[NET_SCHED]: Convert packet schedulers from rtnetlink to new netlink API
[deliverable/linux.git] / net / sched / sch_htb.c
1 /*
2 * net/sched/sch_htb.c Hierarchical token bucket, feed tree version
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: Martin Devera, <devik@cdi.cz>
10 *
11 * Credits (in time order) for older HTB versions:
12 * Stef Coene <stef.coene@docum.org>
13 * HTB support at LARTC mailing list
14 * Ondrej Kraus, <krauso@barr.cz>
15 * found missing INIT_QDISC(htb)
16 * Vladimir Smelhaus, Aamer Akhter, Bert Hubert
17 * helped a lot to locate nasty class stall bug
18 * Andi Kleen, Jamal Hadi, Bert Hubert
19 * code review and helpful comments on shaping
20 * Tomasz Wrona, <tw@eter.tym.pl>
21 * created test case so that I was able to fix nasty bug
22 * Wilfried Weissmann
23 * spotted bug in dequeue code and helped with fix
24 * Jiri Fojtasek
25 * fixed requeue routine
26 * and many others. thanks.
27 *
28 * $Id: sch_htb.c,v 1.25 2003/12/07 11:08:25 devik Exp devik $
29 */
30 #include <linux/module.h>
31 #include <linux/types.h>
32 #include <linux/kernel.h>
33 #include <linux/string.h>
34 #include <linux/errno.h>
35 #include <linux/skbuff.h>
36 #include <linux/list.h>
37 #include <linux/compiler.h>
38 #include <linux/rbtree.h>
39 #include <net/netlink.h>
40 #include <net/pkt_sched.h>
41
42 /* HTB algorithm.
43 Author: devik@cdi.cz
44 ========================================================================
45 HTB is like TBF with multiple classes. It is also similar to CBQ because
46 it allows to assign priority to each class in hierarchy.
47 In fact it is another implementation of Floyd's formal sharing.
48
49 Levels:
50 Each class is assigned level. Leaf has ALWAYS level 0 and root
51 classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level
52 one less than their parent.
53 */
54
55 #define HTB_HSIZE 16 /* classid hash size */
56 #define HTB_HYSTERESIS 1 /* whether to use mode hysteresis for speedup */
57 #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */
58
59 #if HTB_VER >> 16 != TC_HTB_PROTOVER
60 #error "Mismatched sch_htb.c and pkt_sch.h"
61 #endif
62
63 /* used internaly to keep status of single class */
64 enum htb_cmode {
65 HTB_CANT_SEND, /* class can't send and can't borrow */
66 HTB_MAY_BORROW, /* class can't send but may borrow */
67 HTB_CAN_SEND /* class can send */
68 };
69
70 /* interior & leaf nodes; props specific to leaves are marked L: */
71 struct htb_class {
72 /* general class parameters */
73 u32 classid;
74 struct gnet_stats_basic bstats;
75 struct gnet_stats_queue qstats;
76 struct gnet_stats_rate_est rate_est;
77 struct tc_htb_xstats xstats; /* our special stats */
78 int refcnt; /* usage count of this class */
79
80 /* topology */
81 int level; /* our level (see above) */
82 struct htb_class *parent; /* parent class */
83 struct hlist_node hlist; /* classid hash list item */
84 struct list_head sibling; /* sibling list item */
85 struct list_head children; /* children list */
86
87 union {
88 struct htb_class_leaf {
89 struct Qdisc *q;
90 int prio;
91 int aprio;
92 int quantum;
93 int deficit[TC_HTB_MAXDEPTH];
94 struct list_head drop_list;
95 } leaf;
96 struct htb_class_inner {
97 struct rb_root feed[TC_HTB_NUMPRIO]; /* feed trees */
98 struct rb_node *ptr[TC_HTB_NUMPRIO]; /* current class ptr */
99 /* When class changes from state 1->2 and disconnects from
100 parent's feed then we lost ptr value and start from the
101 first child again. Here we store classid of the
102 last valid ptr (used when ptr is NULL). */
103 u32 last_ptr_id[TC_HTB_NUMPRIO];
104 } inner;
105 } un;
106 struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */
107 struct rb_node pq_node; /* node for event queue */
108 psched_time_t pq_key;
109
110 int prio_activity; /* for which prios are we active */
111 enum htb_cmode cmode; /* current mode of the class */
112
113 /* class attached filters */
114 struct tcf_proto *filter_list;
115 int filter_cnt;
116
117 int warned; /* only one warning about non work conserving .. */
118
119 /* token bucket parameters */
120 struct qdisc_rate_table *rate; /* rate table of the class itself */
121 struct qdisc_rate_table *ceil; /* ceiling rate (limits borrows too) */
122 long buffer, cbuffer; /* token bucket depth/rate */
123 psched_tdiff_t mbuffer; /* max wait time */
124 long tokens, ctokens; /* current number of tokens */
125 psched_time_t t_c; /* checkpoint time */
126
127 int prio; /* For parent to leaf return possible here */
128 int quantum; /* we do backup. Finally full replacement */
129 /* of un.leaf originals should be done. */
130 };
131
132 static inline long L2T(struct htb_class *cl, struct qdisc_rate_table *rate,
133 int size)
134 {
135 long result = qdisc_l2t(rate, size);
136 return result;
137 }
138
139 struct htb_sched {
140 struct list_head root; /* root classes list */
141 struct hlist_head hash[HTB_HSIZE]; /* hashed by classid */
142 struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */
143
144 /* self list - roots of self generating tree */
145 struct rb_root row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
146 int row_mask[TC_HTB_MAXDEPTH];
147 struct rb_node *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
148 u32 last_ptr_id[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
149
150 /* self wait list - roots of wait PQs per row */
151 struct rb_root wait_pq[TC_HTB_MAXDEPTH];
152
153 /* time of nearest event per level (row) */
154 psched_time_t near_ev_cache[TC_HTB_MAXDEPTH];
155
156 /* whether we hit non-work conserving class during this dequeue; we use */
157 int nwc_hit; /* this to disable mindelay complaint in dequeue */
158
159 int defcls; /* class where unclassified flows go to */
160
161 /* filters for qdisc itself */
162 struct tcf_proto *filter_list;
163 int filter_cnt;
164
165 int rate2quantum; /* quant = rate / rate2quantum */
166 psched_time_t now; /* cached dequeue time */
167 struct qdisc_watchdog watchdog;
168
169 /* non shaped skbs; let them go directly thru */
170 struct sk_buff_head direct_queue;
171 int direct_qlen; /* max qlen of above */
172
173 long direct_pkts;
174 };
175
176 /* compute hash of size HTB_HSIZE for given handle */
177 static inline int htb_hash(u32 h)
178 {
179 #if HTB_HSIZE != 16
180 #error "Declare new hash for your HTB_HSIZE"
181 #endif
182 h ^= h >> 8; /* stolen from cbq_hash */
183 h ^= h >> 4;
184 return h & 0xf;
185 }
186
187 /* find class in global hash table using given handle */
188 static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch)
189 {
190 struct htb_sched *q = qdisc_priv(sch);
191 struct hlist_node *p;
192 struct htb_class *cl;
193
194 if (TC_H_MAJ(handle) != sch->handle)
195 return NULL;
196
197 hlist_for_each_entry(cl, p, q->hash + htb_hash(handle), hlist) {
198 if (cl->classid == handle)
199 return cl;
200 }
201 return NULL;
202 }
203
204 /**
205 * htb_classify - classify a packet into class
206 *
207 * It returns NULL if the packet should be dropped or -1 if the packet
208 * should be passed directly thru. In all other cases leaf class is returned.
