Commit | Line | Data |
---|---|---|
87990467 | 1 | /* |
1da177e4 LT |
2 | * net/sched/sch_htb.c Hierarchical token bucket, feed tree version |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Authors: Martin Devera, <devik@cdi.cz> | |
10 | * | |
11 | * Credits (in time order) for older HTB versions: | |
12 | * Stef Coene <stef.coene@docum.org> | |
13 | * HTB support at LARTC mailing list | |
14 | * Ondrej Kraus, <krauso@barr.cz> | |
15 | * found missing INIT_QDISC(htb) | |
16 | * Vladimir Smelhaus, Aamer Akhter, Bert Hubert | |
17 | * helped a lot to locate nasty class stall bug | |
18 | * Andi Kleen, Jamal Hadi, Bert Hubert | |
19 | * code review and helpful comments on shaping | |
20 | * Tomasz Wrona, <tw@eter.tym.pl> | |
21 | * created test case so that I was able to fix nasty bug | |
22 | * Wilfried Weissmann | |
23 | * spotted bug in dequeue code and helped with fix | |
24 | * Jiri Fojtasek | |
25 | * fixed requeue routine | |
26 | * and many others. thanks. | |
27 | * | |
28 | * $Id: sch_htb.c,v 1.25 2003/12/07 11:08:25 devik Exp devik $ | |
29 | */ | |
1da177e4 LT |
30 | #include <linux/module.h> |
31 | #include <asm/uaccess.h> | |
32 | #include <asm/system.h> | |
33 | #include <linux/bitops.h> | |
34 | #include <linux/types.h> | |
35 | #include <linux/kernel.h> | |
36 | #include <linux/sched.h> | |
37 | #include <linux/string.h> | |
38 | #include <linux/mm.h> | |
39 | #include <linux/socket.h> | |
40 | #include <linux/sockios.h> | |
41 | #include <linux/in.h> | |
42 | #include <linux/errno.h> | |
43 | #include <linux/interrupt.h> | |
44 | #include <linux/if_ether.h> | |
45 | #include <linux/inet.h> | |
46 | #include <linux/netdevice.h> | |
47 | #include <linux/etherdevice.h> | |
48 | #include <linux/notifier.h> | |
49 | #include <net/ip.h> | |
50 | #include <net/route.h> | |
51 | #include <linux/skbuff.h> | |
52 | #include <linux/list.h> | |
53 | #include <linux/compiler.h> | |
54 | #include <net/sock.h> | |
55 | #include <net/pkt_sched.h> | |
56 | #include <linux/rbtree.h> | |
57 | ||
58 | /* HTB algorithm. | |
59 | Author: devik@cdi.cz | |
60 | ======================================================================== | |
61 | HTB is like TBF with multiple classes. It is also similar to CBQ because | |
62 | it allows to assign priority to each class in hierarchy. | |
63 | In fact it is another implementation of Floyd's formal sharing. | |
64 | ||
65 | Levels: | |
66 | Each class is assigned level. Leaf has ALWAYS level 0 and root | |
67 | classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level | |
68 | one less than their parent. | |
69 | */ | |
70 | ||
87990467 SH |
71 | #define HTB_HSIZE 16 /* classid hash size */ |
72 | #define HTB_EWMAC 2 /* rate average over HTB_EWMAC*HTB_HSIZE sec */ | |
73 | #define HTB_RATECM 1 /* whether to use rate computer */ | |
74 | #define HTB_HYSTERESIS 1 /* whether to use mode hysteresis for speedup */ | |
75 | #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */ | |
1da177e4 LT |
76 | |
77 | #if HTB_VER >> 16 != TC_HTB_PROTOVER | |
78 | #error "Mismatched sch_htb.c and pkt_sch.h" | |
79 | #endif | |
80 | ||
1da177e4 LT |
81 | /* used internaly to keep status of single class */ |
82 | enum htb_cmode { | |
87990467 SH |
83 | HTB_CANT_SEND, /* class can't send and can't borrow */ |
84 | HTB_MAY_BORROW, /* class can't send but may borrow */ | |
85 | HTB_CAN_SEND /* class can send */ | |
1da177e4 LT |
86 | }; |
87 | ||
88 | /* interior & leaf nodes; props specific to leaves are marked L: */ | |
87990467 SH |
89 | struct htb_class { |
90 | /* general class parameters */ | |
91 | u32 classid; | |
92 | struct gnet_stats_basic bstats; | |
93 | struct gnet_stats_queue qstats; | |
94 | struct gnet_stats_rate_est rate_est; | |
95 | struct tc_htb_xstats xstats; /* our special stats */ | |
96 | int refcnt; /* usage count of this class */ | |
1da177e4 LT |
97 | |
98 | #ifdef HTB_RATECM | |
87990467 SH |
99 | /* rate measurement counters */ |
100 | unsigned long rate_bytes, sum_bytes; | |
101 | unsigned long rate_packets, sum_packets; | |
1da177e4 LT |
102 | #endif |
103 | ||
87990467 SH |
104 | /* topology */ |
105 | int level; /* our level (see above) */ | |
106 | struct htb_class *parent; /* parent class */ | |
0cef296d | 107 | struct hlist_node hlist; /* classid hash list item */ |
87990467 SH |
108 | struct list_head sibling; /* sibling list item */ |
109 | struct list_head children; /* children list */ | |
110 | ||
111 | union { | |
112 | struct htb_class_leaf { | |
113 | struct Qdisc *q; | |
114 | int prio; | |
115 | int aprio; | |
116 | int quantum; | |
117 | int deficit[TC_HTB_MAXDEPTH]; | |
118 | struct list_head drop_list; | |
119 | } leaf; | |
120 | struct htb_class_inner { | |
121 | struct rb_root feed[TC_HTB_NUMPRIO]; /* feed trees */ | |
122 | struct rb_node *ptr[TC_HTB_NUMPRIO]; /* current class ptr */ | |
123 | /* When class changes from state 1->2 and disconnects from | |
124 | parent's feed then we lost ptr value and start from the | |
125 | first child again. Here we store classid of the | |
126 | last valid ptr (used when ptr is NULL). */ | |
127 | u32 last_ptr_id[TC_HTB_NUMPRIO]; | |
128 | } inner; | |
129 | } un; | |
130 | struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ | |
131 | struct rb_node pq_node; /* node for event queue */ | |
132 | unsigned long pq_key; /* the same type as jiffies global */ | |
133 | ||
134 | int prio_activity; /* for which prios are we active */ | |
135 | enum htb_cmode cmode; /* current mode of the class */ | |
136 | ||
137 | /* class attached filters */ | |
138 | struct tcf_proto *filter_list; | |
139 | int filter_cnt; | |
140 | ||
141 | int warned; /* only one warning about non work conserving .. */ | |
142 | ||
143 | /* token bucket parameters */ | |
144 | struct qdisc_rate_table *rate; /* rate table of the class itself */ | |
145 | struct qdisc_rate_table *ceil; /* ceiling rate (limits borrows too) */ | |
146 | long buffer, cbuffer; /* token bucket depth/rate */ | |
147 | psched_tdiff_t mbuffer; /* max wait time */ | |
148 | long tokens, ctokens; /* current number of tokens */ | |
149 | psched_time_t t_c; /* checkpoint time */ | |
1da177e4 LT |
150 | }; |
151 | ||
152 | /* TODO: maybe compute rate when size is too large .. or drop ? */ | |
87990467 SH |
153 | static inline long L2T(struct htb_class *cl, struct qdisc_rate_table *rate, |
154 | int size) | |
155 | { | |
156 | int slot = size >> rate->rate.cell_log; | |
157 | if (slot > 255) { | |
158 | cl->xstats.giants++; | |
159 | slot = 255; | |
160 | } | |
161 | return rate->data[slot]; | |
1da177e4 LT |
162 | } |
163 | ||
87990467 SH |
164 | struct htb_sched { |
165 | struct list_head root; /* root classes list */ | |
0cef296d SH |
166 | struct hlist_head hash[HTB_HSIZE]; /* hashed by classid */ |
167 | struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */ | |
87990467 SH |
168 | |
169 | /* self list - roots of self generating tree */ | |
170 | struct rb_root row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
171 | int row_mask[TC_HTB_MAXDEPTH]; | |
172 | struct rb_node *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
173 | u32 last_ptr_id[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
1da177e4 | 174 | |
87990467 SH |
175 | /* self wait list - roots of wait PQs per row */ |
176 | struct rb_root wait_pq[TC_HTB_MAXDEPTH]; | |
1da177e4 | 177 | |
87990467 SH |
178 | /* time of nearest event per level (row) */ |
179 | unsigned long near_ev_cache[TC_HTB_MAXDEPTH]; | |
1da177e4 | 180 | |
87990467 SH |
181 | /* cached value of jiffies in dequeue */ |
182 | unsigned long jiffies; | |
1da177e4 | 183 | |
87990467 SH |
184 | /* whether we hit non-work conserving class during this dequeue; we use */ |
185 | int nwc_hit; /* this to disable mindelay complaint in dequeue */ | |
1da177e4 | 186 | |
87990467 | 187 | int defcls; /* class where unclassified flows go to */ |
