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 | ||
64153ce0 ED |
68 | static int htb_rate_est = 0; /* htb classes have a default rate estimator */ |
69 | module_param(htb_rate_est, int, 0640); | |
70 | MODULE_PARM_DESC(htb_rate_est, "setup a default rate estimator (4sec 16sec) for htb classes"); | |
71 | ||
1da177e4 LT |
72 | /* used internaly to keep status of single class */ |
73 | enum htb_cmode { | |
87990467 SH |
74 | HTB_CANT_SEND, /* class can't send and can't borrow */ |
75 | HTB_MAY_BORROW, /* class can't send but may borrow */ | |
76 | HTB_CAN_SEND /* class can send */ | |
1da177e4 LT |
77 | }; |
78 | ||
c9364636 ED |
79 | struct htb_prio { |
80 | union { | |
81 | struct rb_root row; | |
82 | struct rb_root feed; | |
83 | }; | |
84 | struct rb_node *ptr; | |
85 | /* When class changes from state 1->2 and disconnects from | |
86 | * parent's feed then we lost ptr value and start from the | |
87 | * first child again. Here we store classid of the | |
88 | * last valid ptr (used when ptr is NULL). | |
89 | */ | |
90 | u32 last_ptr_id; | |
91 | }; | |
92 | ||
ca4ec90b ED |
93 | /* interior & leaf nodes; props specific to leaves are marked L: |
94 | * To reduce false sharing, place mostly read fields at beginning, | |
95 | * and mostly written ones at the end. | |
96 | */ | |
87990467 | 97 | struct htb_class { |
f4c1f3e0 | 98 | struct Qdisc_class_common common; |
ca4ec90b ED |
99 | struct psched_ratecfg rate; |
100 | struct psched_ratecfg ceil; | |
101 | s64 buffer, cbuffer;/* token bucket depth/rate */ | |
102 | s64 mbuffer; /* max wait time */ | |
cbd37556 | 103 | u32 prio; /* these two are used only by leaves... */ |
ca4ec90b ED |
104 | int quantum; /* but stored for parent-to-leaf return */ |
105 | ||
25d8c0d5 | 106 | struct tcf_proto __rcu *filter_list; /* class attached filters */ |
ca4ec90b ED |
107 | int filter_cnt; |
108 | int refcnt; /* usage count of this class */ | |
109 | ||
110 | int level; /* our level (see above) */ | |
111 | unsigned int children; | |
112 | struct htb_class *parent; /* parent class */ | |
113 | ||
45203a3b | 114 | struct gnet_stats_rate_est64 rate_est; |
1da177e4 | 115 | |
ca4ec90b ED |
116 | /* |
117 | * Written often fields | |
118 | */ | |
119 | struct gnet_stats_basic_packed bstats; | |
ca4ec90b | 120 | struct tc_htb_xstats xstats; /* our special stats */ |
87990467 | 121 | |
ca4ec90b ED |
122 | /* token bucket parameters */ |
123 | s64 tokens, ctokens;/* current number of tokens */ | |
124 | s64 t_c; /* checkpoint time */ | |
c19f7a34 | 125 | |
87990467 SH |
126 | union { |
127 | struct htb_class_leaf { | |
87990467 | 128 | struct list_head drop_list; |
c9364636 ED |
129 | int deficit[TC_HTB_MAXDEPTH]; |
130 | struct Qdisc *q; | |
87990467 SH |
131 | } leaf; |
132 | struct htb_class_inner { | |
c9364636 | 133 | struct htb_prio clprio[TC_HTB_NUMPRIO]; |
87990467 SH |
134 | } inner; |
135 | } un; | |
ca4ec90b | 136 | s64 pq_key; |
87990467 | 137 | |
ca4ec90b ED |
138 | int prio_activity; /* for which prios are we active */ |
139 | enum htb_cmode cmode; /* current mode of the class */ | |
140 | struct rb_node pq_node; /* node for event queue */ | |
141 | struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ | |
338ed9b4 ED |
142 | |
143 | unsigned int drops ____cacheline_aligned_in_smp; | |
1da177e4 LT |
144 | }; |
145 | ||
c9364636 ED |
146 | struct htb_level { |
147 | struct rb_root wait_pq; | |
148 | struct htb_prio hprio[TC_HTB_NUMPRIO]; | |
149 | }; | |
150 | ||
87990467 | 151 | struct htb_sched { |
f4c1f3e0 | 152 | struct Qdisc_class_hash clhash; |
c9364636 ED |
153 | int defcls; /* class where unclassified flows go to */ |
154 | int rate2quantum; /* quant = rate / rate2quantum */ | |
1da177e4 | 155 | |
c9364636 | 156 | /* filters for qdisc itself */ |
25d8c0d5 | 157 | struct tcf_proto __rcu *filter_list; |
1da177e4 | 158 | |
c9364636 ED |
159 | #define HTB_WARN_TOOMANYEVENTS 0x1 |
160 | unsigned int warned; /* only one warning */ | |
161 | int direct_qlen; | |
162 | struct work_struct work; | |
1da177e4 | 163 | |
c9364636 ED |
164 | /* non shaped skbs; let them go directly thru */ |
165 | struct sk_buff_head direct_queue; | |
166 | long direct_pkts; | |
1da177e4 | 167 | |
c9364636 | 168 | struct qdisc_watchdog watchdog; |
1da177e4 | 169 | |
c9364636 ED |
170 | s64 now; /* cached dequeue time */ |
171 | struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */ | |
1da177e4 | 172 | |
c9364636 ED |
173 | /* time of nearest event per level (row) */ |
174 | s64 near_ev_cache[TC_HTB_MAXDEPTH]; | |
87990467 | 175 | |
c9364636 | 176 | int row_mask[TC_HTB_MAXDEPTH]; |
e82181de | 177 | |
c9364636 | 178 | struct htb_level hlevel[TC_HTB_MAXDEPTH]; |
1da177e4 LT |
179 | }; |
180 | ||
1da177e4 | 181 | /* find class in global hash table using given handle */ |
87990467 | 182 | static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch) |
1da177e4 LT |
183 | { |
184 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 185 | struct Qdisc_class_common *clc; |
0cef296d | 186 | |
f4c1f3e0 PM |
187 | clc = qdisc_class_find(&q->clhash, handle); |
188 | if (clc == NULL) | |
1da177e4 | 189 | return NULL; |
f4c1f3e0 | 190 | return container_of(clc, struct htb_class, common); |
1da177e4 LT |
191 | } |
192 | ||
193 | /** | |
194 | * htb_classify - classify a packet into class | |
195 | * | |
196 | * It returns NULL if the packet should be dropped or -1 if the packet | |
197 | * should be passed directly thru. In all other cases leaf class is returned. | |
198 | * We allow direct class selection by classid in priority. The we examine | |
199 | * filters in qdisc and in inner nodes (if higher filter points to the inner | |
200 | * node). If we end up with classid MAJOR:0 we enqueue the skb into special | |
10297b99 | 201 | * internal fifo (direct). These packets then go directly thru. If we still |
25985edc | 202 | * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful |
1da177e4 LT |
203 | * then finish and return direct queue. |
204 | */ | |
cc7ec456 | 205 | #define HTB_DIRECT ((struct htb_class *)-1L) |
1da177e4 | 206 | |
87990467 SH |
207 | static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, |
208 | int *qerr) | |
1da177e4 LT |
209 | { |
210 | struct htb_sched *q = qdisc_priv(sch); | |
211 | struct htb_class *cl; | |
212 | struct tcf_result res; | |
213 | struct tcf_proto *tcf; | |
214 | int result; | |
215 | ||
216 | /* allow to select class by setting skb->priority to valid classid; | |
cc7ec456 ED |
217 | * note that nfmark can be used too by attaching filter fw with no |
218 | * rules in it | |
219 | */ | |
1da177e4 | 220 | if (skb->priority == sch->handle) |
87990467 | 221 | return HTB_DIRECT; /* X:0 (direct flow) selected */ |
cc7ec456 | 222 | cl = htb_find(skb->priority, sch); |
29824310 HM |
223 | if (cl) { |
224 | if (cl->level == 0) | |
225 | return cl; | |
226 | /* Start with inner filter chain if a non-leaf class is selected */ | |
25d8c0d5 | 227 | tcf = rcu_dereference_bh(cl->filter_list); |
29824310 | 228 | } else { |
25d8c0d5 | 229 | tcf = rcu_dereference_bh(q->filter_list); |
29824310 | 230 | } |
1da177e4 | 231 | |
c27f339a | 232 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
3b3ae880 | 233 | while (tcf && (result = tc_classify(skb, tcf, &res, false)) >= 0) { |
1da177e4 LT |
234 | #ifdef CONFIG_NET_CLS_ACT |
235 | switch (result) { | |
236 | case TC_ACT_QUEUED: | |
87990467 | 237 | case TC_ACT_STOLEN: |
378a2f09 | 238 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
1da177e4 LT |
239 | case TC_ACT_SHOT: |
240 | return NULL; | |
241 | } | |
1da177e4 | 242 | #endif |
cc7ec456 ED |
243 | cl = (void *)res.class; |
244 | if (!cl) { | |
1da177e4 | 245 | if (res.classid == sch->handle) |
87990467 | 246 | return HTB_DIRECT; /* X:0 (direct flow) */ |
cc7ec456 ED |
247 | cl = htb_find(res.classid, sch); |