209 * We allow direct class selection by classid in priority. The we examine
210 * filters in qdisc and in inner nodes (if higher filter points to the inner
211 * node). If we end up with classid MAJOR:0 we enqueue the skb into special
212 * internal fifo (direct). These packets then go directly thru. If we still
213 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessfull
214 * then finish and return direct queue.
215 */
216 #define HTB_DIRECT (struct htb_class*)-1
217
218 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch,
219 int *qerr)
220 {
221 struct htb_sched *q = qdisc_priv(sch);
222 struct htb_class *cl;
223 struct tcf_result res;
224 struct tcf_proto *tcf;
225 int result;
226
227 /* allow to select class by setting skb->priority to valid classid;
228 note that nfmark can be used too by attaching filter fw with no
229 rules in it */
230 if (skb->priority == sch->handle)
231 return HTB_DIRECT; /* X:0 (direct flow) selected */
232 if ((cl = htb_find(skb->priority, sch)) != NULL && cl->level == 0)
233 return cl;
234
235 *qerr = NET_XMIT_BYPASS;
236 tcf = q->filter_list;
237 while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
238 #ifdef CONFIG_NET_CLS_ACT
239 switch (result) {
240 case TC_ACT_QUEUED:
241 case TC_ACT_STOLEN:
242 *qerr = NET_XMIT_SUCCESS;
243 case TC_ACT_SHOT:
244 return NULL;
245 }
246 #endif
247 if ((cl = (void *)res.class) == NULL) {
248 if (res.classid == sch->handle)
249 return HTB_DIRECT; /* X:0 (direct flow) */
250 if ((cl = htb_find(res.classid, sch)) == NULL)
251 break; /* filter selected invalid classid */
252 }
253 if (!cl->level)
254 return cl; /* we hit leaf; return it */
255
256 /* we have got inner class; apply inner filter chain */
257 tcf = cl->filter_list;
258 }
259 /* classification failed; try to use default class */
260 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
261 if (!cl || cl->level)
262 return HTB_DIRECT; /* bad default .. this is safe bet */
263 return cl;
264 }
265
266 /**
267 * htb_add_to_id_tree - adds class to the round robin list
268 *
269 * Routine adds class to the list (actually tree) sorted by classid.
270 * Make sure that class is not already on such list for given prio.
271 */
272 static void htb_add_to_id_tree(struct rb_root *root,
273 struct htb_class *cl, int prio)
274 {
275 struct rb_node **p = &root->rb_node, *parent = NULL;
276
277 while (*p) {
278 struct htb_class *c;
279 parent = *p;
280 c = rb_entry(parent, struct htb_class, node[prio]);
281
282 if (cl->classid > c->classid)
283 p = &parent->rb_right;
284 else
285 p = &parent->rb_left;
286 }
287 rb_link_node(&cl->node[prio], parent, p);
288 rb_insert_color(&cl->node[prio], root);
289 }
290
291 /**
292 * htb_add_to_wait_tree - adds class to the event queue with delay
293 *
294 * The class is added to priority event queue to indicate that class will
295 * change its mode in cl->pq_key microseconds. Make sure that class is not
296 * already in the queue.
297 */
298 static void htb_add_to_wait_tree(struct htb_sched *q,
299 struct htb_class *cl, long delay)
300 {
301 struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL;
302
303 cl->pq_key = q->now + delay;
304 if (cl->pq_key == q->now)
305 cl->pq_key++;
306
307 /* update the nearest event cache */
308 if (q->near_ev_cache[cl->level] > cl->pq_key)
309 q->near_ev_cache[cl->level] = cl->pq_key;
310
311 while (*p) {
312 struct htb_class *c;
313 parent = *p;
314 c = rb_entry(parent, struct htb_class, pq_node);
315 if (cl->pq_key >= c->pq_key)
316 p = &parent->rb_right;
317 else
318 p = &parent->rb_left;
319 }
320 rb_link_node(&cl->pq_node, parent, p);
321 rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]);
322 }
323
324 /**
325 * htb_next_rb_node - finds next node in binary tree
326 *
327 * When we are past last key we return NULL.
328 * Average complexity is 2 steps per call.
329 */
330 static inline void htb_next_rb_node(struct rb_node **n)
331 {
332 *n = rb_next(*n);
333 }
334
335 /**
336 * htb_add_class_to_row - add class to its row
337 *
338 * The class is added to row at priorities marked in mask.
339 * It does nothing if mask == 0.
340 */
341 static inline void htb_add_class_to_row(struct htb_sched *q,
342 struct htb_class *cl, int mask)
343 {
344 q->row_mask[cl->level] |= mask;
345 while (mask) {
346 int prio = ffz(~mask);
347 mask &= ~(1 << prio);
348 htb_add_to_id_tree(q->row[cl->level] + prio, cl, prio);
349 }
350 }
351
352 /* If this triggers, it is a bug in this code, but it need not be fatal */
353 static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root)
354 {
355 if (RB_EMPTY_NODE(rb)) {
356 WARN_ON(1);
357 } else {
358 rb_erase(rb, root);
359 RB_CLEAR_NODE(rb);
360 }
361 }
362
363
364 /**
365 * htb_remove_class_from_row - removes class from its row
366 *
367 * The class is removed from row at priorities marked in mask.
368 * It does nothing if mask == 0.
369 */
370 static inline void htb_remove_class_from_row(struct htb_sched *q,
371 struct htb_class *cl, int mask)
372 {
373 int m = 0;
374
375 while (mask) {
376 int prio = ffz(~mask);
377
378 mask &= ~(1 << prio);
379 if (q->ptr[cl->level][prio] == cl->node + prio)
380 htb_next_rb_node(q->ptr[cl->level] + prio);
381
382 htb_safe_rb_erase(cl->node + prio, q->row[cl->level] + prio);
383 if (!q->row[cl->level][prio].rb_node)
384 m |= 1 << prio;
385 }
386 q->row_mask[cl->level] &= ~m;
387 }
388
389 /**
390 * htb_activate_prios - creates active classe's feed chain
391 *
392 * The class is connected to ancestors and/or appropriate rows
393 * for priorities it is participating on. cl->cmode must be new
394 * (activated) mode. It does nothing if cl->prio_activity == 0.
