2 * net/sched/sch_htb.c Hierarchical token bucket, feed tree version
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.
11 * Credits (in time order) for older HTB versions:
13 * HTB support at LARTC mailing list
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
21 * created test case so that I was able to fix nasty bug
23 * spotted bug in dequeue code and helped with fix
25 * fixed requeue routine
26 * and many others. thanks.
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/pkt_sched.h>
45 ========================================================================
46 HTB is like TBF with multiple classes. It is also similar to CBQ because
47 it allows to assign priority to each class in hierarchy.
48 In fact it is another implementation of Floyd's formal sharing.
51 Each class is assigned level. Leaf has ALWAYS level 0 and root
52 classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level
53 one less than their parent.
56 static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */
57 #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */
59 #if HTB_VER >> 16 != TC_HTB_PROTOVER
60 #error "Mismatched sch_htb.c and pkt_sch.h"
63 /* Module parameter and sysfs export */
64 module_param (htb_hysteresis, int, 0640);
65 MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate");
67 /* used internaly to keep status of single class */
69 HTB_CANT_SEND, /* class can't send and can't borrow */
70 HTB_MAY_BORROW, /* class can't send but may borrow */
71 HTB_CAN_SEND /* class can send */
74 /* interior & leaf nodes; props specific to leaves are marked L: */
76 struct Qdisc_class_common common;
77 /* general class parameters */
78 struct gnet_stats_basic_packed bstats;
79 struct gnet_stats_queue qstats;
80 struct gnet_stats_rate_est rate_est;
81 struct tc_htb_xstats xstats; /* our special stats */
82 int refcnt; /* usage count of this class */
85 int level; /* our level (see above) */
86 unsigned int children;
87 struct htb_class *parent; /* parent class */
89 int prio; /* these two are used only by leaves... */
90 int quantum; /* but stored for parent-to-leaf return */
93 struct htb_class_leaf {
95 int deficit[TC_HTB_MAXDEPTH];
96 struct list_head drop_list;
98 struct htb_class_inner {
99 struct rb_root feed[TC_HTB_NUMPRIO]; /* feed trees */
100 struct rb_node *ptr[TC_HTB_NUMPRIO]; /* current class ptr */
101 /* When class changes from state 1->2 and disconnects from
102 * parent's feed then we lost ptr value and start from the
103 * first child again. Here we store classid of the
104 * last valid ptr (used when ptr is NULL).
106 u32 last_ptr_id[TC_HTB_NUMPRIO];
109 struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */
110 struct rb_node pq_node; /* node for event queue */
111 psched_time_t pq_key;
113 int prio_activity; /* for which prios are we active */
114 enum htb_cmode cmode; /* current mode of the class */
116 /* class attached filters */
117 struct tcf_proto *filter_list;
120 /* token bucket parameters */
121 struct qdisc_rate_table *rate; /* rate table of the class itself */
122 struct qdisc_rate_table *ceil; /* ceiling rate (limits borrows too) */
123 long buffer, cbuffer; /* token bucket depth/rate */
124 psched_tdiff_t mbuffer; /* max wait time */
125 long tokens, ctokens; /* current number of tokens */
126 psched_time_t t_c; /* checkpoint time */
130 struct Qdisc_class_hash clhash;
131 struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */
133 /* self list - roots of self generating tree */
134 struct rb_root row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
135 int row_mask[TC_HTB_MAXDEPTH];
136 struct rb_node *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
137 u32 last_ptr_id[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
139 /* self wait list - roots of wait PQs per row */
140 struct rb_root wait_pq[TC_HTB_MAXDEPTH];
142 /* time of nearest event per level (row) */
143 psched_time_t near_ev_cache[TC_HTB_MAXDEPTH];
145 int defcls; /* class where unclassified flows go to */
147 /* filters for qdisc itself */
148 struct tcf_proto *filter_list;
150 int rate2quantum; /* quant = rate / rate2quantum */
151 psched_time_t now; /* cached dequeue time */
152 struct qdisc_watchdog watchdog;
154 /* non shaped skbs; let them go directly thru */
155 struct sk_buff_head direct_queue;
156 int direct_qlen; /* max qlen of above */
160 #define HTB_WARN_TOOMANYEVENTS 0x1
161 unsigned int warned; /* only one warning */
162 struct work_struct work;
165 /* find class in global hash table using given handle */
166 static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch)
168 struct htb_sched *q = qdisc_priv(sch);
169 struct Qdisc_class_common *clc;
171 clc = qdisc_class_find(&q->clhash, handle);
174 return container_of(clc, struct htb_class, common);
178 * htb_classify - classify a packet into class
180 * It returns NULL if the packet should be dropped or -1 if the packet
181 * should be passed directly thru. In all other cases leaf class is returned.
182 * We allow direct class selection by classid in priority. The we examine
183 * filters in qdisc and in inner nodes (if higher filter points to the inner
184 * node). If we end up with classid MAJOR:0 we enqueue the skb into special
185 * internal fifo (direct). These packets then go directly thru. If we still
186 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful
187 * then finish and return direct queue.