248 | if (!cl) | |
87990467 | 249 | break; /* filter selected invalid classid */ |
1da177e4 LT |
250 | } |
251 | if (!cl->level) | |
87990467 | 252 | return cl; /* we hit leaf; return it */ |
1da177e4 LT |
253 | |
254 | /* we have got inner class; apply inner filter chain */ | |
25d8c0d5 | 255 | tcf = rcu_dereference_bh(cl->filter_list); |
1da177e4 LT |
256 | } |
257 | /* classification failed; try to use default class */ | |
87990467 | 258 | cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch); |
1da177e4 | 259 | if (!cl || cl->level) |
87990467 | 260 | return HTB_DIRECT; /* bad default .. this is safe bet */ |
1da177e4 LT |
261 | return cl; |
262 | } | |
263 | ||
1da177e4 LT |
264 | /** |
265 | * htb_add_to_id_tree - adds class to the round robin list | |
266 | * | |
267 | * Routine adds class to the list (actually tree) sorted by classid. | |
268 | * Make sure that class is not already on such list for given prio. | |
269 | */ | |
87990467 SH |
270 | static void htb_add_to_id_tree(struct rb_root *root, |
271 | struct htb_class *cl, int prio) | |
1da177e4 LT |
272 | { |
273 | struct rb_node **p = &root->rb_node, *parent = NULL; | |
3bf72957 | 274 | |
1da177e4 | 275 | while (*p) { |
87990467 SH |
276 | struct htb_class *c; |
277 | parent = *p; | |
1da177e4 | 278 | c = rb_entry(parent, struct htb_class, node[prio]); |
3bf72957 | 279 | |
f4c1f3e0 | 280 | if (cl->common.classid > c->common.classid) |
1da177e4 | 281 | p = &parent->rb_right; |
87990467 | 282 | else |
1da177e4 LT |
283 | p = &parent->rb_left; |
284 | } | |
285 | rb_link_node(&cl->node[prio], parent, p); | |
286 | rb_insert_color(&cl->node[prio], root); | |
287 | } | |
288 | ||
289 | /** | |
290 | * htb_add_to_wait_tree - adds class to the event queue with delay | |
291 | * | |
292 | * The class is added to priority event queue to indicate that class will | |
293 | * change its mode in cl->pq_key microseconds. Make sure that class is not | |
294 | * already in the queue. | |
295 | */ | |
87990467 | 296 | static void htb_add_to_wait_tree(struct htb_sched *q, |
56b765b7 | 297 | struct htb_class *cl, s64 delay) |
1da177e4 | 298 | { |
c9364636 | 299 | struct rb_node **p = &q->hlevel[cl->level].wait_pq.rb_node, *parent = NULL; |
3bf72957 | 300 | |
fb983d45 PM |
301 | cl->pq_key = q->now + delay; |
302 | if (cl->pq_key == q->now) | |
1da177e4 LT |
303 | cl->pq_key++; |
304 | ||
305 | /* update the nearest event cache */ | |
fb983d45 | 306 | if (q->near_ev_cache[cl->level] > cl->pq_key) |
1da177e4 | 307 | q->near_ev_cache[cl->level] = cl->pq_key; |
87990467 | 308 | |
1da177e4 | 309 | while (*p) { |
87990467 SH |
310 | struct htb_class *c; |
311 | parent = *p; | |
1da177e4 | 312 | c = rb_entry(parent, struct htb_class, pq_node); |
fb983d45 | 313 | if (cl->pq_key >= c->pq_key) |
1da177e4 | 314 | p = &parent->rb_right; |
87990467 | 315 | else |
1da177e4 LT |
316 | p = &parent->rb_left; |
317 | } | |
318 | rb_link_node(&cl->pq_node, parent, p); | |
c9364636 | 319 | rb_insert_color(&cl->pq_node, &q->hlevel[cl->level].wait_pq); |
1da177e4 LT |
320 | } |
321 | ||
322 | /** | |
323 | * htb_next_rb_node - finds next node in binary tree | |
324 | * | |
325 | * When we are past last key we return NULL. | |
326 | * Average complexity is 2 steps per call. | |
327 | */ | |
3696f625 | 328 | static inline void htb_next_rb_node(struct rb_node **n) |
1da177e4 LT |
329 | { |
330 | *n = rb_next(*n); | |
331 | } | |
332 | ||
333 | /** | |
334 | * htb_add_class_to_row - add class to its row | |
335 | * | |
336 | * The class is added to row at priorities marked in mask. | |
337 | * It does nothing if mask == 0. | |
338 | */ | |
87990467 SH |
339 | static inline void htb_add_class_to_row(struct htb_sched *q, |
340 | struct htb_class *cl, int mask) | |
1da177e4 | 341 | { |
1da177e4 LT |
342 | q->row_mask[cl->level] |= mask; |
343 | while (mask) { | |
344 | int prio = ffz(~mask); | |
345 | mask &= ~(1 << prio); | |
c9364636 | 346 | htb_add_to_id_tree(&q->hlevel[cl->level].hprio[prio].row, cl, prio); |
1da177e4 LT |
347 | } |
348 | } | |
349 | ||
3696f625 SH |
350 | /* If this triggers, it is a bug in this code, but it need not be fatal */ |
351 | static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root) | |
352 | { | |
81771b3b | 353 | if (RB_EMPTY_NODE(rb)) { |
3696f625 SH |
354 | WARN_ON(1); |
355 | } else { | |
356 | rb_erase(rb, root); | |
357 | RB_CLEAR_NODE(rb); | |
358 | } | |
359 | } | |
360 | ||
361 | ||
1da177e4 LT |
362 | /** |
363 | * htb_remove_class_from_row - removes class from its row | |
364 | * | |
365 | * The class is removed from row at priorities marked in mask. | |
366 | * It does nothing if mask == 0. | |
367 | */ | |
87990467 SH |
368 | static inline void htb_remove_class_from_row(struct htb_sched *q, |
369 | struct htb_class *cl, int mask) | |
1da177e4 LT |
370 | { |
371 | int m = 0; | |
c9364636 | 372 | struct htb_level *hlevel = &q->hlevel[cl->level]; |
3bf72957 | 373 | |
1da177e4 LT |
374 | while (mask) { |
375 | int prio = ffz(~mask); | |
c9364636 | 376 | struct htb_prio *hprio = &hlevel->hprio[prio]; |
3696f625 | 377 | |
1da177e4 | 378 | mask &= ~(1 << prio); |
c9364636 ED |
379 | if (hprio->ptr == cl->node + prio) |
380 | htb_next_rb_node(&hprio->ptr); | |
3696f625 | 381 | |
c9364636 ED |
382 | htb_safe_rb_erase(cl->node + prio, &hprio->row); |
383 | if (!hprio->row.rb_node) | |
1da177e4 LT |
384 | m |= 1 << prio; |
385 | } | |
1da177e4 LT |
386 | q->row_mask[cl->level] &= ~m; |
387 | } | |
388 | ||
389 | /** | |
390 | * htb_activate_prios - creates active classe's feed chain | |
391 | * | |
392 | * The class is connected to ancestors and/or appropriate rows | |
10297b99 | 393 | * for priorities it is participating on. cl->cmode must be new |
1da177e4 LT |
394 | * (activated) mode. It does nothing if cl->prio_activity == 0. |
395 | */ | |
87990467 | 396 | static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
397 | { |
398 | struct htb_class *p = cl->parent; | |
87990467 | 399 | long m, mask = cl->prio_activity; |
1da177e4 LT |
400 | |
401 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
402 | m = mask; |
403 | while (m) { | |
1da177e4 LT |
404 | int prio = ffz(~m); |
405 | m &= ~(1 << prio); | |
87990467 | 406 | |
c9364636 | 407 | if (p->un.inner.clprio[prio].feed.rb_node) |
1da177e4 | 408 | /* parent already has its feed in use so that |
cc7ec456 ED |
409 | * reset bit in mask as parent is already ok |
410 | */ | |
1da177e4 | 411 | mask &= ~(1 << prio); |
87990467 | 412 | |
c9364636 | 413 | htb_add_to_id_tree(&p->un.inner.clprio[prio].feed, cl, prio); |
1da177e4 | 414 | } |
1da177e4 | 415 | p->prio_activity |= mask; |
87990467 SH |
416 | cl = p; |
417 | p = cl->parent; | |
3bf72957 | 418 | |
1da177e4 LT |
419 | } |
420 | if (cl->cmode == HTB_CAN_SEND && mask) | |
87990467 | 421 | htb_add_class_to_row(q, cl, mask); |
1da177e4 LT |
422 | } |
423 | ||
424 | /** | |
425 | * htb_deactivate_prios - remove class from feed chain | |
426 | * | |
10297b99 | 427 | * cl->cmode must represent old mode (before deactivation). It does |
1da177e4 LT |
428 | * nothing if cl->prio_activity == 0. Class is removed from all feed |
429 | * chains and rows. | |
430 | */ | |
431 | static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) | |
432 | { | |
433 | struct htb_class *p = cl->parent; | |
87990467 | 434 | long m, mask = cl->prio_activity; |
1da177e4 LT |
435 | |
436 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
437 | m = mask; |
438 | mask = 0; | |
1da177e4 LT |
439 | while (m) { |
440 | int prio = ffz(~m); | |
441 | m &= ~(1 << prio); | |
87990467 | 442 | |
c9364636 | 443 | if (p->un.inner.clprio[prio].ptr == cl->node + prio) { |
1da177e4 | 444 | /* we are removing child which is pointed to from |
cc7ec456 ED |
445 | * parent feed - forget the pointer but remember |
446 | * classid | |
447 | */ | |
c9364636 ED |
