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 | |
10297b99 | 14 | * Ondrej Kraus, <krauso@barr.cz> |
1da177e4 LT |
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. | |
1da177e4 | 27 | */ |
1da177e4 | 28 | #include <linux/module.h> |
47083fc0 | 29 | #include <linux/moduleparam.h> |
1da177e4 LT |
30 | #include <linux/types.h> |
31 | #include <linux/kernel.h> | |
1da177e4 | 32 | #include <linux/string.h> |
1da177e4 | 33 | #include <linux/errno.h> |
1da177e4 LT |
34 | #include <linux/skbuff.h> |
35 | #include <linux/list.h> | |
36 | #include <linux/compiler.h> | |
0ba48053 | 37 | #include <linux/rbtree.h> |
dc5fc579 | 38 | #include <net/netlink.h> |
1da177e4 | 39 | #include <net/pkt_sched.h> |
1da177e4 LT |
40 | |
41 | /* HTB algorithm. | |
42 | Author: devik@cdi.cz | |
43 | ======================================================================== | |
44 | HTB is like TBF with multiple classes. It is also similar to CBQ because | |
10297b99 | 45 | it allows to assign priority to each class in hierarchy. |
1da177e4 LT |
46 | In fact it is another implementation of Floyd's formal sharing. |
47 | ||
48 | Levels: | |
10297b99 | 49 | Each class is assigned level. Leaf has ALWAYS level 0 and root |
1da177e4 LT |
50 | classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level |
51 | one less than their parent. | |
52 | */ | |
53 | ||
47083fc0 | 54 | static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */ |
87990467 | 55 | #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */ |
1da177e4 LT |
56 | |
57 | #if HTB_VER >> 16 != TC_HTB_PROTOVER | |
58 | #error "Mismatched sch_htb.c and pkt_sch.h" | |
59 | #endif | |
60 | ||
47083fc0 JDB |
61 | /* Module parameter and sysfs export */ |
62 | module_param (htb_hysteresis, int, 0640); | |
63 | MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate"); | |
64 | ||
1da177e4 LT |
65 | /* used internaly to keep status of single class */ |
66 | enum htb_cmode { | |
87990467 SH |
67 | HTB_CANT_SEND, /* class can't send and can't borrow */ |
68 | HTB_MAY_BORROW, /* class can't send but may borrow */ | |
69 | HTB_CAN_SEND /* class can send */ | |
1da177e4 LT |
70 | }; |
71 | ||
72 | /* interior & leaf nodes; props specific to leaves are marked L: */ | |
87990467 | 73 | struct htb_class { |
f4c1f3e0 | 74 | struct Qdisc_class_common common; |
87990467 | 75 | /* general class parameters */ |
87990467 SH |
76 | struct gnet_stats_basic bstats; |
77 | struct gnet_stats_queue qstats; | |
78 | struct gnet_stats_rate_est rate_est; | |
79 | struct tc_htb_xstats xstats; /* our special stats */ | |
80 | int refcnt; /* usage count of this class */ | |
1da177e4 | 81 | |
87990467 SH |
82 | /* topology */ |
83 | int level; /* our level (see above) */ | |
42077599 | 84 | unsigned int children; |
87990467 | 85 | struct htb_class *parent; /* parent class */ |
87990467 SH |
86 | |
87 | union { | |
88 | struct htb_class_leaf { | |
89 | struct Qdisc *q; | |
90 | int prio; | |
91 | int aprio; | |
92 | int quantum; | |
93 | int deficit[TC_HTB_MAXDEPTH]; | |
94 | struct list_head drop_list; | |
95 | } leaf; | |
96 | struct htb_class_inner { | |
97 | struct rb_root feed[TC_HTB_NUMPRIO]; /* feed trees */ | |
98 | struct rb_node *ptr[TC_HTB_NUMPRIO]; /* current class ptr */ | |
99 | /* When class changes from state 1->2 and disconnects from | |
100 | parent's feed then we lost ptr value and start from the | |
101 | first child again. Here we store classid of the | |
102 | last valid ptr (used when ptr is NULL). */ | |
103 | u32 last_ptr_id[TC_HTB_NUMPRIO]; | |
104 | } inner; | |
105 | } un; | |
106 | struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ | |
107 | struct rb_node pq_node; /* node for event queue */ | |
fb983d45 | 108 | psched_time_t pq_key; |
87990467 SH |
109 | |
110 | int prio_activity; /* for which prios are we active */ | |
111 | enum htb_cmode cmode; /* current mode of the class */ | |
112 | ||
113 | /* class attached filters */ | |
114 | struct tcf_proto *filter_list; | |
115 | int filter_cnt; | |
116 | ||
117 | int warned; /* only one warning about non work conserving .. */ | |
118 | ||
119 | /* token bucket parameters */ | |
120 | struct qdisc_rate_table *rate; /* rate table of the class itself */ | |
121 | struct qdisc_rate_table *ceil; /* ceiling rate (limits borrows too) */ | |
122 | long buffer, cbuffer; /* token bucket depth/rate */ | |
123 | psched_tdiff_t mbuffer; /* max wait time */ | |
124 | long tokens, ctokens; /* current number of tokens */ | |
125 | psched_time_t t_c; /* checkpoint time */ | |
160d5e10 JP |
126 | |
127 | int prio; /* For parent to leaf return possible here */ | |
128 | int quantum; /* we do backup. Finally full replacement */ | |
129 | /* of un.leaf originals should be done. */ | |
1da177e4 LT |
130 | }; |
131 | ||
87990467 SH |
132 | static inline long L2T(struct htb_class *cl, struct qdisc_rate_table *rate, |
133 | int size) | |
134 | { | |
e9bef55d JDB |
135 | long result = qdisc_l2t(rate, size); |
136 | return result; | |
1da177e4 LT |
137 | } |
138 | ||
87990467 | 139 | struct htb_sched { |
f4c1f3e0 | 140 | struct Qdisc_class_hash clhash; |
0cef296d | 141 | struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */ |
87990467 SH |
142 | |
143 | /* self list - roots of self generating tree */ | |
144 | struct rb_root row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
145 | int row_mask[TC_HTB_MAXDEPTH]; | |
146 | struct rb_node *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
147 | u32 last_ptr_id[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
1da177e4 | 148 | |
87990467 SH |
149 | /* self wait list - roots of wait PQs per row */ |
150 | struct rb_root wait_pq[TC_HTB_MAXDEPTH]; | |
1da177e4 | 151 | |
87990467 | 152 | /* time of nearest event per level (row) */ |
fb983d45 | 153 | psched_time_t near_ev_cache[TC_HTB_MAXDEPTH]; |
1da177e4 | 154 | |
87990467 SH |
155 | /* whether we hit non-work conserving class during this dequeue; we use */ |
156 | int nwc_hit; /* this to disable mindelay complaint in dequeue */ | |
1da177e4 | 157 | |
87990467 | 158 | int defcls; /* class where unclassified flows go to */ |
1da177e4 | 159 | |
87990467 SH |
160 | /* filters for qdisc itself */ |
161 | struct tcf_proto *filter_list; | |
1da177e4 | 162 | |
87990467 SH |
163 | int rate2quantum; /* quant = rate / rate2quantum */ |
164 | psched_time_t now; /* cached dequeue time */ | |
fb983d45 | 165 | struct qdisc_watchdog watchdog; |
1da177e4 | 166 | |
87990467 SH |
167 | /* non shaped skbs; let them go directly thru */ |
168 | struct sk_buff_head direct_queue; | |
169 | int direct_qlen; /* max qlen of above */ | |
170 | ||
171 | long direct_pkts; | |
1da177e4 LT |
172 | }; |
173 | ||
1da177e4 | 174 | /* find class in global hash table using given handle */ |
87990467 | 175 | static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch) |
1da177e4 LT |
176 | { |
177 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 178 | struct Qdisc_class_common *clc; |
0cef296d | 179 | |
f4c1f3e0 PM |
180 | clc = qdisc_class_find(&q->clhash, handle); |
181 | if (clc == NULL) | |
1da177e4 | 182 | return NULL; |
f4c1f3e0 | 183 | return container_of(clc, struct htb_class, common); |
1da177e4 LT |
184 | } |
185 | ||
186 | /** | |
187 | * htb_classify - classify a packet into class | |
188 | * | |
189 | * It returns NULL if the packet should be dropped or -1 if the packet | |
190 | * should be passed directly thru. In all other cases leaf class is returned. | |
191 | * We allow direct class selection by classid in priority. The we examine | |
192 | * filters in qdisc and in inner nodes (if higher filter points to the inner | |
193 | * node). If we end up with classid MAJOR:0 we enqueue the skb into special | |