1da177e4 | 188 | |
87990467 SH |
189 | /* filters for qdisc itself */ |
190 | struct tcf_proto *filter_list; | |
191 | int filter_cnt; | |
1da177e4 | 192 | |
87990467 SH |
193 | int rate2quantum; /* quant = rate / rate2quantum */ |
194 | psched_time_t now; /* cached dequeue time */ | |
195 | struct timer_list timer; /* send delay timer */ | |
1da177e4 | 196 | #ifdef HTB_RATECM |
87990467 SH |
197 | struct timer_list rttim; /* rate computer timer */ |
198 | int recmp_bucket; /* which hash bucket to recompute next */ | |
1da177e4 | 199 | #endif |
1da177e4 | 200 | |
87990467 SH |
201 | /* non shaped skbs; let them go directly thru */ |
202 | struct sk_buff_head direct_queue; | |
203 | int direct_qlen; /* max qlen of above */ | |
204 | ||
205 | long direct_pkts; | |
1da177e4 LT |
206 | }; |
207 | ||
208 | /* compute hash of size HTB_HSIZE for given handle */ | |
87990467 | 209 | static inline int htb_hash(u32 h) |
1da177e4 LT |
210 | { |
211 | #if HTB_HSIZE != 16 | |
87990467 | 212 | #error "Declare new hash for your HTB_HSIZE" |
1da177e4 | 213 | #endif |
87990467 SH |
214 | h ^= h >> 8; /* stolen from cbq_hash */ |
215 | h ^= h >> 4; | |
216 | return h & 0xf; | |
1da177e4 LT |
217 | } |
218 | ||
219 | /* find class in global hash table using given handle */ | |
87990467 | 220 | static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch) |
1da177e4 LT |
221 | { |
222 | struct htb_sched *q = qdisc_priv(sch); | |
0cef296d SH |
223 | struct hlist_node *p; |
224 | struct htb_class *cl; | |
225 | ||
87990467 | 226 | if (TC_H_MAJ(handle) != sch->handle) |
1da177e4 | 227 | return NULL; |
87990467 | 228 | |
0cef296d | 229 | hlist_for_each_entry(cl, p, q->hash + htb_hash(handle), hlist) { |
1da177e4 LT |
230 | if (cl->classid == handle) |
231 | return cl; | |
232 | } | |
233 | return NULL; | |
234 | } | |
235 | ||
236 | /** | |
237 | * htb_classify - classify a packet into class | |
238 | * | |
239 | * It returns NULL if the packet should be dropped or -1 if the packet | |
240 | * should be passed directly thru. In all other cases leaf class is returned. | |
241 | * We allow direct class selection by classid in priority. The we examine | |
242 | * filters in qdisc and in inner nodes (if higher filter points to the inner | |
243 | * node). If we end up with classid MAJOR:0 we enqueue the skb into special | |
244 | * internal fifo (direct). These packets then go directly thru. If we still | |
245 | * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessfull | |
246 | * then finish and return direct queue. | |
247 | */ | |
248 | #define HTB_DIRECT (struct htb_class*)-1 | |
249 | static inline u32 htb_classid(struct htb_class *cl) | |
250 | { | |
251 | return (cl && cl != HTB_DIRECT) ? cl->classid : TC_H_UNSPEC; | |
252 | } | |
253 | ||
87990467 SH |
254 | static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, |
255 | int *qerr) | |
1da177e4 LT |
256 | { |
257 | struct htb_sched *q = qdisc_priv(sch); | |
258 | struct htb_class *cl; | |
259 | struct tcf_result res; | |
260 | struct tcf_proto *tcf; | |
261 | int result; | |
262 | ||
263 | /* allow to select class by setting skb->priority to valid classid; | |
264 | note that nfmark can be used too by attaching filter fw with no | |
265 | rules in it */ | |
266 | if (skb->priority == sch->handle) | |
87990467 SH |
267 | return HTB_DIRECT; /* X:0 (direct flow) selected */ |
268 | if ((cl = htb_find(skb->priority, sch)) != NULL && cl->level == 0) | |
1da177e4 LT |
269 | return cl; |
270 | ||
29f1df6c | 271 | *qerr = NET_XMIT_BYPASS; |
1da177e4 LT |
272 | tcf = q->filter_list; |
273 | while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) { | |
274 | #ifdef CONFIG_NET_CLS_ACT | |
275 | switch (result) { | |
276 | case TC_ACT_QUEUED: | |
87990467 | 277 | case TC_ACT_STOLEN: |
1da177e4 LT |
278 | *qerr = NET_XMIT_SUCCESS; |
279 | case TC_ACT_SHOT: | |
280 | return NULL; | |
281 | } | |
282 | #elif defined(CONFIG_NET_CLS_POLICE) | |
283 | if (result == TC_POLICE_SHOT) | |
284 | return HTB_DIRECT; | |
285 | #endif | |
87990467 | 286 | if ((cl = (void *)res.class) == NULL) { |
1da177e4 | 287 | if (res.classid == sch->handle) |
87990467 SH |
288 | return HTB_DIRECT; /* X:0 (direct flow) */ |
289 | if ((cl = htb_find(res.classid, sch)) == NULL) | |
290 | break; /* filter selected invalid classid */ | |
1da177e4 LT |
291 | } |
292 | if (!cl->level) | |
87990467 | 293 | return cl; /* we hit leaf; return it */ |
1da177e4 LT |
294 | |
295 | /* we have got inner class; apply inner filter chain */ | |
296 | tcf = cl->filter_list; | |
297 | } | |
298 | /* classification failed; try to use default class */ | |
87990467 | 299 | cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch); |
1da177e4 | 300 | if (!cl || cl->level) |
87990467 | 301 | return HTB_DIRECT; /* bad default .. this is safe bet */ |
1da177e4 LT |
302 | return cl; |
303 | } | |
304 | ||
1da177e4 LT |
305 | /** |
306 | * htb_add_to_id_tree - adds class to the round robin list | |
307 | * | |
308 | * Routine adds class to the list (actually tree) sorted by classid. | |
309 | * Make sure that class is not already on such list for given prio. | |
310 | */ | |
87990467 SH |
311 | static void htb_add_to_id_tree(struct rb_root *root, |
312 | struct htb_class *cl, int prio) | |
1da177e4 LT |
313 | { |
314 | struct rb_node **p = &root->rb_node, *parent = NULL; | |
3bf72957 | 315 | |
1da177e4 | 316 | while (*p) { |
87990467 SH |
317 | struct htb_class *c; |
318 | parent = *p; | |
1da177e4 | 319 | c = rb_entry(parent, struct htb_class, node[prio]); |
3bf72957 | 320 | |
1da177e4 LT |
321 | if (cl->classid > c->classid) |
322 | p = &parent->rb_right; | |
87990467 | 323 | else |
1da177e4 LT |
324 | p = &parent->rb_left; |
325 | } | |
326 | rb_link_node(&cl->node[prio], parent, p); | |
327 | rb_insert_color(&cl->node[prio], root); | |
328 | } | |
329 | ||
330 | /** | |
331 | * htb_add_to_wait_tree - adds class to the event queue with delay | |
332 | * | |
333 | * The class is added to priority event queue to indicate that class will | |
334 | * change its mode in cl->pq_key microseconds. Make sure that class is not | |
335 | * already in the queue. | |
336 | */ | |
87990467 SH |
337 | static void htb_add_to_wait_tree(struct htb_sched *q, |
338 | struct htb_class *cl, long delay) | |
1da177e4 LT |
339 | { |
340 | struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL; | |
3bf72957 | 341 | |
1da177e4 LT |
342 | cl->pq_key = q->jiffies + PSCHED_US2JIFFIE(delay); |
343 | if (cl->pq_key == q->jiffies) | |
344 | cl->pq_key++; | |
345 | ||
346 | /* update the nearest event cache */ | |
347 | if (time_after(q->near_ev_cache[cl->level], cl->pq_key)) | |
348 | q->near_ev_cache[cl->level] = cl->pq_key; | |
87990467 | 349 | |
1da177e4 | 350 | while (*p) { |
87990467 SH |
351 | struct htb_class *c; |
352 | parent = *p; | |
1da177e4 LT |
353 | c = rb_entry(parent, struct htb_class, pq_node); |
354 | if (time_after_eq(cl->pq_key, c->pq_key)) | |
355 | p = &parent->rb_right; | |
87990467 | 356 | else |
1da177e4 LT |
357 | p = &parent->rb_left; |
358 | } | |
359 | rb_link_node(&cl->pq_node, parent, p); | |
360 | rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]); | |
361 | } | |
362 | ||
363 | /** | |
364 | * htb_next_rb_node - finds next node in binary tree | |
365 | * | |
366 | * When we are past last key we return NULL. | |
367 | * Average complexity is 2 steps per call. | |
368 | */ | |
3696f625 | 369 | static inline void htb_next_rb_node(struct rb_node **n) |
1da177e4 LT |
370 | { |
371 | *n = rb_next(*n); | |
372 | } | |
373 | ||
374 | /** | |
375 | * htb_add_class_to_row - add class to its row | |
376 | * | |
377 | * The class is added to row at priorities marked in mask. | |
378 | * It does nothing if mask == 0. | |
379 | */ | |
87990467 SH |
380 | static inline void htb_add_class_to_row(struct htb_sched *q, |
381 | struct htb_class *cl, int mask) | |