395 */
396 static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl)
397 {
398 struct htb_class *p = cl->parent;
399 long m, mask = cl->prio_activity;
400
401 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
402 m = mask;
403 while (m) {
404 int prio = ffz(~m);
405 m &= ~(1 << prio);
406
407 if (p->un.inner.feed[prio].rb_node)
408 /* parent already has its feed in use so that
409 reset bit in mask as parent is already ok */
410 mask &= ~(1 << prio);
411
412 htb_add_to_id_tree(p->un.inner.feed + prio, cl, prio);
413 }
414 p->prio_activity |= mask;
415 cl = p;
416 p = cl->parent;
417
418 }
419 if (cl->cmode == HTB_CAN_SEND && mask)
420 htb_add_class_to_row(q, cl, mask);
421 }
422
423 /**
424 * htb_deactivate_prios - remove class from feed chain
425 *
426 * cl->cmode must represent old mode (before deactivation). It does
427 * nothing if cl->prio_activity == 0. Class is removed from all feed
428 * chains and rows.
429 */
430 static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl)
431 {
432 struct htb_class *p = cl->parent;
433 long m, mask = cl->prio_activity;
434
435 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
436 m = mask;
437 mask = 0;
438 while (m) {
439 int prio = ffz(~m);
440 m &= ~(1 << prio);
441
442 if (p->un.inner.ptr[prio] == cl->node + prio) {
443 /* we are removing child which is pointed to from
444 parent feed - forget the pointer but remember
445 classid */
446 p->un.inner.last_ptr_id[prio] = cl->classid;
447 p->un.inner.ptr[prio] = NULL;
448 }
449
450 htb_safe_rb_erase(cl->node + prio, p->un.inner.feed + prio);
451
452 if (!p->un.inner.feed[prio].rb_node)
453 mask |= 1 << prio;
454 }
455
456 p->prio_activity &= ~mask;
457 cl = p;
458 p = cl->parent;
459
460 }
461 if (cl->cmode == HTB_CAN_SEND && mask)
462 htb_remove_class_from_row(q, cl, mask);
463 }
464
465 #if HTB_HYSTERESIS
466 static inline long htb_lowater(const struct htb_class *cl)
467 {
468 return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0;
469 }
470 static inline long htb_hiwater(const struct htb_class *cl)
471 {
472 return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0;
473 }
474 #else
475 #define htb_lowater(cl) (0)
476 #define htb_hiwater(cl) (0)
477 #endif
478
479 /**
480 * htb_class_mode - computes and returns current class mode
481 *
482 * It computes cl's mode at time cl->t_c+diff and returns it. If mode
483 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
484 * from now to time when cl will change its state.
485 * Also it is worth to note that class mode doesn't change simply
486 * at cl->{c,}tokens == 0 but there can rather be hysteresis of
487 * 0 .. -cl->{c,}buffer range. It is meant to limit number of
488 * mode transitions per time unit. The speed gain is about 1/6.
489 */
490 static inline enum htb_cmode
491 htb_class_mode(struct htb_class *cl, long *diff)
492 {
493 long toks;
494
495 if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) {
496 *diff = -toks;
497 return HTB_CANT_SEND;
498 }
499
500 if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl))
501 return HTB_CAN_SEND;
502
503 *diff = -toks;
504 return HTB_MAY_BORROW;
505 }
506
507 /**
508 * htb_change_class_mode - changes classe's mode
509 *
510 * This should be the only way how to change classe's mode under normal
511 * cirsumstances. Routine will update feed lists linkage, change mode
512 * and add class to the wait event queue if appropriate. New mode should
513 * be different from old one and cl->pq_key has to be valid if changing
514 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
515 */
516 static void
517 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff)
518 {
519 enum htb_cmode new_mode = htb_class_mode(cl, diff);
520
521 if (new_mode == cl->cmode)
522 return;
523
524 if (cl->prio_activity) { /* not necessary: speed optimization */
525 if (cl->cmode != HTB_CANT_SEND)
526 htb_deactivate_prios(q, cl);
527 cl->cmode = new_mode;
528 if (new_mode != HTB_CANT_SEND)
529 htb_activate_prios(q, cl);
530 } else
531 cl->cmode = new_mode;
532 }
533
534 /**
535 * htb_activate - inserts leaf cl into appropriate active feeds
536 *
537 * Routine learns (new) priority of leaf and activates feed chain
538 * for the prio. It can be called on already active leaf safely.
539 * It also adds leaf into droplist.
540 */
541 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl)
542 {
543 BUG_TRAP(!cl->level && cl->un.leaf.q && cl->un.leaf.q->q.qlen);
544
545 if (!cl->prio_activity) {
546 cl->prio_activity = 1 << (cl->un.leaf.aprio = cl->un.leaf.prio);
547 htb_activate_prios(q, cl);
548 list_add_tail(&cl->un.leaf.drop_list,
549 q->drops + cl->un.leaf.aprio);
550 }
551 }
552
553 /**
554 * htb_deactivate - remove leaf cl from active feeds
555 *
556 * Make sure that leaf is active. In the other words it can't be called
557 * with non-active leaf. It also removes class from the drop list.
558 */
559 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl)
560 {
561 BUG_TRAP(cl->prio_activity);
562
563 htb_deactivate_prios(q, cl);
564 cl->prio_activity = 0;
565 list_del_init(&cl->un.leaf.drop_list);
566 }
567
568 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
569 {
570 int ret;
571 struct htb_sched *q = qdisc_priv(sch);
572 struct htb_class *cl = htb_classify(skb, sch, &ret);
573
574 if (cl == HTB_DIRECT) {
575 /* enqueue to helper queue */
576 if (q->direct_queue.qlen < q->direct_qlen) {
577 __skb_queue_tail(&q->direct_queue, skb);
578 q->direct_pkts++;
579 } else {
580 kfree_skb(skb);
581 sch->qstats.drops++;
582 return NET_XMIT_DROP;
583 }
584 #ifdef CONFIG_NET_CLS_ACT
585 } else if (!cl) {
586 if (ret == NET_XMIT_BYPASS)
587 sch->qstats.drops++;
588 kfree_skb(skb);
589 return ret;
590 #endif
591 } else if (cl->un.leaf.q->enqueue(skb, cl->un.leaf.q) !=
592 NET_XMIT_SUCCESS) {
593 sch->qstats.drops++;
594 cl->qstats.drops++;
595 return NET_XMIT_DROP;
596 } else {
597 cl->bstats.packets +=
598 skb_is_gso(skb)?skb_shinfo(skb)->gso_segs:1;
599 cl->bstats.bytes += skb->len;
600 htb_activate(q, cl);
601 }
602
603 sch->q.qlen++;
604 sch->bstats.packets += skb_is_gso(skb)?skb_shinfo(skb)->gso_segs:1;
605 sch->bstats.bytes += skb->len;
606 return NET_XMIT_SUCCESS;
607 }
608
609 /* TODO: requeuing packet charges it to policers again !! */
610 static int htb_requeue(struct sk_buff *skb, struct Qdisc *sch)
611 {
612 struct htb_sched *q = qdisc_priv(sch);
613 int ret = NET_XMIT_SUCCESS;
614 struct htb_class *cl = htb_classify(skb, sch, &ret);
615 struct sk_buff *tskb;
616
617 if (cl == HTB_DIRECT || !cl) {
618 /* enqueue to helper queue */
619 if (q->direct_queue.qlen < q->direct_qlen && cl) {
620 __skb_queue_head(&q->direct_queue, skb);
621 } else {
622 __skb_queue_head(&q->direct_queue, skb);
623 tskb = __skb_dequeue_tail(&q->direct_queue);
624 kfree_skb(tskb);
625 sch->qstats.drops++;
626 return NET_XMIT_CN;
627 }
628 } else if (cl->un.leaf.q->ops->requeue(skb, cl->un.leaf.q) !=
629 NET_XMIT_SUCCESS) {
630 sch->qstats.drops++;
631 cl->qstats.drops++;
632 return NET_XMIT_DROP;
633 } else
634 htb_activate(q, cl);
635
636 sch->q.qlen++;
637 sch->qstats.requeues++;
638 return NET_XMIT_SUCCESS;
639 }
640
641 /**
642 * htb_charge_class - charges amount "bytes" to leaf and ancestors
643 *
644 * Routine assumes that packet "bytes" long was dequeued from leaf cl
645 * borrowing from "level". It accounts bytes to ceil leaky bucket for
646 * leaf and all ancestors and to rate bucket for ancestors at levels
647 * "level" and higher. It also handles possible change of mode resulting
648 * from the update. Note that mode can also increase here (MAY_BORROW to
649 * CAN_SEND) because we can use more precise clock that event queue here.