10297b99 | 194 | * internal fifo (direct). These packets then go directly thru. If we still |
1da177e4 LT |
195 | * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessfull |
196 | * then finish and return direct queue. | |
197 | */ | |
198 | #define HTB_DIRECT (struct htb_class*)-1 | |
1da177e4 | 199 | |
87990467 SH |
200 | static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, |
201 | int *qerr) | |
1da177e4 LT |
202 | { |
203 | struct htb_sched *q = qdisc_priv(sch); | |
204 | struct htb_class *cl; | |
205 | struct tcf_result res; | |
206 | struct tcf_proto *tcf; | |
207 | int result; | |
208 | ||
209 | /* allow to select class by setting skb->priority to valid classid; | |
210 | note that nfmark can be used too by attaching filter fw with no | |
211 | rules in it */ | |
212 | if (skb->priority == sch->handle) | |
87990467 SH |
213 | return HTB_DIRECT; /* X:0 (direct flow) selected */ |
214 | if ((cl = htb_find(skb->priority, sch)) != NULL && cl->level == 0) | |
1da177e4 LT |
215 | return cl; |
216 | ||
c27f339a | 217 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 LT |
218 | tcf = q->filter_list; |
219 | while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) { | |
220 | #ifdef CONFIG_NET_CLS_ACT | |
221 | switch (result) { | |
222 | case TC_ACT_QUEUED: | |
87990467 | 223 | case TC_ACT_STOLEN: |
378a2f09 | 224 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
1da177e4 LT |
225 | case TC_ACT_SHOT: |
226 | return NULL; | |
227 | } | |
1da177e4 | 228 | #endif |
87990467 | 229 | if ((cl = (void *)res.class) == NULL) { |
1da177e4 | 230 | if (res.classid == sch->handle) |
87990467 SH |
231 | return HTB_DIRECT; /* X:0 (direct flow) */ |
232 | if ((cl = htb_find(res.classid, sch)) == NULL) | |
233 | break; /* filter selected invalid classid */ | |
1da177e4 LT |
234 | } |
235 | if (!cl->level) | |
87990467 | 236 | return cl; /* we hit leaf; return it */ |
1da177e4 LT |
237 | |
238 | /* we have got inner class; apply inner filter chain */ | |
239 | tcf = cl->filter_list; | |
240 | } | |
241 | /* classification failed; try to use default class */ | |
87990467 | 242 | cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch); |
1da177e4 | 243 | if (!cl || cl->level) |
87990467 | 244 | return HTB_DIRECT; /* bad default .. this is safe bet */ |
1da177e4 LT |
245 | return cl; |
246 | } | |
247 | ||
1da177e4 LT |
248 | /** |
249 | * htb_add_to_id_tree - adds class to the round robin list | |
250 | * | |
251 | * Routine adds class to the list (actually tree) sorted by classid. | |
252 | * Make sure that class is not already on such list for given prio. | |
253 | */ | |
87990467 SH |
254 | static void htb_add_to_id_tree(struct rb_root *root, |
255 | struct htb_class *cl, int prio) | |
1da177e4 LT |
256 | { |
257 | struct rb_node **p = &root->rb_node, *parent = NULL; | |
3bf72957 | 258 | |
1da177e4 | 259 | while (*p) { |
87990467 SH |
260 | struct htb_class *c; |
261 | parent = *p; | |
1da177e4 | 262 | c = rb_entry(parent, struct htb_class, node[prio]); |
3bf72957 | 263 | |
f4c1f3e0 | 264 | if (cl->common.classid > c->common.classid) |
1da177e4 | 265 | p = &parent->rb_right; |
87990467 | 266 | else |
1da177e4 LT |
267 | p = &parent->rb_left; |
268 | } | |
269 | rb_link_node(&cl->node[prio], parent, p); | |
270 | rb_insert_color(&cl->node[prio], root); | |
271 | } | |
272 | ||
273 | /** | |
274 | * htb_add_to_wait_tree - adds class to the event queue with delay | |
275 | * | |
276 | * The class is added to priority event queue to indicate that class will | |
277 | * change its mode in cl->pq_key microseconds. Make sure that class is not | |
278 | * already in the queue. | |
279 | */ | |
87990467 SH |
280 | static void htb_add_to_wait_tree(struct htb_sched *q, |
281 | struct htb_class *cl, long delay) | |
1da177e4 LT |
282 | { |
283 | struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL; | |
3bf72957 | 284 | |
fb983d45 PM |
285 | cl->pq_key = q->now + delay; |
286 | if (cl->pq_key == q->now) | |
1da177e4 LT |
287 | cl->pq_key++; |
288 | ||
289 | /* update the nearest event cache */ | |
fb983d45 | 290 | if (q->near_ev_cache[cl->level] > cl->pq_key) |
1da177e4 | 291 | q->near_ev_cache[cl->level] = cl->pq_key; |
87990467 | 292 | |
1da177e4 | 293 | while (*p) { |
87990467 SH |
294 | struct htb_class *c; |
295 | parent = *p; | |
1da177e4 | 296 | c = rb_entry(parent, struct htb_class, pq_node); |
fb983d45 | 297 | if (cl->pq_key >= c->pq_key) |
1da177e4 | 298 | p = &parent->rb_right; |
87990467 | 299 | else |
1da177e4 LT |
300 | p = &parent->rb_left; |
301 | } | |
302 | rb_link_node(&cl->pq_node, parent, p); | |
303 | rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]); | |
304 | } | |
305 | ||
306 | /** | |
307 | * htb_next_rb_node - finds next node in binary tree | |
308 | * | |
309 | * When we are past last key we return NULL. | |
310 | * Average complexity is 2 steps per call. | |
311 | */ | |
3696f625 | 312 | static inline void htb_next_rb_node(struct rb_node **n) |
1da177e4 LT |
313 | { |
314 | *n = rb_next(*n); | |
315 | } | |
316 | ||
317 | /** | |
318 | * htb_add_class_to_row - add class to its row | |
319 | * | |
320 | * The class is added to row at priorities marked in mask. | |
321 | * It does nothing if mask == 0. | |
322 | */ | |
87990467 SH |
323 | static inline void htb_add_class_to_row(struct htb_sched *q, |
324 | struct htb_class *cl, int mask) | |
1da177e4 | 325 | { |
1da177e4 LT |
326 | q->row_mask[cl->level] |= mask; |
327 | while (mask) { | |
328 | int prio = ffz(~mask); | |
329 | mask &= ~(1 << prio); | |
87990467 | 330 | htb_add_to_id_tree(q->row[cl->level] + prio, cl, prio); |
1da177e4 LT |
331 | } |
332 | } | |
333 | ||
3696f625 SH |
334 | /* If this triggers, it is a bug in this code, but it need not be fatal */ |
335 | static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root) | |
336 | { | |
81771b3b | 337 | if (RB_EMPTY_NODE(rb)) { |
3696f625 SH |
338 | WARN_ON(1); |
339 | } else { | |
340 | rb_erase(rb, root); | |
341 | RB_CLEAR_NODE(rb); | |
342 | } | |
343 | } | |
344 | ||
345 | ||
1da177e4 LT |
346 | /** |
347 | * htb_remove_class_from_row - removes class from its row | |
348 | * | |
349 | * The class is removed from row at priorities marked in mask. | |
350 | * It does nothing if mask == 0. | |
351 | */ | |
87990467 SH |
352 | static inline void htb_remove_class_from_row(struct htb_sched *q, |
353 | struct htb_class *cl, int mask) | |
1da177e4 LT |
354 | { |
355 | int m = 0; | |
3bf72957 | 356 | |
1da177e4 LT |
357 | while (mask) { |
358 | int prio = ffz(~mask); | |
3696f625 | 359 | |
1da177e4 | 360 | mask &= ~(1 << prio); |
87990467 SH |
361 | if (q->ptr[cl->level][prio] == cl->node + prio) |
362 | htb_next_rb_node(q->ptr[cl->level] + prio); | |
3696f625 SH |
363 | |
364 | htb_safe_rb_erase(cl->node + prio, q->row[cl->level] + prio); | |
87990467 | 365 | if (!q->row[cl->level][prio].rb_node) |
1da177e4 LT |
366 | m |= 1 << prio; |
367 | } | |
1da177e4 LT |
368 | q->row_mask[cl->level] &= ~m; |
369 | } | |
370 | ||
371 | /** | |
372 | * htb_activate_prios - creates active classe's feed chain | |
373 | * | |
374 | * The class is connected to ancestors and/or appropriate rows | |
10297b99 | 375 | * for priorities it is participating on. cl->cmode must be new |
1da177e4 LT |
376 | * (activated) mode. It does nothing if cl->prio_activity == 0. |
377 | */ | |
87990467 | 378 | static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
379 | { |
380 | struct htb_class *p = cl->parent; | |
87990467 | 381 | long m, mask = cl->prio_activity; |
1da177e4 LT |
382 | |
383 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
384 | m = mask; |
385 | while (m) { | |
1da177e4 LT |
386 | int prio = ffz(~m); |
387 | m &= ~(1 << prio); | |