1da177e4 | 382 | { |
1da177e4 LT |
383 | q->row_mask[cl->level] |= mask; |
384 | while (mask) { | |
385 | int prio = ffz(~mask); | |
386 | mask &= ~(1 << prio); | |
87990467 | 387 | htb_add_to_id_tree(q->row[cl->level] + prio, cl, prio); |
1da177e4 LT |
388 | } |
389 | } | |
390 | ||
3696f625 SH |
391 | /* If this triggers, it is a bug in this code, but it need not be fatal */ |
392 | static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root) | |
393 | { | |
81771b3b | 394 | if (RB_EMPTY_NODE(rb)) { |
3696f625 SH |
395 | WARN_ON(1); |
396 | } else { | |
397 | rb_erase(rb, root); | |
398 | RB_CLEAR_NODE(rb); | |
399 | } | |
400 | } | |
401 | ||
402 | ||
1da177e4 LT |
403 | /** |
404 | * htb_remove_class_from_row - removes class from its row | |
405 | * | |
406 | * The class is removed from row at priorities marked in mask. | |
407 | * It does nothing if mask == 0. | |
408 | */ | |
87990467 SH |
409 | static inline void htb_remove_class_from_row(struct htb_sched *q, |
410 | struct htb_class *cl, int mask) | |
1da177e4 LT |
411 | { |
412 | int m = 0; | |
3bf72957 | 413 | |
1da177e4 LT |
414 | while (mask) { |
415 | int prio = ffz(~mask); | |
3696f625 | 416 | |
1da177e4 | 417 | mask &= ~(1 << prio); |
87990467 SH |
418 | if (q->ptr[cl->level][prio] == cl->node + prio) |
419 | htb_next_rb_node(q->ptr[cl->level] + prio); | |
3696f625 SH |
420 | |
421 | htb_safe_rb_erase(cl->node + prio, q->row[cl->level] + prio); | |
87990467 | 422 | if (!q->row[cl->level][prio].rb_node) |
1da177e4 LT |
423 | m |= 1 << prio; |
424 | } | |
1da177e4 LT |
425 | q->row_mask[cl->level] &= ~m; |
426 | } | |
427 | ||
428 | /** | |
429 | * htb_activate_prios - creates active classe's feed chain | |
430 | * | |
431 | * The class is connected to ancestors and/or appropriate rows | |
432 | * for priorities it is participating on. cl->cmode must be new | |
433 | * (activated) mode. It does nothing if cl->prio_activity == 0. | |
434 | */ | |
87990467 | 435 | static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
436 | { |
437 | struct htb_class *p = cl->parent; | |
87990467 | 438 | long m, mask = cl->prio_activity; |
1da177e4 LT |
439 | |
440 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
441 | m = mask; |
442 | while (m) { | |
1da177e4 LT |
443 | int prio = ffz(~m); |
444 | m &= ~(1 << prio); | |
87990467 | 445 | |
1da177e4 LT |
446 | if (p->un.inner.feed[prio].rb_node) |
447 | /* parent already has its feed in use so that | |
448 | reset bit in mask as parent is already ok */ | |
449 | mask &= ~(1 << prio); | |
87990467 SH |
450 | |
451 | htb_add_to_id_tree(p->un.inner.feed + prio, cl, prio); | |
1da177e4 | 452 | } |
1da177e4 | 453 | p->prio_activity |= mask; |
87990467 SH |
454 | cl = p; |
455 | p = cl->parent; | |
3bf72957 | 456 | |
1da177e4 LT |
457 | } |
458 | if (cl->cmode == HTB_CAN_SEND && mask) | |
87990467 | 459 | htb_add_class_to_row(q, cl, mask); |
1da177e4 LT |
460 | } |
461 | ||
462 | /** | |
463 | * htb_deactivate_prios - remove class from feed chain | |
464 | * | |
465 | * cl->cmode must represent old mode (before deactivation). It does | |
466 | * nothing if cl->prio_activity == 0. Class is removed from all feed | |
467 | * chains and rows. | |
468 | */ | |
469 | static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) | |
470 | { | |
471 | struct htb_class *p = cl->parent; | |
87990467 | 472 | long m, mask = cl->prio_activity; |
1da177e4 LT |
473 | |
474 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
475 | m = mask; |
476 | mask = 0; | |
1da177e4 LT |
477 | while (m) { |
478 | int prio = ffz(~m); | |
479 | m &= ~(1 << prio); | |
87990467 SH |
480 | |
481 | if (p->un.inner.ptr[prio] == cl->node + prio) { | |
1da177e4 LT |
482 | /* we are removing child which is pointed to from |
483 | parent feed - forget the pointer but remember | |
484 | classid */ | |
485 | p->un.inner.last_ptr_id[prio] = cl->classid; | |
486 | p->un.inner.ptr[prio] = NULL; | |
487 | } | |
87990467 | 488 | |
3696f625 | 489 | htb_safe_rb_erase(cl->node + prio, p->un.inner.feed + prio); |
87990467 SH |
490 | |
491 | if (!p->un.inner.feed[prio].rb_node) | |
1da177e4 LT |
492 | mask |= 1 << prio; |
493 | } | |
3bf72957 | 494 | |
1da177e4 | 495 | p->prio_activity &= ~mask; |
87990467 SH |
496 | cl = p; |
497 | p = cl->parent; | |
3bf72957 | 498 | |
1da177e4 | 499 | } |
87990467 SH |
500 | if (cl->cmode == HTB_CAN_SEND && mask) |
501 | htb_remove_class_from_row(q, cl, mask); | |
1da177e4 LT |
502 | } |
503 | ||
18a63e86 SH |
504 | #if HTB_HYSTERESIS |
505 | static inline long htb_lowater(const struct htb_class *cl) | |
506 | { | |
507 | return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0; | |
508 | } | |
509 | static inline long htb_hiwater(const struct htb_class *cl) | |
510 | { | |
511 | return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0; | |
512 | } | |
513 | #else | |
514 | #define htb_lowater(cl) (0) | |
515 | #define htb_hiwater(cl) (0) | |
516 | #endif | |
517 | ||
1da177e4 LT |
518 | /** |
519 | * htb_class_mode - computes and returns current class mode | |
520 | * | |
521 | * It computes cl's mode at time cl->t_c+diff and returns it. If mode | |
522 | * is not HTB_CAN_SEND then cl->pq_key is updated to time difference | |
523 | * from now to time when cl will change its state. | |
524 | * Also it is worth to note that class mode doesn't change simply | |
525 | * at cl->{c,}tokens == 0 but there can rather be hysteresis of | |
526 | * 0 .. -cl->{c,}buffer range. It is meant to limit number of | |
527 | * mode transitions per time unit. The speed gain is about 1/6. | |
528 | */ | |
87990467 SH |
529 | static inline enum htb_cmode |
530 | htb_class_mode(struct htb_class *cl, long *diff) | |
1da177e4 | 531 | { |
87990467 | 532 | long toks; |
1da177e4 | 533 | |
87990467 SH |
534 | if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) { |
535 | *diff = -toks; | |
536 | return HTB_CANT_SEND; | |
537 | } | |
18a63e86 | 538 | |
87990467 SH |
539 | if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl)) |
540 | return HTB_CAN_SEND; | |
1da177e4 | 541 | |
87990467 SH |
542 | *diff = -toks; |
543 | return HTB_MAY_BORROW; | |
1da177e4 LT |
544 | } |
545 | ||
546 | /** | |
547 | * htb_change_class_mode - changes classe's mode | |
548 | * | |
549 | * This should be the only way how to change classe's mode under normal | |
550 | * cirsumstances. Routine will update feed lists linkage, change mode | |
551 | * and add class to the wait event queue if appropriate. New mode should | |
552 | * be different from old one and cl->pq_key has to be valid if changing | |
553 | * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). | |
554 | */ | |
87990467 | 555 | static void |
1da177e4 | 556 | htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff) |
87990467 SH |
557 | { |
558 | enum htb_cmode new_mode = htb_class_mode(cl, diff); | |
1da177e4 LT |
559 | |
560 | if (new_mode == cl->cmode) | |
87990467 SH |
561 | return; |
562 | ||
563 | if (cl->prio_activity) { /* not necessary: speed optimization */ | |
564 | if (cl->cmode != HTB_CANT_SEND) | |
565 | htb_deactivate_prios(q, cl); | |
1da177e4 | 566 | cl->cmode = new_mode; |
87990467 SH |
567 | if (new_mode != HTB_CANT_SEND) |
568 | htb_activate_prios(q, cl); | |
569 | } else | |
1da177e4 LT |
570 | cl->cmode = new_mode; |
571 | } | |
572 | ||
573 | /** | |
574 | * htb_activate - inserts leaf cl into appropriate active feeds | |
575 | * | |
576 | * Routine learns (new) priority of leaf and activates feed chain | |
577 | * for the prio. It can be called on already active leaf safely. | |
578 | * It also adds leaf into droplist. | |
579 | */ | |
87990467 | 580 | static inline void htb_activate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
581 | { |
582 | BUG_TRAP(!cl->level && cl->un.leaf.q && cl->un.leaf.q->q.qlen); | |
3bf72957 | 583 | |
1da177e4 LT |
584 | if (!cl->prio_activity) { |