650 * In such case we remove class from event queue first.
651 */
652 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl,
653 int level, struct sk_buff *skb)
654 {
655 int bytes = skb->len;
656 long toks, diff;
657 enum htb_cmode old_mode;
658
659 #define HTB_ACCNT(T,B,R) toks = diff + cl->T; \
660 if (toks > cl->B) toks = cl->B; \
661 toks -= L2T(cl, cl->R, bytes); \
662 if (toks <= -cl->mbuffer) toks = 1-cl->mbuffer; \
663 cl->T = toks
664
665 while (cl) {
666 diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
667 if (cl->level >= level) {
668 if (cl->level == level)
669 cl->xstats.lends++;
670 HTB_ACCNT(tokens, buffer, rate);
671 } else {
672 cl->xstats.borrows++;
673 cl->tokens += diff; /* we moved t_c; update tokens */
674 }
675 HTB_ACCNT(ctokens, cbuffer, ceil);
676 cl->t_c = q->now;
677
678 old_mode = cl->cmode;
679 diff = 0;
680 htb_change_class_mode(q, cl, &diff);
681 if (old_mode != cl->cmode) {
682 if (old_mode != HTB_CAN_SEND)
683 htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level);
684 if (cl->cmode != HTB_CAN_SEND)
685 htb_add_to_wait_tree(q, cl, diff);
686 }
687
688 /* update byte stats except for leaves which are already updated */
689 if (cl->level) {
690 cl->bstats.bytes += bytes;
691 cl->bstats.packets += skb_is_gso(skb)?
692 skb_shinfo(skb)->gso_segs:1;
693 }
694 cl = cl->parent;
695 }
696 }
697
698 /**
699 * htb_do_events - make mode changes to classes at the level
700 *
701 * Scans event queue for pending events and applies them. Returns time of
702 * next pending event (0 for no event in pq).
703 * Note: Applied are events whose have cl->pq_key <= q->now.
704 */
705 static psched_time_t htb_do_events(struct htb_sched *q, int level)
706 {
707 int i;
708
709 for (i = 0; i < 500; i++) {
710 struct htb_class *cl;
711 long diff;
712 struct rb_node *p = rb_first(&q->wait_pq[level]);
713
714 if (!p)
715 return 0;
716
717 cl = rb_entry(p, struct htb_class, pq_node);
718 if (cl->pq_key > q->now)
719 return cl->pq_key;
720
721 htb_safe_rb_erase(p, q->wait_pq + level);
722 diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
723 htb_change_class_mode(q, cl, &diff);
724 if (cl->cmode != HTB_CAN_SEND)
725 htb_add_to_wait_tree(q, cl, diff);
726 }
727 if (net_ratelimit())
728 printk(KERN_WARNING "htb: too many events !\n");
729 return q->now + PSCHED_TICKS_PER_SEC / 10;
730 }
731
732 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
733 is no such one exists. */
734 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n,
735 u32 id)
736 {
737 struct rb_node *r = NULL;
738 while (n) {
739 struct htb_class *cl =
740 rb_entry(n, struct htb_class, node[prio]);
741 if (id == cl->classid)
742 return n;
743
744 if (id > cl->classid) {
745 n = n->rb_right;
746 } else {
747 r = n;
748 n = n->rb_left;
749 }
750 }
751 return r;
752 }
753
754 /**
755 * htb_lookup_leaf - returns next leaf class in DRR order
756 *
757 * Find leaf where current feed pointers points to.