189 #define HTB_DIRECT ((struct htb_class *)-1L)
191 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch,
194 struct htb_sched *q = qdisc_priv(sch);
195 struct htb_class *cl;
196 struct tcf_result res;
197 struct tcf_proto *tcf;
200 /* allow to select class by setting skb->priority to valid classid;
201 * note that nfmark can be used too by attaching filter fw with no
204 if (skb->priority == sch->handle)
205 return HTB_DIRECT; /* X:0 (direct flow) selected */
206 cl = htb_find(skb->priority, sch);
207 if (cl && cl->level == 0)
210 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
211 tcf = q->filter_list;
212 while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
213 #ifdef CONFIG_NET_CLS_ACT
217 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
222 cl = (void *)res.class;
224 if (res.classid == sch->handle)
225 return HTB_DIRECT; /* X:0 (direct flow) */
226 cl = htb_find(res.classid, sch);
228 break; /* filter selected invalid classid */
231 return cl; /* we hit leaf; return it */
233 /* we have got inner class; apply inner filter chain */
234 tcf = cl->filter_list;
236 /* classification failed; try to use default class */
237 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
238 if (!cl || cl->level)
239 return HTB_DIRECT; /* bad default .. this is safe bet */
244 * htb_add_to_id_tree - adds class to the round robin list
246 * Routine adds class to the list (actually tree) sorted by classid.
247 * Make sure that class is not already on such list for given prio.
249 static void htb_add_to_id_tree(struct rb_root *root,
250 struct htb_class *cl, int prio)
252 struct rb_node **p = &root->rb_node, *parent = NULL;
257 c = rb_entry(parent, struct htb_class, node[prio]);
259 if (cl->common.classid > c->common.classid)
260 p = &parent->rb_right;
262 p = &parent->rb_left;
264 rb_link_node(&cl->node[prio], parent, p);
265 rb_insert_color(&cl->node[prio], root);
269 * htb_add_to_wait_tree - adds class to the event queue with delay
271 * The class is added to priority event queue to indicate that class will
272 * change its mode in cl->pq_key microseconds. Make sure that class is not
273 * already in the queue.
275 static void htb_add_to_wait_tree(struct htb_sched *q,
276 struct htb_class *cl, long delay)
278 struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL;
280 cl->pq_key = q->now + delay;
281 if (cl->pq_key == q->now)
284 /* update the nearest event cache */
285 if (q->near_ev_cache[cl->level] > cl->pq_key)
286 q->near_ev_cache[cl->level] = cl->pq_key;
291 c = rb_entry(parent, struct htb_class, pq_node);
292 if (cl->pq_key >= c->pq_key)
293 p = &parent->rb_right;
295 p = &parent->rb_left;
297 rb_link_node(&cl->pq_node, parent, p);
298 rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]);
302 * htb_next_rb_node - finds next node in binary tree
304 * When we are past last key we return NULL.
305 * Average complexity is 2 steps per call.
307 static inline void htb_next_rb_node(struct rb_node **n)
313 * htb_add_class_to_row - add class to its row
315 * The class is added to row at priorities marked in mask.
316 * It does nothing if mask == 0.
318 static inline void htb_add_class_to_row(struct htb_sched *q,
319 struct htb_class *cl, int mask)
321 q->row_mask[cl->level] |= mask;
323 int prio = ffz(~mask);
324 mask &= ~(1 << prio);
325 htb_add_to_id_tree(q->row[cl->level] + prio, cl, prio);
329 /* If this triggers, it is a bug in this code, but it need not be fatal */
330 static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root)
332 if (RB_EMPTY_NODE(rb)) {
342 * htb_remove_class_from_row - removes class from its row
344 * The class is removed from row at priorities marked in mask.
345 * It does nothing if mask == 0.
347 static inline void htb_remove_class_from_row(struct htb_sched *q,
348 struct htb_class *cl, int mask)
353 int prio = ffz(~mask);
355 mask &= ~(1 << prio);
356 if (q->ptr[cl->level][prio] == cl->node + prio)
357 htb_next_rb_node(q->ptr[cl->level] + prio);
359 htb_safe_rb_erase(cl->node + prio, q->row[cl->level] + prio);
360 if (!q->row[cl->level][prio].rb_node)
363 q->row_mask[cl->level] &= ~m;
367 * htb_activate_prios - creates active classe's feed chain
369 * The class is connected to ancestors and/or appropriate rows
370 * for priorities it is participating on. cl->cmode must be new
371 * (activated) mode. It does nothing if cl->prio_activity == 0.