87990467 | 388 | |
1da177e4 LT |
389 | if (p->un.inner.feed[prio].rb_node) |
390 | /* parent already has its feed in use so that | |
391 | reset bit in mask as parent is already ok */ | |
392 | mask &= ~(1 << prio); | |
87990467 SH |
393 | |
394 | htb_add_to_id_tree(p->un.inner.feed + prio, cl, prio); | |
1da177e4 | 395 | } |
1da177e4 | 396 | p->prio_activity |= mask; |
87990467 SH |
397 | cl = p; |
398 | p = cl->parent; | |
3bf72957 | 399 | |
1da177e4 LT |
400 | } |
401 | if (cl->cmode == HTB_CAN_SEND && mask) | |
87990467 | 402 | htb_add_class_to_row(q, cl, mask); |
1da177e4 LT |
403 | } |
404 | ||
405 | /** | |
406 | * htb_deactivate_prios - remove class from feed chain | |
407 | * | |
10297b99 | 408 | * cl->cmode must represent old mode (before deactivation). It does |
1da177e4 LT |
409 | * nothing if cl->prio_activity == 0. Class is removed from all feed |
410 | * chains and rows. | |
411 | */ | |
412 | static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) | |
413 | { | |
414 | struct htb_class *p = cl->parent; | |
87990467 | 415 | long m, mask = cl->prio_activity; |
1da177e4 LT |
416 | |
417 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
418 | m = mask; |
419 | mask = 0; | |
1da177e4 LT |
420 | while (m) { |
421 | int prio = ffz(~m); | |
422 | m &= ~(1 << prio); | |
87990467 SH |
423 | |
424 | if (p->un.inner.ptr[prio] == cl->node + prio) { | |
1da177e4 LT |
425 | /* we are removing child which is pointed to from |
426 | parent feed - forget the pointer but remember | |
427 | classid */ | |
f4c1f3e0 | 428 | p->un.inner.last_ptr_id[prio] = cl->common.classid; |
1da177e4 LT |
429 | p->un.inner.ptr[prio] = NULL; |
430 | } | |
87990467 | 431 | |
3696f625 | 432 | htb_safe_rb_erase(cl->node + prio, p->un.inner.feed + prio); |
87990467 SH |
433 | |
434 | if (!p->un.inner.feed[prio].rb_node) | |
1da177e4 LT |
435 | mask |= 1 << prio; |
436 | } | |
3bf72957 | 437 | |
1da177e4 | 438 | p->prio_activity &= ~mask; |
87990467 SH |
439 | cl = p; |
440 | p = cl->parent; | |
3bf72957 | 441 | |
1da177e4 | 442 | } |
87990467 SH |
443 | if (cl->cmode == HTB_CAN_SEND && mask) |
444 | htb_remove_class_from_row(q, cl, mask); | |
1da177e4 LT |
445 | } |
446 | ||
18a63e86 SH |
447 | static inline long htb_lowater(const struct htb_class *cl) |
448 | { | |
47083fc0 JDB |
449 | if (htb_hysteresis) |
450 | return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0; | |
451 | else | |
452 | return 0; | |
18a63e86 SH |
453 | } |
454 | static inline long htb_hiwater(const struct htb_class *cl) | |
455 | { | |
47083fc0 JDB |
456 | if (htb_hysteresis) |
457 | return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0; | |
458 | else | |
459 | return 0; | |
18a63e86 | 460 | } |
47083fc0 | 461 | |
18a63e86 | 462 | |
1da177e4 LT |
463 | /** |
464 | * htb_class_mode - computes and returns current class mode | |
465 | * | |
466 | * It computes cl's mode at time cl->t_c+diff and returns it. If mode | |
467 | * is not HTB_CAN_SEND then cl->pq_key is updated to time difference | |
10297b99 | 468 | * from now to time when cl will change its state. |
1da177e4 | 469 | * Also it is worth to note that class mode doesn't change simply |
10297b99 | 470 | * at cl->{c,}tokens == 0 but there can rather be hysteresis of |
1da177e4 LT |
471 | * 0 .. -cl->{c,}buffer range. It is meant to limit number of |
472 | * mode transitions per time unit. The speed gain is about 1/6. | |
473 | */ | |
87990467 SH |
474 | static inline enum htb_cmode |
475 | htb_class_mode(struct htb_class *cl, long *diff) | |
1da177e4 | 476 | { |
87990467 | 477 | long toks; |
1da177e4 | 478 | |
87990467 SH |
479 | if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) { |
480 | *diff = -toks; | |
481 | return HTB_CANT_SEND; | |
482 | } | |
18a63e86 | 483 | |
87990467 SH |
484 | if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl)) |
485 | return HTB_CAN_SEND; | |
1da177e4 | 486 | |
87990467 SH |
487 | *diff = -toks; |
488 | return HTB_MAY_BORROW; | |
1da177e4 LT |
489 | } |
490 | ||
491 | /** | |
492 | * htb_change_class_mode - changes classe's mode | |
493 | * | |
494 | * This should be the only way how to change classe's mode under normal | |
495 | * cirsumstances. Routine will update feed lists linkage, change mode | |
496 | * and add class to the wait event queue if appropriate. New mode should | |
497 | * be different from old one and cl->pq_key has to be valid if changing | |
498 | * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). | |
499 | */ | |
87990467 | 500 | static void |
1da177e4 | 501 | htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff) |
87990467 SH |
502 | { |
503 | enum htb_cmode new_mode = htb_class_mode(cl, diff); | |
1da177e4 LT |
504 | |
505 | if (new_mode == cl->cmode) | |
87990467 SH |
506 | return; |
507 | ||
508 | if (cl->prio_activity) { /* not necessary: speed optimization */ | |
509 | if (cl->cmode != HTB_CANT_SEND) | |
510 | htb_deactivate_prios(q, cl); | |
1da177e4 | 511 | cl->cmode = new_mode; |
87990467 SH |
512 | if (new_mode != HTB_CANT_SEND) |
513 | htb_activate_prios(q, cl); | |
514 | } else | |
1da177e4 LT |
515 | cl->cmode = new_mode; |
516 | } | |
517 | ||
518 | /** | |
10297b99 | 519 | * htb_activate - inserts leaf cl into appropriate active feeds |
1da177e4 LT |
520 | * |
521 | * Routine learns (new) priority of leaf and activates feed chain | |
522 | * for the prio. It can be called on already active leaf safely. | |
523 | * It also adds leaf into droplist. | |
524 | */ | |
87990467 | 525 | static inline void htb_activate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 526 | { |
547b792c | 527 | WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen); |
3bf72957 | 528 | |
1da177e4 LT |
529 | if (!cl->prio_activity) { |
530 | cl->prio_activity = 1 << (cl->un.leaf.aprio = cl->un.leaf.prio); | |
87990467 SH |
531 | htb_activate_prios(q, cl); |
532 | list_add_tail(&cl->un.leaf.drop_list, | |
533 | q->drops + cl->un.leaf.aprio); | |
1da177e4 LT |
534 | } |
535 | } | |
536 | ||
537 | /** | |
10297b99 | 538 | * htb_deactivate - remove leaf cl from active feeds |
1da177e4 LT |
539 | * |
540 | * Make sure that leaf is active. In the other words it can't be called | |
541 | * with non-active leaf. It also removes class from the drop list. | |
542 | */ | |
87990467 | 543 | static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 544 | { |
547b792c | 545 | WARN_ON(!cl->prio_activity); |
3bf72957 | 546 | |
87990467 | 547 | htb_deactivate_prios(q, cl); |
1da177e4 LT |
548 | cl->prio_activity = 0; |
549 | list_del_init(&cl->un.leaf.drop_list); | |
550 | } | |
551 | ||
552 | static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch) | |
553 | { | |
f30ab418 | 554 | int uninitialized_var(ret); |
87990467 SH |
555 | struct htb_sched *q = qdisc_priv(sch); |
556 | struct htb_class *cl = htb_classify(skb, sch, &ret); | |
557 | ||
558 | if (cl == HTB_DIRECT) { | |
559 | /* enqueue to helper queue */ | |
560 | if (q->direct_queue.qlen < q->direct_qlen) { | |
561 | __skb_queue_tail(&q->direct_queue, skb); | |
562 | q->direct_pkts++; | |
563 | } else { | |
564 | kfree_skb(skb); | |
565 | sch->qstats.drops++; | |
566 | return NET_XMIT_DROP; | |
567 | } | |
1da177e4 | 568 | #ifdef CONFIG_NET_CLS_ACT |
87990467 | 569 | } else if (!cl) { |
c27f339a | 570 | if (ret & __NET_XMIT_BYPASS) |
87990467 SH |
571 | sch->qstats.drops++; |
572 | kfree_skb(skb); | |
573 | return ret; | |
1da177e4 | 574 | #endif |
378a2f09 JP |
575 | } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q)) != NET_XMIT_SUCCESS) { |
576 | if (net_xmit_drop_count(ret)) { | |
577 | sch->qstats.drops++; | |
578 | cl->qstats.drops++; | |
579 | } | |
69747650 | 580 | return ret; |
87990467 | 581 | } else { |
c9726d68 RM |