758 */
759 static struct htb_class *htb_lookup_leaf(struct rb_root *tree, int prio,
760 struct rb_node **pptr, u32 * pid)
761 {
762 int i;
763 struct {
764 struct rb_node *root;
765 struct rb_node **pptr;
766 u32 *pid;
767 } stk[TC_HTB_MAXDEPTH], *sp = stk;
768
769 BUG_TRAP(tree->rb_node);
770 sp->root = tree->rb_node;
771 sp->pptr = pptr;
772 sp->pid = pid;
773
774 for (i = 0; i < 65535; i++) {
775 if (!*sp->pptr && *sp->pid) {
776 /* ptr was invalidated but id is valid - try to recover
777 the original or next ptr */
778 *sp->pptr =
779 htb_id_find_next_upper(prio, sp->root, *sp->pid);
780 }
781 *sp->pid = 0; /* ptr is valid now so that remove this hint as it
782 can become out of date quickly */
783 if (!*sp->pptr) { /* we are at right end; rewind & go up */
784 *sp->pptr = sp->root;
785 while ((*sp->pptr)->rb_left)
786 *sp->pptr = (*sp->pptr)->rb_left;
787 if (sp > stk) {
788 sp--;
789 BUG_TRAP(*sp->pptr);
790 if (!*sp->pptr)
791 return NULL;
792 htb_next_rb_node(sp->pptr);
793 }
794 } else {
795 struct htb_class *cl;
796 cl = rb_entry(*sp->pptr, struct htb_class, node[prio]);
797 if (!cl->level)
798 return cl;
799 (++sp)->root = cl->un.inner.feed[prio].rb_node;
800 sp->pptr = cl->un.inner.ptr + prio;
801 sp->pid = cl->un.inner.last_ptr_id + prio;
802 }
803 }
804 BUG_TRAP(0);
805 return NULL;
806 }
807
808 /* dequeues packet at given priority and level; call only if
809 you are sure that there is active class at prio/level */
810 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, int prio,
811 int level)
812 {
813 struct sk_buff *skb = NULL;
814 struct htb_class *cl, *start;
815 /* look initial class up in the row */
816 start = cl = htb_lookup_leaf(q->row[level] + prio, prio,
817 q->ptr[level] + prio,
818 q->last_ptr_id[level] + prio);
819
820 do {
821 next:
822 BUG_TRAP(cl);
823 if (!cl)
824 return NULL;
825
826 /* class can be empty - it is unlikely but can be true if leaf
827 qdisc drops packets in enqueue routine or if someone used
828 graft operation on the leaf since last dequeue;
829 simply deactivate and skip such class */
830 if (unlikely(cl->un.leaf.q->q.qlen == 0)) {
831 struct htb_class *next;
832 htb_deactivate(q, cl);
833
834 /* row/level might become empty */
835 if ((q->row_mask[level] & (1 << prio)) == 0)
836 return NULL;
837
838 next = htb_lookup_leaf(q->row[level] + prio,
839 prio, q->ptr[level] + prio,
840 q->last_ptr_id[level] + prio);
841
842 if (cl == start) /* fix start if we just deleted it */
843 start = next;
844 cl = next;
845 goto next;
846 }
847
848 skb = cl->un.leaf.q->dequeue(cl->un.leaf.q);
849 if (likely(skb != NULL))
850 break;
851 if (!cl->warned) {
852 printk(KERN_WARNING
853 "htb: class %X isn't work conserving ?!\n",
854 cl->classid);
855 cl->warned = 1;
856 }
857 q->nwc_hit++;
858 htb_next_rb_node((level ? cl->parent->un.inner.ptr : q->
859 ptr[0]) + prio);
860 cl = htb_lookup_leaf(q->row[level] + prio, prio,
861 q->ptr[level] + prio,
862 q->last_ptr_id[level] + prio);
863
864 } while (cl != start);
865
866 if (likely(skb != NULL)) {
867 if ((cl->un.leaf.deficit[level] -= skb->len) < 0) {
868 cl->un.leaf.deficit[level] += cl->un.leaf.quantum;
869 htb_next_rb_node((level ? cl->parent->un.inner.ptr : q->
870 ptr[0]) + prio);
871 }
872 /* this used to be after charge_class but this constelation
873 gives us slightly better performance */
874 if (!cl->un.leaf.q->q.qlen)
875 htb_deactivate(q, cl);
876 htb_charge_class(q, cl, level, skb);
877 }
878 return skb;
879 }
880
881 static struct sk_buff *htb_dequeue(struct Qdisc *sch)
882 {
883 struct sk_buff *skb = NULL;
884 struct htb_sched *q = qdisc_priv(sch);
885 int level;
886 psched_time_t next_event;
887
888 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
889 skb = __skb_dequeue(&q->direct_queue);
890 if (skb != NULL) {
891 sch->flags &= ~TCQ_F_THROTTLED;
892 sch->q.qlen--;
893 return skb;
894 }
895
896 if (!sch->q.qlen)
897 goto fin;
898 q->now = psched_get_time();
899
900 next_event = q->now + 5 * PSCHED_TICKS_PER_SEC;
901 q->nwc_hit = 0;
902 for (level = 0; level < TC_HTB_MAXDEPTH; level++) {
903 /* common case optimization - skip event handler quickly */
904 int m;
905 psched_time_t event;
906
907 if (q->now >= q->near_ev_cache[level]) {
908 event = htb_do_events(q, level);
909 if (!event)
910 event = q->now + PSCHED_TICKS_PER_SEC;
911 q->near_ev_cache[level] = event;
912 } else
913 event = q->near_ev_cache[level];
914
915 if (event && next_event > event)
916 next_event = event;
917
918 m = ~q->row_mask[level];
919 while (m != (int)(-1)) {
920 int prio = ffz(m);
921 m |= 1 << prio;
922 skb = htb_dequeue_tree(q, prio, level);
923 if (likely(skb != NULL)) {
924 sch->q.qlen--;
925 sch->flags &= ~TCQ_F_THROTTLED;
926 goto fin;
927 }
928 }
929 }
930 sch->qstats.overlimits++;
931 qdisc_watchdog_schedule(&q->watchdog, next_event);
932 fin:
933 return skb;
934 }
935
936 /* try to drop from each class (by prio) until one succeed */
937 static unsigned int htb_drop(struct Qdisc *sch)
938 {
939 struct htb_sched *q = qdisc_priv(sch);
940 int prio;
941
942 for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) {
943 struct list_head *p;
944 list_for_each(p, q->drops + prio) {
945 struct htb_class *cl = list_entry(p, struct htb_class,
946 un.leaf.drop_list);
947 unsigned int len;
948 if (cl->un.leaf.q->ops->drop &&
949 (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) {
950 sch->q.qlen--;
951 if (!cl->un.leaf.q->q.qlen)
952 htb_deactivate(q, cl);
953 return len;
954 }
955 }
956 }
957 return 0;
958 }
959
960 /* reset all classes */
961 /* always caled under BH & queue lock */
962 static void htb_reset(struct Qdisc *sch)
963 {
964 struct htb_sched *q = qdisc_priv(sch);
965 int i;
966
967 for (i = 0; i < HTB_HSIZE; i++) {
968 struct hlist_node *p;
969 struct htb_class *cl;
970
971 hlist_for_each_entry(cl, p, q->hash + i, hlist) {
972 if (cl->level)
973 memset(&cl->un.inner, 0, sizeof(cl->un.inner));
974 else {