373 static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl)
375 struct htb_class *p = cl->parent;
376 long m, mask = cl->prio_activity;
378 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
384 if (p->un.inner.feed[prio].rb_node)
385 /* parent already has its feed in use so that
386 * reset bit in mask as parent is already ok
388 mask &= ~(1 << prio);
390 htb_add_to_id_tree(p->un.inner.feed + prio, cl, prio);
392 p->prio_activity |= mask;
397 if (cl->cmode == HTB_CAN_SEND && mask)
398 htb_add_class_to_row(q, cl, mask);
402 * htb_deactivate_prios - remove class from feed chain
404 * cl->cmode must represent old mode (before deactivation). It does
405 * nothing if cl->prio_activity == 0. Class is removed from all feed
408 static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl)
410 struct htb_class *p = cl->parent;
411 long m, mask = cl->prio_activity;
413 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
420 if (p->un.inner.ptr[prio] == cl->node + prio) {
421 /* we are removing child which is pointed to from
422 * parent feed - forget the pointer but remember
425 p->un.inner.last_ptr_id[prio] = cl->common.classid;
426 p->un.inner.ptr[prio] = NULL;
429 htb_safe_rb_erase(cl->node + prio, p->un.inner.feed + prio);
431 if (!p->un.inner.feed[prio].rb_node)
435 p->prio_activity &= ~mask;
440 if (cl->cmode == HTB_CAN_SEND && mask)
441 htb_remove_class_from_row(q, cl, mask);
444 static inline long htb_lowater(const struct htb_class *cl)
447 return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0;
451 static inline long htb_hiwater(const struct htb_class *cl)
454 return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0;
461 * htb_class_mode - computes and returns current class mode
463 * It computes cl's mode at time cl->t_c+diff and returns it. If mode
464 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
465 * from now to time when cl will change its state.
466 * Also it is worth to note that class mode doesn't change simply
467 * at cl->{c,}tokens == 0 but there can rather be hysteresis of
468 * 0 .. -cl->{c,}buffer range. It is meant to limit number of
469 * mode transitions per time unit. The speed gain is about 1/6.
471 static inline enum htb_cmode
472 htb_class_mode(struct htb_class *cl, long *diff)
476 if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) {
478 return HTB_CANT_SEND;
481 if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl))
485 return HTB_MAY_BORROW;
489 * htb_change_class_mode - changes classe's mode
491 * This should be the only way how to change classe's mode under normal
492 * cirsumstances. Routine will update feed lists linkage, change mode
493 * and add class to the wait event queue if appropriate. New mode should
494 * be different from old one and cl->pq_key has to be valid if changing
495 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
498 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff)
500 enum htb_cmode new_mode = htb_class_mode(cl, diff);
502 if (new_mode == cl->cmode)
505 if (cl->prio_activity) { /* not necessary: speed optimization */
506 if (cl->cmode != HTB_CANT_SEND)
507 htb_deactivate_prios(q, cl);
508 cl->cmode = new_mode;
509 if (new_mode != HTB_CANT_SEND)
510 htb_activate_prios(q, cl);
512 cl->cmode = new_mode;
516 * htb_activate - inserts leaf cl into appropriate active feeds
518 * Routine learns (new) priority of leaf and activates feed chain
519 * for the prio. It can be called on already active leaf safely.
520 * It also adds leaf into droplist.
522 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl)
524 WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen);
526 if (!cl->prio_activity) {
527 cl->prio_activity = 1 << cl->prio;
528 htb_activate_prios(q, cl);
529 list_add_tail(&cl->un.leaf.drop_list,
530 q->drops + cl->prio);
535 * htb_deactivate - remove leaf cl from active feeds
537 * Make sure that leaf is active. In the other words it can't be called
538 * with non-active leaf. It also removes class from the drop list.
540 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl)
542 WARN_ON(!cl->prio_activity);
544 htb_deactivate_prios(q, cl);
545 cl->prio_activity = 0;
546 list_del_init(&cl->un.leaf.drop_list);
549 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
551 int uninitialized_var(ret);
552 struct htb_sched *q = qdisc_priv(sch);
553 struct htb_class *cl = htb_classify(skb, sch, &ret);
555 if (cl == HTB_DIRECT) {
556 /* enqueue to helper queue */
557 if (q->direct_queue.qlen < q->direct_qlen) {
558 __skb_queue_tail(&q->direct_queue, skb);
561 return qdisc_drop(skb, sch);
563 #ifdef CONFIG_NET_CLS_ACT
565 if (ret & __NET_XMIT_BYPASS)
570 } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q)) != NET_XMIT_SUCCESS) {
571 if (net_xmit_drop_count(ret)) {
581 return NET_XMIT_SUCCESS;
584 static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, long diff)
586 long toks = diff + cl->tokens;
588 if (toks > cl->buffer)
590 toks -= (long) qdisc_l2t(cl->rate, bytes);
591 if (toks <= -cl->mbuffer)
592 toks = 1 - cl->mbuffer;
597 static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, long diff)
599 long toks = diff + cl->ctokens;
601 if (toks > cl->cbuffer)
603 toks -= (long) qdisc_l2t(cl->ceil, bytes);
604 if (toks <= -cl->mbuffer)
605 toks = 1 - cl->mbuffer;
611 * htb_charge_class - charges amount "bytes" to leaf and ancestors
613 * Routine assumes that packet "bytes" long was dequeued from leaf cl
614 * borrowing from "level". It accounts bytes to ceil leaky bucket for
615 * leaf and all ancestors and to rate bucket for ancestors at levels
616 * "level" and higher. It also handles possible change of mode resulting
617 * from the update. Note that mode can also increase here (MAY_BORROW to
618 * CAN_SEND) because we can use more precise clock that event queue here.
