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> |
1224736d | 38 | #include <linux/workqueue.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
dc5fc579 | 40 | #include <net/netlink.h> |
292f1c7f | 41 | #include <net/sch_generic.h> |
1da177e4 | 42 | #include <net/pkt_sched.h> |
1da177e4 LT |
43 | |
44 | /* HTB algorithm. | |
45 | Author: devik@cdi.cz | |
46 | ======================================================================== | |
47 | HTB is like TBF with multiple classes. It is also similar to CBQ because | |
10297b99 | 48 | it allows to assign priority to each class in hierarchy. |
1da177e4 LT |
49 | In fact it is another implementation of Floyd's formal sharing. |
50 | ||
51 | Levels: | |
10297b99 | 52 | Each class is assigned level. Leaf has ALWAYS level 0 and root |
1da177e4 LT |
53 | classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level |
54 | one less than their parent. | |
55 | */ | |
56 | ||
47083fc0 | 57 | static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */ |
87990467 | 58 | #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */ |
1da177e4 LT |
59 | |
60 | #if HTB_VER >> 16 != TC_HTB_PROTOVER | |
61 | #error "Mismatched sch_htb.c and pkt_sch.h" | |
62 | #endif | |
63 | ||
47083fc0 JDB |
64 | /* Module parameter and sysfs export */ |
65 | module_param (htb_hysteresis, int, 0640); | |
66 | MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate"); | |
67 | ||
1da177e4 LT |
68 | /* used internaly to keep status of single class */ |
69 | enum htb_cmode { | |
87990467 SH |
70 | HTB_CANT_SEND, /* class can't send and can't borrow */ |
71 | HTB_MAY_BORROW, /* class can't send but may borrow */ | |
72 | HTB_CAN_SEND /* class can send */ | |
1da177e4 LT |
73 | }; |
74 | ||
75 | /* interior & leaf nodes; props specific to leaves are marked L: */ | |
87990467 | 76 | struct htb_class { |
f4c1f3e0 | 77 | struct Qdisc_class_common common; |
87990467 | 78 | /* general class parameters */ |
c1a8f1f1 | 79 | struct gnet_stats_basic_packed bstats; |
87990467 SH |
80 | struct gnet_stats_queue qstats; |
81 | struct gnet_stats_rate_est rate_est; | |
82 | struct tc_htb_xstats xstats; /* our special stats */ | |
83 | int refcnt; /* usage count of this class */ | |
1da177e4 | 84 | |
87990467 SH |
85 | /* topology */ |
86 | int level; /* our level (see above) */ | |
42077599 | 87 | unsigned int children; |
87990467 | 88 | struct htb_class *parent; /* parent class */ |
87990467 | 89 | |
c19f7a34 JP |
90 | int prio; /* these two are used only by leaves... */ |
91 | int quantum; /* but stored for parent-to-leaf return */ | |
92 | ||
87990467 SH |
93 | union { |
94 | struct htb_class_leaf { | |
95 | struct Qdisc *q; | |
87990467 SH |
96 | int deficit[TC_HTB_MAXDEPTH]; |
97 | struct list_head drop_list; | |
98 | } leaf; | |
99 | struct htb_class_inner { | |
100 | struct rb_root feed[TC_HTB_NUMPRIO]; /* feed trees */ | |
101 | struct rb_node *ptr[TC_HTB_NUMPRIO]; /* current class ptr */ | |
102 | /* When class changes from state 1->2 and disconnects from | |
cc7ec456 ED |
103 | * parent's feed then we lost ptr value and start from the |
104 | * first child again. Here we store classid of the | |
105 | * last valid ptr (used when ptr is NULL). | |
106 | */ | |
87990467 SH |
107 | u32 last_ptr_id[TC_HTB_NUMPRIO]; |
108 | } inner; | |
109 | } un; | |
110 | struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ | |
111 | struct rb_node pq_node; /* node for event queue */ | |
fb983d45 | 112 | psched_time_t pq_key; |
87990467 SH |
113 | |
114 | int prio_activity; /* for which prios are we active */ | |
115 | enum htb_cmode cmode; /* current mode of the class */ | |
116 | ||
117 | /* class attached filters */ | |
118 | struct tcf_proto *filter_list; | |
119 | int filter_cnt; | |
120 | ||
87990467 | 121 | /* token bucket parameters */ |
292f1c7f JP |
122 | struct psched_ratecfg rate; |
123 | struct psched_ratecfg ceil; | |
56b765b7 | 124 | s64 buffer, cbuffer; /* token bucket depth/rate */ |
87990467 | 125 | psched_tdiff_t mbuffer; /* max wait time */ |
56b765b7 | 126 | s64 tokens, ctokens; /* current number of tokens */ |
87990467 | 127 | psched_time_t t_c; /* checkpoint time */ |
1da177e4 LT |
128 | }; |
129 | ||
87990467 | 130 | struct htb_sched { |
f4c1f3e0 | 131 | struct Qdisc_class_hash clhash; |
0cef296d | 132 | struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */ |
87990467 SH |
133 | |
134 | /* self list - roots of self generating tree */ | |
135 | struct rb_root row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
136 | int row_mask[TC_HTB_MAXDEPTH]; | |
137 | struct rb_node *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
138 | u32 last_ptr_id[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
1da177e4 | 139 | |
87990467 SH |
140 | /* self wait list - roots of wait PQs per row */ |
141 | struct rb_root wait_pq[TC_HTB_MAXDEPTH]; | |
1da177e4 | 142 | |
87990467 | 143 | /* time of nearest event per level (row) */ |
fb983d45 | 144 | psched_time_t near_ev_cache[TC_HTB_MAXDEPTH]; |
1da177e4 | 145 | |
87990467 | 146 | int defcls; /* class where unclassified flows go to */ |
1da177e4 | 147 | |
87990467 SH |
148 | /* filters for qdisc itself */ |
149 | struct tcf_proto *filter_list; | |
1da177e4 | 150 | |
87990467 SH |
151 | int rate2quantum; /* quant = rate / rate2quantum */ |
152 | psched_time_t now; /* cached dequeue time */ | |
fb983d45 | 153 | struct qdisc_watchdog watchdog; |
1da177e4 | 154 | |
87990467 SH |
155 | /* non shaped skbs; let them go directly thru */ |
156 | struct sk_buff_head direct_queue; | |
157 | int direct_qlen; /* max qlen of above */ | |
158 | ||
159 | long direct_pkts; | |
e82181de JP |
160 | |
161 | #define HTB_WARN_TOOMANYEVENTS 0x1 | |
162 | unsigned int warned; /* only one warning */ | |
1224736d | 163 | struct work_struct work; |
1da177e4 LT |
164 | }; |
165 | ||
1da177e4 | 166 | /* find class in global hash table using given handle */ |
87990467 | 167 | static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch) |
1da177e4 LT |
168 | { |
169 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 170 | struct Qdisc_class_common *clc; |
0cef296d | 171 | |
f4c1f3e0 PM |
172 | clc = qdisc_class_find(&q->clhash, handle); |
173 | if (clc == NULL) | |
1da177e4 | 174 | return NULL; |
f4c1f3e0 | 175 | return container_of(clc, struct htb_class, common); |
1da177e4 LT |
176 | } |
177 | ||
178 | /** | |
179 | * htb_classify - classify a packet into class | |
180 | * | |
181 | * It returns NULL if the packet should be dropped or -1 if the packet | |
182 | * should be passed directly thru. In all other cases leaf class is returned. | |
183 | * We allow direct class selection by classid in priority. The we examine | |
184 | * filters in qdisc and in inner nodes (if higher filter points to the inner | |
185 | * node). If we end up with classid MAJOR:0 we enqueue the skb into special | |
10297b99 | 186 | * internal fifo (direct). These packets then go directly thru. If we still |
25985edc | 187 | * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful |
1da177e4 LT |
188 | * then finish and return direct queue. |
189 | */ | |
cc7ec456 | 190 | #define HTB_DIRECT ((struct htb_class *)-1L) |
1da177e4 | 191 | |
87990467 SH |
192 | static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, |
193 | int *qerr) | |
1da177e4 LT |
194 | { |
195 | struct htb_sched *q = qdisc_priv(sch); | |
196 | struct htb_class *cl; | |
197 | struct tcf_result res; | |
198 | struct tcf_proto *tcf; | |
199 | int result; | |
200 | ||
201 | /* allow to select class by setting skb->priority to valid classid; | |
cc7ec456 ED |
202 | * note that nfmark can be used too by attaching filter fw with no |
203 | * rules in it | |
204 | */ | |
1da177e4 | 205 | if (skb->priority == sch->handle) |
87990467 | 206 | return HTB_DIRECT; /* X:0 (direct flow) selected */ |
