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> |
cf1facda | 43 | #include <net/pkt_cls.h> |
1da177e4 LT |
44 | |
45 | /* HTB algorithm. | |
46 | Author: devik@cdi.cz | |
47 | ======================================================================== | |
48 | HTB is like TBF with multiple classes. It is also similar to CBQ because | |
10297b99 | 49 | it allows to assign priority to each class in hierarchy. |
1da177e4 LT |
50 | In fact it is another implementation of Floyd's formal sharing. |
51 | ||
52 | Levels: | |
10297b99 | 53 | Each class is assigned level. Leaf has ALWAYS level 0 and root |
1da177e4 LT |
54 | classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level |
55 | one less than their parent. | |
56 | */ | |
57 | ||
47083fc0 | 58 | static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */ |
87990467 | 59 | #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */ |
1da177e4 LT |
60 | |
61 | #if HTB_VER >> 16 != TC_HTB_PROTOVER | |
62 | #error "Mismatched sch_htb.c and pkt_sch.h" | |
63 | #endif | |
64 | ||
47083fc0 JDB |
65 | /* Module parameter and sysfs export */ |
66 | module_param (htb_hysteresis, int, 0640); | |
67 | MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate"); | |
68 | ||
64153ce0 ED |
69 | static int htb_rate_est = 0; /* htb classes have a default rate estimator */ |
70 | module_param(htb_rate_est, int, 0640); | |
71 | MODULE_PARM_DESC(htb_rate_est, "setup a default rate estimator (4sec 16sec) for htb classes"); | |
72 | ||
1da177e4 LT |
73 | /* used internaly to keep status of single class */ |
74 | enum htb_cmode { | |
87990467 SH |
75 | HTB_CANT_SEND, /* class can't send and can't borrow */ |
76 | HTB_MAY_BORROW, /* class can't send but may borrow */ | |
77 | HTB_CAN_SEND /* class can send */ | |
1da177e4 LT |
78 | }; |
79 | ||
c9364636 ED |
80 | struct htb_prio { |
81 | union { | |
82 | struct rb_root row; | |
83 | struct rb_root feed; | |
84 | }; | |
85 | struct rb_node *ptr; | |
86 | /* When class changes from state 1->2 and disconnects from | |
87 | * parent's feed then we lost ptr value and start from the | |
88 | * first child again. Here we store classid of the | |
89 | * last valid ptr (used when ptr is NULL). | |
90 | */ | |
91 | u32 last_ptr_id; | |
92 | }; | |
93 | ||
ca4ec90b ED |
94 | /* interior & leaf nodes; props specific to leaves are marked L: |
95 | * To reduce false sharing, place mostly read fields at beginning, | |
96 | * and mostly written ones at the end. | |
97 | */ | |
87990467 | 98 | struct htb_class { |
f4c1f3e0 | 99 | struct Qdisc_class_common common; |
ca4ec90b ED |
100 | struct psched_ratecfg rate; |
101 | struct psched_ratecfg ceil; | |
102 | s64 buffer, cbuffer;/* token bucket depth/rate */ | |
103 | s64 mbuffer; /* max wait time */ | |
cbd37556 | 104 | u32 prio; /* these two are used only by leaves... */ |
ca4ec90b ED |
105 | int quantum; /* but stored for parent-to-leaf return */ |
106 | ||
25d8c0d5 | 107 | struct tcf_proto __rcu *filter_list; /* class attached filters */ |
6529eaba | 108 | struct tcf_block *block; |
ca4ec90b ED |
109 | int filter_cnt; |
110 | int refcnt; /* usage count of this class */ | |
111 | ||
112 | int level; /* our level (see above) */ | |
113 | unsigned int children; | |
114 | struct htb_class *parent; /* parent class */ | |
115 | ||
1c0d32fd | 116 | struct net_rate_estimator __rcu *rate_est; |
1da177e4 | 117 | |
ca4ec90b ED |
118 | /* |
119 | * Written often fields | |
120 | */ | |
121 | struct gnet_stats_basic_packed bstats; | |
ca4ec90b | 122 | struct tc_htb_xstats xstats; /* our special stats */ |
87990467 | 123 | |
ca4ec90b ED |
124 | /* token bucket parameters */ |
125 | s64 tokens, ctokens;/* current number of tokens */ | |
126 | s64 t_c; /* checkpoint time */ | |
c19f7a34 | 127 | |
87990467 SH |
128 | union { |
129 | struct htb_class_leaf { | |
87990467 | 130 | struct list_head drop_list; |
c9364636 ED |
131 | int deficit[TC_HTB_MAXDEPTH]; |
132 | struct Qdisc *q; | |
87990467 SH |
133 | } leaf; |
134 | struct htb_class_inner { | |
c9364636 | 135 | struct htb_prio clprio[TC_HTB_NUMPRIO]; |
87990467 SH |
136 | } inner; |
137 | } un; | |
ca4ec90b | 138 | s64 pq_key; |
87990467 | 139 | |
ca4ec90b ED |
140 | int prio_activity; /* for which prios are we active */ |
141 | enum htb_cmode cmode; /* current mode of the class */ | |
142 | struct rb_node pq_node; /* node for event queue */ | |
143 | struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ | |
338ed9b4 ED |
144 | |
145 | unsigned int drops ____cacheline_aligned_in_smp; | |
1da177e4 LT |
146 | }; |
147 | ||
c9364636 ED |
148 | struct htb_level { |
149 | struct rb_root wait_pq; | |
150 | struct htb_prio hprio[TC_HTB_NUMPRIO]; | |
151 | }; | |
152 | ||
87990467 | 153 | struct htb_sched { |
f4c1f3e0 | 154 | struct Qdisc_class_hash clhash; |
c9364636 ED |
155 | int defcls; /* class where unclassified flows go to */ |
156 | int rate2quantum; /* quant = rate / rate2quantum */ | |
1da177e4 | 157 | |
c9364636 | 158 | /* filters for qdisc itself */ |
25d8c0d5 | 159 | struct tcf_proto __rcu *filter_list; |
6529eaba | 160 | struct tcf_block *block; |
1da177e4 | 161 | |
c9364636 ED |
162 | #define HTB_WARN_TOOMANYEVENTS 0x1 |
163 | unsigned int warned; /* only one warning */ | |
164 | int direct_qlen; | |
165 | struct work_struct work; | |
1da177e4 | 166 | |
c9364636 | 167 | /* non shaped skbs; let them go directly thru */ |
48da34b7 | 168 | struct qdisc_skb_head direct_queue; |
c9364636 | 169 | long direct_pkts; |
1da177e4 | 170 | |
c9364636 | 171 | struct qdisc_watchdog watchdog; |
1da177e4 | 172 | |
c9364636 ED |
173 | s64 now; /* cached dequeue time */ |
174 | struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */ | |
1da177e4 | 175 | |
c9364636 ED |
176 | /* time of nearest event per level (row) */ |
177 | s64 near_ev_cache[TC_HTB_MAXDEPTH]; | |
87990467 | 178 | |
c9364636 | 179 | int row_mask[TC_HTB_MAXDEPTH]; |
e82181de | 180 | |
c9364636 | 181 | struct htb_level hlevel[TC_HTB_MAXDEPTH]; |
1da177e4 LT |
182 | }; |
183 | ||
1da177e4 | 184 | /* find class in global hash table using given handle */ |
87990467 | 185 | static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch) |
1da177e4 LT |
186 | { |
187 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 188 | struct Qdisc_class_common *clc; |
0cef296d | 189 | |
f4c1f3e0 PM |
190 | clc = qdisc_class_find(&q->clhash, handle); |
191 | if (clc == NULL) | |
1da177e4 | 192 | return NULL; |
f4c1f3e0 | 193 | return container_of(clc, struct htb_class, common); |
1da177e4 LT |
194 | } |
195 | ||
196 | /** | |
197 | * htb_classify - classify a packet into class | |
198 | * | |
199 | * It returns NULL if the packet should be dropped or -1 if the packet | |
200 | * should be passed directly thru. In all other cases leaf class is returned. | |
201 | * We allow direct class selection by classid in priority. The we examine | |
202 | * filters in qdisc and in inner nodes (if higher filter points to the inner | |
203 | * node). If we end up with classid MAJOR:0 we enqueue the skb into special | |
10297b99 | 204 | * internal fifo (direct). These packets then go directly thru. If we still |
25985edc | 205 | * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful |
1da177e4 LT |
206 | * then finish and return direct queue. |
207 | */ | |
cc7ec456 | 208 | #define HTB_DIRECT ((struct htb_class *)-1L) |
1da177e4 | 209 | |
87990467 SH |
210 | static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, |
211 | int *qerr) | |
1da177e4 LT |
212 | { |
213 | struct htb_sched *q = qdisc_priv(sch); | |
214 | struct htb_class *cl; | |
215 | struct tcf_result res; | |
216 | struct tcf_proto *tcf; | |
217 | int result; | |
218 | ||
219 | /* allow to select class by setting skb->priority to valid classid; | |
cc7ec456 ED |
220 | * note that nfmark can be used too by attaching filter fw with no |
221 | * rules in it | |
222 | */ | |
1da177e4 | 223 | if (skb->priority == sch->handle) |
87990467 | 224 | return HTB_DIRECT; /* X:0 (direct flow) selected */ |
cc7ec456 | 225 | cl = htb_find(skb->priority, sch); |
29824310 HM |
226 | if (cl) { |