619 * In such case we remove class from event queue first.
621 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl,
622 int level, struct sk_buff *skb)
624 int bytes = qdisc_pkt_len(skb);
625 enum htb_cmode old_mode;
629 diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
630 if (cl->level >= level) {
631 if (cl->level == level)
633 htb_accnt_tokens(cl, bytes, diff);
635 cl->xstats.borrows++;
636 cl->tokens += diff; /* we moved t_c; update tokens */
638 htb_accnt_ctokens(cl, bytes, diff);
641 old_mode = cl->cmode;
643 htb_change_class_mode(q, cl, &diff);
644 if (old_mode != cl->cmode) {
645 if (old_mode != HTB_CAN_SEND)
646 htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level);
647 if (cl->cmode != HTB_CAN_SEND)
648 htb_add_to_wait_tree(q, cl, diff);
651 /* update basic stats except for leaves which are already updated */
653 bstats_update(&cl->bstats, skb);
660 * htb_do_events - make mode changes to classes at the level
662 * Scans event queue for pending events and applies them. Returns time of
663 * next pending event (0 for no event in pq, q->now for too many events).
664 * Note: Applied are events whose have cl->pq_key <= q->now.
666 static psched_time_t htb_do_events(struct htb_sched *q, int level,
669 /* don't run for longer than 2 jiffies; 2 is used instead of
670 * 1 to simplify things when jiffy is going to be incremented
673 unsigned long stop_at = start + 2;
674 while (time_before(jiffies, stop_at)) {
675 struct htb_class *cl;
677 struct rb_node *p = rb_first(&q->wait_pq[level]);
682 cl = rb_entry(p, struct htb_class, pq_node);
683 if (cl->pq_key > q->now)
686 htb_safe_rb_erase(p, q->wait_pq + level);
687 diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
688 htb_change_class_mode(q, cl, &diff);
689 if (cl->cmode != HTB_CAN_SEND)
690 htb_add_to_wait_tree(q, cl, diff);
693 /* too much load - let's continue after a break for scheduling */
694 if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) {
695 pr_warning("htb: too many events!\n");
696 q->warned |= HTB_WARN_TOOMANYEVENTS;
702 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
703 * is no such one exists.
705 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n,
708 struct rb_node *r = NULL;
710 struct htb_class *cl =
711 rb_entry(n, struct htb_class, node[prio]);
713 if (id > cl->common.classid) {
715 } else if (id < cl->common.classid) {
726 * htb_lookup_leaf - returns next leaf class in DRR order
728 * Find leaf where current feed pointers points to.
730 static struct htb_class *htb_lookup_leaf(struct rb_root *tree, int prio,
731 struct rb_node **pptr, u32 * pid)
735 struct rb_node *root;
736 struct rb_node **pptr;
738 } stk[TC_HTB_MAXDEPTH], *sp = stk;
740 BUG_ON(!tree->rb_node);
741 sp->root = tree->rb_node;
745 for (i = 0; i < 65535; i++) {
746 if (!*sp->pptr && *sp->pid) {
747 /* ptr was invalidated but id is valid - try to recover
748 * the original or next ptr
751 htb_id_find_next_upper(prio, sp->root, *sp->pid);
753 *sp->pid = 0; /* ptr is valid now so that remove this hint as it
754 * can become out of date quickly
756 if (!*sp->pptr) { /* we are at right end; rewind & go up */
757 *sp->pptr = sp->root;
758 while ((*sp->pptr)->rb_left)
759 *sp->pptr = (*sp->pptr)->rb_left;
766 htb_next_rb_node(sp->pptr);
769 struct htb_class *cl;
770 cl = rb_entry(*sp->pptr, struct htb_class, node[prio]);
773 (++sp)->root = cl->un.inner.feed[prio].rb_node;
774 sp->pptr = cl->un.inner.ptr + prio;
775 sp->pid = cl->un.inner.last_ptr_id + prio;
782 /* dequeues packet at given priority and level; call only if
783 * you are sure that there is active class at prio/level
785 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, int prio,
788 struct sk_buff *skb = NULL;
789 struct htb_class *cl, *start;
790 /* look initial class up in the row */
791 start = cl = htb_lookup_leaf(q->row[level] + prio, prio,
792 q->ptr[level] + prio,
793 q->last_ptr_id[level] + prio);
800 /* class can be empty - it is unlikely but can be true if leaf
801 * qdisc drops packets in enqueue routine or if someone used
802 * graft operation on the leaf since last dequeue;
803 * simply deactivate and skip such class
805 if (unlikely(cl->un.leaf.q->q.qlen == 0)) {
806 struct htb_class *next;
807 htb_deactivate(q, cl);
809 /* row/level might become empty */
810 if ((q->row_mask[level] & (1 << prio)) == 0)
813 next = htb_lookup_leaf(q->row[level] + prio,
814 prio, q->ptr[level] + prio,
815 q->last_ptr_id[level] + prio);
817 if (cl == start) /* fix start if we just deleted it */
823 skb = cl->un.leaf.q->dequeue(cl->un.leaf.q);
824 if (likely(skb != NULL))
