct cbq_class *cl) | |
1087 | { | |
1088 | q->nclasses[cl->priority]--; | |
1089 | q->quanta[cl->priority] -= cl->weight; | |
1090 | cbq_normalize_quanta(q, cl->priority); | |
1091 | } | |
1092 | ||
1093 | static void cbq_addprio(struct cbq_sched_data *q, struct cbq_class *cl) | |
1094 | { | |
1095 | q->nclasses[cl->priority]++; | |
1096 | q->quanta[cl->priority] += cl->weight; | |
1097 | cbq_normalize_quanta(q, cl->priority); | |
1098 | } | |
1099 | ||
1100 | static int cbq_set_wrr(struct cbq_class *cl, struct tc_cbq_wrropt *wrr) | |
1101 | { | |
1102 | struct cbq_sched_data *q = qdisc_priv(cl->qdisc); | |
1103 | ||
1104 | if (wrr->allot) | |
1105 | cl->allot = wrr->allot; | |
1106 | if (wrr->weight) | |
1107 | cl->weight = wrr->weight; | |
1108 | if (wrr->priority) { | |
cc7ec456 | 1109 | cl->priority = wrr->priority - 1; |
1da177e4 LT |
1110 | cl->cpriority = cl->priority; |
1111 | if (cl->priority >= cl->priority2) | |
cc7ec456 | 1112 | cl->priority2 = TC_CBQ_MAXPRIO - 1; |
1da177e4 LT |
1113 | } |
1114 | ||
1115 | cbq_addprio(q, cl); | |
1116 | return 0; | |
1117 | } | |
1118 | ||
1da177e4 LT |
1119 | static int cbq_set_fopt(struct cbq_class *cl, struct tc_cbq_fopt *fopt) |
1120 | { | |
1121 | cbq_change_defmap(cl, fopt->split, fopt->defmap, fopt->defchange); | |
1122 | return 0; | |
1123 | } | |
1124 | ||
27a3421e PM |
1125 | static const struct nla_policy cbq_policy[TCA_CBQ_MAX + 1] = { |
1126 | [TCA_CBQ_LSSOPT] = { .len = sizeof(struct tc_cbq_lssopt) }, | |
1127 | [TCA_CBQ_WRROPT] = { .len = sizeof(struct tc_cbq_wrropt) }, | |
1128 | [TCA_CBQ_FOPT] = { .len = sizeof(struct tc_cbq_fopt) }, | |
1129 | [TCA_CBQ_OVL_STRATEGY] = { .len = sizeof(struct tc_cbq_ovl) }, | |
1130 | [TCA_CBQ_RATE] = { .len = sizeof(struct tc_ratespec) }, | |
1131 | [TCA_CBQ_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
1132 | [TCA_CBQ_POLICE] = { .len = sizeof(struct tc_cbq_police) }, | |
1133 | }; | |
1134 | ||
1e90474c | 1135 | static int cbq_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
1136 | { |
1137 | struct cbq_sched_data *q = qdisc_priv(sch); | |
1e90474c | 1138 | struct nlattr *tb[TCA_CBQ_MAX + 1]; |
1da177e4 | 1139 | struct tc_ratespec *r; |
cee63723 PM |
1140 | int err; |
1141 | ||
fceb6435 | 1142 | err = nla_parse_nested(tb, TCA_CBQ_MAX, opt, cbq_policy, NULL); |
cee63723 PM |
1143 | if (err < 0) |
1144 | return err; | |
1da177e4 | 1145 | |
27a3421e | 1146 | if (tb[TCA_CBQ_RTAB] == NULL || tb[TCA_CBQ_RATE] == NULL) |
1da177e4 LT |
1147 | return -EINVAL; |
1148 | ||
1e90474c | 1149 | r = nla_data(tb[TCA_CBQ_RATE]); |
1da177e4 | 1150 | |
1e90474c | 1151 | if ((q->link.R_tab = qdisc_get_rtab(r, tb[TCA_CBQ_RTAB])) == NULL) |
1da177e4 LT |
1152 | return -EINVAL; |
1153 | ||
d77fea2e PM |
1154 | err = qdisc_class_hash_init(&q->clhash); |
1155 | if (err < 0) | |
1156 | goto put_rtab; | |
1157 | ||
1da177e4 LT |
1158 | q->link.refcnt = 1; |
1159 | q->link.sibling = &q->link; | |
d77fea2e | 1160 | q->link.common.classid = sch->handle; |
1da177e4 | 1161 | q->link.qdisc = sch; |
3511c913 CG |
1162 | q->link.q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
1163 | sch->handle); | |
1164 | if (!q->link.q) | |
1da177e4 | 1165 | q->link.q = &noop_qdisc; |
49b49971 JK |
1166 | else |
1167 | qdisc_hash_add(q->link.q, true); | |