448 | p->un.inner.clprio[prio].last_ptr_id = cl->common.classid; |
449 | p->un.inner.clprio[prio].ptr = NULL; | |
1da177e4 | 450 | } |
87990467 | 451 | |
c9364636 ED |
452 | htb_safe_rb_erase(cl->node + prio, |
453 | &p->un.inner.clprio[prio].feed); | |
87990467 | 454 | |
c9364636 | 455 | if (!p->un.inner.clprio[prio].feed.rb_node) |
1da177e4 LT |
456 | mask |= 1 << prio; |
457 | } | |
3bf72957 | 458 | |
1da177e4 | 459 | p->prio_activity &= ~mask; |
87990467 SH |
460 | cl = p; |
461 | p = cl->parent; | |
3bf72957 | 462 | |
1da177e4 | 463 | } |
87990467 SH |
464 | if (cl->cmode == HTB_CAN_SEND && mask) |
465 | htb_remove_class_from_row(q, cl, mask); | |
1da177e4 LT |
466 | } |
467 | ||
56b765b7 | 468 | static inline s64 htb_lowater(const struct htb_class *cl) |
18a63e86 | 469 | { |
47083fc0 JDB |
470 | if (htb_hysteresis) |
471 | return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0; | |
472 | else | |
473 | return 0; | |
18a63e86 | 474 | } |
56b765b7 | 475 | static inline s64 htb_hiwater(const struct htb_class *cl) |
18a63e86 | 476 | { |
47083fc0 JDB |
477 | if (htb_hysteresis) |
478 | return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0; | |
479 | else | |
480 | return 0; | |
18a63e86 | 481 | } |
47083fc0 | 482 | |
18a63e86 | 483 | |
1da177e4 LT |
484 | /** |
485 | * htb_class_mode - computes and returns current class mode | |
486 | * | |
487 | * It computes cl's mode at time cl->t_c+diff and returns it. If mode | |
488 | * is not HTB_CAN_SEND then cl->pq_key is updated to time difference | |
10297b99 | 489 | * from now to time when cl will change its state. |
1da177e4 | 490 | * Also it is worth to note that class mode doesn't change simply |
10297b99 | 491 | * at cl->{c,}tokens == 0 but there can rather be hysteresis of |
1da177e4 LT |
492 | * 0 .. -cl->{c,}buffer range. It is meant to limit number of |
493 | * mode transitions per time unit. The speed gain is about 1/6. | |
494 | */ | |
87990467 | 495 | static inline enum htb_cmode |
56b765b7 | 496 | htb_class_mode(struct htb_class *cl, s64 *diff) |
1da177e4 | 497 | { |
56b765b7 | 498 | s64 toks; |
1da177e4 | 499 | |
87990467 SH |
500 | if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) { |
501 | *diff = -toks; | |
502 | return HTB_CANT_SEND; | |
503 | } | |
18a63e86 | 504 | |
87990467 SH |
505 | if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl)) |
506 | return HTB_CAN_SEND; | |
1da177e4 | 507 | |
87990467 SH |
508 | *diff = -toks; |
509 | return HTB_MAY_BORROW; | |
1da177e4 LT |
510 | } |
511 | ||
512 | /** | |
513 | * htb_change_class_mode - changes classe's mode | |
514 | * | |
515 | * This should be the only way how to change classe's mode under normal | |
516 | * cirsumstances. Routine will update feed lists linkage, change mode | |
517 | * and add class to the wait event queue if appropriate. New mode should | |
518 | * be different from old one and cl->pq_key has to be valid if changing | |
519 | * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). | |
520 | */ | |
87990467 | 521 | static void |
56b765b7 | 522 | htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, s64 *diff) |
87990467 SH |
523 | { |
524 | enum htb_cmode new_mode = htb_class_mode(cl, diff); | |
1da177e4 LT |
525 | |
526 | if (new_mode == cl->cmode) | |
87990467 SH |
527 | return; |
528 | ||
529 | if (cl->prio_activity) { /* not necessary: speed optimization */ | |
530 | if (cl->cmode != HTB_CANT_SEND) | |
531 | htb_deactivate_prios(q, cl); | |
1da177e4 | 532 | cl->cmode = new_mode; |
87990467 SH |
533 | if (new_mode != HTB_CANT_SEND) |
534 | htb_activate_prios(q, cl); | |
535 | } else | |
1da177e4 LT |
536 | cl->cmode = new_mode; |
537 | } | |
538 | ||
539 | /** | |
10297b99 | 540 | * htb_activate - inserts leaf cl into appropriate active feeds |
1da177e4 LT |
541 | * |
542 | * Routine learns (new) priority of leaf and activates feed chain | |
543 | * for the prio. It can be called on already active leaf safely. | |
544 | * It also adds leaf into droplist. | |
545 | */ | |
87990467 | 546 | static inline void htb_activate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 547 | { |
547b792c | 548 | WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen); |
3bf72957 | 549 | |
1da177e4 | 550 | if (!cl->prio_activity) { |
c19f7a34 | 551 | cl->prio_activity = 1 << cl->prio; |
87990467 SH |
552 | htb_activate_prios(q, cl); |
553 | list_add_tail(&cl->un.leaf.drop_list, | |
c19f7a34 | 554 | q->drops + cl->prio); |
1da177e4 LT |
555 | } |
556 | } | |
557 | ||
558 | /** | |
10297b99 | 559 | * htb_deactivate - remove leaf cl from active feeds |
1da177e4 LT |
560 | * |
561 | * Make sure that leaf is active. In the other words it can't be called | |
562 | * with non-active leaf. It also removes class from the drop list. | |
563 | */ | |
87990467 | 564 | static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 565 | { |
547b792c | 566 | WARN_ON(!cl->prio_activity); |
3bf72957 | 567 | |
87990467 | 568 | htb_deactivate_prios(q, cl); |
1da177e4 LT |
569 | cl->prio_activity = 0; |
570 | list_del_init(&cl->un.leaf.drop_list); | |
571 | } | |
572 | ||
520ac30f ED |
573 | static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
574 | struct sk_buff **to_free) | |
1da177e4 | 575 | { |
f30ab418 | 576 | int uninitialized_var(ret); |
87990467 SH |
577 | struct htb_sched *q = qdisc_priv(sch); |
578 | struct htb_class *cl = htb_classify(skb, sch, &ret); | |
579 | ||
580 | if (cl == HTB_DIRECT) { | |
581 | /* enqueue to helper queue */ | |
582 | if (q->direct_queue.qlen < q->direct_qlen) { | |
583 | __skb_queue_tail(&q->direct_queue, skb); | |
584 | q->direct_pkts++; | |
585 | } else { | |
520ac30f | 586 | return qdisc_drop(skb, sch, to_free); |
87990467 | 587 | } |
1da177e4 | 588 | #ifdef CONFIG_NET_CLS_ACT |
87990467 | 589 | } else if (!cl) { |
c27f339a | 590 | if (ret & __NET_XMIT_BYPASS) |
25331d6c | 591 | qdisc_qstats_drop(sch); |
520ac30f | 592 | __qdisc_drop(skb, to_free); |
87990467 | 593 | return ret; |
1da177e4 | 594 | #endif |
520ac30f ED |
595 | } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q, |
596 | to_free)) != NET_XMIT_SUCCESS) { | |
378a2f09 | 597 | if (net_xmit_drop_count(ret)) { |
25331d6c | 598 | qdisc_qstats_drop(sch); |
338ed9b4 | 599 | cl->drops++; |
378a2f09 | 600 | } |
69747650 | 601 | return ret; |
87990467 | 602 | } else { |
87990467 SH |
603 | htb_activate(q, cl); |
604 | } | |
605 | ||
431e3a8e | 606 | qdisc_qstats_backlog_inc(sch, skb); |
87990467 | 607 | sch->q.qlen++; |
87990467 | 608 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
609 | } |
610 | ||
56b765b7 | 611 | static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, s64 diff) |
59e4220a | 612 | { |
56b765b7 | 613 | s64 toks = diff + cl->tokens; |
59e4220a JP |
614 | |
615 | if (toks > cl->buffer) | |
616 | toks = cl->buffer; | |
292f1c7f | 617 | toks -= (s64) psched_l2t_ns(&cl->rate, bytes); |
59e4220a JP |
618 | if (toks <= -cl->mbuffer) |
619 | toks = 1 - cl->mbuffer; | |
620 | ||
621 | cl->tokens = toks; | |
622 | } | |
623 | ||
56b765b7 | 624 | static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, s64 diff) |
59e4220a | 625 | { |
56b765b7 | 626 | s64 toks = diff + cl->ctokens; |
59e4220a JP |
627 | |
628 | if (toks > cl->cbuffer) | |
629 | toks = cl->cbuffer; | |
292f1c7f | 630 | toks -= (s64) psched_l2t_ns(&cl->ceil, bytes); |
59e4220a JP |
631 | if (toks <= -cl->mbuffer) |
632 | toks = 1 - cl->mbuffer; | |
633 | ||
634 | cl->ctokens = toks; | |
635 | } | |
636 | ||
1da177e4 LT |
637 | /** |
638 | * htb_charge_class - charges amount "bytes" to leaf and ancestors | |
639 | * | |
640 | * Routine assumes that packet "bytes" long was dequeued from leaf cl | |
641 | * borrowing from "level". It accounts bytes to ceil leaky bucket for | |
642 | * leaf and all ancestors and to rate bucket for ancestors at levels | |