cc7ec456 ED |
207 | cl = htb_find(skb->priority, sch); |
208 | if (cl && cl->level == 0) | |
1da177e4 LT |
209 | return cl; |
210 | ||
c27f339a | 211 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 LT |
212 | tcf = q->filter_list; |
213 | while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) { | |
214 | #ifdef CONFIG_NET_CLS_ACT | |
215 | switch (result) { | |
216 | case TC_ACT_QUEUED: | |
87990467 | 217 | case TC_ACT_STOLEN: |
378a2f09 | 218 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
1da177e4 LT |
219 | case TC_ACT_SHOT: |
220 | return NULL; | |
221 | } | |
1da177e4 | 222 | #endif |
cc7ec456 ED |
223 | cl = (void *)res.class; |
224 | if (!cl) { | |
1da177e4 | 225 | if (res.classid == sch->handle) |
87990467 | 226 | return HTB_DIRECT; /* X:0 (direct flow) */ |
cc7ec456 ED |
227 | cl = htb_find(res.classid, sch); |
228 | if (!cl) | |
87990467 | 229 | break; /* filter selected invalid classid */ |
1da177e4 LT |
230 | } |
231 | if (!cl->level) | |
87990467 | 232 | return cl; /* we hit leaf; return it */ |
1da177e4 LT |
233 | |
234 | /* we have got inner class; apply inner filter chain */ | |
235 | tcf = cl->filter_list; | |
236 | } | |
237 | /* classification failed; try to use default class */ | |
87990467 | 238 | cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch); |
1da177e4 | 239 | if (!cl || cl->level) |
87990467 | 240 | return HTB_DIRECT; /* bad default .. this is safe bet */ |
1da177e4 LT |
241 | return cl; |
242 | } | |
243 | ||
1da177e4 LT |
244 | /** |
245 | * htb_add_to_id_tree - adds class to the round robin list | |
246 | * | |
247 | * Routine adds class to the list (actually tree) sorted by classid. | |
248 | * Make sure that class is not already on such list for given prio. | |
249 | */ | |
87990467 SH |
250 | static void htb_add_to_id_tree(struct rb_root *root, |
251 | struct htb_class *cl, int prio) | |
1da177e4 LT |
252 | { |
253 | struct rb_node **p = &root->rb_node, *parent = NULL; | |
3bf72957 | 254 | |
1da177e4 | 255 | while (*p) { |
87990467 SH |
256 | struct htb_class *c; |
257 | parent = *p; | |
1da177e4 | 258 | c = rb_entry(parent, struct htb_class, node[prio]); |
3bf72957 | 259 | |
f4c1f3e0 | 260 | if (cl->common.classid > c->common.classid) |
1da177e4 | 261 | p = &parent->rb_right; |
87990467 | 262 | else |
1da177e4 LT |
263 | p = &parent->rb_left; |
264 | } | |
265 | rb_link_node(&cl->node[prio], parent, p); | |
266 | rb_insert_color(&cl->node[prio], root); | |
267 | } | |
268 | ||
269 | /** | |
270 | * htb_add_to_wait_tree - adds class to the event queue with delay | |
271 | * | |
272 | * The class is added to priority event queue to indicate that class will | |
273 | * change its mode in cl->pq_key microseconds. Make sure that class is not | |
274 | * already in the queue. | |
275 | */ | |
87990467 | 276 | static void htb_add_to_wait_tree(struct htb_sched *q, |
56b765b7 | 277 | struct htb_class *cl, s64 delay) |
1da177e4 LT |
278 | { |
279 | struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL; | |
3bf72957 | 280 | |
fb983d45 PM |
281 | cl->pq_key = q->now + delay; |
282 | if (cl->pq_key == q->now) | |
1da177e4 LT |
283 | cl->pq_key++; |
284 | ||
285 | /* update the nearest event cache */ | |
fb983d45 | 286 | if (q->near_ev_cache[cl->level] > cl->pq_key) |
1da177e4 | 287 | q->near_ev_cache[cl->level] = cl->pq_key; |
87990467 | 288 | |
1da177e4 | 289 | while (*p) { |
87990467 SH |
290 | struct htb_class *c; |
291 | parent = *p; | |
1da177e4 | 292 | c = rb_entry(parent, struct htb_class, pq_node); |
fb983d45 | 293 | if (cl->pq_key >= c->pq_key) |
1da177e4 | 294 | p = &parent->rb_right; |
87990467 | 295 | else |
1da177e4 LT |
296 | p = &parent->rb_left; |
297 | } | |
298 | rb_link_node(&cl->pq_node, parent, p); | |
299 | rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]); | |
300 | } | |
301 | ||
302 | /** | |
303 | * htb_next_rb_node - finds next node in binary tree | |
304 | * | |
305 | * When we are past last key we return NULL. | |
306 | * Average complexity is 2 steps per call. | |
307 | */ | |
3696f625 | 308 | static inline void htb_next_rb_node(struct rb_node **n) |
1da177e4 LT |
309 | { |
310 | *n = rb_next(*n); | |
311 | } | |
312 | ||
313 | /** | |
314 | * htb_add_class_to_row - add class to its row | |
315 | * | |
316 | * The class is added to row at priorities marked in mask. | |
317 | * It does nothing if mask == 0. | |
318 | */ | |
87990467 SH |
319 | static inline void htb_add_class_to_row(struct htb_sched *q, |
320 | struct htb_class *cl, int mask) | |
1da177e4 | 321 | { |
1da177e4 LT |
322 | q->row_mask[cl->level] |= mask; |
323 | while (mask) { | |
324 | int prio = ffz(~mask); | |
325 | mask &= ~(1 << prio); | |
87990467 | 326 | htb_add_to_id_tree(q->row[cl->level] + prio, cl, prio); |
1da177e4 LT |
327 | } |
328 | } | |
329 | ||
3696f625 SH |
330 | /* If this triggers, it is a bug in this code, but it need not be fatal */ |
331 | static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root) | |
332 | { | |
81771b3b | 333 | if (RB_EMPTY_NODE(rb)) { |
3696f625 SH |
334 | WARN_ON(1); |
335 | } else { | |
336 | rb_erase(rb, root); | |
337 | RB_CLEAR_NODE(rb); | |
338 | } | |
339 | } | |
340 | ||
341 | ||
1da177e4 LT |
342 | /** |
343 | * htb_remove_class_from_row - removes class from its row | |
344 | * | |
345 | * The class is removed from row at priorities marked in mask. | |
346 | * It does nothing if mask == 0. | |
347 | */ | |
87990467 SH |
348 | static inline void htb_remove_class_from_row(struct htb_sched *q, |
349 | struct htb_class *cl, int mask) | |
1da177e4 LT |
350 | { |
351 | int m = 0; | |
3bf72957 | 352 | |
1da177e4 LT |
353 | while (mask) { |
354 | int prio = ffz(~mask); | |
3696f625 | 355 | |
1da177e4 | 356 | mask &= ~(1 << prio); |
87990467 SH |
357 | if (q->ptr[cl->level][prio] == cl->node + prio) |
358 | htb_next_rb_node(q->ptr[cl->level] + prio); | |
3696f625 SH |
359 | |
360 | htb_safe_rb_erase(cl->node + prio, q->row[cl->level] + prio); | |
87990467 | 361 | if (!q->row[cl->level][prio].rb_node) |
1da177e4 LT |
362 | m |= 1 << prio; |
363 | } | |
1da177e4 LT |
364 | q->row_mask[cl->level] &= ~m; |
365 | } | |
366 | ||
367 | /** | |
368 | * htb_activate_prios - creates active classe's feed chain | |
369 | * | |
370 | * The class is connected to ancestors and/or appropriate rows | |
10297b99 | 371 | * for priorities it is participating on. cl->cmode must be new |
1da177e4 LT |
372 | * (activated) mode. It does nothing if cl->prio_activity == 0. |
373 | */ | |
87990467 | 374 | static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
375 | { |
376 | struct htb_class *p = cl->parent; | |
87990467 | 377 | long m, mask = cl->prio_activity; |
1da177e4 LT |
378 | |
379 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
380 | m = mask; |
381 | while (m) { | |
1da177e4 LT |
382 | int prio = ffz(~m); |
383 | m &= ~(1 << prio); | |
87990467 | 384 | |
1da177e4 LT |
385 | if (p->un.inner.feed[prio].rb_node) |
386 | /* parent already has its feed in use so that | |
cc7ec456 ED |
387 | * reset bit in mask as parent is already ok |
388 | */ | |
1da177e4 | 389 | mask &= ~(1 << prio); |
87990467 SH |
390 | |
391 | htb_add_to_id_tree(p->un.inner.feed + prio, cl, prio); | |
1da177e4 | 392 | } |
1da177e4 | 393 | p->prio_activity |= mask; |
87990467 SH |
394 | cl = p; |
395 | p = cl->parent; | |
3bf72957 | 396 | |
1da177e4 LT |
397 | } |
398 | if (cl->cmode == HTB_CAN_SEND && mask) | |
87990467 | 399 | htb_add_class_to_row(q, cl, mask); |
1da177e4 LT |
400 | } |
401 | ||
402 | /** | |
403 | * htb_deactivate_prios - remove class from feed chain | |
404 | * | |
10297b99 | 405 | * cl->cmode must represent old mode (before deactivation). It does |
1da177e4 LT |
406 | * nothing if cl->prio_activity == 0. Class is removed from all feed |
