Version:  2.0.40 2.2.26 2.4.37 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5

Linux/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 #include <linux/module.h>
 29 #include <linux/moduleparam.h>
 30 #include <linux/types.h>
 31 #include <linux/kernel.h>
 32 #include <linux/string.h>
 33 #include <linux/errno.h>
 34 #include <linux/skbuff.h>
 35 #include <linux/list.h>
 36 #include <linux/compiler.h>
 37 #include <linux/rbtree.h>
 38 #include <linux/workqueue.h>
 39 #include <linux/slab.h>
 40 #include <net/netlink.h>
 41 #include <net/sch_generic.h>
 42 #include <net/pkt_sched.h>
 43 
 44 /* HTB algorithm.
 45     Author: devik@cdi.cz
 46     ========================================================================
 47     HTB is like TBF with multiple classes. It is also similar to CBQ because
 48     it allows to assign priority to each class in hierarchy.
 49     In fact it is another implementation of Floyd's formal sharing.
 50 
 51     Levels:
 52     Each class is assigned level. Leaf has ALWAYS level 0 and root
 53     classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level
 54     one less than their parent.
 55 */
 56 
 57 static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */
 58 #define HTB_VER 0x30011         /* major must be matched with number suplied by TC as version */
 59 
 60 #if HTB_VER >> 16 != TC_HTB_PROTOVER
 61 #error "Mismatched sch_htb.c and pkt_sch.h"
 62 #endif
 63 
 64 /* Module parameter and sysfs export */
 65 module_param    (htb_hysteresis, int, 0640);
 66 MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate");
 67 
 68 static int htb_rate_est = 0; /* htb classes have a default rate estimator */
 69 module_param(htb_rate_est, int, 0640);
 70 MODULE_PARM_DESC(htb_rate_est, "setup a default rate estimator (4sec 16sec) for htb classes");
 71 
 72 /* used internaly to keep status of single class */
 73 enum htb_cmode {
 74         HTB_CANT_SEND,          /* class can't send and can't borrow */
 75         HTB_MAY_BORROW,         /* class can't send but may borrow */
 76         HTB_CAN_SEND            /* class can send */
 77 };
 78 
 79 struct htb_prio {
 80         union {
 81                 struct rb_root  row;
 82                 struct rb_root  feed;
 83         };
 84         struct rb_node  *ptr;
 85         /* When class changes from state 1->2 and disconnects from
 86          * parent's feed then we lost ptr value and start from the
 87          * first child again. Here we store classid of the
 88          * last valid ptr (used when ptr is NULL).
 89          */
 90         u32             last_ptr_id;
 91 };
 92 
 93 /* interior & leaf nodes; props specific to leaves are marked L:
 94  * To reduce false sharing, place mostly read fields at beginning,
 95  * and mostly written ones at the end.
 96  */
 97 struct htb_class {
 98         struct Qdisc_class_common common;
 99         struct psched_ratecfg   rate;
100         struct psched_ratecfg   ceil;
101         s64                     buffer, cbuffer;/* token bucket depth/rate */
102         s64                     mbuffer;        /* max wait time */
103         u32                     prio;           /* these two are used only by leaves... */
104         int                     quantum;        /* but stored for parent-to-leaf return */
105 
106         struct tcf_proto __rcu  *filter_list;   /* class attached filters */
107         int                     filter_cnt;
108         int                     refcnt;         /* usage count of this class */
109 
110         int                     level;          /* our level (see above) */
111         unsigned int            children;
112         struct htb_class        *parent;        /* parent class */
113 
114         struct gnet_stats_rate_est64 rate_est;
115 
116         /*
117          * Written often fields
118          */
119         struct gnet_stats_basic_packed bstats;
120         struct gnet_stats_queue qstats;
121         struct tc_htb_xstats    xstats; /* our special stats */
122 
123         /* token bucket parameters */
124         s64                     tokens, ctokens;/* current number of tokens */
125         s64                     t_c;            /* checkpoint time */
126 
127         union {
128                 struct htb_class_leaf {
129                         struct list_head drop_list;
130                         int             deficit[TC_HTB_MAXDEPTH];
131                         struct Qdisc    *q;
132                 } leaf;
133                 struct htb_class_inner {
134                         struct htb_prio clprio[TC_HTB_NUMPRIO];
135                 } inner;
136         } un;
137         s64                     pq_key;
138 
139         int                     prio_activity;  /* for which prios are we active */
140         enum htb_cmode          cmode;          /* current mode of the class */
141         struct rb_node          pq_node;        /* node for event queue */
142         struct rb_node          node[TC_HTB_NUMPRIO];   /* node for self or feed tree */
143 };
144 
145 struct htb_level {
146         struct rb_root  wait_pq;
147         struct htb_prio hprio[TC_HTB_NUMPRIO];
148 };
149 
150 struct htb_sched {
151         struct Qdisc_class_hash clhash;
152         int                     defcls;         /* class where unclassified flows go to */
153         int                     rate2quantum;   /* quant = rate / rate2quantum */
154 
155         /* filters for qdisc itself */
156         struct tcf_proto __rcu  *filter_list;
157 
158 #define HTB_WARN_TOOMANYEVENTS  0x1
159         unsigned int            warned; /* only one warning */
160         int                     direct_qlen;
161         struct work_struct      work;
162 
163         /* non shaped skbs; let them go directly thru */
164         struct sk_buff_head     direct_queue;
165         long                    direct_pkts;
166 
167         struct qdisc_watchdog   watchdog;
168 
169         s64                     now;    /* cached dequeue time */
170         struct list_head        drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */
171 
172         /* time of nearest event per level (row) */
173         s64                     near_ev_cache[TC_HTB_MAXDEPTH];
174 
175         int                     row_mask[TC_HTB_MAXDEPTH];
176 
177         struct htb_level        hlevel[TC_HTB_MAXDEPTH];
178 };
179 
180 /* find class in global hash table using given handle */
181 static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch)
182 {
183         struct htb_sched *q = qdisc_priv(sch);
184         struct Qdisc_class_common *clc;
185 
186         clc = qdisc_class_find(&q->clhash, handle);
187         if (clc == NULL)
188                 return NULL;
189         return container_of(clc, struct htb_class, common);
190 }
191 
192 /**
193  * htb_classify - classify a packet into class
194  *
195  * It returns NULL if the packet should be dropped or -1 if the packet
196  * should be passed directly thru. In all other cases leaf class is returned.
197  * We allow direct class selection by classid in priority. The we examine
198  * filters in qdisc and in inner nodes (if higher filter points to the inner
199  * node). If we end up with classid MAJOR:0 we enqueue the skb into special
200  * internal fifo (direct). These packets then go directly thru. If we still
201  * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful
202  * then finish and return direct queue.