643 | * "level" and higher. It also handles possible change of mode resulting | |
644 | * from the update. Note that mode can also increase here (MAY_BORROW to | |
645 | * CAN_SEND) because we can use more precise clock that event queue here. | |
646 | * In such case we remove class from event queue first. | |
647 | */ | |
87990467 | 648 | static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, |
c9726d68 | 649 | int level, struct sk_buff *skb) |
87990467 | 650 | { |
0abf77e5 | 651 | int bytes = qdisc_pkt_len(skb); |
1da177e4 | 652 | enum htb_cmode old_mode; |
56b765b7 | 653 | s64 diff; |
1da177e4 LT |
654 | |
655 | while (cl) { | |
56b765b7 | 656 | diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); |
1da177e4 | 657 | if (cl->level >= level) { |
87990467 SH |
658 | if (cl->level == level) |
659 | cl->xstats.lends++; | |
59e4220a | 660 | htb_accnt_tokens(cl, bytes, diff); |
1da177e4 LT |
661 | } else { |
662 | cl->xstats.borrows++; | |
87990467 | 663 | cl->tokens += diff; /* we moved t_c; update tokens */ |
1da177e4 | 664 | } |
59e4220a | 665 | htb_accnt_ctokens(cl, bytes, diff); |
1da177e4 | 666 | cl->t_c = q->now; |
1da177e4 | 667 | |
87990467 SH |
668 | old_mode = cl->cmode; |
669 | diff = 0; | |
670 | htb_change_class_mode(q, cl, &diff); | |
1da177e4 LT |
671 | if (old_mode != cl->cmode) { |
672 | if (old_mode != HTB_CAN_SEND) | |
c9364636 | 673 | htb_safe_rb_erase(&cl->pq_node, &q->hlevel[cl->level].wait_pq); |
1da177e4 | 674 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 675 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 676 | } |
1da177e4 | 677 | |
bfe0d029 ED |
678 | /* update basic stats except for leaves which are already updated */ |
679 | if (cl->level) | |
680 | bstats_update(&cl->bstats, skb); | |
681 | ||
1da177e4 LT |
682 | cl = cl->parent; |
683 | } | |
684 | } | |
685 | ||
686 | /** | |
687 | * htb_do_events - make mode changes to classes at the level | |
688 | * | |
fb983d45 | 689 | * Scans event queue for pending events and applies them. Returns time of |
1224736d | 690 | * next pending event (0 for no event in pq, q->now for too many events). |
fb983d45 | 691 | * Note: Applied are events whose have cl->pq_key <= q->now. |
1da177e4 | 692 | */ |
c9364636 | 693 | static s64 htb_do_events(struct htb_sched *q, const int level, |
5343a7f8 | 694 | unsigned long start) |
1da177e4 | 695 | { |
8f3ea33a | 696 | /* don't run for longer than 2 jiffies; 2 is used instead of |
cc7ec456 ED |
697 | * 1 to simplify things when jiffy is going to be incremented |
698 | * too soon | |
699 | */ | |
a73be040 | 700 | unsigned long stop_at = start + 2; |
c9364636 ED |
701 | struct rb_root *wait_pq = &q->hlevel[level].wait_pq; |
702 | ||
8f3ea33a | 703 | while (time_before(jiffies, stop_at)) { |
1da177e4 | 704 | struct htb_class *cl; |
56b765b7 | 705 | s64 diff; |
c9364636 | 706 | struct rb_node *p = rb_first(wait_pq); |
30bdbe39 | 707 | |
87990467 SH |
708 | if (!p) |
709 | return 0; | |
1da177e4 LT |
710 | |
711 | cl = rb_entry(p, struct htb_class, pq_node); | |
fb983d45 PM |
712 | if (cl->pq_key > q->now) |
713 | return cl->pq_key; | |
714 | ||
c9364636 | 715 | htb_safe_rb_erase(p, wait_pq); |
56b765b7 | 716 | diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); |
87990467 | 717 | htb_change_class_mode(q, cl, &diff); |
1da177e4 | 718 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 719 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 720 | } |
1224736d JP |
721 | |
722 | /* too much load - let's continue after a break for scheduling */ | |
e82181de | 723 | if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) { |
c17988a9 | 724 | pr_warn("htb: too many events!\n"); |
e82181de JP |
725 | q->warned |= HTB_WARN_TOOMANYEVENTS; |
726 | } | |
1224736d JP |
727 | |
728 | return q->now; | |
1da177e4 LT |
729 | } |
730 | ||
731 | /* Returns class->node+prio from id-tree where classe's id is >= id. NULL | |
cc7ec456 ED |
732 | * is no such one exists. |
733 | */ | |
87990467 SH |
734 | static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n, |
735 | u32 id) | |
1da177e4 LT |
736 | { |
737 | struct rb_node *r = NULL; | |
738 | while (n) { | |
87990467 SH |
739 | struct htb_class *cl = |
740 | rb_entry(n, struct htb_class, node[prio]); | |
87990467 | 741 | |
f4c1f3e0 | 742 | if (id > cl->common.classid) { |
1da177e4 | 743 | n = n->rb_right; |
1b5c0077 | 744 | } else if (id < cl->common.classid) { |
1da177e4 LT |
745 | r = n; |
746 | n = n->rb_left; | |
1b5c0077 JP |
747 | } else { |
748 | return n; | |
1da177e4 LT |
749 | } |
750 | } | |
751 | return r; | |
752 | } | |
753 | ||
754 | /** | |
755 | * htb_lookup_leaf - returns next leaf class in DRR order | |
756 | * | |
757 | * Find leaf where current feed pointers points to. | |
758 | */ | |
c9364636 | 759 | static struct htb_class *htb_lookup_leaf(struct htb_prio *hprio, const int prio) |
1da177e4 LT |
760 | { |
761 | int i; | |
762 | struct { | |
763 | struct rb_node *root; | |
764 | struct rb_node **pptr; | |
765 | u32 *pid; | |
87990467 SH |
766 | } stk[TC_HTB_MAXDEPTH], *sp = stk; |
767 | ||
c9364636 ED |
768 | BUG_ON(!hprio->row.rb_node); |
769 | sp->root = hprio->row.rb_node; | |
770 | sp->pptr = &hprio->ptr; | |
771 | sp->pid = &hprio->last_ptr_id; | |
1da177e4 LT |
772 | |
773 | for (i = 0; i < 65535; i++) { | |
87990467 | 774 | if (!*sp->pptr && *sp->pid) { |
10297b99 | 775 | /* ptr was invalidated but id is valid - try to recover |
cc7ec456 ED |
776 | * the original or next ptr |
777 | */ | |
87990467 SH |
778 | *sp->pptr = |
779 | htb_id_find_next_upper(prio, sp->root, *sp->pid); | |
1da177e4 | 780 | } |
87990467 | 781 | *sp->pid = 0; /* ptr is valid now so that remove this hint as it |
cc7ec456 ED |
782 | * can become out of date quickly |
783 | */ | |
87990467 | 784 | if (!*sp->pptr) { /* we are at right end; rewind & go up */ |
1da177e4 | 785 | *sp->pptr = sp->root; |
87990467 | 786 | while ((*sp->pptr)->rb_left) |
1da177e4 LT |
787 | *sp->pptr = (*sp->pptr)->rb_left; |
788 | if (sp > stk) { | |
789 | sp--; | |
512bb43e JP |
790 | if (!*sp->pptr) { |
791 | WARN_ON(1); | |
87990467 | 792 | return NULL; |
512bb43e | 793 | } |
87990467 | 794 | htb_next_rb_node(sp->pptr); |
1da177e4 LT |
795 | } |
796 | } else { | |
797 | struct htb_class *cl; | |
c9364636 ED |
798 | struct htb_prio *clp; |
799 | ||
87990467 SH |
800 | cl = rb_entry(*sp->pptr, struct htb_class, node[prio]); |
801 | if (!cl->level) | |
1da177e4 | 802 | return cl; |
c9364636 ED |
803 | clp = &cl->un.inner.clprio[prio]; |
804 | (++sp)->root = clp->feed.rb_node; | |
805 | sp->pptr = &clp->ptr; | |
806 | sp->pid = &clp->last_ptr_id; | |
1da177e4 LT |
807 | } |
808 | } | |
547b792c | 809 | WARN_ON(1); |
1da177e4 LT |
810 | return NULL; |
811 | } | |
812 | ||
813 | /* dequeues packet at given priority and level; call only if | |
cc7ec456 ED |
814 | * you are sure that there is active class at prio/level |
815 | */ | |
c9364636 ED |
816 | static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, const int prio, |
817 | const int level) | |
1da177e4 LT |
818 | { |
819 | struct sk_buff *skb = NULL; | |
87990467 | 820 | struct htb_class *cl, *start; |
c9364636 ED |
821 | struct htb_level *hlevel = &q->hlevel[level]; |
822 | struct htb_prio *hprio = &hlevel->hprio[prio]; | |
823 | ||
1da177e4 | 824 | /* look initial class up in the row */ |
c9364636 | 825 | start = cl = htb_lookup_leaf(hprio, prio); |
87990467 | 826 | |
1da177e4 LT |
827 | do { |
828 | next: | |
512bb43e | 829 | if (unlikely(!cl)) |
87990467 | 830 | return NULL; |
1da177e4 LT |
831 | |
832 | /* class can be empty - it is unlikely but can be true if leaf | |
cc7ec456 ED |
833 | * qdisc drops packets in enqueue routine or if someone used |
834 | * graft operation on the leaf since last dequeue; | |
835 | * simply deactivate and skip such class | |
836 | */ | |
1da177e4 LT |