407 | * chains and rows. | |
408 | */ | |
409 | static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) | |
410 | { | |
411 | struct htb_class *p = cl->parent; | |
87990467 | 412 | long m, mask = cl->prio_activity; |
1da177e4 LT |
413 | |
414 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
415 | m = mask; |
416 | mask = 0; | |
1da177e4 LT |
417 | while (m) { |
418 | int prio = ffz(~m); | |
419 | m &= ~(1 << prio); | |
87990467 SH |
420 | |
421 | if (p->un.inner.ptr[prio] == cl->node + prio) { | |
1da177e4 | 422 | /* we are removing child which is pointed to from |
cc7ec456 ED |
423 | * parent feed - forget the pointer but remember |
424 | * classid | |
425 | */ | |
f4c1f3e0 | 426 | p->un.inner.last_ptr_id[prio] = cl->common.classid; |
1da177e4 LT |
427 | p->un.inner.ptr[prio] = NULL; |
428 | } | |
87990467 | 429 | |
3696f625 | 430 | htb_safe_rb_erase(cl->node + prio, p->un.inner.feed + prio); |
87990467 SH |
431 | |
432 | if (!p->un.inner.feed[prio].rb_node) | |
1da177e4 LT |
433 | mask |= 1 << prio; |
434 | } | |
3bf72957 | 435 | |
1da177e4 | 436 | p->prio_activity &= ~mask; |
87990467 SH |
437 | cl = p; |
438 | p = cl->parent; | |
3bf72957 | 439 | |
1da177e4 | 440 | } |
87990467 SH |
441 | if (cl->cmode == HTB_CAN_SEND && mask) |
442 | htb_remove_class_from_row(q, cl, mask); | |
1da177e4 LT |
443 | } |
444 | ||
56b765b7 | 445 | static inline s64 htb_lowater(const struct htb_class *cl) |
18a63e86 | 446 | { |
47083fc0 JDB |
447 | if (htb_hysteresis) |
448 | return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0; | |
449 | else | |
450 | return 0; | |
18a63e86 | 451 | } |
56b765b7 | 452 | static inline s64 htb_hiwater(const struct htb_class *cl) |
18a63e86 | 453 | { |
47083fc0 JDB |
454 | if (htb_hysteresis) |
455 | return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0; | |
456 | else | |
457 | return 0; | |
18a63e86 | 458 | } |
47083fc0 | 459 | |
18a63e86 | 460 | |
1da177e4 LT |
461 | /** |
462 | * htb_class_mode - computes and returns current class mode | |
463 | * | |
464 | * It computes cl's mode at time cl->t_c+diff and returns it. If mode | |
465 | * is not HTB_CAN_SEND then cl->pq_key is updated to time difference | |
10297b99 | 466 | * from now to time when cl will change its state. |
1da177e4 | 467 | * Also it is worth to note that class mode doesn't change simply |
10297b99 | 468 | * at cl->{c,}tokens == 0 but there can rather be hysteresis of |
1da177e4 LT |
469 | * 0 .. -cl->{c,}buffer range. It is meant to limit number of |
470 | * mode transitions per time unit. The speed gain is about 1/6. | |
471 | */ | |
87990467 | 472 | static inline enum htb_cmode |
56b765b7 | 473 | htb_class_mode(struct htb_class *cl, s64 *diff) |
1da177e4 | 474 | { |
56b765b7 | 475 | s64 toks; |
1da177e4 | 476 | |
87990467 SH |
477 | if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) { |
478 | *diff = -toks; | |
479 | return HTB_CANT_SEND; | |
480 | } | |
18a63e86 | 481 | |
87990467 SH |
482 | if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl)) |
483 | return HTB_CAN_SEND; | |
1da177e4 | 484 | |
87990467 SH |
485 | *diff = -toks; |
486 | return HTB_MAY_BORROW; | |
1da177e4 LT |
487 | } |
488 | ||
489 | /** | |
490 | * htb_change_class_mode - changes classe's mode | |
491 | * | |
492 | * This should be the only way how to change classe's mode under normal | |
493 | * cirsumstances. Routine will update feed lists linkage, change mode | |
494 | * and add class to the wait event queue if appropriate. New mode should | |
495 | * be different from old one and cl->pq_key has to be valid if changing | |
496 | * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). | |
497 | */ | |
87990467 | 498 | static void |
56b765b7 | 499 | htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, s64 *diff) |
87990467 SH |
500 | { |
501 | enum htb_cmode new_mode = htb_class_mode(cl, diff); | |
1da177e4 LT |
502 | |
503 | if (new_mode == cl->cmode) | |
87990467 SH |
504 | return; |
505 | ||
506 | if (cl->prio_activity) { /* not necessary: speed optimization */ | |
507 | if (cl->cmode != HTB_CANT_SEND) | |
508 | htb_deactivate_prios(q, cl); | |
1da177e4 | 509 | cl->cmode = new_mode; |
87990467 SH |
510 | if (new_mode != HTB_CANT_SEND) |
511 | htb_activate_prios(q, cl); | |
512 | } else | |
1da177e4 LT |
513 | cl->cmode = new_mode; |
514 | } | |
515 | ||
516 | /** | |
10297b99 | 517 | * htb_activate - inserts leaf cl into appropriate active feeds |
1da177e4 LT |
518 | * |
519 | * Routine learns (new) priority of leaf and activates feed chain | |
520 | * for the prio. It can be called on already active leaf safely. | |
521 | * It also adds leaf into droplist. | |
522 | */ | |
87990467 | 523 | static inline void htb_activate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 524 | { |
547b792c | 525 | WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen); |
3bf72957 | 526 | |
1da177e4 | 527 | if (!cl->prio_activity) { |
c19f7a34 | 528 | cl->prio_activity = 1 << cl->prio; |
87990467 SH |
529 | htb_activate_prios(q, cl); |
530 | list_add_tail(&cl->un.leaf.drop_list, | |
c19f7a34 | 531 | q->drops + cl->prio); |
1da177e4 LT |
532 | } |
533 | } | |
534 | ||
535 | /** | |
10297b99 | 536 | * htb_deactivate - remove leaf cl from active feeds |
1da177e4 LT |
537 | * |
538 | * Make sure that leaf is active. In the other words it can't be called | |
539 | * with non-active leaf. It also removes class from the drop list. | |
540 | */ | |
87990467 | 541 | static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 542 | { |
547b792c | 543 | WARN_ON(!cl->prio_activity); |
3bf72957 | 544 | |
87990467 | 545 | htb_deactivate_prios(q, cl); |
1da177e4 LT |
546 | cl->prio_activity = 0; |
547 | list_del_init(&cl->un.leaf.drop_list); | |
548 | } | |
549 | ||
550 | static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch) | |
551 | { | |
f30ab418 | 552 | int uninitialized_var(ret); |
87990467 SH |
553 | struct htb_sched *q = qdisc_priv(sch); |
554 | struct htb_class *cl = htb_classify(skb, sch, &ret); | |
555 | ||
556 | if (cl == HTB_DIRECT) { | |
557 | /* enqueue to helper queue */ | |
558 | if (q->direct_queue.qlen < q->direct_qlen) { | |
559 | __skb_queue_tail(&q->direct_queue, skb); | |
560 | q->direct_pkts++; | |
561 | } else { | |
17045755 | 562 | return qdisc_drop(skb, sch); |
87990467 | 563 | } |
1da177e4 | 564 | #ifdef CONFIG_NET_CLS_ACT |
87990467 | 565 | } else if (!cl) { |
c27f339a | 566 | if (ret & __NET_XMIT_BYPASS) |
87990467 SH |
567 | sch->qstats.drops++; |
568 | kfree_skb(skb); | |
569 | return ret; | |
1da177e4 | 570 | #endif |
378a2f09 JP |
571 | } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q)) != NET_XMIT_SUCCESS) { |
572 | if (net_xmit_drop_count(ret)) { | |
573 | sch->qstats.drops++; | |
574 | cl->qstats.drops++; | |
575 | } | |
69747650 | 576 | return ret; |
87990467 | 577 | } else { |
87990467 SH |
578 | htb_activate(q, cl); |
579 | } | |
580 | ||
581 | sch->q.qlen++; | |
87990467 | 582 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
583 | } |
584 | ||
56b765b7 | 585 | static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, s64 diff) |
59e4220a | 586 | { |
56b765b7 | 587 | s64 toks = diff + cl->tokens; |
59e4220a JP |
588 | |
589 | if (toks > cl->buffer) | |
590 | toks = cl->buffer; | |
292f1c7f | 591 | toks -= (s64) psched_l2t_ns(&cl->rate, bytes); |
59e4220a JP |
592 | if (toks <= -cl->mbuffer) |
593 | toks = 1 - cl->mbuffer; | |
594 | ||
595 | cl->tokens = toks; | |
596 | } | |
597 | ||
56b765b7 | 598 | static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, s64 diff) |
59e4220a | 599 | { |
56b765b7 | 600 | s64 toks = diff + cl->ctokens; |
59e4220a JP |
601 | |
602 | if (toks > cl->cbuffer) | |
603 | toks = cl->cbuffer; | |
292f1c7f | 604 | toks -= (s64) psched_l2t_ns(&cl->ceil, bytes); |
59e4220a JP |