227 | if (cl->level == 0) | |
228 | return cl; | |
229 | /* Start with inner filter chain if a non-leaf class is selected */ | |
25d8c0d5 | 230 | tcf = rcu_dereference_bh(cl->filter_list); |
29824310 | 231 | } else { |
25d8c0d5 | 232 | tcf = rcu_dereference_bh(q->filter_list); |
29824310 | 233 | } |
1da177e4 | 234 | |
c27f339a | 235 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
87d83093 | 236 | while (tcf && (result = tcf_classify(skb, tcf, &res, false)) >= 0) { |
1da177e4 LT |
237 | #ifdef CONFIG_NET_CLS_ACT |
238 | switch (result) { | |
239 | case TC_ACT_QUEUED: | |
87990467 | 240 | case TC_ACT_STOLEN: |
e25ea21f | 241 | case TC_ACT_TRAP: |
378a2f09 | 242 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
1da177e4 LT |
243 | case TC_ACT_SHOT: |
244 | return NULL; | |
245 | } | |
1da177e4 | 246 | #endif |
cc7ec456 ED |
247 | cl = (void *)res.class; |
248 | if (!cl) { | |
1da177e4 | 249 | if (res.classid == sch->handle) |
87990467 | 250 | return HTB_DIRECT; /* X:0 (direct flow) */ |
cc7ec456 ED |
251 | cl = htb_find(res.classid, sch); |
252 | if (!cl) | |
87990467 | 253 | break; /* filter selected invalid classid */ |
1da177e4 LT |
254 | } |
255 | if (!cl->level) | |
87990467 | 256 | return cl; /* we hit leaf; return it */ |
1da177e4 LT |
257 | |
258 | /* we have got inner class; apply inner filter chain */ | |
25d8c0d5 | 259 | tcf = rcu_dereference_bh(cl->filter_list); |
1da177e4 LT |
260 | } |
261 | /* classification failed; try to use default class */ | |
87990467 | 262 | cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch); |
1da177e4 | 263 | if (!cl || cl->level) |
87990467 | 264 | return HTB_DIRECT; /* bad default .. this is safe bet */ |
1da177e4 LT |
265 | return cl; |
266 | } | |
267 | ||
1da177e4 LT |
268 | /** |
269 | * htb_add_to_id_tree - adds class to the round robin list | |
270 | * | |
271 | * Routine adds class to the list (actually tree) sorted by classid. | |
272 | * Make sure that class is not already on such list for given prio. | |
273 | */ | |
87990467 SH |
274 | static void htb_add_to_id_tree(struct rb_root *root, |
275 | struct htb_class *cl, int prio) | |
1da177e4 LT |
276 | { |
277 | struct rb_node **p = &root->rb_node, *parent = NULL; | |
3bf72957 | 278 | |
1da177e4 | 279 | while (*p) { |
87990467 SH |
280 | struct htb_class *c; |
281 | parent = *p; | |
1da177e4 | 282 | c = rb_entry(parent, struct htb_class, node[prio]); |
3bf72957 | 283 | |
f4c1f3e0 | 284 | if (cl->common.classid > c->common.classid) |
1da177e4 | 285 | p = &parent->rb_right; |
87990467 | 286 | else |
1da177e4 LT |
287 | p = &parent->rb_left; |
288 | } | |
289 | rb_link_node(&cl->node[prio], parent, p); | |
290 | rb_insert_color(&cl->node[prio], root); | |
291 | } | |
292 | ||
293 | /** | |
294 | * htb_add_to_wait_tree - adds class to the event queue with delay | |
295 | * | |
296 | * The class is added to priority event queue to indicate that class will | |
297 | * change its mode in cl->pq_key microseconds. Make sure that class is not | |
298 | * already in the queue. | |
299 | */ | |
87990467 | 300 | static void htb_add_to_wait_tree(struct htb_sched *q, |
56b765b7 | 301 | struct htb_class *cl, s64 delay) |
1da177e4 | 302 | { |
c9364636 | 303 | struct rb_node **p = &q->hlevel[cl->level].wait_pq.rb_node, *parent = NULL; |
3bf72957 | 304 | |
fb983d45 PM |
305 | cl->pq_key = q->now + delay; |
306 | if (cl->pq_key == q->now) | |
1da177e4 LT |
307 | cl->pq_key++; |
308 | ||
309 | /* update the nearest event cache */ | |
fb983d45 | 310 | if (q->near_ev_cache[cl->level] > cl->pq_key) |
1da177e4 | 311 | q->near_ev_cache[cl->level] = cl->pq_key; |
87990467 | 312 | |
1da177e4 | 313 | while (*p) { |
87990467 SH |
314 | struct htb_class *c; |
315 | parent = *p; | |
1da177e4 | 316 | c = rb_entry(parent, struct htb_class, pq_node); |
fb983d45 | 317 | if (cl->pq_key >= c->pq_key) |
1da177e4 | 318 | p = &parent->rb_right; |
87990467 | 319 | else |
1da177e4 LT |
320 | p = &parent->rb_left; |
321 | } | |
322 | rb_link_node(&cl->pq_node, parent, p); | |
c9364636 | 323 | rb_insert_color(&cl->pq_node, &q->hlevel[cl->level].wait_pq); |
1da177e4 LT |
324 | } |
325 | ||
326 | /** | |
327 | * htb_next_rb_node - finds next node in binary tree | |
328 | * | |
329 | * When we are past last key we return NULL. | |
330 | * Average complexity is 2 steps per call. | |
331 | */ | |
3696f625 | 332 | static inline void htb_next_rb_node(struct rb_node **n) |
1da177e4 LT |
333 | { |
334 | *n = rb_next(*n); | |
335 | } | |
336 | ||
337 | /** | |
338 | * htb_add_class_to_row - add class to its row | |
339 | * | |
340 | * The class is added to row at priorities marked in mask. | |
341 | * It does nothing if mask == 0. | |
342 | */ | |
87990467 SH |
343 | static inline void htb_add_class_to_row(struct htb_sched *q, |
344 | struct htb_class *cl, int mask) | |
1da177e4 | 345 | { |
1da177e4 LT |
346 | q->row_mask[cl->level] |= mask; |
347 | while (mask) { | |
348 | int prio = ffz(~mask); | |
349 | mask &= ~(1 << prio); | |
c9364636 | 350 | htb_add_to_id_tree(&q->hlevel[cl->level].hprio[prio].row, cl, prio); |
1da177e4 LT |
351 | } |
352 | } | |
353 | ||
3696f625 SH |
354 | /* If this triggers, it is a bug in this code, but it need not be fatal */ |
355 | static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root) | |
356 | { | |
81771b3b | 357 | if (RB_EMPTY_NODE(rb)) { |
3696f625 SH |
358 | WARN_ON(1); |
359 | } else { | |
360 | rb_erase(rb, root); | |
361 | RB_CLEAR_NODE(rb); | |
362 | } | |
363 | } | |
364 | ||
365 | ||
1da177e4 LT |
366 | /** |
367 | * htb_remove_class_from_row - removes class from its row | |
368 | * | |
369 | * The class is removed from row at priorities marked in mask. | |
370 | * It does nothing if mask == 0. | |
371 | */ | |
87990467 SH |
372 | static inline void htb_remove_class_from_row(struct htb_sched *q, |
373 | struct htb_class *cl, int mask) | |
1da177e4 LT |
374 | { |
375 | int m = 0; | |
c9364636 | 376 | struct htb_level *hlevel = &q->hlevel[cl->level]; |
3bf72957 | 377 | |
1da177e4 LT |
378 | while (mask) { |
379 | int prio = ffz(~mask); | |
c9364636 | 380 | struct htb_prio *hprio = &hlevel->hprio[prio]; |
3696f625 | 381 | |
1da177e4 | 382 | mask &= ~(1 << prio); |
c9364636 ED |
383 | if (hprio->ptr == cl->node + prio) |
384 | htb_next_rb_node(&hprio->ptr); | |
3696f625 | 385 | |
c9364636 ED |
386 | htb_safe_rb_erase(cl->node + prio, &hprio->row); |
387 | if (!hprio->row.rb_node) | |
1da177e4 LT |
388 | m |= 1 << prio; |
389 | } | |
1da177e4 LT |
390 | q->row_mask[cl->level] &= ~m; |
391 | } | |
392 | ||
393 | /** | |
394 | * htb_activate_prios - creates active classe's feed chain | |
395 | * | |
396 | * The class is connected to ancestors and/or appropriate rows | |
10297b99 | 397 | * for priorities it is participating on. cl->cmode must be new |
1da177e4 LT |
398 | * (activated) mode. It does nothing if cl->prio_activity == 0. |
399 | */ | |
87990467 | 400 | static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
401 | { |
402 | struct htb_class *p = cl->parent; | |
87990467 | 403 | long m, mask = cl->prio_activity; |
1da177e4 LT |
404 | |
405 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
406 | m = mask; |
407 | while (m) { | |
1da177e4 LT |
408 | int prio = ffz(~m); |
409 | m &= ~(1 << prio); | |
87990467 | 410 | |
c9364636 | 411 | if (p->un.inner.clprio[prio].feed.rb_node) |
1da177e4 | 412 | /* parent already has its feed in use so that |
cc7ec456 ED |
413 | * reset bit in mask as parent is already ok |
414 | */ | |
1da177e4 | 415 | mask &= ~(1 << prio); |
87990467 | 416 | |
c9364636 | 417 | htb_add_to_id_tree(&p->un.inner.clprio[prio].feed, cl, prio); |
1da177e4 | 418 | } |
1da177e4 | 419 | p->prio_activity |= mask; |
87990467 SH |
420 | cl = p; |
421 | p = cl->parent; | |
3bf72957 | 422 | |
1da177e4 LT |
423 | } |
424 | if (cl->cmode == HTB_CAN_SEND && mask) | |
87990467 | 425 | htb_add_class_to_row(q, cl, mask); |
1da177e4 LT |
426 | } |
427 | ||
428 | /** | |
429 | * htb_deactivate_prios - remove class from feed chain | |
430 | * | |
10297b99 | 431 | * cl->cmode must represent old mode (before deactivation). It does |
1da177e4 LT |