975 if (cl->un.leaf.q)
976 qdisc_reset(cl->un.leaf.q);
977 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
978 }
979 cl->prio_activity = 0;
980 cl->cmode = HTB_CAN_SEND;
981
982 }
983 }
984 qdisc_watchdog_cancel(&q->watchdog);
985 __skb_queue_purge(&q->direct_queue);
986 sch->q.qlen = 0;
987 memset(q->row, 0, sizeof(q->row));
988 memset(q->row_mask, 0, sizeof(q->row_mask));
989 memset(q->wait_pq, 0, sizeof(q->wait_pq));
990 memset(q->ptr, 0, sizeof(q->ptr));
991 for (i = 0; i < TC_HTB_NUMPRIO; i++)
992 INIT_LIST_HEAD(q->drops + i);
993 }
994
995 static int htb_init(struct Qdisc *sch, struct nlattr *opt)
996 {
997 struct htb_sched *q = qdisc_priv(sch);
998 struct nlattr *tb[TCA_HTB_INIT + 1];
999 struct tc_htb_glob *gopt;
1000 int i;
1001 if (!opt || nla_parse_nested(tb, TCA_HTB_INIT, opt, NULL) ||
1002 tb[TCA_HTB_INIT] == NULL ||
1003 nla_len(tb[TCA_HTB_INIT]) < sizeof(*gopt)) {
1004 printk(KERN_ERR "HTB: hey probably you have bad tc tool ?\n");
1005 return -EINVAL;
1006 }
1007 gopt = nla_data(tb[TCA_HTB_INIT]);
1008 if (gopt->version != HTB_VER >> 16) {
1009 printk(KERN_ERR
1010 "HTB: need tc/htb version %d (minor is %d), you have %d\n",
1011 HTB_VER >> 16, HTB_VER & 0xffff, gopt->version);
1012 return -EINVAL;
1013 }
1014
1015 INIT_LIST_HEAD(&q->root);
1016 for (i = 0; i < HTB_HSIZE; i++)
1017 INIT_HLIST_HEAD(q->hash + i);
1018 for (i = 0; i < TC_HTB_NUMPRIO; i++)
1019 INIT_LIST_HEAD(q->drops + i);
1020
1021 qdisc_watchdog_init(&q->watchdog, sch);
1022 skb_queue_head_init(&q->direct_queue);
1023
1024 q->direct_qlen = sch->dev->tx_queue_len;
1025 if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */
1026 q->direct_qlen = 2;
1027
1028 if ((q->rate2quantum = gopt->rate2quantum) < 1)
1029 q->rate2quantum = 1;
1030 q->defcls = gopt->defcls;
1031
1032 return 0;
1033 }
1034
1035 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1036 {
1037 struct htb_sched *q = qdisc_priv(sch);
1038 unsigned char *b = skb_tail_pointer(skb);
1039 struct nlattr *nla;
1040 struct tc_htb_glob gopt;
1041 spin_lock_bh(&sch->dev->queue_lock);
1042 gopt.direct_pkts = q->direct_pkts;
1043
1044 gopt.version = HTB_VER;
1045 gopt.rate2quantum = q->rate2quantum;
1046 gopt.defcls = q->defcls;
1047 gopt.debug = 0;
1048 nla = (struct nlattr *)b;
1049 NLA_PUT(skb, TCA_OPTIONS, 0, NULL);
1050 NLA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt);
1051 nla->nla_len = skb_tail_pointer(skb) - b;
1052 spin_unlock_bh(&sch->dev->queue_lock);
1053 return skb->len;
1054 nla_put_failure:
1055 spin_unlock_bh(&sch->dev->queue_lock);
1056 nlmsg_trim(skb, skb_tail_pointer(skb));
1057 return -1;
1058 }
1059
1060 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1061 struct sk_buff *skb, struct tcmsg *tcm)
1062 {
1063 struct htb_class *cl = (struct htb_class *)arg;
1064 unsigned char *b = skb_tail_pointer(skb);
1065 struct nlattr *nla;
1066 struct tc_htb_opt opt;
1067
1068 spin_lock_bh(&sch->dev->queue_lock);
1069 tcm->tcm_parent = cl->parent ? cl->parent->classid : TC_H_ROOT;
1070 tcm->tcm_handle = cl->classid;
1071 if (!cl->level && cl->un.leaf.q)
1072 tcm->tcm_info = cl->un.leaf.q->handle;
1073
1074 nla = (struct nlattr *)b;
1075 NLA_PUT(skb, TCA_OPTIONS, 0, NULL);
1076
1077 memset(&opt, 0, sizeof(opt));
1078
1079 opt.rate = cl->rate->rate;
1080 opt.buffer = cl->buffer;
1081 opt.ceil = cl->ceil->rate;
1082 opt.cbuffer = cl->cbuffer;
1083 opt.quantum = cl->un.leaf.quantum;
1084 opt.prio = cl->un.leaf.prio;
1085 opt.level = cl->level;
1086 NLA_PUT(skb, TCA_HTB_PARMS, sizeof(opt), &opt);
1087 nla->nla_len = skb_tail_pointer(skb) - b;
1088 spin_unlock_bh(&sch->dev->queue_lock);
1089 return skb->len;
1090 nla_put_failure:
1091 spin_unlock_bh(&sch->dev->queue_lock);
1092 nlmsg_trim(skb, b);
1093 return -1;
1094 }
1095
1096 static int
1097 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d)
1098 {
1099 struct htb_class *cl = (struct htb_class *)arg;
1100
1101 if (!cl->level && cl->un.leaf.q)
1102 cl->qstats.qlen = cl->un.leaf.q->q.qlen;
1103 cl->xstats.tokens = cl->tokens;
1104 cl->xstats.ctokens = cl->ctokens;
1105
1106 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
1107 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1108 gnet_stats_copy_queue(d, &cl->qstats) < 0)
1109 return -1;
1110
1111 return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1112 }
1113
1114 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1115 struct Qdisc **old)
1116 {
1117 struct htb_class *cl = (struct htb_class *)arg;
1118
1119 if (cl && !cl->level) {
1120 if (new == NULL &&
1121 (new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
1122 cl->classid))
1123 == NULL)
1124 return -ENOBUFS;
1125 sch_tree_lock(sch);
1126 if ((*old = xchg(&cl->un.leaf.q, new)) != NULL) {
1127 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1128 qdisc_reset(*old);
1129 }
1130 sch_tree_unlock(sch);
1131 return 0;
1132 }
1133 return -ENOENT;
1134 }
1135
1136 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg)
1137 {
1138 struct htb_class *cl = (struct htb_class *)arg;
1139 return (cl && !cl->level) ? cl->un.leaf.q : NULL;
1140 }
1141
1142 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg)
1143 {
1144 struct htb_class *cl = (struct htb_class *)arg;
1145
1146 if (cl->un.leaf.q->q.qlen == 0)
1147 htb_deactivate(qdisc_priv(sch), cl);
1148 }
1149
1150 static unsigned long htb_get(struct Qdisc *sch, u32 classid)
1151 {
1152 struct htb_class *cl = htb_find(classid, sch);
1153 if (cl)
1154 cl->refcnt++;
1155 return (unsigned long)cl;
1156 }
1157
1158 static inline int htb_parent_last_child(struct htb_class *cl)
1159 {
1160 if (!cl->parent)
1161 /* the root class */
1162 return 0;
1163
1164 if (!(cl->parent->children.next == &cl->sibling &&
1165 cl->parent->children.prev == &cl->sibling))
1166 /* not the last child */
1167 return 0;
1168
1169 return 1;
1170 }
1171
1172 static void htb_parent_to_leaf(struct htb_class *cl, struct Qdisc *new_q)
1173 {