827 qdisc_warn_nonwc("htb", cl->un.leaf.q);
828 htb_next_rb_node((level ? cl->parent->un.inner.ptr : q->
830 cl = htb_lookup_leaf(q->row[level] + prio, prio,
831 q->ptr[level] + prio,
832 q->last_ptr_id[level] + prio);
834 } while (cl != start);
836 if (likely(skb != NULL)) {
837 bstats_update(&cl->bstats, skb);
838 cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb);
839 if (cl->un.leaf.deficit[level] < 0) {
840 cl->un.leaf.deficit[level] += cl->quantum;
841 htb_next_rb_node((level ? cl->parent->un.inner.ptr : q->
844 /* this used to be after charge_class but this constelation
845 * gives us slightly better performance
847 if (!cl->un.leaf.q->q.qlen)
848 htb_deactivate(q, cl);
849 htb_charge_class(q, cl, level, skb);
854 static struct sk_buff *htb_dequeue(struct Qdisc *sch)
857 struct htb_sched *q = qdisc_priv(sch);
859 psched_time_t next_event;
860 unsigned long start_at;
862 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
863 skb = __skb_dequeue(&q->direct_queue);
866 qdisc_bstats_update(sch, skb);
867 qdisc_unthrottled(sch);
874 q->now = psched_get_time();
877 next_event = q->now + 5 * PSCHED_TICKS_PER_SEC;
879 for (level = 0; level < TC_HTB_MAXDEPTH; level++) {
880 /* common case optimization - skip event handler quickly */
884 if (q->now >= q->near_ev_cache[level]) {
885 event = htb_do_events(q, level, start_at);
887 event = q->now + PSCHED_TICKS_PER_SEC;
888 q->near_ev_cache[level] = event;
890 event = q->near_ev_cache[level];
892 if (next_event > event)
895 m = ~q->row_mask[level];
896 while (m != (int)(-1)) {
900 skb = htb_dequeue_tree(q, prio, level);
901 if (likely(skb != NULL))
905 sch->qstats.overlimits++;
906 if (likely(next_event > q->now))
907 qdisc_watchdog_schedule(&q->watchdog, next_event);
909 schedule_work(&q->work);
914 /* try to drop from each class (by prio) until one succeed */
915 static unsigned int htb_drop(struct Qdisc *sch)
917 struct htb_sched *q = qdisc_priv(sch);
920 for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) {
922 list_for_each(p, q->drops + prio) {
923 struct htb_class *cl = list_entry(p, struct htb_class,
926 if (cl->un.leaf.q->ops->drop &&
927 (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) {
929 if (!cl->un.leaf.q->q.qlen)
930 htb_deactivate(q, cl);
938 /* reset all classes */
939 /* always caled under BH & queue lock */
940 static void htb_reset(struct Qdisc *sch)
942 struct htb_sched *q = qdisc_priv(sch);
943 struct htb_class *cl;
944 struct hlist_node *n;
947 for (i = 0; i < q->clhash.hashsize; i++) {
948 hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) {
950 memset(&cl->un.inner, 0, sizeof(cl->un.inner));
953 qdisc_reset(cl->un.leaf.q);
954 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
956 cl->prio_activity = 0;
957 cl->cmode = HTB_CAN_SEND;
961 qdisc_watchdog_cancel(&q->watchdog);
962 __skb_queue_purge(&q->direct_queue);
964 memset(q->row, 0, sizeof(q->row));
965 memset(q->row_mask, 0, sizeof(q->row_mask));
966 memset(q->wait_pq, 0, sizeof(q->wait_pq));
967 memset(q->ptr, 0, sizeof(q->ptr));
968 for (i = 0; i < TC_HTB_NUMPRIO; i++)
969 INIT_LIST_HEAD(q->drops + i);
972 static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = {
973 [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) },
974 [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) },
975 [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
976 [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
979 static void htb_work_func(struct work_struct *work)
981 struct htb_sched *q = container_of(work, struct htb_sched, work);
982 struct Qdisc *sch = q->watchdog.qdisc;
984 __netif_schedule(qdisc_root(sch));
987 static int htb_init(struct Qdisc *sch, struct nlattr *opt)
989 struct htb_sched *q = qdisc_priv(sch);
990 struct nlattr *tb[TCA_HTB_INIT + 1];
991 struct tc_htb_glob *gopt;
998 err = nla_parse_nested(tb, TCA_HTB_INIT, opt, htb_policy);
1002 if (tb[TCA_HTB_INIT] == NULL) {
1003 pr_err("HTB: hey probably you have bad tc tool ?\n");
1006 gopt = nla_data(tb[TCA_HTB_INIT]);
1007 if (gopt->version != HTB_VER >> 16) {
1008 pr_err("HTB: need tc/htb version %d (minor is %d), you have %d\n",
1009 HTB_VER >> 16, HTB_VER & 0xffff, gopt->version);
1013 err = qdisc_class_hash_init(&q->clhash);
1016 for (i = 0; i < TC_HTB_NUMPRIO; i++)
1017 INIT_LIST_HEAD(q->drops + i);
1019 qdisc_watchdog_init(&q->watchdog, sch);
1020 INIT_WORK(&q->work, htb_work_func);
1021 skb_queue_head_init(&q->direct_queue);
1023 q->direct_qlen = qdisc_dev(sch)->tx_queue_len;
1024 if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */
1027 if ((q->rate2quantum = gopt->rate2quantum) < 1)
1028 q->rate2quantum = 1;