585 | cl->prio_activity = 1 << (cl->un.leaf.aprio = cl->un.leaf.prio); | |
87990467 SH |
586 | htb_activate_prios(q, cl); |
587 | list_add_tail(&cl->un.leaf.drop_list, | |
588 | q->drops + cl->un.leaf.aprio); | |
1da177e4 LT |
589 | } |
590 | } | |
591 | ||
592 | /** | |
593 | * htb_deactivate - remove leaf cl from active feeds | |
594 | * | |
595 | * Make sure that leaf is active. In the other words it can't be called | |
596 | * with non-active leaf. It also removes class from the drop list. | |
597 | */ | |
87990467 | 598 | static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
599 | { |
600 | BUG_TRAP(cl->prio_activity); | |
3bf72957 | 601 | |
87990467 | 602 | htb_deactivate_prios(q, cl); |
1da177e4 LT |
603 | cl->prio_activity = 0; |
604 | list_del_init(&cl->un.leaf.drop_list); | |
605 | } | |
606 | ||
607 | static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch) | |
608 | { | |
87990467 SH |
609 | int ret; |
610 | struct htb_sched *q = qdisc_priv(sch); | |
611 | struct htb_class *cl = htb_classify(skb, sch, &ret); | |
612 | ||
613 | if (cl == HTB_DIRECT) { | |
614 | /* enqueue to helper queue */ | |
615 | if (q->direct_queue.qlen < q->direct_qlen) { | |
616 | __skb_queue_tail(&q->direct_queue, skb); | |
617 | q->direct_pkts++; | |
618 | } else { | |
619 | kfree_skb(skb); | |
620 | sch->qstats.drops++; | |
621 | return NET_XMIT_DROP; | |
622 | } | |
1da177e4 | 623 | #ifdef CONFIG_NET_CLS_ACT |
87990467 SH |
624 | } else if (!cl) { |
625 | if (ret == NET_XMIT_BYPASS) | |
626 | sch->qstats.drops++; | |
627 | kfree_skb(skb); | |
628 | return ret; | |
1da177e4 | 629 | #endif |
87990467 SH |
630 | } else if (cl->un.leaf.q->enqueue(skb, cl->un.leaf.q) != |
631 | NET_XMIT_SUCCESS) { | |
632 | sch->qstats.drops++; | |
633 | cl->qstats.drops++; | |
634 | return NET_XMIT_DROP; | |
635 | } else { | |
636 | cl->bstats.packets++; | |
637 | cl->bstats.bytes += skb->len; | |
638 | htb_activate(q, cl); | |
639 | } | |
640 | ||
641 | sch->q.qlen++; | |
642 | sch->bstats.packets++; | |
643 | sch->bstats.bytes += skb->len; | |
644 | return NET_XMIT_SUCCESS; | |
1da177e4 LT |
645 | } |
646 | ||
647 | /* TODO: requeuing packet charges it to policers again !! */ | |
648 | static int htb_requeue(struct sk_buff *skb, struct Qdisc *sch) | |
649 | { | |
87990467 SH |
650 | struct htb_sched *q = qdisc_priv(sch); |
651 | int ret = NET_XMIT_SUCCESS; | |
652 | struct htb_class *cl = htb_classify(skb, sch, &ret); | |
653 | struct sk_buff *tskb; | |
654 | ||
655 | if (cl == HTB_DIRECT || !cl) { | |
656 | /* enqueue to helper queue */ | |
657 | if (q->direct_queue.qlen < q->direct_qlen && cl) { | |
658 | __skb_queue_head(&q->direct_queue, skb); | |
659 | } else { | |
660 | __skb_queue_head(&q->direct_queue, skb); | |
661 | tskb = __skb_dequeue_tail(&q->direct_queue); | |
662 | kfree_skb(tskb); | |
663 | sch->qstats.drops++; | |
664 | return NET_XMIT_CN; | |
665 | } | |
666 | } else if (cl->un.leaf.q->ops->requeue(skb, cl->un.leaf.q) != | |
667 | NET_XMIT_SUCCESS) { | |
668 | sch->qstats.drops++; | |
669 | cl->qstats.drops++; | |
670 | return NET_XMIT_DROP; | |
671 | } else | |
672 | htb_activate(q, cl); | |
673 | ||
674 | sch->q.qlen++; | |
675 | sch->qstats.requeues++; | |
676 | return NET_XMIT_SUCCESS; | |
1da177e4 LT |
677 | } |
678 | ||
679 | static void htb_timer(unsigned long arg) | |
680 | { | |
87990467 SH |
681 | struct Qdisc *sch = (struct Qdisc *)arg; |
682 | sch->flags &= ~TCQ_F_THROTTLED; | |
683 | wmb(); | |
684 | netif_schedule(sch->dev); | |
1da177e4 LT |
685 | } |
686 | ||
687 | #ifdef HTB_RATECM | |
688 | #define RT_GEN(D,R) R+=D-(R/HTB_EWMAC);D=0 | |
689 | static void htb_rate_timer(unsigned long arg) | |
690 | { | |
87990467 | 691 | struct Qdisc *sch = (struct Qdisc *)arg; |
1da177e4 | 692 | struct htb_sched *q = qdisc_priv(sch); |
0cef296d SH |
693 | struct hlist_node *p; |
694 | struct htb_class *cl; | |
695 | ||
1da177e4 LT |
696 | |
697 | /* lock queue so that we can muck with it */ | |
9ac961ee | 698 | spin_lock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
699 | |
700 | q->rttim.expires = jiffies + HZ; | |
701 | add_timer(&q->rttim); | |
702 | ||
703 | /* scan and recompute one bucket at time */ | |
87990467 | 704 | if (++q->recmp_bucket >= HTB_HSIZE) |
1da177e4 | 705 | q->recmp_bucket = 0; |
3bf72957 | 706 | |
0cef296d | 707 | hlist_for_each_entry(cl,p, q->hash + q->recmp_bucket, hlist) { |
87990467 SH |
708 | RT_GEN(cl->sum_bytes, cl->rate_bytes); |
709 | RT_GEN(cl->sum_packets, cl->rate_packets); | |
1da177e4 | 710 | } |
9ac961ee | 711 | spin_unlock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
712 | } |
713 | #endif | |
714 | ||
715 | /** | |
716 | * htb_charge_class - charges amount "bytes" to leaf and ancestors | |
717 | * | |
718 | * Routine assumes that packet "bytes" long was dequeued from leaf cl | |
719 | * borrowing from "level". It accounts bytes to ceil leaky bucket for | |
720 | * leaf and all ancestors and to rate bucket for ancestors at levels | |
721 | * "level" and higher. It also handles possible change of mode resulting | |
722 | * from the update. Note that mode can also increase here (MAY_BORROW to | |
723 | * CAN_SEND) because we can use more precise clock that event queue here. | |
724 | * In such case we remove class from event queue first. | |
725 | */ | |
87990467 SH |
726 | static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, |
727 | int level, int bytes) | |
728 | { | |
729 | long toks, diff; | |
1da177e4 | 730 | enum htb_cmode old_mode; |
1da177e4 LT |
731 | |
732 | #define HTB_ACCNT(T,B,R) toks = diff + cl->T; \ | |
733 | if (toks > cl->B) toks = cl->B; \ | |
734 | toks -= L2T(cl, cl->R, bytes); \ | |
735 | if (toks <= -cl->mbuffer) toks = 1-cl->mbuffer; \ | |
736 | cl->T = toks | |
737 | ||
738 | while (cl) { | |
87990467 | 739 | diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32) cl->mbuffer); |
1da177e4 | 740 | if (cl->level >= level) { |
87990467 SH |
741 | if (cl->level == level) |
742 | cl->xstats.lends++; | |
743 | HTB_ACCNT(tokens, buffer, rate); | |
1da177e4 LT |
744 | } else { |
745 | cl->xstats.borrows++; | |
87990467 | 746 | cl->tokens += diff; /* we moved t_c; update tokens */ |
1da177e4 | 747 | } |
87990467 | 748 | HTB_ACCNT(ctokens, cbuffer, ceil); |
1da177e4 | 749 | cl->t_c = q->now; |
1da177e4 | 750 | |
87990467 SH |
751 | old_mode = cl->cmode; |
752 | diff = 0; | |
753 | htb_change_class_mode(q, cl, &diff); | |
1da177e4 LT |
754 | if (old_mode != cl->cmode) { |
755 | if (old_mode != HTB_CAN_SEND) | |
3696f625 | 756 | htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level); |
1da177e4 | 757 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 758 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 759 | } |
1da177e4 LT |
760 | #ifdef HTB_RATECM |
761 | /* update rate counters */ | |
87990467 SH |
762 | cl->sum_bytes += bytes; |
763 | cl->sum_packets++; | |
1da177e4 LT |
764 | #endif |
765 | ||
766 | /* update byte stats except for leaves which are already updated */ | |
767 | if (cl->level) { | |
768 | cl->bstats.bytes += bytes; | |
769 | cl->bstats.packets++; | |
770 | } | |
771 | cl = cl->parent; | |
772 | } | |
773 | } | |
774 | ||
775 | /** | |
776 | * htb_do_events - make mode changes to classes at the level | |
777 | * | |
778 | * Scans event queue for pending events and applies them. Returns jiffies to | |
779 | * next pending event (0 for no event in pq). | |
780 | * Note: Aplied are events whose have cl->pq_key <= jiffies. | |
781 | */ | |
87990467 | 782 | static long htb_do_events(struct htb_sched *q, int level) |
1da177e4 LT |
783 | { |
784 | int i; | |
3bf72957 | 785 | |
1da177e4 LT |
786 | for (i = 0; i < 500; i++) { |
787 | struct htb_class *cl; | |
788 | long diff; | |
30bdbe39 AM |
789 | struct rb_node *p = rb_first(&q->wait_pq[level]); |
790 | ||
87990467 SH |
791 | if (!p) |
792 | return 0; | |
1da177e4 LT |
793 | |
794 | cl = rb_entry(p, struct htb_class, pq_node); | |