203  */
204 #define HTB_DIRECT ((struct htb_class *)-1L)
205 
206 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch,
207                                       int *qerr)
208 {
209         struct htb_sched *q = qdisc_priv(sch);
210         struct htb_class *cl;
211         struct tcf_result res;
212         struct tcf_proto *tcf;
213         int result;
214 
215         /* allow to select class by setting skb->priority to valid classid;
216          * note that nfmark can be used too by attaching filter fw with no
217          * rules in it
218          */
219         if (skb->priority == sch->handle)
220                 return HTB_DIRECT;      /* X:0 (direct flow) selected */
221         cl = htb_find(skb->priority, sch);
222         if (cl) {
223                 if (cl->level == 0)
224                         return cl;
225                 /* Start with inner filter chain if a non-leaf class is selected */
226                 tcf = rcu_dereference_bh(cl->filter_list);
227         } else {
228                 tcf = rcu_dereference_bh(q->filter_list);
229         }
230 
231         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
232         while (tcf && (result = tc_classify(skb, tcf, &res, false)) >= 0) {
233 #ifdef CONFIG_NET_CLS_ACT
234                 switch (result) {
235                 case TC_ACT_QUEUED:
236                 case TC_ACT_STOLEN:
237                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
238                 case TC_ACT_SHOT:
239                         return NULL;
240                 }
241 #endif
242                 cl = (void *)res.class;
243                 if (!cl) {
244                         if (res.classid == sch->handle)
245                                 return HTB_DIRECT;      /* X:0 (direct flow) */
246                         cl = htb_find(res.classid, sch);
247                         if (!cl)
248                                 break;  /* filter selected invalid classid */
249                 }
250                 if (!cl->level)
251                         return cl;      /* we hit leaf; return it */
252 
253                 /* we have got inner class; apply inner filter chain */
254                 tcf = rcu_dereference_bh(cl->filter_list);
255         }
256         /* classification failed; try to use default class */
257         cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
258         if (!cl || cl->level)
259                 return HTB_DIRECT;      /* bad default .. this is safe bet */
260         return cl;
261 }
262 
263 /**
264  * htb_add_to_id_tree - adds class to the round robin list
265  *
266  * Routine adds class to the list (actually tree) sorted by classid.
267  * Make sure that class is not already on such list for given prio.
268  */
269 static void htb_add_to_id_tree(struct rb_root *root,
270                                struct htb_class *cl, int prio)
271 {
272         struct rb_node **p = &root->rb_node, *parent = NULL;
273 
274         while (*p) {
275                 struct htb_class *c;
276                 parent = *p;
277                 c = rb_entry(parent, struct htb_class, node[prio]);
278 
279                 if (cl->common.classid > c->common.classid)
280                         p = &parent->rb_right;
281                 else
282                         p = &parent->rb_left;
283         }
284         rb_link_node(&cl->node[prio], parent, p);
285         rb_insert_color(&cl->node[prio], root);
286 }
287 
288 /**
289  * htb_add_to_wait_tree - adds class to the event queue with delay
290  *
291  * The class is added to priority event queue to indicate that class will
292  * change its mode in cl->pq_key microseconds. Make sure that class is not
293  * already in the queue.
294  */
295 static void htb_add_to_wait_tree(struct htb_sched *q,
296                                  struct htb_class *cl, s64 delay)
297 {
298         struct rb_node **p = &q->hlevel[cl->level].wait_pq.rb_node, *parent = NULL;
299 
300         cl->pq_key = q->now + delay;
301         if (cl->pq_key == q->now)
302                 cl->pq_key++;
303 
304         /* update the nearest event cache */
305         if (q->near_ev_cache[cl->level] > cl->pq_key)
306                 q->near_ev_cache[cl->level] = cl->pq_key;
307 
308         while (*p) {
309                 struct htb_class *c;
310                 parent = *p;
311                 c = rb_entry(parent, struct htb_class, pq_node);
312                 if (cl->pq_key >= c->pq_key)
313                         p = &parent->rb_right;
314                 else
315                         p = &parent->rb_left;
316         }
317         rb_link_node(&cl->pq_node, parent, p);
318         rb_insert_color(&cl->pq_node, &q->hlevel[cl->level].wait_pq);
319 }
320 
321 /**
322  * htb_next_rb_node - finds next node in binary tree
323  *
324  * When we are past last key we return NULL.
325  * Average complexity is 2 steps per call.
326  */
327 static inline void htb_next_rb_node(struct rb_node **n)
328 {
329         *n = rb_next(*n);
330 }
331 
332 /**
333  * htb_add_class_to_row - add class to its row
334  *
335  * The class is added to row at priorities marked in mask.
336  * It does nothing if mask == 0.
337  */
338 static inline void htb_add_class_to_row(struct htb_sched *q,
339                                         struct htb_class *cl, int mask)
340 {
341         q->row_mask[cl->level] |= mask;
342         while (mask) {
343                 int prio = ffz(~mask);
344                 mask &= ~(1 << prio);
345                 htb_add_to_id_tree(&q->hlevel[cl->level].hprio[prio].row, cl, prio);
346         }
347 }
348 
349 /* If this triggers, it is a bug in this code, but it need not be fatal */
350 static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root)
351 {
352         if (RB_EMPTY_NODE(rb)) {
353                 WARN_ON(1);
354         } else {
355                 rb_erase(rb, root);
356                 RB_CLEAR_NODE(rb);
357         }
358 }
359 
360 
361 /**
362  * htb_remove_class_from_row - removes class from its row
363  *
364  * The class is removed from row at priorities marked in mask.
365  * It does nothing if mask == 0.
366  */
367 static inline void htb_remove_class_from_row(struct htb_sched *q,
368                                                  struct htb_class *cl, int mask)
369 {
370         int m = 0;
371         struct htb_level *hlevel = &q->hlevel[cl->level];
372 
373         while (mask) {
374                 int prio = ffz(~mask);
375                 struct htb_prio *hprio = &hlevel->hprio[prio];
376 
377                 mask &= ~(1 << prio);
378                 if (hprio->ptr == cl->node + prio)
379                         htb_next_rb_node(&hprio->ptr);
380 
381                 htb_safe_rb_erase(cl->node + prio, &hprio->row);
382                 if (!hprio->row.rb_node)
383                         m |= 1 << prio;
384         }
385         q->row_mask[cl->level] &= ~m;
386 }
387 
388 /**
389  * htb_activate_prios - creates active classe's feed chain
390  *
391  * The class is connected to ancestors and/or appropriate rows
392  * for priorities it is participating on. cl->cmode must be new
393  * (activated) mode. It does nothing if cl->prio_activity == 0.