605 | if (toks <= -cl->mbuffer) |
606 | toks = 1 - cl->mbuffer; | |
607 | ||
608 | cl->ctokens = toks; | |
609 | } | |
610 | ||
1da177e4 LT |
611 | /** |
612 | * htb_charge_class - charges amount "bytes" to leaf and ancestors | |
613 | * | |
614 | * Routine assumes that packet "bytes" long was dequeued from leaf cl | |
615 | * borrowing from "level". It accounts bytes to ceil leaky bucket for | |
616 | * leaf and all ancestors and to rate bucket for ancestors at levels | |
617 | * "level" and higher. It also handles possible change of mode resulting | |
618 | * from the update. Note that mode can also increase here (MAY_BORROW to | |
619 | * CAN_SEND) because we can use more precise clock that event queue here. | |
620 | * In such case we remove class from event queue first. | |
621 | */ | |
87990467 | 622 | static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, |
c9726d68 | 623 | int level, struct sk_buff *skb) |
87990467 | 624 | { |
0abf77e5 | 625 | int bytes = qdisc_pkt_len(skb); |
1da177e4 | 626 | enum htb_cmode old_mode; |
56b765b7 | 627 | s64 diff; |
1da177e4 LT |
628 | |
629 | while (cl) { | |
56b765b7 | 630 | diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); |
1da177e4 | 631 | if (cl->level >= level) { |
87990467 SH |
632 | if (cl->level == level) |
633 | cl->xstats.lends++; | |
59e4220a | 634 | htb_accnt_tokens(cl, bytes, diff); |
1da177e4 LT |
635 | } else { |
636 | cl->xstats.borrows++; | |
87990467 | 637 | cl->tokens += diff; /* we moved t_c; update tokens */ |
1da177e4 | 638 | } |
59e4220a | 639 | htb_accnt_ctokens(cl, bytes, diff); |
1da177e4 | 640 | cl->t_c = q->now; |
1da177e4 | 641 | |
87990467 SH |
642 | old_mode = cl->cmode; |
643 | diff = 0; | |
644 | htb_change_class_mode(q, cl, &diff); | |
1da177e4 LT |
645 | if (old_mode != cl->cmode) { |
646 | if (old_mode != HTB_CAN_SEND) | |
3696f625 | 647 | htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level); |
1da177e4 | 648 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 649 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 650 | } |
1da177e4 | 651 | |
bfe0d029 ED |
652 | /* update basic stats except for leaves which are already updated */ |
653 | if (cl->level) | |
654 | bstats_update(&cl->bstats, skb); | |
655 | ||
1da177e4 LT |
656 | cl = cl->parent; |
657 | } | |
658 | } | |
659 | ||
660 | /** | |
661 | * htb_do_events - make mode changes to classes at the level | |
662 | * | |
fb983d45 | 663 | * Scans event queue for pending events and applies them. Returns time of |
1224736d | 664 | * next pending event (0 for no event in pq, q->now for too many events). |
fb983d45 | 665 | * Note: Applied are events whose have cl->pq_key <= q->now. |
1da177e4 | 666 | */ |
a73be040 JP |
667 | static psched_time_t htb_do_events(struct htb_sched *q, int level, |
668 | unsigned long start) | |
1da177e4 | 669 | { |
8f3ea33a | 670 | /* don't run for longer than 2 jiffies; 2 is used instead of |
cc7ec456 ED |
671 | * 1 to simplify things when jiffy is going to be incremented |
672 | * too soon | |
673 | */ | |
a73be040 | 674 | unsigned long stop_at = start + 2; |
8f3ea33a | 675 | while (time_before(jiffies, stop_at)) { |
1da177e4 | 676 | struct htb_class *cl; |
56b765b7 | 677 | s64 diff; |
30bdbe39 AM |
678 | struct rb_node *p = rb_first(&q->wait_pq[level]); |
679 | ||
87990467 SH |
680 | if (!p) |
681 | return 0; | |
1da177e4 LT |
682 | |
683 | cl = rb_entry(p, struct htb_class, pq_node); | |
fb983d45 PM |
684 | if (cl->pq_key > q->now) |
685 | return cl->pq_key; | |
686 | ||
3696f625 | 687 | htb_safe_rb_erase(p, q->wait_pq + level); |
56b765b7 | 688 | diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); |
87990467 | 689 | htb_change_class_mode(q, cl, &diff); |
1da177e4 | 690 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 691 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 692 | } |
1224736d JP |
693 | |
694 | /* too much load - let's continue after a break for scheduling */ | |
e82181de | 695 | if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) { |
cc7ec456 | 696 | pr_warning("htb: too many events!\n"); |
e82181de JP |
697 | q->warned |= HTB_WARN_TOOMANYEVENTS; |
698 | } | |
1224736d JP |
699 | |
700 | return q->now; | |
1da177e4 LT |
701 | } |
702 | ||
703 | /* Returns class->node+prio from id-tree where classe's id is >= id. NULL | |
cc7ec456 ED |
704 | * is no such one exists. |
705 | */ | |
87990467 SH |
706 | static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n, |
707 | u32 id) | |
1da177e4 LT |
708 | { |
709 | struct rb_node *r = NULL; | |
710 | while (n) { | |
87990467 SH |
711 | struct htb_class *cl = |
712 | rb_entry(n, struct htb_class, node[prio]); | |
87990467 | 713 | |
f4c1f3e0 | 714 | if (id > cl->common.classid) { |
1da177e4 | 715 | n = n->rb_right; |
1b5c0077 | 716 | } else if (id < cl->common.classid) { |
1da177e4 LT |
717 | r = n; |
718 | n = n->rb_left; | |
1b5c0077 JP |
719 | } else { |
720 | return n; | |
1da177e4 LT |
721 | } |
722 | } | |
723 | return r; | |
724 | } | |
725 | ||
726 | /** | |
727 | * htb_lookup_leaf - returns next leaf class in DRR order | |
728 | * | |
729 | * Find leaf where current feed pointers points to. | |
730 | */ | |
87990467 SH |
731 | static struct htb_class *htb_lookup_leaf(struct rb_root *tree, int prio, |
732 | struct rb_node **pptr, u32 * pid) | |
1da177e4 LT |
733 | { |
734 | int i; | |
735 | struct { | |
736 | struct rb_node *root; | |
737 | struct rb_node **pptr; | |
738 | u32 *pid; | |
87990467 SH |
739 | } stk[TC_HTB_MAXDEPTH], *sp = stk; |
740 | ||
512bb43e | 741 | BUG_ON(!tree->rb_node); |
1da177e4 LT |
742 | sp->root = tree->rb_node; |
743 | sp->pptr = pptr; | |
744 | sp->pid = pid; | |
745 | ||
746 | for (i = 0; i < 65535; i++) { | |
87990467 | 747 | if (!*sp->pptr && *sp->pid) { |
10297b99 | 748 | /* ptr was invalidated but id is valid - try to recover |
cc7ec456 ED |
749 | * the original or next ptr |
750 | */ | |
87990467 SH |
751 | *sp->pptr = |
752 | htb_id_find_next_upper(prio, sp->root, *sp->pid); | |
1da177e4 | 753 | } |
87990467 | 754 | *sp->pid = 0; /* ptr is valid now so that remove this hint as it |
cc7ec456 ED |
755 | * can become out of date quickly |
756 | */ | |
87990467 | 757 | if (!*sp->pptr) { /* we are at right end; rewind & go up */ |
1da177e4 | 758 | *sp->pptr = sp->root; |
87990467 | 759 | while ((*sp->pptr)->rb_left) |
1da177e4 LT |
760 | *sp->pptr = (*sp->pptr)->rb_left; |
761 | if (sp > stk) { | |
762 | sp--; | |
512bb43e JP |
763 | if (!*sp->pptr) { |
764 | WARN_ON(1); | |
87990467 | 765 | return NULL; |
512bb43e | 766 | } |
87990467 | 767 | htb_next_rb_node(sp->pptr); |
1da177e4 LT |
768 | } |
769 | } else { | |
770 | struct htb_class *cl; | |
87990467 SH |
771 | cl = rb_entry(*sp->pptr, struct htb_class, node[prio]); |
772 | if (!cl->level) | |
1da177e4 LT |
773 | return cl; |
774 | (++sp)->root = cl->un.inner.feed[prio].rb_node; | |
87990467 SH |
775 | sp->pptr = cl->un.inner.ptr + prio; |
776 | sp->pid = cl->un.inner.last_ptr_id + prio; | |
1da177e4 LT |
777 | } |
778 | } | |
547b792c | 779 | WARN_ON(1); |
1da177e4 LT |
780 | return NULL; |
781 | } | |
782 | ||
783 | /* dequeues packet at given priority and level; call only if | |
cc7ec456 ED |
784 | * you are sure that there is active class at prio/level |
785 | */ | |
87990467 SH |
786 | static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, int prio, |
787 | int level) | |
1da177e4 LT |
788 | { |
789 | struct sk_buff *skb = NULL; | |
87990467 | 790 | struct htb_class *cl, *start; |
1da177e4 | 791 | /* look initial class up in the row */ |
87990467 SH |
792 | start = cl = htb_lookup_leaf(q->row[level] + prio, prio, |
793 | q->ptr[level] + prio, | |
794 | q->last_ptr_id[level] + prio); | |
795 | ||
1da177e4 LT |
796 | do { |
797 | next: | |
512bb43e | 798 | if (unlikely(!cl)) |
87990467 | 799 | return NULL; |