837 | if (unlikely(cl->un.leaf.q->q.qlen == 0)) { |
838 | struct htb_class *next; | |
87990467 | 839 | htb_deactivate(q, cl); |
1da177e4 LT |
840 | |
841 | /* row/level might become empty */ | |
842 | if ((q->row_mask[level] & (1 << prio)) == 0) | |
87990467 | 843 | return NULL; |
1da177e4 | 844 | |
c9364636 | 845 | next = htb_lookup_leaf(hprio, prio); |
87990467 SH |
846 | |
847 | if (cl == start) /* fix start if we just deleted it */ | |
1da177e4 LT |
848 | start = next; |
849 | cl = next; | |
850 | goto next; | |
851 | } | |
87990467 SH |
852 | |
853 | skb = cl->un.leaf.q->dequeue(cl->un.leaf.q); | |
854 | if (likely(skb != NULL)) | |
1da177e4 | 855 | break; |
633fe66e | 856 | |
b00355db | 857 | qdisc_warn_nonwc("htb", cl->un.leaf.q); |
c9364636 ED |
858 | htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr: |
859 | &q->hlevel[0].hprio[prio].ptr); | |
860 | cl = htb_lookup_leaf(hprio, prio); | |
1da177e4 LT |
861 | |
862 | } while (cl != start); | |
863 | ||
864 | if (likely(skb != NULL)) { | |
196d97f6 | 865 | bstats_update(&cl->bstats, skb); |
0abf77e5 JK |
866 | cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb); |
867 | if (cl->un.leaf.deficit[level] < 0) { | |
c19f7a34 | 868 | cl->un.leaf.deficit[level] += cl->quantum; |
c9364636 ED |
869 | htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr : |
870 | &q->hlevel[0].hprio[prio].ptr); | |
1da177e4 LT |
871 | } |
872 | /* this used to be after charge_class but this constelation | |
cc7ec456 ED |
873 | * gives us slightly better performance |
874 | */ | |
1da177e4 | 875 | if (!cl->un.leaf.q->q.qlen) |
87990467 | 876 | htb_deactivate(q, cl); |
c9726d68 | 877 | htb_charge_class(q, cl, level, skb); |
1da177e4 LT |
878 | } |
879 | return skb; | |
880 | } | |
881 | ||
1da177e4 LT |
882 | static struct sk_buff *htb_dequeue(struct Qdisc *sch) |
883 | { | |
9190b3b3 | 884 | struct sk_buff *skb; |
1da177e4 LT |
885 | struct htb_sched *q = qdisc_priv(sch); |
886 | int level; | |
5343a7f8 | 887 | s64 next_event; |
a73be040 | 888 | unsigned long start_at; |
1da177e4 LT |
889 | |
890 | /* try to dequeue direct packets as high prio (!) to minimize cpu work */ | |
87990467 SH |
891 | skb = __skb_dequeue(&q->direct_queue); |
892 | if (skb != NULL) { | |
9190b3b3 ED |
893 | ok: |
894 | qdisc_bstats_update(sch, skb); | |
431e3a8e | 895 | qdisc_qstats_backlog_dec(sch, skb); |
1da177e4 LT |
896 | sch->q.qlen--; |
897 | return skb; | |
898 | } | |
899 | ||
87990467 SH |
900 | if (!sch->q.qlen) |
901 | goto fin; | |
d2de875c | 902 | q->now = ktime_get_ns(); |
a73be040 | 903 | start_at = jiffies; |
1da177e4 | 904 | |
d2fe85da | 905 | next_event = q->now + 5LLU * NSEC_PER_SEC; |
633fe66e | 906 | |
1da177e4 LT |
907 | for (level = 0; level < TC_HTB_MAXDEPTH; level++) { |
908 | /* common case optimization - skip event handler quickly */ | |
909 | int m; | |
c9364636 | 910 | s64 event = q->near_ev_cache[level]; |
fb983d45 | 911 | |
c9364636 | 912 | if (q->now >= event) { |
a73be040 | 913 | event = htb_do_events(q, level, start_at); |
2e4b3b0e | 914 | if (!event) |
56b765b7 | 915 | event = q->now + NSEC_PER_SEC; |
2e4b3b0e | 916 | q->near_ev_cache[level] = event; |
c9364636 | 917 | } |
fb983d45 | 918 | |
c0851347 | 919 | if (next_event > event) |
fb983d45 | 920 | next_event = event; |
87990467 | 921 | |
1da177e4 LT |
922 | m = ~q->row_mask[level]; |
923 | while (m != (int)(-1)) { | |
87990467 | 924 | int prio = ffz(m); |
cc7ec456 | 925 | |
1da177e4 | 926 | m |= 1 << prio; |
87990467 | 927 | skb = htb_dequeue_tree(q, prio, level); |
9190b3b3 ED |
928 | if (likely(skb != NULL)) |
929 | goto ok; | |
1da177e4 LT |
930 | } |
931 | } | |
25331d6c | 932 | qdisc_qstats_overlimit(sch); |
a9efad8b | 933 | if (likely(next_event > q->now)) |
45f50bed | 934 | qdisc_watchdog_schedule_ns(&q->watchdog, next_event); |
a9efad8b | 935 | else |
1224736d | 936 | schedule_work(&q->work); |
1da177e4 | 937 | fin: |
1da177e4 LT |
938 | return skb; |
939 | } | |
940 | ||
1da177e4 LT |
941 | /* reset all classes */ |
942 | /* always caled under BH & queue lock */ | |
87990467 | 943 | static void htb_reset(struct Qdisc *sch) |
1da177e4 LT |
944 | { |
945 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 946 | struct htb_class *cl; |
f4c1f3e0 | 947 | unsigned int i; |
0cef296d | 948 | |
f4c1f3e0 | 949 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 950 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { |
1da177e4 | 951 | if (cl->level) |
87990467 | 952 | memset(&cl->un.inner, 0, sizeof(cl->un.inner)); |
1da177e4 | 953 | else { |
87990467 | 954 | if (cl->un.leaf.q) |
1da177e4 LT |
955 | qdisc_reset(cl->un.leaf.q); |
956 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); | |
957 | } | |
958 | cl->prio_activity = 0; | |
959 | cl->cmode = HTB_CAN_SEND; | |
1da177e4 LT |
960 | } |
961 | } | |
fb983d45 | 962 | qdisc_watchdog_cancel(&q->watchdog); |
a5a9f534 | 963 | __qdisc_reset_queue(&q->direct_queue); |
1da177e4 | 964 | sch->q.qlen = 0; |
431e3a8e | 965 | sch->qstats.backlog = 0; |
c9364636 | 966 | memset(q->hlevel, 0, sizeof(q->hlevel)); |
87990467 | 967 | memset(q->row_mask, 0, sizeof(q->row_mask)); |
1da177e4 | 968 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 969 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 LT |
970 | } |
971 | ||
27a3421e PM |
972 | static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = { |
973 | [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) }, | |
974 | [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) }, | |
975 | [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
976 | [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
6906f4ed | 977 | [TCA_HTB_DIRECT_QLEN] = { .type = NLA_U32 }, |
df62cdf3 ED |
978 | [TCA_HTB_RATE64] = { .type = NLA_U64 }, |
979 | [TCA_HTB_CEIL64] = { .type = NLA_U64 }, | |
27a3421e PM |
980 | }; |
981 | ||
1224736d JP |
982 | static void htb_work_func(struct work_struct *work) |
983 | { | |
984 | struct htb_sched *q = container_of(work, struct htb_sched, work); | |
985 | struct Qdisc *sch = q->watchdog.qdisc; | |
986 | ||
0ee13627 | 987 | rcu_read_lock(); |
1224736d | 988 | __netif_schedule(qdisc_root(sch)); |
0ee13627 | 989 | rcu_read_unlock(); |
1224736d JP |
990 | } |
991 | ||
1e90474c | 992 | static int htb_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
993 | { |
994 | struct htb_sched *q = qdisc_priv(sch); | |
6906f4ed | 995 | struct nlattr *tb[TCA_HTB_MAX + 1]; |
1da177e4 | 996 | struct tc_htb_glob *gopt; |
cee63723 | 997 | int err; |
1da177e4 | 998 | int i; |
cee63723 PM |
999 | |
1000 | if (!opt) | |
1001 | return -EINVAL; | |
1002 | ||
6906f4ed | 1003 | err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy); |
cee63723 PM |
1004 | if (err < 0) |
1005 | return err; | |
1006 | ||
6906f4ed | 1007 | if (!tb[TCA_HTB_INIT]) |
1da177e4 | 1008 | return -EINVAL; |
6906f4ed | 1009 | |
1e90474c | 1010 | gopt = nla_data(tb[TCA_HTB_INIT]); |
6906f4ed | 1011 | if (gopt->version != HTB_VER >> 16) |
1da177e4 | 1012 | return -EINVAL; |
1da177e4 | 1013 | |
f4c1f3e0 PM |
1014 | err = qdisc_class_hash_init(&q->clhash); |
1015 | if (err < 0) | |
1016 | return err; | |
1da177e4 | 1017 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 1018 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 | 1019 | |
fb983d45 | 1020 | qdisc_watchdog_init(&q->watchdog, sch); |
1224736d | 1021 | INIT_WORK(&q->work, htb_work_func); |
ab34f648 | 1022 | __skb_queue_head_init(&q->direct_queue); |
1da177e4 | 1023 | |
6906f4ed ED |
1024 | if (tb[TCA_HTB_DIRECT_QLEN]) |
1025 | q->direct_qlen = nla_get_u32(tb[TCA_HTB_DIRECT_QLEN]); | |
348e3435 | 1026 | else |
6906f4ed | 1027 | q->direct_qlen = qdisc_dev(sch)->tx_queue_len; |
348e3435 | 1028 | |
1da177e4 LT |
1029 | if ((q->rate2quantum = gopt->rate2quantum) < 1) |
1030 | q->rate2quantum = 1; | |
1031 | q->defcls = gopt->defcls; | |