394  */
395 static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl)
396 {
397         struct htb_class *p = cl->parent;
398         long m, mask = cl->prio_activity;
399 
400         while (cl->cmode == HTB_MAY_BORROW && p && mask) {
401                 m = mask;
402                 while (m) {
403                         int prio = ffz(~m);
404                         m &= ~(1 << prio);
405 
406                         if (p->un.inner.clprio[prio].feed.rb_node)
407                                 /* parent already has its feed in use so that
408                                  * reset bit in mask as parent is already ok
409                                  */
410                                 mask &= ~(1 << prio);
411 
412                         htb_add_to_id_tree(&p->un.inner.clprio[prio].feed, 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.clprio[prio].ptr == cl->node + prio) {
443                                 /* we are removing child which is pointed to from
444                                  * parent feed - forget the pointer but remember
445                                  * classid
446                                  */
447                                 p->un.inner.clprio[prio].last_ptr_id = cl->common.classid;
448                                 p->un.inner.clprio[prio].ptr = NULL;
449                         }
450 
451                         htb_safe_rb_erase(cl->node + prio,
452                                           &p->un.inner.clprio[prio].feed);
453 
454                         if (!p->un.inner.clprio[prio].feed.rb_node)
455                                 mask |= 1 << prio;
456                 }
457 
458                 p->prio_activity &= ~mask;
459                 cl = p;
460                 p = cl->parent;
461 
462         }
463         if (cl->cmode == HTB_CAN_SEND && mask)
464                 htb_remove_class_from_row(q, cl, mask);
465 }
466 
467 static inline s64 htb_lowater(const struct htb_class *cl)
468 {
469         if (htb_hysteresis)
470                 return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0;
471         else
472                 return 0;
473 }
474 static inline s64 htb_hiwater(const struct htb_class *cl)
475 {
476         if (htb_hysteresis)
477                 return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0;
478         else
479                 return 0;
480 }
481 
482 
483 /**
484  * htb_class_mode - computes and returns current class mode
485  *
486  * It computes cl's mode at time cl->t_c+diff and returns it. If mode
487  * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
488  * from now to time when cl will change its state.
489  * Also it is worth to note that class mode doesn't change simply
490  * at cl->{c,}tokens == 0 but there can rather be hysteresis of
491  * 0 .. -cl->{c,}buffer range. It is meant to limit number of
492  * mode transitions per time unit. The speed gain is about 1/6.
493  */
494 static inline enum htb_cmode
495 htb_class_mode(struct htb_class *cl, s64 *diff)
496 {
497         s64 toks;
498 
499         if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) {
500                 *diff = -toks;
501                 return HTB_CANT_SEND;
502         }
503 
504         if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl))
505                 return HTB_CAN_SEND;
506 
507         *diff = -toks;
508         return HTB_MAY_BORROW;
509 }
510 
511 /**
512  * htb_change_class_mode - changes classe's mode
513  *
514  * This should be the only way how to change classe's mode under normal
515  * cirsumstances. Routine will update feed lists linkage, change mode
516  * and add class to the wait event queue if appropriate. New mode should
517  * be different from old one and cl->pq_key has to be valid if changing
518  * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
519  */
520 static void
521 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, s64 *diff)
522 {
523         enum htb_cmode new_mode = htb_class_mode(cl, diff);
524 
525         if (new_mode == cl->cmode)
526                 return;
527 
528         if (cl->prio_activity) {        /* not necessary: speed optimization */
529                 if (cl->cmode != HTB_CANT_SEND)
530                         htb_deactivate_prios(q, cl);
531                 cl->cmode = new_mode;
532                 if (new_mode != HTB_CANT_SEND)
533                         htb_activate_prios(q, cl);
534         } else
535                 cl->cmode = new_mode;
536 }
537 
538 /**
539  * htb_activate - inserts leaf cl into appropriate active feeds
540  *
541  * Routine learns (new) priority of leaf and activates feed chain
542  * for the prio. It can be called on already active leaf safely.
543  * It also adds leaf into droplist.
544  */
545 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl)
546 {
547         WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen);
548 
549         if (!cl->prio_activity) {
550                 cl->prio_activity = 1 << cl->prio;
551                 htb_activate_prios(q, cl);
552                 list_add_tail(&cl->un.leaf.drop_list,
553                               q->drops + cl->prio);
554         }
555 }
556 
557 /**
558  * htb_deactivate - remove leaf cl from active feeds
559  *
560  * Make sure that leaf is active. In the other words it can't be called
561  * with non-active leaf. It also removes class from the drop list.
562  */
563 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl)
564 {
565         WARN_ON(!cl->prio_activity);
566 
567         htb_deactivate_prios(q, cl);
568         cl->prio_activity = 0;
569         list_del_init(&cl->un.leaf.drop_list);
570 }
571 
572 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
573 {
574         int uninitialized_var(ret);
575         struct htb_sched *q = qdisc_priv(sch);
576         struct htb_class *cl = htb_classify(skb, sch, &ret);
577 
578         if (cl == HTB_DIRECT) {
579                 /* enqueue to helper queue */
580                 if (q->direct_queue.qlen < q->direct_qlen) {
581                         __skb_queue_tail(&q->direct_queue, skb);
582                         q->direct_pkts++;
583                 } else {
584                         return qdisc_drop(skb, sch);
585                 }
586 #ifdef CONFIG_NET_CLS_ACT
587         } else if (!cl) {
588                 if (ret & __NET_XMIT_BYPASS)
589                         qdisc_qstats_drop(sch);
590                 kfree_skb(skb);
591                 return ret;
592 #endif
593         } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q)) != NET_XMIT_SUCCESS) {
594                 if (net_xmit_drop_count(ret)) {
595                         qdisc_qstats_drop(sch);
596                         cl->qstats.drops++;
597                 }
598                 return ret;
599         } else {
600                 htb_activate(q, cl);
601         }
602 
603         sch->q.qlen++;
604         return NET_XMIT_SUCCESS;
605 }
606 
607 static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, s64 diff)
608 {
609         s64 toks = diff + cl->tokens;
610 
611         if (toks > cl->buffer)
612                 toks = cl->buffer;
613         toks -= (s64) psched_l2t_ns(&cl->rate, bytes);
614         if (toks <= -cl->mbuffer)
615                 toks = 1 - cl->mbuffer;
616 
617         cl->tokens = toks;
618 }
619 
620 static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, s64 diff)
621 {
622         s64 toks = diff + cl->ctokens;
623 
624         if (toks > cl->cbuffer)
625                 toks = cl->cbuffer;
626         toks -= (s64) psched_l2t_ns(&cl->ceil, bytes);
627         if (toks <= -cl->mbuffer)
628                 toks = 1 - cl->mbuffer;
629 
630         cl->ctokens = toks;
631 }
632 
633 /**
634  * htb_charge_class - charges amount "bytes" to leaf and ancestors
635  *
636  * Routine assumes that packet "bytes" long was dequeued from leaf cl
637  * borrowing from "level". It accounts bytes to ceil leaky bucket for
638  * leaf and all ancestors and to rate bucket for ancestors at levels
639  * "level" and higher. It also handles possible change of mode resulting
640  * from the update. Note that mode can also increase here (MAY_BORROW to
641  * CAN_SEND) because we can use more precise clock that event queue here.
642  * In such case we remove class from event queue first.