1da177e4 | 1168 | |
cc7ec456 ED |
1169 | q->link.priority = TC_CBQ_MAXPRIO - 1; |
1170 | q->link.priority2 = TC_CBQ_MAXPRIO - 1; | |
1171 | q->link.cpriority = TC_CBQ_MAXPRIO - 1; | |
5ce2d488 | 1172 | q->link.allot = psched_mtu(qdisc_dev(sch)); |
1da177e4 LT |
1173 | q->link.quantum = q->link.allot; |
1174 | q->link.weight = q->link.R_tab->rate.rate; | |
1175 | ||
1176 | q->link.ewma_log = TC_CBQ_DEF_EWMA; | |
1177 | q->link.avpkt = q->link.allot/2; | |
1178 | q->link.minidle = -0x7FFFFFFF; | |
1da177e4 | 1179 | |
88a99354 | 1180 | qdisc_watchdog_init(&q->watchdog, sch); |
4a8e320c | 1181 | hrtimer_init(&q->delay_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED); |
1da177e4 LT |
1182 | q->delay_timer.function = cbq_undelay; |
1183 | q->toplevel = TC_CBQ_MAXLEVEL; | |
3bebcda2 | 1184 | q->now = psched_get_time(); |
1da177e4 LT |
1185 | |
1186 | cbq_link_class(&q->link); | |
1187 | ||
1e90474c PM |
1188 | if (tb[TCA_CBQ_LSSOPT]) |
1189 | cbq_set_lss(&q->link, nla_data(tb[TCA_CBQ_LSSOPT])); | |
1da177e4 LT |
1190 | |
1191 | cbq_addprio(q, &q->link); | |
1192 | return 0; | |
d77fea2e PM |
1193 | |
1194 | put_rtab: | |
1195 | qdisc_put_rtab(q->link.R_tab); | |
1196 | return err; | |
1da177e4 LT |
1197 | } |
1198 | ||
cc7ec456 | 1199 | static int cbq_dump_rate(struct sk_buff *skb, struct cbq_class *cl) |
1da177e4 | 1200 | { |
27a884dc | 1201 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 | 1202 | |
1b34ec43 DM |
1203 | if (nla_put(skb, TCA_CBQ_RATE, sizeof(cl->R_tab->rate), &cl->R_tab->rate)) |
1204 | goto nla_put_failure; | |
1da177e4 LT |
1205 | return skb->len; |
1206 | ||
1e90474c | 1207 | nla_put_failure: |
dc5fc579 | 1208 | nlmsg_trim(skb, b); |
1da177e4 LT |
1209 | return -1; |
1210 | } | |
1211 | ||
cc7ec456 | 1212 | static int cbq_dump_lss(struct sk_buff *skb, struct cbq_class *cl) |
1da177e4 | 1213 | { |
27a884dc | 1214 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
1215 | struct tc_cbq_lssopt opt; |
1216 | ||
1217 | opt.flags = 0; | |
1218 | if (cl->borrow == NULL) | |
1219 | opt.flags |= TCF_CBQ_LSS_BOUNDED; | |
1220 | if (cl->share == NULL) | |
1221 | opt.flags |= TCF_CBQ_LSS_ISOLATED; | |
1222 | opt.ewma_log = cl->ewma_log; | |
1223 | opt.level = cl->level; | |
1224 | opt.avpkt = cl->avpkt; | |
1225 | opt.maxidle = cl->maxidle; | |
1226 | opt.minidle = (u32)(-cl->minidle); | |
1227 | opt.offtime = cl->offtime; | |
1228 | opt.change = ~0; | |
1b34ec43 DM |
1229 | if (nla_put(skb, TCA_CBQ_LSSOPT, sizeof(opt), &opt)) |
1230 | goto nla_put_failure; | |
1da177e4 LT |
1231 | return skb->len; |
1232 | ||
1e90474c | 1233 | nla_put_failure: |
dc5fc579 | 1234 | nlmsg_trim(skb, b); |
1da177e4 LT |
1235 | return -1; |
1236 | } | |
1237 | ||
cc7ec456 | 1238 | static int cbq_dump_wrr(struct sk_buff *skb, struct cbq_class *cl) |
1da177e4 | 1239 | { |
27a884dc | 1240 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
1241 | struct tc_cbq_wrropt opt; |
1242 | ||
a0db856a | 1243 | memset(&opt, 0, sizeof(opt)); |
1da177e4 LT |
1244 | opt.flags = 0; |
1245 | opt.allot = cl->allot; | |
cc7ec456 ED |
1246 | opt.priority = cl->priority + 1; |
1247 | opt.cpriority = cl->cpriority + 1; | |
1da177e4 | 1248 | opt.weight = cl->weight; |
1b34ec43 DM |
1249 | if (nla_put(skb, TCA_CBQ_WRROPT, sizeof(opt), &opt)) |