795 | if (time_after(cl->pq_key, q->jiffies)) { | |
1da177e4 LT |
796 | return cl->pq_key - q->jiffies; |
797 | } | |
3696f625 | 798 | htb_safe_rb_erase(p, q->wait_pq + level); |
87990467 SH |
799 | diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32) cl->mbuffer); |
800 | htb_change_class_mode(q, cl, &diff); | |
1da177e4 | 801 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 802 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 LT |
803 | } |
804 | if (net_ratelimit()) | |
805 | printk(KERN_WARNING "htb: too many events !\n"); | |
87990467 | 806 | return HZ / 10; |
1da177e4 LT |
807 | } |
808 | ||
809 | /* Returns class->node+prio from id-tree where classe's id is >= id. NULL | |
810 | is no such one exists. */ | |
87990467 SH |
811 | static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n, |
812 | u32 id) | |
1da177e4 LT |
813 | { |
814 | struct rb_node *r = NULL; | |
815 | while (n) { | |
87990467 SH |
816 | struct htb_class *cl = |
817 | rb_entry(n, struct htb_class, node[prio]); | |
818 | if (id == cl->classid) | |
819 | return n; | |
820 | ||
1da177e4 LT |
821 | if (id > cl->classid) { |
822 | n = n->rb_right; | |
823 | } else { | |
824 | r = n; | |
825 | n = n->rb_left; | |
826 | } | |
827 | } | |
828 | return r; | |
829 | } | |
830 | ||
831 | /** | |
832 | * htb_lookup_leaf - returns next leaf class in DRR order | |
833 | * | |
834 | * Find leaf where current feed pointers points to. | |
835 | */ | |
87990467 SH |
836 | static struct htb_class *htb_lookup_leaf(struct rb_root *tree, int prio, |
837 | struct rb_node **pptr, u32 * pid) | |
1da177e4 LT |
838 | { |
839 | int i; | |
840 | struct { | |
841 | struct rb_node *root; | |
842 | struct rb_node **pptr; | |
843 | u32 *pid; | |
87990467 SH |
844 | } stk[TC_HTB_MAXDEPTH], *sp = stk; |
845 | ||
1da177e4 LT |
846 | BUG_TRAP(tree->rb_node); |
847 | sp->root = tree->rb_node; | |
848 | sp->pptr = pptr; | |
849 | sp->pid = pid; | |
850 | ||
851 | for (i = 0; i < 65535; i++) { | |
87990467 | 852 | if (!*sp->pptr && *sp->pid) { |
1da177e4 LT |
853 | /* ptr was invalidated but id is valid - try to recover |
854 | the original or next ptr */ | |
87990467 SH |
855 | *sp->pptr = |
856 | htb_id_find_next_upper(prio, sp->root, *sp->pid); | |
1da177e4 | 857 | } |
87990467 SH |
858 | *sp->pid = 0; /* ptr is valid now so that remove this hint as it |
859 | can become out of date quickly */ | |
860 | if (!*sp->pptr) { /* we are at right end; rewind & go up */ | |
1da177e4 | 861 | *sp->pptr = sp->root; |
87990467 | 862 | while ((*sp->pptr)->rb_left) |
1da177e4 LT |
863 | *sp->pptr = (*sp->pptr)->rb_left; |
864 | if (sp > stk) { | |
865 | sp--; | |
87990467 SH |
866 | BUG_TRAP(*sp->pptr); |
867 | if (!*sp->pptr) | |
868 | return NULL; | |
869 | htb_next_rb_node(sp->pptr); | |
1da177e4 LT |
870 | } |
871 | } else { | |
872 | struct htb_class *cl; | |
87990467 SH |
873 | cl = rb_entry(*sp->pptr, struct htb_class, node[prio]); |
874 | if (!cl->level) | |
1da177e4 LT |
875 | return cl; |
876 | (++sp)->root = cl->un.inner.feed[prio].rb_node; | |
87990467 SH |
877 | sp->pptr = cl->un.inner.ptr + prio; |
878 | sp->pid = cl->un.inner.last_ptr_id + prio; | |
1da177e4 LT |
879 | } |
880 | } | |
881 | BUG_TRAP(0); | |
882 | return NULL; | |
883 | } | |
884 | ||
885 | /* dequeues packet at given priority and level; call only if | |
886 | you are sure that there is active class at prio/level */ | |
87990467 SH |
887 | static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, int prio, |
888 | int level) | |
1da177e4 LT |
889 | { |
890 | struct sk_buff *skb = NULL; | |
87990467 | 891 | struct htb_class *cl, *start; |
1da177e4 | 892 | /* look initial class up in the row */ |
87990467 SH |
893 | start = cl = htb_lookup_leaf(q->row[level] + prio, prio, |
894 | q->ptr[level] + prio, | |
895 | q->last_ptr_id[level] + prio); | |
896 | ||
1da177e4 LT |
897 | do { |
898 | next: | |
87990467 SH |
899 | BUG_TRAP(cl); |
900 | if (!cl) | |
901 | return NULL; | |
1da177e4 LT |
902 | |
903 | /* class can be empty - it is unlikely but can be true if leaf | |
904 | qdisc drops packets in enqueue routine or if someone used | |
905 | graft operation on the leaf since last dequeue; | |
906 | simply deactivate and skip such class */ | |
907 | if (unlikely(cl->un.leaf.q->q.qlen == 0)) { | |
908 | struct htb_class *next; | |
87990467 | 909 | htb_deactivate(q, cl); |
1da177e4 LT |
910 | |
911 | /* row/level might become empty */ | |
912 | if ((q->row_mask[level] & (1 << prio)) == 0) | |
87990467 | 913 | return NULL; |
1da177e4 | 914 | |
87990467 SH |
915 | next = htb_lookup_leaf(q->row[level] + prio, |
916 | prio, q->ptr[level] + prio, | |
917 | q->last_ptr_id[level] + prio); | |
918 | ||
919 | if (cl == start) /* fix start if we just deleted it */ | |
1da177e4 LT |
920 | start = next; |
921 | cl = next; | |
922 | goto next; | |
923 | } | |
87990467 SH |
924 | |
925 | skb = cl->un.leaf.q->dequeue(cl->un.leaf.q); | |
926 | if (likely(skb != NULL)) | |
1da177e4 LT |
927 | break; |
928 | if (!cl->warned) { | |
87990467 SH |
929 | printk(KERN_WARNING |
930 | "htb: class %X isn't work conserving ?!\n", | |
931 | cl->classid); | |
1da177e4 LT |
932 | cl->warned = 1; |
933 | } | |
934 | q->nwc_hit++; | |
87990467 SH |
935 | htb_next_rb_node((level ? cl->parent->un.inner.ptr : q-> |
936 | ptr[0]) + prio); | |
937 | cl = htb_lookup_leaf(q->row[level] + prio, prio, | |
938 | q->ptr[level] + prio, | |
939 | q->last_ptr_id[level] + prio); | |
1da177e4 LT |
940 | |
941 | } while (cl != start); | |
942 | ||
943 | if (likely(skb != NULL)) { | |
944 | if ((cl->un.leaf.deficit[level] -= skb->len) < 0) { | |
1da177e4 | 945 | cl->un.leaf.deficit[level] += cl->un.leaf.quantum; |
87990467 SH |
946 | htb_next_rb_node((level ? cl->parent->un.inner.ptr : q-> |
947 | ptr[0]) + prio); | |
1da177e4 LT |
948 | } |
949 | /* this used to be after charge_class but this constelation | |
950 | gives us slightly better performance */ | |
951 | if (!cl->un.leaf.q->q.qlen) | |
87990467 SH |
952 | htb_deactivate(q, cl); |
953 | htb_charge_class(q, cl, level, skb->len); | |
1da177e4 LT |
954 | } |
955 | return skb; | |
956 | } | |
957 | ||
87990467 | 958 | static void htb_delay_by(struct Qdisc *sch, long delay) |
1da177e4 LT |
959 | { |
960 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 SH |
961 | if (delay <= 0) |
962 | delay = 1; | |
963 | if (unlikely(delay > 5 * HZ)) { | |
1da177e4 LT |
964 | if (net_ratelimit()) |
965 | printk(KERN_INFO "HTB delay %ld > 5sec\n", delay); | |
87990467 | 966 | delay = 5 * HZ; |
1da177e4 LT |
967 | } |
968 | /* why don't use jiffies here ? because expires can be in past */ | |
969 | mod_timer(&q->timer, q->jiffies + delay); | |
970 | sch->flags |= TCQ_F_THROTTLED; | |
971 | sch->qstats.overlimits++; | |
1da177e4 LT |
972 | } |
973 | ||
974 | static struct sk_buff *htb_dequeue(struct Qdisc *sch) | |
975 | { | |
976 | struct sk_buff *skb = NULL; | |
977 | struct htb_sched *q = qdisc_priv(sch); | |
978 | int level; | |
979 | long min_delay; | |
1da177e4 LT |
980 | |
981 | q->jiffies = jiffies; | |
1da177e4 LT |
982 | |
983 | /* try to dequeue direct packets as high prio (!) to minimize cpu work */ | |
87990467 SH |
984 | skb = __skb_dequeue(&q->direct_queue); |
985 | if (skb != NULL) { | |
1da177e4 LT |
986 | sch->flags &= ~TCQ_F_THROTTLED; |
987 | sch->q.qlen--; | |
988 | return skb; | |
989 | } | |
990 | ||
87990467 SH |
991 | if (!sch->q.qlen) |
992 | goto fin; | |
1da177e4 LT |
993 | PSCHED_GET_TIME(q->now); |
994 | ||
995 | min_delay = LONG_MAX; | |
996 | q->nwc_hit = 0; | |
997 | for (level = 0; level < TC_HTB_MAXDEPTH; level++) { | |
998 | /* common case optimization - skip event handler quickly */ | |
999 | int m; | |
1000 | long delay; | |
1001 | if (time_after_eq(q->jiffies, q->near_ev_cache[level])) { | |
87990467 SH |
1002 | delay = htb_do_events(q, level); |