432 | * nothing if cl->prio_activity == 0. Class is removed from all feed |
433 | * chains and rows. | |
434 | */ | |
435 | static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) | |
436 | { | |
437 | struct htb_class *p = cl->parent; | |
87990467 | 438 | long m, mask = cl->prio_activity; |
1da177e4 LT |
439 | |
440 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
441 | m = mask; |
442 | mask = 0; | |
1da177e4 LT |
443 | while (m) { |
444 | int prio = ffz(~m); | |
445 | m &= ~(1 << prio); | |
87990467 | 446 | |
c9364636 | 447 | if (p->un.inner.clprio[prio].ptr == cl->node + prio) { |
1da177e4 | 448 | /* we are removing child which is pointed to from |
cc7ec456 ED |
449 | * parent feed - forget the pointer but remember |
450 | * classid | |
451 | */ | |
c9364636 ED |
452 | p->un.inner.clprio[prio].last_ptr_id = cl->common.classid; |
453 | p->un.inner.clprio[prio].ptr = NULL; | |
1da177e4 | 454 | } |
87990467 | 455 | |
c9364636 ED |
456 | htb_safe_rb_erase(cl->node + prio, |
457 | &p->un.inner.clprio[prio].feed); | |
87990467 | 458 | |
c9364636 | 459 | if (!p->un.inner.clprio[prio].feed.rb_node) |
1da177e4 LT |
460 | mask |= 1 << prio; |
461 | } | |
3bf72957 | 462 | |
1da177e4 | 463 | p->prio_activity &= ~mask; |
87990467 SH |
464 | cl = p; |
465 | p = cl->parent; | |
3bf72957 | 466 | |
1da177e4 | 467 | } |
87990467 SH |
468 | if (cl->cmode == HTB_CAN_SEND && mask) |
469 | htb_remove_class_from_row(q, cl, mask); | |
1da177e4 LT |
470 | } |
471 | ||
56b765b7 | 472 | static inline s64 htb_lowater(const struct htb_class *cl) |
18a63e86 | 473 | { |
47083fc0 JDB |
474 | if (htb_hysteresis) |
475 | return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0; | |
476 | else | |
477 | return 0; | |
18a63e86 | 478 | } |
56b765b7 | 479 | static inline s64 htb_hiwater(const struct htb_class *cl) |
18a63e86 | 480 | { |
47083fc0 JDB |
481 | if (htb_hysteresis) |
482 | return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0; | |
483 | else | |
484 | return 0; | |
18a63e86 | 485 | } |
47083fc0 | 486 | |
18a63e86 | 487 | |
1da177e4 LT |
488 | /** |
489 | * htb_class_mode - computes and returns current class mode | |
490 | * | |
491 | * It computes cl's mode at time cl->t_c+diff and returns it. If mode | |
492 | * is not HTB_CAN_SEND then cl->pq_key is updated to time difference | |
10297b99 | 493 | * from now to time when cl will change its state. |
1da177e4 | 494 | * Also it is worth to note that class mode doesn't change simply |
10297b99 | 495 | * at cl->{c,}tokens == 0 but there can rather be hysteresis of |
1da177e4 LT |
496 | * 0 .. -cl->{c,}buffer range. It is meant to limit number of |
497 | * mode transitions per time unit. The speed gain is about 1/6. | |
498 | */ | |
87990467 | 499 | static inline enum htb_cmode |
56b765b7 | 500 | htb_class_mode(struct htb_class *cl, s64 *diff) |
1da177e4 | 501 | { |
56b765b7 | 502 | s64 toks; |
1da177e4 | 503 | |
87990467 SH |
504 | if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) { |
505 | *diff = -toks; | |
506 | return HTB_CANT_SEND; | |
507 | } | |
18a63e86 | 508 | |
87990467 SH |
509 | if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl)) |
510 | return HTB_CAN_SEND; | |
1da177e4 | 511 | |
87990467 SH |
512 | *diff = -toks; |
513 | return HTB_MAY_BORROW; | |
1da177e4 LT |
514 | } |
515 | ||
516 | /** | |
517 | * htb_change_class_mode - changes classe's mode | |
518 | * | |
519 | * This should be the only way how to change classe's mode under normal | |
520 | * cirsumstances. Routine will update feed lists linkage, change mode | |
521 | * and add class to the wait event queue if appropriate. New mode should | |
522 | * be different from old one and cl->pq_key has to be valid if changing | |
523 | * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). | |
524 | */ | |
87990467 | 525 | static void |
56b765b7 | 526 | htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, s64 *diff) |
87990467 SH |
527 | { |
528 | enum htb_cmode new_mode = htb_class_mode(cl, diff); | |
1da177e4 LT |
529 | |
530 | if (new_mode == cl->cmode) | |
87990467 SH |
531 | return; |
532 | ||
533 | if (cl->prio_activity) { /* not necessary: speed optimization */ | |
534 | if (cl->cmode != HTB_CANT_SEND) | |
535 | htb_deactivate_prios(q, cl); | |
1da177e4 | 536 | cl->cmode = new_mode; |
87990467 SH |
537 | if (new_mode != HTB_CANT_SEND) |
538 | htb_activate_prios(q, cl); | |
539 | } else | |
1da177e4 LT |
540 | cl->cmode = new_mode; |
541 | } | |
542 | ||
543 | /** | |
10297b99 | 544 | * htb_activate - inserts leaf cl into appropriate active feeds |
1da177e4 LT |
545 | * |
546 | * Routine learns (new) priority of leaf and activates feed chain | |
547 | * for the prio. It can be called on already active leaf safely. | |
548 | * It also adds leaf into droplist. | |
549 | */ | |
87990467 | 550 | static inline void htb_activate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 551 | { |
547b792c | 552 | WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen); |
3bf72957 | 553 | |
1da177e4 | 554 | if (!cl->prio_activity) { |
c19f7a34 | 555 | cl->prio_activity = 1 << cl->prio; |
87990467 SH |
556 | htb_activate_prios(q, cl); |
557 | list_add_tail(&cl->un.leaf.drop_list, | |
c19f7a34 | 558 | q->drops + cl->prio); |
1da177e4 LT |
559 | } |
560 | } | |
561 | ||
562 | /** | |
10297b99 | 563 | * htb_deactivate - remove leaf cl from active feeds |
1da177e4 LT |
564 | * |
565 | * Make sure that leaf is active. In the other words it can't be called | |
566 | * with non-active leaf. It also removes class from the drop list. | |
567 | */ | |
87990467 | 568 | static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 569 | { |
547b792c | 570 | WARN_ON(!cl->prio_activity); |
3bf72957 | 571 | |
87990467 | 572 | htb_deactivate_prios(q, cl); |
1da177e4 LT |
573 | cl->prio_activity = 0; |
574 | list_del_init(&cl->un.leaf.drop_list); | |
575 | } | |
576 | ||
48da34b7 FW |
577 | static void htb_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch, |
578 | struct qdisc_skb_head *qh) | |
579 | { | |
580 | struct sk_buff *last = qh->tail; | |
581 | ||
582 | if (last) { | |
583 | skb->next = NULL; | |
584 | last->next = skb; | |
585 | qh->tail = skb; | |
586 | } else { | |
587 | qh->tail = skb; | |
588 | qh->head = skb; | |
589 | } | |
590 | qh->qlen++; | |
591 | } | |
592 | ||
520ac30f ED |
593 | static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
594 | struct sk_buff **to_free) | |
1da177e4 | 595 | { |
f30ab418 | 596 | int uninitialized_var(ret); |
87990467 SH |
597 | struct htb_sched *q = qdisc_priv(sch); |
598 | struct htb_class *cl = htb_classify(skb, sch, &ret); | |
599 | ||
600 | if (cl == HTB_DIRECT) { | |
601 | /* enqueue to helper queue */ | |
602 | if (q->direct_queue.qlen < q->direct_qlen) { | |
48da34b7 | 603 | htb_enqueue_tail(skb, sch, &q->direct_queue); |
87990467 SH |
604 | q->direct_pkts++; |
605 | } else { | |
520ac30f | 606 | return qdisc_drop(skb, sch, to_free); |
87990467 | 607 | } |
1da177e4 | 608 | #ifdef CONFIG_NET_CLS_ACT |
87990467 | 609 | } else if (!cl) { |
c27f339a | 610 | if (ret & __NET_XMIT_BYPASS) |
25331d6c | 611 | qdisc_qstats_drop(sch); |
520ac30f | 612 | __qdisc_drop(skb, to_free); |
87990467 | 613 | return ret; |
1da177e4 | 614 | #endif |
520ac30f ED |
615 | } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q, |
616 | to_free)) != NET_XMIT_SUCCESS) { | |
378a2f09 | 617 | if (net_xmit_drop_count(ret)) { |
25331d6c | 618 | qdisc_qstats_drop(sch); |
338ed9b4 | 619 | cl->drops++; |
378a2f09 | 620 | } |
69747650 | 621 | return ret; |
87990467 | 622 | } else { |
87990467 SH |
623 | htb_activate(q, cl); |
624 | } | |
625 | ||
431e3a8e | 626 | qdisc_qstats_backlog_inc(sch, skb); |
87990467 | 627 | sch->q.qlen++; |
87990467 | 628 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
629 | } |
630 | ||
56b765b7 | 631 | static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, s64 diff) |
59e4220a | 632 | { |
56b765b7 | 633 | s64 toks = diff + cl->tokens; |
59e4220a JP |
634 | |
635 | if (toks > cl->buffer) | |
636 | toks = cl->buffer; | |
292f1c7f | 637 | toks -= (s64) psched_l2t_ns(&cl->rate, bytes); |