1250 | goto nla_put_failure; | |
1da177e4 LT |
1251 | return skb->len; |
1252 | ||
1e90474c | 1253 | nla_put_failure: |
dc5fc579 | 1254 | nlmsg_trim(skb, b); |
1da177e4 LT |
1255 | return -1; |
1256 | } | |
1257 | ||
cc7ec456 | 1258 | static int cbq_dump_fopt(struct sk_buff *skb, struct cbq_class *cl) |
1da177e4 | 1259 | { |
27a884dc | 1260 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
1261 | struct tc_cbq_fopt opt; |
1262 | ||
1263 | if (cl->split || cl->defmap) { | |
d77fea2e | 1264 | opt.split = cl->split ? cl->split->common.classid : 0; |
1da177e4 LT |
1265 | opt.defmap = cl->defmap; |
1266 | opt.defchange = ~0; | |
1b34ec43 DM |
1267 | if (nla_put(skb, TCA_CBQ_FOPT, sizeof(opt), &opt)) |
1268 | goto nla_put_failure; | |
1da177e4 LT |
1269 | } |
1270 | return skb->len; | |
1271 | ||
1e90474c | 1272 | nla_put_failure: |
dc5fc579 | 1273 | nlmsg_trim(skb, b); |
1da177e4 LT |
1274 | return -1; |
1275 | } | |
1276 | ||
1da177e4 LT |
1277 | static int cbq_dump_attr(struct sk_buff *skb, struct cbq_class *cl) |
1278 | { | |
1279 | if (cbq_dump_lss(skb, cl) < 0 || | |
1280 | cbq_dump_rate(skb, cl) < 0 || | |
1281 | cbq_dump_wrr(skb, cl) < 0 || | |
1da177e4 LT |
1282 | cbq_dump_fopt(skb, cl) < 0) |
1283 | return -1; | |
1284 | return 0; | |
1285 | } | |
1286 | ||
1287 | static int cbq_dump(struct Qdisc *sch, struct sk_buff *skb) | |
1288 | { | |
1289 | struct cbq_sched_data *q = qdisc_priv(sch); | |
4b3550ef | 1290 | struct nlattr *nest; |
1da177e4 | 1291 | |
4b3550ef PM |
1292 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1293 | if (nest == NULL) | |
1294 | goto nla_put_failure; | |
1da177e4 | 1295 | if (cbq_dump_attr(skb, &q->link) < 0) |
1e90474c | 1296 | goto nla_put_failure; |
d59b7d80 | 1297 | return nla_nest_end(skb, nest); |
1da177e4 | 1298 | |
1e90474c | 1299 | nla_put_failure: |
4b3550ef | 1300 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1301 | return -1; |
1302 | } | |
1303 | ||
1304 | static int | |
1305 | cbq_dump_stats(struct Qdisc *sch, struct gnet_dump *d) | |
1306 | { | |
1307 | struct cbq_sched_data *q = qdisc_priv(sch); | |
1308 | ||
1309 | q->link.xstats.avgidle = q->link.avgidle; | |
1310 | return gnet_stats_copy_app(d, &q->link.xstats, sizeof(q->link.xstats)); | |
1311 | } | |
1312 | ||
1313 | static int | |
1314 | cbq_dump_class(struct Qdisc *sch, unsigned long arg, | |
1315 | struct sk_buff *skb, struct tcmsg *tcm) | |
1316 | { | |
cc7ec456 | 1317 | struct cbq_class *cl = (struct cbq_class *)arg; |
4b3550ef | 1318 | struct nlattr *nest; |
1da177e4 LT |
1319 | |
1320 | if (cl->tparent) | |
d77fea2e | 1321 | tcm->tcm_parent = cl->tparent->common.classid; |
1da177e4 LT |
1322 | else |
1323 | tcm->tcm_parent = TC_H_ROOT; | |
d77fea2e | 1324 | tcm->tcm_handle = cl->common.classid; |
1da177e4 LT |
1325 | tcm->tcm_info = cl->q->handle; |
1326 | ||
4b3550ef PM |
1327 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1328 | if (nest == NULL) | |
1329 | goto nla_put_failure; | |
1da177e4 | 1330 | if (cbq_dump_attr(skb, cl) < 0) |
1e90474c | 1331 | goto nla_put_failure; |
d59b7d80 | 1332 | return nla_nest_end(skb, nest); |
1da177e4 | 1333 | |
1e90474c | 1334 | nla_put_failure: |
4b3550ef | 1335 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1336 | return -1; |
1337 | } | |
1338 | ||