643  */
644 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl,
645                              int level, struct sk_buff *skb)
646 {
647         int bytes = qdisc_pkt_len(skb);
648         enum htb_cmode old_mode;
649         s64 diff;
650 
651         while (cl) {
652                 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer);
653                 if (cl->level >= level) {
654                         if (cl->level == level)
655                                 cl->xstats.lends++;
656                         htb_accnt_tokens(cl, bytes, diff);
657                 } else {
658                         cl->xstats.borrows++;
659                         cl->tokens += diff;     /* we moved t_c; update tokens */
660                 }
661                 htb_accnt_ctokens(cl, bytes, diff);
662                 cl->t_c = q->now;
663 
664                 old_mode = cl->cmode;
665                 diff = 0;
666                 htb_change_class_mode(q, cl, &diff);
667                 if (old_mode != cl->cmode) {
668                         if (old_mode != HTB_CAN_SEND)
669                                 htb_safe_rb_erase(&cl->pq_node, &q->hlevel[cl->level].wait_pq);
670                         if (cl->cmode != HTB_CAN_SEND)
671                                 htb_add_to_wait_tree(q, cl, diff);
672                 }
673 
674                 /* update basic stats except for leaves which are already updated */
675                 if (cl->level)
676                         bstats_update(&cl->bstats, skb);
677 
678                 cl = cl->parent;
679         }
680 }
681 
682 /**
683  * htb_do_events - make mode changes to classes at the level
684  *
685  * Scans event queue for pending events and applies them. Returns time of
686  * next pending event (0 for no event in pq, q->now for too many events).
687  * Note: Applied are events whose have cl->pq_key <= q->now.
688  */
689 static s64 htb_do_events(struct htb_sched *q, const int level,
690                          unsigned long start)
691 {
692         /* don't run for longer than 2 jiffies; 2 is used instead of
693          * 1 to simplify things when jiffy is going to be incremented
694          * too soon
695          */
696         unsigned long stop_at = start + 2;
697         struct rb_root *wait_pq = &q->hlevel[level].wait_pq;
698 
699         while (time_before(jiffies, stop_at)) {
700                 struct htb_class *cl;
701                 s64 diff;
702                 struct rb_node *p = rb_first(wait_pq);
703 
704                 if (!p)
705                         return 0;
706 
707                 cl = rb_entry(p, struct htb_class, pq_node);
708                 if (cl->pq_key > q->now)
709                         return cl->pq_key;
710 
711                 htb_safe_rb_erase(p, wait_pq);
712                 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer);
713                 htb_change_class_mode(q, cl, &diff);
714                 if (cl->cmode != HTB_CAN_SEND)
715                         htb_add_to_wait_tree(q, cl, diff);
716         }
717 
718         /* too much load - let's continue after a break for scheduling */
719         if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) {
720                 pr_warn("htb: too many events!\n");
721                 q->warned |= HTB_WARN_TOOMANYEVENTS;
722         }
723 
724         return q->now;
725 }
726 
727 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
728  * is no such one exists.
729  */
730 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n,
731                                               u32 id)
732 {
733         struct rb_node *r = NULL;
734         while (n) {
735                 struct htb_class *cl =
736                     rb_entry(n, struct htb_class, node[prio]);
737 
738                 if (id > cl->common.classid) {
739                         n = n->rb_right;
740                 } else if (id < cl->common.classid) {
741                         r = n;
742                         n = n->rb_left;
743                 } else {
744                         return n;
745                 }
746         }
747         return r;
748 }
749 
750 /**
751  * htb_lookup_leaf - returns next leaf class in DRR order
752  *
753  * Find leaf where current feed pointers points to.
754  */
755 static struct htb_class *htb_lookup_leaf(struct htb_prio *hprio, const int prio)
756 {
757         int i;
758         struct {
759                 struct rb_node *root;
760                 struct rb_node **pptr;
761                 u32 *pid;
762         } stk[TC_HTB_MAXDEPTH], *sp = stk;
763 
764         BUG_ON(!hprio->row.rb_node);
765         sp->root = hprio->row.rb_node;
766         sp->pptr = &hprio->ptr;
767         sp->pid = &hprio->last_ptr_id;
768 
769         for (i = 0; i < 65535; i++) {
770                 if (!*sp->pptr && *sp->pid) {
771                         /* ptr was invalidated but id is valid - try to recover
772                          * the original or next ptr
773                          */
774                         *sp->pptr =
775                             htb_id_find_next_upper(prio, sp->root, *sp->pid);
776                 }
777                 *sp->pid = 0;   /* ptr is valid now so that remove this hint as it
778                                  * can become out of date quickly
779                                  */
780                 if (!*sp->pptr) {       /* we are at right end; rewind & go up */
781                         *sp->pptr = sp->root;
782                         while ((*sp->pptr)->rb_left)
783                                 *sp->pptr = (*sp->pptr)->rb_left;
784                         if (sp > stk) {
785                                 sp--;
786                                 if (!*sp->pptr) {
787                                         WARN_ON(1);
788                                         return NULL;
789                                 }
790                                 htb_next_rb_node(sp->pptr);
791                         }
792                 } else {
793                         struct htb_class *cl;
794                         struct htb_prio *clp;
795 
796                         cl = rb_entry(*sp->pptr, struct htb_class, node[prio]);
797                         if (!cl->level)
798                                 return cl;
799                         clp = &cl->un.inner.clprio[prio];
800                         (++sp)->root = clp->feed.rb_node;
801                         sp->pptr = &clp->ptr;
802                         sp->pid = &clp->last_ptr_id;
803                 }
804         }
805         WARN_ON(1);
806         return NULL;
807 }
808 
809 /* dequeues packet at given priority and level; call only if
810  * you are sure that there is active class at prio/level
811  */
812 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, const int prio,
813                                         const int level)
814 {
815         struct sk_buff *skb = NULL;
816         struct htb_class *cl, *start;
817         struct htb_level *hlevel = &q->hlevel[level];
818         struct htb_prio *hprio = &hlevel->hprio[prio];
819 
820         /* look initial class up in the row */
821         start = cl = htb_lookup_leaf(hprio, prio);
822 
823         do {
824 next:
825                 if (unlikely(!cl))
826                         return NULL;
827 
828                 /* class can be empty - it is unlikely but can be true if leaf
829                  * qdisc drops packets in enqueue routine or if someone used
830                  * graft operation on the leaf since last dequeue;
831                  * simply deactivate and skip such class
832                  */
833                 if (unlikely(cl->un.leaf.q->q.qlen == 0)) {
834                         struct htb_class *next;
835                         htb_deactivate(q, cl);
836 
837                         /* row/level might become empty */
838                         if ((q->row_mask[level] & (1 << prio)) == 0)
839                                 return NULL;
840 
841                         next = htb_lookup_leaf(hprio, prio);
842 
843                         if (cl == start)        /* fix start if we just deleted it */
844                                 start = next;
845                         cl = next;
846                         goto next;
847                 }
848 
849                 skb = cl->un.leaf.q->dequeue(cl->un.leaf.q);
850                 if (likely(skb != NULL))