59e4220a JP |
638 | if (toks <= -cl->mbuffer) |
639 | toks = 1 - cl->mbuffer; | |
640 | ||
641 | cl->tokens = toks; | |
642 | } | |
643 | ||
56b765b7 | 644 | static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, s64 diff) |
59e4220a | 645 | { |
56b765b7 | 646 | s64 toks = diff + cl->ctokens; |
59e4220a JP |
647 | |
648 | if (toks > cl->cbuffer) | |
649 | toks = cl->cbuffer; | |
292f1c7f | 650 | toks -= (s64) psched_l2t_ns(&cl->ceil, bytes); |
59e4220a JP |
651 | if (toks <= -cl->mbuffer) |
652 | toks = 1 - cl->mbuffer; | |
653 | ||
654 | cl->ctokens = toks; | |
655 | } | |
656 | ||
1da177e4 LT |
657 | /** |
658 | * htb_charge_class - charges amount "bytes" to leaf and ancestors | |
659 | * | |
660 | * Routine assumes that packet "bytes" long was dequeued from leaf cl | |
661 | * borrowing from "level". It accounts bytes to ceil leaky bucket for | |
662 | * leaf and all ancestors and to rate bucket for ancestors at levels | |
663 | * "level" and higher. It also handles possible change of mode resulting | |
664 | * from the update. Note that mode can also increase here (MAY_BORROW to | |
665 | * CAN_SEND) because we can use more precise clock that event queue here. | |
666 | * In such case we remove class from event queue first. | |
667 | */ | |
87990467 | 668 | static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, |
c9726d68 | 669 | int level, struct sk_buff *skb) |
87990467 | 670 | { |
0abf77e5 | 671 | int bytes = qdisc_pkt_len(skb); |
1da177e4 | 672 | enum htb_cmode old_mode; |
56b765b7 | 673 | s64 diff; |
1da177e4 LT |
674 | |
675 | while (cl) { | |
56b765b7 | 676 | diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); |
1da177e4 | 677 | if (cl->level >= level) { |
87990467 SH |
678 | if (cl->level == level) |
679 | cl->xstats.lends++; | |
59e4220a | 680 | htb_accnt_tokens(cl, bytes, diff); |
1da177e4 LT |
681 | } else { |
682 | cl->xstats.borrows++; | |
87990467 | 683 | cl->tokens += diff; /* we moved t_c; update tokens */ |
1da177e4 | 684 | } |
59e4220a | 685 | htb_accnt_ctokens(cl, bytes, diff); |
1da177e4 | 686 | cl->t_c = q->now; |
1da177e4 | 687 | |
87990467 SH |
688 | old_mode = cl->cmode; |
689 | diff = 0; | |
690 | htb_change_class_mode(q, cl, &diff); | |
1da177e4 LT |
691 | if (old_mode != cl->cmode) { |
692 | if (old_mode != HTB_CAN_SEND) | |
c9364636 | 693 | htb_safe_rb_erase(&cl->pq_node, &q->hlevel[cl->level].wait_pq); |
1da177e4 | 694 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 695 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 696 | } |
1da177e4 | 697 | |
bfe0d029 ED |
698 | /* update basic stats except for leaves which are already updated */ |
699 | if (cl->level) | |
700 | bstats_update(&cl->bstats, skb); | |
701 | ||
1da177e4 LT |
702 | cl = cl->parent; |
703 | } | |
704 | } | |
705 | ||
706 | /** | |
707 | * htb_do_events - make mode changes to classes at the level | |
708 | * | |
fb983d45 | 709 | * Scans event queue for pending events and applies them. Returns time of |
1224736d | 710 | * next pending event (0 for no event in pq, q->now for too many events). |
fb983d45 | 711 | * Note: Applied are events whose have cl->pq_key <= q->now. |
1da177e4 | 712 | */ |
c9364636 | 713 | static s64 htb_do_events(struct htb_sched *q, const int level, |
5343a7f8 | 714 | unsigned long start) |
1da177e4 | 715 | { |
8f3ea33a | 716 | /* don't run for longer than 2 jiffies; 2 is used instead of |
cc7ec456 ED |
717 | * 1 to simplify things when jiffy is going to be incremented |
718 | * too soon | |
719 | */ | |
a73be040 | 720 | unsigned long stop_at = start + 2; |
c9364636 ED |
721 | struct rb_root *wait_pq = &q->hlevel[level].wait_pq; |
722 | ||
8f3ea33a | 723 | while (time_before(jiffies, stop_at)) { |
1da177e4 | 724 | struct htb_class *cl; |
56b765b7 | 725 | s64 diff; |
c9364636 | 726 | struct rb_node *p = rb_first(wait_pq); |
30bdbe39 | 727 | |
87990467 SH |
728 | if (!p) |
729 | return 0; | |
1da177e4 LT |
730 | |
731 | cl = rb_entry(p, struct htb_class, pq_node); | |
fb983d45 PM |
732 | if (cl->pq_key > q->now) |
733 | return cl->pq_key; | |
734 | ||
c9364636 | 735 | htb_safe_rb_erase(p, wait_pq); |
56b765b7 | 736 | diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); |
87990467 | 737 | htb_change_class_mode(q, cl, &diff); |
1da177e4 | 738 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 739 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 740 | } |
1224736d JP |
741 | |
742 | /* too much load - let's continue after a break for scheduling */ | |
e82181de | 743 | if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) { |
c17988a9 | 744 | pr_warn("htb: too many events!\n"); |
e82181de JP |
745 | q->warned |= HTB_WARN_TOOMANYEVENTS; |
746 | } | |
1224736d JP |
747 | |
748 | return q->now; | |
1da177e4 LT |
749 | } |
750 | ||
751 | /* Returns class->node+prio from id-tree where classe's id is >= id. NULL | |
cc7ec456 ED |
752 | * is no such one exists. |
753 | */ | |
87990467 SH |
754 | static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n, |
755 | u32 id) | |
1da177e4 LT |
756 | { |
757 | struct rb_node *r = NULL; | |
758 | while (n) { | |
87990467 SH |
759 | struct htb_class *cl = |
760 | rb_entry(n, struct htb_class, node[prio]); | |
87990467 | 761 | |
f4c1f3e0 | 762 | if (id > cl->common.classid) { |
1da177e4 | 763 | n = n->rb_right; |
1b5c0077 | 764 | } else if (id < cl->common.classid) { |
1da177e4 LT |
765 | r = n; |
766 | n = n->rb_left; | |
1b5c0077 JP |
767 | } else { |
768 | return n; | |
1da177e4 LT |
769 | } |
770 | } | |
771 | return r; | |
772 | } | |
773 | ||
774 | /** | |
775 | * htb_lookup_leaf - returns next leaf class in DRR order | |
776 | * | |
777 | * Find leaf where current feed pointers points to. | |
778 | */ | |
c9364636 | 779 | static struct htb_class *htb_lookup_leaf(struct htb_prio *hprio, const int prio) |
1da177e4 LT |
780 | { |
781 | int i; | |
782 | struct { | |
783 | struct rb_node *root; | |
784 | struct rb_node **pptr; | |
785 | u32 *pid; | |
87990467 SH |
786 | } stk[TC_HTB_MAXDEPTH], *sp = stk; |
787 | ||
c9364636 ED |
788 | BUG_ON(!hprio->row.rb_node); |
789 | sp->root = hprio->row.rb_node; | |
790 | sp->pptr = &hprio->ptr; | |
791 | sp->pid = &hprio->last_ptr_id; | |
1da177e4 LT |
792 | |
793 | for (i = 0; i < 65535; i++) { | |
87990467 | 794 | if (!*sp->pptr && *sp->pid) { |
10297b99 | 795 | /* ptr was invalidated but id is valid - try to recover |
cc7ec456 ED |
796 | * the original or next ptr |
797 | */ | |
87990467 SH |
798 | *sp->pptr = |
799 | htb_id_find_next_upper(prio, sp->root, *sp->pid); | |
1da177e4 | 800 | } |
87990467 | 801 | *sp->pid = 0; /* ptr is valid now so that remove this hint as it |
cc7ec456 ED |
802 | * can become out of date quickly |
803 | */ | |
87990467 | 804 | if (!*sp->pptr) { /* we are at right end; rewind & go up */ |
1da177e4 | 805 | *sp->pptr = sp->root; |
87990467 | 806 | while ((*sp->pptr)->rb_left) |
1da177e4 LT |
807 | *sp->pptr = (*sp->pptr)->rb_left; |
808 | if (sp > stk) { | |
809 | sp--; | |
512bb43e JP |
810 | if (!*sp->pptr) { |
811 | WARN_ON(1); | |
87990467 | 812 | return NULL; |
512bb43e | 813 | } |
87990467 | 814 | htb_next_rb_node(sp->pptr); |
1da177e4 LT |
815 | } |
816 | } else { | |
817 | struct htb_class *cl; | |
c9364636 ED |
818 | struct htb_prio *clp; |
819 | ||
87990467 SH |
820 | cl = rb_entry(*sp->pptr, struct htb_class, node[prio]); |
821 | if (!cl->level) | |
1da177e4 | 822 | return cl; |
c9364636 ED |
823 | clp = &cl->un.inner.clprio[prio]; |
824 | (++sp)->root = clp->feed.rb_node; | |
825 | sp->pptr = &clp->ptr; | |
826 | sp->pid = &clp->last_ptr_id; | |
1da177e4 LT |
827 | } |
828 | } | |
547b792c | 829 | WARN_ON(1); |
1da177e4 LT |
830 | return NULL; |
831 | } | |
832 | ||
833 | /* dequeues packet at given priority and level; call only if | |
cc7ec456 ED |
834 | * you are sure that there is active class at prio/level |
835 | */ | |
c9364636 ED |
836 | static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, const int prio, |
837 | const int level) | |
1da177e4 LT |
838 | { |
839 | struct sk_buff *skb = NULL; | |
87990467 | 840 | struct htb_class *cl, *start; |