1339 | static int | |
1340 | cbq_dump_class_stats(struct Qdisc *sch, unsigned long arg, | |
1341 | struct gnet_dump *d) | |
1342 | { | |
1343 | struct cbq_sched_data *q = qdisc_priv(sch); | |
cc7ec456 | 1344 | struct cbq_class *cl = (struct cbq_class *)arg; |
1da177e4 | 1345 | |
1da177e4 LT |
1346 | cl->xstats.avgidle = cl->avgidle; |
1347 | cl->xstats.undertime = 0; | |
1348 | ||
a084980d | 1349 | if (cl->undertime != PSCHED_PASTPERFECT) |
8edc0c31 | 1350 | cl->xstats.undertime = cl->undertime - q->now; |
1da177e4 | 1351 | |
edb09eb1 ED |
1352 | if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), |
1353 | d, NULL, &cl->bstats) < 0 || | |
1c0d32fd | 1354 | gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || |
b0ab6f92 | 1355 | gnet_stats_copy_queue(d, NULL, &cl->qstats, cl->q->q.qlen) < 0) |
1da177e4 LT |
1356 | return -1; |
1357 | ||
1358 | return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); | |
1359 | } | |
1360 | ||
1361 | static int cbq_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
1362 | struct Qdisc **old) | |
1363 | { | |
cc7ec456 | 1364 | struct cbq_class *cl = (struct cbq_class *)arg; |
1da177e4 | 1365 | |
5b9a9ccf | 1366 | if (new == NULL) { |
3511c913 | 1367 | new = qdisc_create_dflt(sch->dev_queue, |
5b9a9ccf PM |
1368 | &pfifo_qdisc_ops, cl->common.classid); |
1369 | if (new == NULL) | |
1370 | return -ENOBUFS; | |
1da177e4 | 1371 | } |
5b9a9ccf | 1372 | |
86a7996c | 1373 | *old = qdisc_replace(sch, new, &cl->q); |
5b9a9ccf | 1374 | return 0; |
1da177e4 LT |
1375 | } |
1376 | ||
cc7ec456 | 1377 | static struct Qdisc *cbq_leaf(struct Qdisc *sch, unsigned long arg) |
1da177e4 | 1378 | { |
cc7ec456 | 1379 | struct cbq_class *cl = (struct cbq_class *)arg; |
1da177e4 | 1380 | |
5b9a9ccf | 1381 | return cl->q; |
1da177e4 LT |
1382 | } |
1383 | ||
a37ef2e3 JP |
1384 | static void cbq_qlen_notify(struct Qdisc *sch, unsigned long arg) |
1385 | { | |
1386 | struct cbq_class *cl = (struct cbq_class *)arg; | |
1387 | ||
1388 | if (cl->q->q.qlen == 0) | |
1389 | cbq_deactivate_class(cl); | |
1390 | } | |
1391 | ||
1da177e4 LT |
1392 | static unsigned long cbq_get(struct Qdisc *sch, u32 classid) |
1393 | { | |
1394 | struct cbq_sched_data *q = qdisc_priv(sch); | |
1395 | struct cbq_class *cl = cbq_class_lookup(q, classid); | |
1396 | ||
1397 | if (cl) { | |
1398 | cl->refcnt++; | |
1399 | return (unsigned long)cl; | |
1400 | } | |
1401 | return 0; | |
1402 | } | |
1403 | ||
1da177e4 LT |
1404 | static void cbq_destroy_class(struct Qdisc *sch, struct cbq_class *cl) |
1405 | { | |
1406 | struct cbq_sched_data *q = qdisc_priv(sch); | |
1407 | ||
547b792c | 1408 | WARN_ON(cl->filters); |
1da177e4 | 1409 | |
6529eaba | 1410 | tcf_block_put(cl->block); |
1da177e4 LT |
1411 | qdisc_destroy(cl->q); |
1412 | qdisc_put_rtab(cl->R_tab); | |
1c0d32fd | 1413 | gen_kill_estimator(&cl->rate_est); |
1da177e4 LT |
1414 | if (cl != &q->link) |
1415 | kfree(cl); | |
1416 | } | |
1417 | ||
cc7ec456 | 1418 | static void cbq_destroy(struct Qdisc *sch) |
1da177e4 LT |
1419 | { |
1420 | struct cbq_sched_data *q = qdisc_priv(sch); | |
b67bfe0d | 1421 | struct hlist_node *next; |
1da177e4 | 1422 | struct cbq_class *cl; |
cc7ec456 | 1423 | unsigned int h; |
1da177e4 | 1424 | |
c3bc7cff | 1425 | #ifdef CONFIG_NET_CLS_ACT |
1da177e4 LT |