1032 | ||
1033 | return 0; | |
1034 | } | |
1035 | ||
1036 | static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) | |
1037 | { | |
1038 | struct htb_sched *q = qdisc_priv(sch); | |
4b3550ef | 1039 | struct nlattr *nest; |
1da177e4 | 1040 | struct tc_htb_glob gopt; |
4b3550ef | 1041 | |
6f542efc ED |
1042 | /* Its safe to not acquire qdisc lock. As we hold RTNL, |
1043 | * no change can happen on the qdisc parameters. | |
1044 | */ | |
1da177e4 | 1045 | |
4b3550ef | 1046 | gopt.direct_pkts = q->direct_pkts; |
1da177e4 LT |
1047 | gopt.version = HTB_VER; |
1048 | gopt.rate2quantum = q->rate2quantum; | |
1049 | gopt.defcls = q->defcls; | |
3bf72957 | 1050 | gopt.debug = 0; |
4b3550ef PM |
1051 | |
1052 | nest = nla_nest_start(skb, TCA_OPTIONS); | |
1053 | if (nest == NULL) | |
1054 | goto nla_put_failure; | |
6906f4ed ED |
1055 | if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt) || |
1056 | nla_put_u32(skb, TCA_HTB_DIRECT_QLEN, q->direct_qlen)) | |
1b34ec43 | 1057 | goto nla_put_failure; |
4b3550ef | 1058 | |
6f542efc | 1059 | return nla_nest_end(skb, nest); |
4b3550ef | 1060 | |
1e90474c | 1061 | nla_put_failure: |
4b3550ef | 1062 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1063 | return -1; |
1064 | } | |
1065 | ||
1066 | static int htb_dump_class(struct Qdisc *sch, unsigned long arg, | |
87990467 | 1067 | struct sk_buff *skb, struct tcmsg *tcm) |
1da177e4 | 1068 | { |
87990467 | 1069 | struct htb_class *cl = (struct htb_class *)arg; |
4b3550ef | 1070 | struct nlattr *nest; |
1da177e4 LT |
1071 | struct tc_htb_opt opt; |
1072 | ||
6f542efc ED |
1073 | /* Its safe to not acquire qdisc lock. As we hold RTNL, |
1074 | * no change can happen on the class parameters. | |
1075 | */ | |
f4c1f3e0 PM |
1076 | tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT; |
1077 | tcm->tcm_handle = cl->common.classid; | |
1da177e4 LT |
1078 | if (!cl->level && cl->un.leaf.q) |
1079 | tcm->tcm_info = cl->un.leaf.q->handle; | |
1080 | ||
4b3550ef PM |
1081 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1082 | if (nest == NULL) | |
1083 | goto nla_put_failure; | |
1da177e4 | 1084 | |
87990467 | 1085 | memset(&opt, 0, sizeof(opt)); |
1da177e4 | 1086 | |
01cb71d2 | 1087 | psched_ratecfg_getrate(&opt.rate, &cl->rate); |
9c10f411 | 1088 | opt.buffer = PSCHED_NS2TICKS(cl->buffer); |
01cb71d2 | 1089 | psched_ratecfg_getrate(&opt.ceil, &cl->ceil); |
9c10f411 | 1090 | opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer); |
c19f7a34 JP |
1091 | opt.quantum = cl->quantum; |
1092 | opt.prio = cl->prio; | |
87990467 | 1093 | opt.level = cl->level; |
1b34ec43 DM |
1094 | if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt)) |
1095 | goto nla_put_failure; | |
df62cdf3 | 1096 | if ((cl->rate.rate_bytes_ps >= (1ULL << 32)) && |
2a51c1e8 ND |
1097 | nla_put_u64_64bit(skb, TCA_HTB_RATE64, cl->rate.rate_bytes_ps, |
1098 | TCA_HTB_PAD)) | |
df62cdf3 ED |
1099 | goto nla_put_failure; |
1100 | if ((cl->ceil.rate_bytes_ps >= (1ULL << 32)) && | |
2a51c1e8 ND |
1101 | nla_put_u64_64bit(skb, TCA_HTB_CEIL64, cl->ceil.rate_bytes_ps, |
1102 | TCA_HTB_PAD)) | |
df62cdf3 | 1103 | goto nla_put_failure; |
4b3550ef | 1104 | |
6f542efc | 1105 | return nla_nest_end(skb, nest); |
4b3550ef | 1106 | |
1e90474c | 1107 | nla_put_failure: |
4b3550ef | 1108 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1109 | return -1; |
1110 | } | |
1111 | ||
1112 | static int | |
87990467 | 1113 | htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d) |
1da177e4 | 1114 | { |
87990467 | 1115 | struct htb_class *cl = (struct htb_class *)arg; |
338ed9b4 ED |
1116 | struct gnet_stats_queue qs = { |
1117 | .drops = cl->drops, | |
1118 | }; | |
64015853 | 1119 | __u32 qlen = 0; |
1da177e4 | 1120 | |
338ed9b4 | 1121 | if (!cl->level && cl->un.leaf.q) { |
64015853 | 1122 | qlen = cl->un.leaf.q->q.qlen; |
338ed9b4 ED |
1123 | qs.backlog = cl->un.leaf.q->qstats.backlog; |
1124 | } | |
0564bf0a KK |
1125 | cl->xstats.tokens = clamp_t(s64, PSCHED_NS2TICKS(cl->tokens), |
1126 | INT_MIN, INT_MAX); | |
1127 | cl->xstats.ctokens = clamp_t(s64, PSCHED_NS2TICKS(cl->ctokens), | |
1128 | INT_MIN, INT_MAX); | |
1da177e4 | 1129 | |
edb09eb1 ED |
1130 | if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), |
1131 | d, NULL, &cl->bstats) < 0 || | |
d250a5f9 | 1132 | gnet_stats_copy_rate_est(d, NULL, &cl->rate_est) < 0 || |
338ed9b4 | 1133 | gnet_stats_copy_queue(d, NULL, &qs, qlen) < 0) |
1da177e4 LT |
1134 | return -1; |
1135 | ||
1136 | return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); | |
1137 | } | |
1138 | ||
1139 | static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
87990467 | 1140 | struct Qdisc **old) |
1da177e4 | 1141 | { |
87990467 | 1142 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 | 1143 | |
5b9a9ccf PM |
1144 | if (cl->level) |
1145 | return -EINVAL; | |
1146 | if (new == NULL && | |
3511c913 | 1147 | (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
5b9a9ccf PM |
1148 | cl->common.classid)) == NULL) |
1149 | return -ENOBUFS; | |
1150 | ||
86a7996c | 1151 | *old = qdisc_replace(sch, new, &cl->un.leaf.q); |
5b9a9ccf | 1152 | return 0; |
1da177e4 LT |
1153 | } |
1154 | ||
87990467 | 1155 | static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg) |
1da177e4 | 1156 | { |
87990467 | 1157 | struct htb_class *cl = (struct htb_class *)arg; |
5b9a9ccf | 1158 | return !cl->level ? cl->un.leaf.q : NULL; |
1da177e4 LT |
1159 | } |
1160 | ||
256d61b8 PM |
1161 | static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg) |
1162 | { | |
1163 | struct htb_class *cl = (struct htb_class *)arg; | |
1164 | ||
1165 | if (cl->un.leaf.q->q.qlen == 0) | |
1166 | htb_deactivate(qdisc_priv(sch), cl); | |
1167 | } | |
1168 | ||
1da177e4 LT |
1169 | static unsigned long htb_get(struct Qdisc *sch, u32 classid) |
1170 | { | |
87990467 SH |
1171 | struct htb_class *cl = htb_find(classid, sch); |
1172 | if (cl) | |
1da177e4 LT |
1173 | cl->refcnt++; |
1174 | return (unsigned long)cl; | |
1175 | } | |
1176 | ||
160d5e10 JP |
1177 | static inline int htb_parent_last_child(struct htb_class *cl) |
1178 | { | |
1179 | if (!cl->parent) | |
1180 | /* the root class */ | |
1181 | return 0; | |
42077599 | 1182 | if (cl->parent->children > 1) |
160d5e10 JP |
1183 | /* not the last child */ |
1184 | return 0; | |
160d5e10 JP |
1185 | return 1; |
1186 | } | |
1187 | ||
3ba08b00 JP |
1188 | static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl, |
1189 | struct Qdisc *new_q) | |
160d5e10 JP |
1190 | { |
1191 | struct htb_class *parent = cl->parent; | |
1192 | ||
547b792c | 1193 | WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity); |
160d5e10 | 1194 | |
3ba08b00 | 1195 | if (parent->cmode != HTB_CAN_SEND) |
c9364636 ED |
1196 | htb_safe_rb_erase(&parent->pq_node, |
1197 | &q->hlevel[parent->level].wait_pq); | |
3ba08b00 | 1198 | |
160d5e10 JP |
1199 | parent->level = 0; |
1200 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1201 | INIT_LIST_HEAD(&parent->un.leaf.drop_list); | |
1202 | parent->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
160d5e10 JP |
1203 | parent->tokens = parent->buffer; |
1204 | parent->ctokens = parent->cbuffer; | |
d2de875c | 1205 | parent->t_c = ktime_get_ns(); |
160d5e10 JP |
1206 | parent->cmode = HTB_CAN_SEND; |
1207 | } | |
1208 | ||
87990467 | 1209 | static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl) |
1da177e4 | 1210 | { |
1da177e4 | 1211 | if (!cl->level) { |
547b792c | 1212 | WARN_ON(!cl->un.leaf.q); |
1da177e4 LT |
1213 | qdisc_destroy(cl->un.leaf.q); |
1214 | } | |
ee39e10c | 1215 | gen_kill_estimator(&cl->bstats, &cl->rate_est); |
ff31ab56 | 1216 | tcf_destroy_chain(&cl->filter_list); |
1da177e4 LT |
1217 | kfree(cl); |
1218 | } | |
1219 | ||
87990467 | 1220 | static void htb_destroy(struct Qdisc *sch) |
1da177e4 LT |
1221 | { |