c9364636 ED |
841 | struct htb_level *hlevel = &q->hlevel[level]; |
842 | struct htb_prio *hprio = &hlevel->hprio[prio]; | |
843 | ||
1da177e4 | 844 | /* look initial class up in the row */ |
c9364636 | 845 | start = cl = htb_lookup_leaf(hprio, prio); |
87990467 | 846 | |
1da177e4 LT |
847 | do { |
848 | next: | |
512bb43e | 849 | if (unlikely(!cl)) |
87990467 | 850 | return NULL; |
1da177e4 LT |
851 | |
852 | /* class can be empty - it is unlikely but can be true if leaf | |
cc7ec456 ED |
853 | * qdisc drops packets in enqueue routine or if someone used |
854 | * graft operation on the leaf since last dequeue; | |
855 | * simply deactivate and skip such class | |
856 | */ | |
1da177e4 LT |
857 | if (unlikely(cl->un.leaf.q->q.qlen == 0)) { |
858 | struct htb_class *next; | |
87990467 | 859 | htb_deactivate(q, cl); |
1da177e4 LT |
860 | |
861 | /* row/level might become empty */ | |
862 | if ((q->row_mask[level] & (1 << prio)) == 0) | |
87990467 | 863 | return NULL; |
1da177e4 | 864 | |
c9364636 | 865 | next = htb_lookup_leaf(hprio, prio); |
87990467 SH |
866 | |
867 | if (cl == start) /* fix start if we just deleted it */ | |
1da177e4 LT |
868 | start = next; |
869 | cl = next; | |
870 | goto next; | |
871 | } | |
87990467 SH |
872 | |
873 | skb = cl->un.leaf.q->dequeue(cl->un.leaf.q); | |
874 | if (likely(skb != NULL)) | |
1da177e4 | 875 | break; |
633fe66e | 876 | |
b00355db | 877 | qdisc_warn_nonwc("htb", cl->un.leaf.q); |
c9364636 ED |
878 | htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr: |
879 | &q->hlevel[0].hprio[prio].ptr); | |
880 | cl = htb_lookup_leaf(hprio, prio); | |
1da177e4 LT |
881 | |
882 | } while (cl != start); | |
883 | ||
884 | if (likely(skb != NULL)) { | |
196d97f6 | 885 | bstats_update(&cl->bstats, skb); |
0abf77e5 JK |
886 | cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb); |
887 | if (cl->un.leaf.deficit[level] < 0) { | |
c19f7a34 | 888 | cl->un.leaf.deficit[level] += cl->quantum; |
c9364636 ED |
889 | htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr : |
890 | &q->hlevel[0].hprio[prio].ptr); | |
1da177e4 LT |
891 | } |
892 | /* this used to be after charge_class but this constelation | |
cc7ec456 ED |
893 | * gives us slightly better performance |
894 | */ | |
1da177e4 | 895 | if (!cl->un.leaf.q->q.qlen) |
87990467 | 896 | htb_deactivate(q, cl); |
c9726d68 | 897 | htb_charge_class(q, cl, level, skb); |
1da177e4 LT |
898 | } |
899 | return skb; | |
900 | } | |
901 | ||
1da177e4 LT |
902 | static struct sk_buff *htb_dequeue(struct Qdisc *sch) |
903 | { | |
9190b3b3 | 904 | struct sk_buff *skb; |
1da177e4 LT |
905 | struct htb_sched *q = qdisc_priv(sch); |
906 | int level; | |
5343a7f8 | 907 | s64 next_event; |
a73be040 | 908 | unsigned long start_at; |
1da177e4 LT |
909 | |
910 | /* try to dequeue direct packets as high prio (!) to minimize cpu work */ | |
48da34b7 | 911 | skb = __qdisc_dequeue_head(&q->direct_queue); |
87990467 | 912 | if (skb != NULL) { |
9190b3b3 ED |
913 | ok: |
914 | qdisc_bstats_update(sch, skb); | |
431e3a8e | 915 | qdisc_qstats_backlog_dec(sch, skb); |
1da177e4 LT |
916 | sch->q.qlen--; |
917 | return skb; | |
918 | } | |
919 | ||
87990467 SH |
920 | if (!sch->q.qlen) |
921 | goto fin; | |
d2de875c | 922 | q->now = ktime_get_ns(); |
a73be040 | 923 | start_at = jiffies; |
1da177e4 | 924 | |
d2fe85da | 925 | next_event = q->now + 5LLU * NSEC_PER_SEC; |
633fe66e | 926 | |
1da177e4 LT |
927 | for (level = 0; level < TC_HTB_MAXDEPTH; level++) { |
928 | /* common case optimization - skip event handler quickly */ | |
929 | int m; | |
c9364636 | 930 | s64 event = q->near_ev_cache[level]; |
fb983d45 | 931 | |
c9364636 | 932 | if (q->now >= event) { |
a73be040 | 933 | event = htb_do_events(q, level, start_at); |
2e4b3b0e | 934 | if (!event) |
56b765b7 | 935 | event = q->now + NSEC_PER_SEC; |
2e4b3b0e | 936 | q->near_ev_cache[level] = event; |
c9364636 | 937 | } |
fb983d45 | 938 | |
c0851347 | 939 | if (next_event > event) |
fb983d45 | 940 | next_event = event; |
87990467 | 941 | |
1da177e4 LT |
942 | m = ~q->row_mask[level]; |
943 | while (m != (int)(-1)) { | |
87990467 | 944 | int prio = ffz(m); |
cc7ec456 | 945 | |
1da177e4 | 946 | m |= 1 << prio; |
87990467 | 947 | skb = htb_dequeue_tree(q, prio, level); |
9190b3b3 ED |
948 | if (likely(skb != NULL)) |
949 | goto ok; | |
1da177e4 LT |
950 | } |
951 | } | |
25331d6c | 952 | qdisc_qstats_overlimit(sch); |
a9efad8b | 953 | if (likely(next_event > q->now)) |
45f50bed | 954 | qdisc_watchdog_schedule_ns(&q->watchdog, next_event); |
a9efad8b | 955 | else |
1224736d | 956 | schedule_work(&q->work); |
1da177e4 | 957 | fin: |
1da177e4 LT |
958 | return skb; |
959 | } | |
960 | ||
1da177e4 LT |
961 | /* reset all classes */ |
962 | /* always caled under BH & queue lock */ | |
87990467 | 963 | static void htb_reset(struct Qdisc *sch) |
1da177e4 LT |
964 | { |
965 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 966 | struct htb_class *cl; |
f4c1f3e0 | 967 | unsigned int i; |
0cef296d | 968 | |
f4c1f3e0 | 969 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 970 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { |
1da177e4 | 971 | if (cl->level) |
87990467 | 972 | memset(&cl->un.inner, 0, sizeof(cl->un.inner)); |
1da177e4 | 973 | else { |
87990467 | 974 | if (cl->un.leaf.q) |
1da177e4 LT |
975 | qdisc_reset(cl->un.leaf.q); |
976 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); | |
977 | } | |
978 | cl->prio_activity = 0; | |
979 | cl->cmode = HTB_CAN_SEND; | |
1da177e4 LT |
980 | } |
981 | } | |
fb983d45 | 982 | qdisc_watchdog_cancel(&q->watchdog); |
a5a9f534 | 983 | __qdisc_reset_queue(&q->direct_queue); |
1da177e4 | 984 | sch->q.qlen = 0; |
431e3a8e | 985 | sch->qstats.backlog = 0; |
c9364636 | 986 | memset(q->hlevel, 0, sizeof(q->hlevel)); |
87990467 | 987 | memset(q->row_mask, 0, sizeof(q->row_mask)); |
1da177e4 | 988 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 989 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 LT |
990 | } |
991 | ||
27a3421e PM |
992 | static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = { |
993 | [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) }, | |
994 | [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) }, | |
995 | [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
996 | [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
6906f4ed | 997 | [TCA_HTB_DIRECT_QLEN] = { .type = NLA_U32 }, |
df62cdf3 ED |
998 | [TCA_HTB_RATE64] = { .type = NLA_U64 }, |
999 | [TCA_HTB_CEIL64] = { .type = NLA_U64 }, | |
27a3421e PM |
1000 | }; |
1001 | ||
1224736d JP |
1002 | static void htb_work_func(struct work_struct *work) |
1003 | { | |
1004 | struct htb_sched *q = container_of(work, struct htb_sched, work); | |
1005 | struct Qdisc *sch = q->watchdog.qdisc; | |
1006 | ||
0ee13627 | 1007 | rcu_read_lock(); |
1224736d | 1008 | __netif_schedule(qdisc_root(sch)); |
0ee13627 | 1009 | rcu_read_unlock(); |
1224736d JP |
1010 | } |
1011 | ||
1e90474c | 1012 | static int htb_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
1013 | { |
1014 | struct htb_sched *q = qdisc_priv(sch); | |
6906f4ed | 1015 | struct nlattr *tb[TCA_HTB_MAX + 1]; |
1da177e4 | 1016 | struct tc_htb_glob *gopt; |
cee63723 | 1017 | int err; |
1da177e4 | 1018 | int i; |
cee63723 PM |
1019 | |
1020 | if (!opt) | |
1021 | return -EINVAL; | |
1022 | ||
6529eaba JP |
1023 | err = tcf_block_get(&q->block, &q->filter_list); |
1024 | if (err) | |
1025 | return err; | |
1026 | ||
fceb6435 | 1027 | err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy, NULL); |
cee63723 PM |
1028 | if (err < 0) |
1029 | return err; | |
1030 | ||
6906f4ed | 1031 | if (!tb[TCA_HTB_INIT]) |
1da177e4 | 1032 | return -EINVAL; |
6906f4ed | 1033 | |
1e90474c | 1034 | gopt = nla_data(tb[TCA_HTB_INIT]); |
6906f4ed | 1035 | if (gopt->version != HTB_VER >> 16) |
1da177e4 | 1036 | return -EINVAL; |
1da177e4 | 1037 | |
f4c1f3e0 PM |
1038 | err = qdisc_class_hash_init(&q->clhash); |
1039 | if (err < 0) | |
1040 | return err; | |
1da177e4 | 1041 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 1042 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 | 1043 | |