1029 q->defcls = gopt->defcls;
1034 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1036 spinlock_t *root_lock = qdisc_root_sleeping_lock(sch);
1037 struct htb_sched *q = qdisc_priv(sch);
1038 struct nlattr *nest;
1039 struct tc_htb_glob gopt;
1041 spin_lock_bh(root_lock);
1043 gopt.direct_pkts = q->direct_pkts;
1044 gopt.version = HTB_VER;
1045 gopt.rate2quantum = q->rate2quantum;
1046 gopt.defcls = q->defcls;
1049 nest = nla_nest_start(skb, TCA_OPTIONS);
1051 goto nla_put_failure;
1052 if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt))
1053 goto nla_put_failure;
1054 nla_nest_end(skb, nest);
1056 spin_unlock_bh(root_lock);
1060 spin_unlock_bh(root_lock);
1061 nla_nest_cancel(skb, nest);
1065 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1066 struct sk_buff *skb, struct tcmsg *tcm)
1068 struct htb_class *cl = (struct htb_class *)arg;
1069 spinlock_t *root_lock = qdisc_root_sleeping_lock(sch);
1070 struct nlattr *nest;
1071 struct tc_htb_opt opt;
1073 spin_lock_bh(root_lock);
1074 tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT;
1075 tcm->tcm_handle = cl->common.classid;
1076 if (!cl->level && cl->un.leaf.q)
1077 tcm->tcm_info = cl->un.leaf.q->handle;
1079 nest = nla_nest_start(skb, TCA_OPTIONS);
1081 goto nla_put_failure;
1083 memset(&opt, 0, sizeof(opt));
1085 opt.rate = cl->rate->rate;
1086 opt.buffer = cl->buffer;
1087 opt.ceil = cl->ceil->rate;
1088 opt.cbuffer = cl->cbuffer;
1089 opt.quantum = cl->quantum;
1090 opt.prio = cl->prio;
1091 opt.level = cl->level;
1092 if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt))
1093 goto nla_put_failure;
1095 nla_nest_end(skb, nest);
1096 spin_unlock_bh(root_lock);
1100 spin_unlock_bh(root_lock);
1101 nla_nest_cancel(skb, nest);
1106 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d)
1108 struct htb_class *cl = (struct htb_class *)arg;
1110 if (!cl->level && cl->un.leaf.q)
1111 cl->qstats.qlen = cl->un.leaf.q->q.qlen;
1112 cl->xstats.tokens = cl->tokens;
1113 cl->xstats.ctokens = cl->ctokens;
1115 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
1116 gnet_stats_copy_rate_est(d, NULL, &cl->rate_est) < 0 ||
1117 gnet_stats_copy_queue(d, &cl->qstats) < 0)
1120 return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1123 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1126 struct htb_class *cl = (struct htb_class *)arg;
1131 (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
1132 cl->common.classid)) == NULL)
1136 *old = cl->un.leaf.q;
1137 cl->un.leaf.q = new;
1139 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1142 sch_tree_unlock(sch);
1146 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg)
1148 struct htb_class *cl = (struct htb_class *)arg;
1149 return !cl->level ? cl->un.leaf.q : NULL;
1152 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg)
1154 struct htb_class *cl = (struct htb_class *)arg;
1156 if (cl->un.leaf.q->q.qlen == 0)
1157 htb_deactivate(qdisc_priv(sch), cl);
1160 static unsigned long htb_get(struct Qdisc *sch, u32 classid)
1162 struct htb_class *cl = htb_find(classid, sch);
1165 return (unsigned long)cl;
1168 static inline int htb_parent_last_child(struct htb_class *cl)
1171 /* the root class */
1173 if (cl->parent->children > 1)
1174 /* not the last child */
1179 static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl,
1180 struct Qdisc *new_q)
1182 struct htb_class *parent = cl->parent;
1184 WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity);
1186 if (parent->cmode != HTB_CAN_SEND)
1187 htb_safe_rb_erase(&parent->pq_node, q->wait_pq + parent->level);
1190 memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1191 INIT_LIST_HEAD(&parent->un.leaf.drop_list);
1192 parent->un.leaf.q = new_q ? new_q : &noop_qdisc;
1193 parent->tokens = parent->buffer;
1194 parent->ctokens = parent->cbuffer;
1195 parent->t_c = psched_get_time();
1196 parent->cmode = HTB_CAN_SEND;
1199 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
1202 WARN_ON(!cl->un.leaf.q);
1203 qdisc_destroy(cl->un.leaf.q);
1205 gen_kill_estimator(&cl->bstats, &cl->rate_est);
1206 qdisc_put_rtab(cl->rate);
1207 qdisc_put_rtab(cl->ceil);
1209 tcf_destroy_chain(&cl->filter_list);
1213 static void htb_destroy(struct Qdisc *sch)
1215 struct htb_sched *q = qdisc_priv(sch);
1216 struct hlist_node *n, *next;
1217 struct htb_class *cl;
1220 cancel_work_sync(&q->work);
1221 qdisc_watchdog_cancel(&q->watchdog);
1222 /* This line used to be after htb_destroy_class call below
1223 * and surprisingly it worked in 2.4. But it must precede it
1224 * because filter need its target class alive to be able to call
1225 * unbind_filter on it (without Oops).