1426 | q->rx_class = NULL; |
1427 | #endif | |
1428 | /* | |
1429 | * Filters must be destroyed first because we don't destroy the | |
1430 | * classes from root to leafs which means that filters can still | |
1431 | * be bound to classes which have been destroyed already. --TGR '04 | |
1432 | */ | |
d77fea2e | 1433 | for (h = 0; h < q->clhash.hashsize; h++) { |
b67bfe0d | 1434 | hlist_for_each_entry(cl, &q->clhash.hash[h], common.hnode) |
6529eaba | 1435 | tcf_block_put(cl->block); |
b00b4bf9 | 1436 | } |
d77fea2e | 1437 | for (h = 0; h < q->clhash.hashsize; h++) { |
b67bfe0d | 1438 | hlist_for_each_entry_safe(cl, next, &q->clhash.hash[h], |
d77fea2e | 1439 | common.hnode) |
1da177e4 | 1440 | cbq_destroy_class(sch, cl); |
1da177e4 | 1441 | } |
d77fea2e | 1442 | qdisc_class_hash_destroy(&q->clhash); |
1da177e4 LT |
1443 | } |
1444 | ||
1445 | static void cbq_put(struct Qdisc *sch, unsigned long arg) | |
1446 | { | |
cc7ec456 | 1447 | struct cbq_class *cl = (struct cbq_class *)arg; |
1da177e4 LT |
1448 | |
1449 | if (--cl->refcnt == 0) { | |
c3bc7cff | 1450 | #ifdef CONFIG_NET_CLS_ACT |
102396ae | 1451 | spinlock_t *root_lock = qdisc_root_sleeping_lock(sch); |
1da177e4 LT |
1452 | struct cbq_sched_data *q = qdisc_priv(sch); |
1453 | ||
7698b4fc | 1454 | spin_lock_bh(root_lock); |
1da177e4 LT |
1455 | if (q->rx_class == cl) |
1456 | q->rx_class = NULL; | |
7698b4fc | 1457 | spin_unlock_bh(root_lock); |
1da177e4 LT |
1458 | #endif |
1459 | ||
1460 | cbq_destroy_class(sch, cl); | |
1461 | } | |
1462 | } | |
1463 | ||
1464 | static int | |
1e90474c | 1465 | cbq_change_class(struct Qdisc *sch, u32 classid, u32 parentid, struct nlattr **tca, |
1da177e4 LT |
1466 | unsigned long *arg) |
1467 | { | |
1468 | int err; | |
1469 | struct cbq_sched_data *q = qdisc_priv(sch); | |
cc7ec456 | 1470 | struct cbq_class *cl = (struct cbq_class *)*arg; |
1e90474c PM |
1471 | struct nlattr *opt = tca[TCA_OPTIONS]; |
1472 | struct nlattr *tb[TCA_CBQ_MAX + 1]; | |
1da177e4 LT |
1473 | struct cbq_class *parent; |
1474 | struct qdisc_rate_table *rtab = NULL; | |
1475 | ||
cee63723 | 1476 | if (opt == NULL) |
1da177e4 LT |
1477 | return -EINVAL; |
1478 | ||
fceb6435 | 1479 | err = nla_parse_nested(tb, TCA_CBQ_MAX, opt, cbq_policy, NULL); |
cee63723 PM |
1480 | if (err < 0) |
1481 | return err; | |
1482 | ||
dd47c1fa | 1483 | if (tb[TCA_CBQ_OVL_STRATEGY] || tb[TCA_CBQ_POLICE]) |
c3498d34 FW |
1484 | return -EOPNOTSUPP; |
1485 | ||
1da177e4 LT |
1486 | if (cl) { |
1487 | /* Check parent */ | |
1488 | if (parentid) { | |
d77fea2e PM |
1489 | if (cl->tparent && |
1490 | cl->tparent->common.classid != parentid) | |
1da177e4 LT |
1491 | return -EINVAL; |
1492 | if (!cl->tparent && parentid != TC_H_ROOT) | |
1493 | return -EINVAL; | |
1494 | } | |
1495 | ||
1e90474c | 1496 | if (tb[TCA_CBQ_RATE]) { |
71bcb09a SH |
1497 | rtab = qdisc_get_rtab(nla_data(tb[TCA_CBQ_RATE]), |
1498 | tb[TCA_CBQ_RTAB]); | |
1da177e4 LT |
1499 | if (rtab == NULL) |
1500 | return -EINVAL; | |
1501 | } | |
1502 | ||
71bcb09a | 1503 | if (tca[TCA_RATE]) { |
22e0f8b9 JF |
1504 | err = gen_replace_estimator(&cl->bstats, NULL, |
1505 | &cl->rate_est, | |
edb09eb1 ED |
1506 | NULL, |
1507 | qdisc_root_sleeping_running(sch), | |
71bcb09a SH |
1508 | tca[TCA_RATE]); |
1509 | if (err) { | |