fb983d45 | 1044 | qdisc_watchdog_init(&q->watchdog, sch); |
1224736d | 1045 | INIT_WORK(&q->work, htb_work_func); |
48da34b7 | 1046 | qdisc_skb_head_init(&q->direct_queue); |
1da177e4 | 1047 | |
6906f4ed ED |
1048 | if (tb[TCA_HTB_DIRECT_QLEN]) |
1049 | q->direct_qlen = nla_get_u32(tb[TCA_HTB_DIRECT_QLEN]); | |
348e3435 | 1050 | else |
6906f4ed | 1051 | q->direct_qlen = qdisc_dev(sch)->tx_queue_len; |
348e3435 | 1052 | |
1da177e4 LT |
1053 | if ((q->rate2quantum = gopt->rate2quantum) < 1) |
1054 | q->rate2quantum = 1; | |
1055 | q->defcls = gopt->defcls; | |
1056 | ||
1057 | return 0; | |
1058 | } | |
1059 | ||
1060 | static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) | |
1061 | { | |
1062 | struct htb_sched *q = qdisc_priv(sch); | |
4b3550ef | 1063 | struct nlattr *nest; |
1da177e4 | 1064 | struct tc_htb_glob gopt; |
4b3550ef | 1065 | |
6f542efc ED |
1066 | /* Its safe to not acquire qdisc lock. As we hold RTNL, |
1067 | * no change can happen on the qdisc parameters. | |
1068 | */ | |
1da177e4 | 1069 | |
4b3550ef | 1070 | gopt.direct_pkts = q->direct_pkts; |
1da177e4 LT |
1071 | gopt.version = HTB_VER; |
1072 | gopt.rate2quantum = q->rate2quantum; | |
1073 | gopt.defcls = q->defcls; | |
3bf72957 | 1074 | gopt.debug = 0; |
4b3550ef PM |
1075 | |
1076 | nest = nla_nest_start(skb, TCA_OPTIONS); | |
1077 | if (nest == NULL) | |
1078 | goto nla_put_failure; | |
6906f4ed ED |
1079 | if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt) || |
1080 | nla_put_u32(skb, TCA_HTB_DIRECT_QLEN, q->direct_qlen)) | |
1b34ec43 | 1081 | goto nla_put_failure; |
4b3550ef | 1082 | |
6f542efc | 1083 | return nla_nest_end(skb, nest); |
4b3550ef | 1084 | |
1e90474c | 1085 | nla_put_failure: |
4b3550ef | 1086 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1087 | return -1; |
1088 | } | |
1089 | ||
1090 | static int htb_dump_class(struct Qdisc *sch, unsigned long arg, | |
87990467 | 1091 | struct sk_buff *skb, struct tcmsg *tcm) |
1da177e4 | 1092 | { |
87990467 | 1093 | struct htb_class *cl = (struct htb_class *)arg; |
4b3550ef | 1094 | struct nlattr *nest; |
1da177e4 LT |
1095 | struct tc_htb_opt opt; |
1096 | ||
6f542efc ED |
1097 | /* Its safe to not acquire qdisc lock. As we hold RTNL, |
1098 | * no change can happen on the class parameters. | |
1099 | */ | |
f4c1f3e0 PM |
1100 | tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT; |
1101 | tcm->tcm_handle = cl->common.classid; | |
1da177e4 LT |
1102 | if (!cl->level && cl->un.leaf.q) |
1103 | tcm->tcm_info = cl->un.leaf.q->handle; | |
1104 | ||
4b3550ef PM |
1105 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1106 | if (nest == NULL) | |
1107 | goto nla_put_failure; | |
1da177e4 | 1108 | |
87990467 | 1109 | memset(&opt, 0, sizeof(opt)); |
1da177e4 | 1110 | |
01cb71d2 | 1111 | psched_ratecfg_getrate(&opt.rate, &cl->rate); |
9c10f411 | 1112 | opt.buffer = PSCHED_NS2TICKS(cl->buffer); |
01cb71d2 | 1113 | psched_ratecfg_getrate(&opt.ceil, &cl->ceil); |
9c10f411 | 1114 | opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer); |
c19f7a34 JP |
1115 | opt.quantum = cl->quantum; |
1116 | opt.prio = cl->prio; | |
87990467 | 1117 | opt.level = cl->level; |
1b34ec43 DM |
1118 | if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt)) |
1119 | goto nla_put_failure; | |
df62cdf3 | 1120 | if ((cl->rate.rate_bytes_ps >= (1ULL << 32)) && |
2a51c1e8 ND |
1121 | nla_put_u64_64bit(skb, TCA_HTB_RATE64, cl->rate.rate_bytes_ps, |
1122 | TCA_HTB_PAD)) | |
df62cdf3 ED |
1123 | goto nla_put_failure; |
1124 | if ((cl->ceil.rate_bytes_ps >= (1ULL << 32)) && | |
2a51c1e8 ND |
1125 | nla_put_u64_64bit(skb, TCA_HTB_CEIL64, cl->ceil.rate_bytes_ps, |
1126 | TCA_HTB_PAD)) | |
df62cdf3 | 1127 | goto nla_put_failure; |
4b3550ef | 1128 | |
6f542efc | 1129 | return nla_nest_end(skb, nest); |
4b3550ef | 1130 | |
1e90474c | 1131 | nla_put_failure: |
4b3550ef | 1132 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1133 | return -1; |
1134 | } | |
1135 | ||
1136 | static int | |
87990467 | 1137 | htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d) |
1da177e4 | 1138 | { |
87990467 | 1139 | struct htb_class *cl = (struct htb_class *)arg; |
338ed9b4 ED |
1140 | struct gnet_stats_queue qs = { |
1141 | .drops = cl->drops, | |
1142 | }; | |
64015853 | 1143 | __u32 qlen = 0; |
1da177e4 | 1144 | |
338ed9b4 | 1145 | if (!cl->level && cl->un.leaf.q) { |
64015853 | 1146 | qlen = cl->un.leaf.q->q.qlen; |
338ed9b4 ED |
1147 | qs.backlog = cl->un.leaf.q->qstats.backlog; |
1148 | } | |
0564bf0a KK |
1149 | cl->xstats.tokens = clamp_t(s64, PSCHED_NS2TICKS(cl->tokens), |
1150 | INT_MIN, INT_MAX); | |
1151 | cl->xstats.ctokens = clamp_t(s64, PSCHED_NS2TICKS(cl->ctokens), | |
1152 | INT_MIN, INT_MAX); | |
1da177e4 | 1153 | |
edb09eb1 ED |
1154 | if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), |
1155 | d, NULL, &cl->bstats) < 0 || | |
1c0d32fd | 1156 | gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || |
338ed9b4 | 1157 | gnet_stats_copy_queue(d, NULL, &qs, qlen) < 0) |
1da177e4 LT |
1158 | return -1; |
1159 | ||
1160 | return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); | |
1161 | } | |
1162 | ||
1163 | static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
87990467 | 1164 | struct Qdisc **old) |
1da177e4 | 1165 | { |
87990467 | 1166 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 | 1167 | |
5b9a9ccf PM |
1168 | if (cl->level) |
1169 | return -EINVAL; | |
1170 | if (new == NULL && | |
3511c913 | 1171 | (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
5b9a9ccf PM |
1172 | cl->common.classid)) == NULL) |
1173 | return -ENOBUFS; | |
1174 | ||
86a7996c | 1175 | *old = qdisc_replace(sch, new, &cl->un.leaf.q); |
5b9a9ccf | 1176 | return 0; |
1da177e4 LT |
1177 | } |
1178 | ||
87990467 | 1179 | static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg) |
1da177e4 | 1180 | { |
87990467 | 1181 | struct htb_class *cl = (struct htb_class *)arg; |
5b9a9ccf | 1182 | return !cl->level ? cl->un.leaf.q : NULL; |
1da177e4 LT |
1183 | } |
1184 | ||
256d61b8 PM |
1185 | static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg) |
1186 | { | |
1187 | struct htb_class *cl = (struct htb_class *)arg; | |
1188 | ||
1189 | if (cl->un.leaf.q->q.qlen == 0) | |
1190 | htb_deactivate(qdisc_priv(sch), cl); | |
1191 | } | |
1192 | ||
1da177e4 LT |
1193 | static unsigned long htb_get(struct Qdisc *sch, u32 classid) |
1194 | { | |
87990467 SH |
1195 | struct htb_class *cl = htb_find(classid, sch); |
1196 | if (cl) | |
1da177e4 LT |
1197 | cl->refcnt++; |
1198 | return (unsigned long)cl; | |
1199 | } | |
1200 | ||
160d5e10 JP |
1201 | static inline int htb_parent_last_child(struct htb_class *cl) |
1202 | { | |
1203 | if (!cl->parent) | |
1204 | /* the root class */ | |
1205 | return 0; | |
42077599 | 1206 | if (cl->parent->children > 1) |
160d5e10 JP |
1207 | /* not the last child */ |
1208 | return 0; | |
160d5e10 JP |
1209 | return 1; |
1210 | } | |
1211 | ||
3ba08b00 JP |
1212 | static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl, |
1213 | struct Qdisc *new_q) | |
160d5e10 JP |
1214 | { |
1215 | struct htb_class *parent = cl->parent; | |
1216 | ||
547b792c | 1217 | WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity); |
160d5e10 | 1218 | |
3ba08b00 | 1219 | if (parent->cmode != HTB_CAN_SEND) |
c9364636 ED |
1220 | htb_safe_rb_erase(&parent->pq_node, |
1221 | &q->hlevel[parent->level].wait_pq); | |
3ba08b00 | 1222 | |
160d5e10 JP |
1223 | parent->level = 0; |
1224 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1225 | INIT_LIST_HEAD(&parent->un.leaf.drop_list); | |
1226 | parent->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
160d5e10 JP |
1227 | parent->tokens = parent->buffer; |
1228 | parent->ctokens = parent->cbuffer; | |
d2de875c | 1229 | parent->t_c = ktime_get_ns(); |
160d5e10 JP |
1230 | parent->cmode = HTB_CAN_SEND; |
1231 | } | |
1232 | ||
87990467 | 1233 | static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl) |
1da177e4 | 1234 | { |
1da177e4 | 1235 | if (!cl->level) { |
547b792c | 1236 | WARN_ON(!cl->un.leaf.q); |
1da177e4 LT |
1237 | qdisc_destroy(cl->un.leaf.q); |