1174 struct htb_class *parent = cl->parent;
1175
1176 BUG_TRAP(!cl->level && cl->un.leaf.q && !cl->prio_activity);
1177
1178 parent->level = 0;
1179 memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1180 INIT_LIST_HEAD(&parent->un.leaf.drop_list);
1181 parent->un.leaf.q = new_q ? new_q : &noop_qdisc;
1182 parent->un.leaf.quantum = parent->quantum;
1183 parent->un.leaf.prio = parent->prio;
1184 parent->tokens = parent->buffer;
1185 parent->ctokens = parent->cbuffer;
1186 parent->t_c = psched_get_time();
1187 parent->cmode = HTB_CAN_SEND;
1188 }
1189
1190 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
1191 {
1192 struct htb_sched *q = qdisc_priv(sch);
1193
1194 if (!cl->level) {
1195 BUG_TRAP(cl->un.leaf.q);
1196 qdisc_destroy(cl->un.leaf.q);
1197 }
1198 gen_kill_estimator(&cl->bstats, &cl->rate_est);
1199 qdisc_put_rtab(cl->rate);
1200 qdisc_put_rtab(cl->ceil);
1201
1202 tcf_destroy_chain(cl->filter_list);
1203
1204 while (!list_empty(&cl->children))
1205 htb_destroy_class(sch, list_entry(cl->children.next,
1206 struct htb_class, sibling));
1207
1208 /* note: this delete may happen twice (see htb_delete) */
1209 hlist_del_init(&cl->hlist);
1210 list_del(&cl->sibling);
1211
1212 if (cl->prio_activity)
1213 htb_deactivate(q, cl);
1214
1215 if (cl->cmode != HTB_CAN_SEND)
1216 htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level);
1217
1218 kfree(cl);
1219 }
1220
1221 /* always caled under BH & queue lock */
1222 static void htb_destroy(struct Qdisc *sch)
1223 {
1224 struct htb_sched *q = qdisc_priv(sch);
1225
1226 qdisc_watchdog_cancel(&q->watchdog);
1227 /* This line used to be after htb_destroy_class call below
1228 and surprisingly it worked in 2.4. But it must precede it
1229 because filter need its target class alive to be able to call
1230 unbind_filter on it (without Oops). */
1231 tcf_destroy_chain(q->filter_list);
1232
1233 while (!list_empty(&q->root))
1234 htb_destroy_class(sch, list_entry(q->root.next,
1235 struct htb_class, sibling));
1236
1237 __skb_queue_purge(&q->direct_queue);
1238 }
1239
1240 static int htb_delete(struct Qdisc *sch, unsigned long arg)
1241 {
1242 struct htb_sched *q = qdisc_priv(sch);
1243 struct htb_class *cl = (struct htb_class *)arg;
1244 unsigned int qlen;
1245 struct Qdisc *new_q = NULL;
1246 int last_child = 0;
1247
1248 // TODO: why don't allow to delete subtree ? references ? does
1249 // tc subsys quarantee us that in htb_destroy it holds no class
1250 // refs so that we can remove children safely there ?
1251 if (!list_empty(&cl->children) || cl->filter_cnt)
1252 return -EBUSY;
1253
1254 if (!cl->level && htb_parent_last_child(cl)) {
1255 new_q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
1256 cl->parent->classid);
1257 last_child = 1;
1258 }
1259
1260 sch_tree_lock(sch);
1261
1262 if (!cl->level) {
1263 qlen = cl->un.leaf.q->q.qlen;
1264 qdisc_reset(cl->un.leaf.q);
1265 qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen);
1266 }
1267
1268 /* delete from hash and active; remainder in destroy_class */
1269 hlist_del_init(&cl->hlist);
1270
1271 if (cl->prio_activity)
1272 htb_deactivate(q, cl);
1273
1274 if (last_child)
1275 htb_parent_to_leaf(cl, new_q);
1276
1277 if (--cl->refcnt == 0)
1278 htb_destroy_class(sch, cl);
1279
1280 sch_tree_unlock(sch);
1281 return 0;
1282 }
1283
1284 static void htb_put(struct Qdisc *sch, unsigned long arg)
1285 {
1286 struct htb_class *cl = (struct htb_class *)arg;
1287
1288 if (--cl->refcnt == 0)
1289 htb_destroy_class(sch, cl);
1290 }
1291
1292 static int htb_change_class(struct Qdisc *sch, u32 classid,
1293 u32 parentid, struct nlattr **tca,
1294 unsigned long *arg)
1295 {
1296 int err = -EINVAL;
1297 struct htb_sched *q = qdisc_priv(sch);
1298 struct htb_class *cl = (struct htb_class *)*arg, *parent;
1299 struct nlattr *opt = tca[TCA_OPTIONS];
1300 struct qdisc_rate_table *rtab = NULL, *ctab = NULL;
1301 struct nlattr *tb[TCA_HTB_RTAB + 1];
1302 struct tc_htb_opt *hopt;
1303
1304 /* extract all subattrs from opt attr */
1305 if (!opt || nla_parse_nested(tb, TCA_HTB_RTAB, opt, NULL) ||
1306 tb[TCA_HTB_PARMS] == NULL ||
1307 nla_len(tb[TCA_HTB_PARMS]) < sizeof(*hopt))
1308 goto failure;
1309
1310 parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch);
1311
1312 hopt = nla_data(tb[TCA_HTB_PARMS]);
1313
1314 rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB]);
1315 ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB]);
1316 if (!rtab || !ctab)
1317 goto failure;
1318
1319 if (!cl) { /* new class */
1320 struct Qdisc *new_q;
1321 int prio;
1322 struct {
1323 struct nlattr nla;
1324 struct gnet_estimator opt;
1325 } est = {
1326 .nla = {
1327 .nla_len = nla_attr_size(sizeof(est.opt)),
1328 .nla_type = TCA_RATE,
1329 },
1330 .opt = {
1331 /* 4s interval, 16s averaging constant */
1332 .interval = 2,
1333 .ewma_log = 2,
1334 },
1335 };
1336
1337 /* check for valid classid */
1338 if (!classid || TC_H_MAJ(classid ^ sch->handle)
1339 || htb_find(classid, sch))
1340 goto failure;
1341
1342 /* check maximal depth */
1343 if (parent && parent->parent && parent->parent->level < 2) {
1344 printk(KERN_ERR "htb: tree is too deep\n");
1345 goto failure;
1346 }
1347 err = -ENOBUFS;
1348 if ((cl = kzalloc(sizeof(*cl), GFP_KERNEL)) == NULL)
1349 goto failure;
1350
1351 gen_new_estimator(&cl->bstats, &cl->rate_est,
1352 &sch->dev->queue_lock,
1353 tca[TCA_RATE] ? : &est.nla);
1354 cl->refcnt = 1;
1355 INIT_LIST_HEAD(&cl->sibling);
1356 INIT_HLIST_NODE(&cl->hlist);
1357 INIT_LIST_HEAD(&cl->children);
1358 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
1359 RB_CLEAR_NODE(&cl->pq_node);
1360
1361 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++)
1362 RB_CLEAR_NODE(&cl->node[prio]);
1363
1364 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1365 so that can't be used inside of sch_tree_lock
1366 -- thanks to Karlis Peisenieks */
1367 new_q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, classid);