1003 | q->near_ev_cache[level] = | |
1004 | q->jiffies + (delay ? delay : HZ); | |
1da177e4 | 1005 | } else |
87990467 SH |
1006 | delay = q->near_ev_cache[level] - q->jiffies; |
1007 | ||
1008 | if (delay && min_delay > delay) | |
1da177e4 LT |
1009 | min_delay = delay; |
1010 | m = ~q->row_mask[level]; | |
1011 | while (m != (int)(-1)) { | |
87990467 | 1012 | int prio = ffz(m); |
1da177e4 | 1013 | m |= 1 << prio; |
87990467 | 1014 | skb = htb_dequeue_tree(q, prio, level); |
1da177e4 LT |
1015 | if (likely(skb != NULL)) { |
1016 | sch->q.qlen--; | |
1017 | sch->flags &= ~TCQ_F_THROTTLED; | |
1018 | goto fin; | |
1019 | } | |
1020 | } | |
1021 | } | |
87990467 | 1022 | htb_delay_by(sch, min_delay > 5 * HZ ? 5 * HZ : min_delay); |
1da177e4 | 1023 | fin: |
1da177e4 LT |
1024 | return skb; |
1025 | } | |
1026 | ||
1027 | /* try to drop from each class (by prio) until one succeed */ | |
87990467 | 1028 | static unsigned int htb_drop(struct Qdisc *sch) |
1da177e4 LT |
1029 | { |
1030 | struct htb_sched *q = qdisc_priv(sch); | |
1031 | int prio; | |
1032 | ||
1033 | for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) { | |
1034 | struct list_head *p; | |
87990467 | 1035 | list_for_each(p, q->drops + prio) { |
1da177e4 LT |
1036 | struct htb_class *cl = list_entry(p, struct htb_class, |
1037 | un.leaf.drop_list); | |
1038 | unsigned int len; | |
87990467 SH |
1039 | if (cl->un.leaf.q->ops->drop && |
1040 | (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) { | |
1da177e4 LT |
1041 | sch->q.qlen--; |
1042 | if (!cl->un.leaf.q->q.qlen) | |
87990467 | 1043 | htb_deactivate(q, cl); |
1da177e4 LT |
1044 | return len; |
1045 | } | |
1046 | } | |
1047 | } | |
1048 | return 0; | |
1049 | } | |
1050 | ||
1051 | /* reset all classes */ | |
1052 | /* always caled under BH & queue lock */ | |
87990467 | 1053 | static void htb_reset(struct Qdisc *sch) |
1da177e4 LT |
1054 | { |
1055 | struct htb_sched *q = qdisc_priv(sch); | |
1056 | int i; | |
1da177e4 LT |
1057 | |
1058 | for (i = 0; i < HTB_HSIZE; i++) { | |
0cef296d SH |
1059 | struct hlist_node *p; |
1060 | struct htb_class *cl; | |
1061 | ||
1062 | hlist_for_each_entry(cl, p, q->hash + i, hlist) { | |
1da177e4 | 1063 | if (cl->level) |
87990467 | 1064 | memset(&cl->un.inner, 0, sizeof(cl->un.inner)); |
1da177e4 | 1065 | else { |
87990467 | 1066 | if (cl->un.leaf.q) |
1da177e4 LT |
1067 | qdisc_reset(cl->un.leaf.q); |
1068 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); | |
1069 | } | |
1070 | cl->prio_activity = 0; | |
1071 | cl->cmode = HTB_CAN_SEND; | |
1da177e4 LT |
1072 | |
1073 | } | |
1074 | } | |
1075 | sch->flags &= ~TCQ_F_THROTTLED; | |
1076 | del_timer(&q->timer); | |
1077 | __skb_queue_purge(&q->direct_queue); | |
1078 | sch->q.qlen = 0; | |
87990467 SH |
1079 | memset(q->row, 0, sizeof(q->row)); |
1080 | memset(q->row_mask, 0, sizeof(q->row_mask)); | |
1081 | memset(q->wait_pq, 0, sizeof(q->wait_pq)); | |
1082 | memset(q->ptr, 0, sizeof(q->ptr)); | |
1da177e4 | 1083 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 1084 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 LT |
1085 | } |
1086 | ||
1087 | static int htb_init(struct Qdisc *sch, struct rtattr *opt) | |
1088 | { | |
1089 | struct htb_sched *q = qdisc_priv(sch); | |
1090 | struct rtattr *tb[TCA_HTB_INIT]; | |
1091 | struct tc_htb_glob *gopt; | |
1092 | int i; | |
1da177e4 | 1093 | if (!opt || rtattr_parse_nested(tb, TCA_HTB_INIT, opt) || |
87990467 SH |
1094 | tb[TCA_HTB_INIT - 1] == NULL || |
1095 | RTA_PAYLOAD(tb[TCA_HTB_INIT - 1]) < sizeof(*gopt)) { | |
1da177e4 LT |
1096 | printk(KERN_ERR "HTB: hey probably you have bad tc tool ?\n"); |
1097 | return -EINVAL; | |
1098 | } | |
87990467 | 1099 | gopt = RTA_DATA(tb[TCA_HTB_INIT - 1]); |
1da177e4 | 1100 | if (gopt->version != HTB_VER >> 16) { |
87990467 SH |
1101 | printk(KERN_ERR |
1102 | "HTB: need tc/htb version %d (minor is %d), you have %d\n", | |
1103 | HTB_VER >> 16, HTB_VER & 0xffff, gopt->version); | |
1da177e4 LT |
1104 | return -EINVAL; |
1105 | } | |
1da177e4 LT |
1106 | |
1107 | INIT_LIST_HEAD(&q->root); | |
1108 | for (i = 0; i < HTB_HSIZE; i++) | |
0cef296d | 1109 | INIT_HLIST_HEAD(q->hash + i); |
1da177e4 | 1110 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 1111 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 LT |
1112 | |
1113 | init_timer(&q->timer); | |
1114 | skb_queue_head_init(&q->direct_queue); | |
1115 | ||
1116 | q->direct_qlen = sch->dev->tx_queue_len; | |
87990467 | 1117 | if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */ |
1da177e4 LT |
1118 | q->direct_qlen = 2; |
1119 | q->timer.function = htb_timer; | |
1120 | q->timer.data = (unsigned long)sch; | |
1121 | ||
1122 | #ifdef HTB_RATECM | |
1123 | init_timer(&q->rttim); | |
1124 | q->rttim.function = htb_rate_timer; | |
1125 | q->rttim.data = (unsigned long)sch; | |
1126 | q->rttim.expires = jiffies + HZ; | |
1127 | add_timer(&q->rttim); | |
1128 | #endif | |
1129 | if ((q->rate2quantum = gopt->rate2quantum) < 1) | |
1130 | q->rate2quantum = 1; | |
1131 | q->defcls = gopt->defcls; | |
1132 | ||
1133 | return 0; | |
1134 | } | |
1135 | ||
1136 | static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) | |
1137 | { | |
1138 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1139 | unsigned char *b = skb->tail; |
1da177e4 LT |
1140 | struct rtattr *rta; |
1141 | struct tc_htb_glob gopt; | |
9ac961ee | 1142 | spin_lock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
1143 | gopt.direct_pkts = q->direct_pkts; |
1144 | ||
1da177e4 LT |
1145 | gopt.version = HTB_VER; |
1146 | gopt.rate2quantum = q->rate2quantum; | |
1147 | gopt.defcls = q->defcls; | |
3bf72957 | 1148 | gopt.debug = 0; |
87990467 | 1149 | rta = (struct rtattr *)b; |
1da177e4 LT |
1150 | RTA_PUT(skb, TCA_OPTIONS, 0, NULL); |
1151 | RTA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt); | |
1152 | rta->rta_len = skb->tail - b; | |
9ac961ee | 1153 | spin_unlock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
1154 | return skb->len; |
1155 | rtattr_failure: | |
9ac961ee | 1156 | spin_unlock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
1157 | skb_trim(skb, skb->tail - skb->data); |
1158 | return -1; | |
1159 | } | |
1160 | ||
1161 | static int htb_dump_class(struct Qdisc *sch, unsigned long arg, | |
87990467 | 1162 | struct sk_buff *skb, struct tcmsg *tcm) |
1da177e4 | 1163 | { |
87990467 SH |
1164 | struct htb_class *cl = (struct htb_class *)arg; |
1165 | unsigned char *b = skb->tail; | |
1da177e4 LT |
1166 | struct rtattr *rta; |
1167 | struct tc_htb_opt opt; | |
1168 | ||
9ac961ee | 1169 | spin_lock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
1170 | tcm->tcm_parent = cl->parent ? cl->parent->classid : TC_H_ROOT; |
1171 | tcm->tcm_handle = cl->classid; | |
1172 | if (!cl->level && cl->un.leaf.q) | |
1173 | tcm->tcm_info = cl->un.leaf.q->handle; | |
1174 | ||
87990467 | 1175 | rta = (struct rtattr *)b; |
1da177e4 LT |
1176 | RTA_PUT(skb, TCA_OPTIONS, 0, NULL); |
1177 | ||
87990467 | 1178 | memset(&opt, 0, sizeof(opt)); |
1da177e4 | 1179 | |
87990467 SH |
1180 | opt.rate = cl->rate->rate; |
1181 | opt.buffer = cl->buffer; | |
1182 | opt.ceil = cl->ceil->rate; | |
1183 | opt.cbuffer = cl->cbuffer; | |
1184 | opt.quantum = cl->un.leaf.quantum; | |
1185 | opt.prio = cl->un.leaf.prio; | |
1186 | opt.level = cl->level; | |
1da177e4 LT |
1187 | RTA_PUT(skb, TCA_HTB_PARMS, sizeof(opt), &opt); |
1188 | rta->rta_len = skb->tail - b; | |
9ac961ee | 1189 | spin_unlock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
1190 | return skb->len; |
1191 | rtattr_failure: | |
9ac961ee | 1192 | spin_unlock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
1193 | skb_trim(skb, b - skb->data); |
1194 | return -1; | |
1195 | } | |
1196 | ||
1197 | static int | |
87990467 | 1198 | htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d) |
1da177e4 | 1199 | { |
87990467 | 1200 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1201 | |