1da177e4 LT |
800 | |
801 | /* class can be empty - it is unlikely but can be true if leaf | |
cc7ec456 ED |
802 | * qdisc drops packets in enqueue routine or if someone used |
803 | * graft operation on the leaf since last dequeue; | |
804 | * simply deactivate and skip such class | |
805 | */ | |
1da177e4 LT |
806 | if (unlikely(cl->un.leaf.q->q.qlen == 0)) { |
807 | struct htb_class *next; | |
87990467 | 808 | htb_deactivate(q, cl); |
1da177e4 LT |
809 | |
810 | /* row/level might become empty */ | |
811 | if ((q->row_mask[level] & (1 << prio)) == 0) | |
87990467 | 812 | return NULL; |
1da177e4 | 813 | |
87990467 SH |
814 | next = htb_lookup_leaf(q->row[level] + prio, |
815 | prio, q->ptr[level] + prio, | |
816 | q->last_ptr_id[level] + prio); | |
817 | ||
818 | if (cl == start) /* fix start if we just deleted it */ | |
1da177e4 LT |
819 | start = next; |
820 | cl = next; | |
821 | goto next; | |
822 | } | |
87990467 SH |
823 | |
824 | skb = cl->un.leaf.q->dequeue(cl->un.leaf.q); | |
825 | if (likely(skb != NULL)) | |
1da177e4 | 826 | break; |
633fe66e | 827 | |
b00355db | 828 | qdisc_warn_nonwc("htb", cl->un.leaf.q); |
87990467 SH |
829 | htb_next_rb_node((level ? cl->parent->un.inner.ptr : q-> |
830 | ptr[0]) + prio); | |
831 | cl = htb_lookup_leaf(q->row[level] + prio, prio, | |
832 | q->ptr[level] + prio, | |
833 | q->last_ptr_id[level] + prio); | |
1da177e4 LT |
834 | |
835 | } while (cl != start); | |
836 | ||
837 | if (likely(skb != NULL)) { | |
196d97f6 | 838 | bstats_update(&cl->bstats, skb); |
0abf77e5 JK |
839 | cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb); |
840 | if (cl->un.leaf.deficit[level] < 0) { | |
c19f7a34 | 841 | cl->un.leaf.deficit[level] += cl->quantum; |
87990467 SH |
842 | htb_next_rb_node((level ? cl->parent->un.inner.ptr : q-> |
843 | ptr[0]) + prio); | |
1da177e4 LT |
844 | } |
845 | /* this used to be after charge_class but this constelation | |
cc7ec456 ED |
846 | * gives us slightly better performance |
847 | */ | |
1da177e4 | 848 | if (!cl->un.leaf.q->q.qlen) |
87990467 | 849 | htb_deactivate(q, cl); |
c9726d68 | 850 | htb_charge_class(q, cl, level, skb); |
1da177e4 LT |
851 | } |
852 | return skb; | |
853 | } | |
854 | ||
1da177e4 LT |
855 | static struct sk_buff *htb_dequeue(struct Qdisc *sch) |
856 | { | |
9190b3b3 | 857 | struct sk_buff *skb; |
1da177e4 LT |
858 | struct htb_sched *q = qdisc_priv(sch); |
859 | int level; | |
fb983d45 | 860 | psched_time_t next_event; |
a73be040 | 861 | unsigned long start_at; |
1da177e4 LT |
862 | |
863 | /* try to dequeue direct packets as high prio (!) to minimize cpu work */ | |
87990467 SH |
864 | skb = __skb_dequeue(&q->direct_queue); |
865 | if (skb != NULL) { | |
9190b3b3 ED |
866 | ok: |
867 | qdisc_bstats_update(sch, skb); | |
fd245a4a | 868 | qdisc_unthrottled(sch); |
1da177e4 LT |
869 | sch->q.qlen--; |
870 | return skb; | |
871 | } | |
872 | ||
87990467 SH |
873 | if (!sch->q.qlen) |
874 | goto fin; | |
56b765b7 | 875 | q->now = ktime_to_ns(ktime_get()); |
a73be040 | 876 | start_at = jiffies; |
1da177e4 | 877 | |
d2fe85da | 878 | next_event = q->now + 5LLU * NSEC_PER_SEC; |
633fe66e | 879 | |
1da177e4 LT |
880 | for (level = 0; level < TC_HTB_MAXDEPTH; level++) { |
881 | /* common case optimization - skip event handler quickly */ | |
882 | int m; | |
fb983d45 PM |
883 | psched_time_t event; |
884 | ||
885 | if (q->now >= q->near_ev_cache[level]) { | |
a73be040 | 886 | event = htb_do_events(q, level, start_at); |
2e4b3b0e | 887 | if (!event) |
56b765b7 | 888 | event = q->now + NSEC_PER_SEC; |
2e4b3b0e | 889 | q->near_ev_cache[level] = event; |
1da177e4 | 890 | } else |
fb983d45 PM |
891 | event = q->near_ev_cache[level]; |
892 | ||
c0851347 | 893 | if (next_event > event) |
fb983d45 | 894 | next_event = event; |
87990467 | 895 | |
1da177e4 LT |
896 | m = ~q->row_mask[level]; |
897 | while (m != (int)(-1)) { | |
87990467 | 898 | int prio = ffz(m); |
cc7ec456 | 899 | |
1da177e4 | 900 | m |= 1 << prio; |
87990467 | 901 | skb = htb_dequeue_tree(q, prio, level); |
9190b3b3 ED |
902 | if (likely(skb != NULL)) |
903 | goto ok; | |
1da177e4 LT |
904 | } |
905 | } | |
fb983d45 | 906 | sch->qstats.overlimits++; |
56b765b7 V |
907 | if (likely(next_event > q->now)) { |
908 | if (!test_bit(__QDISC_STATE_DEACTIVATED, | |
909 | &qdisc_root_sleeping(q->watchdog.qdisc)->state)) { | |
910 | ktime_t time = ns_to_ktime(next_event); | |
911 | qdisc_throttled(q->watchdog.qdisc); | |
912 | hrtimer_start(&q->watchdog.timer, time, | |
913 | HRTIMER_MODE_ABS); | |
914 | } | |
915 | } else { | |
1224736d | 916 | schedule_work(&q->work); |
56b765b7 | 917 | } |
1da177e4 | 918 | fin: |
1da177e4 LT |
919 | return skb; |
920 | } | |
921 | ||
922 | /* try to drop from each class (by prio) until one succeed */ | |
87990467 | 923 | static unsigned int htb_drop(struct Qdisc *sch) |
1da177e4 LT |
924 | { |
925 | struct htb_sched *q = qdisc_priv(sch); | |
926 | int prio; | |
927 | ||
928 | for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) { | |
929 | struct list_head *p; | |
87990467 | 930 | list_for_each(p, q->drops + prio) { |
1da177e4 LT |
931 | struct htb_class *cl = list_entry(p, struct htb_class, |
932 | un.leaf.drop_list); | |
933 | unsigned int len; | |
87990467 SH |
934 | if (cl->un.leaf.q->ops->drop && |
935 | (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) { | |
1da177e4 LT |
936 | sch->q.qlen--; |
937 | if (!cl->un.leaf.q->q.qlen) | |
87990467 | 938 | htb_deactivate(q, cl); |
1da177e4 LT |
939 | return len; |
940 | } | |
941 | } | |
942 | } | |
943 | return 0; | |
944 | } | |
945 | ||
946 | /* reset all classes */ | |
947 | /* always caled under BH & queue lock */ | |
87990467 | 948 | static void htb_reset(struct Qdisc *sch) |
1da177e4 LT |
949 | { |
950 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 PM |
951 | struct htb_class *cl; |
952 | struct hlist_node *n; | |
953 | unsigned int i; | |
0cef296d | 954 | |
f4c1f3e0 PM |
955 | for (i = 0; i < q->clhash.hashsize; i++) { |
956 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { | |
1da177e4 | 957 | if (cl->level) |
87990467 | 958 | memset(&cl->un.inner, 0, sizeof(cl->un.inner)); |
1da177e4 | 959 | else { |
87990467 | 960 | if (cl->un.leaf.q) |
1da177e4 LT |
961 | qdisc_reset(cl->un.leaf.q); |
962 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); | |
963 | } | |
964 | cl->prio_activity = 0; | |
965 | cl->cmode = HTB_CAN_SEND; | |
1da177e4 LT |
966 | |
967 | } | |
968 | } | |
fb983d45 | 969 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
970 | __skb_queue_purge(&q->direct_queue); |
971 | sch->q.qlen = 0; | |
87990467 SH |
972 | memset(q->row, 0, sizeof(q->row)); |
973 | memset(q->row_mask, 0, sizeof(q->row_mask)); | |
974 | memset(q->wait_pq, 0, sizeof(q->wait_pq)); | |
975 | memset(q->ptr, 0, sizeof(q->ptr)); | |
1da177e4 | 976 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 977 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 LT |
978 | } |
979 | ||
27a3421e PM |
980 | static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = { |
981 | [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) }, | |
982 | [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) }, | |
983 | [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
984 | [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
985 | }; | |
986 | ||
1224736d JP |
987 | static void htb_work_func(struct work_struct *work) |
988 | { | |
989 | struct htb_sched *q = container_of(work, struct htb_sched, work); | |
990 | struct Qdisc *sch = q->watchdog.qdisc; | |
991 | ||
992 | __netif_schedule(qdisc_root(sch)); | |
993 | } | |
994 | ||
1e90474c | 995 | static int htb_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
996 | { |
997 | struct htb_sched *q = qdisc_priv(sch); | |
1e90474c | 998 | struct nlattr *tb[TCA_HTB_INIT + 1]; |