851                         break;
852 
853                 qdisc_warn_nonwc("htb", cl->un.leaf.q);
854                 htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr:
855                                          &q->hlevel[0].hprio[prio].ptr);
856                 cl = htb_lookup_leaf(hprio, prio);
857 
858         } while (cl != start);
859 
860         if (likely(skb != NULL)) {
861                 bstats_update(&cl->bstats, skb);
862                 cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb);
863                 if (cl->un.leaf.deficit[level] < 0) {
864                         cl->un.leaf.deficit[level] += cl->quantum;
865                         htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr :
866                                                  &q->hlevel[0].hprio[prio].ptr);
867                 }
868                 /* this used to be after charge_class but this constelation
869                  * gives us slightly better performance
870                  */
871                 if (!cl->un.leaf.q->q.qlen)
872                         htb_deactivate(q, cl);
873                 htb_charge_class(q, cl, level, skb);
874         }
875         return skb;
876 }
877 
878 static struct sk_buff *htb_dequeue(struct Qdisc *sch)
879 {
880         struct sk_buff *skb;
881         struct htb_sched *q = qdisc_priv(sch);
882         int level;
883         s64 next_event;
884         unsigned long start_at;
885 
886         /* try to dequeue direct packets as high prio (!) to minimize cpu work */
887         skb = __skb_dequeue(&q->direct_queue);
888         if (skb != NULL) {
889 ok:
890                 qdisc_bstats_update(sch, skb);
891                 qdisc_unthrottled(sch);
892                 sch->q.qlen--;
893                 return skb;
894         }
895 
896         if (!sch->q.qlen)
897                 goto fin;
898         q->now = ktime_get_ns();
899         start_at = jiffies;
900 
901         next_event = q->now + 5LLU * NSEC_PER_SEC;
902 
903         for (level = 0; level < TC_HTB_MAXDEPTH; level++) {
904                 /* common case optimization - skip event handler quickly */
905                 int m;
906                 s64 event = q->near_ev_cache[level];
907 
908                 if (q->now >= event) {
909                         event = htb_do_events(q, level, start_at);
910                         if (!event)
911                                 event = q->now + NSEC_PER_SEC;
912                         q->near_ev_cache[level] = event;
913                 }
914 
915                 if (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 
922                         m |= 1 << prio;
923                         skb = htb_dequeue_tree(q, prio, level);
924                         if (likely(skb != NULL))
925                                 goto ok;
926                 }
927         }
928         qdisc_qstats_overlimit(sch);
929         if (likely(next_event > q->now)) {
930                 if (!test_bit(__QDISC_STATE_DEACTIVATED,
931                               &qdisc_root_sleeping(q->watchdog.qdisc)->state)) {
932                         ktime_t time = ns_to_ktime(next_event);
933                         qdisc_throttled(q->watchdog.qdisc);
934                         hrtimer_start(&q->watchdog.timer, time,
935                                       HRTIMER_MODE_ABS_PINNED);
936                 }
937         } else {
938                 schedule_work(&q->work);
939         }
940 fin:
941         return skb;
942 }
943 
944 /* try to drop from each class (by prio) until one succeed */
945 static unsigned int htb_drop(struct Qdisc *sch)
946 {
947         struct htb_sched *q = qdisc_priv(sch);
948         int prio;
949 
950         for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) {
951                 struct list_head *p;
952                 list_for_each(p, q->drops + prio) {
953                         struct htb_class *cl = list_entry(p, struct htb_class,
954                                                           un.leaf.drop_list);
955                         unsigned int len;
956                         if (cl->un.leaf.q->ops->drop &&
957                             (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) {
958                                 sch->q.qlen--;
959                                 if (!cl->un.leaf.q->q.qlen)
960                                         htb_deactivate(q, cl);
961                                 return len;
962                         }
963                 }
964         }
965         return 0;
966 }
967 
968 /* reset all classes */
969 /* always caled under BH & queue lock */
970 static void htb_reset(struct Qdisc *sch)
971 {
972         struct htb_sched *q = qdisc_priv(sch);
973         struct htb_class *cl;
974         unsigned int i;
975 
976         for (i = 0; i < q->clhash.hashsize; i++) {
977                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
978                         if (cl->level)
979                                 memset(&cl->un.inner, 0, sizeof(cl->un.inner));
980                         else {
981                                 if (cl->un.leaf.q)
982                                         qdisc_reset(cl->un.leaf.q);
983                                 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
984                         }
985                         cl->prio_activity = 0;
986                         cl->cmode = HTB_CAN_SEND;
987 
988                 }
989         }
990         qdisc_watchdog_cancel(&q->watchdog);
991         __skb_queue_purge(&q->direct_queue);
992         sch->q.qlen = 0;
993         memset(q->hlevel, 0, sizeof(q->hlevel));
994         memset(q->row_mask, 0, sizeof(q->row_mask));
995         for (i = 0; i < TC_HTB_NUMPRIO; i++)
996                 INIT_LIST_HEAD(q->drops + i);
997 }
998 
999 static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = {
1000         [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) },
1001         [TCA_HTB_INIT]  = { .len = sizeof(struct tc_htb_glob) },
1002         [TCA_HTB_CTAB]  = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1003         [TCA_HTB_RTAB]  = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1004         [TCA_HTB_DIRECT_QLEN] = { .type = NLA_U32 },
1005         [TCA_HTB_RATE64] = { .type = NLA_U64 },
1006         [TCA_HTB_CEIL64] = { .type = NLA_U64 },
1007 };
1008 
1009 static void htb_work_func(struct work_struct *work)
1010 {
1011         struct htb_sched *q = container_of(work, struct htb_sched, work);
1012         struct Qdisc *sch = q->watchdog.qdisc;
1013 
1014         __netif_schedule(qdisc_root(sch));
1015 }
1016 
1017 static int htb_init(struct Qdisc *sch, struct nlattr *opt)
1018 {
1019         struct htb_sched *q = qdisc_priv(sch);
1020         struct nlattr *tb[TCA_HTB_MAX + 1];
1021         struct tc_htb_glob *gopt;
1022         int err;
1023         int i;
1024 
1025         if (!opt)
1026                 return -EINVAL;
1027 
1028         err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy);
1029         if (err < 0)
1030                 return err;
1031 
1032         if (!tb[TCA_HTB_INIT])
1033                 return -EINVAL;
1034 
1035         gopt = nla_data(tb[TCA_HTB_INIT]);
1036         if (gopt->version != HTB_VER >> 16)
1037                 return -EINVAL;
1038 
1039         err = qdisc_class_hash_init(&q->clhash);
1040         if (err < 0)
1041                 return err;
1042         for (i = 0; i < TC_HTB_NUMPRIO; i++)
1043                 INIT_LIST_HEAD(q->drops + i);
1044 
1045         qdisc_watchdog_init(&q->watchdog, sch);
1046         INIT_WORK(&q->work, htb_work_func);
1047         __skb_queue_head_init(&q->direct_queue);
1048 
1049         if (tb[TCA_HTB_DIRECT_QLEN])
1050                 q->direct_qlen = nla_get_u32(tb[TCA_HTB_DIRECT_QLEN]);
1051         else
1052                 q->direct_qlen = qdisc_dev(sch)->tx_queue_len;
1053 
1054         if ((q->rate2quantum = gopt->rate2quantum) < 1)
1055                 q->rate2quantum = 1;
1056         q->defcls = gopt->defcls;
1057 
1058         return 0;
1059 }
1060 
1061 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1062 {
1063         struct htb_sched *q = qdisc_priv(sch);
1064         struct nlattr *nest;
1065         struct tc_htb_glob gopt;
1066 
1067         /* Its safe to not acquire qdisc lock. As we hold RTNL,
1068          * no change can happen on the qdisc parameters.