79c11f2e | 1510 | qdisc_put_rtab(rtab); |
71bcb09a SH |
1511 | return err; |
1512 | } | |
1513 | } | |
1514 | ||
1da177e4 LT |
1515 | /* Change class parameters */ |
1516 | sch_tree_lock(sch); | |
1517 | ||
1518 | if (cl->next_alive != NULL) | |
1519 | cbq_deactivate_class(cl); | |
1520 | ||
1521 | if (rtab) { | |
b94c8afc PM |
1522 | qdisc_put_rtab(cl->R_tab); |
1523 | cl->R_tab = rtab; | |
1da177e4 LT |
1524 | } |
1525 | ||
1e90474c PM |
1526 | if (tb[TCA_CBQ_LSSOPT]) |
1527 | cbq_set_lss(cl, nla_data(tb[TCA_CBQ_LSSOPT])); | |
1da177e4 | 1528 | |
1e90474c | 1529 | if (tb[TCA_CBQ_WRROPT]) { |
1da177e4 | 1530 | cbq_rmprio(q, cl); |
1e90474c | 1531 | cbq_set_wrr(cl, nla_data(tb[TCA_CBQ_WRROPT])); |
1da177e4 LT |
1532 | } |
1533 | ||
1e90474c PM |
1534 | if (tb[TCA_CBQ_FOPT]) |
1535 | cbq_set_fopt(cl, nla_data(tb[TCA_CBQ_FOPT])); | |
1da177e4 LT |
1536 | |
1537 | if (cl->q->q.qlen) | |
1538 | cbq_activate_class(cl); | |
1539 | ||
1540 | sch_tree_unlock(sch); | |
1541 | ||
1da177e4 LT |
1542 | return 0; |
1543 | } | |
1544 | ||
1545 | if (parentid == TC_H_ROOT) | |
1546 | return -EINVAL; | |
1547 | ||
1e90474c PM |
1548 | if (tb[TCA_CBQ_WRROPT] == NULL || tb[TCA_CBQ_RATE] == NULL || |
1549 | tb[TCA_CBQ_LSSOPT] == NULL) | |
1da177e4 LT |
1550 | return -EINVAL; |
1551 | ||
1e90474c | 1552 | rtab = qdisc_get_rtab(nla_data(tb[TCA_CBQ_RATE]), tb[TCA_CBQ_RTAB]); |
1da177e4 LT |
1553 | if (rtab == NULL) |
1554 | return -EINVAL; | |
1555 | ||
1556 | if (classid) { | |
1557 | err = -EINVAL; | |
cc7ec456 ED |
1558 | if (TC_H_MAJ(classid ^ sch->handle) || |
1559 | cbq_class_lookup(q, classid)) | |
1da177e4 LT |
1560 | goto failure; |
1561 | } else { | |
1562 | int i; | |
cc7ec456 | 1563 | classid = TC_H_MAKE(sch->handle, 0x8000); |
1da177e4 | 1564 | |
cc7ec456 | 1565 | for (i = 0; i < 0x8000; i++) { |
1da177e4 LT |
1566 | if (++q->hgenerator >= 0x8000) |
1567 | q->hgenerator = 1; | |
1568 | if (cbq_class_lookup(q, classid|q->hgenerator) == NULL) | |
1569 | break; | |
1570 | } | |
1571 | err = -ENOSR; | |
1572 | if (i >= 0x8000) | |
1573 | goto failure; | |
1574 | classid = classid|q->hgenerator; | |
1575 | } | |
1576 | ||
1577 | parent = &q->link; | |
1578 | if (parentid) { | |
1579 | parent = cbq_class_lookup(q, parentid); | |
1580 | err = -EINVAL; | |
1581 | if (parent == NULL) | |
1582 | goto failure; | |
1583 | } | |
1584 | ||
1585 | err = -ENOBUFS; | |
0da974f4 | 1586 | cl = kzalloc(sizeof(*cl), GFP_KERNEL); |
1da177e4 LT |
1587 | if (cl == NULL) |
1588 | goto failure; | |
71bcb09a | 1589 | |
6529eaba JP |
1590 | err = tcf_block_get(&cl->block, &cl->filter_list); |
1591 | if (err) { | |
1592 | kfree(cl); | |
1593 | return err; | |
1594 | } | |
1595 | ||
71bcb09a | 1596 | if (tca[TCA_RATE]) { |
22e0f8b9 | 1597 | err = gen_new_estimator(&cl->bstats, NULL, &cl->rate_est, |
edb09eb1 ED |
1598 | NULL, |
1599 | qdisc_root_sleeping_running(sch), | |
71bcb09a SH |
1600 | tca[TCA_RATE]); |
1601 | if (err) { | |
6529eaba | 1602 | tcf_block_put(cl->block); |
71bcb09a SH |
1603 | kfree(cl); |
1604 | goto failure; | |
1605 | } | |
1606 | } | |
1607 | ||
1da177e4 LT |
1608 | cl->R_tab = rtab; |
1609 | rtab = NULL; | |
1610 | cl->refcnt = 1; | |
3511c913 CG |
1611 | cl->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, classid); |