1202 | #ifdef HTB_RATECM | |
87990467 SH |
1203 | cl->rate_est.bps = cl->rate_bytes / (HTB_EWMAC * HTB_HSIZE); |
1204 | cl->rate_est.pps = cl->rate_packets / (HTB_EWMAC * HTB_HSIZE); | |
1da177e4 LT |
1205 | #endif |
1206 | ||
1207 | if (!cl->level && cl->un.leaf.q) | |
1208 | cl->qstats.qlen = cl->un.leaf.q->q.qlen; | |
1209 | cl->xstats.tokens = cl->tokens; | |
1210 | cl->xstats.ctokens = cl->ctokens; | |
1211 | ||
1212 | if (gnet_stats_copy_basic(d, &cl->bstats) < 0 || | |
1213 | gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || | |
1214 | gnet_stats_copy_queue(d, &cl->qstats) < 0) | |
1215 | return -1; | |
1216 | ||
1217 | return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); | |
1218 | } | |
1219 | ||
1220 | static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
87990467 | 1221 | struct Qdisc **old) |
1da177e4 | 1222 | { |
87990467 | 1223 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1224 | |
1225 | if (cl && !cl->level) { | |
9f9afec4 PM |
1226 | if (new == NULL && |
1227 | (new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, | |
1228 | cl->classid)) | |
87990467 SH |
1229 | == NULL) |
1230 | return -ENOBUFS; | |
1da177e4 LT |
1231 | sch_tree_lock(sch); |
1232 | if ((*old = xchg(&cl->un.leaf.q, new)) != NULL) { | |
1233 | if (cl->prio_activity) | |
87990467 | 1234 | htb_deactivate(qdisc_priv(sch), cl); |
1da177e4 LT |
1235 | |
1236 | /* TODO: is it correct ? Why CBQ doesn't do it ? */ | |
87990467 | 1237 | sch->q.qlen -= (*old)->q.qlen; |
1da177e4 LT |
1238 | qdisc_reset(*old); |
1239 | } | |
1240 | sch_tree_unlock(sch); | |
1241 | return 0; | |
1242 | } | |
1243 | return -ENOENT; | |
1244 | } | |
1245 | ||
87990467 | 1246 | static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg) |
1da177e4 | 1247 | { |
87990467 | 1248 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1249 | return (cl && !cl->level) ? cl->un.leaf.q : NULL; |
1250 | } | |
1251 | ||
1252 | static unsigned long htb_get(struct Qdisc *sch, u32 classid) | |
1253 | { | |
87990467 SH |
1254 | struct htb_class *cl = htb_find(classid, sch); |
1255 | if (cl) | |
1da177e4 LT |
1256 | cl->refcnt++; |
1257 | return (unsigned long)cl; | |
1258 | } | |
1259 | ||
1260 | static void htb_destroy_filters(struct tcf_proto **fl) | |
1261 | { | |
1262 | struct tcf_proto *tp; | |
1263 | ||
1264 | while ((tp = *fl) != NULL) { | |
1265 | *fl = tp->next; | |
1266 | tcf_destroy(tp); | |
1267 | } | |
1268 | } | |
1269 | ||
87990467 | 1270 | static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl) |
1da177e4 LT |
1271 | { |
1272 | struct htb_sched *q = qdisc_priv(sch); | |
814a175e | 1273 | |
1da177e4 LT |
1274 | if (!cl->level) { |
1275 | BUG_TRAP(cl->un.leaf.q); | |
1da177e4 LT |
1276 | qdisc_destroy(cl->un.leaf.q); |
1277 | } | |
1278 | qdisc_put_rtab(cl->rate); | |
1279 | qdisc_put_rtab(cl->ceil); | |
87990467 SH |
1280 | |
1281 | htb_destroy_filters(&cl->filter_list); | |
1282 | ||
1283 | while (!list_empty(&cl->children)) | |
1284 | htb_destroy_class(sch, list_entry(cl->children.next, | |
1285 | struct htb_class, sibling)); | |
1da177e4 LT |
1286 | |
1287 | /* note: this delete may happen twice (see htb_delete) */ | |
da33e3eb | 1288 | hlist_del_init(&cl->hlist); |
1da177e4 | 1289 | list_del(&cl->sibling); |
87990467 | 1290 | |
1da177e4 | 1291 | if (cl->prio_activity) |
87990467 SH |
1292 | htb_deactivate(q, cl); |
1293 | ||
1da177e4 | 1294 | if (cl->cmode != HTB_CAN_SEND) |
3696f625 | 1295 | htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level); |
87990467 | 1296 | |
1da177e4 LT |
1297 | kfree(cl); |
1298 | } | |
1299 | ||
1300 | /* always caled under BH & queue lock */ | |
87990467 | 1301 | static void htb_destroy(struct Qdisc *sch) |
1da177e4 LT |
1302 | { |
1303 | struct htb_sched *q = qdisc_priv(sch); | |
1da177e4 | 1304 | |
87990467 | 1305 | del_timer_sync(&q->timer); |
1da177e4 | 1306 | #ifdef HTB_RATECM |
87990467 | 1307 | del_timer_sync(&q->rttim); |
1da177e4 LT |
1308 | #endif |
1309 | /* This line used to be after htb_destroy_class call below | |
1310 | and surprisingly it worked in 2.4. But it must precede it | |
1311 | because filter need its target class alive to be able to call | |
1312 | unbind_filter on it (without Oops). */ | |
1313 | htb_destroy_filters(&q->filter_list); | |
87990467 SH |
1314 | |
1315 | while (!list_empty(&q->root)) | |
1316 | htb_destroy_class(sch, list_entry(q->root.next, | |
1317 | struct htb_class, sibling)); | |
1da177e4 LT |
1318 | |
1319 | __skb_queue_purge(&q->direct_queue); | |
1320 | } | |
1321 | ||
1322 | static int htb_delete(struct Qdisc *sch, unsigned long arg) | |
1323 | { | |
1324 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1325 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1326 | |
1327 | // TODO: why don't allow to delete subtree ? references ? does | |
1328 | // tc subsys quarantee us that in htb_destroy it holds no class | |
1329 | // refs so that we can remove children safely there ? | |
1330 | if (!list_empty(&cl->children) || cl->filter_cnt) | |
1331 | return -EBUSY; | |
87990467 | 1332 | |
1da177e4 | 1333 | sch_tree_lock(sch); |
87990467 | 1334 | |
1da177e4 | 1335 | /* delete from hash and active; remainder in destroy_class */ |
da33e3eb | 1336 | hlist_del_init(&cl->hlist); |
0cef296d | 1337 | |
814a175e PM |
1338 | if (!cl->level) { |
1339 | sch->q.qlen -= cl->un.leaf.q->q.qlen; | |
1340 | qdisc_reset(cl->un.leaf.q); | |
1341 | } | |
1342 | ||
1da177e4 | 1343 | if (cl->prio_activity) |
87990467 | 1344 | htb_deactivate(q, cl); |
1da177e4 LT |
1345 | |
1346 | if (--cl->refcnt == 0) | |
87990467 | 1347 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1348 | |
1349 | sch_tree_unlock(sch); | |
1350 | return 0; | |
1351 | } | |
1352 | ||
1353 | static void htb_put(struct Qdisc *sch, unsigned long arg) | |
1354 | { | |
87990467 | 1355 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1356 | |
1357 | if (--cl->refcnt == 0) | |
87990467 | 1358 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1359 | } |
1360 | ||
87990467 SH |
1361 | static int htb_change_class(struct Qdisc *sch, u32 classid, |
1362 | u32 parentid, struct rtattr **tca, | |
1363 | unsigned long *arg) | |
1da177e4 LT |
1364 | { |
1365 | int err = -EINVAL; | |
1366 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 SH |
1367 | struct htb_class *cl = (struct htb_class *)*arg, *parent; |
1368 | struct rtattr *opt = tca[TCA_OPTIONS - 1]; | |
1da177e4 LT |
1369 | struct qdisc_rate_table *rtab = NULL, *ctab = NULL; |
1370 | struct rtattr *tb[TCA_HTB_RTAB]; | |
1371 | struct tc_htb_opt *hopt; | |
1372 | ||
1373 | /* extract all subattrs from opt attr */ | |
1374 | if (!opt || rtattr_parse_nested(tb, TCA_HTB_RTAB, opt) || | |
87990467 SH |
1375 | tb[TCA_HTB_PARMS - 1] == NULL || |
1376 | RTA_PAYLOAD(tb[TCA_HTB_PARMS - 1]) < sizeof(*hopt)) | |
1da177e4 | 1377 | goto failure; |
1da177e4 | 1378 | |
87990467 SH |
1379 | parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch); |
1380 | ||
1381 | hopt = RTA_DATA(tb[TCA_HTB_PARMS - 1]); | |
3bf72957 | 1382 | |
87990467 SH |
1383 | rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB - 1]); |
1384 | ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB - 1]); | |
1385 | if (!rtab || !ctab) | |
1386 | goto failure; | |
1da177e4 | 1387 | |
87990467 | 1388 | if (!cl) { /* new class */ |
1da177e4 | 1389 | struct Qdisc *new_q; |
3696f625 SH |
1390 | int prio; |
1391 | ||
1da177e4 | 1392 | /* check for valid classid */ |
87990467 SH |
1393 | if (!classid || TC_H_MAJ(classid ^ sch->handle) |
1394 | || htb_find(classid, sch)) | |
1da177e4 LT |
1395 | goto failure; |
1396 | ||
1397 | /* check maximal depth */ | |
1398 | if (parent && parent->parent && parent->parent->level < 2) { | |
1399 | printk(KERN_ERR "htb: tree is too deep\n"); | |
1400 | goto failure; | |
1401 | } | |
1402 | err = -ENOBUFS; | |