1222 | struct htb_sched *q = qdisc_priv(sch); | |
b67bfe0d | 1223 | struct hlist_node *next; |
fbd8f137 PM |
1224 | struct htb_class *cl; |
1225 | unsigned int i; | |
1da177e4 | 1226 | |
1224736d | 1227 | cancel_work_sync(&q->work); |
fb983d45 | 1228 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 | 1229 | /* This line used to be after htb_destroy_class call below |
cc7ec456 ED |
1230 | * and surprisingly it worked in 2.4. But it must precede it |
1231 | * because filter need its target class alive to be able to call | |
1232 | * unbind_filter on it (without Oops). | |
1233 | */ | |
ff31ab56 | 1234 | tcf_destroy_chain(&q->filter_list); |
87990467 | 1235 | |
f4c1f3e0 | 1236 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 1237 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) |
fbd8f137 PM |
1238 | tcf_destroy_chain(&cl->filter_list); |
1239 | } | |
f4c1f3e0 | 1240 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 1241 | hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i], |
f4c1f3e0 | 1242 | common.hnode) |
fbd8f137 PM |
1243 | htb_destroy_class(sch, cl); |
1244 | } | |
f4c1f3e0 | 1245 | qdisc_class_hash_destroy(&q->clhash); |
a5a9f534 | 1246 | __qdisc_reset_queue(&q->direct_queue); |
1da177e4 LT |
1247 | } |
1248 | ||
1249 | static int htb_delete(struct Qdisc *sch, unsigned long arg) | |
1250 | { | |
1251 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1252 | struct htb_class *cl = (struct htb_class *)arg; |
160d5e10 JP |
1253 | struct Qdisc *new_q = NULL; |
1254 | int last_child = 0; | |
1da177e4 | 1255 | |
a071d272 YY |
1256 | /* TODO: why don't allow to delete subtree ? references ? does |
1257 | * tc subsys guarantee us that in htb_destroy it holds no class | |
1258 | * refs so that we can remove children safely there ? | |
1259 | */ | |
42077599 | 1260 | if (cl->children || cl->filter_cnt) |
1da177e4 | 1261 | return -EBUSY; |
87990467 | 1262 | |
160d5e10 | 1263 | if (!cl->level && htb_parent_last_child(cl)) { |
3511c913 | 1264 | new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
bb949fbd | 1265 | cl->parent->common.classid); |
160d5e10 JP |
1266 | last_child = 1; |
1267 | } | |
1268 | ||
1da177e4 | 1269 | sch_tree_lock(sch); |
87990467 | 1270 | |
814a175e | 1271 | if (!cl->level) { |
2ccccf5f WC |
1272 | unsigned int qlen = cl->un.leaf.q->q.qlen; |
1273 | unsigned int backlog = cl->un.leaf.q->qstats.backlog; | |
1274 | ||
814a175e | 1275 | qdisc_reset(cl->un.leaf.q); |
2ccccf5f | 1276 | qdisc_tree_reduce_backlog(cl->un.leaf.q, qlen, backlog); |
814a175e PM |
1277 | } |
1278 | ||
f4c1f3e0 PM |
1279 | /* delete from hash and active; remainder in destroy_class */ |
1280 | qdisc_class_hash_remove(&q->clhash, &cl->common); | |
26b284de JP |
1281 | if (cl->parent) |
1282 | cl->parent->children--; | |
c38c83cb | 1283 | |
1da177e4 | 1284 | if (cl->prio_activity) |
87990467 | 1285 | htb_deactivate(q, cl); |
1da177e4 | 1286 | |
fbd8f137 | 1287 | if (cl->cmode != HTB_CAN_SEND) |
c9364636 ED |
1288 | htb_safe_rb_erase(&cl->pq_node, |
1289 | &q->hlevel[cl->level].wait_pq); | |
fbd8f137 | 1290 | |
160d5e10 | 1291 | if (last_child) |
3ba08b00 | 1292 | htb_parent_to_leaf(q, cl, new_q); |
160d5e10 | 1293 | |
7cd0a638 JP |
1294 | BUG_ON(--cl->refcnt == 0); |
1295 | /* | |
1296 | * This shouldn't happen: we "hold" one cops->get() when called | |
1297 | * from tc_ctl_tclass; the destroy method is done from cops->put(). | |
1298 | */ | |
1da177e4 LT |
1299 | |
1300 | sch_tree_unlock(sch); | |
1301 | return 0; | |
1302 | } | |
1303 | ||
1304 | static void htb_put(struct Qdisc *sch, unsigned long arg) | |
1305 | { | |
87990467 | 1306 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1307 | |
1308 | if (--cl->refcnt == 0) | |
87990467 | 1309 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1310 | } |
1311 | ||
87990467 | 1312 | static int htb_change_class(struct Qdisc *sch, u32 classid, |
1e90474c | 1313 | u32 parentid, struct nlattr **tca, |
87990467 | 1314 | unsigned long *arg) |
1da177e4 LT |
1315 | { |
1316 | int err = -EINVAL; | |
1317 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1318 | struct htb_class *cl = (struct htb_class *)*arg, *parent; |
1e90474c | 1319 | struct nlattr *opt = tca[TCA_OPTIONS]; |
6906f4ed | 1320 | struct nlattr *tb[TCA_HTB_MAX + 1]; |
1da177e4 | 1321 | struct tc_htb_opt *hopt; |
df62cdf3 | 1322 | u64 rate64, ceil64; |
1da177e4 LT |
1323 | |
1324 | /* extract all subattrs from opt attr */ | |
cee63723 PM |
1325 | if (!opt) |
1326 | goto failure; | |
1327 | ||
e18434c4 | 1328 | err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy); |
cee63723 PM |
1329 | if (err < 0) |
1330 | goto failure; | |
1331 | ||
1332 | err = -EINVAL; | |
27a3421e | 1333 | if (tb[TCA_HTB_PARMS] == NULL) |
1da177e4 | 1334 | goto failure; |
1da177e4 | 1335 | |
87990467 SH |
1336 | parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch); |
1337 | ||
1e90474c | 1338 | hopt = nla_data(tb[TCA_HTB_PARMS]); |
196d97f6 | 1339 | if (!hopt->rate.rate || !hopt->ceil.rate) |
87990467 | 1340 | goto failure; |
1da177e4 | 1341 | |
8a8e3d84 | 1342 | /* Keeping backward compatible with rate_table based iproute2 tc */ |
6b1dd856 YY |
1343 | if (hopt->rate.linklayer == TC_LINKLAYER_UNAWARE) |
1344 | qdisc_put_rtab(qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB])); | |
1345 | ||
1346 | if (hopt->ceil.linklayer == TC_LINKLAYER_UNAWARE) | |
1347 | qdisc_put_rtab(qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB])); | |
8a8e3d84 | 1348 | |
87990467 | 1349 | if (!cl) { /* new class */ |
1da177e4 | 1350 | struct Qdisc *new_q; |
3696f625 | 1351 | int prio; |
ee39e10c | 1352 | struct { |
1e90474c | 1353 | struct nlattr nla; |
ee39e10c PM |
1354 | struct gnet_estimator opt; |
1355 | } est = { | |
1e90474c PM |
1356 | .nla = { |
1357 | .nla_len = nla_attr_size(sizeof(est.opt)), | |
1358 | .nla_type = TCA_RATE, | |
ee39e10c PM |
1359 | }, |
1360 | .opt = { | |
1361 | /* 4s interval, 16s averaging constant */ | |
1362 | .interval = 2, | |
1363 | .ewma_log = 2, | |
1364 | }, | |
1365 | }; | |
3696f625 | 1366 | |
1da177e4 | 1367 | /* check for valid classid */ |
f64f9e71 JP |
1368 | if (!classid || TC_H_MAJ(classid ^ sch->handle) || |
1369 | htb_find(classid, sch)) | |
1da177e4 LT |
1370 | goto failure; |
1371 | ||
1372 | /* check maximal depth */ | |
1373 | if (parent && parent->parent && parent->parent->level < 2) { | |
cc7ec456 | 1374 | pr_err("htb: tree is too deep\n"); |
1da177e4 LT |
1375 | goto failure; |
1376 | } | |
1377 | err = -ENOBUFS; | |
cc7ec456 ED |
1378 | cl = kzalloc(sizeof(*cl), GFP_KERNEL); |
1379 | if (!cl) | |
1da177e4 | 1380 | goto failure; |
87990467 | 1381 | |
64153ce0 | 1382 | if (htb_rate_est || tca[TCA_RATE]) { |
22e0f8b9 JF |
1383 | err = gen_new_estimator(&cl->bstats, NULL, |
1384 | &cl->rate_est, | |
edb09eb1 ED |
1385 | NULL, |
1386 | qdisc_root_sleeping_running(sch), | |
64153ce0 ED |
1387 | tca[TCA_RATE] ? : &est.nla); |
1388 | if (err) { | |
1389 | kfree(cl); | |
1390 | goto failure; | |
1391 | } | |
71bcb09a SH |
1392 | } |
1393 | ||
1da177e4 | 1394 | cl->refcnt = 1; |
42077599 | 1395 | cl->children = 0; |
1da177e4 | 1396 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); |
3696f625 SH |
1397 | RB_CLEAR_NODE(&cl->pq_node); |
1398 | ||
1399 | for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) | |
1400 | RB_CLEAR_NODE(&cl->node[prio]); | |
1da177e4 LT |
1401 | |
1402 | /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL) | |
cc7ec456 ED |
1403 | * so that can't be used inside of sch_tree_lock |
1404 | * -- thanks to Karlis Peisenieks | |
1405 | */ | |
3511c913 | 1406 | new_q = qdisc_create_dflt(sch->dev_queue, |
bb949fbd | 1407 | &pfifo_qdisc_ops, classid); |
1da177e4 LT |
1408 | sch_tree_lock(sch); |
1409 | if (parent && !parent->level) { | |
256d61b8 | 1410 | unsigned int qlen = parent->un.leaf.q->q.qlen; |