1612 | if (!cl->q) | |
1da177e4 | 1613 | cl->q = &noop_qdisc; |
49b49971 JK |
1614 | else |
1615 | qdisc_hash_add(cl->q, true); | |
1616 | ||
d77fea2e | 1617 | cl->common.classid = classid; |
1da177e4 LT |
1618 | cl->tparent = parent; |
1619 | cl->qdisc = sch; | |
1620 | cl->allot = parent->allot; | |
1621 | cl->quantum = cl->allot; | |
1622 | cl->weight = cl->R_tab->rate.rate; | |
1da177e4 LT |
1623 | |
1624 | sch_tree_lock(sch); | |
1625 | cbq_link_class(cl); | |
1626 | cl->borrow = cl->tparent; | |
1627 | if (cl->tparent != &q->link) | |
1628 | cl->share = cl->tparent; | |
1629 | cbq_adjust_levels(parent); | |
1630 | cl->minidle = -0x7FFFFFFF; | |
1e90474c PM |
1631 | cbq_set_lss(cl, nla_data(tb[TCA_CBQ_LSSOPT])); |
1632 | cbq_set_wrr(cl, nla_data(tb[TCA_CBQ_WRROPT])); | |
cc7ec456 | 1633 | if (cl->ewma_log == 0) |
1da177e4 | 1634 | cl->ewma_log = q->link.ewma_log; |
cc7ec456 | 1635 | if (cl->maxidle == 0) |
1da177e4 | 1636 | cl->maxidle = q->link.maxidle; |
cc7ec456 | 1637 | if (cl->avpkt == 0) |
1da177e4 | 1638 | cl->avpkt = q->link.avpkt; |
1e90474c PM |
1639 | if (tb[TCA_CBQ_FOPT]) |
1640 | cbq_set_fopt(cl, nla_data(tb[TCA_CBQ_FOPT])); | |
1da177e4 LT |
1641 | sch_tree_unlock(sch); |
1642 | ||
d77fea2e PM |
1643 | qdisc_class_hash_grow(sch, &q->clhash); |
1644 | ||
1da177e4 LT |
1645 | *arg = (unsigned long)cl; |
1646 | return 0; | |
1647 | ||
1648 | failure: | |
1649 | qdisc_put_rtab(rtab); | |
1650 | return err; | |
1651 | } | |
1652 | ||
1653 | static int cbq_delete(struct Qdisc *sch, unsigned long arg) | |
1654 | { | |
1655 | struct cbq_sched_data *q = qdisc_priv(sch); | |
cc7ec456 | 1656 | struct cbq_class *cl = (struct cbq_class *)arg; |
2ccccf5f | 1657 | unsigned int qlen, backlog; |
1da177e4 LT |
1658 | |
1659 | if (cl->filters || cl->children || cl == &q->link) | |
1660 | return -EBUSY; | |
1661 | ||
1662 | sch_tree_lock(sch); | |
1663 | ||
a37ef2e3 | 1664 | qlen = cl->q->q.qlen; |
2ccccf5f | 1665 | backlog = cl->q->qstats.backlog; |
a37ef2e3 | 1666 | qdisc_reset(cl->q); |
2ccccf5f | 1667 | qdisc_tree_reduce_backlog(cl->q, qlen, backlog); |
a37ef2e3 | 1668 | |
1da177e4 LT |
1669 | if (cl->next_alive) |
1670 | cbq_deactivate_class(cl); | |
1671 | ||
1672 | if (q->tx_borrowed == cl) | |
1673 | q->tx_borrowed = q->tx_class; | |
1674 | if (q->tx_class == cl) { | |
1675 | q->tx_class = NULL; | |
1676 | q->tx_borrowed = NULL; | |
1677 | } | |
c3bc7cff | 1678 | #ifdef CONFIG_NET_CLS_ACT |
1da177e4 LT |
1679 | if (q->rx_class == cl) |
1680 | q->rx_class = NULL; | |
1681 | #endif | |
1682 | ||
1683 | cbq_unlink_class(cl); | |
1684 | cbq_adjust_levels(cl->tparent); | |
1685 | cl->defmap = 0; | |
1686 | cbq_sync_defmap(cl); | |
1687 | ||
1688 | cbq_rmprio(q, cl); | |
1689 | sch_tree_unlock(sch); | |
1690 | ||
7cd0a638 JP |
1691 | BUG_ON(--cl->refcnt == 0); |
1692 | /* | |
1693 | * This shouldn't happen: we "hold" one cops->get() when called | |
1694 | * from tc_ctl_tclass; the destroy method is done from cops->put(). | |
1695 | */ | |
1da177e4 LT |
1696 | |
1697 | return 0; | |
1698 | } | |
1699 | ||
6529eaba | 1700 | static struct tcf_block *cbq_tcf_block(struct Qdisc *sch, unsigned long arg) |
1da177e4 LT |
1701 | { |
1702 | struct cbq_sched_data *q = qdisc_priv(sch); | |
1703 | struct cbq_class *cl = (struct cbq_class *)arg; | |