582 | cl->bstats.packets += |
583 | skb_is_gso(skb)?skb_shinfo(skb)->gso_segs:1; | |
0abf77e5 | 584 | cl->bstats.bytes += qdisc_pkt_len(skb); |
87990467 SH |
585 | htb_activate(q, cl); |
586 | } | |
587 | ||
588 | sch->q.qlen++; | |
c9726d68 | 589 | sch->bstats.packets += skb_is_gso(skb)?skb_shinfo(skb)->gso_segs:1; |
0abf77e5 | 590 | sch->bstats.bytes += qdisc_pkt_len(skb); |
87990467 | 591 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
592 | } |
593 | ||
1da177e4 LT |
594 | /** |
595 | * htb_charge_class - charges amount "bytes" to leaf and ancestors | |
596 | * | |
597 | * Routine assumes that packet "bytes" long was dequeued from leaf cl | |
598 | * borrowing from "level". It accounts bytes to ceil leaky bucket for | |
599 | * leaf and all ancestors and to rate bucket for ancestors at levels | |
600 | * "level" and higher. It also handles possible change of mode resulting | |
601 | * from the update. Note that mode can also increase here (MAY_BORROW to | |
602 | * CAN_SEND) because we can use more precise clock that event queue here. | |
603 | * In such case we remove class from event queue first. | |
604 | */ | |
87990467 | 605 | static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, |
c9726d68 | 606 | int level, struct sk_buff *skb) |
87990467 | 607 | { |
0abf77e5 | 608 | int bytes = qdisc_pkt_len(skb); |
87990467 | 609 | long toks, diff; |
1da177e4 | 610 | enum htb_cmode old_mode; |
1da177e4 LT |
611 | |
612 | #define HTB_ACCNT(T,B,R) toks = diff + cl->T; \ | |
613 | if (toks > cl->B) toks = cl->B; \ | |
614 | toks -= L2T(cl, cl->R, bytes); \ | |
615 | if (toks <= -cl->mbuffer) toks = 1-cl->mbuffer; \ | |
616 | cl->T = toks | |
617 | ||
618 | while (cl) { | |
03cc45c0 | 619 | diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer); |
1da177e4 | 620 | if (cl->level >= level) { |
87990467 SH |
621 | if (cl->level == level) |
622 | cl->xstats.lends++; | |
623 | HTB_ACCNT(tokens, buffer, rate); | |
1da177e4 LT |
624 | } else { |
625 | cl->xstats.borrows++; | |
87990467 | 626 | cl->tokens += diff; /* we moved t_c; update tokens */ |
1da177e4 | 627 | } |
87990467 | 628 | HTB_ACCNT(ctokens, cbuffer, ceil); |
1da177e4 | 629 | cl->t_c = q->now; |
1da177e4 | 630 | |
87990467 SH |
631 | old_mode = cl->cmode; |
632 | diff = 0; | |
633 | htb_change_class_mode(q, cl, &diff); | |
1da177e4 LT |
634 | if (old_mode != cl->cmode) { |
635 | if (old_mode != HTB_CAN_SEND) | |
3696f625 | 636 | htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level); |
1da177e4 | 637 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 638 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 639 | } |
1da177e4 LT |
640 | |
641 | /* update byte stats except for leaves which are already updated */ | |
642 | if (cl->level) { | |
643 | cl->bstats.bytes += bytes; | |
c9726d68 RM |
644 | cl->bstats.packets += skb_is_gso(skb)? |
645 | skb_shinfo(skb)->gso_segs:1; | |
1da177e4 LT |
646 | } |
647 | cl = cl->parent; | |
648 | } | |
649 | } | |
650 | ||
651 | /** | |
652 | * htb_do_events - make mode changes to classes at the level | |
653 | * | |
fb983d45 | 654 | * Scans event queue for pending events and applies them. Returns time of |
1da177e4 | 655 | * next pending event (0 for no event in pq). |
fb983d45 | 656 | * Note: Applied are events whose have cl->pq_key <= q->now. |
1da177e4 | 657 | */ |
fb983d45 | 658 | static psched_time_t htb_do_events(struct htb_sched *q, int level) |
1da177e4 | 659 | { |
8f3ea33a MD |
660 | /* don't run for longer than 2 jiffies; 2 is used instead of |
661 | 1 to simplify things when jiffy is going to be incremented | |
662 | too soon */ | |
663 | unsigned long stop_at = jiffies + 2; | |
664 | while (time_before(jiffies, stop_at)) { | |
1da177e4 LT |
665 | struct htb_class *cl; |
666 | long diff; | |
30bdbe39 AM |
667 | struct rb_node *p = rb_first(&q->wait_pq[level]); |
668 | ||
87990467 SH |
669 | if (!p) |
670 | return 0; | |
1da177e4 LT |
671 | |
672 | cl = rb_entry(p, struct htb_class, pq_node); | |
fb983d45 PM |
673 | if (cl->pq_key > q->now) |
674 | return cl->pq_key; | |
675 | ||
3696f625 | 676 | htb_safe_rb_erase(p, q->wait_pq + level); |
03cc45c0 | 677 | diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer); |
87990467 | 678 | htb_change_class_mode(q, cl, &diff); |
1da177e4 | 679 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 680 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 681 | } |
8f3ea33a MD |
682 | /* too much load - let's continue on next jiffie */ |
683 | return q->now + PSCHED_TICKS_PER_SEC / HZ; | |
1da177e4 LT |
684 | } |
685 | ||
686 | /* Returns class->node+prio from id-tree where classe's id is >= id. NULL | |
687 | is no such one exists. */ | |
87990467 SH |
688 | static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n, |
689 | u32 id) | |
1da177e4 LT |
690 | { |
691 | struct rb_node *r = NULL; | |
692 | while (n) { | |
87990467 SH |
693 | struct htb_class *cl = |
694 | rb_entry(n, struct htb_class, node[prio]); | |
f4c1f3e0 | 695 | if (id == cl->common.classid) |
87990467 SH |
696 | return n; |
697 | ||
f4c1f3e0 | 698 | if (id > cl->common.classid) { |
1da177e4 LT |
699 | n = n->rb_right; |
700 | } else { | |
701 | r = n; | |
702 | n = n->rb_left; | |
703 | } | |
704 | } | |
705 | return r; | |
706 | } | |
707 | ||
708 | /** | |
709 | * htb_lookup_leaf - returns next leaf class in DRR order | |
710 | * | |
711 | * Find leaf where current feed pointers points to. | |
712 | */ | |
87990467 SH |
713 | static struct htb_class *htb_lookup_leaf(struct rb_root *tree, int prio, |
714 | struct rb_node **pptr, u32 * pid) | |
1da177e4 LT |
715 | { |
716 | int i; | |
717 | struct { | |
718 | struct rb_node *root; | |
719 | struct rb_node **pptr; | |
720 | u32 *pid; | |
87990467 SH |
721 | } stk[TC_HTB_MAXDEPTH], *sp = stk; |
722 | ||
547b792c | 723 | WARN_ON(!tree->rb_node); |
1da177e4 LT |
724 | sp->root = tree->rb_node; |
725 | sp->pptr = pptr; | |
726 | sp->pid = pid; | |
727 | ||
728 | for (i = 0; i < 65535; i++) { | |
87990467 | 729 | if (!*sp->pptr && *sp->pid) { |
10297b99 | 730 | /* ptr was invalidated but id is valid - try to recover |
1da177e4 | 731 | the original or next ptr */ |
87990467 SH |
732 | *sp->pptr = |
733 | htb_id_find_next_upper(prio, sp->root, *sp->pid); | |
1da177e4 | 734 | } |
87990467 SH |
735 | *sp->pid = 0; /* ptr is valid now so that remove this hint as it |
736 | can become out of date quickly */ | |
737 | if (!*sp->pptr) { /* we are at right end; rewind & go up */ | |
1da177e4 | 738 | *sp->pptr = sp->root; |
87990467 | 739 | while ((*sp->pptr)->rb_left) |
1da177e4 LT |
740 | *sp->pptr = (*sp->pptr)->rb_left; |
741 | if (sp > stk) { | |
742 | sp--; | |
547b792c | 743 | WARN_ON(!*sp->pptr); |
87990467 SH |
744 | if (!*sp->pptr) |
745 | return NULL; | |
746 | htb_next_rb_node(sp->pptr); | |
1da177e4 LT |
747 | } |
748 | } else { | |
749 | struct htb_class *cl; | |
87990467 SH |
750 | cl = rb_entry(*sp->pptr, struct htb_class, node[prio]); |
751 | if (!cl->level) | |
1da177e4 LT |
752 | return cl; |
753 | (++sp)->root = cl->un.inner.feed[prio].rb_node; | |
87990467 SH |
754 | sp->pptr = cl->un.inner.ptr + prio; |
755 | sp->pid = cl->un.inner.last_ptr_id + prio; | |
1da177e4 LT |
756 | } |
757 | } | |
547b792c | 758 | WARN_ON(1); |
1da177e4 LT |
759 | return NULL; |
760 | } | |
761 | ||
762 | /* dequeues packet at given priority and level; call only if | |
763 | you are sure that there is active class at prio/level */ | |
87990467 SH |
764 | static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, int prio, |
765 | int level) | |
1da177e4 LT |
766 | { |
767 | struct sk_buff *skb = NULL; | |
87990467 | 768 | struct htb_class *cl, *start; |
1da177e4 | 769 | /* look initial class up in the row */ |