1da177e4 | 999 | struct tc_htb_glob *gopt; |
cee63723 | 1000 | int err; |
1da177e4 | 1001 | int i; |
cee63723 PM |
1002 | |
1003 | if (!opt) | |
1004 | return -EINVAL; | |
1005 | ||
27a3421e | 1006 | err = nla_parse_nested(tb, TCA_HTB_INIT, opt, htb_policy); |
cee63723 PM |
1007 | if (err < 0) |
1008 | return err; | |
1009 | ||
27a3421e | 1010 | if (tb[TCA_HTB_INIT] == NULL) { |
cc7ec456 | 1011 | pr_err("HTB: hey probably you have bad tc tool ?\n"); |
1da177e4 LT |
1012 | return -EINVAL; |
1013 | } | |
1e90474c | 1014 | gopt = nla_data(tb[TCA_HTB_INIT]); |
1da177e4 | 1015 | if (gopt->version != HTB_VER >> 16) { |
cc7ec456 | 1016 | pr_err("HTB: need tc/htb version %d (minor is %d), you have %d\n", |
87990467 | 1017 | HTB_VER >> 16, HTB_VER & 0xffff, gopt->version); |
1da177e4 LT |
1018 | return -EINVAL; |
1019 | } | |
1da177e4 | 1020 | |
f4c1f3e0 PM |
1021 | err = qdisc_class_hash_init(&q->clhash); |
1022 | if (err < 0) | |
1023 | return err; | |
1da177e4 | 1024 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 1025 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 | 1026 | |
fb983d45 | 1027 | qdisc_watchdog_init(&q->watchdog, sch); |
1224736d | 1028 | INIT_WORK(&q->work, htb_work_func); |
1da177e4 LT |
1029 | skb_queue_head_init(&q->direct_queue); |
1030 | ||
5ce2d488 | 1031 | q->direct_qlen = qdisc_dev(sch)->tx_queue_len; |
87990467 | 1032 | if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */ |
1da177e4 | 1033 | q->direct_qlen = 2; |
1da177e4 | 1034 | |
1da177e4 LT |
1035 | if ((q->rate2quantum = gopt->rate2quantum) < 1) |
1036 | q->rate2quantum = 1; | |
1037 | q->defcls = gopt->defcls; | |
1038 | ||
1039 | return 0; | |
1040 | } | |
1041 | ||
1042 | static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) | |
1043 | { | |
102396ae | 1044 | spinlock_t *root_lock = qdisc_root_sleeping_lock(sch); |
1da177e4 | 1045 | struct htb_sched *q = qdisc_priv(sch); |
4b3550ef | 1046 | struct nlattr *nest; |
1da177e4 | 1047 | struct tc_htb_glob gopt; |
4b3550ef | 1048 | |
7698b4fc | 1049 | spin_lock_bh(root_lock); |
1da177e4 | 1050 | |
4b3550ef | 1051 | gopt.direct_pkts = q->direct_pkts; |
1da177e4 LT |
1052 | gopt.version = HTB_VER; |
1053 | gopt.rate2quantum = q->rate2quantum; | |
1054 | gopt.defcls = q->defcls; | |
3bf72957 | 1055 | gopt.debug = 0; |
4b3550ef PM |
1056 | |
1057 | nest = nla_nest_start(skb, TCA_OPTIONS); | |
1058 | if (nest == NULL) | |
1059 | goto nla_put_failure; | |
1b34ec43 DM |
1060 | if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt)) |
1061 | goto nla_put_failure; | |
4b3550ef PM |
1062 | nla_nest_end(skb, nest); |
1063 | ||
7698b4fc | 1064 | spin_unlock_bh(root_lock); |
1da177e4 | 1065 | return skb->len; |
4b3550ef | 1066 | |
1e90474c | 1067 | nla_put_failure: |
7698b4fc | 1068 | spin_unlock_bh(root_lock); |
4b3550ef | 1069 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1070 | return -1; |
1071 | } | |
1072 | ||
1073 | static int htb_dump_class(struct Qdisc *sch, unsigned long arg, | |
87990467 | 1074 | struct sk_buff *skb, struct tcmsg *tcm) |
1da177e4 | 1075 | { |
87990467 | 1076 | struct htb_class *cl = (struct htb_class *)arg; |
102396ae | 1077 | spinlock_t *root_lock = qdisc_root_sleeping_lock(sch); |
4b3550ef | 1078 | struct nlattr *nest; |
1da177e4 LT |
1079 | struct tc_htb_opt opt; |
1080 | ||
7698b4fc | 1081 | spin_lock_bh(root_lock); |
f4c1f3e0 PM |
1082 | tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT; |
1083 | tcm->tcm_handle = cl->common.classid; | |
1da177e4 LT |
1084 | if (!cl->level && cl->un.leaf.q) |
1085 | tcm->tcm_info = cl->un.leaf.q->handle; | |
1086 | ||
4b3550ef PM |
1087 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1088 | if (nest == NULL) | |
1089 | goto nla_put_failure; | |
1da177e4 | 1090 | |
87990467 | 1091 | memset(&opt, 0, sizeof(opt)); |
1da177e4 | 1092 | |
292f1c7f | 1093 | opt.rate.rate = psched_ratecfg_getrate(&cl->rate); |
9c10f411 | 1094 | opt.buffer = PSCHED_NS2TICKS(cl->buffer); |
292f1c7f | 1095 | opt.ceil.rate = psched_ratecfg_getrate(&cl->ceil); |
9c10f411 | 1096 | opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer); |
c19f7a34 JP |
1097 | opt.quantum = cl->quantum; |
1098 | opt.prio = cl->prio; | |
87990467 | 1099 | opt.level = cl->level; |
1b34ec43 DM |
1100 | if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt)) |
1101 | goto nla_put_failure; | |
4b3550ef PM |
1102 | |
1103 | nla_nest_end(skb, nest); | |
7698b4fc | 1104 | spin_unlock_bh(root_lock); |
1da177e4 | 1105 | return skb->len; |
4b3550ef | 1106 | |
1e90474c | 1107 | nla_put_failure: |
7698b4fc | 1108 | spin_unlock_bh(root_lock); |
4b3550ef | 1109 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1110 | return -1; |
1111 | } | |
1112 | ||
1113 | static int | |
87990467 | 1114 | htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d) |
1da177e4 | 1115 | { |
87990467 | 1116 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 | 1117 | |
1da177e4 LT |
1118 | if (!cl->level && cl->un.leaf.q) |
1119 | cl->qstats.qlen = cl->un.leaf.q->q.qlen; | |
1120 | cl->xstats.tokens = cl->tokens; | |
1121 | cl->xstats.ctokens = cl->ctokens; | |
1122 | ||
1123 | if (gnet_stats_copy_basic(d, &cl->bstats) < 0 || | |
d250a5f9 | 1124 | gnet_stats_copy_rate_est(d, NULL, &cl->rate_est) < 0 || |
1da177e4 LT |
1125 | gnet_stats_copy_queue(d, &cl->qstats) < 0) |
1126 | return -1; | |
1127 | ||
1128 | return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); | |
1129 | } | |
1130 | ||
1131 | static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
87990467 | 1132 | struct Qdisc **old) |
1da177e4 | 1133 | { |
87990467 | 1134 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 | 1135 | |
5b9a9ccf PM |
1136 | if (cl->level) |
1137 | return -EINVAL; | |
1138 | if (new == NULL && | |
3511c913 | 1139 | (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
5b9a9ccf PM |
1140 | cl->common.classid)) == NULL) |
1141 | return -ENOBUFS; | |
1142 | ||
1143 | sch_tree_lock(sch); | |
1144 | *old = cl->un.leaf.q; | |
1145 | cl->un.leaf.q = new; | |
1146 | if (*old != NULL) { | |
1147 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); | |
1148 | qdisc_reset(*old); | |
1da177e4 | 1149 | } |
5b9a9ccf PM |
1150 | sch_tree_unlock(sch); |
1151 | return 0; | |
1da177e4 LT |
1152 | } |
1153 | ||
87990467 | 1154 | static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg) |
1da177e4 | 1155 | { |
87990467 | 1156 | struct htb_class *cl = (struct htb_class *)arg; |
5b9a9ccf | 1157 | return !cl->level ? cl->un.leaf.q : NULL; |
1da177e4 LT |
1158 | } |
1159 | ||
256d61b8 PM |
1160 | static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg) |
1161 | { | |
1162 | struct htb_class *cl = (struct htb_class *)arg; | |
1163 | ||
1164 | if (cl->un.leaf.q->q.qlen == 0) | |
1165 | htb_deactivate(qdisc_priv(sch), cl); | |
1166 | } | |
1167 | ||
1da177e4 LT |
1168 | static unsigned long htb_get(struct Qdisc *sch, u32 classid) |
1169 | { | |
87990467 SH |
1170 | struct htb_class *cl = htb_find(classid, sch); |
1171 | if (cl) | |
1da177e4 LT |
1172 | cl->refcnt++; |
1173 | return (unsigned long)cl; | |
1174 | } | |
1175 | ||
160d5e10 JP |
1176 | static inline int htb_parent_last_child(struct htb_class *cl) |
1177 | { | |
1178 | if (!cl->parent) | |
1179 | /* the root class */ | |
1180 | return 0; | |
42077599 | 1181 | if (cl->parent->children > 1) |
160d5e10 JP |
1182 | /* not the last child */ |
1183 | return 0; | |
160d5e10 JP |
1184 | return 1; |
1185 | } | |
1186 | ||
3ba08b00 JP |
1187 | static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl, |
1188 | struct Qdisc *new_q) | |
160d5e10 JP |
1189 | { |
1190 | struct htb_class *parent = cl->parent; | |
1191 | ||
547b792c | 1192 | WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity); |
160d5e10 | 1193 | |
3ba08b00 JP |