1238 | } | |
1c0d32fd | 1239 | gen_kill_estimator(&cl->rate_est); |
6529eaba | 1240 | tcf_block_put(cl->block); |
1da177e4 LT |
1241 | kfree(cl); |
1242 | } | |
1243 | ||
87990467 | 1244 | static void htb_destroy(struct Qdisc *sch) |
1da177e4 LT |
1245 | { |
1246 | struct htb_sched *q = qdisc_priv(sch); | |
b67bfe0d | 1247 | struct hlist_node *next; |
fbd8f137 PM |
1248 | struct htb_class *cl; |
1249 | unsigned int i; | |
1da177e4 | 1250 | |
1224736d | 1251 | cancel_work_sync(&q->work); |
fb983d45 | 1252 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 | 1253 | /* This line used to be after htb_destroy_class call below |
cc7ec456 ED |
1254 | * and surprisingly it worked in 2.4. But it must precede it |
1255 | * because filter need its target class alive to be able to call | |
1256 | * unbind_filter on it (without Oops). | |
1257 | */ | |
6529eaba | 1258 | tcf_block_put(q->block); |
87990467 | 1259 | |
f4c1f3e0 | 1260 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 1261 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) |
6529eaba | 1262 | tcf_block_put(cl->block); |
fbd8f137 | 1263 | } |
f4c1f3e0 | 1264 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 1265 | hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i], |
f4c1f3e0 | 1266 | common.hnode) |
fbd8f137 PM |
1267 | htb_destroy_class(sch, cl); |
1268 | } | |
f4c1f3e0 | 1269 | qdisc_class_hash_destroy(&q->clhash); |
a5a9f534 | 1270 | __qdisc_reset_queue(&q->direct_queue); |
1da177e4 LT |
1271 | } |
1272 | ||
1273 | static int htb_delete(struct Qdisc *sch, unsigned long arg) | |
1274 | { | |
1275 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1276 | struct htb_class *cl = (struct htb_class *)arg; |
160d5e10 JP |
1277 | struct Qdisc *new_q = NULL; |
1278 | int last_child = 0; | |
1da177e4 | 1279 | |
a071d272 YY |
1280 | /* TODO: why don't allow to delete subtree ? references ? does |
1281 | * tc subsys guarantee us that in htb_destroy it holds no class | |
1282 | * refs so that we can remove children safely there ? | |
1283 | */ | |
42077599 | 1284 | if (cl->children || cl->filter_cnt) |
1da177e4 | 1285 | return -EBUSY; |
87990467 | 1286 | |
160d5e10 | 1287 | if (!cl->level && htb_parent_last_child(cl)) { |
3511c913 | 1288 | new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
bb949fbd | 1289 | cl->parent->common.classid); |
160d5e10 JP |
1290 | last_child = 1; |
1291 | } | |
1292 | ||
1da177e4 | 1293 | sch_tree_lock(sch); |
87990467 | 1294 | |
814a175e | 1295 | if (!cl->level) { |
2ccccf5f WC |
1296 | unsigned int qlen = cl->un.leaf.q->q.qlen; |
1297 | unsigned int backlog = cl->un.leaf.q->qstats.backlog; | |
1298 | ||
814a175e | 1299 | qdisc_reset(cl->un.leaf.q); |
2ccccf5f | 1300 | qdisc_tree_reduce_backlog(cl->un.leaf.q, qlen, backlog); |
814a175e PM |
1301 | } |
1302 | ||
f4c1f3e0 PM |
1303 | /* delete from hash and active; remainder in destroy_class */ |
1304 | qdisc_class_hash_remove(&q->clhash, &cl->common); | |
26b284de JP |
1305 | if (cl->parent) |
1306 | cl->parent->children--; | |
c38c83cb | 1307 | |
1da177e4 | 1308 | if (cl->prio_activity) |
87990467 | 1309 | htb_deactivate(q, cl); |
1da177e4 | 1310 | |
fbd8f137 | 1311 | if (cl->cmode != HTB_CAN_SEND) |
c9364636 ED |
1312 | htb_safe_rb_erase(&cl->pq_node, |
1313 | &q->hlevel[cl->level].wait_pq); | |
fbd8f137 | 1314 | |
160d5e10 | 1315 | if (last_child) |
3ba08b00 | 1316 | htb_parent_to_leaf(q, cl, new_q); |
160d5e10 | 1317 | |
7cd0a638 JP |
1318 | BUG_ON(--cl->refcnt == 0); |
1319 | /* | |
1320 | * This shouldn't happen: we "hold" one cops->get() when called | |
1321 | * from tc_ctl_tclass; the destroy method is done from cops->put(). | |
1322 | */ | |
1da177e4 LT |
1323 | |
1324 | sch_tree_unlock(sch); | |
1325 | return 0; | |
1326 | } | |
1327 | ||
1328 | static void htb_put(struct Qdisc *sch, unsigned long arg) | |
1329 | { | |
87990467 | 1330 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1331 | |
1332 | if (--cl->refcnt == 0) | |
87990467 | 1333 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1334 | } |
1335 | ||
87990467 | 1336 | static int htb_change_class(struct Qdisc *sch, u32 classid, |
1e90474c | 1337 | u32 parentid, struct nlattr **tca, |
87990467 | 1338 | unsigned long *arg) |
1da177e4 LT |
1339 | { |
1340 | int err = -EINVAL; | |
1341 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1342 | struct htb_class *cl = (struct htb_class *)*arg, *parent; |
1e90474c | 1343 | struct nlattr *opt = tca[TCA_OPTIONS]; |
6906f4ed | 1344 | struct nlattr *tb[TCA_HTB_MAX + 1]; |
1da177e4 | 1345 | struct tc_htb_opt *hopt; |
df62cdf3 | 1346 | u64 rate64, ceil64; |
1da177e4 LT |
1347 | |
1348 | /* extract all subattrs from opt attr */ | |
cee63723 PM |
1349 | if (!opt) |
1350 | goto failure; | |
1351 | ||
fceb6435 | 1352 | err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy, NULL); |
cee63723 PM |
1353 | if (err < 0) |
1354 | goto failure; | |
1355 | ||
1356 | err = -EINVAL; | |
27a3421e | 1357 | if (tb[TCA_HTB_PARMS] == NULL) |
1da177e4 | 1358 | goto failure; |
1da177e4 | 1359 | |
87990467 SH |
1360 | parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch); |
1361 | ||
1e90474c | 1362 | hopt = nla_data(tb[TCA_HTB_PARMS]); |
196d97f6 | 1363 | if (!hopt->rate.rate || !hopt->ceil.rate) |
87990467 | 1364 | goto failure; |
1da177e4 | 1365 | |
8a8e3d84 | 1366 | /* Keeping backward compatible with rate_table based iproute2 tc */ |
6b1dd856 YY |
1367 | if (hopt->rate.linklayer == TC_LINKLAYER_UNAWARE) |
1368 | qdisc_put_rtab(qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB])); | |
1369 | ||
1370 | if (hopt->ceil.linklayer == TC_LINKLAYER_UNAWARE) | |
1371 | qdisc_put_rtab(qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB])); | |
8a8e3d84 | 1372 | |
87990467 | 1373 | if (!cl) { /* new class */ |
1da177e4 | 1374 | struct Qdisc *new_q; |
3696f625 | 1375 | int prio; |
ee39e10c | 1376 | struct { |
1e90474c | 1377 | struct nlattr nla; |
ee39e10c PM |
1378 | struct gnet_estimator opt; |
1379 | } est = { | |
1e90474c PM |
1380 | .nla = { |
1381 | .nla_len = nla_attr_size(sizeof(est.opt)), | |
1382 | .nla_type = TCA_RATE, | |
ee39e10c PM |
1383 | }, |
1384 | .opt = { | |
1385 | /* 4s interval, 16s averaging constant */ | |
1386 | .interval = 2, | |
1387 | .ewma_log = 2, | |
1388 | }, | |
1389 | }; | |
3696f625 | 1390 | |
1da177e4 | 1391 | /* check for valid classid */ |
f64f9e71 JP |
1392 | if (!classid || TC_H_MAJ(classid ^ sch->handle) || |
1393 | htb_find(classid, sch)) | |
1da177e4 LT |
1394 | goto failure; |
1395 | ||
1396 | /* check maximal depth */ | |
1397 | if (parent && parent->parent && parent->parent->level < 2) { | |
cc7ec456 | 1398 | pr_err("htb: tree is too deep\n"); |
1da177e4 LT |
1399 | goto failure; |
1400 | } | |
1401 | err = -ENOBUFS; | |
cc7ec456 ED |
1402 | cl = kzalloc(sizeof(*cl), GFP_KERNEL); |
1403 | if (!cl) | |
1da177e4 | 1404 | goto failure; |
87990467 | 1405 | |
6529eaba JP |
1406 | err = tcf_block_get(&cl->block, &cl->filter_list); |
1407 | if (err) { | |
1408 | kfree(cl); | |
1409 | goto failure; | |
1410 | } | |
64153ce0 | 1411 | if (htb_rate_est || tca[TCA_RATE]) { |
22e0f8b9 JF |
1412 | err = gen_new_estimator(&cl->bstats, NULL, |
1413 | &cl->rate_est, | |
edb09eb1 ED |
1414 | NULL, |
1415 | qdisc_root_sleeping_running(sch), | |
64153ce0 ED |
1416 | tca[TCA_RATE] ? : &est.nla); |
1417 | if (err) { | |
6529eaba | 1418 | tcf_block_put(cl->block); |
64153ce0 ED |
1419 | kfree(cl); |
1420 | goto failure; | |
1421 | } | |
71bcb09a SH |
1422 | } |
1423 | ||
1da177e4 | 1424 | cl->refcnt = 1; |
42077599 | 1425 | cl->children = 0; |
1da177e4 | 1426 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); |
3696f625 SH |
1427 | RB_CLEAR_NODE(&cl->pq_node); |
1428 | ||
1429 | for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) | |
1430 | RB_CLEAR_NODE(&cl->node[prio]); | |
1da177e4 LT |
1431 | |
1432 | /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL) | |
cc7ec456 ED |
1433 | * so that can't be used inside of sch_tree_lock |
1434 | * -- thanks to Karlis Peisenieks | |
1435 | */ | |
3511c913 | 1436 | new_q = qdisc_create_dflt(sch->dev_queue, |