87990467 SH |
770 | start = cl = htb_lookup_leaf(q->row[level] + prio, prio, |
771 | q->ptr[level] + prio, | |
772 | q->last_ptr_id[level] + prio); | |
773 | ||
1da177e4 LT |
774 | do { |
775 | next: | |
547b792c | 776 | WARN_ON(!cl); |
87990467 SH |
777 | if (!cl) |
778 | return NULL; | |
1da177e4 LT |
779 | |
780 | /* class can be empty - it is unlikely but can be true if leaf | |
781 | qdisc drops packets in enqueue routine or if someone used | |
10297b99 | 782 | graft operation on the leaf since last dequeue; |
1da177e4 LT |
783 | simply deactivate and skip such class */ |
784 | if (unlikely(cl->un.leaf.q->q.qlen == 0)) { | |
785 | struct htb_class *next; | |
87990467 | 786 | htb_deactivate(q, cl); |
1da177e4 LT |
787 | |
788 | /* row/level might become empty */ | |
789 | if ((q->row_mask[level] & (1 << prio)) == 0) | |
87990467 | 790 | return NULL; |
1da177e4 | 791 | |
87990467 SH |
792 | next = htb_lookup_leaf(q->row[level] + prio, |
793 | prio, q->ptr[level] + prio, | |
794 | q->last_ptr_id[level] + prio); | |
795 | ||
796 | if (cl == start) /* fix start if we just deleted it */ | |
1da177e4 LT |
797 | start = next; |
798 | cl = next; | |
799 | goto next; | |
800 | } | |
87990467 SH |
801 | |
802 | skb = cl->un.leaf.q->dequeue(cl->un.leaf.q); | |
803 | if (likely(skb != NULL)) | |
1da177e4 LT |
804 | break; |
805 | if (!cl->warned) { | |
87990467 SH |
806 | printk(KERN_WARNING |
807 | "htb: class %X isn't work conserving ?!\n", | |
f4c1f3e0 | 808 | cl->common.classid); |
1da177e4 LT |
809 | cl->warned = 1; |
810 | } | |
811 | q->nwc_hit++; | |
87990467 SH |
812 | htb_next_rb_node((level ? cl->parent->un.inner.ptr : q-> |
813 | ptr[0]) + prio); | |
814 | cl = htb_lookup_leaf(q->row[level] + prio, prio, | |
815 | q->ptr[level] + prio, | |
816 | q->last_ptr_id[level] + prio); | |
1da177e4 LT |
817 | |
818 | } while (cl != start); | |
819 | ||
820 | if (likely(skb != NULL)) { | |
0abf77e5 JK |
821 | cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb); |
822 | if (cl->un.leaf.deficit[level] < 0) { | |
1da177e4 | 823 | cl->un.leaf.deficit[level] += cl->un.leaf.quantum; |
87990467 SH |
824 | htb_next_rb_node((level ? cl->parent->un.inner.ptr : q-> |
825 | ptr[0]) + prio); | |
1da177e4 LT |
826 | } |
827 | /* this used to be after charge_class but this constelation | |
828 | gives us slightly better performance */ | |
829 | if (!cl->un.leaf.q->q.qlen) | |
87990467 | 830 | htb_deactivate(q, cl); |
c9726d68 | 831 | htb_charge_class(q, cl, level, skb); |
1da177e4 LT |
832 | } |
833 | return skb; | |
834 | } | |
835 | ||
1da177e4 LT |
836 | static struct sk_buff *htb_dequeue(struct Qdisc *sch) |
837 | { | |
838 | struct sk_buff *skb = NULL; | |
839 | struct htb_sched *q = qdisc_priv(sch); | |
840 | int level; | |
fb983d45 | 841 | psched_time_t next_event; |
1da177e4 LT |
842 | |
843 | /* try to dequeue direct packets as high prio (!) to minimize cpu work */ | |
87990467 SH |
844 | skb = __skb_dequeue(&q->direct_queue); |
845 | if (skb != NULL) { | |
1da177e4 LT |
846 | sch->flags &= ~TCQ_F_THROTTLED; |
847 | sch->q.qlen--; | |
848 | return skb; | |
849 | } | |
850 | ||
87990467 SH |
851 | if (!sch->q.qlen) |
852 | goto fin; | |
3bebcda2 | 853 | q->now = psched_get_time(); |
1da177e4 | 854 | |
fb983d45 | 855 | next_event = q->now + 5 * PSCHED_TICKS_PER_SEC; |
1da177e4 LT |
856 | q->nwc_hit = 0; |
857 | for (level = 0; level < TC_HTB_MAXDEPTH; level++) { | |
858 | /* common case optimization - skip event handler quickly */ | |
859 | int m; | |
fb983d45 PM |
860 | psched_time_t event; |
861 | ||
862 | if (q->now >= q->near_ev_cache[level]) { | |
863 | event = htb_do_events(q, level); | |
2e4b3b0e PM |
864 | if (!event) |
865 | event = q->now + PSCHED_TICKS_PER_SEC; | |
866 | q->near_ev_cache[level] = event; | |
1da177e4 | 867 | } else |
fb983d45 PM |
868 | event = q->near_ev_cache[level]; |
869 | ||
870 | if (event && next_event > event) | |
871 | next_event = event; | |
87990467 | 872 | |
1da177e4 LT |
873 | m = ~q->row_mask[level]; |
874 | while (m != (int)(-1)) { | |
87990467 | 875 | int prio = ffz(m); |
1da177e4 | 876 | m |= 1 << prio; |
87990467 | 877 | skb = htb_dequeue_tree(q, prio, level); |
1da177e4 LT |
878 | if (likely(skb != NULL)) { |
879 | sch->q.qlen--; | |
880 | sch->flags &= ~TCQ_F_THROTTLED; | |
881 | goto fin; | |
882 | } | |
883 | } | |
884 | } | |
fb983d45 PM |
885 | sch->qstats.overlimits++; |
886 | qdisc_watchdog_schedule(&q->watchdog, next_event); | |
1da177e4 | 887 | fin: |
1da177e4 LT |
888 | return skb; |
889 | } | |
890 | ||
891 | /* try to drop from each class (by prio) until one succeed */ | |
87990467 | 892 | static unsigned int htb_drop(struct Qdisc *sch) |
1da177e4 LT |
893 | { |
894 | struct htb_sched *q = qdisc_priv(sch); | |
895 | int prio; | |
896 | ||
897 | for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) { | |
898 | struct list_head *p; | |
87990467 | 899 | list_for_each(p, q->drops + prio) { |
1da177e4 LT |
900 | struct htb_class *cl = list_entry(p, struct htb_class, |
901 | un.leaf.drop_list); | |
902 | unsigned int len; | |
87990467 SH |
903 | if (cl->un.leaf.q->ops->drop && |
904 | (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) { | |
1da177e4 LT |
905 | sch->q.qlen--; |
906 | if (!cl->un.leaf.q->q.qlen) | |
87990467 | 907 | htb_deactivate(q, cl); |
1da177e4 LT |
908 | return len; |
909 | } | |
910 | } | |
911 | } | |
912 | return 0; | |
913 | } | |
914 | ||
915 | /* reset all classes */ | |
916 | /* always caled under BH & queue lock */ | |
87990467 | 917 | static void htb_reset(struct Qdisc *sch) |
1da177e4 LT |
918 | { |
919 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 PM |
920 | struct htb_class *cl; |
921 | struct hlist_node *n; | |
922 | unsigned int i; | |
0cef296d | 923 | |
f4c1f3e0 PM |
924 | for (i = 0; i < q->clhash.hashsize; i++) { |
925 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { | |
1da177e4 | 926 | if (cl->level) |
87990467 | 927 | memset(&cl->un.inner, 0, sizeof(cl->un.inner)); |
1da177e4 | 928 | else { |
87990467 | 929 | if (cl->un.leaf.q) |
1da177e4 LT |
930 | qdisc_reset(cl->un.leaf.q); |
931 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); | |
932 | } | |
933 | cl->prio_activity = 0; | |
934 | cl->cmode = HTB_CAN_SEND; | |
1da177e4 LT |
935 | |
936 | } | |
937 | } | |
fb983d45 | 938 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
939 | __skb_queue_purge(&q->direct_queue); |
940 | sch->q.qlen = 0; | |
87990467 SH |
941 | memset(q->row, 0, sizeof(q->row)); |
942 | memset(q->row_mask, 0, sizeof(q->row_mask)); | |
943 | memset(q->wait_pq, 0, sizeof(q->wait_pq)); | |
944 | memset(q->ptr, 0, sizeof(q->ptr)); | |
1da177e4 | 945 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 946 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 LT |
947 | } |
948 | ||
27a3421e PM |
949 | static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = { |
950 | [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) }, | |
951 | [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) }, | |
952 | [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
953 | [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
954 | }; | |
955 | ||
1e90474c | 956 | static int htb_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
957 | { |
958 | struct htb_sched *q = qdisc_priv(sch); | |
1e90474c | 959 | struct nlattr *tb[TCA_HTB_INIT + 1]; |
1da177e4 | 960 | struct tc_htb_glob *gopt; |
cee63723 | 961 | int err; |
1da177e4 | 962 | int i; |
cee63723 PM |
963 | |
964 | if (!opt) | |
965 | return -EINVAL; | |
966 | ||
27a3421e | 967 | err = nla_parse_nested(tb, TCA_HTB_INIT, opt, htb_policy); |
cee63723 PM |
968 | if (err < 0) |
969 | return err; | |
970 | ||
27a3421e | 971 | if (tb[TCA_HTB_INIT] == NULL) { |