1368 sch_tree_lock(sch);
1369 if (parent && !parent->level) {
1370 unsigned int qlen = parent->un.leaf.q->q.qlen;
1371
1372 /* turn parent into inner node */
1373 qdisc_reset(parent->un.leaf.q);
1374 qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen);
1375 qdisc_destroy(parent->un.leaf.q);
1376 if (parent->prio_activity)
1377 htb_deactivate(q, parent);
1378
1379 /* remove from evt list because of level change */
1380 if (parent->cmode != HTB_CAN_SEND) {
1381 htb_safe_rb_erase(&parent->pq_node, q->wait_pq);
1382 parent->cmode = HTB_CAN_SEND;
1383 }
1384 parent->level = (parent->parent ? parent->parent->level
1385 : TC_HTB_MAXDEPTH) - 1;
1386 memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1387 }
1388 /* leaf (we) needs elementary qdisc */
1389 cl->un.leaf.q = new_q ? new_q : &noop_qdisc;
1390
1391 cl->classid = classid;
1392 cl->parent = parent;
1393
1394 /* set class to be in HTB_CAN_SEND state */
1395 cl->tokens = hopt->buffer;
1396 cl->ctokens = hopt->cbuffer;
1397 cl->mbuffer = 60 * PSCHED_TICKS_PER_SEC; /* 1min */
1398 cl->t_c = psched_get_time();
1399 cl->cmode = HTB_CAN_SEND;
1400
1401 /* attach to the hash list and parent's family */
1402 hlist_add_head(&cl->hlist, q->hash + htb_hash(classid));
1403 list_add_tail(&cl->sibling,
1404 parent ? &parent->children : &q->root);
1405 } else {
1406 if (tca[TCA_RATE])
1407 gen_replace_estimator(&cl->bstats, &cl->rate_est,
1408 &sch->dev->queue_lock,
1409 tca[TCA_RATE]);
1410 sch_tree_lock(sch);
1411 }
1412
1413 /* it used to be a nasty bug here, we have to check that node
1414 is really leaf before changing cl->un.leaf ! */
1415 if (!cl->level) {
1416 cl->un.leaf.quantum = rtab->rate.rate / q->rate2quantum;
1417 if (!hopt->quantum && cl->un.leaf.quantum < 1000) {
1418 printk(KERN_WARNING
1419 "HTB: quantum of class %X is small. Consider r2q change.\n",
1420 cl->classid);
1421 cl->un.leaf.quantum = 1000;
1422 }
1423 if (!hopt->quantum && cl->un.leaf.quantum > 200000) {
1424 printk(KERN_WARNING
1425 "HTB: quantum of class %X is big. Consider r2q change.\n",
1426 cl->classid);
1427 cl->un.leaf.quantum = 200000;
1428 }
1429 if (hopt->quantum)
1430 cl->un.leaf.quantum = hopt->quantum;
1431 if ((cl->un.leaf.prio = hopt->prio) >= TC_HTB_NUMPRIO)
1432 cl->un.leaf.prio = TC_HTB_NUMPRIO - 1;
1433
1434 /* backup for htb_parent_to_leaf */
1435 cl->quantum = cl->un.leaf.quantum;
1436 cl->prio = cl->un.leaf.prio;
1437 }
1438
1439 cl->buffer = hopt->buffer;
1440 cl->cbuffer = hopt->cbuffer;
1441 if (cl->rate)
1442 qdisc_put_rtab(cl->rate);
1443 cl->rate = rtab;
1444 if (cl->ceil)
1445 qdisc_put_rtab(cl->ceil);
1446 cl->ceil = ctab;
1447 sch_tree_unlock(sch);
1448
1449 *arg = (unsigned long)cl;
1450 return 0;
1451
1452 failure:
1453 if (rtab)
1454 qdisc_put_rtab(rtab);
1455 if (ctab)
1456 qdisc_put_rtab(ctab);
1457 return err;
1458 }
1459
1460 static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg)
1461 {
1462 struct htb_sched *q = qdisc_priv(sch);
1463 struct htb_class *cl = (struct htb_class *)arg;
1464 struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list;
1465
1466 return fl;
1467 }
1468
1469 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
1470 u32 classid)
1471 {
1472 struct htb_sched *q = qdisc_priv(sch);
1473 struct htb_class *cl = htb_find(classid, sch);
1474
1475 /*if (cl && !cl->level) return 0;
1476 The line above used to be there to prevent attaching filters to
1477 leaves. But at least tc_index filter uses this just to get class
1478 for other reasons so that we have to allow for it.
1479 ----
1480 19.6.2002 As Werner explained it is ok - bind filter is just
1481 another way to "lock" the class - unlike "get" this lock can
1482 be broken by class during destroy IIUC.
1483 */
1484 if (cl)
1485 cl->filter_cnt++;
1486 else
1487 q->filter_cnt++;
1488 return (unsigned long)cl;
1489 }
1490
1491 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
1492 {
1493 struct htb_sched *q = qdisc_priv(sch);
1494 struct htb_class *cl = (struct htb_class *)arg;
1495
1496 if (cl)
1497 cl->filter_cnt--;
1498 else
1499 q->filter_cnt--;
1500 }
1501
1502 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1503 {
1504 struct htb_sched *q = qdisc_priv(sch);
1505 int i;
1506
1507 if (arg->stop)
1508 return;
1509
1510 for (i = 0; i < HTB_HSIZE; i++) {
1511 struct hlist_node *p;
1512 struct htb_class *cl;
1513
1514 hlist_for_each_entry(cl, p, q->hash + i, hlist) {
1515 if (arg->count < arg->skip) {
1516 arg->count++;
1517 continue;
1518 }
1519 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1520 arg->stop = 1;
1521 return;
1522 }
1523 arg->count++;
1524 }
1525 }
1526 }
1527
1528 static const struct Qdisc_class_ops htb_class_ops = {
1529 .graft = htb_graft,
1530 .leaf = htb_leaf,
1531 .qlen_notify = htb_qlen_notify,
1532 .get = htb_get,
1533 .put = htb_put,
1534 .change = htb_change_class,
1535 .delete = htb_delete,
1536 .walk = htb_walk,
1537 .tcf_chain = htb_find_tcf,
1538 .bind_tcf = htb_bind_filter,
1539 .unbind_tcf = htb_unbind_filter,
1540 .dump = htb_dump_class,
1541 .dump_stats = htb_dump_class_stats,
1542 };
1543
1544 static struct Qdisc_ops htb_qdisc_ops __read_mostly = {
1545 .next = NULL,
1546 .cl_ops = &htb_class_ops,
1547 .id = "htb",
1548 .priv_size = sizeof(struct htb_sched),
1549 .enqueue = htb_enqueue,
1550 .dequeue = htb_dequeue,
1551 .requeue = htb_requeue,
1552 .drop = htb_drop,
1553 .init = htb_init,
1554 .reset = htb_reset,
1555 .destroy = htb_destroy,
1556 .change = NULL /* htb_change */,
1557 .dump = htb_dump,
1558 .owner = THIS_MODULE,
1559 };
1560
1561 static int __init htb_module_init(void)
1562 {
1563 return register_qdisc(&htb_qdisc_ops);
1564 }
1565 static void __exit htb_module_exit(void)
1566 {
1567 unregister_qdisc(&htb_qdisc_ops);
1568 }
1569
1570 module_init(htb_module_init)
1571 module_exit(htb_module_exit)
1572 MODULE_LICENSE("GPL");
This page took 0.091529 seconds and 5 git commands to generate.