1069          */
1070 
1071         gopt.direct_pkts = q->direct_pkts;
1072         gopt.version = HTB_VER;
1073         gopt.rate2quantum = q->rate2quantum;
1074         gopt.defcls = q->defcls;
1075         gopt.debug = 0;
1076 
1077         nest = nla_nest_start(skb, TCA_OPTIONS);
1078         if (nest == NULL)
1079                 goto nla_put_failure;
1080         if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt) ||
1081             nla_put_u32(skb, TCA_HTB_DIRECT_QLEN, q->direct_qlen))
1082                 goto nla_put_failure;
1083 
1084         return nla_nest_end(skb, nest);
1085 
1086 nla_put_failure:
1087         nla_nest_cancel(skb, nest);
1088         return -1;
1089 }
1090 
1091 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1092                           struct sk_buff *skb, struct tcmsg *tcm)
1093 {
1094         struct htb_class *cl = (struct htb_class *)arg;
1095         struct nlattr *nest;
1096         struct tc_htb_opt opt;
1097 
1098         /* Its safe to not acquire qdisc lock. As we hold RTNL,
1099          * no change can happen on the class parameters.
1100          */
1101         tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT;
1102         tcm->tcm_handle = cl->common.classid;
1103         if (!cl->level && cl->un.leaf.q)
1104                 tcm->tcm_info = cl->un.leaf.q->handle;
1105 
1106         nest = nla_nest_start(skb, TCA_OPTIONS);
1107         if (nest == NULL)
1108                 goto nla_put_failure;
1109 
1110         memset(&opt, 0, sizeof(opt));
1111 
1112         psched_ratecfg_getrate(&opt.rate, &cl->rate);
1113         opt.buffer = PSCHED_NS2TICKS(cl->buffer);
1114         psched_ratecfg_getrate(&opt.ceil, &cl->ceil);
1115         opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer);
1116         opt.quantum = cl->quantum;
1117         opt.prio = cl->prio;
1118         opt.level = cl->level;
1119         if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt))
1120                 goto nla_put_failure;
1121         if ((cl->rate.rate_bytes_ps >= (1ULL << 32)) &&
1122             nla_put_u64(skb, TCA_HTB_RATE64, cl->rate.rate_bytes_ps))
1123                 goto nla_put_failure;
1124         if ((cl->ceil.rate_bytes_ps >= (1ULL << 32)) &&
1125             nla_put_u64(skb, TCA_HTB_CEIL64, cl->ceil.rate_bytes_ps))
1126                 goto nla_put_failure;
1127 
1128         return nla_nest_end(skb, nest);
1129 
1130 nla_put_failure:
1131         nla_nest_cancel(skb, nest);
1132         return -1;
1133 }
1134 
1135 static int
1136 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d)
1137 {
1138         struct htb_class *cl = (struct htb_class *)arg;
1139         __u32 qlen = 0;
1140 
1141         if (!cl->level && cl->un.leaf.q)
1142                 qlen = cl->un.leaf.q->q.qlen;
1143         cl->xstats.tokens = PSCHED_NS2TICKS(cl->tokens);
1144         cl->xstats.ctokens = PSCHED_NS2TICKS(cl->ctokens);
1145 
1146         if (gnet_stats_copy_basic(d, NULL, &cl->bstats) < 0 ||
1147             gnet_stats_copy_rate_est(d, NULL, &cl->rate_est) < 0 ||
1148             gnet_stats_copy_queue(d, NULL, &cl->qstats, qlen) < 0)
1149                 return -1;
1150 
1151         return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1152 }
1153 
1154 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1155                      struct Qdisc **old)
1156 {
1157         struct htb_class *cl = (struct htb_class *)arg;
1158 
1159         if (cl->level)
1160                 return -EINVAL;
1161         if (new == NULL &&
1162             (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
1163                                      cl->common.classid)) == NULL)
1164                 return -ENOBUFS;
1165 
1166         sch_tree_lock(sch);
1167         *old = cl->un.leaf.q;
1168         cl->un.leaf.q = new;
1169         if (*old != NULL) {
1170                 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1171                 qdisc_reset(*old);
1172         }
1173         sch_tree_unlock(sch);
1174         return 0;
1175 }
1176 
1177 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg)
1178 {
1179         struct htb_class *cl = (struct htb_class *)arg;
1180         return !cl->level ? cl->un.leaf.q : NULL;
1181 }
1182 
1183 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg)
1184 {
1185         struct htb_class *cl = (struct htb_class *)arg;
1186 
1187         if (cl->un.leaf.q->q.qlen == 0)
1188                 htb_deactivate(qdisc_priv(sch), cl);
1189 }
1190 
1191 static unsigned long htb_get(struct Qdisc *sch, u32 classid)
1192 {
1193         struct htb_class *cl = htb_find(classid, sch);
1194         if (cl)
1195                 cl->refcnt++;
1196         return (unsigned long)cl;
1197 }
1198 
1199 static inline int htb_parent_last_child(struct htb_class *cl)
1200 {
1201         if (!cl->parent)
1202                 /* the root class */
1203                 return 0;
1204         if (cl->parent->children > 1)
1205                 /* not the last child */
1206                 return 0;
1207         return 1;
1208 }
1209 
1210 static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl,
1211                                struct Qdisc *new_q)
1212 {
1213         struct htb_class *parent = cl->parent;
1214 
1215         WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity);
1216 
1217         if (parent->cmode != HTB_CAN_SEND)
1218                 htb_safe_rb_erase(&parent->pq_node,
1219                                   &q->hlevel[parent->level].wait_pq);
1220 
1221         parent->level = 0;
1222         memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1223         INIT_LIST_HEAD(&parent->un.leaf.drop_list);
1224         parent->un.leaf.q = new_q ? new_q : &noop_qdisc;
1225         parent->tokens = parent->buffer;
1226         parent->ctokens = parent->cbuffer;
1227         parent->t_c = ktime_get_ns();
1228         parent->cmode = HTB_CAN_SEND;
1229 }
1230 
1231 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
1232 {
1233         if (!cl->level) {
1234                 WARN_ON(!cl->un.leaf.q);
1235                 qdisc_destroy(cl->un.leaf.q);
1236         }
1237         gen_kill_estimator(&cl->bstats, &cl->rate_est);
1238         tcf_destroy_chain(&cl->filter_list);
1239         kfree(cl);
1240 }
1241 
1242 static void htb_destroy(struct Qdisc *sch)
1243 {
1244         struct htb_sched *q = qdisc_priv(sch);
1245         struct hlist_node *next;
1246         struct htb_class *cl;
1247         unsigned int i;
1248 
1249         cancel_work_sync(&q->work);
1250         qdisc_watchdog_cancel(&q->watchdog);
1251         /* This line used to be after htb_destroy_class call below
1252          * and surprisingly it worked in 2.4. But it must precede it
1253          * because filter need its target class alive to be able to call
1254          * unbind_filter on it (without Oops).