1194 | if (parent->cmode != HTB_CAN_SEND) |
1195 | htb_safe_rb_erase(&parent->pq_node, q->wait_pq + parent->level); | |
1196 | ||
160d5e10 JP |
1197 | parent->level = 0; |
1198 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1199 | INIT_LIST_HEAD(&parent->un.leaf.drop_list); | |
1200 | parent->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
160d5e10 JP |
1201 | parent->tokens = parent->buffer; |
1202 | parent->ctokens = parent->cbuffer; | |
3bebcda2 | 1203 | parent->t_c = psched_get_time(); |
160d5e10 JP |
1204 | parent->cmode = HTB_CAN_SEND; |
1205 | } | |
1206 | ||
87990467 | 1207 | static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl) |
1da177e4 | 1208 | { |
1da177e4 | 1209 | if (!cl->level) { |
547b792c | 1210 | WARN_ON(!cl->un.leaf.q); |
1da177e4 LT |
1211 | qdisc_destroy(cl->un.leaf.q); |
1212 | } | |
ee39e10c | 1213 | gen_kill_estimator(&cl->bstats, &cl->rate_est); |
ff31ab56 | 1214 | tcf_destroy_chain(&cl->filter_list); |
1da177e4 LT |
1215 | kfree(cl); |
1216 | } | |
1217 | ||
87990467 | 1218 | static void htb_destroy(struct Qdisc *sch) |
1da177e4 LT |
1219 | { |
1220 | struct htb_sched *q = qdisc_priv(sch); | |
fbd8f137 PM |
1221 | struct hlist_node *n, *next; |
1222 | struct htb_class *cl; | |
1223 | unsigned int i; | |
1da177e4 | 1224 | |
1224736d | 1225 | cancel_work_sync(&q->work); |
fb983d45 | 1226 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 | 1227 | /* This line used to be after htb_destroy_class call below |
cc7ec456 ED |
1228 | * and surprisingly it worked in 2.4. But it must precede it |
1229 | * because filter need its target class alive to be able to call | |
1230 | * unbind_filter on it (without Oops). | |
1231 | */ | |
ff31ab56 | 1232 | tcf_destroy_chain(&q->filter_list); |
87990467 | 1233 | |
f4c1f3e0 PM |
1234 | for (i = 0; i < q->clhash.hashsize; i++) { |
1235 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) | |
fbd8f137 PM |
1236 | tcf_destroy_chain(&cl->filter_list); |
1237 | } | |
f4c1f3e0 PM |
1238 | for (i = 0; i < q->clhash.hashsize; i++) { |
1239 | hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i], | |
1240 | common.hnode) | |
fbd8f137 PM |
1241 | htb_destroy_class(sch, cl); |
1242 | } | |
f4c1f3e0 | 1243 | qdisc_class_hash_destroy(&q->clhash); |
1da177e4 LT |
1244 | __skb_queue_purge(&q->direct_queue); |
1245 | } | |
1246 | ||
1247 | static int htb_delete(struct Qdisc *sch, unsigned long arg) | |
1248 | { | |
1249 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1250 | struct htb_class *cl = (struct htb_class *)arg; |
256d61b8 | 1251 | unsigned int qlen; |
160d5e10 JP |
1252 | struct Qdisc *new_q = NULL; |
1253 | int last_child = 0; | |
1da177e4 LT |
1254 | |
1255 | // TODO: why don't allow to delete subtree ? references ? does | |
1256 | // tc subsys quarantee us that in htb_destroy it holds no class | |
1257 | // refs so that we can remove children safely there ? | |
42077599 | 1258 | if (cl->children || cl->filter_cnt) |
1da177e4 | 1259 | return -EBUSY; |
87990467 | 1260 | |
160d5e10 | 1261 | if (!cl->level && htb_parent_last_child(cl)) { |
3511c913 | 1262 | new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
bb949fbd | 1263 | cl->parent->common.classid); |
160d5e10 JP |
1264 | last_child = 1; |
1265 | } | |
1266 | ||
1da177e4 | 1267 | sch_tree_lock(sch); |
87990467 | 1268 | |
814a175e | 1269 | if (!cl->level) { |
256d61b8 | 1270 | qlen = cl->un.leaf.q->q.qlen; |
814a175e | 1271 | qdisc_reset(cl->un.leaf.q); |
256d61b8 | 1272 | qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen); |
814a175e PM |
1273 | } |
1274 | ||
f4c1f3e0 PM |
1275 | /* delete from hash and active; remainder in destroy_class */ |
1276 | qdisc_class_hash_remove(&q->clhash, &cl->common); | |
26b284de JP |
1277 | if (cl->parent) |
1278 | cl->parent->children--; | |
c38c83cb | 1279 | |
1da177e4 | 1280 | if (cl->prio_activity) |
87990467 | 1281 | htb_deactivate(q, cl); |
1da177e4 | 1282 | |
fbd8f137 PM |
1283 | if (cl->cmode != HTB_CAN_SEND) |
1284 | htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level); | |
1285 | ||
160d5e10 | 1286 | if (last_child) |
3ba08b00 | 1287 | htb_parent_to_leaf(q, cl, new_q); |
160d5e10 | 1288 | |
7cd0a638 JP |
1289 | BUG_ON(--cl->refcnt == 0); |
1290 | /* | |
1291 | * This shouldn't happen: we "hold" one cops->get() when called | |
1292 | * from tc_ctl_tclass; the destroy method is done from cops->put(). | |
1293 | */ | |
1da177e4 LT |
1294 | |
1295 | sch_tree_unlock(sch); | |
1296 | return 0; | |
1297 | } | |
1298 | ||
1299 | static void htb_put(struct Qdisc *sch, unsigned long arg) | |
1300 | { | |
87990467 | 1301 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1302 | |
1303 | if (--cl->refcnt == 0) | |
87990467 | 1304 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1305 | } |
1306 | ||
87990467 | 1307 | static int htb_change_class(struct Qdisc *sch, u32 classid, |
1e90474c | 1308 | u32 parentid, struct nlattr **tca, |
87990467 | 1309 | unsigned long *arg) |
1da177e4 LT |
1310 | { |
1311 | int err = -EINVAL; | |
1312 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1313 | struct htb_class *cl = (struct htb_class *)*arg, *parent; |
1e90474c | 1314 | struct nlattr *opt = tca[TCA_OPTIONS]; |
e18434c4 | 1315 | struct nlattr *tb[__TCA_HTB_MAX]; |
1da177e4 LT |
1316 | struct tc_htb_opt *hopt; |
1317 | ||
1318 | /* extract all subattrs from opt attr */ | |
cee63723 PM |
1319 | if (!opt) |
1320 | goto failure; | |
1321 | ||
e18434c4 | 1322 | err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy); |
cee63723 PM |
1323 | if (err < 0) |
1324 | goto failure; | |
1325 | ||
1326 | err = -EINVAL; | |
27a3421e | 1327 | if (tb[TCA_HTB_PARMS] == NULL) |
1da177e4 | 1328 | goto failure; |
1da177e4 | 1329 | |
87990467 SH |
1330 | parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch); |
1331 | ||
1e90474c | 1332 | hopt = nla_data(tb[TCA_HTB_PARMS]); |
196d97f6 | 1333 | if (!hopt->rate.rate || !hopt->ceil.rate) |
87990467 | 1334 | goto failure; |
1da177e4 | 1335 | |
87990467 | 1336 | if (!cl) { /* new class */ |
1da177e4 | 1337 | struct Qdisc *new_q; |
3696f625 | 1338 | int prio; |
ee39e10c | 1339 | struct { |
1e90474c | 1340 | struct nlattr nla; |
ee39e10c PM |
1341 | struct gnet_estimator opt; |
1342 | } est = { | |
1e90474c PM |
1343 | .nla = { |
1344 | .nla_len = nla_attr_size(sizeof(est.opt)), | |
1345 | .nla_type = TCA_RATE, | |
ee39e10c PM |
1346 | }, |
1347 | .opt = { | |
1348 | /* 4s interval, 16s averaging constant */ | |
1349 | .interval = 2, | |
1350 | .ewma_log = 2, | |
1351 | }, | |
1352 | }; | |
3696f625 | 1353 | |
1da177e4 | 1354 | /* check for valid classid */ |
f64f9e71 JP |
1355 | if (!classid || TC_H_MAJ(classid ^ sch->handle) || |
1356 | htb_find(classid, sch)) | |
1da177e4 LT |
1357 | goto failure; |
1358 | ||
1359 | /* check maximal depth */ | |
1360 | if (parent && parent->parent && parent->parent->level < 2) { | |
cc7ec456 | 1361 | pr_err("htb: tree is too deep\n"); |
1da177e4 LT |
1362 | goto failure; |
1363 | } | |
1364 | err = -ENOBUFS; | |
cc7ec456 ED |
1365 | cl = kzalloc(sizeof(*cl), GFP_KERNEL); |
1366 | if (!cl) | |
1da177e4 | 1367 | goto failure; |
87990467 | 1368 | |
71bcb09a SH |
1369 | err = gen_new_estimator(&cl->bstats, &cl->rate_est, |
1370 | qdisc_root_sleeping_lock(sch), | |
1371 | tca[TCA_RATE] ? : &est.nla); | |
1372 | if (err) { | |
1373 | kfree(cl); | |
1374 | goto failure; | |
1375 | } | |
1376 | ||
1da177e4 | 1377 | cl->refcnt = 1; |
42077599 | 1378 | cl->children = 0; |
1da177e4 | 1379 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); |
3696f625 SH |
1380 | RB_CLEAR_NODE(&cl->pq_node); |
1381 | ||
1382 | for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) | |
1383 | RB_CLEAR_NODE(&cl->node[prio]); | |
1da177e4 LT |
1384 | |
1385 | /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL) | |
cc7ec456 ED |