1704 | ||
1705 | if (cl == NULL) | |
1706 | cl = &q->link; | |
1707 | ||
6529eaba | 1708 | return cl->block; |
1da177e4 LT |
1709 | } |
1710 | ||
1711 | static unsigned long cbq_bind_filter(struct Qdisc *sch, unsigned long parent, | |
1712 | u32 classid) | |
1713 | { | |
1714 | struct cbq_sched_data *q = qdisc_priv(sch); | |
cc7ec456 | 1715 | struct cbq_class *p = (struct cbq_class *)parent; |
1da177e4 LT |
1716 | struct cbq_class *cl = cbq_class_lookup(q, classid); |
1717 | ||
1718 | if (cl) { | |
1719 | if (p && p->level <= cl->level) | |
1720 | return 0; | |
1721 | cl->filters++; | |
1722 | return (unsigned long)cl; | |
1723 | } | |
1724 | return 0; | |
1725 | } | |
1726 | ||
1727 | static void cbq_unbind_filter(struct Qdisc *sch, unsigned long arg) | |
1728 | { | |
cc7ec456 | 1729 | struct cbq_class *cl = (struct cbq_class *)arg; |
1da177e4 LT |
1730 | |
1731 | cl->filters--; | |
1732 | } | |
1733 | ||
1734 | static void cbq_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
1735 | { | |
1736 | struct cbq_sched_data *q = qdisc_priv(sch); | |
d77fea2e | 1737 | struct cbq_class *cl; |
cc7ec456 | 1738 | unsigned int h; |
1da177e4 LT |
1739 | |
1740 | if (arg->stop) | |
1741 | return; | |
1742 | ||
d77fea2e | 1743 | for (h = 0; h < q->clhash.hashsize; h++) { |
b67bfe0d | 1744 | hlist_for_each_entry(cl, &q->clhash.hash[h], common.hnode) { |
1da177e4 LT |
1745 | if (arg->count < arg->skip) { |
1746 | arg->count++; | |
1747 | continue; | |
1748 | } | |
1749 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
1750 | arg->stop = 1; | |
1751 | return; | |
1752 | } | |
1753 | arg->count++; | |
1754 | } | |
1755 | } | |
1756 | } | |
1757 | ||
20fea08b | 1758 | static const struct Qdisc_class_ops cbq_class_ops = { |
1da177e4 LT |
1759 | .graft = cbq_graft, |
1760 | .leaf = cbq_leaf, | |
a37ef2e3 | 1761 | .qlen_notify = cbq_qlen_notify, |
1da177e4 LT |
1762 | .get = cbq_get, |
1763 | .put = cbq_put, | |
1764 | .change = cbq_change_class, | |
1765 | .delete = cbq_delete, | |
1766 | .walk = cbq_walk, | |
6529eaba | 1767 | .tcf_block = cbq_tcf_block, |
1da177e4 LT |
1768 | .bind_tcf = cbq_bind_filter, |
1769 | .unbind_tcf = cbq_unbind_filter, | |
1770 | .dump = cbq_dump_class, | |
1771 | .dump_stats = cbq_dump_class_stats, | |
1772 | }; | |
1773 | ||
20fea08b | 1774 | static struct Qdisc_ops cbq_qdisc_ops __read_mostly = { |
1da177e4 LT |
1775 | .next = NULL, |
1776 | .cl_ops = &cbq_class_ops, | |
1777 | .id = "cbq", | |
1778 | .priv_size = sizeof(struct cbq_sched_data), | |
1779 | .enqueue = cbq_enqueue, | |
1780 | .dequeue = cbq_dequeue, | |
77be155c | 1781 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1782 | .init = cbq_init, |
1783 | .reset = cbq_reset, | |
1784 | .destroy = cbq_destroy, | |
1785 | .change = NULL, | |
1786 | .dump = cbq_dump, | |
1787 | .dump_stats = cbq_dump_stats, | |
1788 | .owner = THIS_MODULE, | |
1789 | }; | |
1790 | ||
1791 | static int __init cbq_module_init(void) | |
1792 | { | |
1793 | return register_qdisc(&cbq_qdisc_ops); | |
1794 | } | |
10297b99 | 1795 | static void __exit cbq_module_exit(void) |
1da177e4 LT |
1796 | { |
1797 | unregister_qdisc(&cbq_qdisc_ops); | |
1798 | } | |
1799 | module_init(cbq_module_init) | |
1800 | module_exit(cbq_module_exit) | |
1801 | MODULE_LICENSE("GPL"); |