1255          */
1256         tcf_destroy_chain(&q->filter_list);
1257 
1258         for (i = 0; i < q->clhash.hashsize; i++) {
1259                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode)
1260                         tcf_destroy_chain(&cl->filter_list);
1261         }
1262         for (i = 0; i < q->clhash.hashsize; i++) {
1263                 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
1264                                           common.hnode)
1265                         htb_destroy_class(sch, cl);
1266         }
1267         qdisc_class_hash_destroy(&q->clhash);
1268         __skb_queue_purge(&q->direct_queue);
1269 }
1270 
1271 static int htb_delete(struct Qdisc *sch, unsigned long arg)
1272 {
1273         struct htb_sched *q = qdisc_priv(sch);
1274         struct htb_class *cl = (struct htb_class *)arg;
1275         unsigned int qlen;
1276         struct Qdisc *new_q = NULL;
1277         int last_child = 0;
1278 
1279         /* TODO: why don't allow to delete subtree ? references ? does
1280          * tc subsys guarantee us that in htb_destroy it holds no class
1281          * refs so that we can remove children safely there ?
1282          */
1283         if (cl->children || cl->filter_cnt)
1284                 return -EBUSY;
1285 
1286         if (!cl->level && htb_parent_last_child(cl)) {
1287                 new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
1288                                           cl->parent->common.classid);
1289                 last_child = 1;
1290         }
1291 
1292         sch_tree_lock(sch);
1293 
1294         if (!cl->level) {
1295                 qlen = cl->un.leaf.q->q.qlen;
1296                 qdisc_reset(cl->un.leaf.q);
1297                 qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen);
1298         }
1299 
1300         /* delete from hash and active; remainder in destroy_class */
1301         qdisc_class_hash_remove(&q->clhash, &cl->common);
1302         if (cl->parent)
1303                 cl->parent->children--;
1304 
1305         if (cl->prio_activity)
1306                 htb_deactivate(q, cl);
1307 
1308         if (cl->cmode != HTB_CAN_SEND)
1309                 htb_safe_rb_erase(&cl->pq_node,
1310                                   &q->hlevel[cl->level].wait_pq);
1311 
1312         if (last_child)
1313                 htb_parent_to_leaf(q, cl, new_q);
1314 
1315         BUG_ON(--cl->refcnt == 0);
1316         /*
1317          * This shouldn't happen: we "hold" one cops->get() when called
1318          * from tc_ctl_tclass; the destroy method is done from cops->put().
1319          */
1320 
1321         sch_tree_unlock(sch);
1322         return 0;
1323 }
1324 
1325 static void htb_put(struct Qdisc *sch, unsigned long arg)
1326 {
1327         struct htb_class *cl = (struct htb_class *)arg;
1328 
1329         if (--cl->refcnt == 0)
1330                 htb_destroy_class(sch, cl);
1331 }
1332 
1333 static int htb_change_class(struct Qdisc *sch, u32 classid,
1334                             u32 parentid, struct nlattr **tca,
1335                             unsigned long *arg)
1336 {
1337         int err = -EINVAL;
1338         struct htb_sched *q = qdisc_priv(sch);
1339         struct htb_class *cl = (struct htb_class *)*arg, *parent;
1340         struct nlattr *opt = tca[TCA_OPTIONS];
1341         struct nlattr *tb[TCA_HTB_MAX + 1];
1342         struct tc_htb_opt *hopt;
1343         u64 rate64, ceil64;
1344 
1345         /* extract all subattrs from opt attr */
1346         if (!opt)
1347                 goto failure;
1348 
1349         err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy);
1350         if (err < 0)
1351                 goto failure;
1352 
1353         err = -EINVAL;
1354         if (tb[TCA_HTB_PARMS] == NULL)
1355                 goto failure;
1356 
1357         parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch);
1358 
1359         hopt = nla_data(tb[TCA_HTB_PARMS]);
1360         if (!hopt->rate.rate || !hopt->ceil.rate)
1361                 goto failure;
1362 
1363         /* Keeping backward compatible with rate_table based iproute2 tc */
1364         if (hopt->rate.linklayer == TC_LINKLAYER_UNAWARE)
1365                 qdisc_put_rtab(qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB]));
1366 
1367         if (hopt->ceil.linklayer == TC_LINKLAYER_UNAWARE)
1368                 qdisc_put_rtab(qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB]));
1369 
1370         if (!cl) {              /* new class */
1371                 struct Qdisc *new_q;
1372                 int prio;
1373                 struct {
1374                         struct nlattr           nla;
1375                         struct gnet_estimator   opt;
1376                 } est = {
1377                         .nla = {
1378                                 .nla_len        = nla_attr_size(sizeof(est.opt)),
1379                                 .nla_type       = TCA_RATE,
1380                         },
1381                         .opt = {
1382                                 /* 4s interval, 16s averaging constant */
1383                                 .interval       = 2,
1384                                 .ewma_log       = 2,
1385                         },
1386                 };
1387 
1388                 /* check for valid classid */
1389                 if (!classid || TC_H_MAJ(classid ^ sch->handle) ||
1390                     htb_find(classid, sch))
1391                         goto failure;
1392 
1393                 /* check maximal depth */
1394                 if (parent && parent->parent && parent->parent->level < 2) {
1395                         pr_err("htb: tree is too deep\n");
1396                         goto failure;
1397                 }
1398                 err = -ENOBUFS;
1399                 cl = kzalloc(sizeof(*cl), GFP_KERNEL);
1400                 if (!cl)
1401                         goto failure;
1402 
1403                 if (htb_rate_est || tca[TCA_RATE]) {
1404                         err = gen_new_estimator(&cl->bstats, NULL,
1405                                                 &cl->rate_est,
1406                                                 qdisc_root_sleeping_lock(sch),
1407                                                 tca[TCA_RATE] ? : &est.nla);
1408                         if (err) {
1409                                 kfree(cl);
1410                                 goto failure;
1411                         }
1412                 }
1413 
1414                 cl->refcnt = 1;
1415                 cl->children = 0;
1416                 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
1417                 RB_CLEAR_NODE(&cl->pq_node);
1418 
1419                 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++)
1420                         RB_CLEAR_NODE(&cl->node[prio]);
1421 
1422                 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1423                  * so that can't be used inside of sch_tree_lock
1424                  * -- thanks to Karlis Peisenieks
1425                  */
1426                 new_q = qdisc_create_dflt(sch->dev_queue,
1427                                           &pfifo_qdisc_ops, classid);
1428                 sch_tree_lock(sch);
1429                 if (parent && !parent->level) {
1430                         unsigned int qlen = parent->un.leaf.q->q.qlen;
1431 
1432                         /* turn parent into inner node */
1433                         qdisc_reset(parent->un.leaf.q);
1434                         qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen);
1435                         qdisc_destroy(parent->un.leaf.q);
1436                         if (parent->prio_activity)
1437                                 htb_deactivate(q, parent);
1438 
1439                         /* remove from evt list because of level change */
1440                         if (parent->cmode != HTB_CAN_SEND) {
1441                                 htb_safe_rb_erase(&parent->pq_node, &q->hlevel[0].wait_pq);
1442                                 parent->cmode = HTB_CAN_SEND;
1443                         }
1444                         parent->level = (parent->parent ? parent->parent->level
1445                                          : TC_HTB_MAXDEPTH) - 1;
1446                         memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1447                 }
1448                 /* leaf (we) needs elementary qdisc */
1449                 cl->un.leaf.q = new_q ? new_q : &noop_qdisc;
1450 
1451                 cl->common.classid = classid;
1452                 cl->parent = parent;
1453 
1454                 /* set class to be in HTB_CAN_SEND state */
1455                 cl->tokens = PSCHED_TICKS2NS(hopt->buffer);
1456                 cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer);
1457                 cl->mbuffer = 60ULL * NSEC_PER_SEC;     /* 1min */
1458                 cl->t_c = ktime_get_ns();
1459                 cl->cmode = HTB_CAN_SEND;
1460 
1461                 /* attach to the hash list and parent's family */
1462                 qdisc_class_hash_insert(&q->clhash, &cl->common);
1463                 if (parent)
1464                         parent->children++;
1465         } else {
1466                 if (tca[TCA_RATE]) {
1467                         spinlock_t *lock = qdisc_root_sleeping_lock(sch);
1468 
1469                         err = gen_replace_estimator(&cl->bstats, NULL,
1470                                                     &cl->rate_est,
1471                                                     lock,
1472                                                     tca[TCA_RATE]);
1473                         if (err)
1474                                 return err;
1475                 }
1476                 sch_tree_lock(sch);
1477         }
1478 
1479         rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0;
1480 
1481         ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0;
1482 
1483         psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64);
1484         psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64);
1485 
1486         /* it used to be a nasty bug here, we have to check that node
1487          * is really leaf before changing cl->un.leaf !