1386 | * so that can't be used inside of sch_tree_lock |
1387 | * -- thanks to Karlis Peisenieks | |
1388 | */ | |
3511c913 | 1389 | new_q = qdisc_create_dflt(sch->dev_queue, |
bb949fbd | 1390 | &pfifo_qdisc_ops, classid); |
1da177e4 LT |
1391 | sch_tree_lock(sch); |
1392 | if (parent && !parent->level) { | |
256d61b8 PM |
1393 | unsigned int qlen = parent->un.leaf.q->q.qlen; |
1394 | ||
1da177e4 | 1395 | /* turn parent into inner node */ |
256d61b8 PM |
1396 | qdisc_reset(parent->un.leaf.q); |
1397 | qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen); | |
87990467 SH |
1398 | qdisc_destroy(parent->un.leaf.q); |
1399 | if (parent->prio_activity) | |
1400 | htb_deactivate(q, parent); | |
1da177e4 LT |
1401 | |
1402 | /* remove from evt list because of level change */ | |
1403 | if (parent->cmode != HTB_CAN_SEND) { | |
3696f625 | 1404 | htb_safe_rb_erase(&parent->pq_node, q->wait_pq); |
1da177e4 LT |
1405 | parent->cmode = HTB_CAN_SEND; |
1406 | } | |
1407 | parent->level = (parent->parent ? parent->parent->level | |
87990467 SH |
1408 | : TC_HTB_MAXDEPTH) - 1; |
1409 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1da177e4 LT |
1410 | } |
1411 | /* leaf (we) needs elementary qdisc */ | |
1412 | cl->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
1413 | ||
f4c1f3e0 | 1414 | cl->common.classid = classid; |
87990467 | 1415 | cl->parent = parent; |
1da177e4 LT |
1416 | |
1417 | /* set class to be in HTB_CAN_SEND state */ | |
b9a7afde JP |
1418 | cl->tokens = PSCHED_TICKS2NS(hopt->buffer); |
1419 | cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer); | |
00c04af9 | 1420 | cl->mbuffer = 60 * PSCHED_TICKS_PER_SEC; /* 1min */ |
3bebcda2 | 1421 | cl->t_c = psched_get_time(); |
1da177e4 LT |
1422 | cl->cmode = HTB_CAN_SEND; |
1423 | ||
1424 | /* attach to the hash list and parent's family */ | |
f4c1f3e0 | 1425 | qdisc_class_hash_insert(&q->clhash, &cl->common); |
42077599 PM |
1426 | if (parent) |
1427 | parent->children++; | |
ee39e10c | 1428 | } else { |
71bcb09a SH |
1429 | if (tca[TCA_RATE]) { |
1430 | err = gen_replace_estimator(&cl->bstats, &cl->rate_est, | |
1431 | qdisc_root_sleeping_lock(sch), | |
1432 | tca[TCA_RATE]); | |
1433 | if (err) | |
1434 | return err; | |
1435 | } | |
87990467 | 1436 | sch_tree_lock(sch); |
ee39e10c | 1437 | } |
1da177e4 LT |
1438 | |
1439 | /* it used to be a nasty bug here, we have to check that node | |
cc7ec456 ED |
1440 | * is really leaf before changing cl->un.leaf ! |
1441 | */ | |
1da177e4 | 1442 | if (!cl->level) { |
196d97f6 | 1443 | cl->quantum = hopt->rate.rate / q->rate2quantum; |
c19f7a34 | 1444 | if (!hopt->quantum && cl->quantum < 1000) { |
cc7ec456 | 1445 | pr_warning( |
87990467 | 1446 | "HTB: quantum of class %X is small. Consider r2q change.\n", |
f4c1f3e0 | 1447 | cl->common.classid); |
c19f7a34 | 1448 | cl->quantum = 1000; |
1da177e4 | 1449 | } |
c19f7a34 | 1450 | if (!hopt->quantum && cl->quantum > 200000) { |
cc7ec456 | 1451 | pr_warning( |
87990467 | 1452 | "HTB: quantum of class %X is big. Consider r2q change.\n", |
f4c1f3e0 | 1453 | cl->common.classid); |
c19f7a34 | 1454 | cl->quantum = 200000; |
1da177e4 LT |
1455 | } |
1456 | if (hopt->quantum) | |
c19f7a34 JP |
1457 | cl->quantum = hopt->quantum; |
1458 | if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO) | |
1459 | cl->prio = TC_HTB_NUMPRIO - 1; | |
1da177e4 LT |
1460 | } |
1461 | ||
292f1c7f JP |
1462 | psched_ratecfg_precompute(&cl->rate, hopt->rate.rate); |
1463 | psched_ratecfg_precompute(&cl->ceil, hopt->ceil.rate); | |
56b765b7 | 1464 | |
324f5aa5 JP |
1465 | cl->buffer = PSCHED_TICKS2NS(hopt->buffer); |
1466 | cl->cbuffer = PSCHED_TICKS2NS(hopt->buffer); | |
56b765b7 | 1467 | |
1da177e4 LT |
1468 | sch_tree_unlock(sch); |
1469 | ||
f4c1f3e0 PM |
1470 | qdisc_class_hash_grow(sch, &q->clhash); |
1471 | ||
1da177e4 LT |
1472 | *arg = (unsigned long)cl; |
1473 | return 0; | |
1474 | ||
1475 | failure: | |
1da177e4 LT |
1476 | return err; |
1477 | } | |
1478 | ||
1479 | static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg) | |
1480 | { | |
1481 | struct htb_sched *q = qdisc_priv(sch); | |
1482 | struct htb_class *cl = (struct htb_class *)arg; | |
1483 | struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list; | |
3bf72957 | 1484 | |
1da177e4 LT |
1485 | return fl; |
1486 | } | |
1487 | ||
1488 | static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, | |
87990467 | 1489 | u32 classid) |
1da177e4 | 1490 | { |
87990467 | 1491 | struct htb_class *cl = htb_find(classid, sch); |
3bf72957 | 1492 | |
1da177e4 | 1493 | /*if (cl && !cl->level) return 0; |
cc7ec456 ED |
1494 | * The line above used to be there to prevent attaching filters to |
1495 | * leaves. But at least tc_index filter uses this just to get class | |
1496 | * for other reasons so that we have to allow for it. | |
1497 | * ---- | |
1498 | * 19.6.2002 As Werner explained it is ok - bind filter is just | |
1499 | * another way to "lock" the class - unlike "get" this lock can | |
1500 | * be broken by class during destroy IIUC. | |
1da177e4 | 1501 | */ |
87990467 SH |
1502 | if (cl) |
1503 | cl->filter_cnt++; | |
1da177e4 LT |
1504 | return (unsigned long)cl; |
1505 | } | |
1506 | ||
1507 | static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) | |
1508 | { | |
1da177e4 | 1509 | struct htb_class *cl = (struct htb_class *)arg; |
3bf72957 | 1510 | |
87990467 SH |
1511 | if (cl) |
1512 | cl->filter_cnt--; | |
1da177e4 LT |
1513 | } |
1514 | ||
1515 | static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
1516 | { | |
1517 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 PM |
1518 | struct htb_class *cl; |
1519 | struct hlist_node *n; | |
1520 | unsigned int i; | |
1da177e4 LT |
1521 | |
1522 | if (arg->stop) | |
1523 | return; | |
1524 | ||
f4c1f3e0 PM |
1525 | for (i = 0; i < q->clhash.hashsize; i++) { |
1526 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { | |
1da177e4 LT |
1527 | if (arg->count < arg->skip) { |
1528 | arg->count++; | |
1529 | continue; | |
1530 | } | |
1531 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
1532 | arg->stop = 1; | |
1533 | return; | |
1534 | } | |
1535 | arg->count++; | |
1536 | } | |
1537 | } | |
1538 | } | |
1539 | ||
20fea08b | 1540 | static const struct Qdisc_class_ops htb_class_ops = { |
1da177e4 LT |
1541 | .graft = htb_graft, |
1542 | .leaf = htb_leaf, | |
256d61b8 | 1543 | .qlen_notify = htb_qlen_notify, |
1da177e4 LT |
1544 | .get = htb_get, |
1545 | .put = htb_put, | |
1546 | .change = htb_change_class, | |
1547 | .delete = htb_delete, | |
1548 | .walk = htb_walk, | |
1549 | .tcf_chain = htb_find_tcf, | |
1550 | .bind_tcf = htb_bind_filter, | |
1551 | .unbind_tcf = htb_unbind_filter, | |
1552 | .dump = htb_dump_class, | |
1553 | .dump_stats = htb_dump_class_stats, | |
1554 | }; | |
1555 | ||
20fea08b | 1556 | static struct Qdisc_ops htb_qdisc_ops __read_mostly = { |
1da177e4 LT |
1557 | .cl_ops = &htb_class_ops, |
1558 | .id = "htb", | |
1559 | .priv_size = sizeof(struct htb_sched), | |
1560 | .enqueue = htb_enqueue, | |
1561 | .dequeue = htb_dequeue, | |
77be155c | 1562 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1563 | .drop = htb_drop, |
1564 | .init = htb_init, | |
1565 | .reset = htb_reset, | |
1566 | .destroy = htb_destroy, | |
1da177e4 LT |
1567 | .dump = htb_dump, |
1568 | .owner = THIS_MODULE, | |
1569 | }; | |
1570 | ||
1571 | static int __init htb_module_init(void) | |
1572 | { | |
87990467 | 1573 | return register_qdisc(&htb_qdisc_ops); |
1da177e4 | 1574 | } |
87990467 | 1575 | static void __exit htb_module_exit(void) |
1da177e4 | 1576 | { |
87990467 | 1577 | unregister_qdisc(&htb_qdisc_ops); |
1da177e4 | 1578 | } |
87990467 | 1579 | |
1da177e4 LT |
1580 | module_init(htb_module_init) |
1581 | module_exit(htb_module_exit) | |
1582 | MODULE_LICENSE("GPL"); |