1da177e4 LT |
972 | printk(KERN_ERR "HTB: hey probably you have bad tc tool ?\n"); |
973 | return -EINVAL; | |
974 | } | |
1e90474c | 975 | gopt = nla_data(tb[TCA_HTB_INIT]); |
1da177e4 | 976 | if (gopt->version != HTB_VER >> 16) { |
87990467 SH |
977 | printk(KERN_ERR |
978 | "HTB: need tc/htb version %d (minor is %d), you have %d\n", | |
979 | HTB_VER >> 16, HTB_VER & 0xffff, gopt->version); | |
1da177e4 LT |
980 | return -EINVAL; |
981 | } | |
1da177e4 | 982 | |
f4c1f3e0 PM |
983 | err = qdisc_class_hash_init(&q->clhash); |
984 | if (err < 0) | |
985 | return err; | |
1da177e4 | 986 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 987 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 | 988 | |
fb983d45 | 989 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 LT |
990 | skb_queue_head_init(&q->direct_queue); |
991 | ||
5ce2d488 | 992 | q->direct_qlen = qdisc_dev(sch)->tx_queue_len; |
87990467 | 993 | if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */ |
1da177e4 | 994 | q->direct_qlen = 2; |
1da177e4 | 995 | |
1da177e4 LT |
996 | if ((q->rate2quantum = gopt->rate2quantum) < 1) |
997 | q->rate2quantum = 1; | |
998 | q->defcls = gopt->defcls; | |
999 | ||
1000 | return 0; | |
1001 | } | |
1002 | ||
1003 | static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) | |
1004 | { | |
102396ae | 1005 | spinlock_t *root_lock = qdisc_root_sleeping_lock(sch); |
1da177e4 | 1006 | struct htb_sched *q = qdisc_priv(sch); |
4b3550ef | 1007 | struct nlattr *nest; |
1da177e4 | 1008 | struct tc_htb_glob gopt; |
4b3550ef | 1009 | |
7698b4fc | 1010 | spin_lock_bh(root_lock); |
1da177e4 | 1011 | |
4b3550ef | 1012 | gopt.direct_pkts = q->direct_pkts; |
1da177e4 LT |
1013 | gopt.version = HTB_VER; |
1014 | gopt.rate2quantum = q->rate2quantum; | |
1015 | gopt.defcls = q->defcls; | |
3bf72957 | 1016 | gopt.debug = 0; |
4b3550ef PM |
1017 | |
1018 | nest = nla_nest_start(skb, TCA_OPTIONS); | |
1019 | if (nest == NULL) | |
1020 | goto nla_put_failure; | |
1e90474c | 1021 | NLA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt); |
4b3550ef PM |
1022 | nla_nest_end(skb, nest); |
1023 | ||
7698b4fc | 1024 | spin_unlock_bh(root_lock); |
1da177e4 | 1025 | return skb->len; |
4b3550ef | 1026 | |
1e90474c | 1027 | nla_put_failure: |
7698b4fc | 1028 | spin_unlock_bh(root_lock); |
4b3550ef | 1029 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1030 | return -1; |
1031 | } | |
1032 | ||
1033 | static int htb_dump_class(struct Qdisc *sch, unsigned long arg, | |
87990467 | 1034 | struct sk_buff *skb, struct tcmsg *tcm) |
1da177e4 | 1035 | { |
87990467 | 1036 | struct htb_class *cl = (struct htb_class *)arg; |
102396ae | 1037 | spinlock_t *root_lock = qdisc_root_sleeping_lock(sch); |
4b3550ef | 1038 | struct nlattr *nest; |
1da177e4 LT |
1039 | struct tc_htb_opt opt; |
1040 | ||
7698b4fc | 1041 | spin_lock_bh(root_lock); |
f4c1f3e0 PM |
1042 | tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT; |
1043 | tcm->tcm_handle = cl->common.classid; | |
1da177e4 LT |
1044 | if (!cl->level && cl->un.leaf.q) |
1045 | tcm->tcm_info = cl->un.leaf.q->handle; | |
1046 | ||
4b3550ef PM |
1047 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1048 | if (nest == NULL) | |
1049 | goto nla_put_failure; | |
1da177e4 | 1050 | |
87990467 | 1051 | memset(&opt, 0, sizeof(opt)); |
1da177e4 | 1052 | |
87990467 SH |
1053 | opt.rate = cl->rate->rate; |
1054 | opt.buffer = cl->buffer; | |
1055 | opt.ceil = cl->ceil->rate; | |
1056 | opt.cbuffer = cl->cbuffer; | |
1057 | opt.quantum = cl->un.leaf.quantum; | |
1058 | opt.prio = cl->un.leaf.prio; | |
1059 | opt.level = cl->level; | |
1e90474c | 1060 | NLA_PUT(skb, TCA_HTB_PARMS, sizeof(opt), &opt); |
4b3550ef PM |
1061 | |
1062 | nla_nest_end(skb, nest); | |
7698b4fc | 1063 | spin_unlock_bh(root_lock); |
1da177e4 | 1064 | return skb->len; |
4b3550ef | 1065 | |
1e90474c | 1066 | nla_put_failure: |
7698b4fc | 1067 | spin_unlock_bh(root_lock); |
4b3550ef | 1068 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1069 | return -1; |
1070 | } | |
1071 | ||
1072 | static int | |
87990467 | 1073 | htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d) |
1da177e4 | 1074 | { |
87990467 | 1075 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 | 1076 | |
1da177e4 LT |
1077 | if (!cl->level && cl->un.leaf.q) |
1078 | cl->qstats.qlen = cl->un.leaf.q->q.qlen; | |
1079 | cl->xstats.tokens = cl->tokens; | |
1080 | cl->xstats.ctokens = cl->ctokens; | |
1081 | ||
1082 | if (gnet_stats_copy_basic(d, &cl->bstats) < 0 || | |
1083 | gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || | |
1084 | gnet_stats_copy_queue(d, &cl->qstats) < 0) | |
1085 | return -1; | |
1086 | ||
1087 | return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); | |
1088 | } | |
1089 | ||
1090 | static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
87990467 | 1091 | struct Qdisc **old) |
1da177e4 | 1092 | { |
87990467 | 1093 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1094 | |
1095 | if (cl && !cl->level) { | |
9f9afec4 | 1096 | if (new == NULL && |
5ce2d488 | 1097 | (new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, |
bb949fbd | 1098 | &pfifo_qdisc_ops, |
f4c1f3e0 | 1099 | cl->common.classid)) |
87990467 SH |
1100 | == NULL) |
1101 | return -ENOBUFS; | |
1da177e4 | 1102 | sch_tree_lock(sch); |
b94c8afc PM |
1103 | *old = cl->un.leaf.q; |
1104 | cl->un.leaf.q = new; | |
1105 | if (*old != NULL) { | |
256d61b8 | 1106 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); |
1da177e4 LT |
1107 | qdisc_reset(*old); |
1108 | } | |
1109 | sch_tree_unlock(sch); | |
1110 | return 0; | |
1111 | } | |
1112 | return -ENOENT; | |
1113 | } | |
1114 | ||
87990467 | 1115 | static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg) |
1da177e4 | 1116 | { |
87990467 | 1117 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1118 | return (cl && !cl->level) ? cl->un.leaf.q : NULL; |
1119 | } | |
1120 | ||
256d61b8 PM |
1121 | static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg) |
1122 | { | |
1123 | struct htb_class *cl = (struct htb_class *)arg; | |
1124 | ||
1125 | if (cl->un.leaf.q->q.qlen == 0) | |
1126 | htb_deactivate(qdisc_priv(sch), cl); | |
1127 | } | |
1128 | ||
1da177e4 LT |
1129 | static unsigned long htb_get(struct Qdisc *sch, u32 classid) |
1130 | { | |
87990467 SH |
1131 | struct htb_class *cl = htb_find(classid, sch); |
1132 | if (cl) | |
1da177e4 LT |
1133 | cl->refcnt++; |
1134 | return (unsigned long)cl; | |
1135 | } | |
1136 | ||
160d5e10 JP |
1137 | static inline int htb_parent_last_child(struct htb_class *cl) |
1138 | { | |
1139 | if (!cl->parent) | |
1140 | /* the root class */ | |
1141 | return 0; | |
42077599 | 1142 | if (cl->parent->children > 1) |
160d5e10 JP |
1143 | /* not the last child */ |
1144 | return 0; | |
160d5e10 JP |
1145 | return 1; |
1146 | } | |
1147 | ||
3ba08b00 JP |
1148 | static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl, |
1149 | struct Qdisc *new_q) | |
160d5e10 JP |
1150 | { |
1151 | struct htb_class *parent = cl->parent; | |
1152 | ||
547b792c | 1153 | WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity); |
160d5e10 | 1154 | |
3ba08b00 JP |
1155 | if (parent->cmode != HTB_CAN_SEND) |
1156 | htb_safe_rb_erase(&parent->pq_node, q->wait_pq + parent->level); | |
1157 | ||
160d5e10 JP |
1158 | parent->level = 0; |
1159 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1160 | INIT_LIST_HEAD(&parent->un.leaf.drop_list); | |
1161 | parent->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
1162 | parent->un.leaf.quantum = parent->quantum; | |
1163 | parent->un.leaf.prio = parent->prio; | |
1164 | parent->tokens = parent->buffer; | |
1165 | parent->ctokens = parent->cbuffer; | |