2ccccf5f | 1411 | unsigned int backlog = parent->un.leaf.q->qstats.backlog; |
256d61b8 | 1412 | |
1da177e4 | 1413 | /* turn parent into inner node */ |
256d61b8 | 1414 | qdisc_reset(parent->un.leaf.q); |
2ccccf5f | 1415 | qdisc_tree_reduce_backlog(parent->un.leaf.q, qlen, backlog); |
87990467 SH |
1416 | qdisc_destroy(parent->un.leaf.q); |
1417 | if (parent->prio_activity) | |
1418 | htb_deactivate(q, parent); | |
1da177e4 LT |
1419 | |
1420 | /* remove from evt list because of level change */ | |
1421 | if (parent->cmode != HTB_CAN_SEND) { | |
c9364636 | 1422 | htb_safe_rb_erase(&parent->pq_node, &q->hlevel[0].wait_pq); |
1da177e4 LT |
1423 | parent->cmode = HTB_CAN_SEND; |
1424 | } | |
1425 | parent->level = (parent->parent ? parent->parent->level | |
87990467 SH |
1426 | : TC_HTB_MAXDEPTH) - 1; |
1427 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1da177e4 LT |
1428 | } |
1429 | /* leaf (we) needs elementary qdisc */ | |
1430 | cl->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
1431 | ||
f4c1f3e0 | 1432 | cl->common.classid = classid; |
87990467 | 1433 | cl->parent = parent; |
1da177e4 LT |
1434 | |
1435 | /* set class to be in HTB_CAN_SEND state */ | |
b9a7afde JP |
1436 | cl->tokens = PSCHED_TICKS2NS(hopt->buffer); |
1437 | cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer); | |
5343a7f8 | 1438 | cl->mbuffer = 60ULL * NSEC_PER_SEC; /* 1min */ |
d2de875c | 1439 | cl->t_c = ktime_get_ns(); |
1da177e4 LT |
1440 | cl->cmode = HTB_CAN_SEND; |
1441 | ||
1442 | /* attach to the hash list and parent's family */ | |
f4c1f3e0 | 1443 | qdisc_class_hash_insert(&q->clhash, &cl->common); |
42077599 PM |
1444 | if (parent) |
1445 | parent->children++; | |
ee39e10c | 1446 | } else { |
71bcb09a | 1447 | if (tca[TCA_RATE]) { |
22e0f8b9 JF |
1448 | err = gen_replace_estimator(&cl->bstats, NULL, |
1449 | &cl->rate_est, | |
edb09eb1 ED |
1450 | NULL, |
1451 | qdisc_root_sleeping_running(sch), | |
71bcb09a SH |
1452 | tca[TCA_RATE]); |
1453 | if (err) | |
1454 | return err; | |
1455 | } | |
87990467 | 1456 | sch_tree_lock(sch); |
ee39e10c | 1457 | } |
1da177e4 | 1458 | |
1598f7cb YY |
1459 | rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0; |
1460 | ||
1461 | ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0; | |
1462 | ||
1463 | psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64); | |
1464 | psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64); | |
1465 | ||
1da177e4 | 1466 | /* it used to be a nasty bug here, we have to check that node |
cc7ec456 ED |
1467 | * is really leaf before changing cl->un.leaf ! |
1468 | */ | |
1da177e4 | 1469 | if (!cl->level) { |
1598f7cb YY |
1470 | u64 quantum = cl->rate.rate_bytes_ps; |
1471 | ||
1472 | do_div(quantum, q->rate2quantum); | |
1473 | cl->quantum = min_t(u64, quantum, INT_MAX); | |
1474 | ||
c19f7a34 | 1475 | if (!hopt->quantum && cl->quantum < 1000) { |
c17988a9 YY |
1476 | pr_warn("HTB: quantum of class %X is small. Consider r2q change.\n", |
1477 | cl->common.classid); | |
c19f7a34 | 1478 | cl->quantum = 1000; |
1da177e4 | 1479 | } |
c19f7a34 | 1480 | if (!hopt->quantum && cl->quantum > 200000) { |
c17988a9 YY |
1481 | pr_warn("HTB: quantum of class %X is big. Consider r2q change.\n", |
1482 | cl->common.classid); | |
c19f7a34 | 1483 | cl->quantum = 200000; |
1da177e4 LT |
1484 | } |
1485 | if (hopt->quantum) | |
c19f7a34 JP |
1486 | cl->quantum = hopt->quantum; |
1487 | if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO) | |
1488 | cl->prio = TC_HTB_NUMPRIO - 1; | |
1da177e4 LT |
1489 | } |
1490 | ||
324f5aa5 | 1491 | cl->buffer = PSCHED_TICKS2NS(hopt->buffer); |
f3ad857e | 1492 | cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer); |
56b765b7 | 1493 | |
1da177e4 LT |
1494 | sch_tree_unlock(sch); |
1495 | ||
f4c1f3e0 PM |
1496 | qdisc_class_hash_grow(sch, &q->clhash); |
1497 | ||
1da177e4 LT |
1498 | *arg = (unsigned long)cl; |
1499 | return 0; | |
1500 | ||
1501 | failure: | |
1da177e4 LT |
1502 | return err; |
1503 | } | |
1504 | ||
25d8c0d5 JF |
1505 | static struct tcf_proto __rcu **htb_find_tcf(struct Qdisc *sch, |
1506 | unsigned long arg) | |
1da177e4 LT |
1507 | { |
1508 | struct htb_sched *q = qdisc_priv(sch); | |
1509 | struct htb_class *cl = (struct htb_class *)arg; | |
25d8c0d5 | 1510 | struct tcf_proto __rcu **fl = cl ? &cl->filter_list : &q->filter_list; |
3bf72957 | 1511 | |
1da177e4 LT |
1512 | return fl; |
1513 | } | |
1514 | ||
1515 | static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, | |
87990467 | 1516 | u32 classid) |
1da177e4 | 1517 | { |
87990467 | 1518 | struct htb_class *cl = htb_find(classid, sch); |
3bf72957 | 1519 | |
1da177e4 | 1520 | /*if (cl && !cl->level) return 0; |
cc7ec456 ED |
1521 | * The line above used to be there to prevent attaching filters to |
1522 | * leaves. But at least tc_index filter uses this just to get class | |
1523 | * for other reasons so that we have to allow for it. | |
1524 | * ---- | |
1525 | * 19.6.2002 As Werner explained it is ok - bind filter is just | |
1526 | * another way to "lock" the class - unlike "get" this lock can | |
1527 | * be broken by class during destroy IIUC. | |
1da177e4 | 1528 | */ |
87990467 SH |
1529 | if (cl) |
1530 | cl->filter_cnt++; | |
1da177e4 LT |
1531 | return (unsigned long)cl; |
1532 | } | |
1533 | ||
1534 | static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) | |
1535 | { | |
1da177e4 | 1536 | struct htb_class *cl = (struct htb_class *)arg; |
3bf72957 | 1537 | |
87990467 SH |
1538 | if (cl) |
1539 | cl->filter_cnt--; | |
1da177e4 LT |
1540 | } |
1541 | ||
1542 | static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
1543 | { | |
1544 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 1545 | struct htb_class *cl; |
f4c1f3e0 | 1546 | unsigned int i; |
1da177e4 LT |
1547 | |
1548 | if (arg->stop) | |
1549 | return; | |
1550 | ||
f4c1f3e0 | 1551 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 1552 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { |
1da177e4 LT |
1553 | if (arg->count < arg->skip) { |
1554 | arg->count++; | |
1555 | continue; | |
1556 | } | |
1557 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
1558 | arg->stop = 1; | |
1559 | return; | |
1560 | } | |
1561 | arg->count++; | |
1562 | } | |
1563 | } | |
1564 | } | |
1565 | ||
20fea08b | 1566 | static const struct Qdisc_class_ops htb_class_ops = { |
1da177e4 LT |
1567 | .graft = htb_graft, |
1568 | .leaf = htb_leaf, | |
256d61b8 | 1569 | .qlen_notify = htb_qlen_notify, |
1da177e4 LT |
1570 | .get = htb_get, |
1571 | .put = htb_put, | |
1572 | .change = htb_change_class, | |
1573 | .delete = htb_delete, | |
1574 | .walk = htb_walk, | |
1575 | .tcf_chain = htb_find_tcf, | |
1576 | .bind_tcf = htb_bind_filter, | |
1577 | .unbind_tcf = htb_unbind_filter, | |
1578 | .dump = htb_dump_class, | |
1579 | .dump_stats = htb_dump_class_stats, | |
1580 | }; | |
1581 | ||
20fea08b | 1582 | static struct Qdisc_ops htb_qdisc_ops __read_mostly = { |
1da177e4 LT |
1583 | .cl_ops = &htb_class_ops, |
1584 | .id = "htb", | |
1585 | .priv_size = sizeof(struct htb_sched), | |
1586 | .enqueue = htb_enqueue, | |
1587 | .dequeue = htb_dequeue, | |
77be155c | 1588 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1589 | .init = htb_init, |
1590 | .reset = htb_reset, | |
1591 | .destroy = htb_destroy, | |
1da177e4 LT |
1592 | .dump = htb_dump, |
1593 | .owner = THIS_MODULE, | |
1594 | }; | |
1595 | ||
1596 | static int __init htb_module_init(void) | |
1597 | { | |
87990467 | 1598 | return register_qdisc(&htb_qdisc_ops); |
1da177e4 | 1599 | } |
87990467 | 1600 | static void __exit htb_module_exit(void) |
1da177e4 | 1601 | { |
87990467 | 1602 | unregister_qdisc(&htb_qdisc_ops); |
1da177e4 | 1603 | } |
87990467 | 1604 | |
1da177e4 LT |
1605 | module_init(htb_module_init) |
1606 | module_exit(htb_module_exit) | |
1607 | MODULE_LICENSE("GPL"); |