1227 tcf_destroy_chain(&q->filter_list);
1229 for (i = 0; i < q->clhash.hashsize; i++) {
1230 hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode)
1231 tcf_destroy_chain(&cl->filter_list);
1233 for (i = 0; i < q->clhash.hashsize; i++) {
1234 hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i],
1236 htb_destroy_class(sch, cl);
1238 qdisc_class_hash_destroy(&q->clhash);
1239 __skb_queue_purge(&q->direct_queue);
1242 static int htb_delete(struct Qdisc *sch, unsigned long arg)
1244 struct htb_sched *q = qdisc_priv(sch);
1245 struct htb_class *cl = (struct htb_class *)arg;
1247 struct Qdisc *new_q = NULL;
1250 // TODO: why don't allow to delete subtree ? references ? does
1251 // tc subsys quarantee us that in htb_destroy it holds no class
1252 // refs so that we can remove children safely there ?
1253 if (cl->children || cl->filter_cnt)
1256 if (!cl->level && htb_parent_last_child(cl)) {
1257 new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
1258 cl->parent->common.classid);
1265 qlen = cl->un.leaf.q->q.qlen;
1266 qdisc_reset(cl->un.leaf.q);
1267 qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen);
1270 /* delete from hash and active; remainder in destroy_class */
1271 qdisc_class_hash_remove(&q->clhash, &cl->common);
1273 cl->parent->children--;
1275 if (cl->prio_activity)
1276 htb_deactivate(q, cl);
1278 if (cl->cmode != HTB_CAN_SEND)
1279 htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level);
1282 htb_parent_to_leaf(q, cl, new_q);
1284 BUG_ON(--cl->refcnt == 0);
1286 * This shouldn't happen: we "hold" one cops->get() when called
1287 * from tc_ctl_tclass; the destroy method is done from cops->put().
1290 sch_tree_unlock(sch);
1294 static void htb_put(struct Qdisc *sch, unsigned long arg)
1296 struct htb_class *cl = (struct htb_class *)arg;
1298 if (--cl->refcnt == 0)
1299 htb_destroy_class(sch, cl);
1302 static int htb_change_class(struct Qdisc *sch, u32 classid,
1303 u32 parentid, struct nlattr **tca,
1307 struct htb_sched *q = qdisc_priv(sch);
1308 struct htb_class *cl = (struct htb_class *)*arg, *parent;
1309 struct nlattr *opt = tca[TCA_OPTIONS];
1310 struct qdisc_rate_table *rtab = NULL, *ctab = NULL;
1311 struct nlattr *tb[__TCA_HTB_MAX];
1312 struct tc_htb_opt *hopt;
1314 /* extract all subattrs from opt attr */
1318 err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy);
1323 if (tb[TCA_HTB_PARMS] == NULL)
1326 parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch);
1328 hopt = nla_data(tb[TCA_HTB_PARMS]);
1330 rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB]);
1331 ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB]);
1335 if (!cl) { /* new class */
1336 struct Qdisc *new_q;
1340 struct gnet_estimator opt;
1343 .nla_len = nla_attr_size(sizeof(est.opt)),
1344 .nla_type = TCA_RATE,
1347 /* 4s interval, 16s averaging constant */
1353 /* check for valid classid */
1354 if (!classid || TC_H_MAJ(classid ^ sch->handle) ||
1355 htb_find(classid, sch))
1358 /* check maximal depth */
1359 if (parent && parent->parent && parent->parent->level < 2) {
1360 pr_err("htb: tree is too deep\n");
1364 cl = kzalloc(sizeof(*cl), GFP_KERNEL);
1368 err = gen_new_estimator(&cl->bstats, &cl->rate_est,
1369 qdisc_root_sleeping_lock(sch),
1370 tca[TCA_RATE] ? : &est.nla);
1378 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
1379 RB_CLEAR_NODE(&cl->pq_node);
1381 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++)
1382 RB_CLEAR_NODE(&cl->node[prio]);
1384 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1385 * so that can't be used inside of sch_tree_lock
1386 * -- thanks to Karlis Peisenieks
1388 new_q = qdisc_create_dflt(sch->dev_queue,
1389 &pfifo_qdisc_ops, classid);
1391 if (parent && !parent->level) {
1392 unsigned int qlen = parent->un.leaf.q->q.qlen;
1394 /* turn parent into inner node */
1395 qdisc_reset(parent->un.leaf.q);
1396 qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen);
1397 qdisc_destroy(parent->un.leaf.q);
1398 if (parent->prio_activity)
1399 htb_deactivate(q, parent);