1488          */
1489         if (!cl->level) {
1490                 u64 quantum = cl->rate.rate_bytes_ps;
1491 
1492                 do_div(quantum, q->rate2quantum);
1493                 cl->quantum = min_t(u64, quantum, INT_MAX);
1494 
1495                 if (!hopt->quantum && cl->quantum < 1000) {
1496                         pr_warn("HTB: quantum of class %X is small. Consider r2q change.\n",
1497                                 cl->common.classid);
1498                         cl->quantum = 1000;
1499                 }
1500                 if (!hopt->quantum && cl->quantum > 200000) {
1501                         pr_warn("HTB: quantum of class %X is big. Consider r2q change.\n",
1502                                 cl->common.classid);
1503                         cl->quantum = 200000;
1504                 }
1505                 if (hopt->quantum)
1506                         cl->quantum = hopt->quantum;
1507                 if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO)
1508                         cl->prio = TC_HTB_NUMPRIO - 1;
1509         }
1510 
1511         cl->buffer = PSCHED_TICKS2NS(hopt->buffer);
1512         cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer);
1513 
1514         sch_tree_unlock(sch);
1515 
1516         qdisc_class_hash_grow(sch, &q->clhash);
1517 
1518         *arg = (unsigned long)cl;
1519         return 0;
1520 
1521 failure:
1522         return err;
1523 }
1524 
1525 static struct tcf_proto __rcu **htb_find_tcf(struct Qdisc *sch,
1526                                              unsigned long arg)
1527 {
1528         struct htb_sched *q = qdisc_priv(sch);
1529         struct htb_class *cl = (struct htb_class *)arg;
1530         struct tcf_proto __rcu **fl = cl ? &cl->filter_list : &q->filter_list;
1531 
1532         return fl;
1533 }
1534 
1535 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
1536                                      u32 classid)
1537 {
1538         struct htb_class *cl = htb_find(classid, sch);
1539 
1540         /*if (cl && !cl->level) return 0;
1541          * The line above used to be there to prevent attaching filters to
1542          * leaves. But at least tc_index filter uses this just to get class
1543          * for other reasons so that we have to allow for it.
1544          * ----
1545          * 19.6.2002 As Werner explained it is ok - bind filter is just
1546          * another way to "lock" the class - unlike "get" this lock can
1547          * be broken by class during destroy IIUC.
1548          */
1549         if (cl)
1550                 cl->filter_cnt++;
1551         return (unsigned long)cl;
1552 }
1553 
1554 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
1555 {
1556         struct htb_class *cl = (struct htb_class *)arg;
1557 
1558         if (cl)
1559                 cl->filter_cnt--;
1560 }
1561 
1562 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1563 {
1564         struct htb_sched *q = qdisc_priv(sch);
1565         struct htb_class *cl;
1566         unsigned int i;
1567 
1568         if (arg->stop)
1569                 return;
1570 
1571         for (i = 0; i < q->clhash.hashsize; i++) {
1572                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
1573                         if (arg->count < arg->skip) {
1574                                 arg->count++;
1575                                 continue;
1576                         }
1577                         if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1578                                 arg->stop = 1;
1579                                 return;
1580                         }
1581                         arg->count++;
1582                 }
1583         }
1584 }
1585 
1586 static const struct Qdisc_class_ops htb_class_ops = {
1587         .graft          =       htb_graft,
1588         .leaf           =       htb_leaf,
1589         .qlen_notify    =       htb_qlen_notify,
1590         .get            =       htb_get,
1591         .put            =       htb_put,
1592         .change         =       htb_change_class,
1593         .delete         =       htb_delete,
1594         .walk           =       htb_walk,
1595         .tcf_chain      =       htb_find_tcf,
1596         .bind_tcf       =       htb_bind_filter,
1597         .unbind_tcf     =       htb_unbind_filter,
1598         .dump           =       htb_dump_class,
1599         .dump_stats     =       htb_dump_class_stats,
1600 };
1601 
1602 static struct Qdisc_ops htb_qdisc_ops __read_mostly = {
1603         .cl_ops         =       &htb_class_ops,
1604         .id             =       "htb",
1605         .priv_size      =       sizeof(struct htb_sched),
1606         .enqueue        =       htb_enqueue,
1607         .dequeue        =       htb_dequeue,
1608         .peek           =       qdisc_peek_dequeued,
1609         .drop           =       htb_drop,
1610         .init           =       htb_init,
1611         .reset          =       htb_reset,
1612         .destroy        =       htb_destroy,
1613         .dump           =       htb_dump,
1614         .owner          =       THIS_MODULE,
1615 };
1616 
1617 static int __init htb_module_init(void)
1618 {
1619         return register_qdisc(&htb_qdisc_ops);
1620 }
1621 static void __exit htb_module_exit(void)
1622 {
1623         unregister_qdisc(&htb_qdisc_ops);
1624 }
1625 
1626 module_init(htb_module_init)
1627 module_exit(htb_module_exit)
1628 MODULE_LICENSE("GPL");
1629 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us