bb949fbd | 1437 | &pfifo_qdisc_ops, classid); |
1da177e4 LT |
1438 | sch_tree_lock(sch); |
1439 | if (parent && !parent->level) { | |
256d61b8 | 1440 | unsigned int qlen = parent->un.leaf.q->q.qlen; |
2ccccf5f | 1441 | unsigned int backlog = parent->un.leaf.q->qstats.backlog; |
256d61b8 | 1442 | |
1da177e4 | 1443 | /* turn parent into inner node */ |
256d61b8 | 1444 | qdisc_reset(parent->un.leaf.q); |
2ccccf5f | 1445 | qdisc_tree_reduce_backlog(parent->un.leaf.q, qlen, backlog); |
87990467 SH |
1446 | qdisc_destroy(parent->un.leaf.q); |
1447 | if (parent->prio_activity) | |
1448 | htb_deactivate(q, parent); | |
1da177e4 LT |
1449 | |
1450 | /* remove from evt list because of level change */ | |
1451 | if (parent->cmode != HTB_CAN_SEND) { | |
c9364636 | 1452 | htb_safe_rb_erase(&parent->pq_node, &q->hlevel[0].wait_pq); |
1da177e4 LT |
1453 | parent->cmode = HTB_CAN_SEND; |
1454 | } | |
1455 | parent->level = (parent->parent ? parent->parent->level | |
87990467 SH |
1456 | : TC_HTB_MAXDEPTH) - 1; |
1457 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1da177e4 LT |
1458 | } |
1459 | /* leaf (we) needs elementary qdisc */ | |
1460 | cl->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
1461 | ||
f4c1f3e0 | 1462 | cl->common.classid = classid; |
87990467 | 1463 | cl->parent = parent; |
1da177e4 LT |
1464 | |
1465 | /* set class to be in HTB_CAN_SEND state */ | |
b9a7afde JP |
1466 | cl->tokens = PSCHED_TICKS2NS(hopt->buffer); |
1467 | cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer); | |
5343a7f8 | 1468 | cl->mbuffer = 60ULL * NSEC_PER_SEC; /* 1min */ |
d2de875c | 1469 | cl->t_c = ktime_get_ns(); |
1da177e4 LT |
1470 | cl->cmode = HTB_CAN_SEND; |
1471 | ||
1472 | /* attach to the hash list and parent's family */ | |
f4c1f3e0 | 1473 | qdisc_class_hash_insert(&q->clhash, &cl->common); |
42077599 PM |
1474 | if (parent) |
1475 | parent->children++; | |
49b49971 JK |
1476 | if (cl->un.leaf.q != &noop_qdisc) |
1477 | qdisc_hash_add(cl->un.leaf.q, true); | |
ee39e10c | 1478 | } else { |
71bcb09a | 1479 | if (tca[TCA_RATE]) { |
22e0f8b9 JF |
1480 | err = gen_replace_estimator(&cl->bstats, NULL, |
1481 | &cl->rate_est, | |
edb09eb1 ED |
1482 | NULL, |
1483 | qdisc_root_sleeping_running(sch), | |
71bcb09a SH |
1484 | tca[TCA_RATE]); |
1485 | if (err) | |
1486 | return err; | |
1487 | } | |
87990467 | 1488 | sch_tree_lock(sch); |
ee39e10c | 1489 | } |
1da177e4 | 1490 | |
1598f7cb YY |
1491 | rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0; |
1492 | ||
1493 | ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0; | |
1494 | ||
1495 | psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64); | |
1496 | psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64); | |
1497 | ||
1da177e4 | 1498 | /* it used to be a nasty bug here, we have to check that node |
cc7ec456 ED |
1499 | * is really leaf before changing cl->un.leaf ! |
1500 | */ | |
1da177e4 | 1501 | if (!cl->level) { |
1598f7cb YY |
1502 | u64 quantum = cl->rate.rate_bytes_ps; |
1503 | ||
1504 | do_div(quantum, q->rate2quantum); | |
1505 | cl->quantum = min_t(u64, quantum, INT_MAX); | |
1506 | ||
c19f7a34 | 1507 | if (!hopt->quantum && cl->quantum < 1000) { |
c17988a9 YY |
1508 | pr_warn("HTB: quantum of class %X is small. Consider r2q change.\n", |
1509 | cl->common.classid); | |
c19f7a34 | 1510 | cl->quantum = 1000; |
1da177e4 | 1511 | } |
c19f7a34 | 1512 | if (!hopt->quantum && cl->quantum > 200000) { |
c17988a9 YY |
1513 | pr_warn("HTB: quantum of class %X is big. Consider r2q change.\n", |
1514 | cl->common.classid); | |
c19f7a34 | 1515 | cl->quantum = 200000; |
1da177e4 LT |
1516 | } |
1517 | if (hopt->quantum) | |
c19f7a34 JP |
1518 | cl->quantum = hopt->quantum; |
1519 | if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO) | |
1520 | cl->prio = TC_HTB_NUMPRIO - 1; | |
1da177e4 LT |
1521 | } |
1522 | ||
324f5aa5 | 1523 | cl->buffer = PSCHED_TICKS2NS(hopt->buffer); |
f3ad857e | 1524 | cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer); |
56b765b7 | 1525 | |
1da177e4 LT |
1526 | sch_tree_unlock(sch); |
1527 | ||
f4c1f3e0 PM |
1528 | qdisc_class_hash_grow(sch, &q->clhash); |
1529 | ||
1da177e4 LT |
1530 | *arg = (unsigned long)cl; |
1531 | return 0; | |
1532 | ||
1533 | failure: | |
1da177e4 LT |
1534 | return err; |
1535 | } | |
1536 | ||
6529eaba | 1537 | static struct tcf_block *htb_tcf_block(struct Qdisc *sch, unsigned long arg) |
1da177e4 LT |
1538 | { |
1539 | struct htb_sched *q = qdisc_priv(sch); | |
1540 | struct htb_class *cl = (struct htb_class *)arg; | |
3bf72957 | 1541 | |
6529eaba | 1542 | return cl ? cl->block : q->block; |
1da177e4 LT |
1543 | } |
1544 | ||
1545 | static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, | |
87990467 | 1546 | u32 classid) |
1da177e4 | 1547 | { |
87990467 | 1548 | struct htb_class *cl = htb_find(classid, sch); |
3bf72957 | 1549 | |
1da177e4 | 1550 | /*if (cl && !cl->level) return 0; |
cc7ec456 ED |
1551 | * The line above used to be there to prevent attaching filters to |
1552 | * leaves. But at least tc_index filter uses this just to get class | |
1553 | * for other reasons so that we have to allow for it. | |
1554 | * ---- | |
1555 | * 19.6.2002 As Werner explained it is ok - bind filter is just | |
1556 | * another way to "lock" the class - unlike "get" this lock can | |
1557 | * be broken by class during destroy IIUC. | |
1da177e4 | 1558 | */ |
87990467 SH |
1559 | if (cl) |
1560 | cl->filter_cnt++; | |
1da177e4 LT |
1561 | return (unsigned long)cl; |
1562 | } | |
1563 | ||
1564 | static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) | |
1565 | { | |
1da177e4 | 1566 | struct htb_class *cl = (struct htb_class *)arg; |
3bf72957 | 1567 | |
87990467 SH |
1568 | if (cl) |
1569 | cl->filter_cnt--; | |
1da177e4 LT |
1570 | } |
1571 | ||
1572 | static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
1573 | { | |
1574 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 1575 | struct htb_class *cl; |
f4c1f3e0 | 1576 | unsigned int i; |
1da177e4 LT |
1577 | |
1578 | if (arg->stop) | |
1579 | return; | |
1580 | ||
f4c1f3e0 | 1581 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 1582 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { |
1da177e4 LT |
1583 | if (arg->count < arg->skip) { |
1584 | arg->count++; | |
1585 | continue; | |
1586 | } | |
1587 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
1588 | arg->stop = 1; | |
1589 | return; | |
1590 | } | |
1591 | arg->count++; | |
1592 | } | |
1593 | } | |
1594 | } | |
1595 | ||
20fea08b | 1596 | static const struct Qdisc_class_ops htb_class_ops = { |
1da177e4 LT |
1597 | .graft = htb_graft, |
1598 | .leaf = htb_leaf, | |
256d61b8 | 1599 | .qlen_notify = htb_qlen_notify, |
1da177e4 LT |
1600 | .get = htb_get, |
1601 | .put = htb_put, | |
1602 | .change = htb_change_class, | |
1603 | .delete = htb_delete, | |
1604 | .walk = htb_walk, | |
6529eaba | 1605 | .tcf_block = htb_tcf_block, |
1da177e4 LT |
1606 | .bind_tcf = htb_bind_filter, |
1607 | .unbind_tcf = htb_unbind_filter, | |
1608 | .dump = htb_dump_class, | |
1609 | .dump_stats = htb_dump_class_stats, | |
1610 | }; | |
1611 | ||
20fea08b | 1612 | static struct Qdisc_ops htb_qdisc_ops __read_mostly = { |
1da177e4 LT |
1613 | .cl_ops = &htb_class_ops, |
1614 | .id = "htb", | |
1615 | .priv_size = sizeof(struct htb_sched), | |
1616 | .enqueue = htb_enqueue, | |
1617 | .dequeue = htb_dequeue, | |
77be155c | 1618 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1619 | .init = htb_init, |
1620 | .reset = htb_reset, | |
1621 | .destroy = htb_destroy, | |
1da177e4 LT |
1622 | .dump = htb_dump, |
1623 | .owner = THIS_MODULE, | |
1624 | }; | |
1625 | ||
1626 | static int __init htb_module_init(void) | |
1627 | { | |
87990467 | 1628 | return register_qdisc(&htb_qdisc_ops); |
1da177e4 | 1629 | } |
87990467 | 1630 | static void __exit htb_module_exit(void) |
1da177e4 | 1631 | { |
87990467 | 1632 | unregister_qdisc(&htb_qdisc_ops); |
1da177e4 | 1633 | } |
87990467 | 1634 | |
1da177e4 LT |
1635 | module_init(htb_module_init) |
1636 | module_exit(htb_module_exit) | |
1637 | MODULE_LICENSE("GPL"); |