1401 /* remove from evt list because of level change */
1402 if (parent->cmode != HTB_CAN_SEND) {
1403 htb_safe_rb_erase(&parent->pq_node, q->wait_pq);
1404 parent->cmode = HTB_CAN_SEND;
1406 parent->level = (parent->parent ? parent->parent->level
1407 : TC_HTB_MAXDEPTH) - 1;
1408 memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1410 /* leaf (we) needs elementary qdisc */
1411 cl->un.leaf.q = new_q ? new_q : &noop_qdisc;
1413 cl->common.classid = classid;
1414 cl->parent = parent;
1416 /* set class to be in HTB_CAN_SEND state */
1417 cl->tokens = hopt->buffer;
1418 cl->ctokens = hopt->cbuffer;
1419 cl->mbuffer = 60 * PSCHED_TICKS_PER_SEC; /* 1min */
1420 cl->t_c = psched_get_time();
1421 cl->cmode = HTB_CAN_SEND;
1423 /* attach to the hash list and parent's family */
1424 qdisc_class_hash_insert(&q->clhash, &cl->common);
1428 if (tca[TCA_RATE]) {
1429 err = gen_replace_estimator(&cl->bstats, &cl->rate_est,
1430 qdisc_root_sleeping_lock(sch),
1438 /* it used to be a nasty bug here, we have to check that node
1439 * is really leaf before changing cl->un.leaf !
1442 cl->quantum = rtab->rate.rate / q->rate2quantum;
1443 if (!hopt->quantum && cl->quantum < 1000) {
1445 "HTB: quantum of class %X is small. Consider r2q change.\n",
1446 cl->common.classid);
1449 if (!hopt->quantum && cl->quantum > 200000) {
1451 "HTB: quantum of class %X is big. Consider r2q change.\n",
1452 cl->common.classid);
1453 cl->quantum = 200000;
1456 cl->quantum = hopt->quantum;
1457 if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO)
1458 cl->prio = TC_HTB_NUMPRIO - 1;
1461 cl->buffer = hopt->buffer;
1462 cl->cbuffer = hopt->cbuffer;
1464 qdisc_put_rtab(cl->rate);
1467 qdisc_put_rtab(cl->ceil);
1469 sch_tree_unlock(sch);
1471 qdisc_class_hash_grow(sch, &q->clhash);
1473 *arg = (unsigned long)cl;
1478 qdisc_put_rtab(rtab);
1480 qdisc_put_rtab(ctab);
1484 static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg)
1486 struct htb_sched *q = qdisc_priv(sch);
1487 struct htb_class *cl = (struct htb_class *)arg;
1488 struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list;
1493 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
1496 struct htb_class *cl = htb_find(classid, sch);
1498 /*if (cl && !cl->level) return 0;
1499 * The line above used to be there to prevent attaching filters to
1500 * leaves. But at least tc_index filter uses this just to get class
1501 * for other reasons so that we have to allow for it.
1503 * 19.6.2002 As Werner explained it is ok - bind filter is just
1504 * another way to "lock" the class - unlike "get" this lock can
1505 * be broken by class during destroy IIUC.
1509 return (unsigned long)cl;
1512 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
1514 struct htb_class *cl = (struct htb_class *)arg;
1520 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1522 struct htb_sched *q = qdisc_priv(sch);
1523 struct htb_class *cl;
1524 struct hlist_node *n;
1530 for (i = 0; i < q->clhash.hashsize; i++) {
1531 hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) {
1532 if (arg->count < arg->skip) {
1536 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1545 static const struct Qdisc_class_ops htb_class_ops = {
1548 .qlen_notify = htb_qlen_notify,
1551 .change = htb_change_class,
1552 .delete = htb_delete,
1554 .tcf_chain = htb_find_tcf,
1555 .bind_tcf = htb_bind_filter,
1556 .unbind_tcf = htb_unbind_filter,
1557 .dump = htb_dump_class,
1558 .dump_stats = htb_dump_class_stats,
1561 static struct Qdisc_ops htb_qdisc_ops __read_mostly = {
1562 .cl_ops = &htb_class_ops,
1564 .priv_size = sizeof(struct htb_sched),
1565 .enqueue = htb_enqueue,
1566 .dequeue = htb_dequeue,
1567 .peek = qdisc_peek_dequeued,
1571 .destroy = htb_destroy,
1573 .owner = THIS_MODULE,
1576 static int __init htb_module_init(void)
1578 return register_qdisc(&htb_qdisc_ops);
1580 static void __exit htb_module_exit(void)
1582 unregister_qdisc(&htb_qdisc_ops);
1585 module_init(htb_module_init)
1586 module_exit(htb_module_exit)
1587 MODULE_LICENSE("GPL");