3bebcda2 | 1166 | parent->t_c = psched_get_time(); |
160d5e10 JP |
1167 | parent->cmode = HTB_CAN_SEND; |
1168 | } | |
1169 | ||
87990467 | 1170 | static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl) |
1da177e4 | 1171 | { |
1da177e4 | 1172 | if (!cl->level) { |
547b792c | 1173 | WARN_ON(!cl->un.leaf.q); |
1da177e4 LT |
1174 | qdisc_destroy(cl->un.leaf.q); |
1175 | } | |
ee39e10c | 1176 | gen_kill_estimator(&cl->bstats, &cl->rate_est); |
1da177e4 LT |
1177 | qdisc_put_rtab(cl->rate); |
1178 | qdisc_put_rtab(cl->ceil); | |
87990467 | 1179 | |
ff31ab56 | 1180 | tcf_destroy_chain(&cl->filter_list); |
1da177e4 LT |
1181 | kfree(cl); |
1182 | } | |
1183 | ||
1184 | /* always caled under BH & queue lock */ | |
87990467 | 1185 | static void htb_destroy(struct Qdisc *sch) |
1da177e4 LT |
1186 | { |
1187 | struct htb_sched *q = qdisc_priv(sch); | |
fbd8f137 PM |
1188 | struct hlist_node *n, *next; |
1189 | struct htb_class *cl; | |
1190 | unsigned int i; | |
1da177e4 | 1191 | |
fb983d45 | 1192 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 | 1193 | /* This line used to be after htb_destroy_class call below |
10297b99 | 1194 | and surprisingly it worked in 2.4. But it must precede it |
1da177e4 LT |
1195 | because filter need its target class alive to be able to call |
1196 | unbind_filter on it (without Oops). */ | |
ff31ab56 | 1197 | tcf_destroy_chain(&q->filter_list); |
87990467 | 1198 | |
f4c1f3e0 PM |
1199 | for (i = 0; i < q->clhash.hashsize; i++) { |
1200 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) | |
fbd8f137 PM |
1201 | tcf_destroy_chain(&cl->filter_list); |
1202 | } | |
f4c1f3e0 PM |
1203 | for (i = 0; i < q->clhash.hashsize; i++) { |
1204 | hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i], | |
1205 | common.hnode) | |
fbd8f137 PM |
1206 | htb_destroy_class(sch, cl); |
1207 | } | |
f4c1f3e0 | 1208 | qdisc_class_hash_destroy(&q->clhash); |
1da177e4 LT |
1209 | __skb_queue_purge(&q->direct_queue); |
1210 | } | |
1211 | ||
1212 | static int htb_delete(struct Qdisc *sch, unsigned long arg) | |
1213 | { | |
1214 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1215 | struct htb_class *cl = (struct htb_class *)arg; |
256d61b8 | 1216 | unsigned int qlen; |
160d5e10 JP |
1217 | struct Qdisc *new_q = NULL; |
1218 | int last_child = 0; | |
1da177e4 LT |
1219 | |
1220 | // TODO: why don't allow to delete subtree ? references ? does | |
1221 | // tc subsys quarantee us that in htb_destroy it holds no class | |
1222 | // refs so that we can remove children safely there ? | |
42077599 | 1223 | if (cl->children || cl->filter_cnt) |
1da177e4 | 1224 | return -EBUSY; |
87990467 | 1225 | |
160d5e10 | 1226 | if (!cl->level && htb_parent_last_child(cl)) { |
5ce2d488 | 1227 | new_q = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, |
bb949fbd DM |
1228 | &pfifo_qdisc_ops, |
1229 | cl->parent->common.classid); | |
160d5e10 JP |
1230 | last_child = 1; |
1231 | } | |
1232 | ||
1da177e4 | 1233 | sch_tree_lock(sch); |
87990467 | 1234 | |
814a175e | 1235 | if (!cl->level) { |
256d61b8 | 1236 | qlen = cl->un.leaf.q->q.qlen; |
814a175e | 1237 | qdisc_reset(cl->un.leaf.q); |
256d61b8 | 1238 | qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen); |
814a175e PM |
1239 | } |
1240 | ||
f4c1f3e0 PM |
1241 | /* delete from hash and active; remainder in destroy_class */ |
1242 | qdisc_class_hash_remove(&q->clhash, &cl->common); | |
26b284de JP |
1243 | if (cl->parent) |
1244 | cl->parent->children--; | |
c38c83cb | 1245 | |
1da177e4 | 1246 | if (cl->prio_activity) |
87990467 | 1247 | htb_deactivate(q, cl); |
1da177e4 | 1248 | |
fbd8f137 PM |
1249 | if (cl->cmode != HTB_CAN_SEND) |
1250 | htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level); | |
1251 | ||
160d5e10 | 1252 | if (last_child) |
3ba08b00 | 1253 | htb_parent_to_leaf(q, cl, new_q); |
160d5e10 | 1254 | |
1da177e4 | 1255 | if (--cl->refcnt == 0) |
87990467 | 1256 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1257 | |
1258 | sch_tree_unlock(sch); | |
1259 | return 0; | |
1260 | } | |
1261 | ||
1262 | static void htb_put(struct Qdisc *sch, unsigned long arg) | |
1263 | { | |
87990467 | 1264 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1265 | |
1266 | if (--cl->refcnt == 0) | |
87990467 | 1267 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1268 | } |
1269 | ||
87990467 | 1270 | static int htb_change_class(struct Qdisc *sch, u32 classid, |
1e90474c | 1271 | u32 parentid, struct nlattr **tca, |
87990467 | 1272 | unsigned long *arg) |
1da177e4 LT |
1273 | { |
1274 | int err = -EINVAL; | |
1275 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1276 | struct htb_class *cl = (struct htb_class *)*arg, *parent; |
1e90474c | 1277 | struct nlattr *opt = tca[TCA_OPTIONS]; |
1da177e4 | 1278 | struct qdisc_rate_table *rtab = NULL, *ctab = NULL; |
1e90474c | 1279 | struct nlattr *tb[TCA_HTB_RTAB + 1]; |
1da177e4 LT |
1280 | struct tc_htb_opt *hopt; |
1281 | ||
1282 | /* extract all subattrs from opt attr */ | |
cee63723 PM |
1283 | if (!opt) |
1284 | goto failure; | |
1285 | ||
27a3421e | 1286 | err = nla_parse_nested(tb, TCA_HTB_RTAB, opt, htb_policy); |
cee63723 PM |
1287 | if (err < 0) |
1288 | goto failure; | |
1289 | ||
1290 | err = -EINVAL; | |
27a3421e | 1291 | if (tb[TCA_HTB_PARMS] == NULL) |
1da177e4 | 1292 | goto failure; |
1da177e4 | 1293 | |
87990467 SH |
1294 | parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch); |
1295 | ||
1e90474c | 1296 | hopt = nla_data(tb[TCA_HTB_PARMS]); |
3bf72957 | 1297 | |
1e90474c PM |
1298 | rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB]); |
1299 | ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB]); | |
87990467 SH |
1300 | if (!rtab || !ctab) |
1301 | goto failure; | |
1da177e4 | 1302 | |
87990467 | 1303 | if (!cl) { /* new class */ |
1da177e4 | 1304 | struct Qdisc *new_q; |
3696f625 | 1305 | int prio; |
ee39e10c | 1306 | struct { |
1e90474c | 1307 | struct nlattr nla; |
ee39e10c PM |
1308 | struct gnet_estimator opt; |
1309 | } est = { | |
1e90474c PM |
1310 | .nla = { |
1311 | .nla_len = nla_attr_size(sizeof(est.opt)), | |
1312 | .nla_type = TCA_RATE, | |
ee39e10c PM |
1313 | }, |
1314 | .opt = { | |
1315 | /* 4s interval, 16s averaging constant */ | |
1316 | .interval = 2, | |
1317 | .ewma_log = 2, | |
1318 | }, | |
1319 | }; | |
3696f625 | 1320 | |
1da177e4 | 1321 | /* check for valid classid */ |
87990467 SH |
1322 | if (!classid || TC_H_MAJ(classid ^ sch->handle) |
1323 | || htb_find(classid, sch)) | |
1da177e4 LT |
1324 | goto failure; |
1325 | ||
1326 | /* check maximal depth */ | |
1327 | if (parent && parent->parent && parent->parent->level < 2) { | |
1328 | printk(KERN_ERR "htb: tree is too deep\n"); | |
1329 | goto failure; | |
1330 | } | |
1331 | err = -ENOBUFS; | |
0da974f4 | 1332 | if ((cl = kzalloc(sizeof(*cl), GFP_KERNEL)) == NULL) |
1da177e4 | 1333 | goto failure; |
87990467 | 1334 | |
ee39e10c | 1335 | gen_new_estimator(&cl->bstats, &cl->rate_est, |
f6f9b93f | 1336 | qdisc_root_sleeping_lock(sch), |
1e90474c | 1337 | tca[TCA_RATE] ? : &est.nla); |
1da177e4 | 1338 | cl->refcnt = 1; |
42077599 | 1339 | cl->children = 0; |
1da177e4 | 1340 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); |
3696f625 SH |
1341 | RB_CLEAR_NODE(&cl->pq_node); |
1342 | ||
1343 | for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) | |
1344 | RB_CLEAR_NODE(&cl->node[prio]); | |
1da177e4 LT |
1345 | |
1346 | /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL) | |
1347 | so that can't be used inside of sch_tree_lock | |
1348 | -- thanks to Karlis Peisenieks */ | |
5ce2d488 | 1349 | new_q = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, |
bb949fbd | 1350 | &pfifo_qdisc_ops, classid); |
1da177e4 LT |
1351 | sch_tree_lock(sch); |