0da974f4 | 1403 | if ((cl = kzalloc(sizeof(*cl), GFP_KERNEL)) == NULL) |
1da177e4 | 1404 | goto failure; |
87990467 | 1405 | |
1da177e4 LT |
1406 | cl->refcnt = 1; |
1407 | INIT_LIST_HEAD(&cl->sibling); | |
0cef296d | 1408 | INIT_HLIST_NODE(&cl->hlist); |
1da177e4 LT |
1409 | INIT_LIST_HEAD(&cl->children); |
1410 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); | |
3696f625 SH |
1411 | RB_CLEAR_NODE(&cl->pq_node); |
1412 | ||
1413 | for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) | |
1414 | RB_CLEAR_NODE(&cl->node[prio]); | |
1da177e4 LT |
1415 | |
1416 | /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL) | |
1417 | so that can't be used inside of sch_tree_lock | |
1418 | -- thanks to Karlis Peisenieks */ | |
9f9afec4 | 1419 | new_q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, classid); |
1da177e4 LT |
1420 | sch_tree_lock(sch); |
1421 | if (parent && !parent->level) { | |
1422 | /* turn parent into inner node */ | |
1423 | sch->q.qlen -= parent->un.leaf.q->q.qlen; | |
87990467 SH |
1424 | qdisc_destroy(parent->un.leaf.q); |
1425 | if (parent->prio_activity) | |
1426 | htb_deactivate(q, parent); | |
1da177e4 LT |
1427 | |
1428 | /* remove from evt list because of level change */ | |
1429 | if (parent->cmode != HTB_CAN_SEND) { | |
3696f625 | 1430 | htb_safe_rb_erase(&parent->pq_node, q->wait_pq); |
1da177e4 LT |
1431 | parent->cmode = HTB_CAN_SEND; |
1432 | } | |
1433 | parent->level = (parent->parent ? parent->parent->level | |
87990467 SH |
1434 | : TC_HTB_MAXDEPTH) - 1; |
1435 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1da177e4 LT |
1436 | } |
1437 | /* leaf (we) needs elementary qdisc */ | |
1438 | cl->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
1439 | ||
87990467 SH |
1440 | cl->classid = classid; |
1441 | cl->parent = parent; | |
1da177e4 LT |
1442 | |
1443 | /* set class to be in HTB_CAN_SEND state */ | |
1444 | cl->tokens = hopt->buffer; | |
1445 | cl->ctokens = hopt->cbuffer; | |
87990467 | 1446 | cl->mbuffer = PSCHED_JIFFIE2US(HZ * 60); /* 1min */ |
1da177e4 LT |
1447 | PSCHED_GET_TIME(cl->t_c); |
1448 | cl->cmode = HTB_CAN_SEND; | |
1449 | ||
1450 | /* attach to the hash list and parent's family */ | |
0cef296d | 1451 | hlist_add_head(&cl->hlist, q->hash + htb_hash(classid)); |
87990467 SH |
1452 | list_add_tail(&cl->sibling, |
1453 | parent ? &parent->children : &q->root); | |
1454 | } else | |
1455 | sch_tree_lock(sch); | |
1da177e4 LT |
1456 | |
1457 | /* it used to be a nasty bug here, we have to check that node | |
87990467 | 1458 | is really leaf before changing cl->un.leaf ! */ |
1da177e4 LT |
1459 | if (!cl->level) { |
1460 | cl->un.leaf.quantum = rtab->rate.rate / q->rate2quantum; | |
1461 | if (!hopt->quantum && cl->un.leaf.quantum < 1000) { | |
87990467 SH |
1462 | printk(KERN_WARNING |
1463 | "HTB: quantum of class %X is small. Consider r2q change.\n", | |
1464 | cl->classid); | |
1da177e4 LT |
1465 | cl->un.leaf.quantum = 1000; |
1466 | } | |
1467 | if (!hopt->quantum && cl->un.leaf.quantum > 200000) { | |
87990467 SH |
1468 | printk(KERN_WARNING |
1469 | "HTB: quantum of class %X is big. Consider r2q change.\n", | |
1470 | cl->classid); | |
1da177e4 LT |
1471 | cl->un.leaf.quantum = 200000; |
1472 | } | |
1473 | if (hopt->quantum) | |
1474 | cl->un.leaf.quantum = hopt->quantum; | |
1475 | if ((cl->un.leaf.prio = hopt->prio) >= TC_HTB_NUMPRIO) | |
1476 | cl->un.leaf.prio = TC_HTB_NUMPRIO - 1; | |
1477 | } | |
1478 | ||
1479 | cl->buffer = hopt->buffer; | |
1480 | cl->cbuffer = hopt->cbuffer; | |
87990467 SH |
1481 | if (cl->rate) |
1482 | qdisc_put_rtab(cl->rate); | |
1483 | cl->rate = rtab; | |
1484 | if (cl->ceil) | |
1485 | qdisc_put_rtab(cl->ceil); | |
1486 | cl->ceil = ctab; | |
1da177e4 LT |
1487 | sch_tree_unlock(sch); |
1488 | ||
1489 | *arg = (unsigned long)cl; | |
1490 | return 0; | |
1491 | ||
1492 | failure: | |
87990467 SH |
1493 | if (rtab) |
1494 | qdisc_put_rtab(rtab); | |
1495 | if (ctab) | |
1496 | qdisc_put_rtab(ctab); | |
1da177e4 LT |
1497 | return err; |
1498 | } | |
1499 | ||
1500 | static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg) | |
1501 | { | |
1502 | struct htb_sched *q = qdisc_priv(sch); | |
1503 | struct htb_class *cl = (struct htb_class *)arg; | |
1504 | struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list; | |
3bf72957 | 1505 | |
1da177e4 LT |
1506 | return fl; |
1507 | } | |
1508 | ||
1509 | static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, | |
87990467 | 1510 | u32 classid) |
1da177e4 LT |
1511 | { |
1512 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1513 | struct htb_class *cl = htb_find(classid, sch); |
3bf72957 | 1514 | |
1da177e4 | 1515 | /*if (cl && !cl->level) return 0; |
87990467 SH |
1516 | The line above used to be there to prevent attaching filters to |
1517 | leaves. But at least tc_index filter uses this just to get class | |
1518 | for other reasons so that we have to allow for it. | |
1519 | ---- | |
1520 | 19.6.2002 As Werner explained it is ok - bind filter is just | |
1521 | another way to "lock" the class - unlike "get" this lock can | |
1522 | be broken by class during destroy IIUC. | |
1da177e4 | 1523 | */ |
87990467 SH |
1524 | if (cl) |
1525 | cl->filter_cnt++; | |
1526 | else | |
1da177e4 LT |
1527 | q->filter_cnt++; |
1528 | return (unsigned long)cl; | |
1529 | } | |
1530 | ||
1531 | static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) | |
1532 | { | |
1533 | struct htb_sched *q = qdisc_priv(sch); | |
1534 | struct htb_class *cl = (struct htb_class *)arg; | |
3bf72957 | 1535 | |
87990467 SH |
1536 | if (cl) |
1537 | cl->filter_cnt--; | |
1538 | else | |
1da177e4 LT |
1539 | q->filter_cnt--; |
1540 | } | |
1541 | ||
1542 | static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
1543 | { | |
1544 | struct htb_sched *q = qdisc_priv(sch); | |
1545 | int i; | |
1546 | ||
1547 | if (arg->stop) | |
1548 | return; | |
1549 | ||
1550 | for (i = 0; i < HTB_HSIZE; i++) { | |
0cef296d SH |
1551 | struct hlist_node *p; |
1552 | struct htb_class *cl; | |
1553 | ||
1554 | hlist_for_each_entry(cl, p, q->hash + i, hlist) { | |
1da177e4 LT |
1555 | if (arg->count < arg->skip) { |
1556 | arg->count++; | |
1557 | continue; | |
1558 | } | |
1559 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
1560 | arg->stop = 1; | |
1561 | return; | |
1562 | } | |
1563 | arg->count++; | |
1564 | } | |
1565 | } | |
1566 | } | |
1567 | ||
1568 | static struct Qdisc_class_ops htb_class_ops = { | |
1569 | .graft = htb_graft, | |
1570 | .leaf = htb_leaf, | |
1571 | .get = htb_get, | |
1572 | .put = htb_put, | |
1573 | .change = htb_change_class, | |
1574 | .delete = htb_delete, | |
1575 | .walk = htb_walk, | |
1576 | .tcf_chain = htb_find_tcf, | |
1577 | .bind_tcf = htb_bind_filter, | |
1578 | .unbind_tcf = htb_unbind_filter, | |
1579 | .dump = htb_dump_class, | |
1580 | .dump_stats = htb_dump_class_stats, | |
1581 | }; | |
1582 | ||
1583 | static struct Qdisc_ops htb_qdisc_ops = { | |
1584 | .next = NULL, | |
1585 | .cl_ops = &htb_class_ops, | |
1586 | .id = "htb", | |
1587 | .priv_size = sizeof(struct htb_sched), | |
1588 | .enqueue = htb_enqueue, | |
1589 | .dequeue = htb_dequeue, | |
1590 | .requeue = htb_requeue, | |
1591 | .drop = htb_drop, | |
1592 | .init = htb_init, | |
1593 | .reset = htb_reset, | |
1594 | .destroy = htb_destroy, | |
1595 | .change = NULL /* htb_change */, | |
1596 | .dump = htb_dump, | |
1597 | .owner = THIS_MODULE, | |
1598 | }; | |
1599 | ||
1600 | static int __init htb_module_init(void) | |
1601 | { | |
87990467 | 1602 | return register_qdisc(&htb_qdisc_ops); |
1da177e4 | 1603 | } |
87990467 | 1604 | static void __exit htb_module_exit(void) |
1da177e4 | 1605 | { |
87990467 | 1606 | unregister_qdisc(&htb_qdisc_ops); |
1da177e4 | 1607 | } |
87990467 | 1608 | |
1da177e4 LT |
1609 | module_init(htb_module_init) |
1610 | module_exit(htb_module_exit) | |
1611 | MODULE_LICENSE("GPL"); |