1352 | if (parent && !parent->level) { | |
256d61b8 PM |
1353 | unsigned int qlen = parent->un.leaf.q->q.qlen; |
1354 | ||
1da177e4 | 1355 | /* turn parent into inner node */ |
256d61b8 PM |
1356 | qdisc_reset(parent->un.leaf.q); |
1357 | qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen); | |
87990467 SH |
1358 | qdisc_destroy(parent->un.leaf.q); |
1359 | if (parent->prio_activity) | |
1360 | htb_deactivate(q, parent); | |
1da177e4 LT |
1361 | |
1362 | /* remove from evt list because of level change */ | |
1363 | if (parent->cmode != HTB_CAN_SEND) { | |
3696f625 | 1364 | htb_safe_rb_erase(&parent->pq_node, q->wait_pq); |
1da177e4 LT |
1365 | parent->cmode = HTB_CAN_SEND; |
1366 | } | |
1367 | parent->level = (parent->parent ? parent->parent->level | |
87990467 SH |
1368 | : TC_HTB_MAXDEPTH) - 1; |
1369 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1da177e4 LT |
1370 | } |
1371 | /* leaf (we) needs elementary qdisc */ | |
1372 | cl->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
1373 | ||
f4c1f3e0 | 1374 | cl->common.classid = classid; |
87990467 | 1375 | cl->parent = parent; |
1da177e4 LT |
1376 | |
1377 | /* set class to be in HTB_CAN_SEND state */ | |
1378 | cl->tokens = hopt->buffer; | |
1379 | cl->ctokens = hopt->cbuffer; | |
00c04af9 | 1380 | cl->mbuffer = 60 * PSCHED_TICKS_PER_SEC; /* 1min */ |
3bebcda2 | 1381 | cl->t_c = psched_get_time(); |
1da177e4 LT |
1382 | cl->cmode = HTB_CAN_SEND; |
1383 | ||
1384 | /* attach to the hash list and parent's family */ | |
f4c1f3e0 | 1385 | qdisc_class_hash_insert(&q->clhash, &cl->common); |
42077599 PM |
1386 | if (parent) |
1387 | parent->children++; | |
ee39e10c | 1388 | } else { |
1e90474c | 1389 | if (tca[TCA_RATE]) |
ee39e10c | 1390 | gen_replace_estimator(&cl->bstats, &cl->rate_est, |
f6f9b93f | 1391 | qdisc_root_sleeping_lock(sch), |
1e90474c | 1392 | tca[TCA_RATE]); |
87990467 | 1393 | sch_tree_lock(sch); |
ee39e10c | 1394 | } |
1da177e4 LT |
1395 | |
1396 | /* it used to be a nasty bug here, we have to check that node | |
87990467 | 1397 | is really leaf before changing cl->un.leaf ! */ |
1da177e4 LT |
1398 | if (!cl->level) { |
1399 | cl->un.leaf.quantum = rtab->rate.rate / q->rate2quantum; | |
1400 | if (!hopt->quantum && cl->un.leaf.quantum < 1000) { | |
87990467 SH |
1401 | printk(KERN_WARNING |
1402 | "HTB: quantum of class %X is small. Consider r2q change.\n", | |
f4c1f3e0 | 1403 | cl->common.classid); |
1da177e4 LT |
1404 | cl->un.leaf.quantum = 1000; |
1405 | } | |
1406 | if (!hopt->quantum && cl->un.leaf.quantum > 200000) { | |
87990467 SH |
1407 | printk(KERN_WARNING |
1408 | "HTB: quantum of class %X is big. Consider r2q change.\n", | |
f4c1f3e0 | 1409 | cl->common.classid); |
1da177e4 LT |
1410 | cl->un.leaf.quantum = 200000; |
1411 | } | |
1412 | if (hopt->quantum) | |
1413 | cl->un.leaf.quantum = hopt->quantum; | |
1414 | if ((cl->un.leaf.prio = hopt->prio) >= TC_HTB_NUMPRIO) | |
1415 | cl->un.leaf.prio = TC_HTB_NUMPRIO - 1; | |
160d5e10 JP |
1416 | |
1417 | /* backup for htb_parent_to_leaf */ | |
1418 | cl->quantum = cl->un.leaf.quantum; | |
1419 | cl->prio = cl->un.leaf.prio; | |
1da177e4 LT |
1420 | } |
1421 | ||
1422 | cl->buffer = hopt->buffer; | |
1423 | cl->cbuffer = hopt->cbuffer; | |
87990467 SH |
1424 | if (cl->rate) |
1425 | qdisc_put_rtab(cl->rate); | |
1426 | cl->rate = rtab; | |
1427 | if (cl->ceil) | |
1428 | qdisc_put_rtab(cl->ceil); | |
1429 | cl->ceil = ctab; | |
1da177e4 LT |
1430 | sch_tree_unlock(sch); |
1431 | ||
f4c1f3e0 PM |
1432 | qdisc_class_hash_grow(sch, &q->clhash); |
1433 | ||
1da177e4 LT |
1434 | *arg = (unsigned long)cl; |
1435 | return 0; | |
1436 | ||
1437 | failure: | |
87990467 SH |
1438 | if (rtab) |
1439 | qdisc_put_rtab(rtab); | |
1440 | if (ctab) | |
1441 | qdisc_put_rtab(ctab); | |
1da177e4 LT |
1442 | return err; |
1443 | } | |
1444 | ||
1445 | static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg) | |
1446 | { | |
1447 | struct htb_sched *q = qdisc_priv(sch); | |
1448 | struct htb_class *cl = (struct htb_class *)arg; | |
1449 | struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list; | |
3bf72957 | 1450 | |
1da177e4 LT |
1451 | return fl; |
1452 | } | |
1453 | ||
1454 | static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, | |
87990467 | 1455 | u32 classid) |
1da177e4 | 1456 | { |
87990467 | 1457 | struct htb_class *cl = htb_find(classid, sch); |
3bf72957 | 1458 | |
1da177e4 | 1459 | /*if (cl && !cl->level) return 0; |
87990467 SH |
1460 | The line above used to be there to prevent attaching filters to |
1461 | leaves. But at least tc_index filter uses this just to get class | |
1462 | for other reasons so that we have to allow for it. | |
1463 | ---- | |
1464 | 19.6.2002 As Werner explained it is ok - bind filter is just | |
1465 | another way to "lock" the class - unlike "get" this lock can | |
1466 | be broken by class during destroy IIUC. | |
1da177e4 | 1467 | */ |
87990467 SH |
1468 | if (cl) |
1469 | cl->filter_cnt++; | |
1da177e4 LT |
1470 | return (unsigned long)cl; |
1471 | } | |
1472 | ||
1473 | static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) | |
1474 | { | |
1da177e4 | 1475 | struct htb_class *cl = (struct htb_class *)arg; |
3bf72957 | 1476 | |
87990467 SH |
1477 | if (cl) |
1478 | cl->filter_cnt--; | |
1da177e4 LT |
1479 | } |
1480 | ||
1481 | static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
1482 | { | |
1483 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 PM |
1484 | struct htb_class *cl; |
1485 | struct hlist_node *n; | |
1486 | unsigned int i; | |
1da177e4 LT |
1487 | |
1488 | if (arg->stop) | |
1489 | return; | |
1490 | ||
f4c1f3e0 PM |
1491 | for (i = 0; i < q->clhash.hashsize; i++) { |
1492 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { | |
1da177e4 LT |
1493 | if (arg->count < arg->skip) { |
1494 | arg->count++; | |
1495 | continue; | |
1496 | } | |
1497 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
1498 | arg->stop = 1; | |
1499 | return; | |
1500 | } | |
1501 | arg->count++; | |
1502 | } | |
1503 | } | |
1504 | } | |
1505 | ||
20fea08b | 1506 | static const struct Qdisc_class_ops htb_class_ops = { |
1da177e4 LT |
1507 | .graft = htb_graft, |
1508 | .leaf = htb_leaf, | |
256d61b8 | 1509 | .qlen_notify = htb_qlen_notify, |
1da177e4 LT |
1510 | .get = htb_get, |
1511 | .put = htb_put, | |
1512 | .change = htb_change_class, | |
1513 | .delete = htb_delete, | |
1514 | .walk = htb_walk, | |
1515 | .tcf_chain = htb_find_tcf, | |
1516 | .bind_tcf = htb_bind_filter, | |
1517 | .unbind_tcf = htb_unbind_filter, | |
1518 | .dump = htb_dump_class, | |
1519 | .dump_stats = htb_dump_class_stats, | |
1520 | }; | |
1521 | ||
20fea08b | 1522 | static struct Qdisc_ops htb_qdisc_ops __read_mostly = { |
1da177e4 LT |
1523 | .next = NULL, |
1524 | .cl_ops = &htb_class_ops, | |
1525 | .id = "htb", | |
1526 | .priv_size = sizeof(struct htb_sched), | |
1527 | .enqueue = htb_enqueue, | |
1528 | .dequeue = htb_dequeue, | |
77be155c | 1529 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1530 | .drop = htb_drop, |
1531 | .init = htb_init, | |
1532 | .reset = htb_reset, | |
1533 | .destroy = htb_destroy, | |
1534 | .change = NULL /* htb_change */, | |
1535 | .dump = htb_dump, | |
1536 | .owner = THIS_MODULE, | |
1537 | }; | |
1538 | ||
1539 | static int __init htb_module_init(void) | |
1540 | { | |
87990467 | 1541 | return register_qdisc(&htb_qdisc_ops); |
1da177e4 | 1542 | } |
87990467 | 1543 | static void __exit htb_module_exit(void) |
1da177e4 | 1544 | { |
87990467 | 1545 | unregister_qdisc(&htb_qdisc_ops); |
1da177e4 | 1546 | } |
87990467 | 1547 | |
1da177e4 LT |
1548 | module_init(htb_module_init) |
1549 | module_exit(htb_module_exit) | |
1550 | MODULE_LICENSE("GPL"); |